# RAMIRAN 2017: SUSTAINABLE UTILISATION OF MANURES AND RESIDUE RESOURCES IN AGRICULTURE

EDITED BY : Tom Misselbrook, Francisco Javier Salazar and Claudia Wagner-Riddle PUBLISHED IN : Frontiers in Sustainable Food Systems

### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88963-227-5 DOI 10.3389/978-2-88963-227-5

### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# RAMIRAN 2017: SUSTAINABLE UTILISATION OF MANURES AND RESIDUE RESOURCES IN AGRICULTURE

Topic Editors: Tom Misselbrook, Rothamsted Research, UK Francisco Javier Salazar, INIA, Chile Claudia Wagner-Riddle, University of Guelph, Canada

This eBook presents highlight papers from the 17th International conference of the Recycling of Agricultural, Municipal and Industrial Residues to Agriculture Network (RAMIRAN) that was held in Wexford, Ireland in September 2017. The book contains a broad range of papers around this multidisciplinary theme covering topics including regional and national organic resource use planning, impact of livestock diet on manure composition, fate and utilisation of excreta from grazing livestock, anaerobic digestion, overcoming barriers to resource reuse, hygienic aspects of residue recycling and impacts on soil health. The overarching theme being addressed is the sustainable recycling of organic residues to agriculture, to promote effective nutrient use and minimise environmental impact.

Citation: Misselbrook, T., Salazar, F. J., Wagner-Riddle, C., eds. (2019). RAMIRAN 2017: Sustainable Utilisation of Manures and Residue Resources in Agriculture. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-227-5

# Table of Contents

### *06 Editorial: RAMIRAN 2017: Sustainable Utilisation of Manures and Residue Resources in Agriculture*

Tom Misselbrook, Claudia Wagner-Riddle, Karl Richards, Gary Lanigan, William Burchill and Francisco Salazar

### SECTION 1

### REGIONAL AND LOCAL PLANNING

### *09 Modeling Manure Quantity and Quality in Finland*

Sari Luostarinen, Juha Grönroos, Maarit Hellstedt, Jouni Nousiainen and Joonas Munther

*23 Overview of Animal Manure Management for Beef, Pig, and Poultry Farms in France*

Laurence Loyon

*33 A GIS-Based Approach for the Sustainable Management of Livestock Effluents on Alpine Meadows*

Francesco Gubert, Silvia Silvestri, Angelo Pecile and Luca Grandi

*44 Dairy Manure Management Perceptions and Needs in South American Countries*

María A. Herrero, Julio C. P. Palhares, Francisco J. Salazar, Verónica Charlón, María P. Tieri and Ana M. Pereyra

# SECTION 2

### INFLUENCE OF LIVESTOCK DIET ON MANURE COMPOSITION


Jan Dijkstra, André Bannink, Pieter M. Bosma, Egbert A. Lantinga and Joan W. Reijs

*89 Diet Supplementation With Pomegranate Peel Extract Altered Odorants Emission From Fresh and Incubated Calves' Feces*

Vempalli S. Varma, Ariel Shabtay, Moran Yishay, Itzhak Mizrahi, Naama Shterzer, Shiri Freilich, Shlomit Medina, Rotem Agmon and Yael Laor

### SECTION 3

### EXCRETAL RETURNS BY GRAZING LIVESTOCK

*105 Identifying Urine Patches on Intensively Managed Grassland Using Aerial Imagery Captured From Remotely Piloted Aircraft Systems*

Juliette Maire, Simon Gibson-Poole, Nicholas Cowan, Dave S. Reay, Karl G. Richards, Ute Skiba, Robert M. Rees and Gary J. Lanigan

### SECTION 4

### ANAEROBIC DIGESTION AND USE OF DIGESTATE


Antonio R. Sánchez-Rodríguez, Alison M. Carswell, Rory Shaw, John Hunt, Karen Saunders, Joseph Cotton, Dave R. Chadwick, Davey L. Jones and Tom H. Misselbrook


Ivan Dragicevic, Trine A. Sogn and Susanne Eich-Greatorex

## SECTION 5

### OVERCOMING BARRIERS TO ORGANIC RESIDUE RE-USE


# SECTION 6

### HYGIENIC ASPECTS AND SOIL HEALTH


Stephen Nolan, Nicholas R. Waters, Fiona Brennan, Agathe Auer, Owen Fenton, Karl Richards, Declan J. Bolton, Leighton Pritchard, Vincent O'Flaherty and Florence Abram

### *224 Pollution of Surface and Ground Water by Sources Related to Agricultural Activities*

Nada Sasakova, Gabriela Gregova, Daniela Takacova, Jana Mojzisova, Ingrid Papajova, Jan Venglovsky, Tatiana Szaboova and Simona Kovacova

### *235 Improvements in the Quality of Agricultural Soils Following Organic Material Additions Depend on Both the Quantity and Quality of the Materials Applied*

Anne Bhogal, Fiona A. Nicholson, Alison Rollett, Matt Taylor, Audrey Litterick, Mark J. Whittingham and John R. Williams

# Editorial: RAMIRAN 2017: Sustainable Utilisation of Manures and Residue Resources in Agriculture

Tom Misselbrook <sup>1</sup> \*, Claudia Wagner-Riddle<sup>2</sup> , Karl Richards <sup>3</sup> , Gary Lanigan<sup>3</sup> , William Burchill <sup>3</sup> and Francisco Salazar <sup>4</sup>

*<sup>1</sup> Rothamsted Research, Okehampton, United Kingdom, <sup>2</sup> School of Environmental Sciences, University of Guelph, Guelph, ON, Canada, <sup>3</sup> Teagasc, Johnstown Castle, Wexford, Ireland, <sup>4</sup> Instituto de Investigaciones Agropecurarias, INIA Remehue, Osorno, Chile*

Keywords: organic residues, livestock manure, manure management, digestate, circular economy

### **Editorial on the Research Topic**

### **RAMIRAN 2017: Sustainable Utilisation of Manures and Residue Resources in Agriculture**

The recycling of organic residues deriving from on-farm (e.g., livestock manure) or off-farm (e.g., sewage sludge, industrial by-products) is a central part of the circular economy toward developing more sustainable food production systems (e.g., EC, 2014). However, the safe, effective, and efficient use of organic "waste" streams as resources for nutrient provision and soil improvement in agricultural systems require several challenges to be addressed, summarized by Bernal (2017) as (i) to improve nutrient availability and soil cycling; (ii) to develop technologies for nutrient re-use; (iii) to reduce contaminants and improve food safety; (iv) to mitigate environmental emissions; and (v) to enhance soil health and function. Addressing these challenges needs multidisciplinary research within a whole systems context.

The "Recycling of Agricultural, Municipal and Industrial Residues to Agriculture Network" (RAMIRAN) is a research and expertise network focusing on environmental, hygienic, and agronomic issues associated with the use of livestock manure and other organic residues in agriculture, and as such is well-positioned to address these grand challenges, and indeed has been doing so for some years (Misselbrook et al., 2012). The network evolved in 1996 as an expansion of the previous more narrowly focused FAO Animal Waste Network, initially as a predominantly European network but more recently with a much more global make-up and remit. The main aims of RAMIRAN are to promote the exchange of methodologies, materials, and processes; to progress knowledge relating to agronomic, environmental, and hygienic aspects of organic residue recycling in agriculture; to identify research priorities and initiate innovative collaborative activities that make use of the synergies within the international network. One of the main activities of RAMIRAN is to hold a regular international scientific conference. The papers in this Research Topic derive from the 17th International conference, RAMIRAN 2017, held in Wexford, Ireland in September 2017, with contributions at the conference from over 30 countries representing 6 continents.

Livestock manure is a key organic resource for re-use in agriculture, although the increasing specialization and spatial disconnect between livestock and arable production in many parts of the world, together with the comparatively low cost and ease of use of synthetic fertilizers, has resulted in this resource not being effectively utilized and often treated as a waste. A good understanding therefore of the types, quantities, and composition (particularly of plant available nutrients) for this resource are an important first step in planning for their improved utilization at a regional and local scale. There are challenges in collecting accurate data on this at the national scale, as discussed by Luostarinen et al., who provide a case study of a model approach for Finland and argue for a greater international harmonization in approaches. An example of a national data

### Edited and reviewed by:

*Maria Pilar Bernal, Spanish National Research Council, Spain*

\*Correspondence: *Tom Misselbrook tom.misselbrook@rothamsted.ac.uk*

### Specialty section:

*This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems*

> Received: *30 August 2019* Accepted: *09 September 2019* Published: *24 September 2019*

### Citation:

*Misselbrook T, Wagner-Riddle C, Richards K, Lanigan G, Burchill W and Salazar F (2019) Editorial: RAMIRAN 2017: Sustainable Utilisation of Manures and Residue Resources in Agriculture. Front. Sustain. Food Syst. 3:80. doi: 10.3389/fsufs.2019.00080* set is given by Loyon who provides statistics on the types of manure and practices used to store and land-apply the 120 million tons of manure produced per year by cattle, pigs, and poultry in France. Land availability, suitability, and requirement for crop nutrients are also key factors in the planning of effective organic resource utilization and Gubert et al. present a GIS-based approach to the management of digestate spreading on Alpine hay meadows which could have much broader applicability. A manure management survey on a broader regional basis, for South America, specifically for dairy manure was undertaken by Herrero et al. who highlight the generally positive perception of dairy manure as a good fertilizer but also the barriers to achieving effective management. The importance of policy and other stakeholder interventions to improve nutrient use efficiency and reduce environmental pollution is also discussed.

The impact of livestock diet on manure composition, and N excretion in particular, is important to predict subsequent impacts on nitrogen losses and transformations through the manure management chain. The ability to model this impact at the national and farm-scale is crucial both for providing accurate estimates in national emission inventories and for assessing potential dietary strategies for emission mitigation. Bannink et al. describe an improved approach to estimating nitrogen excretion in dairy cows, and in particular the urinary nitrogen excretion, in the Dutch ammonia emission inventory based on a Tier 3 method for enteric methane prediction. A range of dietary strategies for dairy cows and their impact on nitrogen excretion and subsequent nitrogen utilization and losses at the farm scale were assessed by Dijkstra et al., who present developments to an existing model to provide better representation of manure nitrogen availability to crops. Varma et al. demonstrate, however, the unintended consequences that may occur through dietary manipulation, reporting an increase in the emissions of odorous volatile organic compounds from manure from calves fed a diet supplemented with pomegranate peel extract intended to improve calf health.

Recycling of nutrients through the direct excretal returns of grazing animals reduces the requirement for additional nutrient application to the pasture, but the concentration of nutrients, nitrogen in urine patches in particular, poses a risk of losses to the environment (Selbie et al., 2015). A novel remote sensing method for mapping the spatial distribution of urine patches is described by Maire et al. which could be of great use in assessing the influence of different grazing management strategies or other interventions on predicted nitrogen losses and loss risk mapping.

Anaerobic digestion of organic residues as an energy source, waste treatment process and source of potentially valuable organic fertilizer has increased greatly in recent years. For economic viability, the gas yield of the process is of great importance and as described by Chiariotti and Crisà, who investigated specific inoculum for hydrogen production from livestock by-products, are influenced by the archeal and bacterial community composition. Improving the agronomic benefits of the resulting digestate from the anaerobic digestion process through acidification (Sánchez-Rodríguez et al.) or solids separation (Ehmann et al.) are discussed as is the benefit of summer fertigation of dairy slurry compared with autumn injection to cropping systems at risk of nitrate leaching (Gamble et al.). The issue of potential increased heavy metal availability and uptake by plants following applications of digestate to crops is addressed by Dragicevic et al., who report that application of biogas digestates had little effect on plant metal uptake or crop quality.

While livestock manure and digestates represent the majority of organic residue returns to land, there are a range of other materials which can potentially be used. One barrier to their uptake is heterogeneity in both the material itself and in the application of it to the land, giving less confidence in the agronomic benefits compared with the application of synthetic fertilizers. Technologies to improve the consistency of the materials and the precision of their application are therefore required. Toward this, Delin et al. assessed the optimum precise placement of pelleted meat bone meal for a spring oats crop, which showed significant yield benefits over surface broadcast application. Another potential barrier is the concern regarding safe use of materials on crops for either animal feed or human food production. Source separated human urine represents a valuable source of major and micronutrients for application to crops (Vinneras and Jonsson, 2002), but there are concerns regarding the presence of pharmaceuticals and hormones in human urine. Results from the field study reported by Viskari et al. using human urine as a fertilizer for barley, suggest that source separated urine can be safely used as a fertilizer, with no pharmaceuticals or hormones detected in soil or barley grain, despite being present in the applied urine.

Risk of contamination is also a concern addressed in papers by Ashekuzzaman et al., investigating the potential transfer of E. coli from slurry and biosolids application to grazing land, and Nolan et al., studying the survival and fate of pathogens through anaerobic digestion. At a broader scale, Sasakova et al. consider the risks of transfer of microbial pathogens from agricultural activities to surface and ground waters in a catchment in Slovakia and the appropriateness of current regulations.

Finally, the importance of soil health in enabling good nutrient use efficiency and the potential impacts that organic residue applications may have on this are addressed by Bhogal et al., who show that the quantity and quality of the material applied influence the level of improvement in soil organic carbon content, and by association soil biological and physical properties.

Undoubtedly there is still much research to be conducted in this area and the relevance of RAMIRAN is as great now as it ever was in addressing the challenges of sustainable recycling of organic resources to agriculture, of enhancing the circular economy and minimizing the environmental footprint of our food production systems.

### AUTHOR CONTRIBUTIONS

TM, CW-R, KR, GL, WB, and FS all contributed to the writing of this editorial.

### ACKNOWLEDGMENTS

The authors would like to thank all contributors to this Research Topic, particularly the many reviewers who helped to improve

### REFERENCES


the quality of the scientific papers, and to the organizers of the RAMIRAN 2017 conference in Wexford in September 2017 from which these papers derive. The financial sponsorship received for RAMIRAN 2017 is gratefully acknowledged.

Vinneras, B., and Jonsson, H. (2002). The performance and potential of fecal separation and urine diversion to recycle plant nutrients in household wastewater. Bioresour. Technol. 84, 275–282. doi: 10.1016/S0960-8524(02)00054-8

**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Misselbrook, Wagner-Riddle, Richards, Lanigan, Burchill and Salazar. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Modeling Manure Quantity and Quality in Finland

Sari Luostarinen<sup>1</sup> \*, Juha Grönroos <sup>2</sup> , Maarit Hellstedt <sup>3</sup> , Jouni Nousiainen<sup>1</sup> and Joonas Munther <sup>2</sup>

<sup>1</sup> Natural Resources Institute Finland (Luke), Jokioinen, Finland, <sup>2</sup> Finnish Environment Institute SYKE, Helsinki, Finland, <sup>3</sup> Natural Resources Institute Finland (Luke), Seinäjoki, Finland

Data on manure quantity and quality is a prerequisite for planning manure management and regulation. It is the basis for directing manure use into more efficient and environmentally sound actions and for fulfilling the targets of nutrient recycling in a circular economy. Manure data is often scarce, old or badly documented. Some collect it by sampling and analysis, others with calculation systems/models. In Finland, both options are used. The farmers need to have their manure analyzed at least every 5 years. The resulting analyzed data from the farms can be combined into a statistical report on manure quality. However, this dataset has major shortcomings, such as difficulty to identify different animal categories. Thus, a model called the Finnish Normative Manure System was developed. Technically the system works well and its biggest challenges are related to the vast amount of background data needed. There are still data gaps e.g., in bedding use and cleaning water additions and a need to update the excretion calculations. To assist development of such models, international harmonization of the methods would be beneficial. As such manure data is usually the basis for emission inventories and burden sharing, harmonization would also place farms and countries in a more equal position in international contracts on emission reduction. In this paper, the challenges related to manure data provision are discussed in reflection to the experiences gained during the development of the Finnish Normative Manure System.

### Edited by:

Tom Misselbrook, Rothamsted Research (BBSRC), United Kingdom

### Reviewed by:

Theun Valentijn Vellinga, Wageningen University and Research, Netherlands Harald Menzi, Federal Office for the Environment, Switzerland

\*Correspondence:

Sari Luostarinen sari.luostarinen@luke.fi

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 22 February 2018 Accepted: 24 August 2018 Published: 19 September 2018

### Citation:

Luostarinen S, Grönroos J, Hellstedt M, Nousiainen J and Munther J (2018) Modeling Manure Quantity and Quality in Finland. Front. Sustain. Food Syst. 2:60. doi: 10.3389/fsufs.2018.00060 Keywords: manure, manure data, mass balance, model, normative

# INTRODUCTION

Circular economy is being promoted worldwide and e.g., the European Union has a dedicated strategy and stakeholder platform for its increased implementation (COM/2014/0398 final/2, 2014). One important aspect in a circular economy is nutrient recycling. Instead of focusing on mining phosphorus and fixing atmospheric nitrogen as mineral fertilizers, the nutrients already in use should be recovered and reused efficiently.

The main user of nutrients is agriculture. Fertilizers are needed to ensure high yields in crop production which in turn enables efficient food production. Manure has always been used as a recycled fertilizer in agriculture. However, in the developed world, farms usually specialize into either animal or crop production and manure reuse as a fertilizer becomes mainly restricted to animal farms alone. While animal production has simultaneously often concentrated on certain areas and crop production on others, manure reuse has further been limited to only some regions. Excess areas for manure have been established, based on the ratio between production animals and available land at the farm or based on high phosphorus levels in field soils due to excessive

fertilization in the past. Such hot spots contain an increased risk for manure-related harmful environmental impacts. In case such regional concentration of animal farms continues, manure should be transported from these areas to those in need of recycled nutrients to effectively replace mineral fertilizers.

To plan efficient manure management and use on farms and to drive manure use into a more sustainable direction on regional, national and international scales, information on manure quantity and quality is needed. Still, data on manure quantity, quality, regional location and management practices are scarcely systematically collected, reported and updated in many countries. Furthermore, if such data are even partially available, it is usually reported in national languages only. Clear documentation of the data collection methods and results is rarely available and international databases do not exist. There are no joint international guidelines for the methods of collecting such manure data, except for some parameters related to gaseous nitrogen and greenhouse gas emissions and nutrient balances (IPCC, 2006; Eurostat, 2013; EMEP/EEA, 2016). This results in a situation in which some countries have sophisticated methods to measure and/or to calculate national manure quantity and quality, while others continue to use outdated values or do not have much data at all (e.g., Luostarinen and Kaasinen, 2016).

In Finland, a more systematic data collection on manure was initiated in 2008. The growing attention to the need to reuse manure more efficiently led to a situation in which the previously used, oldish and largely expert-estimated information did not suffice. The two governmental research institutes, Natural Resources Institute Finland (Luke, formerly MTT Agrifood Research Finland) and Finnish Environment Institute (SYKE), responsible e.g., for national emission inventories and nutrient balances, took action on several levels.

Firstly, the information on analyzed manure samples from farms were obtained from two important commercial laboratories and new average values for manure nutrient content were drawn. This information (so-called table values) is now an option to farm-specific manure analysis for planning manure fertilizer use as regulated in the national decree (1250/2014/FI) executing the EU Nitrates Directive (91/676/EEC). Secondly, a survey to determine the current manure management practices on farms was developed and implemented in 2013–2014 (e.g., Grönroos et al., 2017). Thirdly, the manure storage capacity requirements per animal category were roughly re-calculated and updated to the national decree of 1250/2014/FI.

During these tasks, the need for a proper calculation system to provide systematically updated manure data for all manurerelated regulation and activities became apparent. Luke and SYKE decided to develop a calculation model for manure quantity and quality called the Finnish Normative Manure System. The development of the large model was and still is a challenging task, but the system has already proved its usefulness in providing versatile manure data needed in several contexts from regulation of manure management to national and regional plans of enhancing manure use. In the future, the use of the system as a basis of manure fertilization will also be discussed.

In this paper, we summarize the basics of the Finnish Normative Manure System, give some results as an example of what kind of data it can deliver, and discuss the challenges and strengths of the approach. We also proceed to the need of international dialogue on how such models should be built and used to ensure equality between farmers and countries under the increasing obligations to promote nutrient recycling and to control agricultural emissions. The goal of the paper is to raise discussion on whether harmonized methods for modeling manure quantity and quality are needed and how the scientific community using such models could better learn from each other.

# MATERIALS AND METHODS

A short overview of the Finnish Normative Manure System is presented here. For more detailed information, two documentation reports are available online (cattle, pigs, poultry, horses, and goats in English: Luostarinen et al., 2017a; fur animals in Finnish: Luostarinen et al., 2017b).

The Finnish Normative Manure System comprises five interlinked units, all built in MS Excel <sup>R</sup> (**Figure 1**). Unit 1 calculates excretion, i.e., the quantity and quality of feces and urine excreted by each animal category included. Unit 2 collects data on the additions such as bedding and water into manure, and the losses of dry matter and water. Unit 3 calculates gaseous losses from manure during housing and storage for each animal category and manure type considered. Unit 4 is responsible for the actual mass balance calculation for each animal category and manure type, the results of which are organized into clear tables and graphs in Unit 5.

The most important animal categories (**Table 1**) are calculated e.g., for agricultural emission inventories. More specific animal categories are also included (**Table 2**) to serve the needs for more detailed information. This may be required e.g., in case of a farm specifically focused on rearing indigenous cattle which are smaller and produce less manure than high-yielding or even average Finnish cattle.

Furthermore, to enable calculation of manure quantities and their nutrient content per certain region or the whole nation, the calculated manure data are multiplied by annual statistics on animal numbers also containing the information on the geographic location of the farms. The results presented in this paper are based on 2016 animal statistics (Finnish Fur Breeders' Association, 2016; Suomen Hippos/Finnish Trotting Breeding Association, 2016; Luke Statistics, 2017).

The calculations proceed as a mass balance (see an example of its main components as a flow chart of slurry; **Figure 2**). Firstly, excretion is calculated as animal feed minus retention in animal, resulting in the amount of excreted feces and urine and their composition regarding nitrogen and the share of total ammoniacal nitrogen (TAN), phosphorus, potassium, dry matter, organic matter. Animal growth, reproduction, genetic type, and product yield (milk, meat, eggs, pelt) are averaged for the production conditions in Finland and according to national feeding recommendations (Luke Statistics, 2014a,b,c,d; Finnish Fur Breeders' Association, 2016; Luke Feed Tables, 2016). The excretion presented here is based on the production and feeding

TABLE 1 | The main animal categories in the Finnish Normative Manure System and used in the emission inventories.


<sup>a</sup>An average of farrowing, gestating and mating sows + piglets until weaning

data from 2014 (Luostarinen et al., 2017a), except those of fur animals which are based on data from 2016 (Luostarinen et al., 2017b).

Secondly, the additions of bedding materials and cleaning water during animal housing are considered per manure type (Luostarinen et al., 2017a,b). The manure types calculated include slurry, farmyard manure, deep litter and separately collected dung and urine. Slurry means a mixture of feces and urine into which cleaning waters of the animal house (and milking equipment) are directed and very little bedding is used. Its dry matter content is low (<12%). Farmyard manure is a solid mixture of feces and bedding into which all urine is adsorbed. It is removed from the animal house regularly (daily). Deep litter is also a solid manure with an even higher content of bedding as it is removed from the animal house only after a production batch (e.g., broilers) or once a year. All urine is adsorbed into the bedding which is added regularly on top of the manure bed. In a housing unit with separately collected dung and urine, some bedding is used and part of the urine may be adsorbed into it producing thus a solid manure type called dung. Most of the urine is, however, collected into a separate storage tank using tilted flooring and urine channels.

The share of these manure types differs between different animal categories (**Table 3**) and is considered when calculating e.g., national manure data. Further, the calculation considers the shares of grazing affecting the amount of manure collected inhouse (**Table 3**). It also uses the shares of manure storage options (different covering) affecting the gaseous losses during manure storage (**Table 4**).

Further, rainwater addition to open manure storages is considered as the annual average precipitation in Finland of 600 mm. Evaporation of water from the storages is considered as the mean annual evaporation rate, being 300 mm for open storages and 100 mm for slurry storages with floating covers. For solid manures (farmyard manure, deep litter, dung), evaporation of water is adjusted according to nationally analyzed dry matter content of manures, as suggested by Poulsen and Kristensen (1998). Also, 10% of dry matter is assumed to be lost due to conversion of organic matter in housing with deep litter systems (Poulsen and Kristensen, 1998) and 10% during storage of all manure types, except urine. For fur animals, 5% of dry matter is assumed lost under the open sheds TABLE 2 | Detailed animal categories included into the Finnish Normative Manure System.


farmyard manure, separately collected dung) do not receive cleaning water, but may receive feces from outdoor yards (see: flow charts per manure type in Luostarinen Luostarinen et al., 2017a).


TABLE 3 | The share of different manure types produced and grazing data in Finland (% of manure per each manure type and animal category; Grönroos et al., 2017).


and 15% during storage. Mineralization and immobilization of nitrogen are also included into the calculation as described in Grönroos et al. (2017).

The loss of gaseous nitrogen (NH3, NOx, N2O, N2) is calculated using the Finnish model for agricultural emissions of gaseous nitrogen and non-methane volatile organic compounds (Grönroos et al., 2017). The calculation follows the instructions of EMEP/EEA (2016) using the total ammoniacal nitrogen (TAN) excreted by each animal as the starting point. Calculation of nitrous oxide and methane emissions follow the principles of IPCC (2006). The amount of carbon dioxide released during manure management is estimated with the method developed by Hamelin (2013).

The reporting unit offers several types of results, examples of which are given in the next section of this paper. Specific results per animal or animal place give values for manure ex animal (excretion), manure ex housing (manure leaving the housing unit in different manure types) and manure ex storage (manure leaving storage and to be applied on fields; **Figures 2**, **3**). Results ex animal include the quantity of feces and urine, and the quantity of nitrogen, phosphorus, potassium, dry matter and organic matter in them [kg/animal(place)/year]. Results ex housing and ex storage include the quantity of the relevant manure types per animal category [t/animal(place)/year], the quantity of nitrogen, ammonium-nitrogen, phosphorus, potassium, dry matter and organic matter (t/animal(place)/year). These results are also presented as concentrations (kg/t of manure). Results per chosen animal population, e.g., all animals in Finland, can also be calculated by multiplying the animal-specific results with animal statistics.

### RESULTS

In this section, some results are presented as an example of what kind of data the Finnish Normative Manure System produces and how the data can be used. For all current datasets per animal category available, the readers should download the documentation reports (cattle, pigs, poultry, horses, and goats in English: Luostarinen et al., 2017a; fur animals in Finnish: Luostarinen et al., 2017b). It is stressed that the development and data collection processes for the system are still ongoing and these are only the first results provided. During this development, more comparisons to analyzed manure data will be made to further validate the model. Some comparisons can

### TABLE 4 | The shares of different storage types and measures used in Finland (Grönroos et al., 2017).



be found in the documentation reports (Luostarinen et al., 2017a,b).

Examples of the animal-specific results are given here for an average Finnish dairy cow (**Table 5**) and broiler (**Table 6**). Similar result tables have been published for most of the animal categories listed in **Table 1** (Luostarinen et al., 2017a,b). The results with more detailed animal categories (**Table 2**) are also available but not yet published.

The results are given per animal place and per year (**Tables 5**, **6**). For cattle, the animal places are usually occupied all year (**Table 5**). However, in case of animals reared in production batches, such as broilers, it should be noticed that the animal place is not occupied all year and the result calculated per full year should be multiplied by 0.65 to consider production pauses (**Table 6**).

While with broilers the only manure type produced is deep litter (**Table 6**), the results for dairy cow (**Table 5**) are given as alternatives per different manure types as all, excluding deep litter, are produced in Finland (**Table 3**). Deep litter is still calculated to compare to other cattle categories. Furthermore, the dairy cow feces collected from exercise yards is added to the manure ex storage, while the share of feces and urine excreted on pasture (**Table 3**) is excluded from the results.

It should also be noted that with dairy cow, bedding materials are added to all manure types during housing (Luostarinen et al., 2017a) and they add to the manure quantity and alter the content of dry matter, organic matter and nutrients. Also cleaning waters from housing and milking equipment dilute slurry and increase its quantity. Similar changes due to bedding addition can be noticed with broiler deep litter. Changes also occur due to addition of rain water in manure ex storage, evaporation of water from deep litter during housing and loss of dry matter and nitrogen during manure management.

National totals of manure quantities are also available, calculated here with the animal statistics of 2016 (ex housing: **Table 7**, ex storage: **Table 8**). Similarly, e.g., total nitrogen and total phosphorus can be calculated for all manures in Finland (example of manure ex storage in **Figure 3**). Such results can also be produced for certain regions, such as provinces and municipalities. The information provided offers an insight into the practical shares of different manure types, and their nutrient content and locations. This supports the planning of their more efficient utilization as such or with processing into new organic fertilizers.

The calculation system functions well. The results on different levels are easily recovered, a function which has not been available in Finland previously. The information provided is based on the best and most updated background data available, thus merging large amounts of data into the type of results which can be used in several different manure-related activities in a harmonized form.

The changes in manure quantity and quality along the manure management chain are easily recognized when comparing the results ex animal, ex housing, and ex storage. Such data has not previously been available in Finland. The information is important e.g. when planning manure processing plants into which manure is usually fed as fresh as possible. Thus, the results ex housing should be used as the basis of all planning. The losses of organic matter and nitrogen in the manure management chain deserve special attention to highlight the need to minimize them with the right actions and to ensure as high a dose of both into the fields as possible to maintain soil organic matter and offer nitrogen for crop

growth. By altering the calculation e.g., by implementing a higher share of covered storage than actually used at the moment, the difference in manure nitrogen content can be determined.

### DISCUSSION

The first published version of the Finnish Normative Manure System (Luostarinen et al., 2017a,b) is discussed here in relation to its original need, experiences during its construction and its ongoing development. Additionally, comparison between using sampled and analyzed manure data and calculated manure data in different functions is considered. Ultimately, the need to develop international guidelines for more harmonized methodologies for providing manure data is discussed.

# The Finnish Normative Manure System and its Uses

The first version of the Finnish Normative Manure System has proved to easily provide the manure data needed in many actions from policymaking to farming. Technically it works well and can be updated fairly simply. However, the requirement for rather detailed background data, existence of some important data gaps and the rather complex MS Excel <sup>R</sup> structure still call for development. Some phenomena appearing during manure management, such as loss of dry matter, is also on the development list.

The data provided by the system has already been used in several functions in Finland. It is coupled with the inventory of air pollutant emissions from agriculture. It is also the best available information on manure quantity, quality and location (when combined with animal statistics), a dataset which is used


ap, animal place; DM, dry matter; OM, organic matter; Ntot, total nitrogen; Nsol, soluble nitrogen; Ptot, total phosphorus;

 Ktot, total potassium; VS, volatile solids (organic matter).

as the basis for planning more effective manure use including e.g., different manure processing options. It provides manure data for an open data source about different organic wastes and byproducts from agriculture, forestry, municipalities and industries in Finland called Biomass Atlas (https://www.luke.fi/biomassaatlas/en/) and for a planning tool for regional nutrient recycling to be used by regional authorities (ready for use in 2018).

In the future, the system will provide manure data for national nutrient balance calculations and updated information e.g., for the requirements of manure storage capacity and animal-specific coefficients determining the threshold number of animals for environmental permitting of animal farms. Further uses could include being the basis for manure fertilization plans (instead of current values derived from large datasets of analyzed manures). The ultimate aim is to harmonize the national manure data used by policymakers, authorities, research, business, education, agricultural advisors, and farmers.

### Development Needs of the Finnish Normative Manure System

At the time of writing, the most important development need in the system is the excretion calculation. Excretion has the largest effect on manure quantity and quality within the system. Yet, there are no harmonized guidelines on how it should be calculated. The need for harmonizing excretion calculation has also been noted elsewhere, especially in relation to nitrogen excretion (Velthof et al., 2015). Also, the difficulty of such harmonization has been noted as it may not be possible to simply copy the calculation system of one country to another. The calculation procedure must always reflect the country-specific animal production. Thus, the role of background data on feeding, growth, product yield, reproduction etc. becomes integral.

In Finland, excretion calculation is the responsibility of Natural Resources Institute Finland (Luke). Before the development of the Finnish Normative Manure System, basically only excretion of nitrogen, phosphorus and organic matter was needed. With the introduction of the Normative Manure System, the parameters to be calculated were increased to quantity of feces and urine and the quantity of total nitrogen, total phosphorus, total potassium, dry matter and organic matter in both. It soon became apparent that the national excretion calculation requires a larger reorganization which is now proceeding in Luke. The task is large and will take some time. Thus, the first results given by the Normative Manure System are not fully documented and subject to change due to introducing the updated excretion calculation.

During the update of excretion calculation, the background data concerning animal production, including actual animal numbers reared and their feeding, growth, and reproduction will all be updated in cooperation with farmers' representatives, feed producers and food industry. It is vital that the information used relates to the actual current practices on farms. This calls for comparisons between using feeding recommendations (Luke Feed Tables, 2016) and feeding information collected on farms. There have been concerns over whether using feeding recommendations as the background data really represents the



TABLE 7 | The manure quantity ex housing in Finland based on the manure data of the Finnish Normative Manure System and animal statistics of 2016.

The manure excreted on pasture is excluded.

TABLE 8 | The manure quantity ex storage in Finland based on the manure data of the Finnish Normative Manure System and animal statistics of 2016.


The manure excreted on pasture is excluded.

feeding used on farms in practice. This is of special interest especially for cattle, the feeding of which is not quite as controlled and coming largely from the feed industry than e.g., with poultry. A separate research project to study this will be conducted during 2018–2019.

Also the animal categories to be calculated will be checked to enable all relevant types of animals to be included. Further, the actual calculation procedure with its algorithms will be re-evaluated and updated. Necessary documentation in English will also be prepared.

A large data gap in the Finnish Normative Manure System is the information on bedding materials and cleaning waters added into different manure types under different housing systems. This information has not been collected for years and the attempt to collect it mostly failed during the 2013 manure management survey on animal farms, due to the farms having problems with estimating their bedding use, Clearer data on bedding use in poultry production and horse stables were received from separate data collections with simplified questions (horses) or direct contact with the farms and their representatives (poultry). An important obstacle for the data collection was that the farms rarely document their bedding use. It may also change depending on bedding material availability and price. Similarly, cleaning water use and its direction to slurry is not usually measured. For the calculation, this is a major shortcoming.

Losses of nitrogen, methane and carbon dioxide are rather straightforward to calculate due to them being based on the international guidelines (IPCC, 2006; EMEP/EEA, 2016). However, the loss of dry matter is currently based on a very rough estimation for deep litter during housing and for all manure types except urine during storage. Better solutions have not been found, but might call for international discussion on how this should be taken into account in manure-related mass balance calculations. Further, the evaporation of water from solid manure types is currently estimated based on manure analysis results as suggested by Poulsen and Kristensen (1998). This method should also be subjected to further discussion and potentially also measurements.

### Use and Limitations of Manure Data Collected by Manure Analysis

In Finland, farmers are required to take manure samples minimum every 5 years and have it analyzed according to methods approved in national legislation (1250/2015/FI). Analyses are done in commercial, accredited laboratories. One of the laboratories have compiled and published averages over 5 year periods (Eurofins Agro Finland, 2017). However, the dataset has not been subjected to statistical analysis. It gives the average, minimum, maximum, and number of samples. From these, it is usually obvious that there are outliers that should be removed from the data set. If, e.g., a slurry maximum for dry matter is 50%, it clearly is not slurry at all. Such statistical analysis would improve the quality of the data set.

The dataset is also not a particularly accurate generalization of manure quality in Finland. The samples are often poorly named by the farmers, which makes it often impossible to distinguish from which animals the sample derives from. It makes a big difference whether a sample is named "poultry manure" or "broiler manure" or even just "manure." Further, there is often inconsistent or no data on the manure type, either. At the time of writing, the laboratories do ask whether the manure sample is from slurry, deep litter, solid manure (meaning farmyard manure or dung) or urine. Categorization of the data into cattle slurry, solid manure and urine, pig slurry, solid manure and urine, poultry manure, horse manure, fur animal manure, and sheep and goat manure should thus be possible in the future. The differences between e.g., dairy and beef cattle, fattening pigs and sows, and horses and ponies still cannot be distinguished.

The ratio between different animal categories on the farms is not collected. Even if the sample would be named as accurately as "dairy cattle slurry," there is no information on how many dairy cows, heifers and calves there were on the farm. All of these animals produce different types of manure depending on their feeding, growth, yield and housing solutions. Also, no data on animal breed, feeding, product yield, bedding material and cleaning water use are connected to the samples. Such background data would improve the usability of the data in different contexts e.g., from official table values for manure nutrient content to different models for manure use and manurerelated emissions.

At the time of writing, there are three commercial laboratories analyzing manure samples in Finland. They use somewhat different analysis methods. It is unclear whether this affects the results and whether some method would be more suitable for manure samples than others. During the development of the Finnish Normative Manure System, some manure samples were taken by the researchers involved and sent to two different laboratories for analysis of dry matter and nutrients. Most results were similar, but e.g., the phosphorus content in pig slurries differed between the two laboratories notably. They were on average 0.6 g P/kg in laboratory 1 and 0.3 g P/kg in laboratory 2 for sows and 1.0 and 0.8 g P/kg, respectively, for fattening pigs (data not shown). With solid manures the difference in sow manure was smaller (laboratory 1: 4.7 g P/kg; laboratory 2: 4.5 g P/kg). A more in-depth analysis of the reasons for such differences was not then made, but it appeared that both the analysis method and the manure type can affect the results. For phosphorus, laboratory 1 used a method described in Plaami and Kumpulainen (1991) and laboratory 2 used standard methods (ISO 5516:1978).

Another factor affecting the analysis results is the heterogeneous nature of manure. As the sample volume used in the measurements is small, the possibility for unrepresentative sampling during both analysis and farm sampling is always present. In Finland, the farmers take the samples themselves. They are offered instructions for sampling by the commercial laboratories, but it is not really known how well the guidance is followed. Poor sampling may thus affect the results per farm and subsequently also the datasets drawn from a larger amount of analysis results.

All in all, manure sampling and analysis may provide feasible data for the farms, if sampling and analysis are representative. Still, it describes only that moment in time on that specific farm. Manure collected e.g. over the winter period and stored for months is inevitably different from the manure collected over a month or two in-between spring and summer spreading. Also, changes in bedding use, manure collection frequency and cleaning water use change the manure. The Finnish required minimum frequency of analysis, every 5 years, may be too long to respond to changes in farm practices. Furthermore, to use such analyzed data for the purposes of emission inventories or other manure data aiming at generalizing the manure management in all Finland is inevitably difficult. The uncertainties listed and the lack of separate data for the specific animal categories render the analysis results unfeasible. It may help in validating the manure calculation so that the results provided can be accepted in their uses.

### Uses of the Data Produced by the Normative Manure System

The data produced by the Finnish Normative Manure System is not exactly the manure produced on individual farms. It uses background information which is an average of the animal production in Finland per animal category. Manure data is needed in many other actions than planning and reporting manure fertilization on specific farms. Due to the challenges in analyzed manure data stated above, not all required data can be collected with sampling and analysis. More useful average national manure data is received via the calculation system.

One important aspect for manure management and use is manure quantity per animal or animal place. This is usually not measured on farms, yet the information is needed when planning and building sufficient storage capacity for the farms. The storage capacity requirements are part of regulation for manure management on farms and set according to a calculated estimation of manure quantity per animal category and manure type.

Further, to distinguish between the different manure types per animal category and to couple this information with animal statistics and animal locations in emission inventories and nutrient balance calculations can only be accomplished via a calculation model. The model also needs manure data from different stages of the manure management chain. They often start from excretion, thus needing the data ex animal. The inventories for gaseous emissions estimate the losses happening afterwards along the manure management chain under average national production conditions per animal category. Thus, they actually already make up part of the mass balance system for calculating also other manure parameters.

Many countries are setting targets for improved nutrient recycling including enhanced manure use. Finland declared its goal to become a model country for nutrient recycling already in 2011 (Ministry of Agriculture and Forestry, 2011). This is also part of the Government Programme of the Finnish Prime Minister Juha Sipilä (Prime Minister's Office Finland, 2015) as one of the key projects aims at a breakthrough to a circular economy and adoption of clean solutions.

To strive toward reaching these targets, information on how much nutrient-rich, recyclable wastes and by-products are produced in Finland, what is their nutrient content and where they are located, is required. To produce this information for manure, the Finnish Normative Manure System has proved a feasible tool. In the spring of 2017, a background paper was made on the current situation of nutrient recycling in Finland (Marttinen et al., 2017). The manure data was derived from the Normative Manure System. The significance of the system and the information it provides becomes clear when comparing the volumes of nutrients to be recycled in different wastes and by-products. Manure makes up over 80% of the fresh mass and approximately 75% of phosphorus and 90% of nitrogen in the different wastes and by-products available (including also unused grass biomasses, sewage sludge, municipal and industrial biowaste). This clearly outlines the need for updated national manure data which should be produced in a controlled, official system using the latest information available.

Further, there are regions in Finland to which much of the animal production and thus also manure is concentrated (Ylivainio et al., 2015; Marttinen et al., 2017). To fully understand the situation in these regions regarding the availability of manure nutrients and the need for fertilizing, the data provided by the Normative Manure System becomes necessary. The information is the basis for planning potential actions in manure management and evaluating the necessity of processing manure into new fertilizer products which can be more cost-efficiently transported to the regions in need of the nutrients.

There are also a lot of research and development projects on nutrient recycling going on in Finland at the time of writing. They often need manure data due to manure being the most important nutrient-rich material to be recycled. To produce information which can be even to some extent compared and compiled into larger entities, the manure data used should be uniform. This is another very important use of the calculated data. Only in studies focusing on case farms may manure analysis provide more accurate information.

### Need for Harmonized Guidelines for Calculation Methods

Many countries have their own calculation systems for manurerelated data (e.g., Luostarinen and Kaasinen, 2016). Such systems or at least their results are available e.g., in Denmark (Poulsen and Kristensen, 1998) 1 , Sweden (e.g., Gustafson et al., 2007), Germany (DLG, 2005, 2014; BMELV, 2007), Estonia (Põllumajandusministerium, 2014) and the Netherlands (Statistics Netherlands, 2012). Often the systems are based on mass balances. The challenge is that the calculation systems are usually not thoroughly, if at all, documented. This makes their comparison difficult and leaves little room for learning from each other. To develop more harmonized manure data and to ensure equality between countries and farmers e.g., with regard to emission targets and their surveillance, such harmonization is needed. One step toward this would be to document the calculations better and subsequently facilitate discussion between those organizations responsible for the national systems. This is a task that might be good for a RAMIRAN task group to also take forward.

An attempt to find and build more harmonized methods for manure mass balance calculations is being conducted at the time of writing in a project called Manure Standards. It is an Interreg project (Baltic Sea Region Programme, project duration 2017– 2019) deriving from the ministerial level decision to produce manure nutrient standards for the Baltic Sea Region as part of the work of Baltic Marine Environment Protection Commission— Helsinki Commission (HELCOM). The project is led by Natural Resources Institute Finland (Luke) and its partnership comprises of 19 organizations either working on research or representing state authorities, farmers and agricultural advisors in all Baltic Sea countries (Finland, Sweden, Denmark, Germany, Poland, Lithuania, Latvia, Estonia, and Russia).

The project aims at testing and developing both manure analysis and manure calculation by producing joint guidelines on (i) how to take representative manure samples and which analysis methods are the most suitable for different types of manure, (ii) how to make basic manure mass balance calculation ex animal, ex housing and ex storage, and (iii) how to use manure data in nutrient bookkeeping as the basis of manure fertilization and which methods for accounting nutrients can be recommended for following up on manure use on national and transnational levels. The methods are developed in international cooperation and tested with the pilot farms in each country, including assessments of economic and environmental impacts of updating manure data. The policymakers in each Baltic Sea country are also closely involved via cooperation with the HELCOM group on Sustainable Agricultural Practices (Agri). The Agri group has members from the relevant ministries dealing with agriculture and the environment in the Baltic Sea Region, DG Environment of the European Commission and also representation of farmers' unions and environmental NGOs.

The Finnish Normative Manure System is also subjected to commenting by the other experts in the project Manure Standards and to comparison to the systems available in other participating countries. This is seen as an important opportunity to discuss the solutions in the calculation and to improve and harmonize the existing calculation systems. Simultaneously a jointly agreed basic calculation tool will be constructed in MS Excel <sup>R</sup> to be used in those countries which currently do not have any tools for manure mass balances and may also otherwise have rather old or incomprehensive manure data. This tool will also be clearly documented and its use instructed. In the future, the harmonizing work of manure calculation systems should be advanced to an even larger reach than the Baltic Sea Region covered in Manure Standards.

### AUTHOR CONTRIBUTIONS

SL is the main author of this article. She led the development of the Finnish Normative Manure System, participated in the required data collection, and was the main author of the documentation reports. She is now responsible for its further development and annual updating and communication. She also coordinates excretion calculation in Finland. JG was responsible for developing the Excel <sup>R</sup> -based calculation system

<sup>1</sup>Normtal, 2018 Available online at: http://anis.au.dk/forskning/sektioner/ husdyrernaering-og-fysiologi/normtal/ (Accessed February 12, 2018).

for the Finnish Normative Manure System and manages the Finnish calculation model for gaseous nitrogen emission from agriculture. He also participated in the required data collection and co-authored this article and the related documentation reports. MH participated in data collection and overall development of the Finnish Normative Manure System as an expert on livestock housing and manure management. She coauthored the related documentation reports and this article. JN was responsible for the excretion calculations. JM participated in the technical development of the Excel <sup>R</sup> -based calculation system for the Finnish Normative Manure System.

### REFERENCES


### ACKNOWLEDGMENTS

The authors wish to thank the Finnish Ministry of the Environment for funding the development of the Finnish Normative Manure System. We are also grateful for the Finnish Ministry of Agriculture and Forestry for supporting the work and leading the project steering group. We extend the thanks also to the Finnish Farmers' representatives and food companies who cooperated with us and offered their expertise to the background data collection.

02\_Meijerimaidon\_tuotanto\_v.px/table/tableViewLayout1/?rxid=21e38463- 29ab-4674-a639-06e6cd96d065


Ylivainio, K., Sarvi, M., Lemola, R., Uusitalo, R., Turtola, E. (2015). Regional P stocks in Soil and in Animal Manure as Compared to P Requirement of plants in Finland. Natural resources and bioeconomy studies 62/2015. Natural Resources Institute Finland: Helsinki. Available online at: http://urn.fi/URN:ISBN:978- 952-487-505-9

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Luostarinen, Grönroos, Hellstedt, Nousiainen and Munther. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Overview of Animal Manure Management for Beef, Pig, and Poultry Farms in France

### Laurence Loyon1,2 \*

<sup>1</sup> UR OPAALE, Irstea, Rennes, France, <sup>2</sup> Université Bretagne Loire, Rennes, France

Edited by:

Claudia Wagner-Riddle, University of Guelph, Canada

### Reviewed by:

Harald Menzi, Federal Office for the Environment, Switzerland David Fangueiro, Instituto Superior de Agronomia (ISA), Portugal

> \*Correspondence: Laurence Loyon laurence.loyon@irstea.fr

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 18 January 2018 Accepted: 19 June 2018 Published: 10 July 2018

### Citation:

Loyon L (2018) Overview of Animal Manure Management for Beef, Pig, and Poultry Farms in France. Front. Sustain. Food Syst. 2:36. doi: 10.3389/fsufs.2018.00036 Livestock manure management is the central issue for many environmental policies relating to water and air quality. However, there is little published data on the methods used in those countries affected by pollution from the livestock sector. This paper brings together the available data relating to manure management in France, specifically for pig, cattle and poultry production. An overview of livestock production and legislation is presented using data from the 2010 Agricultural Census, livestock farm surveys carried out in 2008 and other supporting documents relating to manure treatment (professional surveys, expert reports and technical publications). Cattle, pig, and poultry livestock produce around 120 million tons of manure per year not including those on pasture. This figure is made up from 60.6% solid manure, 38.8% livestock slurry (effluent) and a relatively small amount of poultry droppings. Solid manure is mainly stored in temporary field heaps. In the case of manure storage on the farm, the capacity varies from 45 days to 7.5 months depending on farm size and type of animals, time spent outside the buildings and the geographical location. Covered storage (whether rigid or natural crust) accounts for 17% of stored pig slurry, 45% of cattle slurry, and 39% of poultry slurry. Covered storage of solid manure is rarely used on pig or cattle farms whereas 27% of the solid poultry manure (including poultry droppings) is held in covered storage areas. Treatment applies to 13.6 million tons of the manure produced, mainly by methods based on composting or aerobic treatment. Nitrogen applied as slurry is mostly spread on the soil surface using splash plate tankers (83% in the case of cattle slurry, 63% for pig slurry, and 66% for poultry slurry). Incorporation within 24 h of the nitrogen spread on the soil concerns 28% of cattle, 44% of pig, and 56% of poultry manure. The most common method of manure management is storage (in building and pit) and spreading. The treatment of manure and the use of specific techniques to reduce gaseous emissions (such as frequent manure removal from buildings, storage covers, or injection) are not widely reported.

Keywords: overview, manure management, livestock, poultry, cattle, pig, France

# INTRODUCTION

Livestock activities have an environmental impact when manure is improperly handled due to the pollution from various nutrients and organic compounds (nitrogen, phosphorous, organic matter. . . ), from the emission of ammonia (to water soil and air) and greenhouse gas emission (GG). The gases emitted (NH3, CH4, N2O) result from the breakdown of animal manure containing carbon and nitrogen and are released in the buildings, during subsequent storage and during land spreading (Chadwick et al., 2011; Webb et al., 2012). Water pollution by nitrates (NO<sup>−</sup> 3 ) or by phosphorous (P) in certain intensive livestock production areas arise from spreading manure rich in N and P beyond the capacity of the land. The surplus nitrogen and phosphorous is not used by the crop or soil and is washed out by surface run-off or seepage leading to eutrophication of water sources (Velthof et al., 2014; van Dijk et al., 2016). As a result, the livestock sector is considered as one of the principal sources of pollution leading to global warming (in the case of GG emission) water and soil contamination and the loss of biodiversity (Steinfeld et al., 2006). The scale of these impacts are thus closely linked to (amongst other factors) the management and composition of animal manure (Menzi et al., 2010; Chadwick et al., 2011; Petersen et al., 2013).

As a consequence, the management of livestock manure is a central issue in a series of international protocols, of European directives and national regulations. Effectively, the practical aspects of the methods chosen by farmers can influence the scale of diffuse emissions and the possibility to reduce these losses (Chadwick et al., 2011; Velthof et al., 2014). The European directive on emission ceilings (EC, 2001) resulting in the Gothenburg Protocol (The United Nations Convention on Long-range Transboundary Air Pollution or CLRTP) (UNECE, 1999) targets the control of ammonia emissions. Those emissions of CH<sup>4</sup> and N2O are regulated by the Kyoto Protocol arising from the UN Framework Convention on climatic change (UN, 1997). Water pollution by nitrates and phosphorous is the subject of the EU Nitrates Directive (EEC, 1991) and the European Water Framework Directive (EC, 2000). The signatory countries of international conventions or those targeted by European Directives must measure the existing level of water pollution and make an inventory of current emissions of the listed gases. These measurements and inventories are thus the reference base for reduction objectives that imply the enactment of action programs.

Various published works relating to gas emission inventories or the movement of N and P, underline the need for detailed data due to the large variability of management methods in livestock production. In general terms, inventories and environmental analysis of livestock farms need data on animals, the operation of the farm, the level of manure production, the methods of handling of the manure (whether solid manure—FYM or slurry) and their composition (concentration of nitrogen, phosphorous, and organic matter). It is the acquisition of such data that is often considered the most demanding step in carrying out an inventory or an analysis of the farm. Furthermore, the quality of such data is central as this can improve the accuracy of the material balance and provide a reliable basis for subsequent actions (Milne et al., 2014; Velthof et al., 2015). Finally, the availability of data reflecting different manure management practices and its application in different countries remains limited or somewhat artificial or inconsistent thus rendering comparisons difficult between methods used in different countries affected by air and water pollution (Menzi et al., 2015b; Velthof et al., 2015).

France is one of the major livestock nations in Europe and the farming systems vary widely. In fact, the country makes the biggest contribution to the 1,400 million tons of animal manure estimated for the European Union (Foged et al., 2011). Thus, the purpose of this paper is to bring together, as far as possible, available data on the management of animal manure in France, especially information used in the different inventory tools or in the evaluation of technologies used for the reduction of water and air pollutants. This paper is not set out to provide new data but to assemble, standardize, and complete existing data sets.

### ASSESSMENT OF POLICY, GUIDELINES OPTIONS, AND IMPLICATIONS

### Framework of French Regulations of Livestock

All livestock farms in France fall under French and European regulations that seek to protect both the environment and local inhabitants. Farmers are, depending on the size of the farm, subject either to the RSD or "Règlement Sanitaire Départemental" (Departmental health regulations), that is the "Code de la Santé Publique" (Public Health Code) or subject to legislation for those farms coming under the ICPE or "Installations Classées pour la Protection de l'Environnement" (Livestock farms listed for environmental protection: Environmental Code). Basic nationally prescribed measures set out in specific decrees (MEDDE, 2013a,b,c) that relate to the installation (and management) of farm buildings, effluent storage, and land spreading may be reinforced by local rules depending on the local climate and the vulnerability of the local environment. European rules, reworked into French texts are also applied to those livestock farms targeted by the specific directive relating to the integrated prevention and reduction of pollution (that is the IED Directive, previously known as IPPC) or regionally applied as in the case of the Nitrates Directive and/or the Water Framework Directive (MEDDTL, 2011; MEEM, 2017b). These obligatory reglementations can affect the management of farm manure both directly and indirectly.

### The Legal Status of Livestock Manure

Livestock manure (raw or treated) come under several legal categories (waste, by-product, product) depending on their use and each with different land spreading constraints (Houot et al., 2014). Raw livestock manure managed on the farm are considered as by-products from animal production and must respect the environmental rules set out by the RSD, ICPE, and IED with respect to collection, storage, and land spreading. The outputs from treatment (composting, anaerobic digestion, separation, drying,. . . ) that are carried out on the farm or at an external site (composting center, joint AD facilities, and others), are still considered as farm manure and must be applied to the field following an approved scheme of land spreading. However, these same outputs (composts, digestates, solids from separators, dried material, etc.) may be considered as "organic fertilizers" (for free use of as a commercialized product) if they are standardized, homologated or in agreement with a specification approved by regulation. These fertilizer products are thus used according to the recommendations of the supplier without the need of a land spreading plan. Composts from the solids removed by separation and dried materials are generally put on the market (or made available) under the name "organic fertilizer" by simply following set standards (French Standard "Amendement organique) or "Organic soil improvers," NFU 44-051 or the (French Standard "Engrais organiques" or organic fertilizers, NFU 42-001). Digestates from AD must be approved (a long and costly procedure limiting approvals to just 3 products in 2014) or more recently, they can be used by following a set of procedures detailed under a decree published in 2017 (MAA, 2017). A farmer is allowed to give away, sell, or exchange (for straw) raw or treated effluents (that are neither standardized nor authorized) under a specific contract where the recipient undertakes to spread these effluents on land in full compliance with the rules in place. In the case of exported composts and AD digestate, a sanitary certificate is required.

### Principal Regulations That Govern the Management of Livestock Manure Minimal Distances for Buildings, Storage Tanks, and Land Spreading Operations

Livestock buildings, storage tanks and the spreading of effluent must observe minimal distances from residences or aquatic resources which vary depending on the effluent being spread (compost, raw manure, digestates from AD units...), on the specific regulation (RSD, ICPE, water protection. . . ) and the specific region (rules governing vulnerable areas). The minimal distance is set at 100 m for buildings and storage facilities for all effluents. On the other hand, the minimal distances for land spreading depend on the effluent type (slurry, FYM, treated effluent) and the method of land spreading: thus 10 m for composts, 15 m for injected livestock slurry or that incorporated immediately, but up to 100 m for other products. In vulnerable areas, the spreading of slurry and solid manures is forbidden closer than 35 m from the banks of rivers and streams unless there is a permanent vegetative zone (where the minimal distance is reduced to 10 m).

### Storage of Farm Manure

Livestock farmers must have available adequate storage capacity (measured in cubic meters for slurries or square meters for FYM), sufficient to enable compliance with the minimal storage periods before land spreading. The minimal storage capacity is for 45 days (RSD), 4 months (ICPE) or varies from 4 to 7.5 months in vulnerable zones depending on the animal type, the length of time at pasture and the geographical location. Two software tools (called Pré-Dexel and DeXeL) are recognized by the state for sizing and checking storage capacity for livestock farm manure (MEEM and MAAF, 2016). Legal exceptions are possible if the existing capacity is enough to enable the good agronomic use of applied manure. Field storage is allowed in the case of stable FYM (i.e., without drainage) and poultry manure with over 65% dry matter for periods not exceeding 10 months (ICPE) or 9 months (in vulnerable zones) with the stipulation that there is no reuse of the same site for storage for at least 3 years. Within vulnerable zones, field storage is forbidden from the 15 November to the 15 January except for grasslands or if the heap is placed on a bed of absorbent material (around 10 cm thick and with C/N ratio of less than 25) or if the heap is covered.

### Land Spreading of Manure

Land spreading is forbidden during certain periods or on certain land that would otherwise lead to environmental impact via run off or by leaching of the applied nitrogen and phosphorous (e.g., bare soil, sloping ground, saturated land, frozen ground, etc.). In vulnerable areas, spreading periods are determined with respect to the effluent type in terms of the level of mineralization of the organic nitrogen content, local climatic conditions, and technical limitations (soil firmness, access to the field, etc.). The implementation of a maximum 170 kg N/ha in vulnerable zones is a restriction that can lead the farmer to treat livestock manure to allow legal application on fields. Under the ICPE rules, incorporation after land spreading on bare soil is obligatory within 24 h for cattle FYM and solid pig manure (raw or treated) previous held for at least 2 months in storage (for stabilizing with respect to drainage liquids) and 12 h for all other effluents from the farm, whether raw or treated.

### Manure Treatment

Livestock manure treatment is obligatory in France under the Nitrate Directive for those farms located in high risk zones (known as ZAR or "zones d'actions renforcées") where the maximum applied nitrogen dose in the spread effluent, which can vary between departments and defined vulnerable zones, can be even more severe than the usual 170 kg N/ha. Effluent treatment is also obligatory under the Water Framework Directive in the Loire-Brittany catchment area with the purpose of ensuring an agronomic phosphorous balance. Treatment also becomes obligatory when using effluents or digestates in the case of AD as a soil improver or organic fertilizer defined by French standards (AFNOR, 2016).

# Description of Livestock Production and Manure Management

### Livestock Description

France is one of the main producers of meat in Europe being the largest in the case of cattle production, second for poultry and fourth for pig production (Eurostat, 2016). In 2010, the national agricultural census counted 490 000 farms of which 291 000 (59%) included a livestock activity (Agreste, 2011; Idele, 2013b). Cattle rearing covered 193 000 farm units and numbered in total 19.4 million animals; pig rearing covered 22 300 farms and numbered 13.8 million animals whereas poultry production was represented by 95 300 production farms with a total of 292 million birds. More recent data from 2013 indicates that livestock numbers are unchanged but spread across fewer farms (Agreste, 2013a): 176 500, 17 400, and 67 200 farms in the case of cattle, pig, and poultry respectively.

### Geographic Distribution of Livestock Farms

Livestock numbers are not evenly spread across the country. More than half (55.3%) of cattle, pig, and poultry is found in two regions in the west of France (Brittany and the Pays de la Loire). These two regions contain around 70% of pigs and 60% of poultry numbers. Cattle is more evenly spread accross the French countryside but with nonetheless different regional concentrations of dairy cows (found predominantly in the north of the country) and beef animals (found mostly in the center of the country).

### Utilized Agricultural Area (UAA) Associated With Livestock Farms

Agricultural land dedicated to the three livestock sectors, pig, cattle, and poultry, amounts to 15.2, 0.9, and 1.2 million hectares, respectively (Agreste, 2008a,b,c). For pig production, barely 10% of farms (representing 15% of pig numbers) have no adjoining farmland and a mean size much above the average (1,800 pigs or 310 sows) (Agreste, 2013b). Other farms have more than 50 ha of farmland (averaging 83 ha) of which 55% is in cereal production and oilseed/protein crops. In the case of poultry production, the mean area of farmland was in 2010: 63 ha per broiler farms, 56 ha for egg laying farms and 46 ha for pullets (Itavi, 2013a). Those poultry farms lacking any farmland account for 10% of broiler farms and 13% of egg laying farms (Itavi, 2013a). The average area of farmland of dairy farms was (in 2010) 91 ha of which 36% was given over to forage production, 37% was permanent pasture and 50% was used for cereal production and oilseed/protein crops (Agreste Centre, 2013a). In the case of beef farms, the average farmland area was slightly less at 83 ha of which 26% was used for cereal production and oilseed/protein crops and 71% was as permanent pasture (Agreste Centre, 2013b). More generally, these averages hide large regional differences. For example, in the Brittany region, many poultry farms have little agricultural land (over 30% of farms have less than 10 ha) whereas the 51 % of poultry farms in the Champagne region have more than 100 ha of farmland available.

### Livestock Production and Farm Size

Livestock farms fall under either "standard" production or "quality" production the latter being such as "Plein air" (free range), "Lable Rouge" (Red label), "Biologique" (Organic), or "Appellation d'Origine Contrôlée" (AOC). Those "quality" farms have to follow certain official conditions laid out by the Ministry of Agriculture which includes rearing times, food regime, and so on. The conditions have an impact on the quantity and composition of the animal manure produced. 63% of broiler farms, 14% of turkey farms, and 67% of guinea fowl is currently subject to special regulations governing quality (Itavi, 2010). The organic production concerns mostly broiler and egg production, representing 1.0 and 7.6% of numbers respectively in 2014 (Agence Bio, 2016). In 2014, the pig sector covered by the regulations of Red Label code accounted for less than 3.3% of the total French pig production (Badouard, 2016) and sows managed under organic rules accounted for just 0.9% (Agence Bio, 2016). In the cattle sector, around 3% of beef animals fall under special quality regulations whereas the organic codes cover 4.2% of all french dairy cows (FranceAgrimer, 2016). The number of livestock farms covered by one or other quality codes is projected to increase for all types and especially for dairy and beef cattle farms (FranceAgrimer, 2016).

### **Cattle farm size**

On average, the typical cattle farm had 101 animals in 2010 rising to 110 in 2015 (Agreste, 2016b), but the variation of the mean size from region to region was much greater ranging from 58 to 144 animals in 2010 (Agreste, 2014b). Considering the different herds of cattle, the average size is 45 heads of dairy cows, 34 heads of suckler cows while other animals (divided into <1 year, 1–2 years, and >2 years) are rearing in farms with less than 19 heads (Agreste, 2013a). In the case of milk production, 60% of farms keep only dairy animals whereas 40% have beef animals as well and/or veal calf production (Idele, 2013a). Dairy farms with more than 20 cows are predominantly for breeding (73.2% of farms) or for breeding and fattening including young beef and veal calf (20.8% of farms).

### **Pig farm size**

In 2010, 48% of pig farms had fewer than 20 sows (or fewer than 100 pigs where the pig numbers were less than 1% of all farm animals) and the mean for this sector with very small herds was just 9 pigs (Agreste, 2013b). The bulk of pig numbers are thus held in livestock farms with more than 100 pigs (or 20 sows) and the mean size for this sector is 1,200 pigs, but with large regional differences ranging from a mean of 1,860 pigs per farm in Champagne-Ardenne, down to 150 pigs per farm in Corsica. Depending on their principal activity, pig farms can be divided between breeder/fattening (50% of the total), fattening farms (43% and generally without weaners), and pig breeding farms (6% including those with or without weaning).

### **Poultry farm size**

The size of farms varies widely depending on the type of production and the methods used (whether standard practice or following specific codes relating to quality). In the case of both standard broiler production and egg laying systems (in cages), the farms are especially large scale (Agreste, 2013c). More than 60% of meat and egg poultry production is carried out in farms larger than 20 000 and 50 000 birds respectively. Those poultry farms governed by specific codes relating to quality are generally smaller (Itavi, 2010). As an example, the average size of a poultry house following standard practice is 870 m² whereas the average for a farm applying a quality code is 220 m².

### Manure Management

### **Manure storage**

Livestock farms for cattle, pig, and poultry had together in 2008 (Agreste, 2008a,b,c) around 155 000 slurry stores (mostly away from the animal house in the case of cattle and pig farms) and with a combined storage volume of 47 million m<sup>3</sup> (**Table 1**). Seventeen percentage of the storage pits were covered (in 2008)

TABLE 1 | Storage capacities of French livestock according to the 2008 survey (Agreste, 2008a,b,c).


in the case of pig farms and 10% on cattle farms. Thirty-nine percentage of poultry farms had covers on their external stores. It should be noted in this last case, that for 80% of poultry houses, manure storage is within the building which may be considered as covered. The storage of solid manure was in 2008 carried out at 130 000 stores representing a combined area of 25 million m<sup>2</sup> (2500 ha). Covers for such stores were in place for 21% of cattle FYM stores, 16% of solid pig manure stores but only 21.5% of those stores for poultry manure. A large part of the solid manure (55 million tons), mostly from the cattle sector (52 million tons), was stored in field heaps.

### Manure Treatment

In 2008, 12% of pig farms, 11% of poultry farms, and 7.5% of cattle farms used some sort of treatment for their manure. Manure treatment for the three main farm animal types accounted for 13.6 million tons (Loyon, 2017), that is, 11.3% of the 120 million tons of manure produced annually. The main processes, predominantly used at the farm, were composting (8.5 million tons), aerobic treatment (2.9 million tons of pig slurry), and anaerobic digestion or AD (1 million tons). Other treatments of solid manure including physical-chemical methods, were less common (0.4 million tons). In addition, a large part of poultry droppings is dried in or out of the rearing house.

### Land Spreading of Livestock Manure

The application of livestock manure (whether raw or treated) is mainly done on farmer's land or other land generally close to the farm (Quideau, 2010). Fields available for taking the applications of manure are linked to the crop rotation in practice at the farm (Ramonet et al., 2014). Based on the data given in **Table 2**, of the total nitrogen in the manure from livestock farms destined for land spreading [estimated as around 540 kt N: (Citepa, 2017), and personal communication] 36.5% is spread on grassland, 39.6% on maize ground, and 12.9% on cereal land. Nitrogen from cattle manure is more often spread on grassland than that from piggery manure because of differences in the crop cycle between the farm types whereas nitrogen from poultry manure is mostly spread on cereal land. In certain regions (Brittany, Pays de la Loire, Limousin), livestock manure make up the main source of nitrogen and are spread essentially on maize ground, of which the area included in crop rotation is greater than elsewhere (Agreste, 2014a).

According to the crop survey 2011 (Agreste, 2014a), nitrogen in the form of solid manure is more than 90% surface land spread but up to 67% is not incorporated within 24 h (**Table 3**). Nitrogen applied as slurry is mostly spread on the soil surface using splash plate tankers (83% of the nitrogen tonnage in the case of cattle slurry, 63% for pig slurry, and 66% for poultry slurry) (**Table 4**). Incorporation of the nitrogen content in the following 24 h occurs to 88% of the nitrogen tonnage of applied poultry slurry and to 45% of pig slurry whereas 73% of cattle slurry seems not incorporated. Incorporation within 24 h of nitrogen from solid manure concerns 31% of the nitrogen tonnage for cattle, 33% for pig and 56% for poultry manure. This difference between the animal types is explained by the large proportion of the slurry form produced by pig farms and the related obligation to reduce the odor nuisance (with respect to nearby people) by using the method of incorporation. The applied dose of organic nitrogen varies from 87 kg/ha on rapeseed crops to 154 kg/ha on forage maize.

### Estimation of the Amount of Manure Produced

Manure type (slurry, FYM, or dropping) and the quantities produced at a farm depend on the housing type (slatted floor or bedding) and the stage of animal rearing. Manure management in the building (drying belt, scrapping, flushing, storage pit, etc.) also affect the quantities of manure to be handled. Generally, solid manure (FYM) is stored in field heaps or in manure stores and slurries stored in pits.

### **Cattle production**

Eleven building types have been defined (MAPE and MAP, 2001) in order to estimate the storage capacity according to foor type (bedding or slatted floor), housing method (tied or free, "straw flow"—sloping floor with bedding, straw bedding, cubicle), possible inclusion of a yard for animal exercise covered or exposed, and the amount of straw bedding in the different area accessible by the cattle. The most common system for all animal types is an open house design with FYM production (from deep litter, straw bedding areas or from cubicles) covering 80% of all animals (**Table 5**). Deep litter barns without an exercise yard is predominant in the case of cows with followers and other cattle but less so for dairy cows where straw bedding or cubicles are also common. Slurry-based systems are rarely used except for veal calf production and for dairy cows kept in cubicles with slatted floors. These different housing types produce slurry and/or solid manure more or less of high concentration in terms of the dry matter content (DM) (Degueurce et al., 2016). Only high solid FYM (defined as those with a dry matter between 18 and 25%) and very high solid FYM (over 25% dry matter) may be kept in field heaps. Wet FYM (below 18% dry matter) must be kept in FYM bunkers for at least 2 months before possible storage in the field. (MAPE and MAP, 2001; Idele, 2005).

### **Poultry production**

Slurry is produced principally from farms rearing duck for the table or those force fed (for "foie gras"): otherwise, the slurry


TABLE 2 | Types of crop that receive the nitrogen contained in land-spread cattle, pig, and poultry manure (given as % of N applied) (Agreste, 2014a).

(\*) Soft wheat, hard wheat, barley, triticale.

(\*\*) Rape seed, Sunflower seed, fat peas.

(\*\*\*) Sugar beet, potatoes.

TABLE 3 | Time for incorporation of applied nitrogen (as a % of total nitrogen applied) (Agreste, 2014a).


TABLE 4 | Application method of nitrogen as livestock slurry (as % of total slurry nitrogen applied) (Agreste, 2014a).


system is now virtually inexistent for egg producing hens (Itavi, 2013a). Solid manure arises from broiler production on litter, from pullets and from birds retained for future chick production (Itavi, 2010). Poultry houses operating alternative (non-cage based systems) for egg production (30% of layers) also produce solid manure (Itavi, 2013b). Laying birds that are kept in cages produce droppings that are collected and removed relatively frequently by conveyor belt (with or without drying) directing the manure to storage barns or a drying tunnel or drying building. Otherwise, the system is a building with a basement to collect and store the droppings produced being emptied at the end of a production cycle (one year) or removed more frequently using a scraping system.

### **Pig production**

Buildings are mostly fitted with slatted floors (complete or partial) for all animal types, this system accounting for 91.5% of pig places (Ifip, 2010). Straw-based systems represent less than 8% of animal production. Slurry is held in pit located under the floor of the building for the whole production cycle or emptied more frequently to an external slurry pit. An alternative to this standard approach is the frequent removal of manure (gravity emptying every 15 days, automatic scraping several times each day. . . ) but this remains unusual (estimated as representing less than 1% animal numbers in each case (Martin and Mathias, 2013). In the case of bedding systems, the solid manure is managed by accumulation (during a cycle) or removed by scraping 1 or 2 times a week.

The specific quantity of livestock manure produced (per animal) depends on many factors linked to the animal (feed regime, stage of process, type of production system, and so on) and the housing method used. Default values have been proposed by specialist and these are used by the administration to allow the farmers to estimate the storage capacity required by the regulations (**Table 6**). Based on numbers of animals (different from the number of places) in 2010 given as 19.5 million cattle, 13.9 million pigs, and 221.6 million birds, and applying standard data on amounts of manure per animal, recent estimates (Ifip, 2010; Itavi, 2013b; Degueurce et al., 2016) place the total quantity of manure produced annually as around 120 million tons (**Table 7**) of which 60.6% is as solid manure, 38.8% as slurry, and the remaining 0.6% as poultry droppings. This value is less than the 263 million tons estimated by Foged et al. (2011) due to different quantity of manure produced by animal or place and the distribution of place between slurry and solid manure. Linked to the regional distribution of livestock farms, the largest amount of slurry and solid manure is produced in the "Grand Ouest" of France (Brittany, Pays de la Loire and Lower Normandy, **Figure 1**).

### ENVIRONMENTAL IMPACT

In France, total ammonia emissions amounting for 679 kt in 2015 (Citepa, 2017) arise principally from the handling of livestock manure (64%). The contribution from manure to emissions of methane and nitrous oxide in 2017 amounted to 2300 and 137 kt



(\*) including old animals taken out of production.

TABLE 6 | Reference values of the specific quantities of livestock manure produced by the main animal types.


References: (a) (Ifip, 2014) (b) (MAPE and MAP, 2001) (c) (Itavi, 2013b)

(1) Livestock unit Dairy cow – 1.1 LU; Breeding cow – 0.85 LU; Cattle under 1 year – 0.3 LU; Cattle 1-2 years – 0.6 LU;

Cattle over 2 years – 0.8 LU.

(2) Following the standard production methods or those specified by special qualities (Free range, Red Label, organic, AOC),

(3) Droppings of laying hens at 65% dry matter.

respectively which represent 11 and 4.5% of the national emission of each gas (Citepa, 2017). Manure production from intensive livestock farming in certain areas lead to a surplus of nitrogen (both organic and mineral) estimated nationally in 2013 as 902 kt (MEDDE, 2013d) which equates to an average of 32 kg N/ha of farmland. There is a large variation around this mean between areas of extensive farming (around 15 kg N/ha) and intensive regions (e.g., 69 kg N/ha in Brittany). In 2014 the surplus of phosphorous on average was 1 kg P/ha but 20 kg P/ha in Brittany (Agreste, 2016a).

TABLE 7 | Estimation of the total manure quantities (as raw manure) and corresponding nitrogen and P amounts produced by cattle, pig, and poultry farms (given as tons of raw manure and excluding manure deposited in pasture).


(\*) Citepa (personal communication) (\*\*) estimation made by the author. (1) Degueurce et al. (2016) (2) (Ifip, 2010) (3) (Itavi, 2013b) (4) (Ifip, 2010).

# ACTIONABLE RECOMMENDATIONS AND CONCLUSIONS

The management of livestock manure (120 million tons per year nationally) depends greatly on the animal type, the region and the form of the manure (solid or slurry). The largest source of animal manure is from cattle farms that produce either solid manure (69 million tons per year) or slurry (18.2 million tons per year) across the country. Pig farms mostly produce slurry (25.4 million tons per year) which is principally concentrated in two regions of France (Brittany and the Pays de la Loire). Poultry production concentrated in the west of the country produces manure as solid manure, slurry, or droppings. In the case of cattle farms, 75% of the FYM produced is stored in the field (Loyon, 2015).

Slurry is most often stored in pits before land spreading on farmland. Poultry droppings are often dried and transported to other regions. The management of livestock manure is typically without the intentional use of methods to reduce ammonia emissions. In reality, the use of covers for external stores is

not widespread and likewise the use of advanced spreading equipment (trailing hose, injection). The treatment of manure is above all used as a means to reduce the nitrogen surplus in those regions with a high livestock density, motivated by the demands of regulations linked to the Nitrates Directive. Treatment by composting is often used to enable a reduction in the obligatory minimal distances during landspreading. The use of methanisation (anaerobic digestion) to treat manure is restricted mostly due to financial reasons but also because of legal constraints. Until recently, the agricultural use of digestate required the registration or product standardization to reclassify it as a soil improver or organic fertilizer (Loyon, 2017). However, this constraint was due to be relaxed with the emergence in 2017 of a set of procedures enabling the marketing and use of agricultural digestates as fertilizing products (MAA, 2017). The movement of raw (untreated) animal manure between farms and the application of joint land-spreading plans is rare in France. This approach runs up against logistical issues about collection and a negative reaction from local people (Paillat et al., 2009). However, analysis of the best means of gaining value from livestock manure underline the importance to reformulate the manure as standard product to enable both the transport and satisfactory use on other farms bringing in (if possible) a commercial return as well (MEDDE and MAAF, 2013; Ademe, 2014). Nevertheless, for this strategy to succeed requires modified techniques that are economically viable for all livestock farms, and not just large farms are required (Quideau, 2010). This survey arising from 2010 is leading to the development of the release of new national action plans supported with financial packages that seek the reduction of ammonia emission (MEEM, 2017a) and of factors leading to climate change (MTES, 2017). The new BREF document for livestock farming (EC, 2017) arising from the Industrial Emissions Directive (EC, 2010) seeks to impose on around 3,300 pig and poultry farms in France practices determined as Best Available Technology (BAT) between the present time and 2021. New surveys of 2016 will enable an updating of the situation with manure management at livestock farms in France.

The current state of livestock manure management in France reveals that manure handling varies depending on the farm and the region. The main strategy is storage then local land spreading. In regions with a high animal density, policies of restoring water quality and the reduction of manure nuisance (especially with respect to offensive odor) have limited the agronomic use of animal manure. Newly emerging factors (depletion of mineral sources, energy costs, and economic guidelines) should increase the use of treatment technologies as well as new strategies for joint management of livestock manure.

Failing more recent data (2015 survey being in press), the compilation presented here is an important starting point to understand the French livestock production and the remaining efforts to be made to reduce its environmental impact. However, this paper is based on the analysis of a large number of official and non-official documents. One of the difficulties has been to cross-check the data most often formulated in different formats and based on unreported assumptions. As pointed out in recent publications (Kupper et al., 2015; Menzi et al., 2015a; Velthof et al., 2015; Smith and Williams, 2016; European Commission,

### REFERENCES


2017) the methodologies and data used by EU member states are often not well described. Thus, and whatever the environmental issue, there is a need for a common and harmonized methodology and procedure for collecting the data from reliable sources for the estimation of manure production and the nutrient balance.

### AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and approved it for publication.

de Méthanisation en France. Sciences Eaux &Territoires Hors-série numéro 24, 9.


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Loyon. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A GIS-Based Approach for the Sustainable Management of Livestock Effluents on Alpine Meadows

Francesco Gubert, Silvia Silvestri\*, Angelo Pecile and Luca Grandi

Technology Transfer Centre, Fondazione Edmund Mach, San Michele all'Adige, Italy

### Edited by:

Tom Misselbrook, Rothamsted Research (BBSRC), United Kingdom

### Reviewed by:

Laura Zavattaro, Università degli Studi di Torino, Italy Harald Menzi, Federal Office for the Environment, Switzerland

> \*Correspondence: Silvia Silvestri silvia.silvestri@fmach.it

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 29 January 2018 Accepted: 14 August 2018 Published: 03 September 2018

### Citation:

Gubert F, Silvestri S, Pecile A and Grandi L (2018) A GIS-Based Approach for the Sustainable Management of Livestock Effluents on Alpine Meadows. Front. Sustain. Food Syst. 2:55. doi: 10.3389/fsufs.2018.00055 Sustainable livestock effluent management is becoming an increasingly important issue in mountain areas, with particular regard to the agro-environmental performance of forage production and the social acceptability of organic fertilizer application in mixed urban-rural contexts. The present paper proposes a GIS-based methodological approach to the management and planning of digestate spreading on hay meadows, based on the integration of geo-morphological, agro-botanical and spatio-functional data about cultivated agricultural land. The proposed methodology was tested in a case-study Community of the Italian Alps, with seven dairy farms operating an anaerobic digestion plant. Nitrogen production by cattle was quantified and compared to sustainable nitrogen requirements of cut meadows, computed at the single-plot level through agro-botanical typization of swards. Subsequently, spreading restrictions provided by national and local regulations were spatially implemented. Farm-tailored nitrogen balances and digestate spreading plans were designed to help livestock farms adjust effluent spreading patterns according to meadow type and surface runoff risk. Findings are transferrable to other mountain regions based on cattle farming and grassland management.

Keywords: livestock effluents, alpine environment, organic fertilization, cut meadows, digestate, anaerobic digestion

# INTRODUCTION

The economy of alpine areas is based on tourism and livestock farming, which ensures the attractiveness of rural territories by providing high-quality traditional dairy products and maintaining typical, bio-diverse cultivated landscapes (Bätzing, 2005; Tappeiner et al., 2008; Tasser et al., 2013).

One of the main critical issues related to the co-existence of animal husbandry and other human activities, tourism included, is the odor impact of livestock breeding and effluent spreading (Lohr, 1996; Miner et al., 2000; Schauberger et al., 2001; Copeland, 2007). National and local legislation provide spreading restrictions in order to minimize the odor emissions of organic fertilization, but issues persist.

Anaerobic digestion of livestock effluents represents, in some areas, a viable solution. Scientific and practical evidence demonstrates that anaerobic digestion significantly reduces the odor potential of livestock effluents, with measured reduction values up to 80% compared to untreated slurry (Powers et al., 1999; Immovilli et al., 2008; Hjorth et al., 2009; Riva et al., 2016).

Considering the small size of livestock farms in alpine areas (Streifeneder, 2009), anaerobic digestion plants have to be planned, constructed and run by several farms together, grouped in cooperatives or enterprise networks (Blome-Drees et al., 2016). This is necessary not only to take advantage of economies of scale, but also to tackle odor issues at a larger level.

In addition, creating multiple-farm cooperation in livestock effluent management represents a chance to address the sustainable use of organic fertilizers on alpine meadows. The polarization of agricultural use between valley floors (intensification) and slopy areas (extensification or abandonment) has been generating the progressive degradation of the environmental and agronomic quality of hay meadows in mountain areas, provoking questions about the disequilibrium between nutrient inputs and outputs (Cernusca et al., 1996; Gubert, 2008; Penati et al., 2008; Zimmermann et al., 2010; Scotton et al., 2014). Collecting and processing all livestock effluents of an area in a common anaerobic digestion plant offers the opportunity to comprehensively plan nutrient restitution to agricultural land, thus ensuring a balanced organic fertilization and a progressive improvement of the agro-environmental performance of cut meadows (Holm-Nielsen et al., 2009).

Alpine meadows represent not only an important element of biological diversity and landscape quality (Dietl and Jourquera, 2004), but they are the main forage source for alpine farms, binding milk production to the terroir (Rubino, 2006). Therefore, improving meadow quality means increasing forage selfsufficiency at the farm scale and strengthening the bond between typical cheese production and the agro-environment of origin (Penati et al., 2013; Battaglini et al., 2014).

The overall objective of the present study is to provide livestock farms operating in alpine environments with innovative tools for a sustainable use of livestock effluents, in particular digestate, in order to maintain both meadow biodiversity and productivity. The new methodological approach proposed, based on GIS tools, was developed for an anaerobic digestion plant in Trentino (Eastern Italian Alps) but it is transferrable to any other alpine context with rural economies built on livestock farming. The expected outcome is the definition of farm-tailored nitrogen balances and digestate spreading plans, indicating spatial and temporal patterns of digestate use at the plot scale.

# MATERIALS AND METHODS

### Framework

The area of interest of the present study is the Community of Predazzo, in the Autonomous Province of Trento (Eastern Italian Alps). A cooperative of seven dairy farmers plans the construction of an anaerobic digestion plant, in order to improve the management of animal waste in terms of odor impact of effluent spreading and sustainable nutrient supply to cut meadows. As required by national and local legislation, the anaerobic digestion plant has to be provided with an Agronomic Utilization Plan (AUP) of digestate, certifying the equilibrium between nitrogen produced by livestock and nitrogen needed by cultivated land. The design of the AUP offers the chance to propose a new, transferrable approach for the sustainable management of livestock effluents in an alpine context, where cut meadows represent the dominant type of agricultural land use.

**Figure 1** represents the methodological scheme adopted and the expected outputs. Data collected through direct on-farm survey, regarding in particular herd size and composition, were used to calculate nitrogen production, net of any losses from excretion to the field. On the other hand, spatialized data about cut meadow plots were used to estimate nitrogen requirements of cultivated land, in order to produce farm-scale nitrogen balances and digestate spreading plans with cartographic implementation. Further detail is provided in the following sections.

A similar approach was proposed by Peratoner et al. (2010) for the calculation of forage balances at the regional scale. Combining the extent of grassland areas, subdivided according to meadow type and altitude ranges, with their average yield potential, with livestock population and with mean daily demand of different livestock classes, Peratoner et al. (2010) were able to compute a forage balance and a forage self-supply rate for South Tyrol. The present study goes a step further, providing new methodological tools which allow to tackle agro-environmental issues of livestock farming (such as nutrient management) both at the regional and at the single-farm level.

## Data Collection Through Farm Surveys

Initial data about herd size, herd composition, housing system, current effluent management and summer grazing of cattle were collected through farm surveys, in order to quantify housingrelated nitrogen produced by each farm on a yearly basis. Nitrogen not related to housing, produced during summer grazing, was calculated taking into account the number of days spent by cattle on alpine pastures and was subtracted from the yearly amount, as it is not involved in organic fertilization of hay meadows through effluent spreading. Grazing on hay meadows is not a common practice and was therefore not considered.

Standard field nitrogen values provided by Italian national legislation, specifically the Ministerial Decree April 7, 2006 (MIPAAF, 2006), were applied, as reported in **Table 1**. Field nitrogen refers to the nitrogen reaching the field, net of any losses occurring from excretion to the meadow. Nitrogen efficiency related to spreading patterns was not considered at this stage but was included in the nitrogen balancing step.

### GIS-Based Data Management Framework

Dealing with spatial patterns of digestate spreading requires a common data framework with spatial reference, in which existing information layers, i.e., slope maps, infrastructure maps, cadastral maps, can be integrated with newly processed or collected data, i.e., farm meadow plots, areas with spreading restrictions, agro-botanical meadow types. Such framework was created for the area of interest using a commercial GISapplication, acquiring existing data from the Geo-Cartographic Portal of the Autonomous Province of Trento (PAT, 2015).

### Digitalization of Meadow Plots

One of the preliminary steps was to convert existing alfa-numeric data regarding cut meadows into a spatial layer, which contains the polygons of cadastral plots cultivated by each farm, net of non-productive areas such as trees, roads and buildings, with information about the user and net cultivated surface. This layer represents the base for further data collection and processing. Alpha-numeric data about cut meadows plots, net cultivated surfaces and meadow users were acquired from the Provincial Agency for Payments in Agriculture of the Autonomous Province of Trento (APPAG, 2015). Apart from alpine pastures, which are grazed during the summer months, cut meadows for hay production represent the only type of agricultural land use in the area.

### Agro-Botanical Typization of Meadow Plots

Digitalized meadow plots were characterized on-site from an agro-botanical point of view, using the typology of Trentino's hay meadows published by Scotton et al. (2012). The typology is based on a comprehensive characterization of terrain morphology, climatic conditions, geological and biological features as well as management practices. Agro-botanical surveys were extensively conducted on 500 hectares cut meadows at the beginning of the vegetation season and cartographically implemented.

The agro-botanical classification proposed by Scotton et al. (2012) has a high degree of complexity (44 botanic units and 17 botanic types) and requires some simplification for farmoriented knowledge transfer. As proposed by La Notte et al. (2014), surveyed meadow plots were consequently grouped into three main categories, namely (a) valley floor meadows, classified by geo-morphologic parameters such as altitude and slope, (b) species-rich meadows, identified through agro-botanical surveys, and (c) slope meadows, not in (a) and (b). The three types of meadows have distinctive geo-spatial, botanical and management patterns (Gubert, 2015): valley floor meadows are located around the villages in the valley bottom, have low slope steepness and are intensively fertilized and cut (eutrophic), with poor botanical diversity; slope meadows are located at higher altitudes in steeper areas, have a less intensive cutting and fertilization regime (mesotrophic), with intermediate botanical diversity; species-rich meadows are located in marginal areas, with bad accessibility and low mechanizability, have very low utilization intensity (oligotrophic) and are characterized by an outstanding botanical value.

### Nitrogen Balance of Meadow Types

The three types of meadow identified have different forage productivity and fertilization requirements (Gubert, 2015). In order to determine necessary nitrogen input through organic fertilization, and consequently necessary digestate volumes, a nitrogen balance was calculated for each meadow type, using the nitrogen balance formula proposed by MIPAAF (2006).

$$(Y \ast B) = Nc + Nf + An + (KC \ast FC) + (K0 \ast F0) \tag{1}$$

where:

**Y**, expected forage production, calculated using on-site productivity data provided by Scotton et al. (2012);

**B**, nitrogen removal coefficient, calculated using on-site forage quality data provided by Scotton et al. (2012);

**Nc**, nitrogen availability from previous crops, not applicable for permanent grassland;

**Nf** , nitrogen availability from fertilization of the previous year, not applicable for digestate;

**An**, net natural nitrogen inputs from dry and wet deposition and from the mineralization of the organic matter;

**Kc**, nitrogen efficiency coefficient for mineral fertilization, set equal to 1 as suggested by MIPAAF (2006);

**Fc**, nitrogen input from mineral fertilization, set equal to 0 as fertilization occurs only through digestate;

**K0**, nitrogen efficiency coefficient for organic fertilization, calculated according to MIPAAF (2006) depending on soil type and spreading period;

**F0**, nitrogen input from organic fertilization, which represents the "unknown" of the formula and corresponds to the sustainable amount of nitrogen that can be spread on field with organic fertilizers to cover nitrogen outputs, net of other nitrogen inputs.

Excretions during grazing on hay meadows was not considered, as hay meadows are not subject to grazing. Nitrogen input from mineral fertilization was set equal to zero, as mineral nitrogen is not used by farms on hay meadows due to unfavorable costbenefit ratio.

As mentioned above, expected forage production (Y) and nitrogen removal coefficient (B) were calculated using forage quantity and quality data available for the area of interest, collected on 10 different sampling sites by Scotton et al. (2012). This allows to adapt calculation to the site-specific production potential of hay meadows. Nitrogen input from organic fertilization (F0) was calculated for each meadow type and represents the sustainable yearly amount of nitrogen to be spread through digestate on cut meadows per unit of area (ha).

TABLE 1 | Standard values of fild nitrogen according to the Ministerial Decree April 7, 2006 for dairy cows and heifers, depending on housing system (MIPAAF, 2006).


As to net natural nitrogen inputs (An), dry and wet deposition as well as mineralization of the organic matter were considered. For nitrogen deposition, the standard value of 20 kg nitrogen per hectare and year proposed by MIPAAF (2006) was adopted. Rihm and Achermann (2016) report for Switzerland an average total nitrogen deposition of 16.3 kg per hectare and year, with values between 20 and 30 kg in the Southern Alps. Asel (2015) proposes for Austria a value of 20 kg nitrogen per hectare and year in nitrogen balancing on permanent grasslands. Net nitrogen from mineralization of organic matter was set equal to zero, as a permanent meadow subject to adequate cutting and organic fertilization reaches, on the long run, an equilibrium between nitrogen release and immobilization in the organic matter of the soil (T'Mannetje and Jarvins, 1990). Theiss (1989) and Kasper et al. (2015) show that permanent hay meadows in alpine environments, when fertilized with organic fertilizers, may even become net accumulators of organic nitrogen.

### Cartographic Implementation of Spreading Restrictions

The net meadow area of a plot does not always correspond to the area on which effluent spreading is allowed. National and local legislation define spreading restrictions for livestock effluents, in order to minimize the impact of organic fertilization on other human activities (i.e., mobility, housing, public infrastructures) and on the environment (i.e., contamination of surface and groundwater through leaching and surface run-off). These restrictions reduce the net meadow area on which each farm is allowed to spread organic fertilizers.

Spreading restrictions for liquid effluents<sup>1</sup> are summarized in **Table 2**. In order to identify and measure areas excluded from effluent application, spreading restrictions were spatialized using available layers from the Geo-Cartographic Portal of TABLE 2 | Criteria adopted for the spatial definition of areas with spreading restrictions for liquid livestock effluents.


the Autonomous Province of Trento (PAT, 2015) and, where applicable, implementing the necessary buffer zones. Collected agro-botanical data were used for the localization of humid grasslands. Consequently, the meadow plot shapefile was intersected with the spreading restriction shapefile to highlight meadow areas where spreading is not allowed.

As to spreading limitations related to slope steepness, Italian national legislation sets a slope threshold of 10%, above which no spreading of liquid livestock effluents is allowed, except for specific local circumstances. In the alpine context of interest, around 70% of cut meadows have a slope steepness higher than 10%. For this reason, local legislation derogates to national legislation introducing the concept of "no runoff ": the spreading

<sup>1</sup>National and local legislation define spatial spreading limitations also for solid livestock effluents. These limitations were not considered in this study because (a) digestate is spread as liquid effluent, (b) in the case of solid/liquid separation of digestate, the solid fraction is subject to less strict limitations than the liquid fraction.

of liquid livestock effluents is not allowed if it generates surface runoff, depending on quantity spread and on slope steepness.

For the sake of the present study, a slope threshold of 40% was proposed. It was demonstrated that on meadows with slope steepness between 10 and 40% no surface runoff is expected from liquid digestate spreading, at given spreading volumes per operation and in absence of natural precipitation close to spreading.

For the calculation of the runoff potential of digestate spreading, the Runoff Curve Number method proposed by the USDA Soil Conservation Service was applied (USDA-SCS, 1985). This is an empirical hydrological method which allows to estimate surface runoff of a water precipitation depending on soil texture and soil cover. The spreading of liquid digestate was considered as comparable to a water precipitation, since the produced digestate is expected to have a very low dry-matter content. Representative effluent samples collected in Predazzo and digested at the Edmund Mach Foundation showed a total solids content around 5.5%. Comparable data on total solids content is provided by Landesanstalt für Landwirtschaft (2006) and Hjorth et al. (2010) for slurry-based digestates. In case of higher dry-matter contents, solid/liquid separation represents a viable solution to providing a clarified effluent fraction with water-like runoff behavior (Landesanstalt für Landwirtschaft, 2006).

The Runoff Curve Number method requires, in first instance, the definition of the soil hydrological group (A, B, C or D), which basically depends on soil texture. In the area of interest, soil texture data was available from Scotton et al. (2012) for 12 different meadow sites, with slope steepness between 10 and 40%. According to the USDA-SCS classification, all considered soils could be ascribed to the hydrological group B: soils in this group have moderately low runoff potential when thoroughly wet as water transmission through the soil is unimpeded, have typically loamy sand or sandy loam textures and the saturated hydraulic conductivity in the least transmissive layer between the surface and 50 centimeters ranges from 10.0 to 40.0 micrometers per second (USDA-SCS, 1985).

The soil hydrological group, together with soil cover (cut meadow), determine a table-based, non-dimensional Curve Number (CN) of 58. This value was corrected for available soil moisture by the Antecedent Moisture Condition (AMC) factor, which was taken at the highest level (moist soil) for precautionary reasons. Given an AMC factor of 1.3, the corrected CN was equal to 75.4.

CN is correlated to the potential maximum retention after runoff begins (S), in inches, according to the formula:

$$\text{S} = \frac{1000}{\text{CN}} - 10\tag{2}$$

At a CN value of 75.4, S was equal to 3.3 inches. On its turn, S is correlated to runoff (Q), in inches, according to the formula:

$$\begin{aligned} Q &= 0 \text{ for } P \le Ia\\ Q &= \frac{\left(P - Ia\right)^2}{P - Ia + S} \text{ for } P > Ia \end{aligned} \tag{3}$$

where P is the precipitation (in inches) and Ia the initial abstraction (in inches), or the amount of water before runoff, such as infiltration, or rainfall interception by vegetation. For a precipitation lower than or equal to the initial abstraction, surface runoff can be considered equal to zero.

The initial abstraction Ia was calculated, according to USDA-SCS (1985), with the formula Ia = 0.2<sup>∗</sup> S. However, several studies show that this ratio is too high when compared to field measurements (Jiang, 2001; Hawkins et al., 2002; Mishra and Singh, 2004; Baltas et al., 2007). Therefore, a more appropriate initial abstraction ratio, namely Ia = 0.05<sup>∗</sup> S, was proposed according to Hawkins et al. (2002). A lower initial abstraction ratio determines a lower threshold precipitation amount for runoff and, accordingly, a lower maximum amount of digestate that can be spread on hay meadows. As a consequence, more surface is necessary to spread a given amount of digestate. This choice was supported by the Regional Agency for Environmental Protection as a precautionary measure to minimize the risk of surface-runoff.

Applying the latter formula for the calculation of initial abstraction, the threshold precipitation amount for runoff (P = Ia) on the considered soils was equal to 0.16 inches, corresponding to 4.06 millimeters. This means that the amount of liquid digestate per spreading operation that can be distributed on meadows with a slope steepness between 10 and 40% without any risk of surface runoff is equivalent to 4.06 millimeters, or approximately 41 cubic meters per hectare.

The slope threshold of 40% was integrated in the GIS data framework to compute, together with the other spatialized spreading restrictions, the net meadow surface available for spreading at the single-plot as well as at the farm level.

## Farm Nitrogen Balance and Digestate Spreading Plans

The above described methodological steps allow to define, for each meadow plot, the meadow type, the corresponding yearly sustainable nitrogen input from organic fertilization, the net meadow surface available for spreading and, ultimately, the quantity of effluent which could be spread to cover actual nitrogen requirements on a yearly basis. Additional information about runoff-related threshold volumes of effluents was used to generate an agronomically viable and environmentally sustainable spreading plan, consisting of different volumes and frequencies of slurry application during the growing season depending on meadow type. Spreading plans could be spatially implemented both at the plot and at the farm scale. The comparison between nitrogen produced by cattle and actual nitrogen requirements of cultivated meadows allowed to quantify, for each farm, occurring nitrogen excess / deficit, thus highlighting potentially critical situations.

# RESULTS AND DISCUSSION

### Farm Survey Findings

The seven farms surveyed breed dairy cows predominantly in free housing with slurry-based effluent management. To different extents, all farms bring their cattle on high-altitude pastures during the summer months, thus reducing housingrelated nitrogen loads and effluent volumes by approximately 1/3. The declared total number of dairy cows was equal to 517 units, while the number of young cattle under 24 months was equal to 224 units. Nitrogen reaching the field was equal to 39,557 kg per year, corresponding to 9,922 cubic meters effluent volume. Total effluent stock capacity was equal to 13,368 cubic meters. This data was used not only for calculating nitrogen balances at the farm scale, but also for dimensioning the anaerobic digestion plant and its stock volumes.

The use of standard field nitrogen values proposed by national legislation presents two main limits. On the one hand, nitrogen excretions are calculated for intensive dairy farms of plain areas and may be over-estimated when applied to more extensive farming systems in mountain areas. For instance, the Region Valle d'Aosta in the North-Western Italian Alps has introduced lower nitrogen excretion values for local breeds compared with non-local ones (Francesia et al., 2008; Regione Autonoma Valle d'Aosta, 2017), whereas Steinwidder (2009) has calculated nitrogen excretion depending on milk productivity and protein intake. On the other hand, nitrogen losses occurring from excretion to the field are standardized and cannot be differentiated depending on effluent management practices. Adaptation of standard field nitrogen values to specific situations and to alpine conditions can therefore improve overall accuracy of calculation.

## Digitalization and Agro-Botanical Typization of Meadow Plots

Total meadow surface cultivated by the seven farms, net of non-productive areas, was equal to 210.36 hectares, with a total number of 2,505 cadastral plots (840 square meters per plot on average). Spatial distribution of meadow plots on a single-farm basis confirm a high degree of land fragmentation and dispersion, as already assessed and measured by Bittante (2011) in over 1,000 dairy farms of the Province of Trento. Land parcelization is a peculiar trait of the Southern Alps and is one of the most evident weaknesses of mountain agriculture in these areas (Bätzing, 1992).

Due to the high degree of land parcelization and of land turn-over between farms, agro-botanical surveys were conducted independent of existing utilization patterns on all meadow surfaces of the Community of Predazzo, including agricultural land cultivated by farms not joining the anaerobic digestion

TABLE 3 | Values used in the calculation of the sustainable nitrogen input from organic fertilization (F0) for the three meadow types according to formula (1).


project. At an aggregate level, 20 different agro-botanical types of meadows were recorded on approximately 500 hectares, ranging from oligotrophic, species-rich Nardus stricta grasslands to over-fertilized, almost mono-specific lowland Agropyron repens meadows. Slope areas were dominated by swards of the Centauretransalpinae-Triseteum flavescentis community, while valley bottoms were predominantly characterized by more productive meadows of the Centaureo carniolicae-Arrhenaterum elatioris community. To date, the case study of Predazzo represents the vastest area of cartographic implementation of Scotton et al. (2012) agro-botanical typology in the region.

As previously described, the agro-botanical classification of cut meadows proposed by Scotton et al. (2012) is fairly complex. However, it allows to identify both meadows with outstanding ecological value, because of species-diversity or rarity, and meadows with agronomic problems related to utilization intensity, showing distrophic botanical compositions (e.g., high weed coverage). This information was used to produce farm-tailored "health-maps" of hay meadows, with specific management recommendations for the maintenance of species-rich swards and the agronomic improvement of degraded surfaces.

For the sake of transferability to farmers, the meadow plots managed by the seven farms were subsequently classified in three macro-categories. This classification allowed to cartographically identify 73.37 hectares valley floor meadows (34.9%), 95.63 hectares slope meadows (45.5%) and 41.36 hectares speciesrich meadows (19.6%). As expected, considering the geomorphological and pedological traits of the area, meadows with low intensification potential (slope meadows and species-rich meadows) represented the majority of cultivated land. **Figure 2** shows a cartographic detail of agro-botanical meadow typization in a sample area of Predazzo.

### Nitrogen Balance of Meadow Types

For each meadow type, a nitrogen balance was computed according to Equation (1). **Table 3** reports the values used in the calculation and the resulting sustainable nitrogen input from organic fertilization (F0). Nitrogen exports, calculated multiplying expected meadow productivity by unitary nitrogen content of forage, is in line with average values reported by Scotton et al. (2012) for the entire Province of Trento. Comparable values are described by Buchgraber and Gindl (2004) in Austria and Dietl and Lehmann (2006) in Switzerland for meadows with similar production potentials.

One of the key factors determining the sustainable nitrogen input from organic fertilization is the nitrogen efficiency coefficient (K0), calculated according to a table provided by MIPAAF (2006) depending on soil texture and spreading period. The resulting K<sup>0</sup> value was equal to 50%, corresponding to the minimum level indicated by MIPAAF (2006) for liquid bovine effluents. This means that, on average, for every kg nitrogen spread on meadows through slurry or digestate, only 0.5 kg are actually absorbed by plants, while the remaining 0.5 kg are lost through leaching and volatilization (Stanley, 2014). A mean nitrogen efficiency value of 50% for bovine slurry is reported by Webb et al. (2010) in many nitrates action programmes of European Union Member States. More recent Italian legislation has introduced a minimum nitrogen efficiency coefficient for digestate of 60%, starting from the assumption that anaerobic digestion increases nitrogen availability to plants in digestate when compared to untreated effluents (MIPAAF, 2016). As the present study was carried in 2015, a K<sup>0</sup> value of 50% was considered, as indicated by MIPAAF (2006). Different nitrogen efficiency coefficients, related either to legal thresholds or to different types of effluents, do not affect the proposed method.

# Cartographic Implementation of Spreading Restrictions

The effluent spreading restrictions reported in **Table 2** were cartographically implemented to identify and measure the portions of meadow plots subject to limitations in organic fertilization. **Figure 3** shows a cartographic detail of this implementation in a sample area. All areas excluded from effluent application were subtracted from net meadow surface to compute the agricultural surface on which spreading is allowed. On average, 16% of net meadow surface was found to be excluded from spreading. Slope meadow represented the category most affected by spreading restrictions due to higher average slope steepness (Gubert, 2015). A simulation with a slope-threshold of 10% for effluent spreading, as proposed by MIPAAF (2006), resulted in the exclusion of 85% of net meadow area from effluent application.. This confirms the need to take the specificity of mountain areas into consideration when planning normative tools for the agro-environmental management of livestock effluents.

Effluent spreading restrictions maps were elaborated in detail for each farm, in cooperation with the local Agency for Environmental Protection (APPA), in order to reduce the environmental impact of effluent spreading and to increase its social acceptance. The information layers created for this purpose might be potentially integrated in an open-access web-GIS application, similar to the Wisconsin Manure Management Advisory System, which helps farmers and others who apply nutrients to identify suitable cropland areas for spreading (MMAS, 2014).

# Farm Nitrogen Balance and Digestate Spreading Plans

Information about field nitrogen production, sustainable nitrogen requirements of each meadow class and net meadow surface available for spreading were used to compute a nitrogen balance at the single-farm scale. **Table 4** reports an example for one of the seven farms involved, with a good balance between nitrogen produced by cattle and nitrogen required by meadows. Some farms showed a positive nitrogen balance, with more nitrogen produced than required, and some others a negative balance, with less nitrogen produced than required. In sum, the total nitrogen balance of the seven farms was well-balanced.

The peculiarity of the nitrogen balance developed in the present study is that the sustainable input from organic fertilizers was "constructed" starting from the actual surfaces available for spreading and their agro-botanical characterization, with

TABLE 4 | Example of calculated nitrogen balance for a dairy farm involved in the study.


on-site collected data about forage quantity and quality of hay meadows. This means that spatial distribution patterns of cultivated land as well as site-specific production potentials were taken into account when computing the nitrogen balance, as already suggested by Scotton et al. (2012). Transferability to other areas of the Province of Trento is ensured by data about meadow types and productivity provided by Scotton et al. (2012). For other alpine regions, transferability may be limited by the lack of comprehensive, site-specific information. However, simplification of meadow types as proposed by La Notte et al. (2014), integrated with literature data about production potentials of hay meadows in the region of interest, still allows method implementation, even if with a larger degree

of approximation. Peratoner et al. (2010), for instance, have computed forage balances for South Tyrol starting from average productivity data of different types of meadows in the region.

The second important output of the study is the definition of farm-tailored digestate spreading plans, which consider farmspecific meadow type composition and spatial distribution. The dataset developed allowed to quantify, for each meadow plot managed by a farm, not only the total amount of nitrogen—and consequently of digestate—to be spread during a vegetation season, but also the sustainable digestate volume per operation according to actual vegetation requirements and runoff risk potential. **Table 5** summarizes the digestate spreading recommendations elaborated for the three macro-categories of hay meadows. Given the total effluent volume to be spread per hectare and year, spreading was distributed during the vegetation period according to Buchgraber and Gindl (2004) (decreasing

<sup>2</sup>According to the Ministerial Decree April 7, one adult cattle unit corresponds to 600 kg live weight, producing 83 kg field nitrogen per year.


TABLE 5 | Digestate spreading recommendations elaborated for the different types of meadows.

effluent volumes from spring to autumn), ensuring at the same time the absence of surface runoff (spreading volume lower than 41 cubic meters per hectare and operation). Proposed spreading recommendations can be finally spatialized at the plot level, to deliver seasonal spreading plans for each farm.

### CONCLUSIONS

The methodological approach proposed in the present study allows to tackle the issue of animal effluent spreading in mountain areas, with a specific focus on the environmental and agronomic sustainability of digestate use on alpine hay meadows. Data collected on-site were spatialized and integrated with existing geographic information layers, in order to develop new management and planning tools which are transferable to livestock farmers and help them adjust effluent spreading patterns according to the actual nutrient requirements of cut meadows as well as to the potential risk of surface runoff.

The case-study of Predazzo (Trentino, Eastern Italian Alps) allowed to test a new methodological approach, delivering usable results for the agronomic utilization of digestate in seven dairy farms in an alpine context. The main outcomes are a) farm-scale nitrogen balances, calibrated on size and agro-botanical type of hay meadows managed by the farm, b) farm-tailored digestate spreading plans, providing sustainable spatio-temporal patterns of organic fertilization on agricultural land. The methodological procedure as well as findings about nutrient balancing of alpine meadows and effluent-related surface runoff are transferrable to other mountain regions based on livestock farming and grassland management, and are also applicable for farms without anaerobic digestion to optimize effluent and nutrient management on hay meadows in general.

To date, the present study represents the first implementation of GIS tools for the management of livestock effluents in Trentino's mountain areas. Besides methodological aspects, one of the most important innovation elements is the spatial scale, which enables to deliver agro-botanical and management

### REFERENCES


information about hay meadows at the single-plot level. Further developments regard the implementation at a larger geographic scale (i.e., district or region) and the integration of computed geo-referenced data in existing regional cartographic portals and web-GIS applications. Validation as well as monitoring of results will occur in the next 3 years during practical implementation, in order to verify the quality and effectiveness of the proposed method.

# ETHICS STATEMENT

The authors declare that: no ethics approval was required for the survey as per institutional and national guidelines; they obtained a letter of engagement from the cooperative of farmers, signed by the president of the cooperative after having defined and agreed with them the content of the activity; oral informed consent was obtained from all research participants. They also consented by virtue of survey completion.

### AUTHOR CONTRIBUTIONS

SS conceived and supervised the present study in cooperation with the Cooperative Biodigester Predazzo, involving AP, FG, and LG in the development and implementation phases. SS and LG conducted on-farm surveys and related data elaboration, effluent sampling and laboratory analysis, biodigester design and dimensioning. FG developed the GIS-framework and performed related computations and AP helped supervise the project. FG took the lead in writing the manuscript. All authors provided critical feedback and helped shape the research, analysis and manuscript.

### ACKNOWLEDGMENTS

The authors are particularly grateful to the Cooperative Biodigester Predazzo, specifically to Dr. Alberto Bucci and Franco Morandini, for the fundamental contribution to the outcome of the present study.

Österreich. Degree thesis, University of Natural Resources and Life Sciences, Vienna.


farms. Ital. J. Anim. Sci. 13, 431–443. doi: 10.4081/ijas.201 4.3155


Zootecniche, Paesaggistiche e Ambientali. San Michele all'Adige: Fondazione Edmund Mach.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Gubert, Silvestri, Pecile and Grandi. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Dairy Manure Management Perceptions and Needs in South American Countries

María A. Herrero<sup>1</sup> \*, Julio C. P. Palhares <sup>2</sup> , Francisco J. Salazar <sup>3</sup> , Verónica Charlón<sup>4</sup> , María P. Tieri 4,5 and Ana M. Pereyra<sup>1</sup>

<sup>1</sup> Bases Agrícolas, Producción Animal, Facultad de Ciencias Veterinarias, Universidad de Buenos Aires, Buenos Aires, Argentina, <sup>2</sup> Group Water and Waste Management, Embrapa Pecuária Sudeste, São Carlos, Brazil, <sup>3</sup> Instituto de Investigaciones Agropecuarias, Instituto Nacional de Tecnología Agropecuaria, Osorno, Chile, <sup>4</sup> Producción Animal, Estación Experimental Agropecuaria Rafaela, Instituto Nacional de Tecnología Agropecuaria, Santa Fe, Argentina, <sup>5</sup> Administración Agraria, Universidad Tecnológica Nacional, Rafaela, Argentina

### Edited by:

Lars Stoumann Jensen, University of Copenhagen, Denmark

### Reviewed by:

Yong Hou, China Agricultural University, China Luís Miguel Brito, Polytechnic Institute of Viana do Castelo, Portugal

> \*Correspondence: María A. Herrero malejandra.herrero@gmail.com

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 14 February 2018 Accepted: 28 May 2018 Published: 19 June 2018

### Citation:

Herrero MA, Palhares JCP, Salazar FJ, Charlón V, Tieri MP and Pereyra AM (2018) Dairy Manure Management Perceptions and Needs in South American Countries. Front. Sustain. Food Syst. 2:22. doi: 10.3389/fsufs.2018.00022 Milk production is important in South American countries being based mainly on grazing systems. Dairy slurry management has become an important issue in these production systems because of the large volumes produced and the environmental effects. Thus, manure management regulations are emerging in the region. This research aims to identify priorities for management strategies and technology transfer by assessing perceptions, needs and barriers toward dairy manure management by stakeholders in South American countries. A questionnaire was prepared and distributed in Spanish and Portuguese in different formats: on paper and online (PDF format and SurveyMonkeyTM platform) between March 2015- November 2017. It was divided into two sections, the first addressed issues related to water quality and pollution, odor generation, fertilizer value, pathogens impact and biogas production. Responses were measured across a standard 5-point Likert type scales. Section two addressed needs and hindrances concerning about manure application. A total of 593 surveys were completed: Argentina (n = 308, 52%), Brazil (n = 217, 37%) and Chile (n = 68, 11%). The majority of respondents were dairy farmers (31%), professional advisors and consultants (29%) and representatives of public institutions and researchers (31%). Some differences appear according the country. Overall, a large majority perceive that manure is a good fertilizer (91%), also they believe that it contributes to pathogen's transmission and groundwater and shallow aquifers 'contamination. Stakeholders (60%) perceived biogas production as a good option for manure treatment. Most of respondents (79%) would use manure to replace mineral fertilizers, with little differences between countries (Argentina 79%, Brazil 80% and Chile 68%). The most selected needs were: a management handbook, increased investment in equipment and technologies and better access to laboratory analysis. The most chosen barriers were: cumbersome management, lack of knowledge and of specific laws, with differences between countries and respondents. The survey showed interest in dairy manure management as a source of nutrients for grassland and crops, especially among farmers and advisors whom requested guidelines for responsible management. Policymakers and stakeholders should focus on promoting manure reuse on dairy farms through incentives, technologies and/or appropriate strategies, in order to improve nutrient use and reduce pollution to the wider environment.

Keywords: environmental regulations, organic fertilizer, dairy slurry, dairy stakeholders, technology adoption, waste technologies, dairy grazing systems

### INTRODUCTION

The dairy industry is an important sector in South American countries, where Argentina, Brazil, and Chile possess 70% of the South American dairy herds and produce 73% (43 million liters) of the milk of this region (FAOSTATS, 2016). Dairy farming is mainly pasture-based, in non-irrigated areas with a low proportion of herds in confinement, but with part-time confinement for feeding with silage and concentrates. Cows are milked one to three times per day; after each milking, the parlor is cleaned. Cleaning is generally performed with or without scraping and washing with or without pressure. The average effluent volumes used are 27.7 and 36.6 L day−<sup>1</sup> cow−<sup>1</sup> , respectively, for Argentina and Chile (Salazar et al., 2010). Previous studies indicate that the slurry produced has high contents of rain and cleaning water. As a result, dry matter and nutrient contents are low, and the costs of transporting the slurry to arable farms are relatively high (Salazar et al., 2007; Charlón et al., 2013). The rapid development of the dairy industry has resulted in the increase of the total wastewater discharge. As a result, developing a reasonable and effective water resource management to deal with water pollution challenges associated with the dairy industry has become an urgent issue in different countries such as China (Bai et al., 2017).

The use of livestock manure for fertilization is a valid practice and should be part of the waste management of animal production systems, mainly because of the high nutrient and water contents of manure (Watson and Atkinson, 1999), reducing the need to purchase chemical fertilizers and to irrigate the crops. However, in case of inadequate manure management, there is the risk of point source or diffuse pollution, compromising soil, air, and water quality (e.g., Isermann, 1990; Erisman et al., 2008). In South American countries, dairy slurry is used mainly without treatment, although in some cases, mechanical separation is carried out. The effluent is stored in earth banks and lined lagoons or concrete tanks and applied to the field by surface application (e.g., via slurry tanks or irrigation pumps) throughout the year (Salazar et al., 2010). Higher concentrations of nutrients in the environment are often associated with an over-use of fertilizers and manures, intensive livestock breeding, climatological and edaphic conditions, and inadequate agricultural practices and management (Herrero and Gil, 2008; Oenema, 2015). The generation of high amounts of manure in concentrated areas from livestock production systems requires adequate on-farm and off-farm management. In several countries, an equilibrium between the amount of manure produced and the availability of an agricultural area to recycle this manure is difficult to achieve (Bernal, 2017). However, in most South American countries, dairy intensification is a rather recent process, and the relationship between the amount of manure and crop area is more suitable. In addition, dairy production systems in South America are mainly based on grazing, where most of the feces and urine enter the soil directly, reducing the amount of manure production (Salazar et al., 2010).

In Argentina, Brazil, and Chile, the lack of knowledge and the low adoption rate of waste management technologies are the most conflicting issues concerning environmental conservation (Palhares, 2009; Salazar et al., 2010; Charlón et al., 2017a). In these countries, new regulations for manure management are emerging, and farmers and other professionals show an increasing interest in the agronomic use of animal manure; however, special guidelines for the implementation of good management practices are crucial.

To analyze the possibilities of adopting certain technological innovations in terms of manure management in the South American agricultural sector, it is important to know the needs and constraints of producers, consultants, researchers, students, and governmental agents regarding such implementations. Stakeholder perceptions are important aspects to be evaluated, as stakeholders are mainly responsible for the implementation, evaluation, and adaptation of manure management practices. Agri-environmental programs are basically designed to promote changes in the behavior of farmers, either via amplifying behavior which leads to positive externalities or by restricting behavior which leads to negative externalities (Ahnström, 2009; Blackstock et al., 2010; Wissman et al., 2013).

The adoption of adequate manure management practices also depends on the regulatory context and the local market, for example environmental policies and the agricultural industry structure of the country under consideration, or on other, more complex issues, especially when they relate to environmental aspects (Sunding and Zilberman, 2001; Prokopy et al., 2008). The production system, farm size, and management practices are also aspects to be considered to evaluate. Thoma et al. (2013), investigating management practices at US dairy farms, showed how a large diversity of management practices and technologies used in production units translates into significant differences in environmental impacts. There is thus a high potential to improve the environmental performance of the dairy sector.

In Denmark, incentives and legislative requirements for the processing of organic waste were introduced recently to meet environmental objectives. This situation was an incentive for the evaluation of the perception of producers toward the use of organic waste (Case et al., 2017). It is important to carry out studies to understand the decisions of farmers, professional consultants, and policy makers regarding the adoption of alternative organic crop fertilizers, especially against the background of no regulatory incentives or no history of using these techniques. Generally, surveys to assess perceptions, attitudes, and needs focus on different aspects (Petit and van der Werf, 2003; Herrero et al., 2010; Barnes and Toma, 2012; Wolf et al., 2016; Hou et al., 2018). Also, they are generally used at the regional scale to compare stakeholder perceptions between regions and countries (Hou et al., 2018).

On a global level, only few studies have specifically considered manure and the use of organic fertilizers by farmers and professional consultants (Gebrezgabher et al., 2015). Although South America accounts for 54% of the livestock and 5% of the dairy cows in the world (FAOSTATS, 2016), there are no published studies on these issues.

In this context, we identified, for the first time, the requirements for management strategies and technology transfer by assessing the perceptions, needs, and barriers in terms of dairy manure management by stakeholders in South American countries. The results of our study will contribute to the discussion of upcoming policies and regulations on dairy manure use.

# MATERIALS AND METHODS

We used a descriptive methodology to obtain information about the perceptions, needs, and barriers in terms of dairy manure management, explaining the stakeholders' understanding. Descriptive survey types were used to obtain information concerning the current status of the subject and to demonstrate relationships between different issues (Jackson, 2009; Saunders et al., 2012), obtaining qualitative data. The issues in dairy manure management were defined by expert judgment, facilitating the delimitation of the research theme and the discussion about policies and regulations.

### Country Selection and Context

Argentina, Brazil, and Chile have similar livestock production systems, which are based on pasture grazing with daily confinement for feeding with silage and concentrates. These countries have an increased consumption of animal products, and the milk supply chain has important social relevance in terms of employment and income; currently, the milk sector is undergoing verticalization. There are no or few specific environmental regulations for manure management, and the social pressure to adequately manage dairy manure is increasing, with regional regulations being implemented (e.g., in Córdoba Province, Argentina).

### Survey Questionnaire and Data Collection

Data were collected between March 2015 and November 2017. A questionnaire survey was constructed to determine the perceptions, needs, and barriers in terms of dairy manure management by the participants. We surveyed dairy farmers (DF), professional advisors and consultants (PAC), dairy farm employees (DFE), service and inputs providers in dairy technologies, students and representatives of public institutions working in agro-environmental policies and from dairy companies, and researchers from academic institutions; the latter category was classified as "OTHERS" (Herrero et al., 2010; González-Pereyra et al., 2015; Hou et al., 2018). Prior to the survey, the researcher explained to each respondent that it was a survey to assess their perception of manure management and that the results would be used exclusively for research purposes. No respondents were identified by name in the questionnaires, so the authors cannot relate the participant to the answers; in this way, the ethical and more principles of each respondent were not violated.

The survey was prepared in Spanish and Portuguese in three different formats: on paper, which was handed at conferences, workshops, symposiums, and field days (Argentina); online in a PDF format and sent by e-mail (Argentina and Brazil); and via an online survey using the SurveyMonkeyTM platform (Chile). All formats used the same questions and structure. Survey dissemination strategies differed between countries according to working styles, connectivity, and the possibility of contacting the stakeholders, always meeting the usual ways of collecting information. Such an approach generally facilitates high cooperation rates at a low economic cost (Rojas Tejada et al., 1998).

The survey was divided into two sections. The first section addressed seven issues related to:


Responses were measured across a standard 5-point Likert type scale, ranging from strongly agree (5 points), agree (4 points), neutral (3 points), disagree (2 points) to strongly disagree (1 point) (Likert, 1932; Barnes and Toma, 2012; Sullivan and Artino, 2013).

The second section addressed an option that allowed the respondents to choose whether manure or slurry should be reused as fertilizer. There were also five options in terms of "needs" and five in terms of "barriers," and the respondents could choose one or multiple answers (Bernal, 2017; Case et al., 2017; Hou et al., 2018). The response options were selected based upon previous studies and tested by experts in the field of manure management (Salazar et al., 2007, 2010; Asai et al., 2014; Palhares, 2014; Vankeirsbilck et al., 2016; Charlón et al., 2017b; Palhares et al., 2017).

The preliminary version of the questionnaire was based on the definition of a basic structure that was built by the personal experience of the authors in the different local contexts. The questions in the first section were selected from previous studies (Herrero et al., 2010, 2011). The lists of needs and barriers were based on interviews with key actors and experts in the dairy sector and on previous surveys (Herrero et al., 2011). We then performed a pilot test with a small sample of postgraduate students, rural area consultants, and researchers. Once the answers were evaluated, the final version was designed.

### Statistical Analyses

All data from the different versions were downloaded and compiled in one Microsoft Excel 2010 file. The responses were cleaned and checked for inconsistencies. A descriptive analysis of the results of both sections of the survey was carried out (median, ranges, and frequencies and percentages) for individual countries, type of respondents, and for all results.

For analysis of the first section of the questionnaire (seven Likert questions), both univariate and bivariate analyses were conducted to obtain the frequency of response. These data were then used to construct the typology based on the responses to the statements. Multiple correspondence factorial analysis (MCA), hierarchical classification of Ward based on seven factorial axes, and cluster analysis were performed to obtain significant modalities at p < 0.05. All seven Likert questions related to the environmental issues described below were defined as active variables, while country and the profession were considered as supplementary variables (Lebart et al., 1995, 1998). On the other hand, mean comparisons between the three countries were performed for these seven variables via the Kruskal-Wallis rank sum test (p < 0.05), and Dunn's multiple comparison procedure was applied to detect differences between countries (p < 0.05).

Since in the second section, there were several multipleresponse questions for the selection of needs and barriers, the absolute number of respondents referring to each option was converted to the percentage of the total number of respondents who answered the question in order to allow a comparison between countries (Barnes and Toma, 2012; Hou et al., 2018). To compare countries in terms of different needs and barriers, a Chi<sup>2</sup> homogeneity test was used at p < 0.05. To evaluate the association of needs and barriers with the respondent's profiles, a Chi<sup>2</sup> independence test was used at p < 0.05 (SAS Institute, 2011).

### RESULTS

### Respondent Profiles

A total of 593 surveys were completed, with 308 (52%) from Argentina, 217 (37%) from Brazil, and 68 (11%) from Chile. **Figure 1** shows an overview of all respondents in terms of profile and country. Of all respondents, 77% were dairy farmers (n = 182), professional advisors, and consultants (n = 170), as well as representatives of public institutions involved in agro-environmental policies, dairy companies, and researchers of academic institutions (n = 106).

In Argentina, the highest percentage of respondents (77%) corresponded to dairy farmers (n = 108, 36%), professional advisors and consultants (n = 79, 26%), service and inputs providers, representatives of public authorities, and researchers at academic institutions (n = 73, 24%). In Brazil, the highest percentage of respondents (90%) corresponded to professional advisors and consultants (n = 76, 35%), students (n = 63, 29%), and dairy farmers (n = 56, 26%). In Chile, 81% of the respondents were representatives of public authorities, people working in dairy companies, researchers at academic institutions (n = 22, 32%), dairy farmers (n = 18, 26%), and professional advisors and consultants (n = 15, 22%).

## Manure Management and Environmental Perceptions

The general perceptions of the 593 respondents, expressed as percentage according the 5-point Likert scale, are shown in **Figure 2**. Overall, the large majority of respondents in all regions believed (strongly agree and agree) that manure is a good fertilizer; however, at the same time, there was an agreement that manure can cause problems to the wider environment. More than 40% strongly considered odor as a pollutant, and more than 50% related manure to the transmission of pathogens. It is interesting that most of the respondents believed that manure can cause groundwater contamination, but at the same time, there was a low agreement in that it can affect water quality (**Figure 2**).

Five clusters were defined by a dendrogram. Correspondence analysis for perceptions on manure management and environmental impacts, including countries and respondent's profiles, showed that five groups were statistically significantly different (p < 0.05). Group 1 contained 240 respondents, mostly from Argentina, who strongly agreed or agreed (5 and 4) on six of the seven questions (except biogas as an alternative to treatment). Group 2 contained 41 respondents who were indifferent to all the questions, except for the importance of biogas treatment. This group was not made up of respondents from one particular country. Group 3 contained 239 respondents, mostly from Brazil, who the issues in an intermediate way (opinions 4-3 and 2), except for the function of the effluent as a fertilizer, to which they fully agreed. In group 4, there were 46 people who disagreed with all questions. Group 5 contained 57 individuals who did not respond to five of the seven questions, with the exception of the effluent as a producer of odors that are contaminants and the effluent as a water pollutant.

To evaluate the different perceptions for each environmental aspect considered in the questionnaire by countries and by type of respondents, we calculated mean values and standard deviations (**Tables 1**, **2**). To assess these values, we considered for all the Likert scale responses that: strongly agree = 5 points, agree = 4 points, unsure = 3 points, disagree = 3 points, strongly disagree = 1 point, and no response = 0 point. Significant differences between Argentina and Brazil and between Chile and Brazil were observed for the aspects in questions 1– 5 (p < 0.05), according to the Kruskal Wallis test and Dunn's multiple comparison procedure (**Table 1**).

In Argentina (**Figure 3A**) the largest group of respondents (63, 44, and 63%, respectively) considered the consequences of the effluent lagoons for groundwater contamination and for shallow groundwater in lowland areas and agreed that pathogen transmission could be a consequence of manure management. At the same time, an impressive group of respondents perceived the importance of the use of manure as a fertilizer (50%).

Almost all Brazilian respondents agreed with the use of manure as fertilizer, and about 80% believed that biodigestion

TABLE 1 | Perceptions of manure management and its environmental impact in Argentina, Brazil, and Chile expressed as means values ± standard deviation according the survey response.


<sup>a</sup>Different letters in rows shows significant differences between countries, Dunn test (p < 0.05). Perception Likert 5-point scale from strongly agree = 5 points to strongly disagree = 0 point.


TABLE 2 | Perceptions of manure management according different respondents expressed as median ± standard deviation according the survey response.

<sup>a</sup>Different letters in rows shows significant differences between respondents, Dunn test (p < 0.05). Perception Likert 5-point scale from strongly agree = 5 points to strongly disagree = 0 point. Different Groups of respondents. DF; Dairy Farmers; PAC, Professionals advisors and consultants; DFE, Dairy Farm Employees; SIP, Service and Inputs Providers; ST, Students; OTHERS, representatives of public institutions working in agro-environmental policies and from dairy companies and researchers from academic institutions.

is a good option for manure management (**Figure 3B**). The use of manure as fertilizer and biodigesting are coincident, since the effluent from the biodigester can be used as a fertilizer. The statements 4 and 6 obtained similar percentages of agreement and were directly related, since groundwater contamination with manure can represent a pathway of pathogen transmission. The statement that odors are pollutants obtained the second lowest level of agreement, with slightly more than 60% of the respondents, although a high percentage of the respondents believed that there is no adequate regulation in Brazil.

In Chile, most of the stakeholders strongly agreed or agreed with the use of manure as a source of fertilizer; however, at the same time, the respondents were aware of the fact that inappropriate manure management can cause environmental problems (**Figure 3C**). On the other hand, the same respondents placed less importance on water quality related to manure reuse; for them, manure treatment via anaerobic digestion was not a technology prioritized by farmers.

For the three countries, the impact of the water quality used in the milking parlor on the quality and management of manure as fertilizer had the lowest importance (**Figure 2**). This could be explained by the low salt concentrations in the groundwater of Brazil and Chile (**Figures 3B,C**), esp. in the vicinity of dairy farms. However, this was not the case for Argentina (**Figure 3A**), where groundwater quality is an important aspect, with relatively high salt concentrations in the groundwater under dairy farms.

We found significant differences between Argentina and Brazil and between Chile and Brazil in terms of the aspects in questions 1–5 (p < 0.05), according to the Kruskal Wallis test and Dunn's multiple comparison procedure (**Table 1**).

The mean values for the different groups of respondents did not differ significantly for statements 1, 2, 3, and 6 (**Table 2**), but were significantly different in terms of pathogen transmission (4), biogas treatment (5), and water quality impact (7) (p < 0.05). Statements 1 and 6 are related because they consider groundwater and contamination due to incorrect effluent disposal; regardless of the degree of education of the respondent, this relationship was identified by all profiles. All respondent profiles, esp. the dairy farm employees, also understood that the use of manure as fertilizer is a good practice. In terms of effluent as a transmission vehicle for pathogens, there was a significant difference between DF and PAC and between DF and OTHERS. In the case of statement 5, the most significant difference was observed between ST and OTHERS, being the highest average verified for ST. This group, which attended university, was more exposed to such specific knowledge and therefore more open to new practices and technologies.

### Needs and Barriers

Most of the respondents said that they would use manure to replace mineral fertilizers (79%), with slight differences between countries (Argentina 79%, Brazil 80%, and Chile 68%). One group (15%) would not consider the use of dairy manure on farms, also with differences between countries (Argentina 16%, Brazil 9%, and Chile 32%). The higher value for Chile was probably related to the increasing use of dairy manure on farms. In Argentina and Brazil, 5 and 11% of the respondents, respectively, did not answer these questions.

Regardless of the group of respondents, between 68 and 84% would use manure as fertilizer (**Figure 4**). However, the largest group that would not use manure was composed of dairy farmers (22%), with considerable differences between countries (68% Argentina, 15% Brazil, and 18% Chile). In this group, the lack of knowledge (29% Argentina and 47% Brazil) and the specific costs and regulations for Chile (33%) were identified as barriers. Within the group of hesitant respondents regarding the use of manure, most were also dairy farmers (68%), following the same trend between countries (60% Argentina, 15% Brazil, and 18% Chile) and the same barriers.

The participants were asked to select one or more different options from a list of five needs and barriers in terms of the use of dairy manure and slurry to replace mineral fertilizers (**Table 3**). Across all respondents (n = 593), 571 participants selected one or more needs (Argentina, n = 310; Brazil, n = 214, and Chile, n = 47), and 458 participants selected one or more barriers (Argentina, n = 224; Brazil, n = 188, and Chile, n = 49).

Of all respondents, 21% selected all needs (five options), with similar percentages for the three countries (Argentina 22%, Brazil 20%, and Chile 19%). Only 4% of the respondents did not select any option, corresponding to the 80% to the group that would not reuse manure. The mean total value representing the number of choices selected was 2.83 options for each respondent. For four

options, similarities were observed between the needs selected among the three countries. The most important needs were more equipment for the application and manuals or guides for manure management. Significant differences (p < 0.05; test Chi<sup>2</sup> ) were observed only for the option "Requirement of trained personnel," with higher values for Brazil.

FIGURE 4 | Intentions to reuse effluents or slurries as fertilizers according different groups of respondents in Argentina, Brazil and Chile expressed as number. DF, Dairy Farmers; PAC, Professionals advisors and consultants; DFE, Dairy Farm Employees; SIP, Service and Inputs Providers; ST, Students; OTHERS, representatives of public institutions working in agro-environmental policies and from dairy companies and researchers from academic institutions.


TABLE 3 | Needs and barriers selected for dairy manure management in the three South American Countries.

On average, the respondents selected one or two barriers; none selected all five options and only five individuals from different countries selected four options. Of all respondents, 23% did not select any barrier, with differences between countries (Argentina 28%, Brazil 13%, and Chile 28%). The options "non-existence of laws that impose them" and "the reuse of these wastes ends up being a complicated management" were most frequently selected, with differences between countries. Significant differences were observed for the barriers "lack of interest," "lack of knowledge," and "lack of legislation" (p < 0.05; test Chi<sup>2</sup> ). For all three countries, "high costs" and "cumbersome management" were barriers selected, with similar values between countries. Brazilians presented the lowest percentage of "lack of interest in using manure as fertilizer" and the highest percentage of "lack of knowledge." Although for Argentina, "lack of knowledge" was the first barrier identified, dairy producers mentioned the cost (27%), while professionals, advisors, and consultants mentioned cumbersome management (27%). Also, the lack of rules was recognized as an issue. It should be highlighted that all respondents showed an interest in slurry management (greater than 90%). For Chile, most of the needs were ranked similarly, apart from "trained personnel," which was ranked lower. In relation to barriers, "lack of knowledge" and "interest" were ranked lowest.

Needs and barriers, according to the categories of respondents, are shown in **Figures 5**, **6**. When considering respondents in the three countries in terms of advocating the use of manure as fertilizer (78%), the needs were identified as follows: 79% agreed that is important to have more equipment for the application on land, 60% required a manual or guide for application, 55% requested laboratories for analysis of manure and slurry, 52% believed it to be important to have legislations

and rules to regulate applications, and 49% considered it to be important to have trained personnel on the farms. In the other side the 15% that answered did no show that there is a lack of interest (94%), that they will not use because there are not any rule or law that oblige them to reuse waste (79%), a very expensive technical option (76%), a cumbersome and difficult management (70%), and a lack of knowledge for the application (60%).

Significant differences (p < 0.05; test Chi<sup>2</sup> ) in terms of three (more laboratories, more trained personnel, and specific legislation) of the five needs were observed between the types of respondents for all countries (**Figure 5**). The need for more laboratories was selected by professional advisors and consultants and by service and inputs suppliers. The need for more trained personnel was mainly selected by advisors and consultants. The importance of specific legislations for manure management was recognized by representatives of public institutions, dairy companies, and researchers.

On the other hand, significant differences (p < 0.05; test Chi<sup>2</sup> ) in terms of two ("lack of knowledge" and "lack of legislation") of the five barriers were observed between the different types of respondents for all countries (**Figure 5**). Notable, "lack of knowledge" was selected more frequently by dairy farm employees, while "lack of legislation" was selected more frequently by professionals, advisors, consultants, representatives of public institutions, dairy companies, and researchers.

### DISCUSSION

This is the first study analyzing stakeholders' perceptions on dairy manure management in Latin America, considering three different countries in which the dairy sector is an important industry. Our results contribute to the establishment of adequate programs to improve manure management, to the development of appropriate policies, and to the research and education programs. Similar studies have been carried out by Blok et al. (2015), who stated that this methodology allows a better understanding of the needs and perceptions of stakeholders, facilitating successful innovations for sustainable production and consumption. In a study in Denmark, Case et al. (2017) assessed the impacts of new legislative requirements for the processing of organic waste through the evaluation of the perception of producers toward the use of organic waste, while Hou et al. (2018) focused on how manure management is likely to be affected by a wide range of diverse socio-political and environmental factors.

The types of stakeholders participating in the present study were similar to those interviewed by Hou et al. (2018) across different European countries. All respondents had some experience regarding manure management, in contrast to the stakeholders interviewed in our study. The different percentages of respondents in our study between the three countries are due to the type of audience in the conferences, workshops, symposiums, and field days where the questionnaires were applied. These events had a technical and/or scientific profile, and more than 50% of the total of the respondents represented dairy farmers, professionals, and consultants involved in decisions related to manure management on dairy farms.

Despite the productive similarities between the three countries, we observed some differences in the perception of some issues. In particular, there were differences in terms of the levels of adoption and application of regulations and policies. This indicates that the same program, technology, or policy could achieve similar results in Chile and Argentina, but different ones in Brazil.

In the case of Chile (Salazar et al., 2007) and Argentina (Charlón et al., 2017a), the discussion about manure dairy

management and, consequently, the internalization of the theme by stakeholders is more advanced than in Brazil (Palhares, 2014), probably because in Chile and Argentina, the dairy industry has been more intense over a longer period of time, and the pressure from society is higher. In addition, previous studies (e.g., Hou et al., 2018) have mentioned that the use of manure is related to the availability of land, with higher pressure on countries with small farming areas. As mentioned before, the dairy production in Brazil has only been intensified recently (Palhares, 2014), and the pressure from environmental agencies and the society is therefore lower, facilitating the use of manure as fertilizer.

In Argentina, the strong perception of the importance of using manure as fertilizer could be explained because of the implementation of recent environmental initiatives, focusing on the agronomic use of the slurry. Currently, guidelines for adequate management practices are being developed by the sector industries.

In some countries, there are specific regional regulations related to manure and dairy effluent management (Argentina and Brazil), promoted mostly due to the pressure of society and as a result of pollution incidents. On the other hand, there are also agreements between the government and farmer federations, such as "Cleaner Production Agreements" (Chile), which promote a better use and management of dairy effluent. In addition, dairy companies implemented a bonus for their own producers that meet environmental standards, where effluent management is considered (Argentina and Chile; Charlón et al., 2017a).

In Argentinian dairy areas, water contamination with high concentrations of nitrate is an important environmental issue, and the high salinity in these areas could further affect dairy production and impede the reuse of manure (Carbó et al., 2009; Charlón and Herrero, 2012; Charlón et al., 2017b). In this sense, stakeholders might be more conscious in relation to water contamination, water quality issues, and pathogen transmission compared to stakeholders in Brazil and Chile. Besides, this situation could be related to the diffusion of these issues in the mass media and in local workshops and seminars. Most of the stakeholders know that excess salinity of groundwater may be an important restriction for using cattle manure because of the potential soil salinization in soils in dairy land area in Argentina (Charlón and Herrero, 2012). In Argentina, the local community perceptions about the pollution of surface and groundwater were studied by Peluso and Usunoff (1997). These authors found that in general, the community considers those environmental problems as important, such as the pollution of surface water caused by sewage. Similar results regarding such awareness were reported by Sudarmadi et al. (2001) in terms of the perception of environmental and health problems, both in an educated group and a community group in Indonesia. The authors observed a better understanding of such problems when broad information was supplied by newspapers, television, movies, and the radio.

When comparing the three countries, all Brazilian averages to statements 1 to 4 were lower than those for Argentina and Chile; therefore, the experience of these countries could help to enhance the knowledge and propose policies to dairy manure management in Brazil, thereby improving the stakeholder's perceptions of the environmental impacts. The highest Brazilian valuation in relation to biogas use, which was also the highest average among all means, may be the result of the influence of two current governmental programs. One of them is the National Plan on Climate Change and the National Program for a Low Carbon Agriculture. This plan encourages the adoption of sustainable production systems to ensure GHG emission reductions while raising the incomes of framers, particularly with the adoption of technologies such as the biodigestion of animal wastes. Another program is the Brazilian Electrical Energy Agency regulation, which permits farmers to generate electricity credits through biogas production (ANEEL, 2012). Wissman et al. (2013) mentioned that programs that work with economic incentives can modify decision prioritization by farmers. Anaerobic digestion is widely used where financial incentives are linked to renewable energy policies, making it a profitable activity even at a modest scale (Loyon et al., 2016). Also, the growth of anaerobic digestion in Denmark is largely due to an incentive policy such as investment support for construction of biogas plants and government support strategies to increase interactions between various social groups (Raven and Gregersen, 2007). Another important aspect is the history of biodigesters in Brazil. The first system has been installed in a dairy farm in 1979 and has been part of the sector ever since. In the case of Argentina, the temperate climate, the farm scale, and the lack of appropriate financial support to implement this technology have been factors that make difficult the adoption of this technology (Charlón et al., 2017a). In Chile, anaerobic digesters on dairy farms are extremely rare (c. < 1%) (INIA, 2016). However, recently, a program has been developed to incentive the use of biodigesters on dairy farms. Important restrictions will mainly be the low potential for methane production of dairy slurry, the high cost of anaerobic digestion plants, and the low dry matter content (and organic matter content) of dairy slurries based on grassland systems (Salazar et al., 2007).

We observed a contradiction in terms of the perceptions in Brazil about issues 1 and 6. On the one hand, the three countries strongly agree that "groundwater may be contaminated by effluent lagoons," similar to respondents of Argentina and Chile. However, issue 6, which is linked to the same problem and says "the shallow aquifer may become contaminated by disposing effluents in the lowlands," has not received the same perception in Brazil. Although both issues are linked, they are two forms of water contamination associated with the disposal of manure. This situation could be explained because there is an obligation by Brazilian state environmental legislations that all effluent storage ponds should be waterproofed (Palhares, 2008). This is probably the reason why Brazilians do not consider storage lagoons as a source of groundwater contamination. The impact of unsealed effluent storage lagoons was verified by Drommerhausen et al. (1995), who developed an extensive study. They evaluated the impact of effluent lagoons on different soil types in eight dairy farms in the United States and demonstrated that effluent lagoons receive a continuous load of water with excreta daily; when they are not well constructed and sealed, they may become permanent source of groundwater contamination.

The statement "manure is a good fertilizer" obtained the highest agreement. Hou et al. (2018) showed that the use of treatment manure technologies is low in regions that have sufficient land for manure applications, as is the case in Argentina, Brazil, and Chile, where dairy slurries are mainly applied directly to the soil without treatment (Salazar et al., 2010).

Foged et al. (2011) found that less than 10% of the total animal manure were treated in the EU-27 in 2010. We do not have this type of statistical information for any of the countries in our study, but according to the authors' expertise and observation, the use of treatment technologies is low and will remain low due to technical and/or economic limitations. Regardless of the productive and environmental realities of each country, it can be said that the respondents understand this practice as a way to dispose of manure, reuse water, and recycle nutrients. It is also possible to achieve a livestock sector with low carbon emissions by better nutrient recycling.

The question remains whether manure as fertilizer is being used correctly, as there are several studies that show that misuse results in environmental impacts on water, soil, and air. Knowlton and Ray (2013) mentioned that animal manure can be a valuable resource for farmers, providing nutrients, improving the soil structure, and increasing the vegetative cover to reduce the erosion potential. At the same time, the application of manure nutrients in excess of crop requirements can result in environmental contamination. To mitigate environmental risks, governments in Western Europe (Sutton et al., 2010), North America (Compton et al., 2011; US Environmental Protection Agency Science Advisory Board-USEPA, 2011), and Oceania (Ministry of the Environment, 2012) have enacted legislations to control livestock expansion, manure land-spreading, and other farm practices.

Excessive nutrient accumulation and plant uptake may impact animal health and production (Djekic et al., 2014). The use of manures, wastes, composts, and sludges as fertilizers and soil physical and chemical conditioners is advisable, but only when performed considering the concept of nutrient balance and the four Rs (right product, right rate, right time, right location) (Oenema et al., 2003). Waldrip et al. (2015) showed how imbalances between nutrient imports and managed exports can result in nutrient losses to the environment and additions to soil storage, limiting the sustainability of livestock production systems.

Another aspect is the low pressure from governmental legislations. In this context, only recently, a specific legislation for manure management has been launched in Cordoba, Argentina (Charlón et al., 2017a), and most of the manure management is regulated in these countries according to general environmental legislations and regulations in order to avoid water, soil, and air pollution.

The need considered most important by respondents across all countries was "greater variety of equipment." This situation is understandable, because in these three countries, technologies for manure application are relatively new, and new technologies and equipment are only developed relatively slowly. According to Salazar et al. (2010), dairy slurry is applied using mainly surface equipment (e.g., irrigation pumps and tank spreaders). In the study carried out by Hou et al. (2018), it was observed that in European countries, where manure has been applied to the soil over a large number of continuous years, the dominant needs were "reduction of excessive costs," "pathogen control prior to application," and "adaptation of treatment and management systems to changes in legislation."

The second most important need, "manual or guide for manure application," is directly related to the first barrier "lack of knowledge." This indicates that countries should edit technical materials such as manuals, factsheets, etc., taking into account local conditions to help producers, dairy employees, and consultants to adequately use manure as fertilizer. It is important that these materials are recognized as referential by all stakeholders, otherwise they will not be effective in improving the practice. Due to the particularities of each country, these materials should be produced in particular, but it is interesting that there are minimum contents and performance indicators agreed upon between countries, such as the use of the concept of nutrient balance, the best forms of manure disposal, the potential environmental risks, etc. In this way, countries will be able to generate common indicators that will assist in the evaluation of their own policies and programs as well as in the design of common actions. This common information is also important for proposing research projects and strengthening research networks among countries.

In European countries, a key step in improving the efficiency of agriculture and reducing negative impacts on the environment has been the publication of Codes of Good Agricultural Practice (GAP), which provides guidelines for farmers, taking into account manure management on farms. In addition, a set of more "friendly-reading" publications have been published by the MAFF in the United Kingdom, covering aspects of manure characterization and management for use on crops and grassland (Dampney et al., 2000). Complementary strategies have been implemented for different countries. Another approach has been the development of decision support systems, using electronic calculation worksheets and models to predict the value of manure and potential N losses (e.g., Nicholson et al., 2013). A similar approach could be implemented in South American countries, using current research information and generating guidelines, recommendations, and electronic tools to improve manure management on dairy farms. As a stage prior to specific manure management regulations, the dissemination of GAP could be beneficial for the sector.

In Denmark, the barriers selected by farmers in terms of manure use as organic fertilizer were evaluated by Case et al. (2017). The most important problems were odor nuisance, unreliable nutrient content, difficult planning, expensive machinery, and the absence of a quality certification. In our study, we also encountered some of these barriers (e.g., cumbersome management), although some were rather identified as needs (manure equipment, lab analysis).

In other countries, such as in the EU, where regulations have been in place for years, different stakeholders feel pushed by the new legislation to treat dairy wastes. The use of high amounts of mineral fertilizers per unit area, with increased costs of these alternatives, makes manure a more attractive alternative. On the other hand, in these countries, farmers are generally pressed to export manure from farms with higher animal density to areas of agricultural production (Hou et al., 2018). This situation is different in South American countries, were dairy production systems generally have low stocking rates that do not exceed one cow per hectare and where lands are available for manure spreading. Also, in these countries, crop production for feed is important, and such production is generally located near the dairy production area, enabling the combination of dairy and crop production.

According to these results, transferring knowledge by different technology activities will be important in these three countries. It is necessary to develop educational/training strategies (written and oral) so that farmers, employers, and consultants have technical guidelines on how to manage and transfer knowledge related to dairy manure. It is also important to internalize this issue in the University curricula and in technical discussions of stakeholders. In addition, an important technology transfer target will be farmer federations, milk companies, and the public sector, where scientific base information should be transferred to complete such knowledge. The agricultural economic literature shows that innovations do not occur randomly, but rather that incentives and government policies affect the nature and the rate of innovation and adoption. Both the generation of new technologies and their adoption are affected by intentional public policies (e.g., funding of research and extension activities), unintended policies (e.g., manipulation of commodity prices), and activities of the private sector (Sunding and Zilberman, 2001).

Policy makers and stakeholders should focus on appropriate technologies and "win-win" strategies to effectively generate enthusiasm to reuse manure and slurry on the farms. It is necessary to continue the research and transfer in subjects such as the efficient use of nutrients from manure and effluents, mitigating the possible negative impacts. In this context, information and advice based on research have been important aspects of dealing with the environmental problems associated with agriculture in most of the countries in Europe (Thevenet et al., 1993). As farmers are being subject to increasing pressure from the public to reduce environmental impacts, there is a considerable need to provide extended information. There is also an overriding need for farmers and their advisers to understand and accept the impacts of agriculture on the environment and to have the confidence to use technical solutions developed to reduce nutrient losses. Evidently, research institutes and extension services have to provide farmers with appropriate information and tools (Magette, 2000). Such measures should be implemented through a combination of the different actors to eliminate technological barriers for the convenience of the product (Case et al., 2017).

In Argentina, Brazil, and Chile, farmers already have some experience in the regulation of dairy wastes, but because of the current trend toward intensified production, increasing farm scales, and social issues, these regulations need to be improved considering the productive, social, economic, and environmental characteristics of each country. The results from this survey can support actions and programs to disseminate manure management practices in the dairy sector. In this sense, research plays a fundamental role, since environmental and productive standards should be proposed to guarantee environmental conservation and economic viability.

Finally, there are many opportunities and options for improving the manure management of a dairy farm; however, there is no single model applicable for all farms. In addition, the stakeholder's perceptions could change according to different drivers such as public perception, environmental legislations, or own interests. Efforts should therefore focus on the different dairy production systems and the particular soil and climatic conditions where the farms are located. It is also important that the technology and management practices proposed to the farmers include an economic assessment, especially against the background of the current economic pressures (Magette, 2000).

### AUTHOR'S NOTE

This paper is an international research that was carried out in three countries. The researchers MH and VC (Argentina), JP (Brazil), FS (Chile) are part of the Manure research network in South America and have a broad expertise in the different manure management systems. They are all involved in a collaborative research in manure management. MT is a Ph.D. candidate in nutrient and manure management and its environmental impact.

### AUTHOR CONTRIBUTIONS

In this paper, MH, VC, JP and FS have been involved in the survey development and data collection in each country, in its analysis and in the discussion of the results comparing the different countries and stakeholders. In particular FS was involved in Chilean results, VC in Argentinean results and JP in Brazilian results. MT worked in the acquisition of data in Argentina and in drafting the manuscript. AP was responsible for all the

# REFERENCES


statistics analysis and also she interacted with the all the group of authors in drafting the manuscript. MH was the responsible of the conception and design of the study and worked in the Argentinean interpretation of data and in drafting the whole manuscript.

# FUNDING

CONICyT - CHILE Programa de Cooperación internacional Proyectos de apoyo a la formación de Redes Internacionales entre Centros e Investigación (2016-2017) Red Manure South: research network on effluents and water of dairy grazing systems in South America countries. UBACyT Program 2014-2017 project 20020130100498BA CyT-UBA and INTA, Argentina project PNPA 1126043.

### ACKNOWLEDGMENTS

Chilean Research Council, CONICYT-REDES project 150086 ManureSouth Network. We also want to acknowledge to all the farmers and professionals involved in the three countries. For the professionals who helped with the survey data collection: Marcos Bontá, Ana Valeria Gonzalez and Nicolás Sassano (University of Buenos Aires) and Ezequiel Kern, Gustavo Almada, Monica Moretto, Daniela Faure, Maria Rosa Scala (INTA) from Argentina, and Marion Rodríguez and Alejandra Jiménez in Chile.


in Experiences From Other Countries. Swedish Board of Agriculture, ed S. B. O. Agriculture (Jönköping), 1–50.

Wolf, C. A., Tonsor, G. T., McKendree, M. G., Thomson, D. U., Swanson, J. C. (2016). Public and farmer perceptions of dairy cattle welfare in the United States. J. Dairy Sci. 99, 5892–5903. doi: 10.3168/jds.2015-10619

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Herrero, Palhares, Salazar, Charlón, Tieri and Pereyra. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Tier 3 Method for Enteric Methane in Dairy Cows Applied for Fecal N Digestibility in the Ammonia Inventory

### André Bannink <sup>1</sup> \*, Wouter J. Spek <sup>1</sup> , Jan Dijkstra<sup>2</sup> and Leon B. J. Šebek <sup>1</sup>

<sup>1</sup> Wageningen Livestock Research, Wageningen University and Research, Wageningen, Netherlands, <sup>2</sup> Animal Nutrition Group, Wageningen University and Research, Wageningen, Netherlands

### Edited by:

Tom Misselbrook, Rothamsted Research (BBSRC), United Kingdom

### Reviewed by:

Michel A. Wattiaux, University of Wisconsin-Madison, United States Laura Zavattaro, Università degli Studi di Torino, Italy

> \*Correspondence: André Bannink andre.bannink@wur.nl

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 09 April 2018 Accepted: 18 September 2018 Published: 05 November 2018

### Citation:

Bannink A, Spek WJ, Dijkstra J and Šebek LBJ (2018) A Tier 3 Method for Enteric Methane in Dairy Cows Applied for Fecal N Digestibility in the Ammonia Inventory. Front. Sustain. Food Syst. 2:66. doi: 10.3389/fsufs.2018.00066 The current inventory of N emission from cow excreta relies on fecal N digestibility data in Dutch feeding tables, assuming additivity of dietary ingredients to obtain diet values (CVB model). Alternatively, fecal N digestibility can be estimated by a dynamic, mechanistic model of digestion in the gastrointestinal tract, currently used as Tier 3 for enteric methane prediction in the Netherlands (Tier 3 model). Estimates of in situ rumen degradation characteristics for starch, neutral detergent fiber (NDF) and crude protein used as an input for the Tier 3 model were based on Dutch feeding tables (the protein evaluation system). Both methods were evaluated on independent dataset on fecal N digestibility that was constructed from peer-reviewed papers on N balance data for dairy cows published since 1999 (54 trials, 242 treatment means). Results indicate that observed apparent fecal N digestibility (67.0 ± 6.77%) was systematically over-predicted in particular by the CVB model (73.8 ± 4.35%) compared to the Tier 3 model (69.8 ± 4.52%). For the dataset including only observations from Dutch trials the observed fecal N digestibility (70.4 ± 7.33%) was also systematically over-predicted by the CVB model (76.4 ± 5.27%) but not by the Tier 3 model (69.7 ± 5.81%). Mixed model analysis with study as random factor indicated the slope of the regression between observed and predicted fecal N digestibility to be smaller than 1, in particular for the CVB model (CVB model slope varied between 0.405 and 0.560 and Tier 3 model slope between 0.418 and 0.657). The over-prediction by the CVB model with 6–7%-units of digestibility will lead to an over-predicted ammoniacal N excretion (urinary N) in the ammonia inventory, and biased estimation of N mitigating potential of nutritional measures. The present study demonstrates the benefit of using the Tier 3 model to predict the average level of apparent fecal N digestibility compared to the CVB model. The general estimates of in situ rumen degradation characteristics for starch, NDF and crude protein used as input for the Tier 3 model seemed applicable for the Dutch trials but less so for the non-Dutch trials.

Keywords: models, Tier 3, dairy cows, nitrogen digestibility, nitrogen excretion

# INTRODUCTION

Ammonia emitted from dairy production systems is a major water and air pollutant, leading to eutrophication, acidification and fine particulate matter formation. These emissions are reported annually to the European Commission, according to the Göteborg and Kyoto protocols (e.g., Nielsen et al., 2018; Wakeling et al., 2018). Most inventory efforts adopt the concept of ammonia emission factors specific for an animal category or type of agricultural activity (Paulot et al., 2014). This implies a specific emission factor has to be allocated a priori to every management practice or abatement measure accounted for in the model. Actual modeling of the cause of variation in ammonia emission requires representation of details of the emission process itself. An ammonia emission model has been developed for inventory purpose in the Netherlands (Velthof et al., 2012) including the ammonia emissions from animal excreta. A crucial element in this model is the prediction of the urinary excretion rate of potentially volatile nitrogen, often referred to as total ammoniacal nitrogen (TAN). The proportion of nitrogen (N) containing components in urine that is susceptible to almost instant volatilization varies considerably (Dijkstra et al., 2013, 2018), and mineralization of organic manure N (fecal N) also contributes to TAN (Vonk et al., 2016). For reasons of simplicity we refer to TAN as being the total amount of N excreted with urine, irrespective of the form of N present in urine and excluding a further input to TAN from mineralization of fecal excreted N.

In ammonia inventory methodology, accurate estimates of apparent fecal N digestibility are required to allow calculation of TAN excretion rate. This rate is calculated as the amount of N ingested that is apparently digested at the level of feces (by taking N intake times apparent fecal N digestibility of ingested feed), minus the amount of N retained in milk, body tissues, and offspring. Data on N intake and N retained by the cow can be retrieved from the activity data available in the inventory in the Netherlands (Velthof et al., 2012). Apparent fecal N digestibility is obtained or calculated from values for the dietary components given in Dutch tables of feed values for ruminant nutrition, also indicated as the CVB Feed Table (CVB, 2011; referred to from hereon as CVB model). These values have typically been determined in experiments with wethers rather than with cattle. However, since the introduction of the systems of evaluation of net energy for lactation (Schiemann et al., 1971; Van Es, 1978) differences in apparent fecal N digestibility between wethers and cattle have been documented. The Dutch evaluation system of net energy of lactation (VEM; Van Es, 1978) is part of the CVB Feed Table (CVB, 2011). These values are not directly applied with the purpose of estimating apparent fecal N digestibility in dairy cattle, but they are used in calculations of the energy value of feeds. For these reason some doubts could be raised on the accuracy of the current TAN excretion calculation for dairy cattle in the Netherlands. A preliminary evaluation (unpublished) of apparent fecal N digestibility predicted with the CVB model confirmed these doubts. Evaluation against a dataset of 69 dietary treatments from 13 trials indicated a large systematic over-prediction of apparent fecal N digestibility with 7.5 (±5.4) percent units of digestion, corresponding with 11.4% higher predicted than observed values for apparent fecal N digestibility (**Figure 1**). Prediction error appeared negatively related to the level of apparent fecal N digestibility and to the fraction of roughage in the diet, and positively related to DM intake (R <sup>2</sup> = 0.26, 0.15, and 0.11, respectively).

Hence, it appeared that estimates of apparent fecal N digestibility with the CVB model might be biased. The aim of the present study was to evaluate the CVB model, as well as an alternative, more detailed candidate model, against an independent dataset of rather recent observations on apparent fecal N digestibility in dairy cows documented in peer-reviewed literature. As the alternative candidate model, a Tier 3 approach (from here on referred to as Tier 3 model) was chosen which is already in use to estimate enteric methane in dairy cattle (Bannink et al., 2011) in the greenhouse gas inventory in the Netherlands (Vonk et al., 2016).

### MATERIALS AND METHODS

### Data Collection Evaluation Databases

A literature search of the Scopus on-line database was conducted using the combination of words "dairy cattle OR dairy cows," "digestibility OR digestion." and "protein." The period covered was 2000–2016 and the search resulted in 1,207 articles. In order to be included in the dataset, studies had to provide information with respect to the ingredient composition of the diet, dry matter intake (DMI), and apparent fecal protein digestibility. Furthermore, as the CVB Feed Table (CVB, 2011) was used for recalculation of the diets, only those studies were selected in which the ingredients used were also present in the CVB Feed Table (CVB, 2011). Studies were removed from the database if grass silage was inoculated, cow body weight (BW) was lower than 550 kg and breeds other than Holstein Friesian were involved. Some digestion trials carried out by our own research group in the Netherlands were added to this database. This selection process resulted in an evaluation dataset containing a total of 54 studies containing 58 experiments and 242 treatment observations, including 9 Dutch studies containing 13 experiments and 62 treatment observations. A summary of cow and dietary characteristics for the selected studies is given in **Table 1** for the complete evaluation dataset, and in **Table 2** for the dataset of Dutch studies only. The 54 studies included in this analysis are listed in the footnotes of **Tables 1** and **2**.

Performance of the CVB model and the Tier 3 model was evaluated for the three different datasets: (1) the complete dataset including the diets containing rolled or cracked products, (2) the complete dataset excluding the diets containing rolled or cracked high moisture maize silage because for these products in particular representative data were lacking in the CVB Feed Table (CVB, 2011), and (3) a dataset containing the data from Dutch studies only.

### Recalculation of Diets

Diets composition was recalculated using the CVB Feed Table (CVB, 2011) and, as far as available, analyzed nutrient composition of concentrates and roughages were used as inputs for the recalculation of diets. The unidentified fraction

of the dietary DM (in g/kg DM) was calculated as 1,000 crude protein (CP; excluding ammonia CP)—ammonia—crude fat—crude ash—neutral detergent fiber (NDF)—starch—sugar fermentation products. This unidentified fraction was equally allocated to NDF and starch in cases when dietary starch content was higher than sugar content but equally allocated NDF and sugars in cases when sugar content was higher than the starch content. This was a pragmatic solution to allocate 100% of DM including the unidentified part which also contributes to fermentation, microbial growth, digestion and excretion. In a number of cases, rolled high moisture maize silage was used (involving North American studies) and for this product the values in the CVB Feed Table (CVB, 2011) for corn cob mix silage were adopted.

fecal N digestibility from 47.1 to 78.1%.

### Calculation of Model Input Parameters

The required model input parameters for the Tier 3 model (and required input parameters related to CP for the CVB model) are summarized in **Table 3**. Ruminal in situ fermentation characteristics are required for starch, NDF and CP of the individual feed ingredients. These rumen fermentation characteristics for the individual feedstuffs include the washout fraction, the (non-washout) degradable fraction and the (nonwashout) undegradable fraction of starch, NDF, and CP, as well as the respective ruminal in situ fractional degradation rates of the degradable fraction of starch, NDF, and CP. Values were adopted from those applied in the DVE/OEB2010 system (Van Duinkerken et al., 2011) as part of the CVB Feed Table (CVB, 2011). This feed evaluation system estimates requirements and supply of intestinal digestible protein in dairy cattle.

### CVB Model

For all feedstuffs, the CVB Feed Table (CVB, 2011) contains estimates of the coefficient (%) of apparent fecal digestibility of CP (either as table values for concentrate ingredients, or as predictive equations for roughages; CVB, 2007). Digestibility data for the dietary components or ingredients were weighted according to their contribution to dietary DM.

### Tier 3 Model

The Tier 3 model used in the present study to predict apparent fecal N digestibility, as an alternative to the CVB model, has been used in the greenhouse gas inventory in the Netherlands since 2005 to predict enteric methane emission in dairy cattle (Vonk et al., 2016). The Tier 3 model is a dynamic, mechanistic model describing the fermentative and digestive processes in the gastrointestinal tract of dairy cattle. The model is strongly based on the rumen and fermentation model of Dijkstra et al. (1992). This model was adapted by Mills et al. (2001) on postruminal digestion of nutrients and fermentation in the hindgut. Subsequently, it was adapted by Bannink et al. (2008) on the representation of the stoichiometry of production of volatile fatty acids from fermented substrate (soluble carbohydrates, starch, hemi-cellulose, cellulose and CP). Kebreab et al. (2004) used an extended version of this model, including prediction of nutrient utilization for milk yield according to Dijkstra et al. (1996), for the US greenhouse gas inventory purposes. The Tier 3 model represents fermentation and microbial metabolism processes in the rumen, including variation in microbial protein synthesis related to the type of carbohydrate and N-source available,



<sup>a</sup>Data derived from Agle et al. (2010a,b), Akbari-Afjani et al. (2014), Arndt et al. (2015), Arriola et al. (2011), Bahrami-Yekdangi et al. (2014), Bahrami-Yekdangi et al. (2016), Beckman and Weiss (2005), Benchaar et al. (2013), Boerman et al. (2015), Brito and Broderick (2006), Brito et al. (2009), Broderick et al. (2000), Broderick et al. (2001), Broderick et al. (2002), Broderick and Radloff (2004), Broderick et al. (2009), Broderick and Reynal (2009), Colmenero and Broderick (2006), Dann et al. (2014), Doreau et al. (2014), Eun et al. (2014), Fanchone et al. (2013), Flis and Wattiaux (2005), Fredin et al. (2015), Hassanat et al. (2013), Hatew et al. (2015), Hatew et al. (2016), Hindrichsen et al. (2006), Khezri et al. (2009), Klevenhusen et al. (2011), Kowsar et al. (2008), Maesoomi et al. (2006), Mohammadzadeh et al. (2014), Mosavi et al. (2012), Olijhoek et al. (2016), Petit (2002), Peyrat et al. (2016), Poorkasegaran and Yansari (2014), Rafiee-Yarandi et al. (2016), Ruppert et al. (2003), Sinclair et al. (2015), Spek et al. (2013), Stojanovic et al. (2014), Tas et al. (2005), Valk et al. (2000), Warner et al. (2013a), Warner et al. (2013b), Warner et al. (2015), Warner et al. (2016), Weiss et al. (2009), Weiss et al. (2011), Yang and Beauchemin (2006), Yang and Beauchemin (2007).

<sup>b</sup>The unidentified fraction (g) in 1 kg of dietary DM is calculated as 1,000—CP (excluding ammonia CP)—ammonia—crude fat—crude ash—NDF—starch—sugar—fermentation products.

<sup>c</sup>N balance results includes observations for all dairy cows, including lactating as well as non-lactating cows.

<sup>d</sup>For 155 treatments (36 experiments) urine N was not observed but estimated as apparent fecal N digested minus N in milk.

retention time of substrate, acidity of rumen contents, intraruminal microbial N recycling, recycling of N to the rumen via saliva and through the rumen wall. The model distinguishes bacteria and protozoal metabolism and predicts variation in rumen (and large intestinal) metabolism instead of adopting fixed values (reviewed by Bannink et al., 2016). Representation TABLE 2 | Summary statistics of cows and diets in the evaluation dataset, containing Dutch experiments only<sup>a</sup> .


<sup>a</sup>Data derived from Hatew et al. (2015), Hatew et al. (2016), Spek et al. (2013), Tas et al. (2005), Valk et al. (2000), Warner et al. (2013a), Warner et al. (2013b), Warner et al. (2015) and Warner et al. (2016).

<sup>b</sup>The unidentified fraction (g) in 1 kg of dietary DM is calculated as 1,000—CP (excluding ammonia CP)—ammonia—crude fat—crude ash—NDF—sugar—starch—fermentation products.

<sup>c</sup>For 46 treatments (10 experiments) urine N was not observed but estimated as apparent fecal N digested minus N in milk.

of such aspects is relevant to prediction of variation in outflow of microbial and non-microbial protein from the rumen and its subsequent digestion in the intestines.

For the present study, rather limited adaptations were made to the model. These adaptations did not affect predicted enteric methane emission (Vonk et al., 2016), but they were required for accurate prediction of apparent fecal N digestibility. A representation of endogenous protein in the small intestine was included, similar to that in the DVE/OEB 2010 system (Van Duinkerken et al., 2011). The following equation was used to calculate the production rate of endogenous protein (CPEnd; g/d) from the flow of undigested feed (Van Duinkerken et al., 2011): CPEnd= 50 × Feed ((1-DCOM/100) + FrAsh/1000 × 0.5), with Feed as DM intake (kg/d), DCOM as fecal digestibility of organic matter (% of organic matter intake), FrAsh as the fraction of crude ash in feed (g/kg DM), and assuming 50% fecal digestibility of crude ash. It was assumed that 60% of the ileal outflow of endogenous protein and microbial crude protein to the large intestine is potentially degradable in the large intestine.

## Apparent Fecal N Digestibility in Inventory Methodology

Activity data for the diet and performance of the average dairy cow in the Netherlands allowed to generate model inputs for the Tier 3 model to predict apparent fecal N digestibility (% of N intake). The activity data include statistics on the number of dairy cows in the Netherlands, delivered and recorded amounts and composition of tank milk in two identified regions (i.e., the North and West, and the East and South). The data include for each region an estimate of DM intake by cows based on milk yield and milk composition, components in the cow ration and the feeding of these components. The specific task to collect the statistics for these data is allocated to the WUM working group

TABLE 3 | Model inputs required for prediction of apparent fecal N digestibility with the Dutch Tier 3 model (and some for the CVB model<sup>a</sup> ).


<sup>a</sup>Parameter inputs required with predictive equations for roughages in the CVB model.

<sup>b</sup>Model inputs according to Dijkstra et al. (1992) and the Dutch Tier 3 for prediction of enteric methane in cows. See Bannink et al. (2011) for further explanation. c kd, fractional rate of degradation during in situ incubation in the rumen under standardized conditions.

TABLE 4 | Summary of observed and predicted apparent fecal N digestibility (as % of N intake) for the complete dataset, the complete dataset excluding rolled or cracked high moisture maize silage, and the Dutch dataset.


who interpret statistics of the aforementioned data according to a standardized methodology (CBS, 2010). A yearly estimate is delivered of the proportion of each component in dietary DM (grass herbage, grass silage, maize silage, standard concentrates, protein-rich concentrates, and wet by-products), DM intake, and milk yield and composition. Statistics (across season and across postal code reflecting soil and farm type) on the chemical composition, digestibility and feeding value of roughages are obtained from a commercial laboratory (https://www.eurofins. com/agro) that analyses the majority of silage samples offered for analysis by Dutch dairy farmers as almost all Dutch dairy farmers offer samples their silos for analysis. Furthermore, estimates are made for which part of prepared silages are fed within the year of preparation or are fed in subsequent years. The approach was consistent with that followed in the inventory of enteric methane in dairy cattle in the Netherlands from 1990 till 2016 (Vonk et al., 2016). Required model inputs have been described by Bannink et al. (2011) and are listed in **Table 3**. Values achieved with the CVB model have been drawn from the inventory reports in the Netherlands and were available from 1990 till 2014.

### Statistical Analysis

Data were analyzed based on the three different datasets; (1) the complete dataset of observations without any exclusions, (2) the complete dataset excluding diets containing rolled or cracked products, and (3) the dataset containing observations from Dutch studies only. A separate analysis was carried out for the complete excluding rolled or cracked products because no similar products have been listed in the Dutch Feed Table (CVB, 2011) and assumptions on inputs were hence expected to be rather inaccurate. A separate analysis was carried out for the Dutch dataset because the values adopted in the DVE/OEB 2010 system (Van Duinkerken et al., 2011) on rumen degradation characteristics of feedstuffs and diet components are thought to be most reliable for trials performed in the Netherlands. These values have typically been established in in situ degradation studies conducted under Dutch feeding conditions. These values are likely less applicable to the same type of feedstuffs or dietary components (especially for roughages) tested in other countries due to differences in climate, harvest management, varieties of maize and grass, and post-harvest treatment and conservation.

Data were analyzed using the PROC MIXED procedure of SAS (version 9.3). Model predicted apparent fecal N digestibility (%) based on the Tier 3 model and the CVB model were tested against observed apparent fecal N digestibility. Study effect was included as a random effect in the model. Model predictions of apparent fecal N digestibility were evaluated using two methods as described in Ellis et al. (2010). The square root (RMSPE) of the mean square prediction error (MSPE) was calculated and expressed as percentage of the observed mean. The RMSPE was decomposed into error due to overall bias (ECT), error due to deviation of the regression slope from unity (ER), and error due to the disturbance (random error) (ED) (Bibby and Toutenburg, 1977).

Furthermore, concordance correlation coefficient analysis (CCC) was performed (Lin, 1989) where CCC is calculated as:

$$\text{CCC} = \text{R} \times \text{Cb},\tag{1}$$

where Cb is a bias correction factor and is a measure of accuracy, and the R variable (the Pearson correlation coefficient) gives a measure of precision. A higher CCC value indicates a better prediction of observed values. The Cb is calculated from SD<sup>O</sup> and SD<sup>P</sup> as the standard deviation of observed and predicted values, respectively, and M<sup>O</sup> and M<sup>P</sup> as the mean of observed and predicted values respectively, where υ provides a measure of scale shift (i.e., the change in standard deviation between predicted and observed values), and µ provides a measure of location shift (i.e., under-prediction

with a positive value and over-prediction with a negative value):

υ = SD<sup>O</sup> / SDP, µ = [M<sup>O</sup> – MP] / [SD<sup>O</sup> × SDP] 1/2 , Cb = 2/[υ + 1/υ + µ 2 ].

### RESULTS

### Model Evaluation Results

**Table 4** shows summarizing statistics of observed and predicted values of apparent fecal N digestibility (as % of N intake) for the three datasets. For the complete dataset, observed variation in apparent fecal N digestibility (referring to the SD values reported in **Table 4**) was 50 and 56% greater than predicted by the Tier 3 model and CVB model, respectively. For the complete dataset excluding rolled or cracked products it was 32 and 43% greater, and for the Dutch dataset 26 and 39% greater, respectively. The CVB model over-predicted apparent fecal N digestibility by 6.8% units of digestibility for the complete dataset, by 6.0% units for the complete dataset excluding rolled and cracked products, and

by 6.0% units of digestibility for the Dutch dataset (**Table 4**). The Tier 3 model over-predicted by 2.8 and 1.7% units and under-predicted by 0.7% units for these datasets, respectively. Results have been represented graphically in **Figure 2** for the complete dataset and **Figure 3** for the Dutch dataset.

More qualifying statistics on model prediction performance are given in **Table 5**. The relationship between observed and predicted apparent fecal N digestibility, taking into account the study effect, indicated a better prediction by the Tier 3 model. Slope estimates (between 0.418 and 0.657) were greater and intercept estimates smaller (between 23.4 and 41.6) with the Tier 3 model compared to slope estimates obtained for the CVB model (slope estimates between 0.405 and 0.560; intercept estimates between 37.2 and 46.4; **Table 5**).

Consistent results were obtained with RMSPE and CCC analysis of overall performance for these three datasets. A consistently smaller RMSPE was established with the Tier 3 model compared to the CVB model for the complete dataset, the complete dataset excluding rolled or cracked products and the Dutch dataset (a 1.5, 1.4, and 16% units of digestibility smaller prediction error with the Tier 3 model, respectively, based on RMSPE and observed means in **Table 5**). Predictive performance in terms of RMSPE value was highest with the Dutch dataset and lowest with the complete dataset. The RMSPE for the Tier 3 model was almost totally attributed to error due to disturbance (ED) with % of RMSPE attributed to bias (ECT) and regression (ER) not exceeding 16% (**Table 5**). In contrast, with the CVB model most error was due to the bias (more than 58%; ECT) and the contribution of disturbance error (DE) to total error was about half that of the Tier 3 model.

In line with RMSPE results, the results from CCC analysis indicate that predictive performance of both the Tier 3 model and the CVB model improved in the order of the complete dataset, the complete dataset excluding rolled or cracked products and the Dutch dataset. The CCC value increased, the Pearson correlation coefficient R changed in the direction of 1, as well the bias correction factor Cb and the υ parameter indicating scale shift, whereas the µ parameter indicating location shift became smaller (results for both models in **Table 5**). Simultaneously, the CCC value (a high value indicating better prediction) increased by 0.25 for the Tier 3 model and by 0.15 for the CVB model. With the complete dataset, performance by the CVB model was similar comparable to that of the Tier 3 model with a CCC value of 0.32 (**Table 5**). This changed into a slightly better performance by the Tier 3 model for the complete dataset excluding rolled or cracked products (0.04 higher CCC value), and a better performance with the Dutch dataset (0.09 higher CCC value). For all three datasets tested, precision (R; **Table 5**) was higher for the CVB model, whereas accuracy (Cb; **Table 5**) was higher for the Tier 3 model. Higher accuracy (Cb) of the Tier 3 model remained with differences in Cb value of 0.29, 0.33 and 0.31 for the complete dataset, the complete dataset excluding rolled or cracked products and the Dutch dataset, respectively. Precision remained lower for the Tier 3 model but the difference in Rvalue with the CVB model declined from 0.22 to 0.19 to 0.14, respectively. The standard deviation of predicted values was smaller than that of observed values leading to υ values (scale TABLE 5 | Relationship between observed apparent fecal N digestibility (%) and predicted apparent fecal N digestibility (%) with the Tier 3 model and the CVB model (values and calculations based on studies with wethers fed on maintenance level) for the complete dataset, the complete dataset excluding rolled or cracked high moisture maize silage, and the Dutch dataset.


<sup>a</sup>P-value indicates significance of the estimate being different from zero at level of P < 0.001, indicated by \* .

<sup>b</sup>RMSPE as root of mean square prediction error (MSPE) expressed as a percentage of the observed mean, and in parentheses as % units of apparent fecal N digestibility. <sup>c</sup>Error due to bias, as a percent of total MSPE.

<sup>d</sup>Error due to regression, as a percent of total MSPE.

<sup>e</sup>Error due to disturbance, as a percent of total MSPE.

<sup>f</sup> Concordance correlation coefficient.

<sup>g</sup>Bias correction factor.

<sup>h</sup>Scale shift.

<sup>i</sup>Location shift.

<sup>j</sup>Pearson correlation coefficient.

<sup>k</sup>Observed mean of apparent fecal N digestibility (% of N intake).

shift) higher than 1, more so for the CVB model than for the Tier 3 model however (**Table 5**). Both the CVB model and the Tier 3 model over-predicted apparent fecal N digestibility as indicated by the negative µ values (**Table 5**). Absolute values of µ were 2.4, 3.8, and 8.8 times as large for the CVB model compared to the Tier 3 model with the complete dataset, the complete dataset excluding rolled or cracked products, and the Dutch dataset, respectively. There was essentially (only) a small underprediction of apparent fecal N digestibility with the Tier 3 model (**Tables 4**, **5**).

# Prediction of Apparent Fecal N Digestibility in Inventory

**Figure 4** demonstrates the consequences on predicted apparent fecal N digestibility in the ammonia inventory methodology with the CVB model or the alternative the Tier 3 model (Vonk et al., 2016). The average of predictions by the CVB model for the period of 1990 till 2014 was 5.9% units of digestibility higher than the average of predictions by the Tier 3 model from 1990 till 2016. The annual predictions by the CVB model were 5.6 (±0.93) % units of digestibility higher compared to the Tier 3 model. The results further demonstrate a continuous decline in predicted N digestibility since 1990 following the trend in the activity data of a declining dietary N content (data not shown here; Vonk et al., 2016). The predicted decline in the apparent fecal N digestibility from 1990 till 2010 was about 6.5% units of digestibility. Since 2010 the decline leveled off (despite some remaining variation predicted by the Tier 3 model) together with activity data indicating a rather constant dietary N content (data not shown here; Vonk et al., 2016).

# DISCUSSION

In cattle, utilization of dietary N is relatively inefficient with some 50–85% of consumed N excreted in feces and urine (Moore et al., 2014). The amount of N excreted is related to several factors, with dietary protein content and its apparent digestibility being major determinants. Decreasing dietary protein content is among the most effective strategies to reduce ammonia emissions from dairy manure (Agle et al., 2010b). In a metaanalysis, Bougouin et al. (2016) identified DM intake, milk production, and dietary protein content being key explanatory variables in predicting ammonia emission from dairy housing. The estimation of urine N, or TAN, excretion requires knowledge of dietary N consumption, apparent fecal N digestibility and the amount of N retained in animal products. Activity data in the inventory methodology in the Netherlands already deliver insight into N consumption and N retained by cows in milk, growth and offspring. However, these data do not indicate apparent digestibility of dietary N which hence needs to be predicted.

# Comparison of Tier 3 Model and CVB Model

The average of predicted and observed values for apparent fecal N digestibility were closer for the Tier 3 model compared to the CVB model with the difference becoming <1% unit of digestibility (**Table 4**) for the Dutch dataset. Also the better correspondence between predicted and observed N digestibility values when accounting for study effect (**Table 5**) indicates an improved applicability of the Tier 3 model on account of representation of fermentative and digestive mechanisms. The statistical results consistently indicate a better prediction performance by the Tier 3 model although it may remain hard to distinguish the better capture by the model of the within trial treatment differences in **Figures 2**, **3**. This was also demonstrated by a much smaller RMSPE value and more than 77% of error attributed to disturbance (ED) instead of bias (ECT) and regression (ER) (**Table 5**). Furthermore, CCC analysis indicated that the CVB model was less capable than the Tier 3 model to predict apparent fecal N digestibility for the Dutch dataset in particular. With a lower Cb value the CVB model appeared always less accurate (**Table 5**), although demonstrating a higher R-value indicating a better correlation between predicted and observed values (measure for precision).

The complete dataset was restricted, first by exclusion of rolled or cracked products because estimates of in situ degradation characteristics were highly uncertain and not available in the Dutch Feed Table (CVB, 2011), and second by selecting the Dutch dataset with studies conducted in the Netherlands only. Accuracy of the Tier 3 model was high for the Dutch dataset with a Cb value of 0.97 (**Table 5**), and the majority of observations well predicted. Three observations of exceptionally small values of apparent fecal N digestibility by Warner et al. (2013b) could not be reproduced accurately by the Tier 3 model (observations below 60% and prediction above 70%; **Figure 3**). These observation were obtained for three out of six maize silage treatments used in that particular experiment, with a CP content in dietary DM of 18% and maize silage a third of dietary DM with all 6 treatments. Due to the fact that the Tier 3 model received the same in situ degradation characteristics from the Dutch Feed Table (CVB, 2011) as an input for these six maize silages, it is no surprise that the model could not separate out the two groups of observations.

Both the CVB model and the Tier 3 model predicted less variation than observed which is clearly demonstrated by υ values >1, in particular for the CVB model, and more so for the complete dataset than for the Dutch dataset (**Table 5**). Ellis et al. (2010) compared the RMSPE and CCC statistics in an evaluation study of enteric methane prediction equations that are adopted in farm systems modeling. They demonstrated and discussed that when models are unable to describe adequate amounts of the observed variation, CCC analysis is likely the preferred evaluation tool to be used. When mainly focussing on the results of CCC statistics in the present study, the conclusion remain however that the Tier 3 model outperforms the CVB model based on the results obtained for the Dutch dataset to which the model inputs derived from Dutch Feed Table (CVB, 2011) will comply most.

The results depicted in **Figures 2**, **3** show a large positive bias in predicted apparent fecal N digestibility for the CVB model. This is demonstrated by the stronger negative value of the µ parameter from CCC analysis which indicates a stronger overprediction by the CVB model. Over-prediction is clearly far less with the Tier 3 model and almost absent with the Dutch dataset (**Table 5**). The main reason for the over-prediction with the CVB model is likely that it bases its prediction on digestion data retrieved from wethers instead of dairy cattle. The latter are reported to have a lower apparent fecal N digestibility due to a different contribution of endogenous and microbial N sources to fecal N (Schiemann et al., 1971; Van Es, 1978). Results from Soto-Navarro et al. (2014) suggest that also digestibility data for steers might not be representative for dairy cattle. Apparent fecal N digestibility was reported to be equal or higher than in sheep (2.6, 8.6, and 51.5% units of digestibility for alfalfa, high-quality grass hay and low-quality grass hay, respectively). Therefore, any empirical database to be applied to dairy cows should best be obtained from observations on dairy cows under representative nutritional conditions. Furthermore, the relatively small bias (small ECT values and high Cb values; **Table 5**) with the Tier 3 model for all three datasets, suggests that the Tier 3 model performance in predicting the average level of apparent fecal N digestibility is satisfactory. Accurate prediction of such an average level is of particular importance for the Tier 3 model to be used for the national inventory purposes, as these are based on calculations with averaged and consolidated data at the regional or national level. It is noted that the consistent bias obtained with the CVB model (high ECT values and low Cb values; **Table 5**) could be removed by applying a fixed correction factor based on the present findings. It remains to be demonstrated that such a correction factor holds when evaluating the CVB model against another dataset which is independent from the results obtained in the present study. Moreover, the results from mixed model analysis (**Table 5**), in which bias for each study is accounted for by including a random effect of study, show that the CVB model suffers more from the regression slope differing from the optimal value of 1 than the Tier 3 model. This holds in particular again for the Dutch dataset to which the model inputs used comply most (**Table 5**).

### Predictive Performance of the Tier 3 Model

Standard dietary characteristics were obtained from the Dutch Feed Table (CVB, 2011) and served to calculate model inputs for the Tier 3 model. However, these are likely inaccurate for the wide range of roughage types and feed qualities encountered under non-Dutch conditions. This probably contributed to the poorer prediction of apparent fecal N digestibility for the non-Dutch trials, whereas statistical analysis the Dutch dataset was more satisfactory. Even for these Dutch trials, however, the dietary ingredients and roughages must have differed strongly from the standard in situ degradation characteristics that are listed in the Dutch Feed Table (CVB, 2011; Van Duinkerken et al., 2011). Allowing for variation in these in situ degradation characteristics and adopting more realistic estimates reflecting the treatments reported would probably have increased the capacity of the model to capture observed variation in apparent fecal N digestibility. For example, the model cannot be expected to accurately predict the consequence of variation in differences in roughage quality when standardized in situ degradation characteristics of roughages are used as an input. In the present study we used such standardized input from the Dutch Feed Table (CVB, 2011).

Hence, assumptions on in situ degradation characteristics probably have been too generic to capture the variation in apparent fecal N digestibility that was observed in the various N balance trials selected from literature. Differences and inaccuracies in experimental set-ups and measuring techniques have contributed to this variation. In many studies N excreted with urine was calculated by difference method (**Tables 1**, **2**), whereas in the others a full N balance was determined (including measurement of N excreted with urine). This difference between in studies in quantifying urine N excretion will have contributed strongly to the variation not captured by both models. Nevertheless, the most likely explanation of the lowest prediction capacity for the most complete dataset (**Figure 2**) remains the too narrow range or the bias in values retrieved from Dutch Feed Table (CVB, 2011), not being representative for the range of conditions met in international trials. Improving prediction performance of the Tier 3 model for non-Dutch conditions would require such model inputs to be derived from local, non-Dutch conditions as well. Such an approach was followed in studies that aimed to predict enteric methane emission in dairy cattle in various regions in the US (Kebreab et al., 2008) and digestibility (including apparent fecal N digestibility) of various diet types and production conditions (Hanigan et al., 2013). In these studies, similar dynamic, mechanistic models were used, requiring inputs similar to those of the Tier 3 model used the present study.

### N Digestion Models in Ammonia Inventory

There is an urgent need to account for the effect of the ammonia mitigation measures taken in livestock operations. Both farm accounting tools and Life Cycle Analysis methodology would benefit from a more accurate and more case-specific quantification of sources of emissions (Cederberg et al., 2013), such as the amount of volatile N excreted as a source of ammonia. Both accuracy and precision is needed to identify the level and size of trade-offs between various sources and types of emissions. The highly volatile urine N as a source of ammonia is the ingested amount of digestible N by cattle which is not retained in animal product. This becomes apparent in various literature surveys (e.g., Kebreab et al., 2002). In a companion study, Dijkstra et al. (2018) explored how various dietary measures to mitigate N excretion affect the composition and characteristics of C and N containing fractions in urine and feces. The quantitative terms used to characterize manure correspond with the fermentation and digestibility concepts applied in ruminant feed evaluation. Despite the large impact of dietary N mitigation measures on the proportion of urine N in total N excreted, and on the C:N ratio of manure, inventory methodology seldom represents the variation in these proportions to calculate ammonia emissions (EEA, 2016; Nemecek and Ledgard, 2016). However, under various production conditions the proportion of urine N as well as the volume and frequency of urine excretion may impact immediate ammonia and nitrous oxide emissions from urine N (Ledgard et al., 2015; Selbie et al., 2015). Also with regard to ammonia emission from stored manure, complex mechanisms are responsible for variation in emission rates which includes the amount of urine N and the volume of urine excreted in housing (Sommer et al., 2006).

Despite the complexity of the mechanisms underlying the variation in these emissions, rather constant emission factors are often applied in inventory which in principle lack a relationship with nutritional measures and details on excreta composition, N excretion rate and excreted volumes. All models represent mass flows on a dairy farm. The more detailed ammonia emission models such as the dairy farming systems model developed by Rotz et al. (2014) represent details on the effect of type and fate of excreted N and of excreta volumes on ammonia emission. Excreted urine and fecal N are calculated by functions of animal size, feed intake and protein intake, and milk production, but not protein digestibility characteristics. More recently, Chai et al. (2016) added such detail in a model used for an ammonia inventory on Ontario dairy farms in four ecoregions. The Canadian ammonia emission inventory and survey model was refined by introducing a representation of the effect of dietary mitigation measures. They derived a linear equation to estimate the fraction of urine N, or TAN, in total N excretion from dietary CP content. The range of dietary CP content used (123, 153, and 164 g CP/kg DM with TAN proportion in manure N of 0.42, 0.50, and 0.56, respectively) covers the lower half of the range in the database used in the present study (**Tables 1**, **2**). The relationship between dietary CP content and TAN excretion may be considered intrinsically non-linear however. This nonlinear effect on TAN proportion may be covered by the approach of Velthof et al. (2012), who use a method adopting estimates of apparent fecal N digestibility retrieved from the Dutch Feed Table (CVB, 2011), evaluated in the present study as the CVB model. This method attributes all digested N not retained in animal products to TAN and, therefore, with further increase of dietary CP content, the estimated TAN proportion in total excreted N increases non-linearly. Notwithstanding the fact that current methodologies may capture the non-linear increase of proportion of TAN with total N excretion, it is of importance that variation in apparent fecal N digestibility on proportion of TAN is captured as well. The present study focussed on an independent evaluation and improvement of the CVB model as the method use by Velthof et al. (2012). The Tier 3 model was evaluated as well, as an alternative candidate model which takes details on fermentative and digestive aspects into account. Based on the promising findings in the present study, and the fact that this model is already in use in the greenhouse gas inventory for estimating enteric methane in dairy cattle, the Tier 3 model has replaced the CVB model in the ammonia inventory in the Netherlands since 2015 (**Figure 4**; Vonk et al., 2016). The studies of Dijkstra et al. (2013, 2018) demonstrate that further detailing of the composition of urine and feces (and manure) is possible if needed for the purpose of a more detailed inventory.

### CONCLUSIONS

Upon using the CVB model to predict apparent fecal N digestibility in dairy cows in the ammonia emissions inventory in the Netherlands, a large systematic bias of 6–7% units of digestibility occurs. This bias can almost entirely be prevented by the use of the Tier 3 model which is extant methodology to estimate enteric methane in dairy cattle in the greenhouse inventory in the Netherlands. The more mechanistic representation of fermentation and digestion in the

### REFERENCES


gastro-intestinal tract of dairy cows allows a more accurate and acceptable precision of predicted apparent fecal N digestibility under Dutch feeding conditions. Model performance was less satisfactory on the complete dataset, likely because of less valid standardized inputs to the model (in particular ruminal in situ degradation characteristics) when derived from distinct world regions. Satisfactory prediction of the overall average apparent fecal N digestibility demonstrates applicability of the Tier 3 model for the calculation of TAN excretion in the ammonia emissions inventory.

### AUTHOR CONTRIBUTIONS

AB, LŠ, and JD developed the concepts. AB developed the equations. WS and AB performed the simulations and analyzed the data. AB and WS wrote the original draft of the manuscript and LŠ and JD contributed to discussion and revision of this manuscript.

### ACKNOWLEDGMENTS

The funding of this research by the Ministry of Agriculture, Nature and Food Quality (project BO-20-004-111) is gratefully acknowledged.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Bannink, Spek, Dijkstra and Šebek. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Modeling the Effect of Nutritional Strategies for Dairy Cows on the Composition of Excreta Nitrogen

### Edited by:

Tom Misselbrook, Rothamsted Research (BBSRC), United Kingdom

### Reviewed by:

Maria Alejandra Herrero, Universidad de Buenos Aires, Argentina David R. Yanez-Ruiz, Consejo Superior de Investigaciones Científicas (CSIC), Spain

> \*Correspondence: Jan Dijkstra jan.dijkstra@wur.nl

### †Present Address:

Pieter M. Bosma, Dairy farm Bosma, Winsum, Netherlands Joan W. Reijs, Wageningen Economic Research, Wageningen University & Research, Wageningen, Netherlands

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 16 February 2018 Accepted: 10 September 2018 Published: 10 October 2018

### Citation:

Dijkstra J, Bannink A, Bosma PM, Lantinga EA and Reijs JW (2018) Modeling the Effect of Nutritional Strategies for Dairy Cows on the Composition of Excreta Nitrogen. Front. Sustain. Food Syst. 2:63. doi: 10.3389/fsufs.2018.00063 Jan Dijkstra<sup>1</sup> \*, André Bannink <sup>2</sup> , Pieter M. Bosma1†, Egbert A. Lantinga<sup>3</sup> and Joan W. Reijs 4†

<sup>1</sup> Animal Nutrition Group, Wageningen University & Research, Wageningen, Netherlands, <sup>2</sup> Wageningen Livestock Research, Wageningen University & Research, Wageningen, Netherlands, <sup>3</sup> Farming Systems Ecology, Wageningen University & Research, Wageningen, Netherlands, <sup>4</sup> Department of Animal Sciences, Wageningen University & Research, Wageningen, Netherlands

For an integrated evaluation of the effect of nutritional strategies on the utilization and losses of N at dairy farms, reliable estimates of excreta production and composition are indispensable. An extant, dynamic, mechanistic model of rumen functioning was extended with static equations that describe intestinal digestion to simulate the composition of dairy cow feces and urine as a function of diet composition. The extended model predicts organic matter (OM), carbon (C), and nitrogen (N) output of both feces and urine, classified in different components. Total N excretion was partitioned in three fractions based on the C:N ratio of individual components representing their availability of N following manure application to crops, viz. N<sup>M</sup> (immediately available), N<sup>E</sup> (easily decomposable), and N<sup>R</sup> (resistant). Forty nutritional strategies for stall-fed dairy cows, covering diets with a wide range in protein content and OM digestibility, were evaluated. The simulated ranges in fecal and urinary composition were largely in line with values reported in literature. Diet intake and composition had a substantial effect on simulated total N excretion and excreta composition, mainly because of differences in the level of N<sup>M</sup> excretion and the C:N ratio of the N<sup>R</sup> fraction. Furthermore, it was shown that the type of OM excreted varies considerably between different diets. A simplified simulation of degradation processes during the first 4 months of excreta storage produced average values and ranges of slurry characteristics that were in line with values reported in literature. The simulated variation in slurry characteristics suggested a strong variability in ammonia N losses from the slurry pit and a moderate variability in plant availability of slurry N. Further efforts are required to integrate effects of manure storage conditions on the storage processes. In conclusion, the model can be a tool to predict fecal and urinary composition of cattle, and ultimately to improve the utilization of N from field applied manure as well as to evaluate the effects of different nutritional strategies on the whole-farm N balance.

Keywords: models, dairy cattle, feces, urine, diet composition, manure composition

# INTRODUCTION

Dairy production contributes to environmental pollution from fecal and urinary N as ammonia and nitrous oxides in air and as nitrate, ammonium, and organic N in ground and surface water. Nutrition management is an important tool to reduce this environmental pollution in forage based cattle systems (Misselbrook et al., 2013). The total amount of N excreted in manure can be significantly reduced by lowering the dietary protein content (e.g., Kebreab et al., 2002; Powell and Rotz, 2015). As excessive feed N is mainly excreted with urine, a reduction of the dietary protein content will generally result in a more than proportional reduction of the urinary N excretion (Dijkstra et al., 2013; Powell and Rotz, 2015). Urinary N is more susceptible to losses than fecal N (Selbie et al., 2015). Reductions in dietary protein thus result in significant reductions of gaseous N emissions (e.g., Külling et al., 2001; Misselbrook et al., 2005; Bougouin et al., 2016). However, N utilization and excretion is not exclusively determined by the level of protein in the diet. For example, the output of N in milk of dairy cattle fed diets with similar protein levels depends on dietary carbohydrate composition, with decreased milk N output of high fiber compared with high starch diets (diets iso-energetic) (Cantalapiedra-Hijar et al., 2014). Thus, to minimize N excretion and maximize N utilization at the cow level, a proper balance of energy and N supply to the rumen (Dijkstra et al., 1998) and to splanchnic tissues and mammary gland (Lapierre et al., 2010) is required. Indeed, Kebreab et al. (2010) performed multivariate analysis on fecal, urinary, and milk N excretion in dairy cattle, and reported that addition of diet metabolizability (the concentration of metabolizable energy relative to gross energy of the diet) as a covariate with N intake improved the predictions of N excretion.

Several nutritional-related strategies that may improve N utilization and reduce manure N excretion in dairy cattle have been proposed. These strategies include reduction of the N fertilization level (e.g., Peyraud and Astigarraga, 1998; Warner et al., 2016), later cut of grass (e.g., Brask et al., 2013), using low protein, high energy feeds like maize silage (e.g., Van Gastelen et al., 2015) and adjustment of concentrate composition (e.g., Børsting et al., 2003). However, nutrition not only affects the utilization of N by the cow, but also the composition of the excreta and therefore it interacts with the major part of the processes at the farm level where N is converted and lost. Changes in diet composition affect utilization of N from soilapplied dairy cow slurry (e.g., Sørensen et al., 2003; Reijs et al., 2007; Jost et al., 2013). For this reason, the evaluation of nutritional strategies should go beyond the cow level. For an integrated and profound evaluation of the effect of nutritional strategies on N utilization at farm level, qualification and quantification of excreta composition is a crucial step. Such an evaluation requires a realistic and detailed representation of the cow's complex digestive processes. Several empirical models have been developed to predict output and composition of feces and urine (e.g., Reed et al., 2014; Appuhamy et al., 2018), but such models do not allow for detailed manure characterization and do not reflect the availability of N from urinary and fecal components for subsequent losses or use by plants.

The main objective of this study is to present a model that is capable of evaluating the impact of nutritional strategies on N utilization at the farm level, based on understanding of the complex digestion processes occurring at the cow level. For this purpose, an existing dynamic, mechanistic model of rumen function and subsequent nutrient availability (Dijkstra et al., 1992, 1996) was extended with static equations that describe intestinal digestion. The extended model predicts organic matter (OM), carbon (C), and N output in different fecal and urinary components as a function of diet composition. To illustrate the applicability of the model, excreta composition was simulated for 40 nutritional strategies of stall-fed dairy cows in grass silage based systems, covering a wide range in OM digestibility and dietary protein content. In addition, degradation processes during the first 4 months of excreta storage were simulated in a simplified way to quantify the effect of the nutritional strategies on the composition of field-applied slurry. Based on the simulations, the applicability of the model, the potential variation in excreta composition and its consequences for the composition, utilization, and losses of slurry N during storage are evaluated.

# MATERIALS AND METHODS

### General Structure of the Model

A schematic representation of the model is shown in **Figure 1** and the principal symbols used are listed in **Table 1**. The dynamic and mechanistic model of rumen microbial fermentation processes of Dijkstra et al. (1992) was used to predict the outflow of undigested feed and microbial material from the rumen to the intestines (g day−<sup>1</sup> ) as a function of the chemical composition and rumen degradation characteristics (based on in situ nylon bag incubations) of ingested feedstuffs and of the microbial activity. To obtain quantitative data on fecal composition, the rumen model was expanded with equations that describe the digestion of these rumen outflow components in the small and large intestine, described in detail in a subsequent section.

In the model, fecal excretion consists of 10 different components that are aggregated into four different categories (**Figure 1**), viz. fecal endogenous components (FEC), fecal microbial components (FMC), fecal feed fiber components (FFFC) and fecal other feed components (FOFC). The C and N contents for the different components adopted are given in **Table 1**. A fixed milk composition of 40 g kg−<sup>1</sup> fat, 33 g kg−<sup>1</sup> protein, and 46 g kg−<sup>1</sup> lactose was presumed. Potential milk productions based on total absorbed energy and on available nutrients (lipogenic, glucogenic, and aminogenic) were simulated as described in detail by Dijkstra et al. (1996). The lowest of these four values was taken as the actual simulated milk production. Excretion of urinary N (g day−<sup>1</sup> ) was calculated by assuming zero N retention in the body according to:

$$N\_{urine} = N\_{feed} - N\_{milk} - N\_{fines} \tag{1}$$

If required, for growing cows or cows in late lactation, a positive N retention may be adopted. In line with Bussink and Oenema

(1998), Nurine was divided into urea-like urinary components (UUC) and non-urea-like components (UNUC, **Figure 1**). Urea N was calculated as the difference of Nurine minus the sum of N in other urinary constituents described in another section.

### Intestinal Digestion and Fecal Excretion

Rumen undegradable fiber (Fu) and rumen undegradable protein (Pu) were assumed to be also indigestible in the intestines and completely excreted with the feces (Tamminga et al., 1994). Rumen degradable fiber (Fd) not digested in the rumen was assumed to be indigestible in the small intestine (SI). In the large intestine (LI) the digestion coefficient of Fd was based on the retention time of the material according to:

$$\text{digF}d\_{LI} = \text{kdF}d\_{runmen}/(\text{kdF}d\_{runmen} + k\text{p}\_{LI})\tag{2}$$

where digFdLI is the fraction of Fd outflow from the SI digested in the LI, kdFdrumen is the fractional degradation rate of ingested Fd in the rumen (% h−<sup>1</sup> , **Table 2**) and kpLI (% h−<sup>1</sup> ) is the fractional passage rate in the LI, estimated according to Mills et al. (2001):

$$k p\_{LI} = 1/(-0.2 \times DMI + 13) \times 100\% \tag{3}$$

where DMI is Dry Matter Intake in kg day−<sup>1</sup> . The fraction of rumen degradable starch (Sd) washed out from the rumen and digested in the SI (digSdSI) was related to the fraction of starch escaping rumen fermentation, according to Nocek and Tamminga (1991):

$$\text{digS}d\_{\text{SI}} = -0.728 \times \text{RES} + 0.879 \tag{4}$$

where RES is the total outflow of starch from the rumen, including microbial storage polysaccharides (As), as a fraction of total starch intake. The fraction of starch outflow from the SI and digested in the LI (digSdLI) was estimated according to:

$$\text{digS}d\_{LI} = k d \text{S}d\_{rumen} / (k d \text{S}d\_{rumen} + k p\_{LI}) \tag{5}$$

where kdSdrumen is the fractional degradation rate of ingested Sd (% h−<sup>1</sup> ) in the rumen.

Microbial starch is assumed to be completely digested in the SI. Polysaccharide-free microbial OM in the rumen (RMB) was assumed to consist of protein (61%), nucleic acids (18%), lipid (16%), and cell walls (5%), based on Dijkstra et al. (1992). No distinction was made in the digestibility of N in the different TABLE 1 | Abbreviations used in the model and carbon (C) and nitrogen (N) contents (in dry matter, DM) of components.


<sup>a</sup>RMB equals the sum of amylolytic and fibrolytic polysaccharide-free microbial mass as described by Dijkstra et al. (1992).

<sup>b</sup>Contents of C and N in excreta are calculated based on the proportion of individual components and their respective C and N content.

components of RMB (protein, nucleic acids and cell walls) and the digestion of all N contained in the RMB was set at a constant value of 0.81 (Storm et al., 1983). Storm et al. (1983) observed an OM digestibility in the SI of 0.74 for the total microbial matter, including microbial polysaccharides. As microbial polysaccharides are highly digestible, this digestion coefficient was adjusted to 0.67 for RMB as RMB is defined as polysaccharide-free microbial matter.

Excreted endogenous biomass (EB) was divided into protein (EP; digestive enzymes, desquamated epithelial cells, mucus) and lipids (EL; bile salts). Excretion of EP was based on a net loss of metabolic protein of 50 g kg−<sup>1</sup> ingested indigestible DM (Tamminga et al., 1994). Endogenous lipid excretion was estimated to be 24 g day−<sup>1</sup> (Børsting et al., 1992). The SI digestibility of rumen digestible feed protein (Pd) was set at 0.75 and that of feed, microbial and endogenous lipid at 0.90 (Palmquist et al., 1993). Net lipid digestion in the LI was assumed to be zero, following observations of Drochner and Meyer (1991). The digestibility of feed protein and microbial OM in the LI was assumed to be constant at 10% of the outflow from the SI. The digestibility of endogenous protein was set at a significantly higher value of 40%, assuming that the endogenous protein is more easily fermented as it has not been subject to the digestive processes for the full length of the SI (Van Soest,



<sup>a</sup>HFEC, high fertilized, early cutting stage grass silage; HFLC, high fertilized, late cutting stage grass silage; LFEC, low fertilized, early cutting stage grass silage; LFLC, low fertilized, late cutting stage grass silage; MSIL, maize silage; STR, straw; PBP, pressed beet pulp; POT, potatoes; CONC, concentrate.

<sup>b</sup>FP, Fermentation Products (assumed composition: 60% lactic acid, 30% acetic acid, 5% propionic acid, and 5% butyric acid). Other abbreviations are explained in Table 1. c kd, fractional degradation rate.

<sup>d</sup>NEL, Net Energy Lactation.

<sup>e</sup>DVE, Protein Digested in the Small Intestine; OEB, Degraded Protein Balance in the Rumen, according to Tamminga et al. (1994).

<sup>f</sup> n.a., not available.

1994). The amount of fermentable OM in the LI (FOMLI) in g day−<sup>1</sup> was calculated based on the differences in components flows (Pd, digestible feed protein; EP, endogenous protein; RMB, rumen microbial biomass; Fd, rumen degradable fiber; Sd, rumen degradable starch) into the duodenum (duodoutfl) and in manure (manure) according to:

$$\begin{aligned} \text{FOM}\_{LI} &= 0.55 \times \text{(Pd}\_{duodoutfl} - \text{Pd}\_{manure}) + 0.55\\ &\times \text{(EP}\_{duodoutfl} - \text{EP}\_{manure}) + 0.55\\ &\times \text{(RMB}\_{duodoutfl} - \text{RMB}\_{manure})\\ &+ \text{(Fd}\_{duodoutfl} - \text{Fd}\_{manure}) + \text{(Sd}\_{duodoutfl} - \text{Sd}\_{manure}) \end{aligned} \tag{6}$$

The assumption was made that protein provides 0.55 of pyruvate units per mol fermented substrate compared to hexose (Bannink et al., 2006). Production of volatile fatty acids (Va) in the LI was assumed to be 0.70 g per g of FOMLI (DeMeyer and De Graeve, 1991). Assuming that the Va absorption rate (µmol cm<sup>2</sup> min−<sup>1</sup> ) of the LI is similar to that of the rumen (Ding et al., 1998), it was estimated that 75% of the produced Va was absorbed. The production of microbial N in the LI was estimated at 24 g kg−<sup>1</sup> FOM, i.e., equal to that in the rumen (Tamminga et al., 1994). Based on values in **Table 1**, this implies a production of 178 g of large intestinal polysaccharide-free microbial OM per kg FOMLI.

### Urinary N Constituents Other Than Urea

In general, urine of dairy cows contains only traces of free amino acids. Based on data of Bristow et al. (1992) it was assumed that only 2% of the urinary N consists of free amino acids (Aa). Hippuric acid (Hi) in ruminant urine is mainly a derivative of rumen microbial fermentation of phenolic cinnamic acids (Martin, 1982) and it has been shown that its relative contribution to the total N excreted may vary (Dijkstra et al., 2013). However, quantitative data on the effect of diet composition on Hi excretion are scarce and therefore an average contribution of 5% to total urinary N excretion (Bristow et al., 1992) was assumed for Hi. Urinary creatinine (Crn) excretion is a relatively constant function of body weight (BW) and set at 29 mg kg−<sup>1</sup> BW day−<sup>1</sup> (Valadares et al., 1999). Assuming a BW of 625 kg, this corresponds with a Crn-N excretion of 6.5 g day−<sup>1</sup> . Creatine (Cr) N excretion was estimated at 4.8 g day−<sup>1</sup> based on the ratio between Crn and Cr observed by Bristow et al. (1992). Xanthine plus hypoxanthine (Xa) excretion is relatively small and was taken as 0.5% of total urinary N excretion (Bristow et al., 1992). The excretion of purine derivatives (allantoin, xanthine, hypoxanthine and uric acid) has consistently been related to microbial synthesis in the rumen (Valadares et al., 1999). In our model, the relationship reported by Susmel et al. (1993) was used to predict the total excretion of urinary purine derivatives:

$$\text{UPD} = 17.22 + 0.0082 \times \text{RMP} \tag{7}$$

where UPD and RMP are the amounts of excreted urinary purine derivatives and rumen microbial protein outflow, respectively, in mg day−<sup>1</sup> kg−<sup>1</sup> BW0.75. After subtraction of Xa, the remaining UPD was divided into allantoin (Al) and uric acid (Ua), using a ratio of 85:15 (Bristow et al., 1992; Valadares et al., 1999).

### From Excreta to Slurry Composition

The most common system in the Netherlands is to store feces and urine in a mixed slurry system for a period of ∼4 months in the slurry pit. During this storage period the excreta and added bedding material with a relatively high C:N ratio are subject to both anaerobic and aerobic fermentation processes, affecting their composition. Manure OM is degraded (Whitehead and Raistick, 1993), manure C is lost (Sørensen, 1998), urea-N and


TABLE 3 | Description of selected nutritional strategiesa,b,c and simulated average milk production and feed nitrogen (N) conversion.

<sup>a</sup>All combinations of the described strategies were simulated (n = 40).

<sup>b</sup>Dry Matter Intake (DMI) of the complete rations was estimated using the prediction model for lactating Holstein cows of Zom et al. (2012).

<sup>c</sup>Concentrate composition was assumed to be constant.

<sup>d</sup>Based on differences in fertilization level (F) and cutting moment (C) of grass silage: HFEC, high fertilization and early cut; HFLC, high fertilization and late cut; LFEC, low fertilization and early cut; LFLC, low fertilization and late cut.

<sup>e</sup>Characteristics of the feedstuffs are given in Table 2.

<sup>f</sup> FPCM (kg/d), Fat and Protein Corrected Milk (assumed composition: 40 g kg−<sup>1</sup> fat, 33 g kg−<sup>1</sup> protein, 46 g kg−<sup>1</sup> lactose).

<sup>g</sup>Feed N conversion (%) = milk N output / feed N input × 100.

part of the organic N in manure are transformed into NH<sup>+</sup> 4 - N (Whitehead and Raistick, 1993; Sørensen et al., 2003) and N losses occur through gaseous emissions (Misselbrook et al., 2005). To quantify the effect of these processes on the final slurry composition after storage, it was assumed that all N in UUC was converted into NH<sup>+</sup> 4 -N. Based on results of Sørensen et al. (2003), the transformation of the (other) organic N into NH<sup>+</sup> 4 -N was assumed to be negatively related to the fiber content of the diet according to:

$$MIN\_{organicN} = 50 - 0.075 \times NDF\_{diet} \text{(g } \cdot \text{kg}^{-1} \text{DM)} \tag{8}$$

where MINorganicN is the fraction (%) of organic N (total excreted N–UUC N) that is mineralized and transformed into NH<sup>+</sup> 4 -N during storage.

As no quantitative data were found to differentiate C loss for diet or slurry characteristics, the C loss during storage was set at 13% of total C as found by Sørensen (1998) after 20 weeks at a temperature of 15◦C; a change in ambient temperature would change this fractional loss. The fraction of slurry OM loss was assumed to be equal to the C loss, as Kirchmann and Witter (1992) found no marked difference between OM and C loss. From the results of Külling et al. (2001), it was concluded that nitrous oxide emissions are negligible in slurry based systems compared to N losses as ammonia. Total ammonia N losses in the storage period include both emissions from the stable floor and the storage pit, and were estimated to be 22% of the urea-N (Van Duinkerken et al., 2003). The use of bedding material was set at 1 kg of sawdust (C:N ratio of 450) per cow per day.

### Nutritional Strategies

Forty different nutritional strategies, all based on stall-fed situations, were explored with the model. The various strategies included several types of grass silage (high or low fertlization level of grass and early or late cutting of grass before ensiling), type of grass silage replacement (replacement with maize silage or various by-products), and the level of concentrate feeding (**Table 3**). High (HF) or low (LF) level of inorganic N fertilization, combined with an early (EC) or late (LC) cutting stage were considered to give four different spring cut silages, viz. HFEC, HFLC, LFEC, LFLC. The assumed chemical composition and rumen degradation characteristics of these silages are shown in **Table 2**. It was assumed that grass was fertilized before the first cut with dairy slurry (25 ton ha−<sup>1</sup> ) in combination with a high (100 kg N ha−<sup>1</sup> ) and a low level (50 kg N ha−<sup>1</sup> ) of inorganic fertilizer. A reduction of the fertilization level was expected to result in a decrease of the crude protein level (Heeren et al., 2014) and an increase in the content of water soluble carbohydrates (Wr) (Peyraud and Astigarraga, 1998). Later cutting (from 3,000 to 4,500 kg DM ha−<sup>1</sup> ) was expected to increase the neutral detergent fiber (NDF) content and to decrease the crude ash content (Bosch et al., 1992; Heeren et al., 2014). Rumen protein degradation characteristics were estimated by regression formulae from Tamminga et al. (1991). For the EC silages, the rumen undegradable NDF fraction (Fu) was estimated at 10% of total NDF (Bosch et al., 1992; Bruinenberg et al., 2004). Bosch et al. (1992) and Heeren et al. (2014) showed that with increasing NDF contents, Fu (both absolute and as a fraction of total NDF) increases and the fractional degradation rate of TABLE 4 | Mean values and ranges of intake, diet composition, simulated digestion coefficients, simulated milk production and simulated feed nitrogen (N) conversion of 40 nutritional strategies<sup>a</sup> for dairy cows.


<sup>a</sup>Nutritional strategies and feed characteristics are described in Tables 2, 3.

<sup>b</sup>OEB, Degraded Protein Balance in the Rumen according to Tamminga et al. (1994). <sup>c</sup>FPCM, Fat and Protein Corrected Milk (assumed composition: 40 g kg−<sup>1</sup> fat, 33 g kg−<sup>1</sup> protein, 46 g kg−<sup>1</sup> lactose).

rumen-degradable fiber (kdFd) decreases. The Fu fraction of the LC silages was set at 25% of total NDF, being the average value of two silages with similar NDF contents used by Bosch et al. (1992) and Bruinenberg et al. (2004). The fractional degradation rate of the LC silages was set at 65% of that of the EC silages based on the observed differences in kdFd between grass silages with high and low NDF contents in both experiments (Bosch et al., 1992; Bruinenberg et al., 2004).

The composition of concentrate feed was based on an arbitrarily chosen widely used concentrate feed produced by a Dutch company. Chemical composition of the concentrate ingredients, straw (STR) and industrial by-products (pressed beet pulp, PBP; potatoes, POT) were based on Dutch standards (Anonymous, 2011). Chemical composition of the maize silage (MSIL) was taken as the Dutch average for 2004-2009. Rumen degradation characteristics were estimated from reports on insacco experiments for concentrate ingredients (Tamminga et al., 1990; Van Straalen, 1995), MSIL (Klop and De Visser, 1994), STR (Oosting, 1993; Sinclair et al., 1993), PBP (Tamminga et al., 1990; De Visser et al., 1991; DePeters et al., 1997), and POT (Van Straalen, 1995; Offner et al., 2003) (**Table 2**).

The required input for the rumen fermentation model was completed as described below. Dry matter intake of the complete rations was estimated using the prediction model for lactating Holstein cows of Zom et al. (2012) for a reference cow of 625 kg BW, third parity, mid-lactation (180 d in milk), and 90 d pregnant. Rumen fractional passage rates for fluid (kpf) and solid particles (kps) in % h−<sup>1</sup> , were calculated according to Van Straalen (1995):

$$kpf = -3.40 + 1.224 \times DMI - 0.030 \times DMI^2 + 5.93 \times pR \text{ (9)}$$

$$kps = \text{pR} \times \text{(1.74 + 0.15 \times DMI)} + (1 - \text{pR}) \times \text{(10.1 - 0.96)}$$

$$\times DMI + 0.037 \times DMI^2) \tag{10}$$

where DMI is dry matter intake in kg day−<sup>1</sup> and pR is the fraction of roughage in the diet. Rumen digesta volume (RV; liter) was estimated as:

$$RV = 47.86 + 1.759 \times DM \text{ (adapted from Mills et al., 2001)} \tag{11}$$

The average rumen pH (pH) was set at 6.1 for diets with 100% LC grass silages and a high concentrate level based on Abrahamse et al. (2008). For the other strategies the following adjustments for pH were made based on amount and potential degradability of carbohydrate components: low concentrate level: +0.3, EC silages: −0.1, MSIL: −0.05, STR: +0.1, PBP: −0.1, POT: −0.05. The minimum daily pH (PM) and the time below a critical pH for reduced fiber digestion (TF in h/24 h) were calculated as:

$$PM = \text{pH} - \text{(pH} \times \text{0.05) (Mills et al., 2001)} \tag{12}$$

$$TF = \{-10.59 \times \text{pH}\} + 76.82 \{\text{Erdman, 1998}\}, \text{with } TF = 0 \text{ if } \text{pH} > 7.2 \text{ and } TF = 24 \text{ if } \text{pH} < 5.0. \tag{13}$$

### RESULTS

### Ranges in Simulated Excreta Composition

Simulated intake, dietary characteristics, and digestion coefficients showed large variation between nutritional strategies (**Table 4**). Obviously, this variation resulted in differences between nutritional strategies in energy and nutrient availability for milk production. Simulated FPCM production ranged from 19.1 to 33.8 kg day−<sup>1</sup> , whereas simulated total excretion of OM varied between 3.7 and 6.3 kg day−<sup>1</sup> (**Table 5**), because of a range in apparent OM digestibility from 70 to 82% and range in DMI from16.0 to 22.4 kg/d (**Table 4**).

Simulated fecal and urinary OM excretion showed considerable variation (**Table 5**). Total N excretion ranged from 211 to 558 g N day−<sup>1</sup> . The simulated C:N ratio of the total excreta was highly variable (3.4–10.6). Simulated fecal N excretion was relatively constant (128–177 g N day−<sup>1</sup> ) and the fecal C:N ratio was quite variable (9.6–16.8), while urinary N excretion showed a large variation (81–388 g N day−<sup>1</sup> ) and urinary C:N ratio was almost constant. The major part (on average 61.4%) of the OM in manure was excreted as FFFC, while the largest part (on average 48.2%) of the N excretion was covered by UUC. The undigested feed components (FFFC


TABLE 5 | Mean values and ranges of simulated fecal and urinary organic matter (OM) and nitrogen (N) excretion, distribution of excretion between different components<sup>a</sup> and carbon to nitrogen (C:N) ratio of components for 40 nutritional strategies<sup>b</sup> for dairy cows.

<sup>a</sup>UUC, Urinary Urea-like Components; UNUC, Urinary Non-Urea-like Components; FEC, Fecal Endogenous Components; FMC, Fecal Microbial Components; FFFC, Fecal Feed Fiber Components; FOFC, Fecal Other Feed Components

<sup>b</sup>Nutritional strategies and feed characteristics are described in Tables 2, 3.

TABLE 6 | Proportional composition (%) of organic matter (OM)<sup>a</sup> and nitrogen (N)<sup>b</sup> in dairy cow excreta after simulation of 40 nutritional strategies<sup>c</sup> .


<sup>a</sup>OMNF , Non-Fibrous Organic Matter; OMRDF , Rumen Potential Digestible Fiber; OMRIF , Rumen Indigestible Fiber.

<sup>b</sup>NM, Immediately available Nitrogen (C:N ratio <1); N<sup>E</sup> , Easily decomposable Nitrogen (C:N ratio 2–6); NR, Resistant Nitrogen (C:N ratio >10).

<sup>c</sup>Nutritional strategies and feed characteristics are described in Tables 2, 3.

<sup>d</sup>UUC, Urinary Urea-like Components; UNUC, Urinary Non-Urea-like Components; FEC, Fecal Endogenous Components; FMC, Fecal Microbial Components; FFFC, Fecal Feed Fiber Components; FOFC, Fecal Other Feed Components; Fd, rumen degradable neutral detergent fiber; Fu, rumen undegradable neutral detergent fiber; Pu, rumen undegradable protein.

and FOFC) showed a considerable range in C:N ratio, whereas the C:N ratio of the other components (UUC, UNUC, FEC, and FMC) was much less variable. The distribution of fecal N excretion over the different components showed only a small variation: FFFC (28 ± 2%), FOFC (6 ± 2%), FMC (50 ± 2%), and FEC (16 ± 1%). The simulated fraction of urinary N excreted with UNUC ranged between 10 and 22%.

Based on their C:N ratio, the excreta components were divided into three different fractions representing their availability of N following manure application to crops (NM, NE, and NR; Sluijsmans and Kolenbrander, 1977) (**Table 6**). The immediately available fraction (NM) is represented by UUC with a C:N ratio < 1. On average, 48% of the excreted N was present in this fraction, ranging from 29 to 64%. The easily decomposable fraction (NE) consists of all manure components with a C:N ratio between 2 and 6, being UNUC, FEC and FMC, and covered on average 37% (range 27–49%) of the excreted N. The resistant N fraction (NR) comprises the undigested feed components FOFC and FFFC, with a high but variable C:N ratio (range 12–47). This fraction averaged 15% (range 10–22%) of the total excreted N.

The OM excretion was divided into fiber (FFFC) and non-fiber (OMNF) components (**Table 6**). Within the fiber components, a distinction was made between rumen potentially digestible (OMRDF) and rumen indigestible fiber (OMRIF) as this distinction might reflect differences in the degradability of manure OM during storage and after application to soil. On average, 39% of the OM was excreted with the non-fiber fraction but with a considerable range (28 to 57%). The OMRDF fraction was on average 21% of total OM excretion (range: 10–36%), and the OMRIF fraction was on average 40% of total OM excretion (range: 26–58%).

### Effects of Nutritional Strategies on Excreta Composition and Milk Output

The variation in total N excretion is mainly reflected in the N<sup>M</sup> fraction (**Figure 2**). The simulation results showed a strong decrease of N<sup>M</sup> excretion and an increase in feed N to milk N conversion efficiency when N fertilization is reduced from a high (HF) to a low (LF) level, whereas simulated milk output reduced slightly (**Table 3**). An extended growing period of the silage grass (EC vs. LC) decreased N<sup>M</sup> excretion and improved N conversion efficiency even further (**Figure 2**) but at the expense of a larger reduction in milk output. The inclusion of maize silage in the diet strongly reduced simulated N<sup>M</sup> excretion and had a positive effect on milk output, inducing a large increase in the conversion efficiency of feed N into milk N (**Table 3**). The inclusion of 15% straw in the diet markedly reduced DMI and N intake, resulting in a lower N<sup>M</sup> excretion and a strong reduction of milk output. The inclusion of low protein feeds (PBP, POT) in

FIGURE 2 | Simulated N excretion divided in three different fractions as affected by nutritional strategy, the carbon (C) to nitrogen (N) (C:N) ratio of these fractions and the C:N ratio of the total excreta. The excreta fractions represent NM (immediately available N), NE (easily decomposable N), and NR (resistant N) as described in Table 6. The bars represent average values for all strategies with a given grass silage type (A, n = 10), grass silage replacement (B, n = 8), and concentrate level (C, n = 20), as described in Table 3. HFEC, high fertilized, early cutting stage grass silage; HFLC, high fertilized, late cutting stage grass silage; LFEC, low fertilized, early cutting stage grass silage; LFLC, low fertilized, late cutting stage grass silage; NO, no replacement of grass silage; MSIL, maize silage; STR, straw; PBP, pressed beet pulp; POT, potatoes.

FIGURE 3 | Simulated organic matter (OM) excretion divided in three different fractions as affected by nutritional strategy. The excreta fractions represent OMNF (non-fibrous OM), OMRDF (rumen potentially digestible fiber), and OMRIF (rumen indigestible fiber) as described in Table 6. The bars represent average values for all strategies with a given grass silage type (A, n = 10), grass silage replacement (B, n = 8) and concentrate level (C, n = 20) as described in Table 3. HFEC, high fertilized, early cutting stage grass silage; HFLC, high fertilized, late cutting stage grass silage; LFEC, low fertilized, early cutting stage grass silage; LFLC, low fertilized, late cutting stage grass silage; NO, no replacement of grass silage; MSIL, maize silage; STR, straw; PBP, pressed beet pulp; POT, potatoes.

the diet increased milk output and the conversion efficiency of feed N into milk N, but N<sup>M</sup> excretion decreased only slightly.

In contrast to the large variation in N<sup>M</sup> excretion, simulated variation in N<sup>E</sup> and N<sup>R</sup> excretion was small. The strategies that combine a high DMI with a relatively high rumen degradability of the carbohydrate fractions (EC silages, PBP, CONC 40%) showed a slightly higher N<sup>E</sup> excretion, as a result of an increased microbial synthesis in the rumen and the LI. The variation in Pu fraction between silages was minor, helping to explain that variation in N<sup>R</sup> was small. The N<sup>R</sup> excretion appeared to be rather constant, and thus the variation in C:N ratio of the N<sup>R</sup> fraction (**Figure 2**) can be attributed to differences in C (OM) excretion. In case of diets with LC or maize silage, the high C excretion was induced by a high Fu fraction of the diet. Diets with a high concentrate level (CONC 40%) result in a high C excretion because the ruminal NDF digestion was impaired as a result of high rumen fractional passage rates and low rumen pH.

A high total OM excretion was induced either by a high DMI (CONC 40%, PBP, POT), a low OM digestibility (LC silages), or a combination of both (MSIL, **Figure 3**). A low concentrate level (CONC 20%) or the use of straw (STR) decreased total OM excretion due the relatively low DMI with these strategies. Diets with the highest urinary excretion (HFEC, NO, PBP, POT, 40% CONC) showed the highest OMNF excretion. The excretion of fiber OM (sum of OMRIF and OMRDF) is mainly determined by the amount of undigested NDF and was highest for LC silages,

TABLE 7 | Mean values and ranges of simulated slurry composition after 4 months of storage and simulated ammonia losses for 40 nutritional strategies<sup>a</sup> for dairy cows.


<sup>a</sup>Nutritional strategies and feed characteristics are described in Tables 2, 3. <sup>b</sup>C:N, carbon (C) to nitrogen (N) ratio; OM, organic matter.

MSIL and 40% CONC. The fraction of OMRDF clearly reflects the efficiency of rumen NDF digestion. The lower DMI with the STR and 20% CONC diets indicates a more efficient rumen digestion of potentially rumen degradable NDF because of a lower fractional rate of rumen passage and a higher pH.

### Simulated Slurry Composition

The simulated fraction of slurry N present in ammonium (NM) after 4 months of storage was on average 52%, with a considerable variation between diets (range 34–65%, **Table 7**). Simulated total N content of the slurries was on average 61 g kg−<sup>1</sup> OM (range 38–98) and the largest part of the variation was caused by variation in the simulated NH4-N content (14–64 g kg−<sup>1</sup> OM). The variation in simulated organic N content was considerably smaller (23–35 g kg−<sup>1</sup> OM, **Table 7**). Both C:Ntotal ratio (4.4– 11.9) and C:Norganic ratio (12.4–19.3) showed a large variation. The simulated ammonia-N loss was on average 37 g day−<sup>1</sup> , but ranged from 11 to 73 g day−<sup>1</sup> . These losses accounted for 9.7% (range 5.1–13.3) of the total excreted N (**Table 7**). Lowest slurry organic N contents were simulated for diets based on LC silages (HFLC, LFLC), MSIL, and a high use of concentrates (40% CONC, **Figure 4**). The highest inorganic N contents were simulated for diets based on grass silage HFEC and the lowest values for diets based on grass silage LFLC or on MSIL.

## DISCUSSION

### Evaluation of the Selected Nutritional Strategies

Nutritional strategies to reduce excessive N excretion to the environment often focus on an improvement of the feed N conversion. In this study, the selected strategies were aimed at an increase of the feed N conversion compared to the basic situation where highly fertilized early cut grass silage (HFEC) is fed as the sole forage. Simulated feed N conversion was higher indeed for all strategies that included an adaptation of the silage type and/or a replacement of grass silage (**Table 3**). In line with experimental observations, reducing the fertilization level of grass silage (e.g., Peyraud and Astigarraga, 1998; Warner et al., 2016) and the inclusion of maize silage (e.g., Van Gastelen et al., 2015) showed a strong potential to increase feed N conversion. The other strategies showed only moderate effects as a result of a decreased milk production (LC silages, straw) or an increased feed intake (PBP and POT). The average simulated feed N conversion was similar to that reported in a meta-analysis for North European dairy cattle fed primarily silage based diets (29 vs. 28%, respectively; Huhtanen and Hristov, 2009) whereas the simulated range was somewhat smaller than that reported (23–37 vs. 16–40%, respectively).

To obtain a large range in the dietary protein level, the concentrate composition was not adjusted for the protein level of the forages. This occasionally resulted in low dietary crude protein contents and rumen degradable protein balances (**Table 4**). Still, aminogenic nutrients were never predicted to be in short supply; FPCM production was limited by the availability of energy in most of the situations (n = 38) or occasionally by glucogenic nutrients (n = 2). Furthermore, the range in dietary protein content (**Table 4**) is quite similar to the range Huhtanen and Hristov (2009) reported (101 to 252 g kg−<sup>1</sup> DM), and therefore the simulated nutritional strategies might be interpreted as a realistic representation of diets for lactating dairy cows with respect to the dietary protein content. In practice, the selected forages will often be supplemented with byproducts or concentrates aimed to balance the diet offered to cattle, to avoid nutrient deficiencies.

### Simulation of Excreta Composition Fecal Excretion

Several authors have shown that an increase in N intake results in a moderate, linear increase of excretion of fecal N and milk N combined with a much more pronounced linear (Kebreab et al., 2010) or exponential (Castillo et al., 2000; Kebreab et al., 2001) increase in the excretion of urinary N. Our simulation data reproduce a similar pattern (**Figure 5**). The average level of fecal N excretion (154 g day−<sup>1</sup> ) is well in line with experimental data of Castillo et al. (2000); Kebreab et al. (2001), and Spek et al. (2013; European data). However, the simulated range in fecal N excretion is smaller than observed in some of these trials. The small variation in fecal N excretion may partly be attributed to the limited range in Pu fraction of the grass silages. The Pu fraction (range, 16–19 g kg−<sup>1</sup> DM; **Table 2**) was estimated according to regression equations by Tamminga et al. (1991). In their approach, Pu was the N residue × 6.25, remaining in nylon bags after prolonged rumen incubation (336 h) of 17 different grass silages; using stepwise regression they obtained a Pu prediction equation based on grass silage characteristics. The Pu fractions actually observed in the study of Tamminga et al. (1991) ranged from 7 to 29 g kg−<sup>1</sup> DM, in line with Pu fractions reported by Heeren et al. (2014) (10–29 g kg−<sup>1</sup> DM). Other experiments showed that variation in the Pu fraction of individual grass silages may even be larger (Von Keyserlingk et al., 1996; Bruinenberg et al., 2004). The Pu fraction is excreted with feces and it determines directly the amount of N excreted with FFFC. The simulated N excretion with the FFFC fraction for

the selected strategies showed a limited variation of only 40–48 g N day−<sup>1</sup> (data not shown). When more variation in Pu would have been assumed, this range would have been greater, directly implying a larger range in total fecal N and N<sup>R</sup> excretion.

The fraction of fecal N excreted with FFFC (24–32%, data not shown) is slightly higher than measured fractions of NDF-N in feces by Sørensen et al. (2003, 14–21%) and Powell et al. (2006, 18–29%) after feeding a large range of diets to dairy cows. The simulated proportion of fecal N being present in microbial material ranged from 47 to 55% and was somewhat lower than reported values of 70% by Robinson and Sniffen (1985), 53–73% by Robinson et al. (1987), and 61% by Larsen et al. (2001). Mason et al. (1981b) stated that the main components of the fecal water soluble N have their origins in intestinal excretion. In our study, the fecal N contained in endogenous material (FEMC) amounted up to 13–19%, being of similar magnitude as the fractions of water-soluble N reported for dairy cows (25%, Larsen et al., 2001) and sheep (15–24%; Mason et al., 1981a,b).

The assumptions for LI digestibility resulted in an average apparent N digestion in the LI of 6% of the outflow from the SI (ranging from −1 to 10%) which is considerably lower than that found for sheep (21%, Drochner and Meyer, 1991). This lower value may partly be attributed to the lower retention time for digesta in the LI of dairy cows compared to sheep. Apparent N digestion in the LI ranged from −1 to 15 g/d. This range is only slightly below that of 5–20 g N day−<sup>1</sup> derived from Van der Walt (1993) and it is therefore not likely that the net N digestion in the LI has been underestimated significantly.

The fraction of fecal OM excreted with NDF ranged from 43 to 71% (data not shown) and corresponds reasonably with reported values of 57–61% of fecal OM by Robinson et al. (1987), 32–56% of fecal DM by Sørensen et al. (2003), and 50–60% of fecal DM by Powell et al. (2006). According to Van Soest (1994) the N content of the non-NDF fecal OM is 7%. Our simulated average N content of 8% (data not shown) is in agreement with this figure. In our simulations, microbial OM appeared to contribute most to fecal OM excretion. The simulated non-NDF fecal OM consisted for 9–15% of FEC, 66–75% of FMC and the remainder (14–24%) was FOFC.

### Urinary Excretion

Simulated urinary N excretion was on average 58% of total manure N output, and showed a much larger variation (81–388 g N day−<sup>1</sup> ) than simulated fecal N excretion (128–177 g N day−<sup>1</sup> ). The simulated proportion of urinary N excreted with urea ranged from 62 to 86% with an average of 78%. These values are within the range reported in a review on urine composition (52 to 93%; Dijkstra et al., 2013). Bussink and Oenema (1998) stated that non-urea-like urinary components (UNUC) are generally excreted in fairly constant amounts and is on average 31 ± 4 g N day−<sup>1</sup> . As our assumptions are partly based on the same data sources, our simulations have a similar UNUC-N excretion of 27 ± 6 g N day−<sup>1</sup> . The variation in the simulated UNUC-N excretion is mainly determined by the variation in Hi (5% of total urinary N) and Aa (2% of total urinary N) as the other UNUC constituents were estimated either as a constant value (g day−<sup>1</sup> ) or as very small fractions of total urinary N. In their review on urine composition of cattle, Dijkstra et al. (2013) reported N from Hi to vary between 3.4 and 8.0% of total urinary N. Elevated urinary Hi fractions would result in an increase of the urinary C:N ratio. Hippuric acid is mainly a derivative of rumen microbial fermentation of phenolic acids, which are constituents of plant lignin (Martin, 1982). With advancing plant maturity, both the solubility and degradability of various plant phenolic compounds decrease, which may result in reduced excretion of hippuric acid in urine (Dijkstra et al., 2013). However, experimental data to support this hypothesis are lacking.

### Simulation of Slurry Composition

During storage of liquid manure, OM is subject to both anaerobic and aerobic bacterial degradation. To predict the composition of the slurry that is actually applied to the field, the digestion model presented in this study was extended with simple equations describing these processes. The formulated assumptions resulted in organic and inorganic N contents of stored slurry of on average 29 and 33 g kg−<sup>1</sup> OM, respectively (**Table 7**). These values are in line with the average values derived from a large database (2011– 2013) of Dutch dairy slurries (27 and 30 g kg−<sup>1</sup> OM, respectively) (Velthof et al., 2015; CBGV (Committee Fertilisation Grassland Forage Crops), 2017).

In this study, the proportion of OM degraded during 4 months of storage was estimated at 13% based on Sørensen (1998). This value was obtained at a temperature of 15◦C (Sørensen, 1998). Hindrichsen et al. (2006) reported a far higher OM degradation ranging from 32 to 47% within 14 weeks of anaerobic storage after feeding four different diets. This experiment was, however, conducted at an ambient temperature of 24◦C. Whitehead and Raistick (1993) showed that slurry OM degradation after 3 weeks of storage, ranged from 14 to 34% and increased with slurry temperature (5–35◦C). The lower temperature is close to the average Dutch winter temperature, explaining why the estimated 13% OM degradation provides a reasonable representation of the average Dutch winter situation; applying the model to other regions would likely require this factor to be changed.

Losses of N during the storage period (**Table 7**) ranged from 11 to 73 g cow−<sup>1</sup> day−<sup>1</sup> for the 40 nutritional strategies. These results confirm the strong potential to reduce ammonia emission by means of a reduction of the dietary protein content, as observed before (Paul et al., 1998; Külling et al., 2001), and also reported in a recent meta-analysis on ammonia emissions from dairy cattle housing (Bougouin et al., 2016). However, actual N losses depend on a number of variables like temperature, moisture, air flow, cleaning frequency, urease activity, and urine puddle replacement rate (Hristov et al., 2011; Bougouin et al., 2016). The formulated assumptions, aimed to illustrate the impact of differences in cow excreta composition on losses in the stable and during storage, provide a satisfactory representation of the average Dutch manure storage process. However, it is recognized that an accurate simulation of slurry storage processes, also in different regions, requires a more detailed representation of the effect of several storage conditions including pH, temperature, and exposed surface area (e.g., Rotz et al., 2014).

# Nutritional Strategies and the Composition of Slurry N

The present model simulates a large variation in slurry N content (**Figure 4**). This variation intrinsically affects the plant availability of N after field application. Expressed per kg of slurry N, plant availability is related to the Ninorganic: Ntotal ratio (e.g., Reijs et al., 2007; Cavalli et al., 2016). This ratio ranged from 0.34 to 0.65 for the 40 nutritional strategies with an average of 0.52, and resulted from a marked variation in slurry inorganic N combined with a moderate variation in the organic N content of slurry (**Table 7**). This average Ninorganic: Ntotal ratio corresponds with that (0.49) in a database (2008–2010; Den Boer et al., 2012) of dairy cattle slurries, but with a somewhat smaller variation than that observed (SD of 0.079 and 0.135, respectively).

Organic N in slurry is mainly derived from fecal material. Due to a large variation in fecal OM excretion (3.1–5.8 kg day−<sup>1</sup> , **Table 5**) compared to the relatively smaller variation in fecal N excretion (128–177 g day−<sup>1</sup> ), the simulated fecal N content was positively related to the apparent digestibility of the diet. This is in correspondence with findings of Kyvsgaard et al. (2000) and Sørensen et al. (2003). In the present study, low fecal N contents were simulated for diets that contained LC silages, MSIL, or 40% concentrate feeds. In accordance, these nutritional strategies also showed lowest organic N contents in the slurry (**Figure 4**). The variation in simulated slurry organic N content (from 23 to 35 g kg−<sup>1</sup> OM) was smaller than the variation in fecal N content (from 27 to 49 g kg−<sup>1</sup> OM) due to the fact that high digestible diets (e.g., EC silages) contain also a relatively small NDF fraction, implying a higher mineralization of fecal N during storage. Our simulation results reveal only a limited scope for variation in the organic N content of slurry.

The major part of the variation in slurry N content results from the variation in the inorganic N content of slurry (**Figure 4**). This inorganic N content is determined by the excretion of UUC-N relative to the total OM excretion. Therefore, the highest slurry inorganic N contents are observed when diets are fed that combine a high UUC-N excretion with a high OM digestibility. This combination is highly applicable to the nutritional strategies in this study based on the HFEC silages. The opposite is true for diets based on LFLC grass silage and MSIL: a low excretion of UUC-N coincides with a high OM excretion. As the LFLC diets also contain a high NDF fraction, the simulated mineralization of fecal N during storage was low, resulting in extremely low inorganic N contents and Ninorganic: Ntotal ratio's.

Sørensen et al. (2003) showed that slurry N availability (expressed per kg slurry N) to a barley crop was strongly related to the slurry C:Ntotal ratio. These findings were confirmed on grassland by Reijs et al. (2007) and may be explained by an immobilizing effect of organic manure components with a high C:N ratio (Chadwick et al., 2000; Chrystal et al., 2016). Slurry C:Ntotal ratio is affected by the composition of the diet and reported values range from 7.5 to 10.5 (Paul et al., 1998), from 6.4 to 13.1 (Sørensen et al., 2003), and from 5.1 to 11.4 (Reijs et al., 2007). The simulated range in C:Ntotal ratio (from 4.4 to 11.9) is in line with these results. Following our model simulations, a high C:Ntotal ratio reflects both a high C excretion with FFFC and a low N excretion with UUC. Again, diets that combine a low UUC-N excretion with a low NDF digestion (LFLC, MSIL) show the highest values, whereas lowest values are observed for diets with excessive availability of digestible protein and a highly digestible NDF fraction (HFEC). The other selected strategies did not cause pronounced effects on simulated C:Ntotal ratio (**Figure 4**). The latter results indicate that substantial changes in slurry C:Ntotal ratio and the subsequent plant availability of N, require rather large adjustments in the diet composition, affecting both UUC-N excretion and FFFC-OM excretion.

## The Added Value of the Followed Approach

In this study, a dynamic and mechanistic model of rumen fermentation was used to predict the composition of excreta as a function of diet composition. The results indicate a satisfactory prediction of production and composition of feces and urine, as the simulated ranges in fecal and urinary composition were largely in line with values reported in literature. Some areas are identified to require additional refinement of the model, in particular the prediction of hindgut digestion and of the amount of non-urea-like urinary components. The present model takes into account interactions between different types of nutrients and the interaction with microbial activity. Therefore, its use may significantly improve the prediction of feed digestion in comparison to current static feed evaluation systems (Bannink et al., 2016). This feature is clearly illustrated by the prediction of a reduced digestion of rumen digestible fiber on diets that contain a large fraction of concentrate feeds (**Figure 3**). Our predictions are qualitatively in line with observations of Sørensen et al. (2003) who showed that the content of forage-derived decomposable fiber in the slurry was higher when the diet included concentrates. The model predicts the partitioning of N excretion in feces and urine and contributes to a better understanding of the effect of nutritional strategies on the utilization of N in the cow and the direct losses of N in the slurry storage. Losses during storage of slurry were assumed to occur at typical average ambient temperatures in the winter in the Netherlands. Application of the model to other seasons or regions with different ambient temperature would require modification of the value adopted in the present model to result in changed composition of stored slurry.

Compared to an earlier integrated model (Kebreab et al., 2004) the model presented in this study predicts not only the amount but also the composition of excreta N. Mineralization, immobilization and plant availability of N from soil-applied dairy manure is affected by the composition of the manure (Chadwick et al., 2000; Powell et al., 2006; Chrystal et al., 2016). These effects are often complex and variable for different crops and soils, and therefore the plant availability of N following organic manure application is difficult to predict. Several authors have shown that differences in plant availability of N from soil-applied manure are related to differences in cow nutrition (Kyvsgaard et al., 2000; Sørensen et al., 2003; Powell et al., 2006). The current model helps to understand how differences in manure composition are related to the composition of the diet and therefore it might contribute to a better prediction of plant availability of N following field application of cattle manure.

# CONCLUSIONS

The model represents digestion and enteric microbial metabolism in the cow and helps to understand effects of changes in diet composition on excreta composition. The simulation results demonstrate the substantial effects of diverging diets on total N excretion and the composition of excreta in terms of immediately available N for plant uptake and the C:N ratio of the resistant N fraction. Diets with high fertilized, early cut grass silage resulted in the greatest urinary N excretion and ratio of slurry inorganic N to OM, whilst the opposite occurred with low fertilized, late cut grass silage or upon inclusion of maize silage. Further efforts are required to integrate effects of slurry storage conditions on the storage processes and subsequently on stored slurry composition. The model may significantly contribute to a better utilization of N from field applied manure and it can provide essential information for a more elaborate, integrated evaluation of the effect of different nutritional strategies at the whole-farm level.

# AUTHOR CONTRIBUTIONS

JR, JD, and EL developed the concepts. JR, JD, PB, AB, and EL developed the equations. JR and AB performed the simulations and analyzed the data. JR wrote original draft of manuscript and JD, PB, AB, and EL contributed to discussion and revision of this manuscript.

# ACKNOWLEDGMENTS

The paper is based on a chapter in the PhD thesis of JR (Reijs, 2007). This research was financially supported by the Social Sciences Research Council of the Netherlands Organization of Scientific Research (NWO). We thank Henk Valk (Animal Sciences Group, Wageningen University and Research Centre, the Netherlands) and Peter Sørensen (Department of Agroecology, Danish Institute of Agricultural Sciences, Denmark) for useful suggestions on the estimations of grass silage composition and slurry storage processes.

# REFERENCES


to pasture on N mineralisation and forage growth. New Zealand J. Agric. Res. 59, 324–331. doi: 10.1080/00288233.2016.1188131


Energy and Protein Metabolism and Nutrition, ed. G.M. Crovetto (Wageningen: Wageningen Academic Publishers), 417–425.


synthesis estimated from excretion of total purine derivatives. J. Dairy Sci. 82, 2686–2696. doi: 10.3168/jds.S0022-0302(99)75525-6


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Dijkstra, Bannink, Bosma, Lantinga and Reijs. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Diet Supplementation With Pomegranate Peel Extract Altered Odorants Emission From Fresh and Incubated Calves' Feces

### Edited by:

*Claudia Wagner-Riddle, University of Guelph, Canada*

### Reviewed by:

*Anders Peter S. Adamsen, SEGES Agriculture and Food FmbA, Denmark David Bruce Parker, Plains Area, Agricultural Research Service, United States Department of Agriculture, United States*

### \*Correspondence:

*Ariel Shabtay shabtay@volcani.agri.gov.il Yael Laor laor@volcani.agri.gov.il*

*†These authors have contributed equally to this work.*

### Specialty section:

*This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems*

> Received: *22 February 2018* Accepted: *11 June 2018* Published: *17 July 2018*

### Citation:

*Varma VS, Shabtay A, Yishay M, Mizrahi I, Shterzer N, Freilich S, Medina S, Agmon R and Laor Y (2018) Diet Supplementation With Pomegranate Peel Extract Altered Odorants Emission From Fresh and Incubated Calves' Feces. Front. Sustain. Food Syst. 2:33. doi: 10.3389/fsufs.2018.00033* Vempalli S. Varma1†, Ariel Shabtay <sup>2</sup> \*, Moran Yishay 2†, Itzhak Mizrahi <sup>3</sup> , Naama Shterzer <sup>3</sup> , Shiri Freilich<sup>4</sup> , Shlomit Medina1,4, Rotem Agmon<sup>2</sup> and Yael Laor <sup>1</sup> \*

*<sup>1</sup> Newe Ya'ar Research Center, Institute of Soil, Water and Environmental Sciences, Agricultural Research Organization, Ramat Yishay, Israel, <sup>2</sup> Newe Ya'ar Research Center, Institute of Animal Sciences, Agricultural Research Organization, Ramat Yishay, Israel, <sup>3</sup> Department of Life Sciences and the National Institute for Biotechnology in the Negev, Ben-Gurion University of the Negev, Beer-Sheva, Israel, <sup>4</sup> Newe Ya'ar Research Center, Institute of Plant Sciences, Agricultural Research Organization, Ramat Yishay, Israel*

Emissions of odorous volatile organic compounds (VOCs) from livestock manure are of increasing environmental concern. Recent approaches in cattle nutrition and health make use of Mediterranean fruit byproducts, such as concentrated pomegranate peel extract (CPE) that carry antioxidant activities, which in turn may alter manure properties. This study explored the effect of CPE on odorants emission from beef calves feces. Fourteen calves were randomly assigned to control (*n* = 7) and pomegranate (*n* = 7) treatments. The latter was supplemented with 4% CPE in milk until weaning at the age of 60 d. Following weaning, 4% CPE was added to calves ration, on dry matter basis. The control treatment received only milk or solid feed, respectively. Fresh feces of the four treatments (control/pomegranate; before/after weaning) were sampled twice, 2–3 w before and 4–5 w after weaning, and then incubated (28◦C) for 0, 7, 14, and 30 d. Sub-samples were placed in a flux chamber (37◦C) and VOCs collected on thermal desorption (TD) tubes followed by TD-GC-MS analysis. In all treatments, flux quantities followed the general order of volatile fatty acids (VFAs) > alcohols > phenolic & aromatic > sulfuric > esters > aldehydes. Total VOCs, especially VFA fluxes, peaked on day 7 in correspondence with pH dynamics. The fractional contribution of alcohols, phenolic & aromatic and sulfuric VOCs generally increased during incubation. After weaning, short-chain VFAs flux was 5.2 times higher and the pH was 1.26 units lower in the pomegranate treatment (average on days 7&14), suggesting increased fermentation due to possible effect of CPE on gastrointestinal microflora. An automated ribosomal intergenic spacer analysis of fresh and incubated fecal DNA confirmed association between microbial fingerprinting and short chain VFAs, phenolic and sulfuric VOCs. Odorants emission after weaning, expressed as odor activity values, was 2 times higher in the pomegranate treatment (average on days 7&14) and generally dominated by VFAs,

**89**

while in other treatments the contribution of phenolic and sulfuric odors increased as incubation proceeded. In conclusion, diet supplementation with CPE may be adopted with the purpose of increasing calves health and production, but it may alter odor characteristics of feces and increase feedlot nuisance if not managed properly.

Keywords: diet, odor, beef cattle, manure, volatile organic compound (VOCs), odor activity values (OAVs), automated ribosomal intergenic spacer analysis (ARISA), short chain fatty acids (SCFAs)

### INTRODUCTION

Odor emissions from livestock farms trigger daily issues at the rural-urban interface. Livestock odor results from synergistic contribution of inorganic gases (mainly ammonia and hydrogen sulfide) and a large number of volatile organic compounds (VOCs). Sulfur and nitrogen containing VOCs, volatile fatty acids (VFAs), esters, alcohols, aldehydes, ketones, phenolic and aromatic compounds, aliphatic and halogenated hydrocarbons have all been resolved in livestock manure and barn atmosphere of swine (Schiffman et al., 2001; Sunesson et al., 2001; Cai et al., 2006; Lo et al., 2008) and dairy farms (Filipy et al., 2006; Laor et al., 2008). Sulfuric, VFAs, phenolic and aromatic VOCs are considered among the most offensive and characteristic odorants associated with livestock operations (Wright et al., 2005; Bulliner et al., 2006; Koziel et al., 2006; Laor et al., 2014; Woodbury et al., 2015; Yuan et al., 2017).

Diet formulation in combination with manure management have a principal effect on odor formation. Odorous compounds are mainly generated by microbial conversion of non-utilized dietary nutrients and endogenous products secreted in the gastrointestinal tract under anaerobic conditions (Le et al., 2005). Adjustment of feed ingredients would potentially affect manure odor through more efficient use of nutrients supplied, more nearly meeting animal requirements without providing excess nutrients, changes in fermentation patterns within the animal, or changes in post-excretion decomposition patterns. As odorous compounds are formed or decomposed during manure storage, diet manipulations are expected to affect not only the composition of fresh manure but also the formation of odorants during its storage (Yasuhara et al., 1984; Powers, 1998; Gralapp et al., 2002; Le et al., 2005). Sutton et al. (1999) suggested that the primary odor causing compounds in swine manure are evolved with an excess of degradable proteins and due to lack of specific fermentable carbohydrates during microbial fermentation. Previous studies on swine diet reported decrease in nitrogen excretion upon reducing the amount of dietary proteins (Kerr, 1995; Hobbs et al., 1999; Zervas and Zijlstra, 2002). Studies on cattle manure showed the production of VFAs as a major product of starch fermentation in fresh and aged manure (Miller and Varel, 2001; Archibeque et al., 2005; Miller et al., 2006). Shabtay et al. (2009) showed distinct changes in VOC emissions from fresh and aged feces related to calves' development. At age of 7 weeks, a week before calves weaning, feces seemed to be the most offensive, presumably due to partial dietary switch from milk to solid feed that is taking place at this stage and not fully synchronized with the metabolic development of the rumen, hence, hampering the balance between animal's requirements and diet formulation.

In recent years, a growing awareness of the beneficial effects of pomegranate consumption in the human diet (Aviram et al., 2004) has triggered the development of additional industrial pomegranate products. This, in turn, has led researchers to investigate the effects of pomegranate industrial byproducts (i.e., peels) on ruminant health and production (Shabtay et al., 2008, 2012; Oliveira et al., 2010; Jami et al., 2012; Weyl-Feinstein et al., 2014).This recent interest arises from the need to explore alternative feed sources in order to supply cattle with an available and affordable ration, in light of global and local economic processes and prolonged drought periods. Moreover, pomegranate peel attracts attention especially due to its antioxidative capacities (Tzulker et al., 2007), which has often been associated with a decreased risk of various diseases and mortality. To this end, a commercial and standardized product, concentrated pomegranate peel extract (CPE), has been developed by Gan Shmuel Food Ltd. (Kibbutz Gan-Shmuel, Israel). The final CPE is standardized to ensure a uniform and constant dry matter content, and is available year round for feeding ruminants. Supplementing milk with 3.75% CPE to neonatal Holstein calves (Holstein male calves reared for meat) reduced fecal oocyst count of the intestine parasite Criptosporidium parvum and diarrhea intensity and duration with no deleterious effect on average daily gain (ADG) (Weyl-Feinstein et al., 2014). In another study, supplementing lactating cows with 4% CPE significantly increased digestibility of dry matter (DM), protein, and neutral detergent fiber, as well as milk and energy-corrected milk yields (Jami et al., 2012).

Besides its beneficial effects on ruminants' health and production (Shabtay et al., 2008, 2012; Weyl-Feinstein et al., 2014) dietary CPE was shown to significantly affect rumen bacterial communities (Jami et al., 2012). Thus, we hypothesized that CPE may also influence feces properties with consequences on odor emissions from the cattle feedlot. To address this assumption, the current study explores the effect of CPE introduced into calves' diet on odorants emission from fresh and incubated calves' feces before and after weaning. In Israel, beef cattle feedlots differ in size; housing from 50 to 100 and up to several hundred calves at a time. Being a relatively densely populated country, these feedlots may be located in close proximity to urban areas and thus odor emissions are of daily concern.

### MATERIALS AND METHODS

### Calves' Husbandry and Nutrition

This study took place during 2012 at the research beef cattle feedlot in Newe Ya'ar Research Center of the Agricultural Research Organization (northern Israel). All procedures involving animals were approved by the Israeli committee for animal care and experimentation (AEEC—Animal Experimentation Ethics Committee—Volcani center). Fourteen Holstein calves (young horned Holstein bull calves, not castrated; Holstein male calves reared for meat) were randomly assigned to control (n = 7) and pomegranate (n = 7) treatments. Calves were born at Yagur dairy farm and fed pooled cow colostrum milk from the day of calving up to 3 days. At the age of 3 days calves were transferred (20 km distance) to a shaded nursing barn in Newe Ya'ar and housed in two adjacent pens, one for each treatment, until weaning at the age of 60 days. During this period calves of both treatments were bottle fed with 150 g of milk replacer "Halavit Platina" (23% protein, 18% fat; Koffolk, Phibro, Maabarot, Israel) mixed in 1 L of water twice daily, and supplemented with 4% CPE for the pomegranate treatment. In addition, calves were offered ad libitum suckling starter (cat# 7820, Milobar), containing 17% protein, 3% fat, 7% crude fiber, 7% ash, 1% Ca, 0.4% P, and 0.7% NaCl, on a DM basis and had free access to fresh water. Ten days before weaning, the CPE was supplemented only once a day with milk and the additional amount was poured on top of the suckling starter. After weaning, suckling starter was exclusively served ad libitum for an additional 2 months, on top of which the pomegranate treatment were supplemented with 4% CPE, on DM basis. The CPE of the "Wonderful" cultivar was supplied by Gan Shmuel Food Ltd. (Gan-Shmuel, Israel). It was made by chopping up the pomegranate parts remaining after pomegranate juice production, including peels and residual arils, followed by extraction in water, filtering, evaporation procedures and pasteurization; the final extract was standardized to ensure uniform and constant DM content (Shabtay et al., 2012). Nutritional and chemical values of CPE were (DM = 45%; other properties are expressed as percent of DM): crude protein (CP) = 2%, ash = 5%, total polyphenols (expressed as gallic acid equivalent) = 10%, total punicalagins = 3.3%, ellagic acid = 0.2%, and pH = 3.2. The CPE contained 0.75 g L−<sup>1</sup> of sorbate as a preservative and was kept refrigerated at 4◦C.

Average daily weight gain (ADG) was based on two body weight measurements, once upon arrival to Newe Ya'ar (age of ca. 3 days) and again about 2 months after weaning (age of ca. 4 months).

### Feces Sampling and Incubation

The experimental protocol is described in **Figure 1A**. Fresh feces was sampled following a rectal massage, 2–3 weeks before and 4–5 weeks after weaning. Each sampling period extended over 1 week, during which the fecal samples of each calf were collected separately and immediately stored at −20◦C. Then, the weekly pool collected from each calf was brought to room temperature, hand-mixed thoroughly, and divided into four 200 mL dark glass jars (5 cm dimeter × 10 cm height). One jar was immediately placed at −80◦C ("day 0") and the three other jars (sub-samples) were incubated at 28◦C for 7, 14, and 30 days. This incubation temperature represents average daily summer temperatures during July-August in the Newe Ya'ar region (Israel Meteorological Service, 2018). To avoid spillage during incubation (the manure was inflated due to gas formation), the jars were extended up by a "sleeve" and then loosely closed with a cap, all made by folded aluminum foil. Since the jars were not aerated actively, it presumably possessed both aerobic and anaerobic conditions, as also expected at the feedlot. Moisture content was measured on a weekly basis during incubation by drying sub samples at 60◦C for 24 h and adjusting the whole samples to their initial moisture with deionized water.

The pH was measured directly in the fresh feces on day 0 and then by the end of each incubation period, before transferring the sample to −80◦C (mobile ISFET pH meter IQ 150-77; IQ Scientific Instruments, Loveland, CO, equipped with a dry, non-glass probe; PH77-SS; IQ Scientific Instruments). Nitrogen, starch, and water soluble sugars were analyzed in fresh feces only: Total N content was determined according to method 984.13 of AOAC (1990). Crude protein was calculated as N × 6.25. Water soluble reducing sugars and starch contents were determined by Siap Laboratory (Bet Gamliel, Israel). Water soluble sugars were determined according to Miller (1959) using hydrolyzed sucrose as a standard. Starch contents in food samples (milk powder and dry food) and in fresh feces were determined using an alpha amylase assay after Pinchasov and Noy (1994), modified from Bernfeld (1955). Final starch values were obtained after subtracting the corresponding water soluble sugar values. The starch content in the milk powder was determined from the difference before and after chemical hydrolysis of the powder. Starch digestibility (Sdigest) was calculated from starch and N data, after Zinn et al. (2007) using Equation (1).

Sdigest = 100{1 − [(0.938 − 0.497FN + 0.0853FN<sup>2</sup> )FS/DS]} (1)

Where FN is fecal nitrogen, FS is fecal starch and DS is diet starch; all are represented as percentages of DM.

### VOCs Sampling and Analyses by GC-MS With Thermal Desorption

Before analysis, each specific glass jar was first transferred (remained closed) from −80◦ to 4◦C for a period of 24–48 h (freezing was assumed not to alter short-term VOC emissions from thawed samples based on Miller and Woodbury (2006) and Shabtay et al. (2009). Jars were then transferred to a water bath at 37◦C for 1 h and finally opened and mixed thoroughly. About 20 ml of feces was transferred into an open 7-cm diameter Petri dish which was placed in a flow-through flux chamber, composed of 9-L vessel that was laid within a water bath at 37◦C (**Figure 1B**). This temperature represents maximum daily summer temperatures during July-August in the Newe Ya'ar region (Israel Meteorological Service, 2018) and selected to simulate potential peak emissions during summer.

Prior to sampling, the chamber was flushed with nitrogen at 1 L min−<sup>1</sup> for a period of 27 min (three chamber volumes), thereafter sampling started under the same flow conditions. Air

samples were collected onto sorbent tubes (Markes, Stainless Steel, 3 1/2′′ × 1/4′′, 2 sorbents - Tenax TA & Carbograph 1TD), using a pocket pump (SKC210; SKC, 84, PA) set at 80 mL min−<sup>1</sup> for 10 min. Excess outflow was released through a venting port. Replicates were obtained by repeating this procedure for another feces subsample and the GC/MS results of duplicate tubes were averaged. Loaded sorbent tubes were analyzed using thermal desorption-gas chromatography-mass spectrometry (TD-GC-MS). The TD system consisted of a Markes UNITY 2TM thermal desorber (Markes International Ltd., Llantrisant, UK). Samples were quantified with an Agilent 7890A GC with Agilent 5975C MS detector (Agilent Technologies, Inc., Santa Clara, CA, USA). The tube was desorbed for 5 min at 280◦C with a carrier gas flow of 30 mL min−<sup>1</sup> and trapped on the cold trap maintained at 20◦C (graphitized carbon trap used for sampling VOCs of C4/<sup>5</sup> to C30/32). The cold trap was heated to 300◦C for 3 min with a carrier gas flow of 20 mL min−<sup>1</sup> and was transferred to the column in the GC-MS. The general run parameters used were as follows: injector, 230◦C; column oven, 45◦C initial temperature for 5 min, followed by a ramp of 4◦C/min to 150◦C, and 10◦C/min to 230◦C final, and 5-min hold with a total run time of 43.25 min; carrier gas, He; mass spectrometer ionization energy, 70 eV; m/z, 41 to 300; scan time, 5.4/s. The polar analytical column was an Agilent Innowax, 30 m, 0.25 mm ID capillary column (polyethylene glycol, 0.25µm film thickness).

### Determination of VOC Fluxes

Compounds present in feces emissions were identified using two criteria: (1) matching of observed retention times with those of pure compounds run as standards; (2) matching mass spectrums of unknown compounds using ChemStation (version D.00.00.38) from Agilent and Wiley7N spectral library (if no standard existed, identifications were arbitrarily made for matching quality ≥70%). Under the analytical conditions used, peak areas greater than ∼3,000 counts were considered effective to resolve. Therefore, resolved compounds which were found with peak areas greater than 3,500 were utilized for data analysis. Overall, 60 compounds were resolved; out of which we decided to follow 34 dominant or known odorous compounds, categorized into six groups (Table S1): (i) **VFAs** (12 compounds: acetic acid, propanoic acid, 2-methyl propanoic acid, butanoic acid, 3-methyl butanoic acid, pentanoic acid, 4-methyl pentanoic acid, hexanoic acid, heptanoic acid, octanoic acid, nonanoic acid and decanoic acid); (ii) **Esters** (2 compounds: ethyl hexanoate and hexyl hexanoate); (iii) **Alcohols** (10 compounds: sec-Butyl alcohol, iso-Butyl alcohol, n-Butanol, iso-Pentyl alcohol, 1- Pentanol, 1-Hexanol, 1-Heptanol, 2-Ethyl-1-hexanol, 2-phenyl ethanol, 3-Phenyl-1-propanol; (iv) **Phenolic & Aromatic** (5 compounds: phenol, p-cresol, indole, skatole and p-xylene); (v) **Aldehydes** (3 compounds: octanal, nonanal and decanal); and (vi) **Sulfuric** (2 compounds: dimethyl disulfide and dimethyl trisulfide).

For calibration, analytical grade standard solutions were prepared ranging in concentrations from 2.5 to 30 ng and up to 250–3,500 ng mL−<sup>1</sup> by diluting known masses of pure chemicals with methanol. Each cocktail of calibration analytes (1–3 µl) was injected using a GC syringe onto clean sorbent tubes connected to a calibration solution loading rig (CSLR, Markes International Ltd., Llantrisant, UK) at nitrogen flow of 80 mL min−<sup>1</sup> . All standard loaded tubes were prepared in duplicates and results were averaged. The loaded tubes were analyzed under the same conditions as used for the other samples. Standard curves of peak area counts vs. VOC mass (ng) were fitted using both linear and power regression analyses; both yielding high regression coefficients (R <sup>2</sup> ≥ 0.98 in most cases). However, a power regression generally yielded slightly higher R 2 -values and was thus used for the present study. For compounds of which we did not have pure standards, we used the average regression parameters of the respective VOCs group (Table S1).

The flux, J (µg m−<sup>2</sup> min−<sup>1</sup> ), of specific or group of VOCs was calculated using Equation (2).

$$\mathbf{J} = \mathbf{Q} \, \mathbf{C} / \mathbf{A} \tag{2}$$

Where Q is the flushing inflow (m<sup>3</sup> min−<sup>1</sup> ), C is the VOC concentration in the flushing air (µg m−<sup>3</sup> ) and A is the surface area of the feces inside the Petri dish (m<sup>2</sup> ). The concentration, C, was calculated from the mass of VOCs (based on calibration curves) and the volume of air sample loaded onto the tube.

### Odor Activity Values (OAV)

A single-compound odor threshold (SCOT) is defined as the lowest concentration of a single compound in air that can be detected by the human olfactory sense when compared to a nonodorous sample and is comparable to odor detection threshold or dilution-to-threshold, DT (Parker et al., 2010; Laor et al., 2014). The SCOT values used in this study are the calculated geometric means of listed values found in the comprehensive review of van Gemert (2003), as used by Parker et al. (2012). The OAV (dimensionless) was calculated for each individual VOC using Equation (3).

$$\text{OAV} = \text{C} / \text{SCOT} \tag{3}$$

Where C is the VOC concentration in the flushing air (µg m−<sup>3</sup> ) as used for Equation (2), and SCOT is the geometric mean odor detection threshold of that individual compound (µg m−<sup>3</sup> ).

### DNA Extraction and Automated Ribosomal Intergenic Spacer Analysis (ARISA)

Fresh and incubated feces samples used to analyze VOCs also underwent DNA extraction and microbial fingerprinting by means of automated ribosomal intergenic spacer analysis (ARISA). For each sample, about 0.5 g was transferred into an Eppendorf tube and stored at −20◦C. Before extraction, the samples were brought to room temperature and extracted using QIAmp DNA Stool Mini Kit (QIAGEN GmbH, D-40724 Hilden, Germany). DNA concertation was determined by Nanodrop (NanoDrop Technologies, Montchanin, DE, USA; ND-1000 V3.3.1 software) and ca. 10 ng µL <sup>−</sup><sup>1</sup> was taken for the ARISA.

Two technical replicates of DNA from each sample were subjected to PCR amplification for ARISA (Fisher and Triplett, 1999), using the primers ITSF (5′ -GTCGTAACAAGGTAG CCGTA-3′ ) and ITSRtet (5′ -GCCAAGGCATCCAAC-3′ ), fluorescently labeled with TET, as described previously (Welkie et al., 2010). The ARISA PCRs were carried out in 15 µL volumes containing Fermentas DreamTaq Master Mix (Fermentas, Madison, WI), 0.33µM of each primer and 20 ng of template DNA. The PCR conditions were: 94◦C for 2 min followed by 30 cycles of 94◦C for 1 min, 55◦C for 60 s, and 72◦C for 120 s, and finally one cycle of 72◦C for 5 min. The ARISA PCR products were analyzed using an ABI PRISM 3,100 Genetic Analyzer (Applied Biosystems Inc., Carlsbad, CA) along with a custommade ROX-labeled 250- to 1,150-bp standard (BioVentures Inc., Murfreesboro, TN). Raw data generated were initially analyzed using GeneMarker software (SoftGenetics LLC, State College, PA) according to Kovacs et al. (2010). After performing accurate size calling using the program, all data were exported to Microsoft Excel for further analysis. All operational taxonomic units (OTUs) with fluorescence intensity of ≤10 relative fluorescence units were excluded. The remaining OTUs were binned as described by Brown et al. (2005) with the following parameters: bins of 3 bp (±1 bp) for fragments <700 bp in length, bins of 5 bp for fragments ≥700 and ≤1,000 bp in length, and bins of 10 bp for fragments >1,000 bp. Intensities were then summed for each bin. Relative intensities for each binned OTU in a given sample were calculated and OTUs that constituted less than 0.1% of the total intensity of the sample were excluded.

### Statistical Analyses

For general feces properties, the significance of differences between treatments were analyzed by the Tukey-Kramer HSD test at P ≤ 0.05. In certain cases, specific P-values are reported for t-tests or a regression analysis coefficient. For all VOCs (logtransformed emission values), the significance of the overall effect of treatment and incubation time was analyzed by the standard least squares fit model and compared by Tukey-Kramer HSD test at P ≤ 0.05 (JMP software, SAS Institute Inc., Cary, NC). Spearman's rhos correlation coefficients between the distribution patterns of microbial OTUs obtained by ARISA and emission profiles of all 34 VOCs were calculated using cor.test function and plotted using heatmap 0.3 function in R.

### RESULTS

# Calf Performance, Fresh Feces Characteristics and pH Dynamics During Incubation

The average daily gain, calculated from calving to 4 months of age, was not affected by the dietary supplementation of CPE; ADG of the control treatment (1.08 ± 0.15 kg day−<sup>1</sup> ) (±standard deviation) did not differ from that of the pomegranate treatment (1.10 ± 0.11 kg day−<sup>1</sup> ; P = 0.86). Selected properties of the fresh feces, before and after weaning, are summarized in **Table 1**. Fecal DM did not differ between control and pomegranate treatments before or after weaning. It is noteworthy, however, that the lowest DM value was obtained in the pomegranate treatment after weaning (PA), and it differed significantly only from DM of the pomegranate treatment before weaning (PB). The PA treatment was also characterized by the lowest pH, which differed significantly only from the control treatment before weaning (CB). Organic matter (OM), protein and water soluble sugars contents did not differ among the treatments. Starch content was significantly higher in the PA compared to CA and PB, but comparable to CB. Starch digestibility was compared between PA and CA treatments, using an equation designed for non-suckling calves (Equation 1; Zinn et al., 2007), revealing 7.3% lower values for the PA treatment (89.0% for CA vs. 81.7% for PA; P = 0.001). Interestingly, based on comparison of fecal starch content of the suckling treatments (for which Equation 1 is considered less relevant), it can be assumed that the effect of CPE on starch digestibility had an opposite trend, as values tended to be higher in the CB treatment (P = 0.07; **Table 1**).

The dynamics of feces pH during incubation are presented in **Figure 2**. The pH of the fresh feces (day 0) decreased after 1 week of incubation in three out of the four treatments. The pH of the PA treatment, which initially was the lowest among all other treatments, decreased most substantially on day 7 (pH = 4.77) and then increased gradually to reach a value of 6.40. While similar but less intense trends were observed also in the PB and

TABLE 1 | Selected properties of fresh feces collected before and after weaning (day 0, before incubation).


*Different letters for each property (in a row) denote significantly different values (P* ≤ *0.05). Organic matter, protein and starch are calculated on the basis of dry matter (DM). CB, control before weaning; PB, pomegranate before weaning; CA, control after weaning; PA, pomegranate after weaning.*

CA treatments, the dynamics of the CB treatment was relatively stable.

### VOC Emissions From Fresh and Incubated Feces

**Table 2** and Tables S2, S3 summarize the average emission values of 34 VOCs on days 0, 7, 14, and 30 for the four treatments, either separately or summed in VOC groups; compound concentrations (µg m−<sup>3</sup> ) and fluxes (µg m−<sup>2</sup> min−<sup>1</sup> ) are defined in Equation (2). The statistical significance of treatment and incubation time effects are summarized in **Table 3**. Out of the 34 VOCs, the dynamics of 6 selected compounds are plotted in **Figure 3**. In almost all cases, the fluxes of VOCs were lower on day 0 (significant in 18 out of 34 VOCs; **Table 3**), increased during incubation (days 7 and 14) and reduced thereafter. The effect of CPE diet was substantial on the emissions of short chain VFAs. The dynamics of propanoic and butanoic acids represent that pattern (**Figures 3A,B**), in which the flux was significantly the highest in the PA treatment and peaked on day 7 during incubation (the average total flux of acetic, propanoic and butanoic acids on days 7 and 14 was 5.2 times higher in the pomegranate treatment). On the other hand, CPE diet significantly reduced the emissions of sulfuric VOCs after weaning; dimethyl disulfide (**Figure 3C**) was 30.4 times lower in PA vs. CA on day 7. Similar effects after weaning were observed for p-cresol, for which the flux was 5.01 times lower in PA vs. CA on day 14 (**Figure 3D**) and skatole, for which the flux was 330 times lower in PA vs. CA on day 14 (**Table 2**). The CPE diet did not have a significant effect on most alcohol emissions; yet, nbutanol was significantly higher in PB vs. CB, mainly on day 30; **Table 3**).

The effect of weaning on VOC fluxes is demonstrated by the dynamics of phenol (**Figure 3E**) and indole (**Figure 3F**) which were similar in the control and pomegranate treatments, but were significantly higher before weaning. On day 7, fluxes were 11.6 and 10.1 higher before than after weaning, for phenol and indole, respectively (average of CB and PB vs. CA and PA).

**Figure 4** represents the dynamics of VOC fluxes and odor activity values (OAV), based on the categorized VOC groups.


TABLE

2




TABLE 3 | The significances of the treatment effect and the incubation time effect on VOC Fluxes.

*Log-transformed raw data were analyzed by the standard least squares fit model and compared by Tukey-Kramer HSD test. Different letters (in a row) denote significantly different fluxes (P* ≤ *0.05), whereas A represents the highest values. CB, control before weaning; PB, pomegranate before weaning; CA, control after weaning; PA, pomegranate after weaning.*

Flux quantities followed the general order of VFAs > alcohols > phenolic & aromatic > sulfuric > esters > aldehydes (**Figures 4A,A-1**). Overall, VFAs contributed 69–85% of the total fluxes on day 0, and increased up to 96% on day 7 in the PA treatment. Alcohols were the second group contributing to the total fluxes (with n-butanol being a major compound), and it increased during incubation with no clear reduction over the entire incubation period. Thus, together with VFAs and regardless of treatment and incubation time, these two groups were most dominant and contributed from 63 to nearly 100% of the total fluxes. The contribution of these two groups was nearly 100% on day 7 in the PA treatment. Notably, although alcohol fluxes did not differ significantly between treatments, n-butanol was the most dominant alcoholic compound in all treatments at the beginning of incubation. Then, on day 14, the dominance of 1-pentanol and 1-hexanol became similar to that of n-butanol.

Phenolic & aromatic VOCs were the third group to contribute to the total fluxes with increased contribution as incubation proceeded, especially in the control treatment. This group together with VFAs and alcohols contributed from 88% to nearly

100% of the total fluxes. Sulfuric compounds were the fourth group, contributing from nearly 0–9% of the total emissions. Its fractional contribution to the total fluxes generally increased as incubation proceeded, especially in the control treatment, and was generally higher before weaning.

The impact of VOC fluxes on potential odor emissions is illustrated in **Figures 4B,B-1**, through the dynamics of OAVs. The contribution of VFAs to OAVs revealed similar pattern as the contribution of VFAs to the total fluxes. However, alcohols which was the second group to contribute to the total fluxes, was of minor importance in terms of OAV; instead, the phenolic & aromatic group became highly dominant due to the much lower SCOT values associated with the VOCs in this group. Thus, the contribution of VFAs and phenolic & aromatic VOCs to the total OAV accounted for over 90% in almost all cases and ranged up to 99%. The contribution of phenolic & aromatic VOCs to the total OAV became more dominant as incubation proceeded and was more pronounced in the control treatment, whereas

the OAVs derived from VFAs remained more dominant in the pomegranate treatment group, especially after weaning. The fractional contribution of sulfuric VOCs to the total OAV was generally higher in the control vs. pomegranate treatments. Like the total sulfuric fluxes, it also generally increased as incubation proceeded until day 14, and was generally higher before weaning.

# Correlating VOC Emission Profiles and Microbial OTUs Abundance Patterns

**Figure 5** shows the correlation matrix between the emission profiles of the 34 VOCs monitored in this study and the distribution patterns of microbial OTUs obtained by ARISA. Pairwise correlation values between each VOC-OTU combinations were determined according to profile similarity across parallel feces sub-samples including all treatment and incubation times. A reverse pattern of correlation can be observed between certain VOC groups: whereas VFAs, esters and aldehydes are typically positively correlated with a relatively large group of OTUs (right), phenolic & aromatic (phenol, skatole, indole, p-cresol) and sulfuric VOCs are typically negatively correlated with distribution patterns of most OTUs. A small group of OTUs (left) shows a reverse pattern of correlation in comparison to most OTUs (right), with positive correlation with phenolic and sulfuric VOCs but negative correlation with VFAs, esters and aldehydes. Alcoholic compounds are distributed between these two major categories, with n-butanol, iso-pentyl alcohol and 1-pentanol typically positively correlated but other alcohols (iso-Butyl alcohol, 1-hexanol, 1-heptanol 2-phenylethanol, 3-phenyl-1-propanol) typically negatively correlated with most OTU groups. VFAs generally cluster together (except for long chains VFAs; nonanoic and decanoic acids) with close relation to esters and some alcohols. Other VOC groups also cluster together (aldehydes, phenolic, and sulfuric).

# DISCUSSION

The effect of feed additives on growth success of beef cattle calves is primarily estimated by their average daily gain (ADG). Due to the relatively high polyphenol content of pomegranate peel, their undesired effects on animal performance cannot be excluded, when supplemented in the diet. However, in previous studies (Shabtay et al., 2008; Weyl-Feinstein et al., 2014), we did not observe deleterious effects of either fresh pomegranate peel or CPE on fattened and suckling calves growth rate, respectively. In accordance with these previous observations, the CPE in the present study also did not show any negative effect on ADG. Among the various properties tested in fresh fecal samples, OM content, protein, and water soluble sugars did not differ between treatments. Protein is the most expensive component in the diet. While nitrogen is a key element in ruminants diet, its excretion receives increasing attention due to environmental concerns (Islam et al., 2002). Accurate design of protein requirement is thus cardinal for ensuring the balance between an appropriate supply of the assortments of amino acids essential for growth and maintenance, minimizing diet cost and reducing excretion

of excess nitrogen. The findings reported herein imply that CPE had no effect on calves' performance.

Yet, fecal starch content, which was proved as an accurate indicator of total-tract digestibility of starch by feedlot cattle (Zinn et al., 2002; Corona et al., 2005), was significantly higher in the PA as compared with the other three treatments. Interestingly, while at the post-weaning stage fecal starch content was significantly higher in the PA treatment, at the suckling stage it tended to be higher in the CB treatment, indicating that CPE might have a development-related effect on the efficiency of dietary starch digestion. Indeed, although anatomically ruminant neonates possess the same four stomachs as an adult, their digestive metabolic systems function similarly to those of a young monogastric animal, and the rumen becomes metabolically active only at later age, upon consumption of solid feeds (Teagasc, 2017). We thus hypothesize that while CPE facilitates starch digestion from milk at the hindgut, at the suckling stage, it hampers its fermentation in the rumen at the post-weaning stage, presumably due to modulation of rumen microbe population (Jami et al., 2012) which may promote starch to bypass the rumen.

Starch digestibility is a wide-studied theme since it has economic implications due to the high cost of grains and as it can increase production yield and feed efficiency (Firkins et al., 2001). In dairy cows total-tract digestibility of starch ranges from 70 to 100% (Firkins et al., 2001; Ferraretto et al., 2013). The distribution of fecal starch observed in collective studies from 15 trials conducted at the University of Wisconsin-Madison revealed fecal starch that averaged 3% across all trials, with 60 and 36% of the fecal samples ranging from 0 to 3% and from 4 to 8% starch, respectively (Fredin et al., 2014). These results are in agreement with the findings obtained in the current study (**Table 1**).

A linear negative correlation was obtained in the current study between fecal starch concentration and fecal pH (R = −0.539; P = 0.0037); however, the published literature is inconsistent with regard to this relationship. Unlike studies that reported a negative relationship between fecal starch and fecal pH in calves (current study), steers, sheep and growing heifers (Wheeler and Noller, 1976, 1977), other studies have observed either weak or no relationship (Fredin et al., 2014). Fecal pH may be affected by numerous factors other than starch, mostly related to dietary forage concentration and buffer capacity of feed (Erdman et al., 1982; Gressley et al., 2011). These factors were not examined in the present study, however, it is noteworthy that fecal pH, observed herein, obeyed the accepted trend in growing beef calves to decrease with age (Shabtay et al., 2009). Yet, the slightly but significantly lower fecal pH after weaning in the PA treatment, may indicate some effects on ruminal (Jami et al., 2012) and/or post-ruminal microbial fermentation (Gressley et al., 2011).

The dynamics of pH during incubation (**Figure 2**) is presumably affected by the evolution of VFAs and ammonia throughout feces degradation and it was in accordance with the dynamics observed for VFAs. Peak production of VFAs was associated with pH decrease, especially in the PA and to less extent in CA and PB groups (**Figures 3A,B** and **Table 2**). Similar pH dynamics have also been observed in composting studies of livestock manure, whereas the initial drop reflected the production of organic acids and then increase in pH indicated their degradation and the release of ammonia during mineralization of proteins, peptides and amino acid (Atchley and Clark, 1979; Gigliotti et al., 2012; Kim et al., 2016). Ammonia was not measured in the present study, but was likely involved in this pH dynamics. Miller and Varel (2001) who studied the composition of fresh and aged cattle manure suggested that the accumulation of acid products is self-limiting, as low pH inhibits fermentation; they found that production of VFAs from fresh manure was inhibited when the pH fell below 4.5. In the present study, since the pH dropped only to a minimum of 4.77 it did not necessarily inhibit further VFAs production, unless feces samples had reached lower pH, which was not recorded between days 0 and 7.

The production of VFAs as well as other typical anaerobic fermentation byproducts (such as phenolic and sulfuric VOCs) indicates the presence of anaerobic or semi-anaerobic conditions during incubation. Although incubation was not carried out in sealed vessels, it can be assumed that non-optimal aerobic conditions prevailed within the jars, which were aerated only passively through the loosely covering aluminum foil cap. Moreover, previous studies showed that anaerobic byproducts were formed even in actively aerated waste systems (Brinton, 1998, 2006; Beck-Friis et al., 2003; Wu et al., 2010). It was noted by Brinton (2006) that microorganisms producing VFAs under semi-anaerobic conditions are facultative microorganisms and hence, events of oxygen replenishment will not necessarily disturb their presence; moreover, VFAs produced serve as an energy source for consequent aerobic microbial activity, once favorable aerobic conditions return. Depending on aerobic degradation rates (and hence oxygen utilization rate), it was shown that oxygen diffusion is limited to a range of several hundred micrometers around manure particles and thus a substantial portion remains anaerobic even in well aerated systems (Wang et al., 2015). The peak of VFAs shown in the present work is related to the most active degradation phase during which it is most difficult to sufficiently maintain aerobic conditions within aggregates and micro-environments. Then, when degradation rates slow down as the most energetic material is being consumed, the material undergoes mostly aerobic processes and thus VFAs production stops. Our previous study on calve's feces (Shabtay et al., 2009) also showed the production of anaerobic byproducts, both under aerobic and anaerobic systems (VFAs, sulfuric, phenols, and indoles). However, a large reduction of these compounds was shown during 21 days of incubation under aerobic conditions, whereas substantial increase was observed under anaerobic conditions. Nonetheless, in the present study, the general increase of all major VOC emissions during incubation regardless treatment, supports the conclusion that these byproducts were formed during incubation and do not reflect residual VOCs originally existed in the fresh feces.

The order of VOC fluxes, with VFAs and alcohols comprising together 63 to nearly 100% of the total fluxes (**Figures 4A–B**) is in accordance with the observations of Miller and Varel (2001) who found that accumulation of odorous compounds and fermentation products differed between fresh manure and aged samples but was dominated by VFAs and alcohol production. Alcohols were also suggested as major byproducts of incomplete aerobic degradation during composting and were predominant during the active phase (Smet et al., 1999). Compared to VFAs, which were significantly higher in the pomegranate treatment after weaning, alcohols did not vary significantly among the treatments, although their fractional contribution to the total fluxes varied substantially because of the varied VFAs contribution. The third group of phenolic & aromatic VOCs has been the focus of multiple studies due to their characteristic "barn odor." For land-applied swine slurry, Parker et al. (2013) showed that p-cresol accounted for about 80% of the overall OAV, followed by skatole and VFAs. Bulliner et al. (2006) found p-cresol as a key VOC responsible for the overall characteristic swine odor, and Wright et al. (2005) ranked p-cresol as the first VOC odorant at increasing distance from a commercial cattle feedyard. The transformation of VOC patterns into potential odor emissions (as expressed by OAV) showed similar differences between treatments and times of incubations (**Figure 4B**). VFAs most dominantly contributed to the odor impact due to their highest fluxes and taking into consideration their relatively low odor threshold (except acetic acid). Thus, it is evident that the pomegranate treatment after weaning (PA) had the highest odor emission potential. The fractional contribution of each VOCs group to the total OAV (**Figure 4B-1**) shows that the odor emitted from the PA has more VFA characteristics whereas in other cases the odor will be more affected by phenolic type compounds (phenol, p-cresol, indole, skatole). Sulfuric compounds also contribute to all treatments where both phenolic & aromatic as well as sulfuric VOCs have low odor threshold, thus even low concentrations contributed substantially to odor potential. This is in contrast to alcohols; although comprising a substantial fraction of the total VOCs, they had a small contribution due to their relatively high odor threshold.

It is important to note, however, that the OAV approach used in these kind of studies can be biased because of two main reasons: (1) Large efforts are needed to quantitate and monitor all relevant odorous VOCs and possibly some known odorants of extremely low odor thresholds are not resolved due to analytical artifacts. For example, it was shown by Andersen et al. (2012) that it can be difficult to measure methanethiol, since it easily reacts to form dimethyl disulfide during sampling and/or analysis. The analysis of trimethylamine, another potent odorant, also suffers from analytical difficulties, including low precision and sensitivity (Kim and Kim, 2013). Moreover, inorganic gases, mainly hydrogen sulfide and ammonia, which were not included in the present study, are expected to affect odor annoyance. (2) The use of odor threshold values of single compounds much depends on the selected database. This study followed the same approach of Parker et al. (2012) by taking the geometric means of odor thresholds presented in the large compilation of van Gemert (2003). A clear advantage of using large databases is the increased opportunity to approach true values while the use of geometric means reduces the effect of extreme values. A disadvantage may be related to the non-criticizing approach taken by this compilation, which includes values of unclear quality. On the other hand, databases obtained by one approach (e.g., Nagata, 2003; Leonardos et al., 2012), are inevitably biased toward that single approach. Overall, the OAV approach taken in the present as well as in other studies should be considered with caution, yet it assists with the prioritization of specific odorants in different environmental odor mixtures.

With regards to the main odorants monitored in the present study, starch and proteins in manure are considered as the parent sources for VFAs and phenolic VOCs, respectively, where phenol and p-cresol are main products of tyrosine fermentation and indole and skatole are the principal end-products of tryptophan metabolism (Mackie et al., 1998). Yet, observed differences in these VOCs between treatments cannot be attributed to starch and protein contents measured in the fresh feces (**Table 1**). Indeed, starch content was similar in PA and CB, although VFAs production was substantially higher in PA (total VFAs was 73 times higher in PA vs. CB on day 7). Similarly, the emission of phenolic VOCs was substantially higher in CA vs. PA (total phenolic & aromatic was 4.7 times higher in CA vs. PA on day 7) and thus, it cannot be attributed to the similar protein content in these two treatments. Based on the reverse pattern correlations shown in **Figure 5**, it is postulated that different microbial groups (expressed by OTUs) had an associations with different VOC groups. Thus, it could be that the effect of weaning and CPE diet on gastrointestinal microflora was further expressed during incubation by the development of different microbial populations and the resultant fermentation byproducts. This hypothesis deserves more research as it may have implications on ruminant health and production. Enhanced development of short chain fatty acids producing bacteria in CPE diet may be related to the mechanism by which pomegranate additives act against inflammatory diseases. The importance of short chain fatty acids for proper gastrointestinal function has been explored

REFERENCES


in multiple studies on human (Guarner and Malagelada, 2003) and also on ruminants (Guilloteau et al., 2010) and is highly considered in prebiotic formulations (Blaut, 2002). Thus, besides air emissions that are the focus of the present study, the reported observations open new questions related to the physiological mechanisms involved with recent approaches in cattle nutrition that make use of Mediterranean fruit byproducts.

# CONCLUSIONS

Supplementation of post-weaning calves' diet with CPE can alter the emissions of odorants from excreted feces. Major known odorous VOC groups can be affected by the CPE supplement, where increased VFAs in the pomegranate treatment are mostly responsible for such differences. Emissions of other key barn odorants, such as p-cresol, dimethyl disulfide and skatole, may be reduced in the pomegranate treatment; yet, due to their relatively low concentration as compared to VFAs, they only slightly affect the differences in odor potential between pomegranate and control as expressed by OAV. Possible association between VOCs and certain microbial groups in manure may open new questions related to the mechanisms by which pomegranate additives increase ruminants' health. Overall, diet supplementation with CPE may be adopted due to its health and production promoting traits; yet, holistic nutritional-environmental approaches are deserved, taking into consideration possible subsequent changes in manure odor characteristics and increased feedlot nuisance if not managed properly.

# AUTHOR CONTRIBUTIONS

VV analyzed the VOC data and wrote a full draft of the manuscript. AS led and coordinated the animals experiment and all animal health aspects and contributed to the writing of the manuscript. MY treated the feces samples and performed VOC sampling, TD-GC-MS analyses and DNA extractions. IM and NS conducted the ARISA analyses. SF and SM analyzed the ARISA data, RA managed the animals experiment. YL coordinated all VOC sampling, analyses and data treatment, and took a leading role in manuscript preparation.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fsufs. 2018.00033/full#supplementary-material


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Varma, Shabtay, Yishay, Mizrahi, Shterzer, Freilich, Medina, Agmon and Laor. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Identifying Urine Patches on Intensively Managed Grassland Using Aerial Imagery Captured From Remotely Piloted Aircraft Systems

Juliette Maire1,2,3,4 \*, Simon Gibson-Poole2,3, Nicholas Cowan<sup>4</sup> , Dave S. Reay <sup>3</sup> , Karl G. Richards <sup>1</sup> , Ute Skiba<sup>4</sup> , Robert M. Rees <sup>2</sup> and Gary J. Lanigan<sup>1</sup>

*<sup>1</sup> Soils and Land Use Department, Teagasc, Wexford, Ireland, <sup>2</sup> Future Farming Systems, Scotland's Rural College, Edinburgh, United Kingdom, <sup>3</sup> School of Geosciences, University of Edinburgh, Edinburgh, United Kingdom, <sup>4</sup> Atmospheric Chemistry and Effects, Centre for Ecology and Hydrology, Penicuik, United Kingdom*

### Edited by:

*Tom Misselbrook, Rothamsted Research (BBSRC), United Kingdom*

### Reviewed by:

*Paul Harris, Rothamsted Research (BBSRC), United Kingdom Harald Menzi, Federal Office for the Environment, Switzerland*

\*Correspondence:

*Juliette Maire Juliette.Maire@sruc.ac.uk*

### Specialty section:

*This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems*

> Received: *01 February 2018* Accepted: *05 April 2018* Published: *25 April 2018*

### Citation:

*Maire J, Gibson-Poole S, Cowan N, Reay DS, Richards KG, Skiba U, Rees RM and Lanigan GJ (2018) Identifying Urine Patches on Intensively Managed Grassland Using Aerial Imagery Captured From Remotely Piloted Aircraft Systems. Front. Sustain. Food Syst. 2:10. doi: 10.3389/fsufs.2018.00010* The deposition of livestock urine and feces in grazed fields results in a sizable input of available nitrogen (N) in these soils; therefore significantly increasing potential nitrogen pollution from agricultural areas in the form of nitrous oxide (N2O), ammonia (NH3), and nitrate (NO<sup>3</sup> <sup>−</sup>). Livestock deposition events contributes to high spatial variability within the field and generate uncertainties when assessing the contribution that animal waste has on nitrogen pollution pathways. This study investigated an innovative technique for identifying the spatial coverage of urine deposition in grasslands without the need for manual soil measurements. A Remotely Piloted Aircraft System (RPAS) using a twin camera system was used to identify urine patches in a 5 ha field, which had been grazed by sheep 3 weeks previous to measurements. The imagery was processed using Agisoft Photoscan (Agisoft LLC) to produce true and false color orthomosaic imagery of the entire field. Imagery of five areas (225 m<sup>2</sup> ) within the field were analyzed using a custom R script. For a total of 1,125 m<sup>2</sup> of grassland, 12.2% of the area consisted of what was classified as urine patch. A simple up-scaling method was applied to these data to calculate N2O emissions for the entire field providing an estimate of 1.3–2.0 kg N2O-N ha−<sup>1</sup> emissions from urine and fertilizer inputs.

Keywords: RPAS, UAV, image analysis, feature detection, urine, nitrous oxide, grassland

# 1. INTRODUCTION

In order to improve Nutrient Use Efficiency (NUE) and reduce unnecessary losses in the food supply chain, management of nutrients in agricultural systems has to be considered in its entirety. One potential solution to improve NUE is to use precision farming techniques which take into account the spatial heterogeneity and temporal variability of nutrients already present at the field scale before further fertilizers are applied (Mulla, 2012; Hedley, 2014). In the context of precision agriculture in intensively grazed grassland management, it is of particular interest to study excreta deposited by grazing animals. Nutrient losses from livestock at the field scale are difficult to assess in full due to the randomness of the deposition of urine and dung from grazing animals (Auerswald and Mayer, 2010; Cowan et al., 2015). Past research has focused primarily upon the overall control and management of livestock waste and its impacts on the environment, grass production, and soil quality (Boon et al., 2014; Selbie et al., 2015; Hyde et al., 2016) with little attention paid to spatial heterogeneity.

In the case of sheep urine, the nitrogen (N) content is reported in the literature only sporadically and the uncertainties about these values are large as shown in the meta-analysis reported by Selbie et al. (2015). The nitrogen loading was reported to vary from 500 to 1,089 kg N ha−<sup>1</sup> for sheep urine deposits based on the findings that the urine contains 5–10 g N L−<sup>1</sup> and a volume per urination of 0.5 L would cover an area of 0.03–0.05 m<sup>2</sup> (Haynes and Williams, 1993). In our study, using these values, the amount of urine N deposited represents an equivalent of 2.0–4.8 times the annual amount of N fertilizer (225 kg N ha−<sup>1</sup> ). This excess of applied nitrogen leads to an increased likelihood of N leaching, ammonia (NH3) volatilization, and nitrous oxide (N2O) emission, but also increased grass growth (Hyde et al., 2016; Marsden et al., 2016), as well as increasing nitrogen and carbon pools in the soil urine depositions change soil pH, soil surface temperature, and soil moisture content (Marriott et al., 1987; Moir et al., 2011; Boon et al., 2014; Selbie et al., 2015). All factors are likely to change the N2O emission rate (Clough et al., 2004; Hoogendoorn et al., 2008; De Klein et al., 2014).

Typical apportionment values for deposited sheep urine are estimated as 13% NH<sup>3</sup> volatilization; 2% N2O emission; 20% NO<sup>3</sup> <sup>−</sup> leaching; 41% pasture uptake and 26% gross immobilization of the total deposited urinary nitrogen (Selbie et al., 2015). Monitoring deposited urine in soils is difficult due to the fact that the urine itself is not directly visible. However, urine patches do have visible consequences for the grass growth, most noticeably color and density of the leaves (Dennis et al., 2011). These properties may serve as a useful proxy for tracking urine deposition, but the response in grass growth to urine deposition also depends strongly on soil type, soil moisture content, seasonal and climatic conditions and the nitrogen content of the urine (Clough et al., 2004).

Existing urine deposition detection methods include simple visual observations of variations in vegetation growth and color (Auerswald and Mayer, 2010). Recent advancements in technology have made it possible to detect urine patches by fitting the grazing animals with GPS collars and thermal sensors (Betteridge et al., 2010), or with cameras footage or urine sensors (Misselbrook et al., 2016); all of which typically require considerable investment in time, human and material resources. These methods are usually either only effective over small areas, a small number of grazing animals or require installing sensors on the animals. Moreover, post-grazing methods to detect urine patches manually or electronically are limited in their ability to capture reliable and good quality data (Misselbrook et al., 2016). The method developed in this study could help to mitigate these issues by increasing the frequency of observations, allowing the study of overlapping urination during the same or different grazing events. In other words, a low-cost, high-frequency, non-destructive method that is easy to deploy in the field is required.

In agricultural studies, practices that make use of remote sensing technologies have been widely developed to map a variety of spatial factors such as crop production estimation (Jensen et al., 2007; Hunt et al., 2010), grass nutrient content (Capolupo et al., 2015; Pullanagari et al., 2016), weed distribution (Jensen et al., 2003), soil spatial variability mapping (Stoorvogel et al., 2015), and diseased or damaged crops (Mirik et al., 2006). Remotely Piloted Aircraft Systems (RPAS) can fly at low altitude allowing acquisition of high spatial resolution imagery to observe small individual objects, such as grass patches, and can be deployed even in cloudy conditions for which the acquisition of satellite imagery or helicopter videography become difficult. The effort required to deploy an RPAS platform has greatly reduced in recent years, contributing in some cases to more flexible and affordable experimentation than with other aerial image acquisition systems (Zhang and Kovacs, 2012). Use of other remote sensing techniques (e.g., piloted aircraft, helicopter, satellite platforms) can be limited in its ability to provide adequate field-scale image acquisition, image quality, and spatial and temporal resolutions partly due to cost and sensitivity to weather conditions (Dennis et al., 2013; Ali et al., 2016; Lopes et al., 2017).

In the case of urine patch detection and grass quality studies, good preliminary results have already been obtained using aerial or ground-based imagery (Moir et al., 2011; Dennis et al., 2013; Roten et al., 2017). Nevertheless, development of automated preand post-processing of images covering large areas, enabling automated detection of patches, is still required. The challenge of automating patch-detection presents complex difficulties, such as the light variability effect on similar reflectance properties, the requirement of a high-resolution image, the identification and removal of unwanted plants and object reflectance interfering with the detection. Recently, numerous approaches have been developed to perform feature or land-cover detection on images from satellite imagery (Sammouda et al., 2014), phenology cameras (Filippa et al., 2016), microscopic or X-ray imagery, and remote sensing imagery from RPAS (Hunt et al., 2010; Mulla, 2012; Capolupo et al., 2015). For high resolution remotely sensed imagery (where image pixels are much smaller than the objects to be identified), an object based image analysis (OBIA) technique is more appropriate to use compared to a pixel based approach (Blaschke, 2010). Commonly used software packages that use OBIA techniques include eCognition (Gupta and Bhadauria, 2014) or python scripts combined with OpenCV, however, these programs can carry expensive licenses or may not be user-friendly for most environmental and agricultural science researchers.

The method developed in this study is a remote sensingbased approach, aimed at enabling the collection of a large number of urination events at numerous times in an automated way (Mulla, 2012). This method is based on grass growth response and does not measure the area over which the urine has been deposited (wetted area), but considers the effective area (Buckthought et al., 2016). The effective area of a urine patch includes the wetted area, the diffusional area and the pasture response area. The wetted area has been distinguished from the diffusional and pasture response area which incorporates the diffusive edge of the nutrients and the plants able to access, via their roots, these nutrients (Marsden et al., 2016). Often N2O emission estimates of urine patches focus on the wetted area only and do not account for the diffusional areas (Williams and Haynes, 1994; Hoogendoorn et al., 2008). By measuring nitrogen input from urination and spatially determine their locations, the development of this method has the potential to help farmers to control their fertilizer management, improving NUE, and reducing associated N pollution to the environment.

The objective of this study was to evaluate the potential and the limitations of using a combined tool of RPAS orthoimagery and a script written in R (R Development Core Team, 2016) to allow feature detection. The aim was to provide an efficient tool to map urine patch coverage over grazed grassland in order to improve N2O estimates at the field scale and to better explain field soil spatial variability.

# 2. MATERIALS AND METHODS

Urine patch detection was undertaken by: (1) Collection of pictures in the field using a RPAS; (2) Stitching the collected pictures together to obtain an orthoimage of the entire surveyed area; (3) Automated identification of the urine patches from the pictures; (4) Aggregation of detected urine patch data. This stepwise method was designed to allow the characterization of field scale urine deposition coverage, size, and color.

# 2.1. Remotely Piloted Aircraft Systems (RPAS) and on Board Camera System

The RPAS used in this study was a custom-built, eight-motor multi-rotor system housed in a 1 m diameter Vulcan octocopter frame (VulcanUAV, Mitcheldean, UK; **Figure 1**), controlled via a 3DR Pixhawk autopilot running Arducopter (v3.2.1) firmware (3D Robotics, Berkeley, USA). The autopilot contained an inertial measurement unit, a barometer, a magnetometer and an external GPS for navigation. The RPAS was powered by two 14.8 V, 10,000 mAh lithium polymer batteries which provide a flight time of ∼14 min whilst carrying the dual camera payload of ∼320 g. The dual camera system was housed in a stabilized gimbal and contained two Canon A2200 point and shoot cameras (Canon, Tokyo, Japan). One of the cameras was unmodified, giving a typical red, green, blue (RGB) image, and one was modified to sense near infra-red (NIR) wavelengths of light through the removal of its internal NIR filter and the addition of an acrylic 585 nm long pass filter (Knight Optical, Harrietsham, UK).

FIGURE 1 | (A) Octocopter used for this project mounted with an unmodified (B, left) and a modified Canon A2200 measuring visible light and near infra-red light (B, right).

The spectral sensitivity of the cameras was tested to identify their spectral characteristics, revealing that for the modified camera, NIR was captured across all channels with the blue channel showing the purest signal (Berra et al., 2015). Both cameras used the Canon Hack Development Kit (CHDK) modified firmware (v1.2) and the KAP UAV exposure control script (v3.1) to enable RAW imagery to be acquired when commanded via the autopilot. The script also allows the shutter speed and ISO to vary within a specified range (1/200 to 1/2,000 s and 200 to 400 ISO, respectively). The internal neutral density filter was not used. The aperture (f 2.8) and zoom level (default) were fixed with focus set to infinity and the white balance was calibrated against a gray card before the flight to provide reliable visual results.

### 2.2. Unmanned Aerial Survey

On the 6th of June 2016, four flights were operated to survey the entire field, from two take-off positions which ran perpendicular to the slope of the field to maintain an altitude of 35 m above ground level. The images captured during the four flights were then considered as one dataset. All flights used pre-programmed automatic waypoints facilitated by Mission Planner (http:// ardupilot.org/planner) to ensure an image overlap of 60% and a side overlap of 80% in order to optimize the image stitching. The flight speed was 2 m s−<sup>1</sup> to allow for the camera system to capture images at the rate of one image every ∼6 s. Georectification of the imagery was performed by surveying the center of twelve fixed collars (used for static chambers measurements) distributed within the field using a Piksi (Swift Navigation, San Francisco, USA) real-time kinematic GPS with an expected accuracy of ±13 cm (**Figure 2**).

### 2.3. Field Site

The survey was conducted over a 5 ha intensively managed grazed grassland ∼10 km South of Edinburgh, 190 m above sea level (3◦ 12′W, 55◦ 52′N) over the period March 2016 to June 2016 (Jones et al., 2017). The field is predominantly grazed by sheep, which is annually grazed at 0.7 livestock unit (LSU) per hectare rate. Before the RPAS survey, the field was grazed from the third week of March 2016 to mid-May 2016 (7 weeks), by 100 ewes and was fertilized in early April 2016 with 69 kg of N ha−<sup>1</sup> in the form of urea. After the grazing period, no animals were present in the field. The field was harvested mid-July 2016 and 12 sub-samples were collected for a total area of 1.5 m<sup>2</sup> . The dry matter (DM) grass yield was 8.0 tone ha−<sup>1</sup> with a dry matter content of 243 g kg−<sup>1</sup> . The crude protein content was an average of 79.3 g kg−<sup>1</sup> DM and the metabolized energy 10.8 MJ kg−<sup>1</sup> DM. From the same subsamples, the DM grass yield was found to be significantly different between patch areas, areas where urine was deposited and areas visually not affected by urine (one-way ANOVA, p = 0.0015, n = 12) with an average of 9.5 and 6.0 tone ha−<sup>1</sup> , respectively. The protein content and the metabolized energy of the grass did not show significant differences between the two areas of grass visually assessed as patch area or not affected area. The field consists of an imperfectly drained MacMerry soil series, Rowanhill soil association (Eutric Cambisol) with

a pH (in H2O) of 5.1 and a clay fraction of 20–26% (Jones et al., 2011). The main grass species is Italian ryegrass (Lolium perenne). The long-term average annual rainfall (1981–2010) at this site is 980 mm and the mean daily temperature is 18.8 ◦C in summer (July) and 5.6 ◦C in winter (January) (Jones et al., 2017).

# 2.4. Pre-processing Using Agisoft Photoscan

The RGB and NIR images were initially processed to remove erroneous pixels using the Canon Hack Development Kit CHDK (http://chdk.wikia.com/wiki/PTP\_Extension), followed by further processing using a custom script in ImageJ (Schindelin et al., 2013) converted each image to a 16 bit linear tagged image file format (TIFF) file (white balance set to 1, no gamma correction) using DCRAW software (Coffin, 2016), which utilized a dark image of the same ISO and shutter speed in order to reduce dark current signal noise. Each image was then smoothed using an ImageJs despeckle filter to further remove noise before the PTlens software (T.Niemann, Portland, Oregon, USA) was used to correct lens and edges distortion. The RGB images were processed a second time to produce a better visual set of data in 16 bit TIFF format (gamma corrected, white balance as set for each flight, utilizing highlight recovery options) and sharpened using ImageJs sharpen filter. The TIFF files were geotagged using the GPS information from the RPAS flight log and were then processed using Agisoft Photoscan (Agisoft LLC), using high settings (Highest alignment, High Dense cloud with mild depth filtering) to produce a georeferenced orthomosaic for each dataset: RGB and NIR (**Figure 2**).

### 2.5. Detection of Urine Patches Algorithm 2.5.1. Algorithm Step 1: Cropping of the Orthoimages and NDVI Calculation

The RGB and NIR orthimages of the surveyed field were stacked on a raster layer then clipped to select smaller areas resulting in a more manageable file sizes of 15 by 15 m of grassland (1107 by 1107 pixels, format .tiff, 20.6 MB, resolution of 1.84 cm<sup>2</sup> pixel) (**Figure 2**).

$$NDVI = \frac{NIR \, - \, Red}{NIR \, + \, Red} \tag{1}$$

Red and NIR stands for the red reflectance and near-infrared reflectance.

The NIR images allowed the calculation of the Normalized Difference Vegetation Index (NDVI) based on the data from the blue channel of the modified camera (to give NIR) and the red channel of the un-modified camera (to give the red). NDVI is a ratio using red and NIR reflectance to highlight photosynthesis (Equation 1). NDVI varies between −1.0 and +1.0 and is mostly used for satellite pictures due to its link with differences in vegetation type, biomass and photosynthetic potential. NDVI is commonly used for feature detection in vegetation environments (Jensen et al., 2003; Hunt et al., 2010; Mulla, 2012). Red and NIR cropped images were also studied in addition to NDVI cropped

images to estimate the efficiency of using NDVI to supplement the use of either Red or only NIR images.

### 2.5.2. Algorithms Step 2: Pixel Clustering

To detect urine patches in each picture the clustering method based on pixel segmentation was chosen to be applied to the NDVI raster layer created from the RBG and NIR images. Clustering is the task of grouping a set of pixels in a way that pixels of the same group called a cluster (K) are more similar in term of color characteristics, to each other than to those of the other groups. In this study, the algorithm was written to perform an unsupervised classification using K-means clustering method (Jain, 2010) on each pixel of the NDVI layer. This method is designed to handle large datasets and follows four consecutive steps for each cluster:


The algorithm implemented in R was developed by Hartigan and Wong (1979) for the purpose of partitioning data points into k groups to minimize the distance from the data points to the cluster centroid. In other words, in the Lloyd's algorithm (Equation 2), for each iteration, each pixel is assigned to the cluster with the smallest value of:

$$\text{SS}(K) = \sum\_{i=1}^{n} \sum\_{\boldsymbol{x} \in c\_{i}} (\boldsymbol{x}\_{i} - \boldsymbol{\mu}\_{i})^{2} \tag{2}$$

Where n is the number of pixels, K the given cluster, i is the pixel considered, c<sup>i</sup> is the set of pixel that belong to the cluster k and x<sup>i</sup> − µ<sup>i</sup> the Euclidean distance between the pixel i and the centroid of the cluster K. The selection of the cluster first centroid is normally randomized inside the whole image. But in this study, the starting point to the K-means method was arbitrary set to ensure that the results would be the same if the process was to be repeated. The clustering was performed for a set number of clusters per image that needed to be predetermined. For this purpose, the elbow method (**Figure 3**) which is a hierarchical cluster analysis was performed using a set of dissimilarities for the number of objects (n) being clustered. The method selected was the Ward's minimum variance method (Ward, 1963) that allows the identification of compact, spherical clusters. Through this method, the optimal number of clusters was found to be 4 for the 15 by 15 m squares of grassland.

### 2.5.3. Algorithms Step 3: Isolation of Each Urine Patch

The next step was to isolate the urine patches from each other. For each image, inside the cluster corresponding to the urine patch, the connected adjacent pixels were grouped together to form a patch. For this step, a virtual window of 9 by 9 pixels was created to screen the whole image to remove small groups of pixels which were noise from the clustering step. Then using the same method, gaps inside the same patch were dissolved and pixels belonging to the same patch were connected. Before moving to the next step, the function rasterToPolygon() (from the R package "raster," https://www.rdocumentation.org/ packages/raster/versions/2.6-7) was used to convert each patch as a polygon.

### 2.5.4. Algorithms Step 4: Patch Selection and Calculation of Their Characteristics

To avoid the detection of unwanted objects such as weed patches, small shadows and groups of denser grass, objects <300 cm<sup>2</sup> were discarded, which correspond at the minimal potential size of a urine patch (Selbie et al., 2015; Marsden et al., 2016). Finally, the patch characteristics such as size, centroid coordinates, patch average color values, and shape index (giving an information on the shape of the patch) were calculated and converted to square meters. These values were used to estimate the total coverage of urine patches at the field scale. A step-by-step synthetic diagram of the script is provided in **Figure 3**.

# 3. RESULTS AND DISCUSSION

### 3.1. RPAS and Image Stitching Limitations

The orthoimages (i.e., RGB and NIR) obtained from the RPAS survey undertaken on the 6th of June 2016, were generated using Agisoft Photoscan (Agisoft LLC). The **Figure 2** shows the output of this software: RBG orthoimage and NIR adding the red digital channel and the Digital Elevation Model (DEM) reprensenting the elevation from the sea level. The stitching of the images captured by the RPAS into an orthoimage is necessary to create an image appearing as though it was taken from a uniform altitude, a rectilinear lens, with limiting edge distortions and with accurate details. The image stitching software is limited by the quality of the pictures captured and the weather conditions (e.g., influenced by light and wind speed). The quality of the camera can also be problematic in some cases. To account for the images distorsions, the surveyed area must be at least 10 m wider than the actual study area. Moreover, color calibration of the pictures is required to enable time series monitoring and the comparison between fields. The proximity and the size of the urine patch deposition required a high pixel resolution but also required the images to be as close as possible to true-colors to ensure accurate patch detection. A more detailed review on the challenges and limitations of using RPAS over grassland environments is presented in Von Bueren et al. (2015).

## 3.2. Image Segmentation Using K-Means Algorithm

Orthoimages of the whole field were cropped to 225 m<sup>2</sup> squares (15 by 15 m) of grassland (**Figure 2**). Five of these cropped images, corresponding to locations close to the middle of the field surveyed, were processed using an R script (see location of the first square in **Figure 2**) as a proof of concept.

To automate the patch detection, the K-means method (Hartigan and Wong, 1979), commonly used for image segmentation (Lopes et al., 2017; Singh and Misra, 2017), was

implemented in the R script. The advantage of this algorithm is that it has a low computational complexity, it is an unsupervised learning mechanism and the resulted clusters of this method are not overlapping. This method was able to detect the urine patches on image with high color similarities, patches in close proximity, and overlapping. However, the K-means methods can work efficiently only if the optimal number of clusters is correctly determined. In this script, the elbow method (Strobl et al., 2017) was used to determine that four clusters were the optimal number of clusters required (**Figure 3**). The Kmeans results (**Figure 4B**) were compared visually to the RGB images (**Figure 4A**) to access the certainty of patch detection. The specific cluster corresponding to the urine patch was allocated visually and processed using custom functions to isolate each patch (**Figures 4C,D**). This step is the limitating step in terms of computational complexity, and therefore it is the slowest step in the process (**Table 1**). An object-based detection instead of a pixel-based method could improve the efficiency of this step but would require a supervised initial classification of some of the areas by the script-user (Rastner et al., 2014). Other features may be mistakenly be labeled as patches when in reality they are weed patches, fence poles, or tractors tracks. These issues have not been assessed in this study due to the fact that the study area did not contained any of these items, but it will be important to include a correction in a future version of this script.

The typical wetted area of the sheep urine patch, based on field measurements, is estimated to be between 300 and 500 cm<sup>2</sup> (Selbie et al., 2015; Marsden et al., 2016). Additionally, the effective area of a sheep urine deposition has been shown to not exceed 20 cm beyond the wetted area (Ducau et al., 2003). Therefore, the total area of the visible patch can range between 400 and 1,600 cm<sup>2</sup> (Marsden et al., 2016). The effective area may vary with the volume, the urine nitrogen concentration, soil texture, soil moisture content as well as the topography of the area, vegetation type, and root architecture (Haynes and Williams, 1993; Dennis et al., 2011). Using this information, the script was written to select effective patch areas larger than 300 cm<sup>2</sup> .

### 3.3. Validation of the Use of NDVI

NDVI used in this study was chosen for the analysis based on references belonging to other scientific disciplines such as satellite images analysis (Zha et al., 2001; Colombo et al., 2003), feature detection for self-driving cars (Cho et al., 2014) and other RPAS studies as described in the section 1. It was chosen in

FIGURE 4 | Examples of results from urine patch detection script on a 15 by 15 m square of grassland (example 1 in the Table 3). (A) RGB image, (B) K-means clustering results, (C) selected cluster, and (D) patch isolation results.

this study due to its capability to detect the small differences of the green and red spectra inside the images (Rasmussen et al., 2016).

The NDVI is based on Red digital numbers and NIR reflectance values which required to modify the camera to measure NIR. Multispectral sensors such as the parrot sequoia or red edges (https://www.parrot.com/us/business-solutions/ parrot-sequoia), are often used to measure NIR values. These devices are about five times more expensive than the modified Canon camera used in this project. Moreover, the necessity of having an extra device on the RPAS would have increased payload capacity of the RPAS and cost. The downside of having two different cameras to ensure that each pixel in both orthoimages are corresponding to the exact same location in the field. For this task, the images have been georeferenced using GPS data collected in the field.

The next step was to identify if the Red digital numbers or the NIR data alone could differentiate efficiently the area affected by urine deposition to the rest of the field. After running the script using the NDVI data, a t-test has been used to compare the values of pixels allocated to the patch area and the ones allocated to the "non-patch" area. The Fvalue and the R 2 from the t-tests were used to compare the performance of the color indices (**Figure 5**). The difference between pixels allocated to the patch and non-patch area was significant for the three indices (Red, NIR, NDVI). Nonetheless, in this study, Red and NIR values for pixels allocated to urine patches were significantly different than the non-patch pixels. However, from the F-stats and R 2 values, it is clear that NIR and Red digital numbers did not perform as well as the NDVI values (**Table 2**).

## 3.4. Urine Patch Coverage and Characteristics

The detection of individual urine patches using the RPAS during four flights on one single day allowed us to calculate the area covered by patches over a sub-section of the field of 1,125 m<sup>2</sup> (**Table 3**), which was 12.2 ± 2.2 %. This coverage value has been used as an estimate of the whole field coverage which corresponds to an area covered by urine patch of 0.7 ha for the 5 ha field. This value is at the low end of urine patch estimation compared to repeated estimates (14–31%) over a year of urine depositions from repetition grazed cattle using field observations or RPAS imagery and grazed by cattle (Moir et al., 2011; Dennis et al., 2013; Selbie et al., 2015). This difference is likely due to the short grazing period before the survey (7 weeks), smaller animals (sheep) as well as as well as the urine nutrient content difference (Kelliher et al., 2014). To evaluate the annual

TABLE 1 | Average processing time of the R script for five squares of grassland (225 m<sup>2</sup> each) expressed in seconds ± standard deviation.


TABLE 2 | Summary of the Student's *t*-test performed to test the difference in mean values between pixels belonging to patch and pixel not belonging to patch (non-patch) for the three color indices Red digital number (Red), Near infra-red (NIR), and Normalized Difference Vegetation Index (NDVI).


coverage in our study, it would be necessary to repeat the survey regularly throughout the year. Moir et al. (2011)'s experiment was conducted for 4 years (2003–2007) where the urine patches were identified visually in the spring, summer, and autumn periods each year. This identification was time-consuming and took 12 weeks for each season and was considered to be relevant for the previous 3 months of urine deposition. By comparison, RPAS survey could be undertaken weekly, pre- and post- grazing and will generally take <2 h.

In this study, every grass patch detected using the RPAS survey was considered as a urine patch. However, dung patches are likely to form grass patches unde specific conditions of accumulation of sheep dung. In contrast to cattle dung which forms a grass patch of >50 cm<sup>2</sup> after degradation of the dung (taking up to 12 months), sheep dung is in form of pellets scattered over large areas and are unlikely to generate a discernible pasture response (Williams and Haynes, 1995). Moreover, within the period between the grazing ends and the RPAS survey the sheep dung depositions were probably fully degraded (Williams and Haynes, 1995). Therefore, dung deposition was not visible in the output images. For these reasons, grass patches visible in the output image have been assumed to be due to urine and not dung depositions. In the case of potential prior excreta depositions, in this study, the previous grazing event was over 5 months prior to the grazing period studied. Therefore it was unlikely that previous depositions were visible on the RPAS survey images.

# 3.5. Estimation of N2O Emissions From Patches at the Field Scale

In this study, the urine patch coverage was used to estimate urine and fertilizer induced N2O emissions. A homogeneous amount of urine deposited on the patch has been assumed to calculate the total N2O emissions of the studied field. The calculations have been based on the national greenhouse gas inventory methodology (De Klein et al., 2006). The emission factor of mineral nitrogen fertilizer application and of urine deposition is 1%, i.e., 1% of the nitrogen applied is emitted as N2O. During the period of study, a treatment of 69 kg of N ha−<sup>1</sup> was applied and we estimated that 12.2% of the field was covered by urine patches. The amount of nitrogen in sheep urine is required to use the emission factor but this is poorly reported in the literature (De Klein et al., 2014; Hyde et al., 2016; Marsden et al., 2016). For these calculations, the data summarized by Selbie et al. (2015) on sheep grazing urine N content were considered. The N loading for sheep urine was ranged from 500 to 1,089 kg N ha−<sup>1</sup> . An average value of 800 kg N ha−<sup>1</sup> was assumed. From these data, the total emissions of N2O from both urine and N fertilizer application was estimated at 1.3 and 2.0 kg N2O-N ha−<sup>1</sup> for the period of grazing studied (**Table 4**). The emissions from dung depositions and the potentially combined effect of urine and fertilizer were not considered.

To provide an order of magnitude, the contributions of the N2O emissions from 7 weeks of grazing of 100 ewes and from the mineral fertilizer applied during the same period have been determined. The urine depositions from the grazing ewes contribute of 47–66% of the total N2O emissions (**Table 4**). From these estimations, emissions induced by urine deposition are not

TABLE 3 | Results of the application of the urine patch detection script.


*Urine patch coverage estimation in square meter and percentage of the total area considered using the R script of five squares of grassland of 225 m<sup>2</sup> (1–5) with an image resolution of 1.84 cm.pixel*−*<sup>1</sup> and the sum of the five squares (1,125 m<sup>2</sup> ).*

TABLE 4 | Results from up-scaling of the urine deposition and fertilizer application N2O emissions using the IPCC emission factor of 1% and the value of urine patch coverage estimated over a 7 week period (12.2%).


negligible compared to the N2O emissions induced by fertilizer application.

To improve the up-scaling of the emissions, more specific emission factors are required for sheep urine patches (De Klein et al., 2014; Marsden et al., 2016), as they vary with season, soil properties (texture, pH, moisture content), and as it has been done for cattle urine deposition (Clough et al., 2004; Boon et al., 2014; Krol et al., 2016; Minet et al., 2016).

These calculations were based on the percentage of urine patch coverage calculated previously and so, on the effective area of the urine deposition. The difference in emissions between the wetted and the effective areas was assessed for cattle urine by Marsden et al. (2016). They concluded that the cattle urine patch diffusional area is an extremely important source of emissions from urine patches and needs to be considered when measuring EFs. This study justified using the total area affected by urine deposition instead of the wetted area for N2O emissions calculation.

These results (**Table 4**) are an example of how the urine patch coverage can be used to improve our understanding and estimation of emissions. Furthermore, the automated detection of urine coverage can improve model validation when compared with field data, upscaling from individual patches to field scale as well as allowing the consideration for temporal changes of the emissions.

### 3.6. R Script Efficiency

This study has led to the production of a script written in the R software. This software is largely used in the scientific community as a statistical tool but more and more researchers are using it for spatial and image analysis. OBIA techniques can be implemented to optimize the processing time and increase the accuracy of the detection (Blaschke, 2010). The R-package currently under further development, will allow researchers working in this area to easily replicate a similar analysis. For a 225 m<sup>2</sup> square of grassland, the script will take 48.4 s to process (on a computer with limited power capacity, 4GB RAM, processor Intel <sup>R</sup> , CoreTM, i5-5200 CPU, 2.20 GHz). Each step of the script time processing is shown in **Table 1**. While considering 1 ha of grassland, the script takes about 35 min to run (**Figure 3**). This accounts for the image segmentation, clustering, the calculation of the parameters of each grass patch and plotting the results, but does not include the generation of the orthomosaic. In future, it will be important to expand this code to run with object-based detection software such as eCognition by Trimble or ArcGIS software by Esri. This would prevent multiple counting or miscounting of patches at the edges of the smaller images by processing the entire field image at once. It will also allow the analysis of larger datasets, such farm scale or entire grazing period datasets.

### 4. CONCLUSIONS

In this study, RPAS and R image analysis have proven to be effective when carrying out high-resolution, non-destructive, near real-time, and low-cost assessment of the size and distribution of urine patches from aerial surveys. This process has been automated and kept unsupervised. The process is based on R software which gives the opportunity to researchers to easily adapt this script to their research purposes as well as directly using it for urine patch coverage estimation. An outcome of this study is the R package which facilitates easy and quick processing of the orthoimages collected with a RPAS. The script efficiency has shown promise for analyzing small and homogenous areas which seem to work sufficiently for plot-based experiments or individual occasions. However, for long-term monitoring of grazing and management impacts on grassland, a more efficient software would be required. Using low-cost RPAS, onboard cameras and an open source software, this method offers new perspectives for nutrient management, precision agriculture, and greenhouse gas emissions estimation in grassland systems.

# AUTHOR CONTRIBUTIONS

JM and SG-P designed the method, developed the image analysis approach and conducted data collection. JM wrote the paper with contributions from all co-authors.

### ACKNOWLEDGMENTS

The authors gratefully acknowledge the University of Edinburgh farm manager Wim Bosma for allowing access to the Easter Bush study field. Valuable assistance was also provided by the Biomathematics and Statistics Scotland, BioS. Funding for

### REFERENCES


this study was provided by the Walsh fellowship program by Teagasc, Ireland and the BBSRC–Newton project UK-China Virtual Joint Centre for Improved Nitrogen Agronomy (CINAG). The authors are grateful to the two reviewers and the editor their suggestions that have improved the manuscript.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer PH and handling Editor declared their shared affiliation.

Copyright © 2018 Maire, Gibson-Poole, Cowan, Reay, Richards, Skiba, Rees and Lanigan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Bio-Hydrogen Production From Buffalo Waste With Rumen Inoculum and Metagenomic Characterization of Bacterial and Archaeal Community

### Antonella Chiariotti\* and Alessandra Crisà

CREA Research Center for Animal Production and Aquaculture, Monterotondo, Italy

Edited by:

Claudia Wagner-Riddle, University of Guelph, Canada

### Reviewed by:

Brandon Gilroyed, University of Guelph, Canada Jaak Truu, University of Tartu, Estonia

\*Correspondence: Antonella Chiariotti antonella.chiariotti@crea.gov.it

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 26 January 2018 Accepted: 10 April 2018 Published: 02 May 2018

### Citation:

Chiariotti A and Crisà A (2018) Bio-Hydrogen Production From Buffalo Waste With Rumen Inoculum and Metagenomic Characterization of Bacterial and Archaeal Community. Front. Sustain. Food Syst. 2:13. doi: 10.3389/fsufs.2018.00013 Biogas production from agricultural and industrial wastes delivers two benefits: on one side, the treatment of organic residues prevents the environmental and economic impact of their disposal; on the other side, methane and/or hydrogen are generated. The aims of this study were both to produce bio-hydrogen from buffalo wastes and to investigate the relationship between biogas production and bacterial and archaeal community composition. Anaerobic codigestion of livestock by-products (buffalo sludge and low protein cheese whey-scotta), with buffalo rumen and buffalo sludge as inoculum, was performed. The microbial community was analyzed using next-generation sequencing of 16S rRNA gene amplicons. Codigestion showed to be positive because of both sludge buffering capability and highly degradable carbohydrates content in scotta. Rumen inoculum proved more efficient compared to sludge during fermentation. In fact, cumulated production was higher (120.8 vs. 65.4 ml H<sup>2</sup> g VS−<sup>1</sup> respectively) and the average percentage of hydrogen in biogas was 48.1 (v/v) with maximum peak at 64.6. Moreover, rumen bacterial profile showed higher genera richness. Taxonomic classification showed that among the bacteria, Firmicutes, 23.3% of whom Clostridia; Bacteroidetes, and in particular Bacteroidia; Proteobacteria and Tenericutes, accounted for 88.2% of total sequences. Concerning the Clostridia Family XIII, the C. Incertae Sedis was the most represented (6.6%), and its quantity was twice as much in rumen inoculated hydrogen-producing samples than those non-producing. In the archaeal, community predominated the phylum Euryarcheota, with Methanobrevibacter the most represented, which was higher when hydrogen was produced with rumen inoculum. Studies on buffalo rumen as inoculum for hydrogen production are limited and this paper gives a first overview of microbial community composition by NGS in producing and non-producing samples.

Keywords: waste, bio-hydrogen, rumen, buffalo sludge, low protein cheese whey (scotta), metagenomic, microbial community

# INTRODUCTION

Biogas production by anaerobic digestion (AD) process from agricultural and industrial wastes delivers two benefits: on one side, the treatment of organic residues prevents the environmental and economic impact of their disposal; on the other side, energy carriers (methane and/or hydrogen) are generated. Current interest in the use of hydrogen as fuel, both for industrial applications and for road haulage, is because it has a high-energy density (122–142 kJ/g) and produced pollution is almost nil (Antonopoulou et al., 2008; Venetsaneas et al., 2009). Infact, hydrogen only produces water and traces of N oxides, when used in combustion systems, while producing only water, when used with electrochemical systems (fuel cells). Bio-hydrogen production is depending on various elements such as feedstocks composition, microbial inocula, pretreatments, pH, temperature (Hawkes et al., 2002; Kim et al., 2006; Karadag and Puhakka, 2010; Karadag et al., 2014).

Considerable amount of cheese whey (CW), around 145 mil ton/year (Macwan et al., 2016) are produced worldwide deriving from cheese-making process. CW is rich in lactose and minerals, along with a chemical oxygen demand (COD) concentration up to 80 g L−<sup>1</sup> , therefore, it was considered an valid ingredient for bioenergy production (Sansonetti et al., 2009; Hublin and Zelic, 2013; Perna et al., 2013; Karadag et al., 2014). In Italy, a large amount of CW is produced yearly, also deriving from buffalo mozzarella cheese (4th DOP cheese as production), together with "scotta", a deproteinized by-product, obtained after the production of both cheese and "ricotta" (similar to cottage cheese product, widespread in Anglo-saxon country) (Vasmara and Marchetti, 2017). The scotta has still a high organic load, environmentally pollutant if improperly managed, so cheese manufacturers would be extremely interested to make "scotta" an economic value by-product, also avoiding costly disposal solutions.

Nevertheless, some difficulties in AD of CW have been reported, such as low alkalinity content, rapid acidification and subsequent reactor failure (Ergüder et al., 2001). To address such difficulties, the codigestion with dairy manure has proven effective (Kavacik and Topaloglu, 2010), because of its high suspended solids content and fibrous material, sufficient alkalinity level, and higher buffer capability, so representing a complementary combination with CW (Rico et al., 2015).

AD processes for biogas production are divided in four major phases: hydrolysis, acidogenesis, acetogenesis, and methanogenesis (Batstone et al., 2002; Thauer et al., 2008; Angelidaki et al., 2011). Different bacterial groups accomplish the first three, but methanogenesis is exclusively performed by methanogenic Archaea. So microbial consortia belonging to different environments, such as marine and fresh water sediments, soils, and animal gut, can produce biogas, but they their taxonomic abundance and diversity relies on physical and chemical conditions in addition to substrate availability and kind (Gudelj et al., 2010; Kittelmann et al., 2013; Tapio et al., 2017). According to Johnson et al. (2012), "metabolic specialization provides one plausible explanation for how diversity could be promoted, and is therefore a likely general organizing principle that shapes the assembly of microbial communities."

The potential of application of rumen microorganisms as inocula has been explored so far, mainly for the conversion of lignocellulosic biomass (Hu and Yu, 2005; Sutherland and Varela, 2014; Sawatdeenarunat et al., 2015; Li et al., 2017). Rumen is a well-adapted microbial community, on which ruminants rely on to convert feed into energy-yielding products, such as VFAs, used by the host as an energy source. Rumen contains two groups of prokaryotes (bacteria, dominated by the phyla Firmicutes and Bacteroidetes, and Archaea) and two groups of eukaryotes (protists and fungi) and theirs strong metabolic interactions characterize its environment. Microbial community in the rumen demonstrates according to Weimer (2015), both redundancy (overlap of physiological capabilities across species) and resilience (resistence to, capacity to spring back to shape after perturbation).This flexibility and consequent resilience would make it suitable for application to different industrial contexts such as energy production (Henderson et al., 2015; Weimer and Kohn, 2016).

Although complexity of microbial communities have been studied for a long time, recent advances in metagenomic approach by next generation sequencing (NGS) offer a new tools because of both their molecular detail and their accessibility to a broad scientific community, supplementing culture-based approaches (Schmidt et al., 1991; Tringe and Hugenholtz, 2008). Metagenomic studies have explored a number of environments, including cow rumen (Li et al., 2017; Tapio et al., 2017), human intestinal tract (Qin et al., 2010) and oral cavity (Dewhirst et al., 2010), marine (Venter et al., 2004), freshwater (Roux et al., 2012), soil (De Angelis et al., 2010), and air (Tringe et al., 2008). The sequencing of 16S rRNA gene is the most efficient approach to assess the phylogeny and diversity of a community, especially if the studied environment contains a large fraction of uncharacterized microorganisms.

The laboratory trial reported in this paper was preparatory to the application in a two-stage pilot plant for combined production of hydrogen and methane operating with liquid and semiliquid feedstocks (CREA-ENEA Italian patent n◦ 1416926). The aims of this study were both to produce bio-hydrogen from buffalo wastes (sludge—a mix of manure and urine—and scotta) and to investigate the bacterial and archaeal community structure related to gas production.

The first step was to verify if buffalo rumen fluid (BU) were a better inoculum than buffalo sludge itself (BS) to be used in biohydrogen production and if scotta could be a valid substitute for raw CW as well as sludge for manure. The microbial community (bacterial and archaeal) was analyzed using NGS of 16S rRNA gene amplicons, to identify core groups and specific microbes involved in the dark fermentation of these buffalo wastes.

# MATERIALS AND METHODS

### Experimental Design and Animals

Anaerobic codigestion experiments using buffalo by-products (sludge and scotta) inoculated with BU and BS were performed. Feedstocks were fresh BS, a mixture of manure and urine taken from our farm sewage tank, strained (two layers of cheesecloth) before use, and fresh scotta from a local cheese factory used within an hour.

Inocula were fresh BS (the same as feedstock) and buffalo rumen. The latter (1 L) was taken from three cannulated Mediterranean Buffalo cows fitted with rumen cannula (#3C, 100 mm, Bar-Diamond Inc., Parma, USA) before the morning feeding, strained (three layers of cheesecloth) and pooled.

The 60:40 BS/Scotta ratio (for a total of 37 g L−<sup>1</sup> Volatile Solids (VS)) was used, pH was adjusted at 7 and temperature maintained at mesophilic condition (39◦C) throughout both trials, using inoculum at 15%. The 120 ml batch reactors (70 ml of working volume, in three replicates) were flushed with nitrogen gas to establish anaerobic conditions and maintained in the dark at orbital shaking (40 rpm/min). The trials lasted until the end of hydrogen production (i.e., 12 days).

Because BS inoculum gave very low hydrogen production in first trial (T1), a subsequent trial (T2) was carried out using a matured BS inoculum (BS2), maintained for 3 weeks at room temperature.

Substrate composition is reported in **Table S1**.

An ethics approval for all experimental procedures of this research was obtained from the Italian Welfare Ministry (DGSAF, 588/2017-PR), in accordance with the guidelines established by the European Community Council Directives 86/609/EEC.

### Samples Collection

Samples from every replicate and each sampling time (0, 5, 7, 10, 12 days) were taken for VFAs and lactic acid analysis (0.5 mL) and pH determination. VS (70 mL) were determined at the beginning and end of the experiment. Samples for DNA extraction (1 mL) were taken from each replicates at the beginning of the experiment and at day 7, 10, 12 to compare the two trials.

## Chemical Analysis

The metabolic products of fermentation (VFAs, lactic acid) were analyzed through high-performance liquid chromatography (HPLC) (Waters 2695, HPLC and detector Waters 2487) equipped with an UV-vis detector (λ = 220 nm) and a refractive index detector. The analytical column was Aminex 85 HPX-87H (300 × 7.8 mm and 9µm particle size) of BIO-RAD (California, USA) with a 4 × 30 mm security guard cartridge Carbo-H (Phenomenex, California, USA). Operating conditions were 40◦C, under isocratic conditions, using a solution of 0.008N H2SO<sup>4</sup> as mobile phase (flow rate, 0.6 ml/min filtered through a 0.45µm Millipore Teflon membrane and degassed). The liquid samples were diluted 1:1 (v/v) in H2SO<sup>4</sup> (0.1N) and filtered with 0.45µm Teflon membrane before injection of 20 µL into the HPLC. The external standard analytical curves were prepared at three different concentrations for each acid (lactic, acetic, propionic, i-butyric, butyric, i-valeric, and valeric, pure standards from Sigma, USA) and fitted with weighted leastsquare regression.

Total biogas volume was measured using a water displacement system (Kalia et al., 1994) and cumulated gas production (ml g VS−<sup>1</sup> ) was calculated. Biogas composition (% v/v) in reactors headspace was analyzed using a gas chromatograph (GC, MPS)(Pollution, Bologna, Italy) equipped with a Molsieve column (L10m; i.d. 320µm; film 30µm) + pre-column (L 3 m; i.d. 320µm; film 30µm) with backflush injector for di H2, CH4,O2, N<sup>2</sup> determination.

### DNA Extraction and Sequencing

DNA was extracted from 26 samples, including the inocula, 15 from trial T1 (the BS non-producing samples were pooled) and 11 from trial T2. The samples were centrifuged at 5,000 g for 5 min and the pellet re-suspended in 400 µl of PBS (Sigma, USA) before extraction, using Maxwell <sup>R</sup> 16 tissue DNA purification kit (Promega, Madison, USA), in Maxwell <sup>R</sup> 16 Instrument according to manufacturer guidelines. The quality and the quantity of DNA was determined using NanoPhotometerTM Pearl (Implen GmbH, München, Germany) and Quantity-one fluorimeter (Promega, Madison, USA). The 16S rRNA gene amplicons were generated with the U341F (CCTAYGGGRBGCASCAG) and U806R (GGACTACNNGGGTATCTAAT) primers targeting the hypervariable V3–V4 regions for both bacteria and archaea. Amplicons from individual samples were pooled at equal molar ratios and purified. About 100 ng of each pool was used to construct sequencing libraries using the Ovation Rapid DR Multiplex System 1-96 (NuGEN). The sequencing libraries were pooled and size selected by preparative gel electrophoresis and directly sequenced at 300 bp paired-end reads by using the Illumina MiSeq V3 sequencer (Illumina, S.Diego, USA) for 2.5 million read pairs.

### Statistical Analysis

The chemical data were analyzed using the SAS general linear models procedure (SAS Inst., vers. 9.4; Cary, NC, USA) (SAS Statistical Analysis System Institute, 2012). Least square means (LSM) and pooled standard error of means were obtained.

The analysis of metagenomic data included Krustal-Wallis statistical test (Bonferroni adjusted) on OTU abundance in the different sample groups [initial composition (T0) and producing and non-producing samples (Y,N) according to trial (T1, T2)]. On alpha diversity (Chao1 and Observed species indexes), comparison tests using a two-sided Student's twosample t-test were undertaken. Beta diversity using Unifrac weighted and unweighted distance metrics was calculated and significance tests were performed using a two-sided Student's two-sample t-test (parametric) and nonparametric p-values, calculated using 999 Monte Carlo permutations to underline statistical differences between the above mentioned groups. A hierarchical clustering was obtained by Unweighted Pair Group Method with Arithmetic mean (UPGMA) (Sneath and Sokal, 1973). Jackknifed beta diversity analyses measured robustness of UPGMA clusters.

### Metagenomic Data Analysis

Samples were processed and analyzed with the following procedure: (1) pre-processing of raw sequence reads for quality control (QC). Demultiplexing of all libraries for each sequencing lane using the Illumina bcl2fastq 1.8.4 software; clipping of sequencing adapter from 3′ ends of reads; Combination of forward and reverse reads using BBMerge 34.48; FastQC reports for all FASTQ files; (2) processing of the **combined** reads for the community analysis. Sequences were filtered using Mothur 1.35.1, aligned against the Mothur-Silva16S SEED r119 reference alignment, and subsampled to 80,000 sequences per each sample. Chimeras sequences were removed with Uchime algorithm and taxonomical classification performed against the Silva reference classification. OTUs were picked at a 97% identity level using the average neighbor method. Both OTU diversity and statistical analyses (rarefaction curves, alpha and beta-diversity estimates) were conducted using the Quantitative Insights Into Microbial Ecology (QIIME 1.9.0) open-source pipeline (Caporaso et al., 2010).

# RESULTS AND DISCUSSION

### Batch Hydrogen Production Performances

In T1 trial, biogas production started from day 5 and day 7 when BS and BU inoculum were employed respectively, as reported in **Table 1**. This could be due to the presence of a richer Archea population, which used hydrogen to produce methane, in buffalo rumen inoculum (I-BU) compared to BS inoculum (I-BS) samples (**Figure 1**). As reported in **Table 1** cumulated production was higher with BU inoculum than with BS (120.8 vs. 23.61 ml H<sup>2</sup> g VS−<sup>1</sup> , respectively), nevertheless, VS reduction at the end of the trial was similar (11% approximately). The hydrogen percentage obtained on average was 48.1 (v/v) with BU at day 7, against a 23.7 of BS.

In trial T2, the hydrogen production started at day 5 with 27% (v/v in biogas) and increased to 55.6% at day 7 so being higher in BS2 than BU in trial T1, and resulted in 65.4 cumulated production (ml H<sup>2</sup> g VS−<sup>1</sup> ). In this case, an inoculum, matured for 3 weeks, was used, confirming that aging can improve performances (Kavacik and Topaloglu, 2010; Di Cristofaro et al., 2014). The hydrogen percentage in biogas was higher in T2 trial compared to T1 (for BU samples) and both results were interesting if compared to what reported in literature from livestock manure and wastewater (Tang et al., 2008; Xing et al., 2010; Di Cristofaro et al., 2014) but nonetheless BU inoculum gave the highest overall yield. This could be due to the higher content of total VFAs (46.2 vs. 44.7 g L−<sup>1</sup> , T1, and T2 respectively), and particularly acetic acid (P < 0.01, data shown in **Tables 2**, **3**) which suggest a more efficient fermentation process.

In **Table 2**, pH and VFAs values, measured during trials T1 and T2, are reported. Only BU (T1) and BS2 (T2) inocula were compared, because hydrogen production from BS was too low. The accumulation of VFAs and especially lactic acid, during the first 5 days, caused an intense reduction of pH, which dropped to 4.1 in T1 and to 4.6 in T2. Nevertheless, the buffering capability of buffalo sludge as feedstock allowed the increase and maintenance of pH at an optimum range (5–6) throughout the fermentation process as reported also by different authors (Antonopoulou et al., 2008; Weiland, 2010; Tenca et al., 2011). The rapid increase in lactic acid during the first 5 days (1.33 to 12.16 in T1 and 0.25 to 11.49 g L−<sup>1</sup> in T2, P < 0.01), due to the scotta used as feedstock (Ergüder et al., 2001; Rico et al., 2015), showed statistically significant differences between sampling times but not between trials. As soon as lactic acid concentration started to decrease, hydrogen production began. Acetic acid concentration was the highest among all VFAs, and revealed a significant difference between trials (3.35 vs. 2.33 g L−<sup>1</sup> on average for T1 and T2 respectively, P < 0.01). This is in agreement both with Vasmara and Marchetti (2017) results obtained using scotta permeate and Li et al. (2017) data using rumen fluid inoculated reactors for methane production. Propionic acid concentration showed no statistical difference between trials but within the trials. Butyric acid concentration showed a statistical difference only between day 10 and the other sampling times in T1 trial and between day 10 and 12 in T2. Valeric acid was detectable only when hydrogen was produced and the quantity was significantly higher in T1 compared to T2 (1.1 vs. 0.7 g L−<sup>1</sup> on average, P < 0.01). i-Valeric was detectable in T2 day 10 and 12 samples. Hu and Yu (2005) using rumen microbes for biogas production reported that "by inhibiting methane formation, a potential hydrogen sink was eliminated, and the reducing power was used to produce higher carboxylic acids. This might be responsible for the production of longer chain fatty acids, such as valerate." This observation is in agreement with our hypothesis about a more efficient fermentation in BU samples, which resulted in a higher hydrogen production, some of which could be transformed in valeric acid.

TABLE 1 | H2 average percentage in gas mixture according to sampling times (day 5, 7, 10, and 12), Volatile solids (VS) average reduction, H2 and CH4 cumulated production during trial T1 and T2.


H<sup>2</sup> (% v/v) is expressed as mean ± SD; BS, buffalo sludge; BU, buffalo rumen; BS2, buffalo sludge matured.

TABLE 2 | Statistical analysis of VFAs and lactic acid concentration (g L−<sup>1</sup> ) during trial T1 and T2 according to sampling times (day 0, 5, 7, 10, and 12).


Uppercase letter refers to trial differences. Lowercase letter to sampling times differences; statistical difference is at P < 0.01, nd, not detected. The dark gray columns represents hydrogen-producing samples.

In **Table 3** the results of the statistical analysis between producing (Y) and non-producing samples (N) in trial T1 (buffalo inoculated) and T2 (BS2 inoculated) are reported. Lactic acid samples were different only in T1 trial, where N samples had the highest value because included day 5 samples (12.2 g L−<sup>1</sup> , **Table 3**).

We hypothesized the day 5 samples in T1 did not produce hydrogen, because it included a high number of methanogens, which consumed H<sup>2</sup> to produce methane (data reported in **Table 1**). Acetic acid concentration between Y and N in T1 was also statistically significant, and BU inoculum produced 62% more than BS2 during hydrogen production (3.76 vs. 2.34 g L−<sup>1</sup> on average, P < 0.01). i-butyric acid concentration was statistically different in T1 (0.11 vs. 0.40 g L−<sup>1</sup> , Y, and N respectively) where the quantity decreased when hydrogen production started. It is interesting to note that valeric acid is detectable only when hydrogen is produced and the quantity is significantly higher in T1 samples.



Y, Hydrogen-producing; N, Non-producing Hydrogen; nd, not detected. Lowercase letter refers to a statistical difference at P < 0.01.

# Microbial Community Analysis and Composition

After the pre-processing, 2,003,102 combined reads were retained for the following analysis. Sequences were deposited in the EMBL-EBI (European Nucloeotide Archive) under the study accession PRJEB25602. The 16S rDNA amplicon sequencing depth of the 26 analyzed samples ranged from 31,826 to 140,676 paired reads. Singleton OTUs were removed and the data from each sample were rarefied at 19,769 to keep as many samples as possible. 10,779 OTUs were obtained from all the samples. Krustal-Wallis test did not show any statistically significant differences between the OTUs abundance in the different sample groups.

The first step of our analysis was to obtain a general description of the microbial community composition found in all samples. The majority of the sequences were members of 4 phyla: Firmicutes (39.73%), and in particular Clostridia (23.31%) and Bacilli (12.06%); Bacteroidetes (31.98%), 28.80% of whom Bacteroidia; Proteobacteria (9.92%) and Tenericutes (6.59%), which all together accounted for 88.22% as reported in **Table S2**. Aboundance of minor bacterial phyla (Actinobacteria, Spirochaetae, Flavobacteria) were below 3%.

The first three main phyla are reported as part of a core group by numerous authors (Abendroth et al., 2015; Bassani et al., 2015; Duda et al., 2015; Goux et al., 2015). In AD of cattle manure and sludge, some authors (Rivière et al., 2009; St-Pierre and Wright, 2014; Campanaro et al., 2016) reported, as part of the core group, also Cloroflexi, which, on the contrary, we found at 0.03% only. Maus et al. (2017) also reported Firmicutes, Bacteroidetes as the main bacteria phyla in mesophyllic conditions plants with cow manure as inoculum. The Firmicutes are important because they are implicated in several metabolic processes including carbohydrates degradation, fatty acids utilization, Wood– Ljungdahl pathway (WLP) (homoacetogenesis) or syntrophic acetate oxidation (SAO) as reported by Campanaro et al. (2016).

Six orders have been found, among Firmicutes: Bacillales (5 families, 20 genera), Lactobacillales (5 families, 18 genera), Clostridiales (18 families, 88 genera), Erysipelotrichiales (1 family, 9 genera), Selemononadales (2 families, 17 genera), OPB54 (1 family, 1 genus).

The Bacteroidetes which are known to be proteolytic bacteria, were the second most abundant phylum, including Bacteroidales (20 families, 41 genera), Cytophagales (3 families, 9 genera), Order III (2 families, 2 genera), Flavobacteriales (3 families, 24 genera), Sphingobacteriales (7 families, 16 genera).

The Proteobacteria included Alphaproteobacteria (1.15%), Betaproteobacteria (0.92%), Deltaproteobacteria (0.41%), Epsilonbacteria (0.5%) and Gammaproteobacteria (6.84%). The Tenericutes include Mollicutes and in particular Acholeplasmatales (5.70%).

In a second step we were focused on the analysis of microbiological community structure in hydrogen producing (Y) and non-producing groups (N) both in trial T1 and T2. In **Figure 1** we report the taxonomic analysis at Class level, taking into account frequency above 0.3%, according to: I-BU and I-BS inocula, T0 (beginning of trial) and T1 and T2, Y, and N samples.

The Firmicutes were higher in N samples, in both trial, in particular Bacilli and Negativicutes. Bacilli numbers (97% of which are Lactobacillales) increased markedly from 1.3% in T0 group to 24.9% in T1-N, in accordance to lactic acid increase that we previously discussed (see **Table 2**). Whitin the trials they decreased in Y samples (24.9% to 14.1% in T1-N and T1-Y and 13.0% to 9.5% in T2-N and T2-Y, respectively) (see **Table S3**). This was favored by pH increase due to sludge buffering capacity. Moreover, some researchers observed that the excretion of bacteriocins by Lactobacilli and Prevotella resulted in inhibitory effects of hydrogen production (Castelló et al., 2011). Negativicutes (7.3%), Candidate\_division\_TM7 (2.2%), and Synergistia (0.8%) are more abundant in T2 and in particular N samples, even though the latter are reported to ferment amino acids to acetate and propionate together with other species (Campanaro et al., 2016).

On the contrary, Clostridia were higher in T1 hydrogen producing samples (see **Figure 1** and **Table S3**) as reported in literature (Davila Vazquez et al., 2009; Cheng et al., 2011; Maus et al., 2017). Among the Clostridia, Family XIII, the C. Incertae Sedis was the most represented (6.5% in our trial), in particular in T1-Y, the quantity was twice as much as T1-N; in T2 trial we observed the opposite trend. Matsumoto et al. (2017) reported a positive correlation between the Incertae Sedis bacteria and hydrogen production in human gut.

Clostridium Senso Stricto 1 (1.7%), among Clostridiaceae\_1, reported by Maus et al. (2017) as the most represented, in our trial was the second more abundant genus and we observed higher quantity in T1-N than T1-Y, and lower quantity in T2-N than T2-Y.

Among Bacteroidales the most represented was Prevotella (4.2% on average), particularly high in T2-N (8.8%), nevertheless we observed three fold more bacteria in T1-Y samples (5.9%) compared to T1-N (1.7%). On the contrary, RC9 gut group (2.1%) was more abundant in T1 trial compared to T2, and in Y samples compared to N. BS11 gut group was present in T1 only, which had buffalo rumen inoculum, and in higher quantity in hydrogen-producing samples.

The alpha diversity analysis between groups (initial composition (T0) and producing and non-producing samples (Y,N) according to trial (T1, T2)) using Chao1 index is reported as box-plot in **Figure 2**. A statistically significant difference (P < 0.05) was revealed between groups considering T0 vs. T2-Y, T1-Y vs. T2-N, T1-Y vs. T2-Y. The higher species richness observed in T1 vs. T2 and T0 group, are probably due to the presence of rumen inoculum, to be noticed that its richness is not evident in **Figure 1**, where only frequencies above 0.3% were included.

Statistical significant difference according to beta diversity was obtained only between T1-Y Vs T2-Y with parametric test (data not shown). The result of beta diversity between all samples calculated as hierarchical clustering by UPGMA is presented in **Figure 3**. The Figure shows clearly the clustering of the BU inoculated samples separated from the BS samples with a good jackknife support (> 75%). Moreover the BU inoculum is separated from all the other samples.

All the Archea identified species obtained in this study were attributed predominantly to phylum Euryarcheota (1.74%) as reported in **Table S2**. Considering the groups analysis, Euryarcheota included 5 orders: Halobacteriales, Methanobacteriales, Methanomicrobiales, Methanosarcinales, and Thermoplasmatales as found in previous studies (Rivière et al., 2009; Sun et al., 2015). The most abundant were Methanocorpusculaceae (0.5%) with predominance of genus Methanocorpusculum, and Methanobacteriaceae (0.5%) with three genera Methanobacterium, Methanosphaera, and Methanobrevibacter (**Table S3**). The latter was particularly higher when hydrogen was produced with rumen inoculum (1.5%, **Table S3**). As reported in bovine rumen there is a prevalence of this hydrogenotrophic Archaea, which uses H2-CO<sup>2</sup> as substrate to produce methane (Li et al., 2017).

The Methanomicrobiales included 5 families The Methanosarcinales were represented by Metanosarcinaceae and in particular Methanimicrococcus (0.4%) and Thermoplasmatales by family Incertae Sedis and genus Candidatus Methanomethylophilus (0.2%).

FIGURE 3 | A visualization of bootstrap-supported hierarchical clustering (UPGMA) of the 26 sample's microbial communities. Internal tree nodes color represent jackknifed support (red for 75–100%, yellow for 50–75%, green for 25–50%, and blue < 25%).

## CONCLUSIONS

This paper deals with the bio-hydrogen production from livestock by-products (buffalo sludge and scotta) with buffalo rumen and buffalo sludge as inocula as well as the microbial community composition. Codigestion was positive due both to sludge buffering capability and scotta's content in high degradable carbohydrates. Buffalo rumen has a number of appealing features as lignocellulose degradation and community resilience, which could be exploited in industrial application such as gas production. It proved more efficient compared to buffalo sludge as inoculum when hydrogen cumulated production is concerned, showing also higher VFAs production, particularly acetic acid and confirming bacterial species richness. However, the maturation of BS inoculum, allowed to improve its performances.

The majority of the bacterial identified genera belonged to Firmicutes—in particular Clostridia—, Bacteroidetes, Proteobacteria, and Tenericutes. Clostridia were higher in hydrogen producing samples confirming what is generally reported in literature. In our study the Family XIII, C. Incertae Sedis was the most represented when rumen inoculated samples were tested.

Most of the archaeal sequences belonged to Euryarcheota and the most abundant was Methanobrevibacter, particularly when hydrogen is produced with buffalo inoculum.

Studies on buffalo rumen employed as inoculum for hydrogen production are limited and this paper gives a first overview of the microbial community structure in dark fermentation processes by next generation sequencing.

### REFERENCES


### AUTHOR CONTRIBUTIONS

AnC, contributed to the study conception and design of the study and performed the dark fermentation analysis; AlC contributed to the metagenomic analysis. Both authors have contributed to the analysis and interpretation of data, drafting of manuscript as well as critical revision. Both authors have read and approved the submitted version.

### ACKNOWLEDGMENTS

This work was supported by the Italian Ministry of Agriculture (MiPAAF) under the AGROENER project (D.D. n. 26329, 1 April 2016) - http://agroener.crea.gov.it/. We are also grateful to farm workers for animal care and Emanuela Rossi for literature review.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fsufs. 2018.00013/full#supplementary-material

Table S1 | Substrate chemical composition.

Table S2 | Taxonomy-binned sequences counts per samples. The levels correspond to taxonomic classification (from Phylum to Class).

Table S3 | Bacterial and archaeal taxa abundances according to beginning of trial (T0) and T1 and T2 trials, hydrogen producing (Y) and non-producing (N) samples. The levels correspond to taxonomic classification (from Phylum to Genus)

production in a UASB reactor fed with raw cheese whey inoculated with compost. Water Sci. Technol. 64, 2265–2273. doi: 10.2166/wst.2011.706


whey in batch and UASB reactors. Waste Manag. 21, 643–650. doi: 10.1016/S0956-053X(00)00114-8


in a two-stage continuous process with alternative pH controlling approaches. Bioresour. Technol. 100, 3713–3717. doi: 10.1016/j.biortech.2009.01.025


Xing, Y., Li, Z., Fan, Y., and Hou, H. (2010). Biohydrogen production from dairy manures with acidification pretreatment by anaerobic fermentation. Environ. Sci. Pollut. Res. 17, 392–399. doi: 10.1007/s11356-009-0187-4

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer BG and handling Editor declared their shared affiliation.

Copyright © 2018 Chiariotti and Crisà. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Advanced Processing of Food Waste Based Digestate for Mitigating Nitrogen Losses in a Winter Wheat Crop

Antonio R. Sánchez-Rodríguez 1,2 \*, Alison M. Carswell <sup>3</sup> , Rory Shaw<sup>1</sup> , John Hunt <sup>3</sup> , Karen Saunders <sup>3</sup> , Joseph Cotton<sup>1</sup> , Dave R. Chadwick <sup>1</sup> , Davey L. Jones <sup>1</sup> and Tom H. Misselbrook <sup>3</sup>

### Edited by:

Raul Moral, Universidad Miguel Hernández de Elche, Spain

### Reviewed by:

Raul Zornoza, Universidad Politécnica de Cartagena, Spain Margarita Ros, Centro de Edafología y Biología Aplicada del Segura (CEBAS), Spain

\*Correspondence:

Antonio R. Sánchez-Rodríguez afs42a@bangor.ac.uk; antonio.sanchez@uco.es

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 14 February 2018 Accepted: 18 June 2018 Published: 17 July 2018

### Citation:

Sánchez-Rodríguez AR, Carswell AM, Shaw R, Hunt J, Saunders K, Cotton J, Chadwick DR, Jones DL and Misselbrook TH (2018) Advanced Processing of Food Waste Based Digestate for Mitigating Nitrogen Losses in a Winter Wheat Crop. Front. Sustain. Food Syst. 2:35. doi: 10.3389/fsufs.2018.00035 <sup>1</sup> School of Environment, Natural Resources & Geography, Bangor University, Bangor, United Kingdom, <sup>2</sup> Departamento de Agronomía, Escuela Técnica Superior de Ingeniería Agronómica y de Montes (ETSIAM), Universidad de Córdoba, Córdoba, Spain, <sup>3</sup> Sustainable Agricultural Systems—North Wyke, Rothamsted Research, Devon, United Kingdom

The anaerobic digestion of food waste converts waste products into "green" energy. Additionally, the secondary product from this process is a nutrient-rich digestate, which could provide a viable alternative to synthetically-produced fertilizers. However, like fertilizers, digestate applied to agricultural land can be susceptible to both ammonia (NH3) and nitrous oxide (N2O) losses, having negative environmental impacts, and reducing the amount of N available for crop uptake. Our main aim was to assess potential methods for mitigating N losses from digestate applied to a winter wheat crop and subsequent impact on yield. Plot experiments were conducted at two UK sites, England (North Wyke-NW) and Wales (Henfaes-HF), to assess NH<sup>3</sup> and N2O losses, yield and N offtake following a single band-spread digestate application. Treatments examined were digestate (D), acidified-digestate (AD), digestate with the nitrification inhibitor DMPP (D+NI), AD with DMPP (AD+NI), and a zero-N control (C). Ammonium nitrate (NH4NO3) fertilizer N response plots (from 75 to 300 kg N ha−<sup>1</sup> ) were included to compare yields with the organic N source. Across both sites, cumulative NH3-N losses were 27.6% from D and D+NI plots and 1.5% for AD and AD+NI of the total N applied, a significant reduction of 95% with acidification. Cumulative N2O losses varied between 0.13 and 0.35% of the total N applied and were reduced by 50% with the use of DMPP although the differences were not significant. Grain yields for the digestate treatments were 7.52–9.21 and 7.23–9.23 t DM ha−<sup>1</sup> at HF and NW, respectively. Yields were greater from the plots receiving acidified-digestate relative to the non-acidified treatments but the differences were not significant. The yields obtained for the digestate treatments ranged between 84.2% (D+NI) and 103.6% (D) of the yields produced by the same N rate from an inorganic source at HF. Advanced processing of digestate reduced N losses providing an environmentally sound option for N management.

Keywords: ammonia volatilization, greenhouse gas emissions, N2O, nitrification inhibitors, acidification, digestate

# INTRODUCTION

During the last few decades, the interest in anaerobic digestion in the European Union (EU) has increased due to the development of regulations and guidelines that encourage the production of renewable energy to benefit the environment (Siebert et al., 2008; EU, 2009; BSI, 2010). Anaerobic digestion plants generate biogas (rich in methane), a source of "green" energy, and a liquid byproduct known as digestate, with a high potential as fertilizer or soil conditioner depending on its nutrient content (Nkoa, 2014). The EU has promoted nutrient recovery as part of the circular economy (EU, 2014) encouraging digestate to be valued as an alternative to inorganic and non-renewable fertilizers in agriculture, as a potential source of income rather than a waste or by-product (Alburquerque et al., 2012a,b; Nkoa, 2014; Kataki et al., 2017).

The main feedstocks for biogas plants are energy crops, animal manures, and other organic wastes (Lukehurst et al., 2010) depending on what is locally available. In some countries of the EU, including the UK, anaerobic digestion is the recommended technology for sanitizing food waste from supermarkets, catering, and kitchen waste (Lukehurst et al., 2010), and their treatment through anaerobic digestion is increasing (Styles et al., 2016). Nevertheless, there is a lack of evidence for the agronomic and environmental effects of the application of food waste derived digestate to agricultural land.

Anaerobic digestion modifies the former properties of the feedstocks, affecting N cycling and bioavailability once the digestate is applied to the soil as a source of nutrients for crops. The enhanced microbial degradation of organic matter and emission of carbon (C), particularly as methane, results in an increase in the proportion of total N that is more readily plant available (i.e., in increase in the ratio of ammonium-N (NH<sup>+</sup> 4 -N) to total N, typically to >70%), a decrease in the C:N ratio and a lower organic matter and dry matter (DM) content (Webb and Hawkes, 1985; Möller et al., 2008; Tampio et al., 2016). Anaerobic digestion can significantly reduce greenhouse gas and odor emissions (if fugitive emissions are minimized) in comparison with the feedstock (Massé et al., 2011; Battini et al., 2014), and produces a more sanitized product when the feedstock is manure (Orzi et al., 2015). However, the increase in pH and NH<sup>+</sup> 4 -N content through anaerobic digestion enhance the polluting potential of the digestate during storage (Sommer and Husted, 1995) and following land spreading (Möller, 2015). The main concerns regarding application of digestate and other organic wastes to agricultural land are emissions of N to the environment through ammonia (NH3) volatilization, nitrate (NO<sup>−</sup> 3 ) leaching and greenhouse gas emissions as nitrous oxide (N2O), with associated impacts on air and water quality, ecosystem functioning and human health (Galloway et al., 2003).

Tiwary et al. (2015) reported that 35–65% of the total N applied in digestate can be lost through NH<sup>3</sup> volatilization if the digestate is surface broadcast. Potential methods to reduce NH<sup>3</sup> volatilization include the rapid incorporation of manures and digestates into the soil after application (Möller et al., 2008; Tiwary et al., 2015), soil injection (Riva et al., 2016), bandspreading (Nicholson et al., 2017), and acidification of slurries (Fangueiro et al., 2015a).

Nitrous oxide emissions following digestate application to land are thought to be lower than those emitted from the undigested material because most of the available C has been converted to biogas prior to land application. However, there are contradictory reports from the literature (Möller, 2015) suggesting that emissions are related to the feedstocks and soil properties to which they are applied, e.g., soil organic matter content, soil texture, water content, and aeration (Chantigny et al., 2009; Eickenscheidt et al., 2014). Reported N losses as N2O emissions following the application of food-based digestate vary from 0.45% (Nicholson et al., 2017) to 4–10% (Tiwary et al., 2015) of the total N applied. A method to reduce N2O emissions from manure applications, which may be equally applicable to digestates is the use of nitrification inhibitors (NI), such as 3,4-Dimethylpyrazole phosphate (DMPP) (Owusu-Twum et al., 2017), that delay the process in which NH<sup>+</sup> 4 transforms into NO<sup>−</sup> 3 . Nitrate is a readily mobile form of N, which can be lost by leaching, therefore, keeping N in the form of NH<sup>+</sup> 4 (lessmobile) could prevent NO<sup>−</sup> 3 leaching while minimizing N2O losses (Subbarao et al., 2006; MarkFoged et al., 2011).

The main objective of this study was to compare the efficiency of different N loss mitigation strategies (acidification, use of a nitrification inhibitor, and the combination of both) to reduce N losses (NH<sup>3</sup> volatilization and N2O emissions) and enhance the value of food waste based digestate as a source of N for a winter wheat crop. Our hypothesis was that the acidification of the digestate and the use of a nitrification inhibitor (i.e., DMPP) would decrease N losses in relation to untreated digestate, improving the N use efficiency for crop yield and thereby the potential of digestate as an alternative to an inorganic fertilizer N source.

### MATERIALS AND METHODS

## Site Description and Experimental Design

Two field experiments were conducted on a winter wheat crop over the 2016-2017 UK growing season. The first site was at the Henfaes Research Station (HF), in Abergwyngregyn, North Wales (53◦ 14′ 21.3′′N, 4◦ 0 ′ 50.3′′W; 10 m above sea level). The second site was at Rothamsted Research North Wyke (NW), in Devon, South West England (50◦ 79′ 39.8′′N, 3◦ 95′ 25.1′′E; 180 m above sea level). The former crop was barley at HF and grassland at NW. Both sites have a temperate climate with average annual rainfall of 1,060 and 1,107 mm, respectively. The soil at HF is a free-draining Eutric Cambisol with a sandy clay loam texture and at NW is a free-draining Dystric Cambisol with a clay loam texture (IUSS, 2015). Five representative soil samples were collected from each field site to a depth of 15 cm. Each soil sample was then crumbled by hand, vegetation, roots, and stones manually removed and the soil thoroughly mixed prior to analysis. The main soil characteristics are shown in **Table 1**.

Triticum aestivum (var. KWS Siskin) was drilled on the 10th October 2016 at both sites with a row spacing of 0.1 m. Prior to this, the fields were plowed to 15 cm depth and limed to increase

TABLE 1 | Background soil properties at the Henfaes (HF) and North Wyke (NW) sites.


Values represent means ± standard error (n = 5) and are expressed on a dry matter basis except pH and electrical conductivity (EC). EC, electrical conductivity; DOC, dissolved organic carbon; DON, dissolved organic nitrogen.

the soil pH. Phosphorus (P, as Ca(H2PO4) and potassium (K, as KCl) were applied during the same week of sowing. Kieserite (MgSO4·H2O) was applied in March at both sites. Application rates were based on routine soil analyses and national fertilizer guidelines (DEFRA, 2010), so that these elements were nonlimiting. Herbicides at both sites, and insecticides and fungicides only at NW were also applied according to manufacturers' recommendations. See Table S1 for additional information.

A randomized complete block design was established at each site with one replication in each block equalling five replications per treatment (n = 5), with plot size 14 × 1.2 m at HF and 9 × 2 m at NW. There were four "digestate treatments" and a control:


The target application rate was 190 kg N ha−<sup>1</sup> as digestate, although actual application rate achieved in the field varied (**Table 2**). The digestate was band-spread parallel with crop rows (30 cm between bands) at a rate equivalent to 40 m<sup>3</sup> ha−<sup>1</sup> using 20 l capacity watering cans on April 19th 2017 at HF and March 20th 2017 at NW, at the start of stem elongation and never after early May, according to DEFRA (2010). The digestate remained in bands in the "digestate treatments" at NW but not at HF because of the lower DM content. The plots were divided into two different areas: (1) the harvest area, which was used to determine grain yields and plant production; and (2) the sampling area, which was used for periodic soil sampling, NH<sup>3</sup> volatilization measurements (wind tunnels) and daily N2O emissions (manual or automatic chambers, **Table 2**). At HF, NH<sup>3</sup> emission measurements were made on the main plots, whereas at NW separate "mini-plots" (2 × 0.5 m) were established for these measurements at the prevailing wind (south westerly) edge of the trial site.

Additionally, to be able to calculate the fertilizer replacement rate of the N mitigation digestate treatments, N response plots were included at both sites. Ammonium nitrate (NH4NO3) was applied at four different rates: 75, 150, 225, and 300 kg N ha−<sup>1</sup> split into three applications between March and April 2017 according to the suggestions by DEFRA (2010) for winter wheat. The N response plots were 6.5 × 1.2 m at HF, where they were included in the randomized block design, and 4.5 × 2 m at NW, where they were established in a separate part of the field. Nitrogen response plots were for yield measurement only with no soil or gaseous emission sampling.

**Table 3** gives the main properties of the anaerobic digestate used in the field experiments from six (HF) and 12 (NW) digestate samples. The digestate, based on food waste and without separating solid and liquid fractions, was provided from local anaerobic digestion plants. Half of the digestate used at each site was acidified, to a target pH of 5.5, with concentrated H2SO<sup>4</sup> before application. Approximately 1 l of concentrated H2SO<sup>4</sup> was added in total per 100 l of digestate. The pH of the digestate at application was determined in a 1:6 (v/v) fresh digestate:distilled water suspension and was lower for the acidified digestate than in the non-acidified digestate at both sites, as expected (**Table 3**), although the reduction in pH to <3 for the NW site was greater than anticipated based on previous laboratory tests.

### Soil Sampling

During the experiment, soil was sampled from the sampling area of each plot three times per week for the first 2 weeks after digestate application, two times per week for the next 2 weeks, followed by weekly sampling thereafter. Subsequently, soil samples were taken once per month until the end of the experiment. On each occasion, eight soil samples were taken per plot to 15 cm depth and pooled to provide one representative sample per plot. At NW, soil was sampled proportionally from within and between the digestate bands. At HF, soil was sampled randomly, as there were no distinct digestate bands. Soil samples were stored at 4◦C and in the dark prior to analyses. Soil moisture, pH, EC, NH4+, and NO3− were determined as detailed previously.

# Analytical Methods

### Chemical Properties

Soil pH and electrical conductivity (EC) were determined in a 1:2.5 (w/v) soil:distilled water suspension with standard electrodes using a Model 209 pH meter (Hanna Instruments Ltd., Leighton Buzzard, UK) and a Jenway 4520 conductivity meter (Cole-Palmer Ltd., Stone, UK). Total soil C and N were determined using a TruSpec <sup>R</sup> analyser (Leco Corp., St Joseph, MI) and ground oven-dried soil (105◦C, 24 h). A soil subsample was taken to determine soil moisture and another for mineral N extractions: a 0.5 M K2SO<sup>4</sup> solution was used in a 1:5


TABLE 2 | Plots dimensions, N application rates, and measurements conducted at the Henfaes (HF) and North Wyke (NW) field experiments.

<sup>a</sup>N applied, Control; D, digestate; D+NI, digestate plus nitrification inhibitors; AD, acidified digestate; AD+NI, acidified digestate plus nitrification inhibitors. <sup>b</sup>NUE, nitrogen use efficiency.

<sup>c</sup>EC, electrical conductivity.

TABLE 3 | Properties of the digestate (D) and acidified digestate (AD) used at Henfaes (HF, n = 3, mean ± standard error) and North Wyke (NW, n = 6; mean ± standard error) expressed on a fresh weight basis.


P is the P-value of the ANOVA for all the properties except for pH and dry matter for NW that is the P-value of the Kruskal-Wallis non-parametric ANOVA. na, not-applicable.

soil:extractant ratio (w:v) shaking at 150 rev min−<sup>1</sup> for 30 min and then centrifuging at 10,000 g for 10 min. The supernatant was stored at −20◦C until analyses. Total dissolved organic C (DOC) and total dissolved N (TDN) in the extracts were measured using a Multi N/C 2100/2100 analyser (AnalytikJena AG, Jena, Germany). Dissolved organic N (DON) was calculated by subtracting NH<sup>+</sup> 4 and NO<sup>−</sup> 3 from the TDN value. Ammonium in the extract was determined colorimetrically using the salicylate method of Mulvaney (1996) and NO<sup>−</sup> 3 following the salicylate method of Miranda et al. (2001) in an Epoch <sup>R</sup> microplate spectrophotometer (Bio Tek Instruments Inc., Winooski, VT). Mineralisable N was determined after anaerobic incubation according to Keeney (1982) using 5 g of soil and calculating the differences in NH<sup>+</sup> 4 between the initial concentrations and the concentrations after 7 days of anaerobic incubation. Acetic acid extractable P was used as a proxy for plant-available P, determined after extracting the soil with 0.5 M acetic acid (1:5 w/v, 200 rev min−<sup>1</sup> for 1 h) by the molybdate blue method (Murphy and Riley, 1962) following centrifugation (10,000 g, 10 min).

Total N, and NO<sup>−</sup> 3 -N in the digestates were determined as previously described, and NH<sup>+</sup> 4 -N was significantly higher in the anaerobic digestate for both sites. A digestate sub-sample was oven-dried at 105◦C for 24 h and ground to pass 1 mm sieve to determine the dry matter (DM) content. Dry matter content was greater in the digestate than in the acidified digestate, and greater at NW than at HF (mean values, **Table 3**). A sample of each digestate was digested with concentrated hydrochloric and nitric acid (aqua-regia) to analyse mineral elements by ICP-OES / ICP-MS as detailed in EPA (1996); the acidified digestate had a significantly higher content in Mg at HF and in S at both sites but a lower content in Zn at NW (Table S2).

### Ammonia Volatilization

Ammonia volatilization measurements were made using a system of small wind tunnels as described by Misselbrook et al. (2005). One wind tunnel was placed on each of the "digestate plots" of the four first blocks at HF and on each of the mini-plots at NW directly after the application of the "digestate treatments" (n = 4 for each treatment). Ammonia concentrations of the inlet and outlet air of each wind tunnel were determined using 0.02 M H3PO<sup>4</sup> acid traps (100 ml) changed every day, except for the first day when higher volatilization rates were expected they were changed twice at HF and three times at NW. After each sampling, the acid trap samples were made up to 100 ml with distilled water in the laboratory and a subsample was frozen before analysis for NH<sup>+</sup> 4 -N as described previously. Ammonia fluxes (FNH3, µg m−<sup>2</sup> s −1 ) were calculated according to equation 1:

$$F\_{\rm NH3} = \nu (\rm C\_0 - \rm C\_l) / At \tag{1}$$

where C<sup>o</sup> and C<sup>i</sup> are the outlet and inlet concentrations, respectively, v is the air volume (m<sup>3</sup> ) drawn through the wind tunnel over the sampling period (t, s), and A the area covered by the wind tunnel (m<sup>2</sup> ).

Cumulative NH<sup>3</sup> emissions over the 7 day measurement period were derived by summing the flux from each sampling time. Total N lost through NH<sup>3</sup> volatilization was expressed as a percentage of the total N applied for each treatment to normalize for the different N application rates at the two sites.

### Nitrous Oxide Emissions

Nitrous oxide emissions were measured with a combination of static manual and automated chambers at HF and only manual chambers at NW. Specifically, three replicate plots with one automated chamber (0.5 × 0.5 × 0.2 m) per plot were used for the "digestate" treatment plots at HF, with one manual static chamber (0.5 × 0.5 × 0.3 m) per plot for three control plots (i.e., n = 3 per treatment at HF). At NW, one manual static chamber (0.5 × 0.5 × 0.3 m) was used on each replicate plot for all treatments (n = 5 per treatment). The automatic chambers at HF were linked to an Isotopic N2O Analyser (Los Gatos Research Inc., San Jose, CA, USA) for measurement of N2O concentration. All chambers were installed at least 1 week before digestate application, with edges pushed at least 5 cm into the soil and packing soil around the external edge of the chamber to ensure a proper seal. Gas tight extensions (0.3 m height) were fitted to the chambers during the growing season to accommodate the height of the growing wheat. Readings from 10 (HF) and 5 (NW) SDI 12 soil moisture sensors (Acclima Inc., USA) at 2.5 cm depth and soil bulk density (**Table 1**) were used to calculate water filled pore space (WFPS, Figure S1) to explain daily N2O fluxes.

Sampling from the manual chambers was done at the same frequency as the soil sampling described above, between 10:00 and 12:00 h. Lids were placed on the chambers and gas samples were taken at 20, 40, and 60 min, and 10 ambient air samples taken (5 before and 5 after the sampling period) away from the plot areas as a measure of concentration at time 0 min for each chamber. All gas samples were collected and stored in pre-evacuated vials prior to N2O analysis. All gas samples collected from the manual static chambers were analyzed using a Perkin Elmer 580 Gas Chromatograph fitted with an electroncapture detector and an automated sample injection system and calibrated using certified N2O standards. The installation of the automatic chambers at HF was the same but metal chamber bases were inserted in the soil to a depth of at least 5 cm and the chambers attached to these. Chambers were programmed to close sequentially using pneumatic actuators, for 30 min for gas sampling, resulting in four measurements per chamber per day. Gas was sampled from the chambers via a sampling port at a rate of 1 l min−<sup>1</sup> , and to avoid a negative pressure, the chambers allowed ambient air entry via an air inlet hole of the same diameter as the sampling one, i.e. these were throughflow chambers. Gas samples were delivered to an Isotopic N2O Analyser via 0.17 mm internal diameter PFA tubing, with the same length for all chambers. Nitrous oxide concentrations were recorded at 0.1 Hz during the 30 min chamber closure. N2O concentration data for the first 0.5 min was discarded from calculations to account for the dead volume in the sample lines. Every four chambers, a standard (1.5 ppm N2O) was introduced into the analyser for calibration.

Hourly N2O fluxes (µg N2O-N m−<sup>2</sup> h −1 ) were calculated using linear regression, with the assumption of linearity for manual and automatic chambers. Calculations for the automatic chamber determinations were made using the lm() function in R (version 3.3.2., R Core Team 2016). The manual chamber N2O emissions (FN2O) were calculated as described by de Klein and Harvey (2012); (Excel, Office 2016) using Equation (2):

$$F\_{\rm N2O} = H \times (C\_t - C\_{t0})/t \tag{2}$$

where H is the ratio of chamber volume to soil surface area (l3 to l−<sup>2</sup> ), C is the concentration of N2O within the chamber at the time (t) of sampling and Ct<sup>0</sup> is the N2O concentration measured at 0 min, measured after the chamber had been sealed. Cumulative N2O emissions were calculated for each plot using the area under a curve function "cumtrapz()" from the "pracma" package (Hans Werner Borchers; R Core Team, 2016). Finally, total N lost as N2O was expressed as the percentage of total N applied in each treatment after subtracting the cumulative N2O emissions from the control plots.

### Grain, Plant Production, and Nitrogen Use Efficiency

Grain and plant production were determined from the "harvest area" of each plot at the end of the experiment (8 and 15th August 2017 at NW and HF, respectively). At HF, wheat plants from three 0.4 × 0.4 m quadrats were harvested 2 cm above the ground and grain and straw were separated by hand and weighed. At NW, a Sampo small-plot combine harvester was used to harvest the wheat, separating the grain and straw, which were weighed. A sub-sample from each plot was used to determine grain and straw moisture. Total N was analyzed using a TruSpec <sup>R</sup> analyser (Leco Corp., St Joseph, MI) from ground oven-dried plant tissue (80◦C, 24 h); N offtake by the total crop was calculated by multiplying the N content of the grain and the straw by the grain and straw yield, respectively. Thousand-grain weight (TGW) was also determined by weighing 1,000 oven-dried grains. Grain yield, straw yield and TGW are reported at 85% dry matter.

Nitrogen Use Efficiency of the crop (total for grain and straw, NUEc) and grain (NUE<sup>g</sup> ) were calculated according to Equations (3, 4), respectively:

$$NUE\_c = (N\_t - N\_c) / N\_{\text{applied}} \times 100\tag{3}$$

where N<sup>t</sup> is the crop N offtake from N (digestate or NH4NO3) treatment plots, N<sup>c</sup> is the mean crop N offtake from the control plots and Napplied is the N fertilizer applied to the plots. All units are in kg N ha−<sup>1</sup> .

$$NUE\_{\%} = (\_{\text{Ngt}} - N\_c) / N\_{\text{applied}} \times 100\tag{4}$$

where Ngt is the grain N offtake from N (digestate or NH4NO3) treatment plots, Ngc is the mean grain N offtake from the C plots and Napplied is the N fertilizer applied to the plots. All units are kg N ha−<sup>1</sup> .

### Statistical Analyses

A factorial analysis of variance (ANOVA) with two factors (site: HF and NW; and treatment: C, D, D+NI, AD, AD+NI) and a blocking factor was performed for cumulative NH<sup>3</sup> and N2O losses (expressed as % of the total N applied), grain and straw yield, N offtake (grain, straw and total), TGW, NUEc, and NUEg. Tukey's post-hoc was used to detect differences between sites and treatments. The t-test was performed to examine variation between the different properties of the digestate and acidified digestate used at both sites, except for pH and DM at NW where a Kruskal-Wallis non-parametric ANOVA was used. One-way ANOVA was used to compare NUEc, NUEg, grain yields, and plant production for the different "digestate treatments" and fertilizer N rates at HF. Cumulative NH<sup>3</sup> losses, N offtake by straw, and plant production were log transformed to ensure the requirements for ANOVA. Statistical significance is defined as p < 0.05. In addition, linear (without including the highest dose) and quadratic regressions were derived for yield and total crop production for the fertilizer N response plots (0 to 300 kg N ha−<sup>1</sup> ) to calculate the fertilizer N replacement value of the different digestate treatments. All statistical analyses were performed using SPSS v22.0 (IBM Corp., Armonk, NY).

### RESULTS

### Soil Analyses

Soil pH, total C, total N, C:N ratio and mineralizable N were higher in HF than NW (**Table 1**). Changes in soil pH, NH<sup>+</sup> 4 , and NO<sup>−</sup> 3 during the experiment are presented in **Figure 1** for both sites. Soil pH decreased following addition of the acidified digestate treatments (AD and AD+NI were between 5.2–6.3 at HF and AD+NI between 4.6–5.4 at NW) relative to the non-acidified treatments (C, D, and D+NI; 6.0–6.8 at HF, and C and D between 5.0–6.0 at NW; **Figures 1A,B**). This effect was observed a few days after digestate application and pH remained lower until harvest at both sites, reaching maximum difference 1 month before harvest (around 1.0 pH unit). The application of the digestate also led to changes in soil EC, with the greatest values for AD and AD+NI, followed by D and D+NI and, finally, by C (Figure S2).

Peaks in soil NH<sup>+</sup> 4 -N content were observed in the first month after digestate application (**Figures 1C,D**). Ammonium contents between 150 and 200 mg N kg−<sup>1</sup> were found at NW, which were double that measured at HF (80 mg N kg−<sup>1</sup> ) in this period. Following the initial peaks, a general decrease in soil NH<sup>+</sup> 4 -N content was observed with time, with a faster rate of decrease at HF. Soil NH<sup>+</sup> 4 -N contents were greatest for AD+NI and AD. A similar trend occurred for soil NO<sup>−</sup> 3 -N content (**Figures 1E,F**), however, the greatest NO<sup>−</sup> 3 -N concentrations were observed for treatments without the nitrification inhibitor (D and AD), within the first month following digestate application. Peak soil NO<sup>−</sup> 3 -N contents were ∼90 and 60 mg N kg−<sup>1</sup> for D and AD respectively at NW, and 13 and 10 mg N kg−<sup>1</sup> for D and AD, and C respectively, just 1 day after digestate application at HF. Soil NO<sup>−</sup> 3 -N contents for D+NI and AD+NI were more constant through the whole experiment and their values were comparable with other treatments in the last 2 months at both sites. Soil NH<sup>+</sup> 4 -N and NO<sup>−</sup> 3 -N contents were below 20 mg N kg−<sup>1</sup> for the controls at both sites throughout the experiment (**Figures 1C–F**).

### Nitrogen Losses

The percentage of total N applied lost as NH<sup>3</sup> and N2O averaged across all "digestate treatments" were significantly higher at HF (17.4 and 0.45% of the total N applied, respectively) than at NW (11.6 and 0.13% of the total N applied, respectively; **Table 4**). The majority of the NH<sup>3</sup> loss occurred during the first and second days following digestate application (**Figure 2**). Cumulative NH<sup>3</sup> volatilization losses were significantly reduced by the acidification of the digestate (P < 0.001), being 1.5% of the total N applied for the mean of AD and AD+NI treatments and 27.6% of the total N applied for the mean of D and D+NI treatments across both sites. Mean N2O loss from digestate treatments with the nitrification inhibitor (D+NI and AD+NI) was 0.17 kg N ha−<sup>1</sup> and 0.35 kg N ha−<sup>1</sup> for those without the nitrification inhibitor, a >50% reduction although the differences were not significant (P = 0.097, **Table 4**). The peaks in daily N2O emissions (Figure S3) were related to higher WFPS (Figure S1), especially for the "digestate" treatments at HF. The airline to one of the automatic chambers used to determine N2O fluxes at HF appeared to be blocked (Figure S4, chamber 2 for AD treatment), so its values were replaced by the mean value of the other two chambers from the same treatment for statistical analysis because only three chambers per treatment were used

FIGURE 1 | Time course of soil pH, soil NH<sup>+</sup> 4 and NO<sup>−</sup> 3 contents (means ± standard error) at Henfaes (HF; A,C,E) and North Wyke (NW; B,D, F) following digestate application. C, control; D, digestate; D+NI, digestate plus nitrification inhibitors; AD, acidified digestate; AD+NI, acidified digestate plus nitrification inhibitors; n = 5 for each treatment.

at HF. A similar ranking was obtained for cumulative N2O emissions at both sites (AD > D > AD+NI > D+NI > C, Figure S4).

### Yield, Nitrogen Offtake, Nitrogen Use Efficiency (NUE), and Inorganic Nitrogen Replacement

Grain yield and total crop production were influenced by the site in a different way for the control and digestate treatments (**Table 4**). Higher mean grain yields (P = 0.004) were measured at NW (8.93 ± 0.37 t ha−<sup>1</sup> ) than at HF (7.55 ± 0.44 t ha−<sup>1</sup> ). The same effect was observed for plant production (P < 0.001), 15.36 ± 0.45 t ha−<sup>1</sup> at NW and 11.37 ± 0.50 t ha−<sup>1</sup> at HF. The application of the different digestate treatments resulted in a significant increase in grain yield (P < 0.001) and total crop production (P < 0.001) in relation to the control treatment (grain yield, 5.47 ± 0.64 t ha−<sup>1</sup> , and plant production, 11.09 ± 1.36 t ha−<sup>1</sup> ) without N application but no significant differences were observed between the "digestate" treatments (grain yield,


between 8.31 ± 0.47 t ha−<sup>1</sup> for D+NI and 9.52 ± 0.49 t ha−<sup>1</sup> for AD, and plant production, between 16.21 ± 0.66 t ha−<sup>1</sup> for D+NI and 16.95 ± 1.06 t ha−<sup>1</sup> for AD+NI; **Table 4**). The interaction site × treatment was not significant for grain yield because an analogous trend for the "digestate" treatments was observed at both sites (**Figure 3**), however, it was significant for plant production (P = 0.024), because the highest mean values were observed for AD > AD+NI > D > D+NI > C at HF, and for AD+NI > D+NI > D > AD > C at NW (**Figure 4**). Thousand grain weight was lower (P < 0.001) at NW (43.9 ± 0.5 g) than at HF (47.8 ± 0.4 g) and was reduced (P < 0.001) in the following order in relation to the different "digestate treatments," C >, D+NI >, D >, AD > AD+NI (**Table 4**).

The mean grain N offtake and crop N offtake the means across the 'digestate' treatments were significantly higher at NW than at HF, but NUE<sup>g</sup> and NUE<sup>c</sup> were lower than that at HF (**Table 4**). Digestate application significantly increased grain N offtake and crop N offtake in comparison to the control, as expected (**Table 4**). Highest N offtake values were from AD followed by AD+NI for both grain and crop, and NUE<sup>g</sup> and NUE<sup>c</sup> were also highest for these treatments although differences were not significant (**Table 4**). There were no significant site × treatment interactions for N offtake, NUE<sup>g</sup> or NUE<sup>c</sup> .

Fertilizer replacement value was significantly greatest for AD and significantly least for D+NI for both grain and total crop yield at HF (Table S3) and followed the same order whether fitting a linear (or quadratic function; P < 0.100 for data calculated with both fittings): AD (168 ± 20 kg N ha−<sup>1</sup> ) > AD+NI (154 ± 18 kg N ha−<sup>1</sup> ) > D (137 ± 9 kg N ha−<sup>1</sup> ) > D+NI (111.1 ± 7.7 kg N ha−<sup>1</sup> ), for the linear approach and grain N fertilizer replacement. The differences between the linear and the quadratic approaches for the calculation of the inorganic N replacement by digestate were 4.5% for grain yield and 8.1% for total crop production. However, when fertilizer replacement was calculated as a function of the total N applied per treatment, the differences were not significant between the "digestate treatments" (Table S3) and ranged between 84.2 ± 5.9% for D+NI treatment and 103.6 ± 6.9% for D treatment. At NW, the fertilizer N response plots were severely affected by lodging and data were subsequently not used.

# DISCUSSION Digestate and Soil Characteristics

The properties of the digestate used in our field experiments were comparable to these reported elsewhere (Möller and Müller, 2012; Nkoa, 2014): high pH (>7.0), low DM content, high proportion of total N as NH<sup>+</sup> 4 -N, negligible NO<sup>−</sup> 3 -N content, and similar total N, P and K contents. In general, the application of digestate does not alter soil properties in the short-term but can increase microbial activity and biomass (Melero et al., 2016), N mineralization and NH<sup>3</sup> oxidation (Odlare et al., 2008), soil mineral N content (Möller et al., 2008), hydraulic conductivity,

and decrease soil bulk density (Garg et al., 2005), in relation to undigested feedstocks.

Unfortunately, the N loading rates for the different treatments in our study were not the same, and not exactly the 190 kg N ha−<sup>1</sup> (equivalent) that we targeted. This is something that has been reported in other studies (Pezzolla et al., 2012; Riva et al., 2016). To address this, the N losses were presented as a percentage of the total N applied. The variability of the feedstocks, digestate handling, transport, and storage in the local biogas plants and in tanks in the fields before the application could have caused changes in the digestate N content between the initial sampling time and the time of land application. It is well known that open stores (Wang et al., 2014), and the lack of semi-permeable materials to cover the tanks (Börjesson and Berglund, 2007) and protective gas-tight layers (Battini et al., 2014) can lead to large N losses, predominantly via NH<sup>3</sup> volatilization (Petersen and Sørensen, 2008; Fangueiro et al., 2015a), in comparison with the undigested feedstocks. Moreover, the pH of our digestate was 8.24 ± 0.01 at HF and 8.05 ± 0.01 at NW, and according to Muck and Steenhuis (1982), very high losses of NH<sup>3</sup> from digestate occurs above pH 8.0, and small losses below pH 6.0. The lower pH of the acidified digestate applied to our fields (5.40 ± 0.01 at HF and 2.88 ± 0.06) and the time between acidification and field application of the digestate (1 week at HF and 2 days at NW) would also contribute in part to explain the higher N content in relation to the non-acidified digestate because the equilibrium NH<sup>+</sup> 4 / NH<sup>3</sup> favors volatilization at higher pH (Möller, 2015). The tanks used for the storage of the digestate in our fields before application had a simple thread lid and were only loosely fixed to prevent pressurization of the tanks, so were not gas-tight, which may have contributed to greater N loss via NH<sup>3</sup> volatilization, especially from the non-acidified treatment (Möller, 2015).

The different dry matter (DM) content (%) of the digestate applied at HF (3.08 ± 0.19 for D and 5.08 ± 0.04 for AD) and at NW (7.52 ± 0.08 for D and 9.66 ± 0.03 for AD) explains the variable distribution of the applied digestate at both sites following simulated bandspreading. The higher DM content at NW resulted in discrete bands of digestate, whereas the lower DM content at HF meant that the digestate did not remain within bands resulting in a more homogeneous distribution covering almost the whole surface of the plots that received digestate. The higher surface area of digestate in contact with the air at HF helps to explain the higher N losses at HF (mainly as NH<sup>3</sup> but also as N2O despite the lower WFPS at HF, Figure S1 and **Table 4**), and differences in soil NH<sup>+</sup> 4 -N and NO<sup>−</sup> 3 -N contents between NW and HF (**Figure 1**), especially during the first months after digestate application. In this study, the higher DM content for NW digestate compared with that at HF did not result in higher NH<sup>3</sup> emissions as would be expected for slurries (e.g., Misselbrook et al., 2004), suggesting that other factors (such as the increased emitting surface area) were important in controlling NH<sup>3</sup> volatilization. The greater post-harvest soil NO<sup>−</sup> 3 -N content at NW could indicate more risk of leaching than at HF (**Figures 1E,F**).

# Nitrogen Losses: Acidification and Nitrification Inhibitors

Acidification to a pH of <6.0 reduced NH<sup>3</sup> volatilization to <2.0% of the total N applied (AD and AD+NI plots), a similar reduction to that reported by other authors when non-acidified digestate or slurries were injected into the soil (Fangueiro et al., 2015b; Riva et al., 2016; Baral et al., 2017). These values were significantly lower than when the digestate was not acidified (D and D+NI), resulting in NH<sup>3</sup> losses of more than 27% of the total applied N (**Table 4**). High NH<sup>+</sup> 4 content and pH of the digestate facilitate N losses via NH<sup>3</sup> volatilization (Fangueiro et al., 2015a; Möller, 2015) that can account up to more than a 40% of the total applied N if not managed carefully (e.g., Riva et al., 2016; Nicholson et al., 2017). Our results for the digestate treatments when the digestate was not acidified (D and D+NI) are consistent with these studies. Ammonia is quickly emitted, normally during the first few hours after slurry (Ni et al., 2012) or digestate (**Figure 2** of this experiment; Nicholson et al., 2017) are applied. Consequently, measures to reduce its emission should be focused in the first few hours after application (e.g., rapid incorporation) and on production or storage phases of the digestate, to reduce N losses at the different phases. The large, significant decrease in N losses from NH<sup>3</sup> volatilization we measured following acidification of digestate (ca. 95% reduction compared with non-acidified digestate) demonstrates the effectiveness of this method to control and reduce these emissions, addressing a key knowledge gap identified by Nicholson et al. (2017). Although more experiments under different weather conditions, physico-chemical soil properties and crops are necessary, our study supports the use of acidification of food based digestate, consistent with this technique being called the Best Available Technology (BAT) for reducing NH<sup>3</sup> losses from slurries in some countries (Kai et al., 2008). Rapid soil incorporation has also been shown to reduce NH<sup>3</sup> losses by up to a 85% when following application of food waste based digestate (Tiwary et al., 2015) but it could increase N losses in the form of N2O as observed for slurries (Thorman, 2011).

When the pH of the digestate is >6.00 the high soil NH<sup>+</sup> 4 contents after the application of the digestate stimulate nitrification (Muck and Steenhuis, 1982), and, consequently, N2O emissions. The intensive frequency of N2O sampling and analysis at HF (Figure S3), and the higher mineralizable N measured at HF (**Table 1**) might explain the greater cumulative N2O losses compared to NW, as some N2O peaks may have been missed because of the lower frequency of sampling at NW. Nitrification could have been responsible for most of the N2O emissions because the WFPS was always <50% at both sites (between ≈10 and 25% at HF and between 15 and 50% at NW, Figures S1A,B) and the N2O peaks were related to higher WFPS in soil (Figures S1, S3; Zhu et al., 2013).

Nitrous oxide emissions as a result of denitrification are stimulated after the application of organic amendments with a large content of C (Rochette et al., 2000). Therefore, we do not discard that denitrification was, in part, responsible of some N2O emissions observed after digestate application (Figure S3), although the initial NO<sup>−</sup> 3 -N contents in soil were

at a probability level of 0.05.

lower (**Figures 1E,F**) than in a previous study by Fangueiro et al. (2015b) where high soil NO<sup>−</sup> 3 -N content (c. 80 mg kg−<sup>1</sup> ) resulted in significant N2O emissions. In addition, hot spots where both nitrification and denitrification processes occur are created in soil after the addition of organic manures, including even when bulk WFPS is below 50%, resulting in N2O emissions (MarkFoged et al., 2011; Zhu et al., 2015). Baral et al. (2017) found that the highest N2O emissions were produced at WFPS between 53 and 56% in a field experiment in which spring barley was fertilized with manure and digestate and that coupled nitrification-denitrification was the source of these emissions.

A decrease in the nitrification process was observed for the treatments in which DMPP was added; i.e. higher NH<sup>+</sup> 4 -N and lower NO<sup>−</sup> 3 -N contents were measured at both experimental sites for D+NI and AD+NI treatments during the experiment (**Figures 1C–F**). The addition of DMPP resulted in a reduction of N2O emission of up to a 50% in comparison to the digestate without the nitrification inhibitor (D and AD), although the differences were not significant (P = 0.097, **Table 4**). The use of nitrification inhibitors such as DMPP and dicyandiamide (DCD) have been proved to be an effective strategy to reduce N losses from soils where mineral fertilizers (Liu et al., 2013) or slurries (Vallejo et al., 2005; Fangueiro et al., 2016) are applied. The acidification of slurries has also been shown to delay nitrification in some soils (Fangueiro et al., 2013) but not in others, e.g., soils with a high buffering capacity where the soil pH was not altered after the application of the acidified digestate (Fangueiro et al., 2016). Owusu-Twum et al. (2017) recently demonstrated in a short-term experiment under controlled conditions that acidification of slurries could significantly reduce N2O emissions, but to a lesser extent than when DMPP was used. We found some evidence of a delay in the nitrification process for the acidified digestate, where peak soil NO<sup>−</sup> 3 -N content was observed a few weeks later than for unacidified digestate at HF (**Figure 1E**), and soil NH<sup>+</sup> 4 -N contents were higher for AD than for D on the majority of measurement occasions (**Figures 1C,D**), although this could also be attributed to the initial higher NH<sup>+</sup> 4 -N contents of the acidified digestate (**Table 3**). This inhibition of nitrification could have been caused by the decrease in soil pH after spreading the acidified digestate, an effect that was persistent until the end of the experiment, because the population and activity of denitrifying bacteria is affected by soil pH (Gandhapudi et al., 2006). However, acidification did not alter N2O emissions (these were only affected by the addition of DMPP). The presence of a substantial amount of C and inorganic N could have promoted completed denitrification to N<sup>2</sup> for AD and AD+NI treatments (where the nitrifying bacteria activity could have been inhibited by acidification) as indicated by Pezzolla et al. (2012) with comparable WFPS values for soils amended with digestate. The percentage of applied N lost via N2O in our experiment ranged between 0.13 and 0.45% (**Table 4**), in accordance with 0.10–0.41% calculated by Baral et al. (2017) and with 0.45 ± 0.15% reported by Nicholson et al. (2017) under comparable conditions, all lower than the 1% default IPCC emission factor (IPCC, 2006).

# Nitrogen Uptake, Nitrogen Use Efficiency, Fertilizer Replacement Rates, and Yields

Although grain and crop N offtakes were improved when the applied digestate was acidified, the differences were not significant for NUE<sup>g</sup> or NUE<sup>c</sup> (**Table 4**). The results for HF indicate that digestate can be an effective replacement for inorganic fertilizers such as NH4NO<sup>3</sup> in terms of crop production (**Figures 3**, **4**). These results are in agreement with similar experiments: Walsh et al. (2012) for a grassland, Riva et al. (2016) for a maize silage crop in which they used manure- and cropbased digestates, Furukawa and Hasegawa (2006) for spinach, Haraldsen et al. (2011) for barley, and Pezzolla et al. (2012) for a grassland using food waste based digestate. On the one hand, yields for D and D+NI treatments were similar to those obtained for doses of inorganic N of 136.7 ± 9.1 and 111.1 ± 7.7 kg N ha−<sup>1</sup> , and the mean values were higher when the digestate was acidified, i.e., AD and AD+NI, which produced similar yields to doses of 168.3 ± 20 and 154.2 ± 17.5 kg N ha−<sup>1</sup> at HF. However, no significant differences were found between the different "digestate treatments" (D, D+NI, AD, and AD+NI) when these fertilizer replacement values are based on the total N applied with each "digestate treatment" at HF. The reduction of yields observed in our experiment (only for the mean values, not significantly) when NI were added to the digestate in comparison to the treatments without NI agrees with Misselbrook et al. (2014) but not with the increase in yields reported by Abalos et al. (2014) in their meta analysis. However, in order to achieve effective mitigation of N losses and fertilizer replacement values, digestate should be acidified or rapidly incorporated into the soil following application, as shown in this experiment and by Möller et al. (2008), respectively.

# CONCLUSIONS

Acidification of digestate and the inclusion of a nitrification inhibitor are good strategies for the utilization of food waste based digestates because they contributed to the mitigation of N losses following application to a winter wheat crop. Without acidification, NH<sup>3</sup> volatilization accounted for almost a 30% of the total N applied in digestate. This emission was reduced by 95% with acidification. We demonstrated that wheat yields when acidified digestate was applied at HF (176 kg N ha−<sup>1</sup> ) were similar to these produced by an inorganic N form (NH4NO3) applied at a rate of 154–168 kg N ha−<sup>1</sup> . Acidification of the digestate seems to be an effective technique after digestate spreading, producing higher mean yields and inorganic N replacements than when the digestate is not acidified. Without the acidification of the digestate, NH<sup>3</sup> volatilization accounted for almost a 30% of the total N applied resulting in a serious economic cost and environmental damage. This study encourages the use of digestate from the anaerobic digestion of food waste alongside acidification and with the addition of a nitrification inhibitor, as an environmentally sound option for N management. However, the reduction in soil pH that was measured in the acidified treatments at both sites, suggest that the effect of slurry and digestate acidification on soil quality and function needs to be assessed in the long-term.

# AUTHOR CONTRIBUTIONS

AS-R, AC, RS, DC, DJ, and TM: Conceptualization; AS-R, AC, JH, KS, and JC: Formal analysis; DC, DJ, and TM: Funding acquisition; AS-R, and AC: Investigation; AS-R, AC, RS, JH, KS, and JC: Methodology; DC, DJ, and TM: Supervision; AS-R, Wrote the manuscript; All the authors review and approved the last version of the manuscript.

# ACKNOWLEDGMENTS

This work was supported by the UK-China Virtual Joint Centre for Agricultural Nitrogen (CINAg, BB/N013468/1), which is jointly supported by the Newton Fund, via UK BBSRC and NERC, and the Chinese Ministry of Science and Technology, and also it was undertaken as part of NUCLEUS, a virtual joint centre to deliver enhanced N-use efficiency via an integrated soil–plant systems approach for the United Kingdom and Brazil. NUCLEUS is funded in Brazil by FAPESP— São Paulo Research Foundation [grant number 2015/50305- 8]; FAPEG—Goiás Research Foundation [grant number 2015- 10267001479]; and FAPEMA—Maranhão Research Foundation [grant number RCUK-02771/16]; and in the United Kingdom by the Biotechnology and Biological Sciences Research Council [grant number BB/N013201/1] under the Newton Fund scheme. We would like to thank Sabine Reinsch (CEH-Bangor), Marina Poyatos, Alvaro Uceda, Juan Espinasa, and Carmen Millan for their help applying the digestate and with different samplings, Llinos Hughes and Mark Hughes for technical support, and Inma Robinson (CEH-Bangor) for particle size distribution analysis. Finally, we would like to thank Fre-Energy Ltd (Wrexham, UK) and Andigestion Ltd (Holsworthy, UK) for providing us with the digestate.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fsufs. 2018.00035/full#supplementary-material

# REFERENCES


DEFRA (2010) The Fertilizer Manual (RB 209), 8th Edn. Norwich: TSO, 253.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Sánchez-Rodríguez, Carswell, Shaw, Hunt, Saunders, Cotton, Chadwick, Jones and Misselbrook. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Fertilizing Potential of Separated Biogas Digestates in Annual and Perennial Biomass Production Systems

### Andrea Ehmann\*, Ulrich Thumm and Iris Lewandowski

Department Biobased Products and Energy Crops, Institute of Crop Science, University of Hohenheim, Stuttgart, Germany

### Edited by:

Tom Misselbrook, Rothamsted Research (BBSRC), United Kingdom

### Reviewed by:

M. Angeles de la Rubia, Universidad Autonoma de Madrid, Spain Laura Zavattaro, Università degli Studi di Torino, Italy

> \*Correspondence: Andrea Ehmann a.ehmann@uni-hohenheim.de

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 31 January 2018 Accepted: 06 April 2018 Published: 24 April 2018

### Citation:

Ehmann A, Thumm U and Lewandowski I (2018) Fertilizing Potential of Separated Biogas Digestates in Annual and Perennial Biomass Production Systems. Front. Sustain. Food Syst. 2:12. doi: 10.3389/fsufs.2018.00012 Digestates produced by the increasing number of biogas plants require appropriate treatment or recycling. This study investigates the fertilizing potential of separated biogas digestates. These contain valuable nutrients and can be used in agriculture to close the nutrient cycle. Multi-year field experiments were established at two challenging sites in south-west Germany in 2010; results from 6 years are shown here. The objectives were to determine (1) whether separated digestates can complement or substitute mineral fertilizers and (2) their effect on long-term yield performance in different biomass cropping and fertilization systems. The fertilizing performance was assessed in a split-plot design with four replications using three cropping systems: (1) perennial grassland; (2) intercropping of triticale and clover grass; (3) silage maize. Five N fertilization treatments were applied, each at 150 kg N ha−<sup>1</sup> :


The influences of site, cropping system, year and fertilization treatment were highly significant.The mineral fertilizer and combination "liquid digestate fraction + mineral fertilizer" mostly led to the highest quantitative biomass yields in all cropping systems at both sites. Fertilization with solid digestate fraction produced lowest yields in all fertilized plots, with results very often not significantly different from the untreated control. Maize achieved relatively high yields in years with favorable weather conditions; unfavorable conditions led to low yields. The grassland and intercropping systems were less susceptible to weather conditions, producing a more constant biomass supply irrespective of site, treatment and year. The separated biogas digestates were found to have a comparable effect to mineral fertilizer on biomass yield, but this varied with cropping system. In the intercropping system, complete substitution was possible. The solid fraction is more likely to contribute positively to soil humus in annual systems. In general, the combined application of digestate and mineral fertilizer is highly recommendable to meet crops' short- and long-term N demand, even on challenging sites. In this study, it allowed a mineral fertilizer input reduction of 66%.

Keywords: biogas digestates, fertilization, cropping systems, bioenergy, alternative biogas substrates, nutrient cycles

# INTRODUCTION

Power and heat generated from biogas provide a significant contribution to the increasing amount of bioenergy produced in Europe. Here, more than 17,300 biogas plants were counted in 2015 (EBA, 2016) with a primary production of 654 petajoules (Eurostat, 2015). The biogas sector has experienced a strong impetus in Germany in particular, supported by the German Renewable Energy Sources Act, which was introduced in 2000 and has since been modified several times. In 2017, there were more than 9,300 agricultural biogas plants operating in Germany alone (German Biogas Association, 2017), producing 116 petajoules electric power (BMWi, 2017) and an estimated 65.5 million cubic meters of biogas digestates (Möller and Müller, 2012).

Biogas digestates are the residues left from the anaerobic fermentation of organic matter, such as animal manure and plant biomass specifically grown for this purpose. Through the production of biogas (CH<sup>4</sup> and CO2) in the fermentation process, the amount of carbon is significantly (>50%) reduced (Tambone et al., 2009). Depending on the operating system (including pH and temperature) of the biogas plant, N can also be lost (as NH3) to a certain extent (Reinhold et al., 2004). However, most of the N and all other mineral elements contained in the input substrates remain in the biogas digestates (Vaneeckhaute et al., 2017). These include major plant nutrients such as phosphorus, potassium and calcium. Therefore, it is common practice to use biogas digestates as organic fertilizers (Alburquerque et al., 2012a), which at the same time saves costs for both mineral fertilizer and potential disposal of the digestates (up to 25 e t −1 , Rolink, 2013). The good fertilizing value of biogas digestates in comparison to mineral fertilizer has been confirmed in several studies (Formowitz and Fritz, 2010; Gunnarsson et al., 2010; Walsh et al., 2012; Barbosa et al., 2014). Also, the remaining carbon bound in the organic matter helps to maintain or even increase soil organic matter (Möller, 2015), which is particularly valuable in marginal soils (Nabel et al., 2014). This effect can be considerable in annual cropping systems.

Although the use of biogas digestates as organic fertilizer seems an efficient way of closing nutrient cycles in agriculture and reducing external inputs of mineral fertilizer, several potential drawbacks need to be considered in order to optimize the efficiency and environmental performance of biomass production systems.

The first is the distribution of digestates. They accumulate at biogas plants and their high water content (>90%) limits their ability to be stored and transported. For this reason, many farms separate the digestates on site in order to reduce the water content and volume and increase transportability (Hjorth et al., 2010). The processing is mostly done using screw press separators, a robust and simple on-farm technology. The separated liquid fraction is characterized by high N (mainly in the form of directly plant-available ammonium) and potassium contents and a total solids content of below 5% (Gutser et al., 2005; Möller et al., 2009; Nkoa, 2014). The solid fraction contains approx. 20% of the total N, a third of the total phosphorus and 15% of the potassium and up to 35% total solids (Rolink, 2013; Vaneeckhaute et al., 2017). Farmers often collaborate with the (more or less) neighboring farms supplying them with biomass in order to optimize operation capacity, particularly in larger biogas plants. The biomass suppliers receive digestates in return, thus helping to manage any oversupply.

Second, the composition of biogas digestates can vary due to variations in substrate supply and, as described above, differs between the solid and liquid fractions. For this reason, farmers are often unsure about the performance of digestates as organic fertilizers and various studies have shown that their fertilizing effect is not always as predictable as that of mineral fertilizer (Möller, 2009; Hjorth et al., 2010; Odlare et al., 2011). Some have reported that such variation in organic fertilizers can lead to fluctuation and/or reduction in biomass yield (e.g. Alburquerque et al., 2012b; Sieling et al., 2013). In order to guarantee biomass yield stability, the yield effect of biogas digestates and their liquid and solid fractions needs to be assessed. One option for overcoming this shortcoming of organic fertilizers may be the use of digestates in combination with mineral fertilizer or gradual supplementation of mineral fertilizers by digestates. To test this, two combinations of digestates and mineral fertilizer were included in this study.

Third, decomposition of the organic matter during the fermentation process leads to an enrichment of NH4-N in biogas digestates (Reinhold et al., 2004). This increases the probability of gaseous N being lost during storage and application. To avoid such losses, field applications of digestates should be timed to meet the crops' nutrient demand and low-emission application techniques should be used. Nutrient demands and optimal fertilization systems very much depend on the type of cropping system. Application techniques and timing, and also the fertilizing effects of organic fertilizers all differ between annual cropping systems, such as maize, and perennial cropping systems, such as grassland (Svoboda et al., 2013). In grassland for example, the immediate effect of an organic fertilizer is usually not very pronounced due to the high organic matter content of the soil (Conant et al., 2017).

In farming practice, the nutrient-rich biogas digestates are generally applied as fertilizer to crops grown for biomass to be used as biogas feedstock. Three such crops are considered in this study: silage maize, grass and winter triticale. Silage maize has been by far the most important biogas crop in Central Europe (Herrmann, 2013), especially Germany (73%, FNR, 2017), for quite some time now. There are several reasons for this including high biomass and methane yields, relatively simple production system, good availability of the required technical equipment and low demands for plant protection (Herrmann et al., 2017). Another aspect is the availability of a wide range of varieties for various site conditions and applications. In Germany however, the proportion of maize in biogas substrate has been limited to a maximum of currently 50% (Bundesministerium der Justiz und für Verbraucherschutz, 2017). This stems from ecological concerns, for example the fact that maize is often cultivated in large-scale monocultures (and unfortunately often also in combination with poor farming practices), leading to an anticipated increase in pests in the future as well as landscape image issues. In addition, experience has shown that maize cultivation is highly susceptible to N losses (via leaching and gaseous emissions) and soil erosion (Taube and Herrmann, 2009; Svoboda et al., 2013). This has led to a call for alternatives to maize as biogas substrate and for diversification in crop rotations (von Cossel et al., 2017). As a result, alternative and more environmentally benign biomass supply systems are currently being sought, including semi- to fully perennial cropping systems.

Permanent grassland is a fully perennial cropping system and a frequent form of land use, especially in agriculturally disadvantaged regions (Huyghe et al., 2014). Cool temperatures and/or a limited vegetation period render them less productive for maize cultivation. Those with a good water supply are very suitable for forage cropping. On such sites, grassland can achieve top yields, comparable to or sometimes even outperforming those of silage maize (Hartmann and Sticksel, 2010). The biomass from grassland (and also clover grass) can be used as animal feed. Any that is not used for feed, e.g., the second and potential following cuts, can be ensiled and digested in a biogas plant (Hartmann et al., 2011). At 12%, grass silage is the second most used biogas crop substrate in Germany (FNR, 2017).

Whole-crop cereals, notably winter triticale, are the third most frequently used biogas crop substrate (8%, FNR, 2017). Winter triticale has a high biomass yield potential and, as a winter cereal crop, can form a valuable part of the crop rotation (Sticksel, 2010). Its ability to resist unfavorable biotic and abiotic environmental factors allows good yields even at marginal sites (Martinek et al., 2008). In our study, it was harvested as whole green crop in early summer. This harvest time makes it difficult to grow a second crop in the same year (Sticksel, 2010). Thus, when grown in an intercropping system, it is most efficient to establish clover grass by undersowing in spring. In this study, the intercropping of triticale and clover grass is considered a "semi-perennial system." It has positive effects on soil erosion control and N use efficiency due to the year-round soil coverage and the integration of legumes in the crop rotation.

The objectives of this study were to determine whether separated digestates can complement or substitute mineral fertilizers and whether/how they affect the long-term yield performance in different biomass cropping systems.

The research approach was set up to test the following hypotheses:


These hypotheses were tested by means of a multi-factorial, long-term field experiment allowing a comparison of different fertilization treatments in three cropping systems at two sites. For this purpose, three typical biogas substrate cropping systems (maize; intercropping of winter triticale with clover grass; and grassland) were established on two locations close to a biogas plant. These were chosen to represent an annual, a semi-perennial and a perennial system, respectively. The sites are located at the base and the top of the mountainous region of the Swabian Alb in south-west Germany, both of which display agriculturally challenging conditions (soil quality/growing season, respectively). The fertilizing effects of biogas digestates on these cropping systems were tested using the separated liquid and solid digestate fractions alone and also in combination with mineral fertilizer.

### MATERIALS AND METHODS

### Site Description

In 2010, two multi-year field experiments were established on marginal sites belonging to the field research station of the University of Hohenheim in south-west Germany: one at the base ("Valley," 48.47◦ latitude, 9.27◦ longitude, approximately 480 m above sea level, average annual air temperature 10.0◦C, average annual rainfall 779 mm) and the other at the top ("Hill," 48.47◦ latitude, 9.30◦ longitude, approximately 700 m above sea level, 7.1◦C, 935 mm) of the mountainous region of the Swabian Alb; approximately 35 km south of Stuttgart.

The soil at the "Valley" site is classified as lithoidal clay rendzina with a depth of approximately 0.6 m. The soil at the "Hill" site is a silty clayey loam with a depth of over 1.0 m. The climate data relevant for the field study (2012–2017) are shown in **Figure 1**. Data for the "Valley" site are taken from the nearest weather station at Metzingen, 48.55◦ latitude, 9.30◦ longitude, 391 m above sea level.

### Experimental Approach

The fertilizing performance of separated biogas digestates was assessed using three cropping systems: (1) perennial grassland; (2) intercropping of winter triticale and clover grass; (3) silage maize.

The grassland plots were established in April 2010 using a grassland seed mixture for 3–4 cuts per year (28% Lolium perenne, 19% Festuca pratensis, 19% Phleum pretense, 13% Poa

pratensis, 6% Festuca rubra, 6% Dactylis glomerata, 9% Trifolium repens; LAZBW Aulendorf, Germany) sown at a rate of 32 kg ha−<sup>1</sup> . Reseeding was carried out in August 2014 at a rate of 23 kg ha−<sup>1</sup> using a mixture specifically designed for less favorable areas (32% Lolium perenne, 20% Phleum pratense, 16% Poa pratensis, 16% Dactylis glomerata, 4% Alopecurus pratensis, 12% Trifolium repens; LAZBW Aulendorf, Germany) with the aim of maintaining grass cover and counteracting increasing gaps.

The winter triticale (x Triticosecale var. "Tarzan") plots were generally sown in the first week of October at a rate of 300 seeds m−². Clover grass was undersown in the triticale in March/April of the years 2013, 2015 and 2017 at a rate of 30 kg ha−<sup>1</sup> using a mixture consisting of 83% Italian ryegrass (Lolium multiflorum L. var. "Tarandus") and 17% red clover (Trifolium pratense L. var. "Titus"). For reasons of clarity, it is referred to as "clover grass" instead of "clover grass mixture." After the last clover grass cut, plots were cultivated with a rotary hoe (8 cm) and chisel plow (16 cm) and then prepared for sowing triticale by rotary harrow (12 cm).

The maize (Zea mays L.) plots were sown after seedbed preparation with a rotary harrow (12 cm) at a rate of 13 seeds m−² in rows 0.75 m apart as soon as reasonable, mostly in the first half of May. Varieties were selected according to the vegetation period at each site: "Ronaldinio" (FAO 240) for "Valley" and "Amadeo" (FAO 220) for "Hill." From 2015 onwards, these were switched to newer varieties with the same FAO numbers, respectively ("Frederico" for "Valley" and "Colisee" for "Hill"). Maize seeds were provided by KWS Saat SE, Einbeck, Germany. Soil tillage included stubble cultivation with a chisel plow (16–18 cm) immediately after harvest and plowing (20 cm) later on.

The three crops were fertilized with separated biogas digestates in four different variants (**Table 1**). The digestates were obtained from a 355 kilowattelectric biogas plant at the research station, fed mainly with animal manure and maize silage. Solid/liquid separation was performed with a screw press separator. A mineral fertilizer and an untreated control were included for comparison. All treatments except the control were applied at 150 kg N ha−<sup>1</sup> ; amounts and timing are summarized in **Table 2**. Residual plant-available nitrogen (Nmin), phosphorus, potassium, calcium, magnesium and the pH in the soil were measured every spring and fall to be used for subsequent research analysis (for methods, see Ehmann et al., 2017). Results from the initial soil sampling (0–30 cm) are summarized in **Table 3**.

Before each application, the NH<sup>+</sup> 4 content of the digestates was determined to take account of slight variations over time. Each time, two subsamples were taken; one was analyzed directly using a Quantofix N volumeter (Van Kessel and Reeves, 2000), the other was stored at −18◦C and analyzed later in the lab (DIN 38406-E5-2) to validate the first measurement.

**Table 4** shows the average NH<sup>+</sup> 4 concentrations of the digestates (values for 2012–2017), together with concentrations of other nutrients and pH (values for 2013–2015).

Applications were split into 2-3 portions to suit the crops' requirements as optimally as possible (**Table 2**). In grassland and clover grass, the initial portion was usually applied in spring and the subsequent portions after cutting. Where possible, the digestates were incorporated immediately after


CAN, calcium ammonium nitrate.

application using a harrow (10 cm) to minimize N losses. In the combined treatments, digestates and mineral fertilizer were applied approximately 1 week apart from each other.

The experiments were established in a split-plot design with four replications, resulting in 72 plots (32 m²) at each site. Main plots were the cropping systems and subplots the fertilization treatments. Treatments were randomized for each site separately.

Herbicides and fungicides were only applied when necessary and then according to good agricultural practice.

The grassland plots were cut three (in 2016 two) times per year according to good agricultural practice. The last sparse growth of each year was cut and removed from the plots, but not included in the yield.

For the intercropping plots, the harvesting regime was as follows: in the years 2012, 2014, and 2016, the clover grass was harvested three to four times; in 2013, 2015, and 2017 the winter triticale was harvested wholecrop around the early dough stage (BBCH 83) and the undersown clover grass in September or October.

The maize was harvested wholecrop with a plot-size field chopper around the stage of silage ripeness (BBCH 85) and a dry matter content of 30–35% TS when weather conditions were suitable.

Samples were taken from each cut and the dry matter biomass yield (DMY) determined by drying at 60◦C to constant weight.

### Statistical Analysis

A mixed model was developed for all traits using the following equation (Piepho et al., 2004):

```
L + C + F + L • C + L • F + C • F + L • C • F : Y + Y • L +
Y • L • R + Y • C + Y • F + Y • L • C + Y • L • F + Y
•C • F + Y • L • C • F + R • Y • L + C • R • Y • L
+ C • F • R • Y • L,
```
where C and F denote effects of the treatments "cropping system" and "fertilization," R, L, and Y denote effects of "replicate," "site," and "year," respectively. Interactions between the treatments "site" and "year" are denoted by a dot between the corresponding main effects. "R • Y • L + C • R • Y • L + C • F • R • Y • L" denotes replicate effects and effects of main and subplot error in each combination of site and year. Effects from different years are repeated measurements, therefore a first-order autocorrelation was fitted to them. Crop-by-fertilizerspecific variances were assumed but only fitted to sub-plot errors to avoid convergence problems. Fixed effects are given before the colon. To achieve homogeneous residual variances and normality of residuals, data were log-transformed. Both pre-requirements were checked graphically. Where an F-test revealed significant effects, a multiple t-test (α = 0.05) was performed. To create the letter display, the %mult macro (Piepho, 2012) was used.

Furthermore, cumulated system-by-site-by-fertilizer treatment estimates across years and their standard errors were calculated as a sum of single-year BLUPs (best linear unbiased prediction), or its standard errors, for each combination of system, site and fertilizer treatment. A single-year BLUP here refers to the sum of the least square estimate for one systemby-site-by-fertilizer treatment mean and the corresponding random year main effect and its interaction effects. Yield data was logarithmically transformed, therefore presented values are interpreted as medians. Thus, cumulated yield estimates were also made from the given model.

The data analysis was carried out with SAS software version 9.3 (SAS Institute Inc., Cary, NC, USA).

### RESULTS

Cumulative yields from the 6 years are presented here to compare the long-term yield performance of the cropping systems. **Table 5** shows the results of the statistical analysis of the main effects; these all had a significant influence except the interaction "site<sup>∗</sup> system."

### Perennial Grassland

At the "Valley" site, the highest DMY was obtained with mineral fertilizer, followed by the two combination treatments. The liquid digestates only and solid digestates only treatments were not significantly different from the combination treatments or the control (**Figure 2**). At the "Hill" site, the treatments appeared to be more efficient than at the "Valley" site. This is visible from the difference in DMY between control and treatments, both in cumulative as well as annual DMY. All treatments led to a higher DMY than in the control. The highest DMY was obtained with the two treatments containing solids, which were both significantly better than "liquid" only (**Figure 2**). As to be expected, the first of the usual three cuts made up the largest share of the annual yield.

It was noticeable that in 2017 the DMY was considerably lower at the "Hill" than at the "Valley" site. Here, the effect of decreasing DMY over the years becomes especially visible in the control plots (from 114 dt ha−<sup>1</sup> in 2012 down to 49 dt ha−<sup>1</sup> in 2017), whereas the "solid+" (average 89 dt ha−<sup>1</sup> ) and "solid" (average 88 dt ha−<sup>1</sup> ) plots were most stable. There were no particular fluctuations visible between the years. At the "Valley" site, the "mineral" plots showed the highest tendency toward decreasing yields.

### TABLE 2 | Amounts and timing of fertilization treatments.


CAN, calcium ammonium nitrate.

This general tendency toward declining yields over the years was observed at both sites. It was most likely due to gaps in the grass cover as a consequence of aging plots on the one hand and an infestation with field mice on the other. These gaps increased in frequency and size over time. Although reseeding was performed in August 2014, the plots did not recover satisfactorily as it was too dry during the following weeks. It was observed that the higher-value grass species in the initial seed mix (e.g., including perennial ryegrass Lolium perenne L., meadow fescue Festuca pratensis L., and timothy Phleum pratense L.) disappeared over time and were replaced by species of inferior quality. At the "Valley" site, this was predominantly rough bluegrass (Poa trivialis L.). In addition, the occurrence of broad-leaved dock (Rumex obtusifolius) reduced the quality of botanical composition at this site. Even the frequent cutting did not displace this persistent weed. The plots at the "Valley" site were also invaded by moss. At the "Hill" site, the initially established perennial ryegrass (Lolium perenne L.) was mainly replaced by cocksfoot (Dactylis glomerata L.). In general, the patchiness of the grass cover was less pronounced at the "Hill" than at the "Valley" site.

The dry matter content (DMC) of the grass samples was homogeneous and relatively low. At the "Valley" site, the average DMC of all samples was 20% (cut 1) and 23% (cuts 2–3); at the "Hill" site, 21 and 24%, respectively.

### Intercropping of Winter Triticale and Clover Grass

The DMY of the intercropping system was fairly homogeneous at both sites. This was particularly the case at the "Valley" site, where all treatments performed equally well and, with the exception of "solid," resulted in significantly higher DMY than the control (**Figure 3**).

At the "Hill" site, all treatments increased the yield compared to the control. The highest DMY was obtained with the "liquid+" treatment. This was significantly higher than with the "solid" treatment (**Figure 3**).

TABLE 3 | Initial soil characteristics (0–30 cm) at the two sites in September 2010.


Values in brackets indicate standard deviation.

DM, dry matter.

\*plant-available concentrations; analyzed with CAL extraction followed by flame photometer (P2O5), FIA measurement (K2O) according to OENORM L 1087:2012-12-01 and CaCl<sup>2</sup> extraction followed by AAS measurement (Mg) according to VDLUFA I A 6.2.4.1; soil pH was determined using a glass electrode after CaCl<sup>2</sup> extraction (DIN ISO 10390:2005); TOC (total organic carbon) analyzed according to DIN EN 15936:2012-11.

TABLE 4 | Characteristics of digestates and mineral fertilizer.


FM: fresh matter; DM: dry matter; STD: standard deviation.



DF, Degree of freedom; level of significance was p ≤ 0.05.

The highest yields were obtained at both sites in 2012 and 2013. After this, the yields decreased, but remained at a more or less constant level (120 dt ha−<sup>1</sup> at the "Valley" site, 112 dt ha−<sup>1</sup> at the "Hill" site). As expected, the yield difference between control and treatments was larger for triticale than for clover grass, indicating a more prominent fertilizing effect.

In 2015, the triticale DMY was reduced at the "Hill" site due to infestation with yellow rust (Puccinia striiformis).

**Figure 4** shows that the majority of the triticale plots were harvested too late and the DMC was higher than the optimal value, particularly at the "Valley" site. The DMC varied considerably more at the "Valley" than at the "Hill" site. In 2017, the average DMC at the "Valley" site was 63% with individual maximal values of more than 80%, whereas in the other years values were more within the normal range (46% in 2013, 36% in 2015).

### Silage Maize

At both sites, the highest maize DMY was obtained with mineral fertilizer; however, this was only significant at the "Hill" site (**Figure 5**).

At the "Valley" site, both combination treatments performed as well as the mineral fertilizer.

At the "Hill" site, all treatments with digestates except "solid" were comparable to each other and resulted in the second highest DMY after mineral fertilizer.

In general, the DMY standard deviations were higher and fluctuated more at the "Valley" than at the "Hill" site. This is likely due to the relatively high heterogeneity of the field conditions. In addition, problems with regrowth from the preceding crop Jerusalem artichoke (Helianthus tuberosus) led to massive yield reductions in certain plots, in one replication in particular, despite frequent manual weeding and the occasional herbicide application. It is interesting that the lowest standard deviation at the "Valley" site was found with mineral fertilizer indicating a reliable fertilizing effect, independent of external influences.

digestates and mineral fertilizer in comparison to unfertilized control. Error bars indicate standard deviation of accumulated yields. Means with identical letters are not significantly different from each other (n = 4, p ≤ 0.05). For explanation of treatments, see Table 1.

FIGURE 3 | Mean dry matter yield (DMY) of an intercropping system with winter triticale and undersown clover grass grown at the sites "Valley" and "Hill" from 2012–2017 fertilized with different treatments of separated biogas digestates and mineral fertilizer in comparison to unfertilized control. Checked parts of the columns indicate years in which only clover grass was harvested (2012, 2014, 2016). Error bars indicate standard deviation of accumulated yields. Means with identical letters are not significantly different from each other (n = 4, p ≤ 0.05). For explanation of treatment, see Table 1.

The effect of year was clearly recognizable. 2013 and 2015 were not good years for maize cultivation: wet and cold conditions in spring delayed sowing and/or germination; drought periods with high temperatures in summer months negatively influenced growth (see also **Figure 1** for weather data). In those years, the mineral fertilizer showed the best performance of all treatments. In years with weather conditions favorable for maize cultivation (2014, 2016, 2017) most digestate treatments worked equally well as mineral fertilizer.

The year 2016 was exceptional in that the spring was cold and wet, but there was a short favorable time slot which could be used for sowing. This was followed by a lot of rain in early summer before a dry, hot period set in. At the "Valley" site, this resulted in low DMY (on average 83 dt ha−<sup>1</sup> for fertilized plots), but a very satisfactory yield at the "Hill" site (118 dt ha−<sup>1</sup> ). Here, the higher altitude and thus lower average temperature, together with the deep soil, were an advantage. Consequently, the water supply lasted longer during the heatwave, ensuring better growth than at the "Valley" site.

The separated solid digestate variant led to the lowest yields of all treatments at both sites (**Figure 5**). This was visible in most years, but also for the accumulated yields. At the "Valley" site, it resulted in yields comparable to the control and to the "liquid" treatment (or even lower in absolute values). At the "Hill" site, it had a DMY higher than the control, but lower than the other treatments.

**Figure 6** shows that the majority of the maize plots were harvested with a dry matter content (DMC) within the optimal range of 30 to 35% TS. In general, DMC was lower and fluctuated less at the "Hill" site.

# DISCUSSION

Significant differences in yield performance were found between the annual, intercropping and perennial cropping systems subjected to the treatment variants. Interactions with site and year effects were also observed.

The highest and most stable biomass yields were found in the intercropping system with triticale and clover grass, irrespective of the site, treatment and year. This was followed by perennial grassland, which also proved to be relatively stable with regard to treatment and year, but provided lower DMY. Maize only produced high yields in years with favorable climatic conditions. Particularly in years with unfavorable conditions, the best maize yields were achieved with mineral fertilizer, whereas in normal years the DMY difference between treatments was small. Thus the influence of the year effect also varied between cropping systems.

In general, the "Valley" site had higher DMY, but the "Hill" site provided better conditions for growth during hot, dry periods due to the lower average temperature and longer water supply. We also observed that the treatments were more effective in terms of yield at the "Hill" site, as the DMY was significantly higher than the control on all fertilized plots in all three systems here.

## Effect of Fertilizer Treatments on Yield Performance of the Three Cropping Systems

The annual system responded most sensitively to influences of treatments, site and year. The yields were significantly influenced by all these factors. As a C4 crop, maize reacts relatively strongly to temperature fluctuations and requires favorable temperatures and sufficient water supply for germination and good establishment, especially if sown in late spring (Maton et al., 2007). In this study, the highest maize yields were achieved with mineral fertilizers, particularly in cooler years and at the cooler "Hill" site. This can be explained by the fact that mineral fertilizer application can be timed to provide plant-available N to coincide with the crop's requirements (Möller, 2009). The N availability of mineral fertilizer is also less dependent on climatic conditions, especially temperature and water supply, than organic fertilizer (Agehara and Warncke, 2005) and the share of mineral N is of course higher than in the digestates (**Table 4**). After sowing, maize first needs to build its root system and is highly dependent on rapidly plant-available N at exactly

the right time (Plénet and Lemaire, 1999). This can best be steered by the application of easily soluble mineral fertilizer.

The effect of more rapid N availability from mineral than from organic fertilizers was particularly evident in the years 2013 and 2015 when temperatures were lower than the long-term average and 2015 also had less than average rainfall during the growing season. In these years, the mineral fertilizer had significantly better results, especially at the "Hill" site, and the solids had a lower performance. In 2013 at the "Valley" site, the solids even resulted in a lower DMY than the unfertilized control. This may have been caused by initial N immobilization, which often occurs after the application of organic matter (Gutser et al., 2005).

Mineral fertilizer had the best effect on maize yield, in terms of both amount and stability, over the years. This was the case for application of mineral fertilizer alone as well as in combination with digestates. The crop's short-term demand for plant-available N was met through the mineral fertilizer and later—once the maize had established—N from the digestate had been mineralized and could provide the maize with a sufficient supply. In addition, the combinations provided at least a certain amount of organic matter (OM). This may be valuable as maize leaves a limited amount of crop residues and its cultivation tends to reduce soil organic matter and humus (Karpenstein-Machan, 2013; Komainda et al., 2018). Several studies have suggested that the combination of organic and mineral fertilizers can improve the regulation of N supply and enhance the effect of the two fertilizer types. As such, it is the most effective way of achieving both high yields and at the same time a build-up of soil organic matter (SOM) (Rauhe, 1987; Körschens et al., 1998; Svensson et al., 2004; Gutser et al., 2005; Möller, 2009). However, as simultaneous application can temporarily immobilize mineral N and increase the risk of N2O emissions, it is recommended that digestates and mineral fertilizer are applied with a time delay (Möller et al., 2009). We followed this recommendation in our field experiments.

In addition, the effect of combined mineral and organic fertilizers versus the application of mineral fertilization alone depends on site conditions. At the cooler "Hill" site, where the soil only warms up slowly in spring, the mineral treatment worked significantly better than the combinations. By contrast, at the "Valley" site, the mineral fertilizer and the combinations had comparable effects.

As expected, and observed at both sites, the yield effects of the different fertilizer types were less pronounced in permanent grassland than in the annual cropping system. As grassland is characterized by year-round soil cover, it can better exploit the long-term fertilizing effects of the organic treatments than the other two systems. These long-term effects result from the more continuous N release as well as better water retention and other factors improving soil fertility. However, grassland proved to be the system with the lowest total yields over 6 years. In addition, the aging effect of the plots in this system needs to be considered. For this reason, it is difficult to assess the effects of fertilizer treatment and system separately, as the system itself degrades over time (increasing gaps, reduction/loss of valuable grass species) and yields subsequently decrease. Therefore, the aging effect on yields may mask the fertilizer effects.

For permanent grasslands, there were also clear site effects. At the warmer "Valley" site, mineral fertilizer resulted in the significantly highest DMY. In contrast, at the "Hill" site, both treatments with solids led to the highest DMY during the experimental period. This was somewhat unexpected as the solids treatments had lower yields in the annual and intercropping systems at both sites. At the beginning of the experiment, we had assumed that organic fertilizers would be less effective the more marginal the site conditions are. As N mineralization and OM turnover are influenced by temperature (Davidson and Janssens, 2006), it was surprising to find the good performance of the treatments with solids at the site with lower average temperature and limited vegetation period. This result was undoubtedly a consequence of an interaction between system, treatment and site, but cannot be sufficiently explained by the data collected in this study. Repeated application of solid digestates could have increased the soil pH at this site, which was relatively low at

the beginning of the experiment (5.5). However, an intermediate soil analysis in fall 2014 showed that the pH had decreased to 5.3 on average on all grassland plots. The smallest decrease was found on plots treated with solids (5.4). Nabel et al. (2017) found that the comparative advantage of digestate fertilization over mineral NPK fertilization on biomass yield became increasingly pronounced over time and explained this through the crucial role of soil carbon content for plant growth. This obviously applies more to perennial systems where the soil is not disturbed and becomes more important the more marginal the soil is. This may serve as an explanation for the surprising performance of the solids at the "Hill" site. However, our hypothesis is that the proportion of nutrient supply provided by OM turnover increases with time and thus renders the grassland system increasingly independent of the direct nutrient effect of the fertilizers.

The intercropping system (here two crops grown in rotation) proved to be a stable and robust system that provided constantly high yields. In this system, the soil was almost always covered (except during early development stages of triticale). Unlike maize, generally all fertilizer treatments worked equally well independent of the site or crop. The yields in the intercropping system appeared to profit from the crop rotation effect, mainly from the biological fixation of atmospheric N<sup>2</sup> by the clover in the mixture (not quantified). This is intended to ensure a more constant N supply independent of fertilizer applications, for example during periods of low N availability due to insufficient amounts of mineralized N. The leguminous component of this system differentiates it from the others. Grassland also contains some clover, but in the intercropping system clover is sown afresh every other year resulting in a higher proportion of legumes in the sward and consequently a higher N fixation rate.

The clover grass and triticale both developed intensive root systems; thus the intercropping system produced a considerable amount of crop residues which additionally contributed to the build-up of SOM and the residual supply of mineralized N (Fouda et al., 2013).

In this study, we focused on the effects of the treatments on biomass yield of the cropping systems and mainly limited the explanation of different fertilizer effects to differences in the timeliness of N availability and the capacity of the various fertilizer types to contribute to SOM production. Another aspect that was considered in explaining differences in yield effects of the various fertilizer types was their interaction with the three cropping systems tested here. All cropping systems have their growth peaks at different times, which clearly affects the nutrient demand and uptake during the vegetation period (Herrmann et al., 2017).

## Implications of Different Fertilization Systems

When assessing the suitability of biogas digestates as fertilizers, other aspects in addition to the yield effect need to be considered. Clearly, a farmer who produces biogas needs to dispose of the digestates. In practice, biogas digestates are often separated and used as fertilizer on the farm. However, when other feedstock streams, such as slurry, are co-digested in the biogas plant, the nutrients in the digestates constitute an oversupply at farm level. Therefore, digestates are often transported to other farms. Alternatively, they can be further processed to bio-based mineral fertilizers (Vaneeckhaute et al., 2017). For example, nutrients can be recovered from the liquid fraction by precipitation and filtered off as a mixture of phosphate salts, including struvite (Bilbao et al., 2017; Ehmann et al., 2017). Since this process is costly, the extent to which digestates are directly applied as organic fertilizer or, especially in the case of the liquid fraction, are processed into mineral fertilizer should be carefully considered on a case-by-case basis.

Mineral fertilizer use is always accompanied by the highest costs and environmental impacts, irrespective of whether it is produced chemically (N), from mining (P) or through recovery from biogas digestates (N and P). From a farming practice point of view, mineral fertilizers have the advantage of more predictable N supply on the one hand and easier applicability on the other. The latter is particularly relevant for permanent cropping systems. One major environmental benefit of digestates is that they can help save on mineral N fertilizer, either by complete or partial substitution. In good agricultural practice, gaseous emissions during and after digestate application are kept to a minimum, which was not ensured with the liquid manure spreader used in this study. Application techniques near the soil surface including trailing hoses, trailing shoes and injection would of course reduce gaseous losses (especially in systems and at stages where incorporation is not possible) and at the same time increase the plant-usable N (Möller et al., 2008). The solid fraction should ideally be incorporated into the soil to avoid gaseous N losses (Holly et al., 2017), allow for nutrient release through decomposition and avoid a layer of organic matter remaining on the crop. The application of solids is even more laborious in systems which require multiple cuts over the vegetation period. Although our results showed that solids significantly increased grassland yields, at least at the less favorable "Hill" site, the practicability of solid application remains limited. For this reason, only the liquid fraction is recommended for grassland due to its good infiltration, and also its high N and K but low P contents which correspond well with the nutrient removal by the crops (Messner, 2014).

The application of solid digestates thus appears more appropriate in cropping systems with frequent soil cultivation and on sites where a benefit from OM can be expected. Soil tillage increases the turnover of OM from digestates and crop residues (Blair et al., 2006; Sarker et al., 2018). Although solids were actually not recommendable for maize in terms of their fertilizing effect, their regular application is considered beneficial here for OM replacement (Nkoa, 2014). A study by Nabel et al. (2017) showed that organic fertilization with digestates had a positive influence on soil properties (e.g. increased soil respiration and enhanced water-holding capacity), particularly on marginal sites. The supply of nutrients other than N, including P, K and various microelements, is a further advantage over mineral fertilizer (Risberg, 2015).

In this study, we divided the fertilizer and digestate applications into several doses. In farming practice, this effort may be lowered by reducing the number of fertilizer doses. In grassland, the majority of the N dose would be applied in late winter or early spring, followed by only one more dose later on (Möller et al., 2009). In maize for example, the solids could be applied in one dose before sowing. This may even be possible for the liquids, primarily in the combinations. Lavandier et al. (2011) fertilized silage maize with up to 170 kg N ha−<sup>1</sup> , applied in form of liquid digestate in one dose and found that this did not lead to increased Nmin values.

In this study, grassland proved to be the system with lowest yields and highest workload. Nevertheless, permanent grassland is considered the most environmentally friendly way of producing energy crops (Rippel, 2008) and provides a suitable opportunity to maintain ecologically valuable grasslands that are no longer used for fodder production. This is particularly the case when mineral fertilizer is replaced with digestates, because the grassland productivity can be maintained with lower environmental impact (Walsh et al., 2012).

### CONCLUSION

The first hypothesis underlying this study, that the influence of mineral fertilizer and separated biogas digestates on biomass yield is comparable, was confirmed. However, the recommendations that can be deduced from this vary depending on cropping system and site. All three systems tested revealed their own specific strengths and weaknesses; the same applies to the treatments. For perennial or intercropping systems, separated digestates can be fully recommended. In the intercropping of triticale and clover grass, separated digestates were able to substitute mineral fertilizer completely. Contrary to our expectations, the solids performed very well in terms of yield in interaction with grassland at the "Hill" site. However, it was seen that the use of solids in permanent grassland does not exploit their full potential. A higher benefit from solids is expected from application in annual systems where they can contribute positively to the build-up of OM. Any short-term N demand of crops is better met by a combination of digestates (liquid preferable to solid, due to high content of plantavailable ammonia-N) and mineral fertilizer. The combinations performed equally well as mineral fertilizer alone in most of the systems, sites and years and allowed mineral fertilizer input to be reduced by 66%.

The second hypothesis, that fertilization effects are stronger in annual cropping systems (with tillage) than in perennial cropping systems, could be partly confirmed. If the objective is to maximize yield performance, the preferred option is the use of mineral fertilizer alone or in combination with digestates. Since the application of solid digestates and their incorporation into soil is most difficult in perennial systems, the best balance between the goals of high biomass yield and maintenance/increase of SOM content on the one hand, and the practicability of applying solid digestates on the other, can be achieved in the intercropping system.

The third hypothesis, that fertilizing effects are influenced by site factors, particularly in the case of organic fertilizers, could also be confirmed. The effect of organic fertilizer was found to be unpredictable, especially on cooler sites. To avoid yield fluctuations and N losses on such sites, perennial systems are recommended, as they capture N released at different times in the vegetation period. For these sites, the positive effect of solid biogas digestates on soil fertility and SOM can help improve the long-term stability of biomass production.

In summary, the combined application of organic and mineral fertilizer is the best approach to implement the multiple aims in terms of high yields, low-cost farming and minimal negative environmental impacts. The good performance of the combinations, together with reduced expenses for mineral fertilizer, can help improve farmers' acceptance of organic fertilizers.

### AUTHOR CONTRIBUTIONS

AE: carried out the experiments, was involved in sample analysis and analyzed the data. UT and IL: were responsible for the original idea for the research and initial concept of the experiment. All authors contributed to the preparation of the manuscript.

### ACKNOWLEDGMENTS

AE was supported by a scholarship from the Faculty of Agricultural Sciences, University of Hohenheim. The authors are grateful to Peter Weckherlin for the excellent maintenance and management of the field experiments and to Sven Schabel and Anja Neuberdt for their assistance with fieldwork. Particular thanks go to Nicole Gaudet for proofreading the manuscript and to Jens Möhring for statistical advice. Weather data were kindly provided by the Center for Agricultural Technology Augustenberg, Karlsruhe, Germany; maize seeds by KWS Saat SE, Einbeck, Germany; and Triticale seeds 'Tarzan' by Pflanzenzucht Oberlimpurg, Schwäbisch Hall, Germany.

Selected yield data from the years 2012 and 2013 were presented at the International Conference Progress in Biogas III, 10 to 11 September 2014, Stuttgart, Germany.

### REFERENCES

Agehara, S., and Warncke, D. D. (2005). Soil moisture and temperature effects on nitrogen release from organic nitrogen sources. Soil Sci. Soc. Am. J. 69:1844. doi: 10.2136/sssaj2004.0361

Alburquerque, J. A., de La Fuente, C., and Bernal, M. P. (2012a). Chemical properties of anaerobic digestates affecting C and N dynamics in amended soils. Agric. Ecosyst. Environ. 160, 15–22. doi: 10.1016/j.agee.2011.03.007

Alburquerque, J. A., La Fuente, C., de, Campoy, M., Carrasco, L., Nájera, I., Baixauli, C., et al. (2012b). Agricultural use of digestate for horticultural crop production and improvement of soil properties. Eur. J. Agron. 43, 119–128. doi: 10.1016/j.eja.2012.06.001


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Ehmann, Thumm and Lewandowski. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Summer Fertigation of Dairy Slurry Reduces Soil Nitrate Concentrations and Subsurface Drainage Nitrate Losses Compared to Fall Injection

Joshua D. Gamble<sup>1</sup> , Gary W. Feyereisen<sup>1</sup> \*, Sharon K. Papiernik <sup>2</sup> , Chris D. Wente<sup>3</sup> and John M. Baker <sup>1</sup>

*<sup>1</sup> Soil and Water Management Research Unit, Agricultural Research Service, United States Department of Agriculture, Saint Paul, MN, United States, <sup>2</sup> North Central Agricultural Research Laboratory, Agricultural Research Service, United States Department of Agriculture, Brookings, SD, United States, <sup>3</sup> North Central Soil Conservation Research Laboratory, Agricultural Research Service, United States Department of Agriculture, Morris, MN, United States*

### Edited by:

*Claudia Wagner-Riddle, University of Guelph, Canada*

### Reviewed by:

*Jeff John Schoenau, University of Saskatchewan, Canada Aicam Laacouri, University of Minnesota, United States*

> \*Correspondence: *Gary W. Feyereisen gary.feyereisen@ars.usda.gov*

### Specialty section:

*This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems*

> Received: *31 January 2018* Accepted: *23 April 2018* Published: *15 May 2018*

### Citation:

*Gamble JD, Feyereisen GW, Papiernik SK, Wente CD and Baker JM (2018) Summer Fertigation of Dairy Slurry Reduces Soil Nitrate Concentrations and Subsurface Drainage Nitrate Losses Compared to Fall Injection. Front. Sustain. Food Syst. 2:15. doi: 10.3389/fsufs.2018.00015* Leaching of nitrate (NO3-N) from manure-applied cropping systems can represent a substantial N-loss to the environment for dairy farms, particularly in fields with artificial subsurface drainage. In this on-farm study, we used a Before/After analysis to assess the effectiveness of summer fertigation with reduced manure rates (years 2010-2015) vs. fall injection (2007-2009) of dairy slurry in terms of subsequent corn silage yield, corn N removal, soil NO3-N distribution, and NO3-N losses in subsurface tile drainage from a 65-ha field in Minnesota, USA. Yield was similar between periods (average of 18.8 Mg ha−<sup>1</sup> ), but crop %N, N removal, and manurial N-use efficiency (MNUE) were 15, 12, and 126% greater during the fertigation than injection period. Fertigation reduced spring soil NO3-N concentrations to 60-cm depth by an average of 53% relative to injection, except in the 15–30 cm increment, where no difference was found. Similarly, fall soil NO3-N concentrations from 30 to 90 cm were 48% lower, on average, under fertigation than injection. Weekly flow-weighted mean NO3-N concentration in tile drainage was lower during fertigation (47.7 mg L−<sup>1</sup> ) than injection (56.8 mg L−<sup>1</sup> ), although mean weekly drainage depth was greater during fertigation (2.3 vs. 1.1 mm). This resulted in similar weekly loads between periods (mean of 0.96 kg NO3-N ha−<sup>1</sup> ). For non-snowmelt flow, relationships between drainage and NO3-N load showed log–log slopes of near 1.0 for injection and 0.97 for fertigation, indicating dilution of concentrations with increased flow during fertigation, but not during injection. Differing intercepts indicated a treatment effect of fertigation independent of flow effects, and corresponded to loads of 5.9 kg NO3-N ha−<sup>1</sup> for injection and 4.7 kg NO3-N ha−<sup>1</sup> for fertigation, a reduction of 20% at a 10 mm weekly flow depth. The magnitude of the reduction in load increased to 22% at a 25 mm weekly flow depth. Results suggest that summer fertigation with attendant reduction in application rate is a viable method for reducing drainage NO3-N losses without impacting yield of irrigated silage corn in the U.S. Midwest.

Keywords: dairy manure, nitrate losses, soil nitrate, manure injection, fertigation, subsurface drainage, irrigation

# INTRODUCTION

The US dairy industry has changed dramatically in recent decades, with a trend toward geographic consolidation into fewer, large-scale confinement operations. From 1992 to 2012, the total number of dairy farms declined by about 60%, down to about 44,000 farms, while the number of farms with more than 1,000 cows increased by over 200%, up to over 1,800 farms (USDA-NASS, 2014). By 2012, nearly 50% of all dairy cows in the US were on farms of 1,000 cows or more, up from a mere 10% in 1992. This has resulted in a geographic consolidation of dairy manure production, and has also led to an increased reliance on corn (Zea mays L.) silage for animal feed in place of perennial forages like alfalfa (Medicago sativa L.). In Minnesota, many of the larger dairies are located in the Minnesota River basin, a highly impaired drainage that is the primary source of nitrogen, phosphorus, and sediment exported from the state via the Mississippi River (MPCA, 2014). Much of the cropland in this region is tile-drained, facilitating rapid transport of water and nutrients from the point of application to nearby waterways (David et al., 2010). These factors, coupled with the aforementioned manure consolidation and loss of perennial cover, raise questions about the water quality implications of the trend toward larger operations in the dairy industry.

Manure can be a valuable source of crop nutrients, but it can be challenging to synchronize nutrient availability with plant needs because of variability in the timing of organic N and P mineralization into plant-available inorganic forms. This is a particular challenge for corn silage production, because the N demand of corn is large and must be met within a relatively short time frame. The uncertainty regarding the timing of N availability, along with the narrow window for crop N need, leads producers to err on the high side with respect to manure application rates (Tarkalson et al., 2006). Over time, high-rate manure applications can result in elevated soil NO3- N (Muñoz et al., 2003), which in turn can lead to higher NO3-N concentrations in subsurface drainage (Randall and Mulla, 2001).

While the agglomeration of dairy animals into fewer, but larger, operations within the region may pose a significant risk to downstream water quality, it may also present unique opportunities for improvement. Economies of scale may make it practical to apply practices in large operations that would be prohibitively expensive for smaller operations. Herein we describe an on-farm experiment done in collaboration with a large dairy farm in Minnesota, U.S.A. The cooperating farmer was interested in understanding the environmental impacts of the current manure application practice, fall injection, as well as the viability and environmental impact of a unique alternative application technique—summer fertigation, or applying diluted, screened dairy slurry through an irrigation system. Previous research at the cooperating dairy has evaluated the effects of a rye cover crop on drainage NO3-N and P losses (Krueger et al., 2013), the effects of replacing open surface tile inlets with fine gravel inlets on sediment and P losses (Feyereisen et al., 2015), and changes in soil organic carbon related to agronomic management and terrain (Gamble et al., 2017).

Fall injection of manure following corn silage harvest is a common management practice for dairy producers in the region. Early fall represents the best opportunity to access the field in terms of trafficability, and it also allows producers to shift manure application away from the busy planting season. However, nutrients in fall-applied manure are especially vulnerable to leaching loss over winter and into the wet spring months, particularly in fields with artificial subsurface drainage. Krueger et al. (2013) reported annual losses of up to 89 kg NO3-N ha −1 in drainage from fall manure-injected silage corn at the same dairy, a loss of roughly 20% of the total N applied in manure. Greater synchrony between the timing of manure application and peak crop nutrient demand could improve crop nutrient utilization and reduce N losses in drainage (Magdoff et al., 1997).

In irrigated fields, such as those at our cooperating dairy, fertigation is an appealing alternative to fall injection that could better synchronize nutrient availability with plant needs. Applying manure via an irrigation system requires little machinery traffic across the field, making growing season application a realistic option. It could also allow for lower application rates over time, as a greater fraction of applied N should be utilized for plant growth, rather than lost to the environment. However, little research has been done at the field-scale to evaluate summer fertigation as a viable manure application practice for corn silage production. We assessed the effectiveness of summer fertigation with reduced manure rates vs. fall injection of dairy manure in terms of subsequent corn silage yield, soil nitrate-N dynamics, and nitrate-N losses in subsurface tile drainage effluent from a 65-ha silage-corn production field. We hypothesized that converting to summer fertigation with lower application rates from fall injection would increase crop utilization of manurial N and reduce nitrate-N losses in subsurface tile drainage without reducing silage corn yields.

### MATERIALS AND METHODS

### Site Description

The study was conducted on a 65-ha field on a privately owned dairy farm in west-central Minnesota from September 2006 through October 2015. Soils at the site were formed in calcareous loamy glacial till, characteristic of the prairie pothole soils of the Upper Midwest. Soils on higher landscape positions included a well-drained Forman clay loam and moderately welldrained Aastad clay loam. The soil on side slopes surrounding depressions was primarily a somewhat poorly drained Hamerly clay loam, while depressional areas were characterized by poorly drained Parnell silty clay loam, Parnell-Flom silty clay loam, and Nuttie-Hatley clay soils.

The study field is extensively tile drained (**Figure 1**). Mapping of the study site with light detection and ranging (LiDAR) revealed low relief with elevation varying by approximately 5 m across the 65-ha research site. From the resulting topographic map, a watershed of 26.6 ha was delineated in ArcGIS 10.1 (ESRI, Inc., Redlands, CA) for sampling of subsurface drainage. The drainage system included 18 surface inlets designed to quickly drain depressional areas during spring snow melt and after large

rain events. These were open inlets from the start of the study through the fall of 2009. In October 2009, all of the open inlets in the field were converted to gravel inlets and some additional tile drainage was added to the field. These changes resulted in a slight increase in the watershed area, from 26.6 to 26.7 ha. The drainage discharged into the subsurface drainage system of an adjacent field. Approximately 0.74 ha of the study watershed lay on the adjacent field.

Precipitation was measured at the site beginning on 4 September 2007 using a tipping bucket gauge with a manual rain gauge backup. Frozen precipitation was not measured, so October to April precipitation data were obtained from the University of Minnesota West Central Research and Outreach Center (WCROC), located 16 km away. Any other missing precipitation data throughout the study period were filled with data from the WCROC.

# Agronomic Management

The field was planted to corn (Zea mays L.) each year of the study on dates ranging from 22 April to 22 May, and was harvested for silage each late summer/early fall on dates ranging from 28 August to 29 September (**Table 1**). In the fall of 2006, 2007, and 2008, liquid dairy manure was injected by drag-line injection at 97,000–150,000 L ha−<sup>1</sup> , with dates of application ranging from 28 August to 4 November. In 2006, manure was pumped directly from the stirred lagoon, while some solids were removed before injection in 2007 and 2008. After all applications, the field was tilled to a depth of 23–30 cm with a disk ripper (Ecolo-Tiger 870, Case IH, Racine, WI, USA), except in 2008 when tillage occurred before manure application. Center pivot irrigators with drop nozzles (T3000 Trashbuster, Nelson Irrigation Corp., Walla Walla, WA, USA) were installed in October 2008, and irrigation began in May 2009 and continued each year throughout the study. No manure was applied in 2009 during the transition to fertigation. Starting in 2010, manure was fertigated via centerpivot irrigation system at 46,000–143,000 L ha−<sup>1</sup> (**Table 1**) on dates ranging from 24 June to 9 July. Two applications were made in 2010 as the producer was experimenting with application techniques. During the first application, undiluted manure was applied. During second application, manure was diluted to a 50/50 mixture with well-water, and all subsequent applications were done this way. In 2015, manure was applied to less than a third of the field due to irrigator malfunction, and 84 kg N ha−<sup>1</sup> of mineral fertilizer was applied to the remainder of the field.

# Water Sampling

Subsurface flow measurement and water sampling equipment were deployed in a manhole access to a 38-cm submain at the edge of the field. Details of the water monitoring and sampling system can be found in Krueger et al. (2013) and Feyereisen et al. (2015). Briefly, for the 2007–2009 period, stage was measured immediately upstream and downstream from a weir with bubbler level gauges (Model 4230, Teledyne Isco, Inc., Lincoln, Nebraska), and a rating curve was used to calculate flow rate. In March 2010, an area-velocity sensor (Model 2150, Teledyne Isco, Inc.) was installed. Water samples were collected on a flow-interval basis with an automated sampler (Model 3700 or 6712, Teledyne Isco, Inc., Lincoln, Nebraska). The autosampler was programmed (2007–2009) or triggered by the area-velocity sensor (2010-2015) to draw a 100-mL subsample per 300 m<sup>3</sup> of flow or 0.7 mm of depth (Harmel and King, 2005). Groups of three subsamples were combined into 300-mL glass jars containing 0.6 mL concentrated H2SO<sup>4</sup> (Clesceri et al., 1998). Samples were retrieved weekly, filtered using a 0.2µm nylon filter, and refrigerated. Determination of NO3-N was by cadmium (Cd) reduction with an Alpkem RFA300 (Alpkem Corp., Clackamas, Oregon).

# Crop Yield and N Sampling

In 2007, total wet mass of corn silage was recorded by the farmer by weighing each truckload of corn silage removed from the field, and five approximately 1.5-kg subsamples were collected for dry matter (DM) determination. In all remaining years, corn DM yield was determined by hand harvesting and weighing a 3 m length of row in at least 16 randomly-selected locations, collecting a subsample of three plants in each location.


*† Total N determined via combustion.*

*‡ An irrigator malfunction resulted in a low manure rate so an additional 84 kg N ha*−1*of mineral fertilizer was also applied.*

Subsamples were dried at 65◦C for 48 h for DM determination and ground to pass through a 1 mm sieve for N analysis by dry combustion (Bremner and Mulvaney, 1982) using a Leco TruSpec CHN Analyzer (Leco Corp., St. Joseph, Michigan). Crop N removal was calculated as the product of crop DM yield and N concentration. Manurial N-use efficiency (MNUE) was calculated as the ratio of crop N removal to manure N applied.

### Soil Sampling

Soil samples were collected after corn silage harvest and before manure application, on dates ranging from 25 August to 10 September, and again in spring on dates ranging from 14 May to 9 June. Late summer/early fall samples were collected within the watershed at 21 locations in a grid pattern with approximately 113 m between sample locations (**Figure 1**). Spring soil samples were collected from 15 locations in the watershed from a subset of the fall locations. Samples were collected to a depth of 60 cm in the spring and 90 cm in the fall using a hydraulic sampling probe with a core inner diameter of 6.5 cm, except in fall 2009 when a sampler core with an inner diameter of 3.8 cm was used. Cores were subdivided into 0–5-, 5 to 15-, 15–30-, 30–60-, and 60–90 cm layers, dried at 37◦C, and ground to <0.5 mm for chemical analysis. Soil samples were extracted with 1 M KCl and NO3- N was determined colorimetrically using flow injection analysis (Alpkem Corp., Clackamas, Oregon).

## Experimental Design and Statistical Analysis

Analysis was conducted using the Before/After design of Spooner et al. (1985), splitting the dataset into periods when manure was injected (2007-2009) and when it was applied through fertigation (2010-2015), hereafter referred to as the "injection period" and "fertigation period," respectively. For water samples during the injection period, the corresponding manure application occurred the previous fall. It is important to note that these periods also correspond with the conversion of open tile inlets to closed, gravel inlets in late fall 2009. Previous research suggests no clear difference in NO3-N loading in drainage between open tile inlets and gravel inlets (Smith and Livingston, 2013), though we acknowledge this change in the drainage system as part of any treatment effects observed in our analysis. A transition period from 1 October 2009 to 31 December 2009 during drainage renovation was excluded from analysis here. All statistical analyses described herein were conducted in program R (R Core Team, 2016).

Flow weighted mean (FWM) concentrations were calculated for each week. Loads were calculated by multiplying the sample concentration by the volume of water passing the sampling point during subsampling using the midpoint in flow as the point of divide between samples. Loads and flow depth were then summed by week for further analysis.

### Weekly Mean Flow, Nitrate Concentration, and Nitrate Load Comparisons

We used t-tests to compare average weekly flow depth, FWM NO3-N concentrations, and NO3-N loads between periods (manure injection vs. fertigation). In these comparisons, we used only non-zero weekly values, and data were square transformed (concentrations) or natural log transformed (flow and loads) to meet assumptions of Normality. There were 88 weekly observations during the injection period and 104 weekly observations during the fertigation period with recorded flow. Serial correlation was present in the weekly datasets, so errors were not independent over time. To account for this effect, we adjusted the standard error equation for the t-tests as Ramsey and Schafer (2013):

$$SE = \sqrt{\frac{1+r\_1}{1-r\_1}} \times \frac{s}{\sqrt{n}}\tag{1}$$

where s is the pooled standard deviation, n is the total number of weeks (sample size), and r<sup>1</sup> is the first order serial correlation coefficient. Under the condition where r<sup>1</sup> is zero (i.e., no serial correlation), the above equation becomes the usual standard error equation. We calculated the first order serial correlation coefficient using the acf function in the "stats" package of program R (R Core Team, 2016). Means reported in the text are back-transformed to the original scale.

### Effects of Manure Rate and Timing on Drainage Nitrate Concentrations

We hypothesized that water samples collected following high rates of manure N application would exhibit higher concentrations than those following lower application rates, regardless of manure application method. Further, we hypothesized that concentrations would remain higher over time (weeks) following high application rates. For this analysis, we created variables in the weekly nitrate concentration dataset to account for two factors related to the potential impacts of manure. The first was a categorical variable to account for the N rate of the most recent manure application for each weekly observation, where manure rates were grouped into equally sized intervals of 0–150, 150–300, and 300–450 kg N ha−<sup>1</sup> . The second was a continuous variable that specified the number of weeks since the most recent manure application.

These variables were then regressed onto the weekly FWM nitrate concentrations to determine if and how manure N rates impacted nitrate losses over time. For this analysis, we first fit an intercept-only model to evaluate potential serial correlation in model residuals. This model was then fit with different temporal correlation structures using the gls function in the R package "nlme" (Pinheiro et al., 2013), and results were compared using AIC to determine the correlation structure that best accounted for temporal patterns in the data. These results were confirmed by examining autocorrelation plots of the normalized residuals and plots of the normalized residuals vs. predicted values. Once a suitable correlation structure was selected, all possible explanatory variables were added to the model and significance was determined via t-test of each coefficient in the regression summary output at α = 0.05.

### Nitrate Transport Dynamics

To assess NO3-N transport dynamics, we evaluated the relationship between weekly nitrate load and flow depth for each period. This analysis also allows for comparison of the water quality implications between periods. In this analysis, we fitted a linear regression to the natural logarithm of weekly NO3-N load (kg ha−<sup>1</sup> ) vs. natural logarithm of weekly flow depth (mm). When the slope of this regression line equals one, the nutrient concentration remains constant with varying flow depth (Tomer et al., 2003; Ghane et al., 2016). When the slope is greater than one, high flows lead to increased concentration, and/or low flows lead to lower concentrations. Moreover, the slope is the percent increase in contaminant load induced by 1% increase in flow depth.

For this analysis, we followed the same model-fitting procedure as described above. We first fit an intercept only model, then determined the best correlation structure, then added all explanatory variables and determined the significance of each model coefficient at α = 0.05. We hypothesized that the effect of flow on NO3-N loads would vary by season, so we also fit a fixed effect for season, where "snowmelt" was 1 March to 15 April, "growing season" was 16 April to September 30, and "winter" was 1 October to the following 28 February.

### Soil and Crop Analysis

Crop DM yield and crop N were analyzed with mixed-effects ANOVA treating period (injection or fertigation) as a fixed effect and sampling site within year as a random, repeated measure. MNUE was analyzed in the same manner. Fall residual soil NO3-N, and spring soil NO3-N were also analyzed with mixedeffects ANOVA, treating period and depth as categorical fixed effects, and sampling site within year as a random, repeated measure. Mixed effects ANOVA was conducted with the lme and anova.lme functions in the R package "nlme" (Pinheiro et al., 2013).

# RESULTS AND DISCUSSION

### Weather and Drainage

Mean daily air temperature ranged from −29 to 30◦C over the study period. Mean annual temperature was 0.8◦C higher than the 30-year climatological normal of 5.8◦C in 2007, but 0.6 and 0.7◦C below normal in 2008 and 2009 (**Figure 2**, **Table 2**). Mean annual temperatures were 0.8, 0.2, 2.4, and 1.8◦C above normal in 2010, 2011, 2012, and 2015, but 1.1 and 0.8◦C below normal in 2013 and 2014. Averaged over years, mean annual temperature was 0.1◦C lower than normal during the injection period and 0.5◦C higher than normal during the fertigation period. Annual precipitation during the injection period ranged from a high of 720 mm in 2007 to a low of 562 mm in 2009, although in 2009, 194 mm of the precipitation fell during the transition period. An additional 102 mm of water was applied as irrigation during the summer of 2009. During the fertigation period, annual precipitation ranged from 493 mm in 2012 to 854 mm in 2010. Annual irrigation rates ranged from 0 mm in 2015 to 126 mm in 2010, with an annual average of 68 mm yr−<sup>1</sup> applied as irrigation, which includes water applied during fertigation. Total water applied (precipitation plus irrigation) was, on average, 79 mm yr−<sup>1</sup> higher during the fertigation period (707 mm yr−<sup>1</sup> ) than the injection period (628 mm yr−<sup>1</sup> ), primarily as a result of the additional water applied as irrigation.

Annual drainage during the three-year injection period ranged from 95 to 160 mm yr−<sup>1</sup> , with an annual average of 119 mm. During the six-year fertigation period, annual drainage ranged from 59 to 252 mm yr−<sup>1</sup> , with an annual average of 140 mm. During the transition period in fall 2009, there was 165 mm of drainage that was excluded from subsequent analysis. The average ratio of drainage to total water applied was strikingly similar between periods, averaging 0.19 ± 0.03 (mean ± SE) during the injection period and 0.19 ± 0.04 during the fertigation period. This suggests that the changes to the tile drainage system in 2009 had little impact on the fraction of drainage, and that a before/after pairing is appropriate for the two periods. However, because total water inputs were greater during fertigation, the average weekly flow depth was also greater during the fertigation period (µ = 2.3 mm) than the injection period (µ = 1.1 mm; t = 2.22, df = 172, P = 0.0348) when weeks with no flow were excluded (**Figure 3**).

### Nitrate-N Concentrations in Drainage

Weekly FWM NO3-N concentrations ranged from 5.6 to 87.3 mg L −1 , with the highest concentrations generally occurring during the growing season and lowest generally in snowmelt (**Figure 4**). Weekly FWM NO3-N concentrations exhibited substantial autocorrelation, with r<sup>1</sup> = 0.59. After accounting for this autocorrelation, we found that NO3-N concentration in tile drainage was 16% lower during fertigation (µ = 47.7 mg L−<sup>1</sup> ) than the injection period (µ = 56.8 mg L−<sup>1</sup> ; t = 2.96, df = 172, P = 0.0054, **Figure 3**). However, given the difference in flow between periods, this could be a dilution effect of greater flow

TABLE 2 | Annual precipitation, average air temperature, irrigation, tile drainage, and nitrate-N load for the study field.


*† The transition period began October 1, 2009 and ended December 31, 2009.*

during fertigation. This possibility is explored further in section Nitrate Transport Dynamics below.

A 16% reduction in NO3-N concentration is a substantial improvement, but 47.7 mg L−<sup>1</sup> is still well above the EPA drinking water standard of 10 mg L−<sup>1</sup> . Concentrations exceeding this standard also contribute to downstream acidification, eutrophication, and toxicity to aquatic organisms (Camargo et al., 2005; Camargo and Alonso, 2006). However, by the final two years of the study, average concentration was further reduced to 37 mg L−<sup>1</sup> , suggesting continued improvement as manure rates were reduced and the soil N pool was drawn down.

### Nitrate-N Loads

Total load during the three-year injection period was 166.6 kg NO3-N ha−<sup>1</sup> , with an annual average of 55.5 kg NO3-N ha−<sup>1</sup> yr−<sup>1</sup> . Total load during the six-year fertigation period was 282.3 kg N ha−<sup>1</sup> , with an annual average of 47.1 kg NO3-N ha−<sup>1</sup> yr−<sup>1</sup> . The total load during the transition period (winter 2009) was 90.5 kg NO3-N ha−<sup>1</sup> , which was excluded from further analysis because of the management transition to fertigation and construction on the drainage system. During the injection period, over half of the losses (86 kg ha−<sup>1</sup> , 52% of total) occurred during the growing season of 2008 (**Figure 5**). Total load was low in 2007, and loads were similar among seasons, as were 2009 loads prior to the transition period. During the fertigation period, very few losses were observed during snowmelt and the winter; the majority of loss (89%) occurred during the growing season. Loads were highest during 2010 and 2011 when manure rate remained high. From 2012 to 2015, loads were considerably lower, which corresponded to a reduction in manure rate.

Weekly NO3-N loads ranged from 0 to 25.4 kg NO3-N ha−<sup>1</sup> , with the highest weekly loads generally observed during the growing season and the lowest during winter. Weekly loads exhibited substantial autocorrelation, with r<sup>1</sup> = 0.44. After accounting for this autocorrelation, we found that weekly mean NO3-N load in tile drainage was similar for the fertigation (µ = 1.28 kg NO3-N ha−<sup>1</sup> ) and injection periods (µ = 0.63 kg

NO3-N ha−<sup>1</sup> ; t = 1.84, df = 180, P = 0.1222, **Figure 3**). This was caused by greater flow during fertigation, as NO3-N concentrations were lower during this period.

### Soil Nitrate Concentrations

Preliminary analysis indicated that both spring and fall soil NO3-N concentrations varied with depth. Therefore, all further analyses were conducted by depth increment. In the spring, soil NO3-N concentrations were greater during injection than fertigation for all depth increments except 15–30 cm (**Figure 6**). From 0 to 5 cm, geometric mean spring NO3-N concentration was nearly 70% lower during fertigation (µ = 13.0 ppm) than injection years (µ = 39.9 ppm; F = 10.5, P = 0.0022). Likewise from 5 to 15 cm, NO3-N concentration was 55% lower during fertigation (µ = 14.5 ppm) than injection years (µ = 32.5 ppm; F = 8.7, P = 0.0050). From 30 to 60 cm, NO3-N concentration was 35% lower during fertigation (µ = 12.1 ppm) than during injection years (µ = 18.6 ppm; F = 10.0, P = 0.0027).

Fall-residual soil NO3-N concentrations were similar during injection and fertigation periods for the 0–5-, 5–15-, and 15– 30-cm depths. From 30 to 60 cm, NO3-N concentration was nearly 40% lower during fertigation (µ = 6.3 ppm; F = 27.3, P = < 0.0001) than during injection (µ = 10.3 ppm). From 60 to 90 cm, soil NO3-N concentration was 55% lower during fertigation (µ = 6.6 ppm) than injection years (µ = 14.6 ppm; F = 49.6, P ≤ 0.0001).

These results demonstrate that summer fertigation of moderate rates of manure N can reduce the amount of NO3-N in the soil profile during two critical times of the year relative to fall injection. This appears to be a function of both manure timing and placement. Fall injection resulted in higher spring concentrations, but this N was of limited benefit to the crop, as root development is typically just underway by mid-May (Chaudhary and Prihar, 1974). Much of the N mineralized from the manure was, therefore, subject to loss during the wettest time of the year. In contrast, summer fertigation appeared to provide greater synchrony by supplying N just before peak crop demand,

around the six-leaf stage for corn (Magdoff, 1991). This was evident in fall soil NO3-N concentrations, which were reduced in deeper soil increments during fertigation relative to injection. This suggests either less mineralization at these depths during fertigation or less percolation of mineralized N to depth. Given that soil nitrate concentrations were similar in the upper soil increments between periods, we think these patterns suggest that the timing and placement of manure with fertigation resulted in less percolation of NO3-N below the crop root zone.

### Nitrate Transport Dynamics

The model that best accounted for the autocorrelation in weekly NO3-N loads included an autoregressive moving average (ARMA) correlation structure accounting for correlation up to 1 lag (week), along with a 3-lag moving average. We then fit effects of log-transformed flow, season, and period, as well as all twoand three-way interactions among these variables ("full model," hereafter). Results from the full model show that the effect of flow on NO3-N load was similar for winter and the growing season (t = 0.71, P = 0.4789), but differed during snowmelt (t = 6.67, P < 0.0001; **Table 3**). Therefore, snowmelt data are treated separately hereafter, and growing season and winter data were grouped for further analysis ("non-snowmelt" hereafter).

Analysis of log-log plots of NO3-N load vs. flow rate for the non-snowmelt data showed differences in the effect of flow between periods, as evidenced by significantly different slopes (t = 2.08, P = 0.0397) and intercepts (t = 3.64, P = 0.0004; **Figure 7**). During the injection period, the slope was β = 0.996 ± 0.017 (mean ± 95% CI), indicating that for each 1% increase in flow depth, there was a corresponding 1% increase in NO3- N load (i.e., concentration did not vary by flow rate). For the fertigation period, the slope was β<sup>1</sup> = 0.966 ± 0.023, indicating that for each 1% increase in flow depth, there was a corresponding 0.97% increase in NO3-N load. Thus, high flows resulted in slightly reduced NO3-N concentrations during the fertigation period. Additionally, the intercept was lower during the fertigation period (β<sup>0</sup> = 1.556 ± 0.102) than the injection period (β<sup>0</sup> = 1.769 ± 0.080), which indicates a treatment effect of fertigation, independent of flow effects.

For snowmelt, results showed no difference in the effect of flow on NO3-N loads between periods as evidenced by similar slopes (t = 0.16, P = 0.8746) and intercepts (t = 0.20, P = 0.8467). Across both periods, the slope was β<sup>1</sup> = 0.801 ± 0.45. Given this high variability, the slope for snowmelt was not different from 1, indicating that NO3-N load was proportional to flow depth in spring snowmelt.

The lower slope and intercept of the log NO3-N load vs. log flow rate line for the fertigation period is evidence of a water quality benefit of the fertigation treatment. The lower intercept for the fertigation period indicates lower NO3-N loads when the natural-log flow depth was 0, which corresponds to a weekly flow depth of 10 mm. The back transformed intercept values were 5.9 kg NO3-N ha−<sup>1</sup> for injection and 4.7 kg NO3-N ha−<sup>1</sup> for fertigation, a reduction of 20%. Furthermore, the slope was lower during fertigation, which indicates that the magnitude of the reduction in NO3-N load increases with increasing flow beyond 10 mm depth. However, below 10 mm flow, the magnitude of reduction is lower. For example, at a flow depth of 5 mm



\**Significant at the 0.05 probability level.*

\*\**Significant at the 0.01 probability level.*

\*\*\**Significant at the 0.001 probability level.*

the average loads were 2.96 kg NO3-N ha−<sup>1</sup> for injection and 2.43 kg NO3-N ha−<sup>1</sup> for fertigation, a reduction of only 18%. In contrast, for a weekly flow of 25 mm flow, the magnitude of reduction was 22%, with average loads of 15.0 kg NO3-N ha−<sup>1</sup> for injection and 11.7 kg NO3-N ha−<sup>1</sup> for fertigation. This represents a substantial reduction in nitrate load, as there were 22 weeks that exceeded 25 mm in flow during the fertigation period. These events accounted for 74% (620 mm) of the total flow and 65% (184 kg NO3-N ha−<sup>1</sup> ) of the total load during this period. Given that this represents a 22% reduction in load compared to injection, we would expect a load of 209 kg NO3-N ha−<sup>1</sup> for a similar number of 25 mm events, or an additional 25 kg NO3- N ha−<sup>1</sup> exported, had injection continued to be used in this field. Our analysis demonstrates that loads were lower during the

tile flow depth and NO3-N loads for non-snowmelt and snowmelt periods. When the slope is one, the nutrient concentration remains constant with varying flow depth. When the slope is greater than one, high flows lead to increased concentration, and/or low flows lead to lower concentrations. Moreover, the slope is the percent increase in contaminant load induced by 1% increase in flow depth.

fertigation period than they would have been during injection under the same meteorological and hydrological conditions.

The dilution of concentrations with increasing flow suggests that the supply of NO3-N available for leaching was reduced during fertigation relative to injection. Tomer et al. (2003) note that large flows may flush contaminants and exhibit an increased concentration, or may lead to diluted concentrations, with the outcome dependent upon location and land-use. Ghane et al. (2016) provide compelling evidence that soil N status is one such land-use characteristic that affects nitrate transport dynamics. The authors evaluated relationships between drainage and NO3- N load and reported log-log slopes of 0.99 for unfertilized and 1.07 for mineral fertilized corn fields with subsurface-drainage in central Minnesota. Our findings are similar in that a lower slope was observed when soil NO3-N concentration was lower (i.e., during fertigation). It is intuitive then, that under low soil NO3-N conditions, increases in flow would cause reduced nitrate-N concentrations in drainage as the soil N supply is leached from the profile and becomes increasingly limiting. In contrast, with higher available soil N, the supply of nitrate-N for leachate is not readily exhausted, and increases in flow can result in consistent or increasing nitrate concentrations in drainage.

Despite the water quality benefit, it is important to note that fertigation may increase the risk of N-loss via other pathways. Research has shown that surface-broadcast of liquid manure results in higher losses of N via ammonia volatilization than manure injection, primarily due to lack of soil incorporation with surface methods (Sommer and Hutchings, 2001; Duncan et al., 2017). Dilution of the slurry with an equal part wellwater results in rapid percolation into the soil, which may help mitigate ammonia losses relative to other surface application methods. Denitrification is another potential N-loss pathway worth exploring. Previous research has shown that manure injection can increase cumulative N2O emissions by 84–152% relative to manure broadcast (Duncan et al., 2017), especially for spring manure applications. However, soil saturation during fertigation may increase N2O emissions relative to other surface application methods. These potential tradeoffs among N-loss pathways associated with switching to fertigation of manure require further exploration.

### Effects of Manure Rate and Time on Drainage Nitrate Concentrations

Our hypothesis was that water samples collected following high rates of manure N application would exhibit higher concentrations than those following lower application rates. Further, we hypothesized that concentrations would remain higher over time following high application rates. Regression analysis indicated that the N rate of the most recent manure application (t = 1.49, P = 0.1383) and the weeks since manure application (t = 1.77, P = 0.0787) had no discernible effect on weekly FWM NO3-N concentrations. We can, therefore, reject our hypotheses regarding the effects of manure N rate and timesince-manure-application on weekly nitrate concentrations. Even when manure rates were relatively low, nitrate concentrations in drainage remained high. This likely occurred because of the cumulative loading effect of repeated high-rate manure applications over the first four years of the study, and resulting carry-over of soil N from year to year.

During the first four years of the study, 345-444 kg total N ha−<sup>1</sup> yr−<sup>1</sup> were added to the soil N pool in manure, of which 107–147 kg N ha−<sup>1</sup> yr−<sup>1</sup> was organic N. Slow mineralization of this pool, along with that in SOM, surely contributed to the lack of decline in nitrate concentrations in the weeks following manure application. Furthermore, carry over of organic N into the fertigation years is likely why we observed high nitrate concentrations during 2010 and 2011, despite lower manure rates. So-called "legacy N" can persist in soils for years or even decades (Van Meter et al., 2016). However, decreasing NO3- N concentrations over time suggest that soil N loading was gradually decreasing in response to reduced application rates.

### Crop Yield and N Removal

Crop yield ranged from 10.6 to 21.2 Mg ha−<sup>1</sup> yr−<sup>1</sup> (**Table 4**) and was similar between periods (F = 0.29, P = 0.593), with mean yields of 18.9 and 18.6 Mg DM ha−<sup>1</sup> during the injection and fertigation periods, respectively. In contrast, crop %N and total N removal (kg ha−<sup>1</sup> ) were greater during the fertigation than injection period. Following conversion to fertigation, silage corn N at harvest increased from 1.07 to 1.23% (F = 35.5, P < 0.0001), which corresponded to an increase in N removal from 205 to 230 kg N ha−<sup>1</sup> yr−<sup>1</sup> (F = 6.32, P = 0.0133). This resulted in greater MNUE during fertigation (µ = 1.20 kg/kg) than injection (0.53 kg/kg, F = 68.28, P < 0.0001).

Although there was no difference in yields between periods, data suggest that yields were beginning to decline slightly by 2014 and 2015, down to approximately 17 Mg ha−<sup>1</sup> . Manure rates had been dramatically reduced since 2010, perhaps too much so. Crop MNUE was greater than 1 during much of the fertigation period, suggesting that the crop was mining additional N from the soil. Over time, a slight increase in the manure rates may be required to maintain crop nutrition and yield with fertigation, though tradeoffs with water quality should be closely monitored.

### SUMMARY AND CONCLUSIONS

Switching from fall injection to summer fertigation allowed this producer to reduce manure N rates, thereby reducing tile NO3-N concentrations and loads, without impacting corn silage yield. Greater flows were observed during the fertigation period than the injection period because of increased precipitation and irrigation. This resulted in similar weekly NO3-N loads between periods, but over time resulted in decreases in annual loads, due to lower nitrate-N concentrations in tile drainage during fertigation.

Analysis of NO3-N flow dynamics suggested that the change in concentrations between periods was not solely a result of dilution, but also a treatment effect of fertigation, independent of flow effects. The treatment effect was the result of a combination of factors. Manure N rates were reduced during fertigation, which undoubtedly helped reduce losses, but the timing and placement of manure N was also important. Crop N uptake and MNUE were greater during fertigation than injection, which suggests that summer fertigation is a more efficient means of delivering N to the crop. This greater utilization, in turn, enables lower manure application rates and results in reduced N losses in drainage.

Results from the soil analyses support these findings. Spring soil NO3-N concentrations were lower at most depths during fertigation than injection. Lower spring soil concentrations



*† Average yield estimated at the field scale by producer.*

*‡Manurial N-use efficiency, calculated as the ratio of crop N removal to manure N applied (*Table 2*). Not calculated for 2015, when mineral N was applied because of irrigator malfunction.*

should result in lower growing season N losses in drainage, which we observed here. Fall residual soil NO3-N concentrations were also reduced at deeper soil increments with fertigation relative to injection. This suggests that less NO3-N percolated below the crop root zone into drainage tile under fertigation.

Manure N rate and time since manure application had no discernible effect on weekly NO3-N concentrations, most likely due to the cumulative soil N loading effect of repeated manure applications prior to the study. However, decreasing drainage NO3-N concentrations over time suggest that soil N loading was decreasing in response to reduced application rates.

Fertigation requires specialized equipment to pump manure from storage lagoons to the irrigators, as well as the ability to screen solids from the manure slurry to prevent clogging of irrigator nozzles. These and additional economic and technological factors will likely influence producer decisions regarding the suitability of summer fertigation for their operations. However, with increasing awareness of the downstream effects of nitrate pollution, many producers are seeking solutions to NO3-N loss on their farms. We have demonstrated here that summer fertigation reduced NO3-N losses from an irrigated summer-annual cropping system in the US Midwest. Despite this reduction, NO3-N concentrations in the drainage water remained high relative to environmental quality standards. More research is needed to determine if drainage NO3-N concentrations from fertigated silage corn can be further reduced to acceptable levels without impacting silage yield. Additional consideration should be given to potential tradeoffs related to crop injury, crop disease prevalence, risk of pathogen transport to waterways, and alternative N loss pathways with fertigation relative to injection of dairy slurry.

### REFERENCES


### DISCLAIMER

Mention of trade names or commercial products in this publication is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S. Department of Agriculture. USDA is an equal opportunity provider and employer.

### AUTHOR CONTRIBUTIONS

JB, SP, CW, and GF conceived the experiments. JB, SP, and GF designed the experiments. GF, JB, CW, and SP performed the experiments. JG analyzed the data and wrote the manuscript in consultation with GF, SP, JB, and CW.

### FUNDING

Funding was provided by USDA-ARS through the St. Paul, Minnesota, and Morris, Minnesota, locations.

### ACKNOWLEDGMENTS

The authors thank the operators of the cooperating dairy farm for allowing us field access for sample collection and for performing agronomic management during the study. We also thank Todd Schumacher and William Breiter, USDA-ARS, St. Paul, Minnesota; Steve Wagner, USDA-ARS, Morris, Minnesota; David Schneider, USDA-ARS Brookings, South Dakota; and Eric Krueger, Oklahoma State University.

inlets on sediment and phosphorus subsurface drainage losses. J. Environ. Qual. 44, 594–604. doi: 10.2134/jeq2014.05.0219


dairy manure. Soil Sci. Soc. Am. J. 67, 817–825. doi: 10.2136/sssaj20 03.0817


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Gamble, Feyereisen, Papiernik, Wente and Baker. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Recycling of Biogas Digestates in Crop Production—Soil and Plant Trace Metal Content and Variability

Ivan Dragicevic\*, Trine A. Sogn and Susanne Eich-Greatorex

Faculty of Environmental Sciences and Natural Resource Management, Norwegian University of Life Sciences, Akershus, Norway

Production of biogas and utilization of the resulting digestates as fertilizers has been increasing in Europe in the last few decades. Depending on the feedstock used for the anaerobic digestion process, these organic fertilizers may be a source of different pollutants, such as trace metals. When applied to soils, digestates may influence the natural metal content and enhance the release of metals to the environment since they can be rich in different trace metals and are usually rich in dissolved organic matter. This study focused on investigating metal presence, seasonal variability of their soluble forms and crop uptake in a 2-year field experiment, using two different biogas digestates as fertilizers. The use of digestates as fertilizers was compared to cattle manure, mineral fertilizer and a control without fertilizer addition, with respect to the presence and distribution of the trace metals Cd, Cu, Zn, Ni, Cr, Mn, and Mo, as well as Al in the soil and plant. The results of the study showed that both biogas digestates have caused a 10–20% increase in the total soil concentration of Ni, Cd and Cr compared to control plots without fertilizer addition. Application of biogas digestates had only a minor effect on metal uptake in plants. Overall, the selected application rate of 100 kg/ha of plant available nitrogen has had little effect on plant metal uptake and crop quality and the use of biogas digestates was comparable to the use of animal manure.

Keywords: organic waste, fertilizer, cadmium, chromium, nickel, zinc

# INTRODUCTION

Over the last few decades, the use of different organic residues as fertilizers has increased. For instance, anaerobic digestion for biogas production results in large amounts of liquid digestate, which contains high amounts of nutrients, such as nitrogen, potassium and phosphorus, and micronutrients in plant-available forms. The fertilization benefit of digestates are well documented (Möller and Müller, 2012; Nkoa, 2014) and long-term application of digestates has been seen to improve soil quality, stimulate crop yields, and even influence positively on the soil bacteria growth (Abubaker et al., 2012).

The most common substrates for biogas production are animal manures (Fantozzi and Buratti, 2009), food waste collected from the municipal households and sewage sludge (Maragkaki et al., 2018) or co-digestion of several substrates. Most of the studies have focused on the use of manure or sewage-sludge based digestates with an emphasis on their nutrient and fertilizer value (Ni et al., 2017; Yun et al., 2018). There are few studies with digestates based on source-separated food waste focused on the optimization of digestate quality as fertilizer through co-digestion

### Edited by:

Tom Misselbrook, Rothamsted Research (BBSRC), United Kingdom

### Reviewed by:

Fiona Nicholson, Agricultural Development Advisory Service, United Kingdom Marianne Thomsen, Aarhus University, Denmark Raul Zornoza, Universidad Politécnica de Cartagena, Spain

\*Correspondence:

Ivan Dragicevic ivan.dragicevic@nmbu.no

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 16 February 2018 Accepted: 11 July 2018 Published: 07 August 2018

### Citation:

Dragicevic I, Sogn TA and Eich-Greatorex S (2018) Recycling of Biogas Digestates in Crop Production—Soil and Plant Trace Metal Content and Variability. Front. Sustain. Food Syst. 2:45. doi: 10.3389/fsufs.2018.00045 with other materials, such as sewage sludge (Borowski et al., 2018). Recent findings have shown that food-waste based digestates can be used for growing vegetables and mushrooms, especially when anaerobic digestion is combined with hydroponics (Stoknes et al., 2016).

Modern crop production is focused on both good yield and grain quality, but also on preservation of the environment. There are studies showing that both animal manure (Nicholson et al., 1999; Yang et al., 2017) and sewage sludge (Ashworth and Alloway, 2004; Chen et al., 2017) contain trace metals in variable concentrations. Still, sewage sludge application to agricultural land is strictly regulated because of the trace elements build up in soil, in particular elements such as cadmium, mercury, lead, and chromium. Further, soil leaching studies for trace metals have showed high ecological risk factors with the application of organic wastes such as compost or manure (Cambier et al., 2014). Pollutants may be applied to soil through inorganic or organic fertilizers and may bioaccumulate in soil organisms and plants (Antoniadis et al., 2017a). The organic matter quality and structure in biogas digestates is dependent on the biogas production feedstock and the applied technology, which means that it can have an effect on the trace metal availability. Soil quality is an important factor for crop quality and yield, and will be affected negatively by accumulation of trace metals or other pollutants. One challenge with the field application of biogas digestates is their chemical safety (Törnwall et al., 2017). Naturally, the content of potential contaminants in substrates used for the biogas process determines the content of these contaminants in the digestates. Metal content in biogas digestates and its potential implications for the use as fertilizer has been a subject of increasing interest in recent years (Kupper et al., 2014; Parraga-Aguado et al., 2017). Environmental effects of added trace metals, such as uptake into plants and soil organisms or leaching to water recipients, depends on their mobility in the soil, which in turn is strongly affected by pH and dissolved organic carbon (DOC) content (Antoniadis and Alloway, 2002; Welikala et al., 2018). For instance, Cu, Zn, Cd, and Ni, are generally mobile in soil under common conditions for crop production in a Nordic climate, i.e., in a pH range between ∼6 and 7. On the other hand, As, Sb, and Hg may precipitate in insoluble forms under these conditions. Applying organic fertilizers represents an addition of DOC and may influence the pH in the soil and may thus increase the risk of plant uptake or leaching to water recipients.

In the field experiment described in this study, two digestates based on food waste were used, one digested alone and one co-digested with sewage sludge. The study has focused on the presence and availability/water extractability of the trace metals Cd, Cu, Zn, Ni, Cr, Mn, and Mo, as well as Al in soil after applying biogas digestates and animal manure in a field experiment conducted over two growing seasons. Aluminum was included because one of the digestates studied contains large amounts of Al from phosphate removal in sewage sludge with Al salts. The main hypothesis of this study was that the application of biogas digestates as organic fertilizers would increase the mobility and thus plant availability of trace metals in soil. Results of the study were used to calculate a contamination factor for each metal based on the concentrations from the soil top layer.

## MATERIALS AND METHODS

### Organic Fertilizers

For the field experiment, two commercially produced biogas digestates and cattle manure were used as organic fertilizers. The digestates (DIG) were produced at two municipal biogas plants in the southeastern part of Norway. The respective companies use different feedstock mixtures and different technologies for biogas production. The digestate marked as DIG1 is derived from pretreated (steam-exploded) source-separated household and industry food waste as substrate for the anaerobic digestion process. The second digestate, DIG2, was produced using a mixture of source-separated food waste and sewage sludge (ratio 50:50) as substrate.

The manure was a cattle slurry from the farm at the Norwegian University of Life Sciences in Ås (Norway). Before chemical analysis, all the samples were stored at 4◦C, in the case of NH<sup>4</sup> and NO<sup>3</sup> analysis no longer than 48 h. Selected characteristics of digestates and manure used in both growing seasons are given in **Table 1**.

# Field Experiment

The field experiment was located at Ås, southeastern Norway (59◦ 39′ 52′′N 10◦ 47′ 40′′E) on a loam soil (13% sand, 57% silt, and 30% clay, as determined by the pipette method; Elonen, 1971). The soil was an epistagnic Retisol according to the World Reference Base for Soil Resources (IUSS working group WRB, 2015). It had a pH value of approximately 6.0, a total organic matter content of 4.2% and total nitrogen content of 0.21% prior to the experiment. The field experiment was started in 2014.

The experiment was organized as a randomized complete block design with three replicates, growing barley (Hordeum vulgare L.) in 2015 and oats (Avena sativa L.) in 2016 as test crops. The field was spring-plowed and harrowed before adding the different fertilizers in April 2015 and mid-May 2016, respectively. The fertilizers were incorporated into the soil by harrowing, and fields were sown within 48 h after application of fertilizers. The size of the experimental area was 39 m × 24 m (0.094 ha), and the size of each treatment plot was 3 × 7 m (0.0021 ha).

Treatments consisted of a control without fertilizer addition (NN), mineral fertilizer (MF), animal manure (AM), and the two digestates. The mineral fertilizer used was an NPK fertilizer with an NPK ratio of 22:3:10. The applied rate per hectare was 100 kg N, 14 kg P, and 45 kg K. The amounts of digestates and manure were determined based on plant-available N NO3-N and NH4-N, **Table 1**) measured in the samples and corresponded to 100 kg N ha−<sup>1</sup> . In a recent study it was found that the content of inorganic nitrogen in digestates was a good estimation of the fertilizer value (Sogn et al., 2018). In Norway, farmers are currently obliged to register N amounts but not P addition from organic fertilizers such as manure.

**Table 2** shows total amounts and the main nutrients N, P, and K applied with organic fertilizers in kg ha−<sup>1</sup> .


TABLE 1 | Biogas digestates and animal manure characteristics for 2015 and 2016.

Total carbon (Tot. C) was measured in a dried sample. Values behind ± are standard deviation. \*Calculated based on the Tot N values from dry fertilizer samples (55◦c).

TABLE 2 | Amounts of nitrogen (N), phosphorus (P), potassium (K), and total carbon (Tot. C) added in kg per hectare for both growing seasons.


\*Plant available N.

The average yields for the 2015 seasons were 1.87 t ha−<sup>1</sup> NN (control plot), 5.31 t ha−<sup>1</sup> MF, 4.96 t ha−<sup>1</sup> AM, 6.18 t ha−<sup>1</sup> DIG1 and 5.64 t ha−<sup>1</sup> DIG2. For 2016 season, the yields were 1.2 t ha−<sup>1</sup> for NN (control plot), 4.0 t ha−<sup>1</sup> for MF, 2.2 t ha−<sup>1</sup> for AM, 3.9 t ha−<sup>1</sup> for DIG1 and 4.0 t ha−<sup>1</sup> for DIG2 plots. The data from 2016 was used for calculating total metal content in oat grains.

### Sample Preparation and Chemical Analysis

In general, the metal content in digestates depends on substrate and applied technology for biogas production. The trace metals investigated in this study are either under Norwegian or EU regulation for application of organic fertilizers, or selected based on previous research findings. Aluminum was included in the analysis because of its high content in the sewage sludge used for the production of DIG2, which in turn is due to removal of phosphorus with Al-salts before the digestion process.

The soil pH was determined by using dried soil with a soil to water ratio of 1:2.5. Total carbon content in dried soil, digestate, and manure samples was determined by a dry combustion method using a Leco Carbon Determinator EC12 (Nelson and Sommers, 1982). Water-soluble concentrations of metals were determined by extracting 5 g of soil/fresh digestates/fresh manure with 25 mL deionized water (Ashworth and Alloway, 2004). Soil water extractions are considered to be good in assessing the plant nutrient availability (van der Paauw, 1971; Sonneveld, 1990). Water-soluble concentrations of metals were analyzed within 48 h after preparing and filtering (Milipore, 0.45µm) the extracts. Water extracts of manure and digestates were also analyzed for DOC by using a Shimadzu TOC analyzer. The concentration of NO3-N and NH4-N in the organic fertilizers was determined in fresh samples by extraction with 2 M KCl and flow injection analysis using FIAstar 5000, FOSS.

For trace metal analysis, soil was sampled from the plow layer (0–20 cm) of each plot once a month from June to August. In 2015, plant material was collected at the same time, whereas in 2016, only grain samples were taken at harvest. The plant samples were dried at 55◦C for 5 days and then stored before sample preparation and analysis. Prior to total metal analysis, the soil, digestate, and manure samples were dried (24 h, 105◦C), sieved and ground. A sample of 0.2–0.3 g was digested in concentrated ultrapure nitric acid (HNO3) prior to ICP-MS analysis by stepwise (90 min) heating up to 250◦C using a Milestone Ultraclave. For the analysis of metal concentrations in soils, digestates, plants, and extracts, an Agilent ICP-MS 8800 TripleQ was used. The ICP-MS analysis has followed minor method validation (depending on the sample matrix) while certified reference materials and reference materials were used for quality control of the applied analytical methods. The filtered samples (extracts of soils and fertilizers) were prepared in 10 % ultrapure nitric acid prior to analysis.

## Statistics and Calculations

Analysis of variance (one-way and two-way ANOVA) was carried out to determine the effect of treatment on total metal concentrations in soil and grains. Two-way ANOVA was applied to the water extract concentrations after digestate application with treatment and sampling time as factors. Different means were separated by t-test. Pearson correlation was carried out to determine the pH and DOC correlation to the water-soluble metal concentration in soil. The statistical programs used were R Commander 3.2.3 and Sigma Plot 14.0. The confidence limit was 95% (p < 0.05).

Principal component analysis (PCA) is a dimension reduction technique, which may be used to demonstrate potential source and spatial distribution of metals in field experiments (Zhang et al., 2018). In the present study, the principal components (PC) were identified by the analysis of the correlation matrix for the data set regarding the water-extracted soil concentrations prepared from soils collected from the field experiment. The PCA analysis was used to identify factors underlying our set of variables (fertilizer treatments, sampling time) in order to determine relationships among them. Results were also used to investigated clustering of samples into groups based on their correlations.

A soil contamination factor (Antoniadis et al., 2017b; Shaheen et al., 2017) was calculated in order to evaluate the potential of long-term accumulation of investigated metals by applying the selected rates of fertilizers. The contamination factor, CF, was determined by the following formula (1):

$$CF = \frac{C\_i}{C\_n} \tag{1}$$

Where, c<sup>i</sup> represents the mean total concentration of the metal in soil of the organic fertilizer treatment, and c<sup>n</sup> that of the unfertilized control.

### RESULTS

# Total Metal Concentration in Digestates and Manure

The total concentrations of trace metals and Al in digestates and manure used in 2015 and 2016 are given in **Table 3**. With the exception of Cr, Ni, and Cd (DIG1 2016), trace metal concentrations were below limits set for use in agriculture (Class I). Norwegian regulation in regards to the use of organic fertilizers categorizes all organic wastes in four classes (class 0, 1, 2, and 3). For comparison we have used class 1, since only class 0 and 1 are allowing the use of organic wastes with no or minimum limitations (Authority, 2006). Nickel concentrations in both digestates were higher in 2016 than in 2015. The Cr concentrations, although high, were not increased between 2015 and 2016. Due to the technology used, Al concentration were high in DIG2 in both years. Lead (Pb) concentrations ranged from 0.5 to 6.0 mg kg−<sup>1</sup> (data not shown), which was well below regulation values (Class I fertilizers, 60 mg kg−<sup>1</sup> ).

## Total Concentration of Metals in Soil

Total metal concentrations in the MF treatment were almost identical to those in the control plots (NN), and results from both NN and MF are presented. Total soil trace metal concentrations are given in **Figure A1**.

When compared to control plots, total soil concentrations of Ni, Cr, Cd, Cu, Zn were significantly increased in the digestate treatments. For Cr, there was a significant decrease from June to August in both seasons. For Ni, the same significant decrease can be seen. For Zn, results from 2016 showed significant increase in total Zn concentration with biogas digestates used as treatments.

### Soil pH and DOC

Generally, pH was higher in NN and AM plots than in the digestate plots at all samplings, with the largest differences in the beginning of the season. The results from MF plots are not included in **Figures 1**, **2** since the values were not significantly different from the NN plots. For both growing seasons, the soil pH from control plots (NN) was in the range 5.8–6.0. Soil pH varied over the course of both growing seasons and depended on the treatment type (**Figure 1**). The pH decreased in both growing seasons from levels around 6.0 to the lowest 5.1. The lowest values for 2015 was DIG1 and for 2016 DIG2 in respect to August samples.

In 2016, a significantly lower pH was found for DIG2 compared to the other treatments at all sampling points.

The addition of the organic fertilizers significantly increased DOC concentrations in the soil for the duration of the growing season. In both growing seasons, the effect was most pronounced in the AM treatment (**Figure 2**).

## Water-Soluble Metal Concentrations in Soil

In general, water-soluble concentrations of the trace metals increased in all treatments over the course of the growing season, both in 2015 and 2016 (**Figure 3**). Concentrations were similar in both years with the exception of Cd, where concentrations were higher in all treatments and at all sampling dates in 2016 compared to 2015. As no significant differences in water-soluble concentrations were observed between treatments for Mn and Mo, the data is not included in **Figure 3**, but concentrations followed the same overall pattern with increasing values over the growing season.

There were no differences in Cd, Cu, and Ni concentrations between treatments at the first sampling date in 2015 and in the case of Cd, differences between treatments remained low throughout the season. In 2016, concentrations at the first sampling were generally higher for Cd than in 2016, and concentrations increased more with time in the organic


Limit for organic fertilizer (class 1) are given for those metals that are a part of Norwegian regulation on organic waste use in the same units as for treatments. AM, animal manure; DIG1, food waste based digestate; DIG2, sewage sludge/food waste digestate.

FIGURE 1 | Soil pH in from plots treated with different fertilizer treatments in 2015 and 2016. Letters show significant difference between each of the treatments (p < 0.05).

treatments than in the control. For Cu, higher concentrations were found in the organic and especially in the digestate treatments at the 2nd and 3rd sampling date in 2015. In 2016, however, the manure treatments had lower concentrations even than the control at the first two sampling dates and only DIG1 addition resulted in higher concentrations than the control. The pattern for Ni was similar in 2015 and 2016 with highest concentrations in both digestate treatments at the last two sampling dates, whereas the manure treatment was on a similar level to the control.

Zn, Cr, and Al already showed significant differences in concentrations at the first sampling date. In the case of Zn, the manure treatment showed higher concentrations than the other treatments both at the first sampling date in 2015 and 2016. For most dates, the concentrations in the organic treatments were significantly increased compared to the control. In July and August 2015, highest concentrations were found in the DIG1 treatment, whereas in 2016, the manure treatment remained the one with highest Zn concentrations throughout the season. Cr concentrations were highest in the two digestate treatments at all sampling dates but also generally also enhanced in the manure treatment compared to the control. Addition of the digestate with the high Al content (DIG2) resulted in higher concentrations of water-soluble Al at all sampling dates compared to the control, but the other two organic treatments showed similar levels either early in the growing season (manure) or at the later sampling dates (DIG1). In the latter case, Al concentrations were at the same low level as in the control early in June but increased strongly until August.

Concentrations of the investigated metals from June to August sampling were in general negatively correlated with pH, with the exception of Zn, and positively correlated with DOC, with R-values above 0.75.

Additional statistical consideration of the results from the section of water-soluble metal concentration was used to assess the influence of the addition of different treatments, pH and DOC. The results of two-way ANOVA analysis are given in **Table 4**.

Water-soluble concentrations of Cu, Cr, and Ni were significantly affected by the factor treatment. Between the factors treatment and sampling time, the F-value is indicating that the sampling time had a more significant role especially for Cd, Cu, and Ni.

The PCA analysis was performed on the entire data set of measured concentrations from June to August for the 2016 season. In **Table 5**, the results for component loadings and eigenvalues of each PC are presented. PC1 had an eigenvalue >2 whereas PC2 had a value below 1 (0.8). Since only PCs with eigenvalues equal to or above 1 can be taken into consideration, only PC1 can be used to explain the variability.

The data structure and component loadings of the first two components, PC1 and PC2 are shown by use of a biplot (**Figure 4**). The first two components describe approximately 97% of the variance (PC1 85%, PC2 12%).

For all the metals, PC1 values were positive. Loading values were above 0.4 for all metals except for Zn. Most of the treatments formed a cluster, except for the DIG1 in July and August (**Figure 4**). The clustering of treatments is strongest for June and July sampling while for August the clustering was more dispersed. Clustering in this type of experimental setup usually implies that used treatments have a similar effect on the metal concentration variability.

# Total Metal Concentrations in Plant Material

In 2015, plants metal content was determined at three times during the growing season at the same time as the soil samples were taken. In 2016, grain metal content was analyzed at harvest. Total concentrations of metals in plant material are given in **Figure 5**. For Cd, during the growing season there was no significant difference between the treatments, where a concentration dilution effect is seen from June to August. Grain concentrations from 2016 were below 0.1 mg kg−<sup>1</sup> for all treatments. A different uptake scenario was seen for the plant nutrients Cu and Zn, where total plant concentrations are significantly affected only by the digestates treatments (DIG1 and DIG2). Grain concentrations for Cu were around 0.2 mg kg−<sup>1</sup> while for Zn they were around 25 mg kg−<sup>1</sup> .

For treatments in season 2015, there was no significant difference in Ni concentrations during the growing season. There were no significant differences in Cr concentration for the used treatments in the sampled plant material for each of the sampling points (season 2015). The same dilution effect as seen for Cd can be also seen for Cr. In general, Cr concentration was lower than concentration of other metals detected in the plant material. The highest concentration of Cr in plant material was found in the plots treated with AM. Still, the grain concentrations in 2016 were very low (below 0.2 µg kg−1 of dry plant material).

TABLE 4 | Results of two-way ANOVA analysis with defined factor of significance for the concentration of selected metals measured in soil extracts using treatments and sampling time as factors and their interactions.


TABLE 5 | Component loadings and eigenvalues for each PC of the selected data set.


Calculated amounts of trace metals taken up per hectare combine the concentrations in grains given in **Figure 5** with the yield (section Field Experiment, year 2016) of the different treatments. Thus, it can be seen (on a per hectare basis) what proportion of the trace metals added with the fertilizers is removed by harvesting.

The highest concentration was seen for Zn, while the lowest was seen for Cd, Ni, and Cr. After Zn, Cu was taken up and transported to the grains in highest amounts.

### Contamination Factors for Soil Treated With Biogas Digestates

The potential metal contamination effect was assessed by using data from the total soil concentration from both growing seasons for the calculation of the contamination factor. Calculated contamination factors (CF) are given in **Figure 7**, using the formula from section Statistics and Calculations.

The general classification of the level of contamination available from the literature (Antoniadis et al., 2017a) usually divides CF into three classes, CF < 1, low degree, 1 ≤ CF < 3, moderate contamination and 3 ≤ CF < 6 with high degree of contamination. For 2015 within the experimental sampling the increase can be seen for Ni (CF = 1.05), Cr (CF = 1.2), and Al (CF= 1.3) for DIG2 treatment (**Figure 6**). In the calculation of CF value from 2016 there is increase for Cd (CF = 1.4) and Cr (CF = 1.2) for both DIG treatments. **Figure 6** shows that the rest of the investigated trace metals have a low or moderate level of contamination.

### DISCUSSION

The trace metals in focus in this study were both plant micronutrients (i.e., Cu, Zn, Ni, Mn, and Mo), and those that are in general not considered to be important or even harmful for plant growth (Cd, Cr, and Al). However, both groups of metals can negatively influence both plant development and soil quality (including runoff water) if present in excessive amounts. The addition of digestate has resulted in a significant increase in the total soil metal concentration, especially in the case of Ni and Zn, but also Cr and Cd. The water-extractable concentrations in the soil have significantly increased for Ni, Zn, and Cr with addition of digestate or manure in both growing seasons. A decrease in pH had an important role in the general increase of the water-soluble concentration of cationic forms of investigated metals. Still, the significant change in the pH and the addition of metals with biogas digestates did not influence metal uptake in plant material from 2015 or in the grain samples from 2016. The PCA has shown that the water-extractable concentrations of trace metals from plots treated digestate are similar to the ones measured for animal manure. In addition, clustering of specific sample points during one growing season as presented in PCA results (biplot) also contributes to the claim that digestates exhibit similar effects when compared to AM.

### Influence of Biogas Digestates Fertilization on Trace Metals as Plant Micronutrients

Application of biogas digestates as fertilizers for both growing seasons has significantly increased Zn and Ni total soil content, while Cu, Mn, and Mo were not significantly increased (**Figure A1**).

Measured levels of total soil Zn are a result of the digestates/manure Zn content, which was in a range 150–250 mg kg−<sup>1</sup> (**Table 3**). Under Mediterranean conditions, suggested adequate Zn content is in a range of 100–300 mg kg−<sup>1</sup> soil for cereal production (Brunetti et al., 2012). Still, for the north of Europe the values are usually in range of 47–61 mg/kg, which means that the measured concentrations are within the suitable levels for growing cereal crops (Kabata-Pendias, 2011). A similar effect of Zn addition was seen in a published 2 year field study with the use of organic waste based fertilizer (composted municipal solid waste) where measured values of total Zn concentration in soils were also under 300 mg kg−<sup>1</sup>

and comparable to our experimental results (Yuksel, 2015). The addition of DIG1 treatment significantly increased the total Zn content especially in growing season 2016. Still, measured values for AM treatment were comparable to the values of DIG1. The concentration of Ni in digestates was high. This resulted in significantly higher total Ni soil concentrations (**Figure A1**) when compared to the NN or MF plots. Suitable levels for total Ni concentration in soil is around 50 mg kg−<sup>1</sup> for a normal agricultural activity. The levels in our study were below a toxicity threshold (above 100 mg kg−<sup>1</sup> in soil) and well within the suitable levels (35–55 mg kg−<sup>1</sup> ) for crop production in general (Kabata-Pendias, 2011; Brunetti et al., 2012).

The observed changes for Zn and Ni water-soluble concentrations for DIG treatments were comparable to the levels measured for the AM treatment, especially in the case of Zn. Zn water-extractable concentration in the soil were not significantly higher than the ones measured for AM in season 2016 (**Figure 3**). The mentioned comparison clearly shows that the Zn added through digestates shows similar behavior as the Zn added through AM. Water-soluble Ni concentrations have significantly increased during both growing seasons in the digestate treatments. This may be due to the decrease in pH, as Ni mobility is enhanced under acidic conditions (Zhu et al., 2011). Kim et al. (2015) also confirmed that low pH can increase the release of naturally present Ni from soil.

The Zn concentration in the plant material collected during the 2015 season was significantly higher in the digestate treatments than in all other treatments (**Figure 5**). This reflects the higher availability of Zn as indicated by increased watersoluble concentrations. There were no significant differences in Zn concentration in the grain samples between the digestate and MF or AM treatments. Added digestate treatments had no significant influence on grain Ni concentration and the values were comparable to AM or MF treatments. Despite differences for Ni concentrations in soil-water extracts between treatments, no significant effect of the organic fertilizer treatments on plant Ni concentrations was found (**Figure 5**). The level of Ni concentrations in grains was comparable to that found in other studies with organic fertilizers conducted in Northern Europe (Hamner et al., 2013). Still, the long-term effect of the Zn or Ni accumulation in soil should not be disregarded. Currently, Zn is marked as one of the three metals mostly

contributing to the global environmental pollution through the use of animal manure fertilizing practices (Leclerc and Laurent, 2017). The soil contamination factors for Zn and Ni were below or at the moderate levels for the 2015 season and at moderate contamination level for the 2016 season, which implies a moderate increase in the risk of trace element accumulation. The maximum water-extractable concentration of Ni from soil was around 50 µg L−<sup>1</sup> , which is below the 70 µg L−<sup>1</sup> defined by the World Health Organization (WHO) drinkingwater regulation (World Health Organization, 2006).

Addition of digestates has not increased significantly the total Cu, Mn, and Mo concentration. The Cu, Mn, and Mo are important micronutrients needed for normal plant growth and development, and change in soil concentration may have a positive effect on plant growth. The recommended values of Cu total soil concentration for cereal growth is ranging between 20 and 40 mg kg−<sup>1</sup> , while toxic effects are dependent on the soil type and are usually seen between 60 and 125 mg kg−<sup>1</sup> (Kabata-Pendias, 2011). Values measured in both growing seasons are approximately 10 times lower than the upper toxic limit for Cu in soil for all used treatments, which implies a lower concentration of Cu in soil then recommended. The reference values of Mn for soils used I agriculture in Norway is ∼6 mg kg−<sup>1</sup> , where values measured in our studies was around 2 mg kg−<sup>1</sup> , showing that the addition of biogas digestates has no significant effect on the total Mn concentration in soil. For total Mo concentration in soil the reference value in soils is ∼2 mg kg−<sup>1</sup> , which is not different than the values measured in our 2-year experiment (1.2– 1.4 mg kg−<sup>1</sup> ). Based on the total concentrations it is clear that the addition of both biogas digestates had no significant influence on the reference values of Mo in soil used for plant production (Kabata-Pendias, 2011; Brunetti et al., 2012).

Copper mobility, measured as water-extractable Cu, generally increased from June to August for all organic fertilizers. Copper mobility is mostly influenced by the changes in soil pH and DOC concentration due to the different treatments. Dissolved organic carbon, i.e., the soluble phase of organically bound carbon is known to affect metal mobility both in soil solution and thus in the soil (Pérez-Esteban et al., 2014). Addition of biogas digestate

has increased the amounts of DOC in soil water extracts. In 2016, a significant increase in DOC was seen in the AM treatment in June. This follows the total amounts added with manure that year. In 2016, a higher volume of manure was added (233 L per plot) than in 2015 because of a lower N concentration. An increase in DOC concentration over the growing season with the use of organic fertilizers (composts) has also been confirmed in a recent study (Manninen et al., 2018). The rather low watersoluble concentration of Mo in our study may be explained by the fact that the strongest adsorption of Mo to the soil occurs at around pH 5 (Smith et al., 1997), which is not different from the pH values measured in the samples from our field experiment in August. Despite a pH change from 6.2 to 5.1 from June to August, no significant changes in water-extractable Mo concentrations were observed.

The total Cu concentrations in plants were significantly increased in the digestates treatments especially at the sampling in August 2015 (**Figure 5**) when compared to control plots, but grain Cu concentrations at harvest (year 2016) were not significantly increased. The contamination factor was below 1 for Cu, Mn, and Mo for all treatments (**Figure 7**), which implies a low risk of soil contamination.

### Influence of Biogas Digestates Fertilization on Trace Metals Not Essential to Plants

For this group of tracemetals (Cr, Cd, and Al), the total soil concentration was significantly higher in the plots treated with biogas digestates (**Figure A1**) when compared to NN and MF plots. Use of both digestates significantly increased total Cr and Cd concentrations in soil (season 2016), while only DIG2 increased the total Al content of the soil. This is due to the high concentration of Cr and Al, but also Cd (2016) in the digestates. Cr concentration in the digestates was above 150 mg Cr/kg DM (**Table 3**), which was higher than the limits set by the Norwegian regulation on organic fertilizers used in agriculture (60 mg/kg dry fertilizer, class I). Slightly higher values were also seen for Cd (**Table 3**), while Al content is not regulated for organic fertilizers in Norway. The Cd content in organic wastes vs. that in mineral fertilizers is an important issue of discussion The maximum allowed total concentration of Cd in soils treated with organic waste such as sludge was 3 mg/kg (Brunetti et al., 2012), and this limit was not been exceeded in the two growing seasons. Toxic values for total Cr concentration in agricultural soils may vary between 150 and 400 mg/kg depending on the soil type (Kabata-Pendias and Pendias, 2001; Kabata-Pendias, 2011). In both seasons, total Cr concentration did not exceed 100 mg kg−<sup>1</sup> .

The water-soluble Cd concentrations were low in all treatments in the first season, as Cd additions with the organic fertilizers were low and little Cd was present in the soil initially. It is important to mention that the field experiment was started with the same treatments already in 2014, while the samples were taken and analyzed in the next 2 years (2015 and 2016). This may be important in evaluating potential Cd accumulation in the soils treated with digestates. Research studies with similar levels of Cd in organic fertilizers have also shown that the addition of organic fertilizers (manure) does not cause a significant change in the soil Cd content (Xu et al., 2015). In both growing seasons, Cd was strongly correlated to the pH then to DOC, which is expected and also confirmed in a recent study (van der Sloot et al., 2017), where lower pH was seen to increase the Cd soil waterextracted concentrations. (Van Zwieten et al., 2013) reported that application of organic fertilizers (poultry litter) with pH above 7 has resulted in soil pH decrease during the growing season (Van Zwieten et al., 2013), which has also seen for our experimental set up. There is a significant increase of Cr watersoluble concentrations during both growing seasons (**Figure 3**) for digestate treatments when compared to NN and MF. Still, measured concentrations were below the limit of 50 µg L−<sup>1</sup> of total Cr, which is a recommended value for fresh waters (World Health Organization, 2006). The values measured in both seasons were comparable to those found in a similar field experiment reported by Wierzbowska et al. where values were also below 50 µg L−<sup>1</sup> (Wierzbowska et al., 2016). In the pH range measured for both growing seasons, Cr(III) is expected to be the predominant form of Cr present in the soil solution (Bradl, 2004; Choppala et al., 2018). The same authors are also suggest that lower pH values (below 6.5) considered as a major factor that is increasing the level of Cr(III) sorption to soil particles. This can be used to explain the low water-extractable Cr concentrations found in our study (**Figure 3**).

Aluminum availability in soil and soil pore water is generally influenced by soil pH and DOC. In our study, both pH and DOC showed good correlations to water-extractable Al concentrations. The high total Al concentration in DIG2 was due to the sewage treatment process where Al salts were used for precipitation of P from sewage water prior to anaerobic digestion. Aluminum phosphates are poorly soluble, and this is clearly reflected in similar levels of water-extractable Al in DIG2 and even though no Al salts were added in the process for DIG1. In general, Al does not occur in toxic forms in the soil at pH values suitable for cereal production (Wang et al., 2006).

Low concentrations in both total and water-soluble forms have resulted in low Cd levels in the plant material (**Figure 5**), with a significant increase in Cd grain concentration only in the AM and DIG2 treatments. The amounts of Cd taken up by grains calculated per hectare were very low for all treatments (**Figure 6**). Cd originating from sewage sludge and similar sources is controlled more strictly than that from mineral fertilizers, although the latter can have a high Cd content (Pizzol et al., 2014). Digestate treatments had no significant effect on plant Cd or Cr concentrations and measured concentrations were comparable with AM or MF. The contamination factor for Cd, Cr, and Al though moderate, were highest from all the CFs measured in this study. The contamination factor for Al is of little relevance unless pH levels decrease below the adequate range (below pH 4) for cereal production, which could firstly affect the barely production.

In our field experiment with cereals, biogas digestates were applied at the same fertilization rates (100 kg /ha available N). At these rates, there was little difference in the effects of the applied fertilizers on soil or crop metal concentrations. Concentrations measured in the cereal grains were well below maximum allowed values found in human exposure studies with cereal crops (Huang et al., 2008). Still, research on accumulation rates over a longer period of application may give valuable insight into soil processes that the use of organic fertilizers can induce in soil. Our field experiment was only conducted for 3 years, and

### REFERENCES


measurements only exist for the last 2 years. Longer time series will be necessary for assessing the plant uptake of trace metals after repeated additions of digestates.

# CONCLUSIONS

In our study, we have reported for the first time results from field experiment conducted through two growing seasons to investigate the application of two different commercial biogas digestates based on food waste and food waste/sewage sludge mixture. The main hypothesis that the application of biogas digestates as organic fertilizers would increase the mobility and thus plant availability of trace metals in soil was not entirely confirmed. While total concentrations of Ni, Cd, Zn, and Cr were increased in soil upon application of digestates, their water-extracted concentrations were below the defined limits established by WHO. In addition, total Cu, Mn, and Mo concentrations were not significantly increased in the soil. Both pH and DOC were important factors in determining the mobility of the trace elements, in most cases increasing the watersoluble concentrations of trace elements from June to August. Plant concentrations did not indicate an increased uptake of trace metals into the cereal crops due to digestate application. Based on these results, the use of digestates can be compared to the use of animal manure or mineral fertilizers with respect to trace metal accumulation in soil and grain uptake. Still, a period of 2 years is too short to conclude on potential trace metal accumulation in soils due to the use of digestates as fertilizers.

# AUTHOR CONTRIBUTIONS

In regards to the work done regarding this manuscript, ID main tasks were sampling during the field experiment, sample analytical lab work, statistical considerations, manuscript draft, and final version preparation. SE-G main tasks were planning and organizing the field experiment, experiment yield measurements, help with the sampling and work on the draft and manuscript final version, while TS main tasks were planning the field experiment and work on the draft and final version of manuscript.

# FUNDING

This research was partly funded by the Norwegian Research Council (Project no. 228747/E20, BiogasFuel).


Elonen, P. (1971). Particle-size analysis of soil. Acta Agralia Fennica 122, 1–122.


World Health Organization (2006). Guidelines for Drinking-Water Quality.


model, principle components analysis and geo statistical analysis. Ecotoxicol. Environ. Saf. 159, 354–362. doi: 10.1016/j.ecoenv.2v.2018.04.072

Zhu, R., Wu, M., and Yang, J. (2011). Mobilities and leachabilities of heavy metals in sludge with humus soil. J. Environ. Sci. 23, 247–254. doi: 10.1016/S1001-0742(10)60399-3

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Dragicevic, Sogn and Eich-Greatorex. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# APPENDIX

FIGURE A1 | Variability of total metal concentrations in soil samples taken during the growing seasons 2015 and 2016. The values are presented as means ± SD. Letters are showing significance of differences between the treatments (p < 0.05)

# Optimal Placement of Meat Bone Meal Pellets to Spring Oats

### Sofia Delin<sup>1</sup> \*, Lena Engström<sup>1</sup> and Anneli Lundkvist <sup>2</sup>

<sup>1</sup> Department of Soil and Environment, Swedish University of Agricultural Sciences, Skara, Sweden, <sup>2</sup> Department of Crop Production Ecology, Swedish University of Agricultural Sciences, Uppsala, Sweden

New technology makes it possible to apply organic fertilizers with higher precision, and organic producers want to know how to exploit these new possibilities to make their production more efficient. This study investigated the effect of band application (in different positions) of pelleted organic fertilizer, compared with broadcasting, on grain yield and weed density in spring oats (Avena sativa L.). Six microplot field experiments were carried out on silty clay and sandy loam in Sweden during the growing season of 2014–2016. In oats seeded at 25 cm row spacing, pelleted meat bone meal was band-applied at one of three distances from the crop row (0, 4, and 12.5 cm) and at two or three incorporation depths (1 and 4 cm on silty clay and 1, 4, and 6 cm on loamy sand). These treatments were compared with broadcast spreading, mineral nitrogen fertilizer, and an unfertilized control. On both soil types, fertilizer placement 4 cm from the crop and 4–6 cm incorporation depth gave the highest yield and crop nitrogen uptake. Yield in this treatment was 800 kg ha−<sup>1</sup> higher on clay soil and 1,100 kg ha−<sup>1</sup> higher on sandy loam compared with the same organic fertilizer applied by broadcasting, an 80–150% yield increase. On the sandy loam, distance from the crop row had a more significant effect on grain yield (p < 0.001) than soil incorporation depth (p = 0.07). On the silty clay, crop yield was significantly influenced by incorporation depth (p = 0.003) and distance from the crop row (p = 0.04). In five experiments, mineral N fertilizer equivalent (MFE) increased from on average 63% with broadcasting to 85% with placement 4 cm from the crop row and 4 cm incorporation depth. Weed biomass was significantly affected by fertilizer placement on the clay soil, with higher weed biomass with deeper incorporation (p = 0.045) and greater distance from the crop row (p = 0.049). On the sandy loam, there was a tendency for larger weed plants at greater distance from the crop row (p = 0.13) except when seeds and pellets were placed together, which gave the highest weed weight, probably due to lower competition from the crop in this treatment.

Keywords: meat bone meal, organic fertilizer, fertilizer banding, soil incorporation, organic grain production

### INTRODUCTION

Due to their physical properties, organic fertilizers are usually difficult to apply with good precision. Use of pelleted organic fertilizers is therefore an attractive alternative for organic farmers. Pellets can be applied with machines that provide a uniform distribution in the field and are not as limited in time to perfect soil conditions as fertilizer products that require heavy machinery. If the pellets are sufficiently robust, they can be applied with a seeder and incorporated in rows with

### Edited by:

Claudia Wagner-Riddle, University of Guelph, Canada

### Reviewed by:

Eliana Lanfranca Tassi, Istituto per lo studio degli ecosistemi (ISE), Italy Peter Sørensen, Aarhus University, Denmark

> \*Correspondence: Sofia Delin sofia.delin@slu.se

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 01 February 2018 Accepted: 31 May 2018 Published: 19 June 2018

### Citation:

Delin S, Engström L and Lundkvist A (2018) Optimal Placement of Meat Bone Meal Pellets to Spring Oats. Front. Sustain. Food Syst. 2:27. doi: 10.3389/fsufs.2018.00027 the same precision as seeds or granulated mineral fertilizers. The dose, time of application, and placement can thereby be much better optimized than for other organic fertilizers, which are often heavy, sticky, and difficult to distribute. Modern machinery equipped with RTK GPS and designed for seeding and weed hoeing with high precision is becoming more frequent in Swedish organic production. This gives farmers the possibility to place the pellets with high precision and many farmers want to know more about optimal placement. Meat and bone meal (MBM) is currently a common pelleted fertilizer in Sweden, although products with chicken manure were more common in the past. In the future, and in other countries, other organic materials could be used.

When pellets are broadcast, soil incorporation is often performed with a harrow to mix the pellets with a large volume of soil, but this typically leaves a large fraction of the pellets close to the soil surface. Application in bands with a seeder can achieve deep incorporation of all pellets, and also keeps the pellets more concentrated and less mixed with the soil. The way in which organic fertilizers are placed in soil during application can affect the biological turnover of nitrogen (Sørensen and Jensen, 1995; Sørensen and Amato, 2002). For example, if the fertilizer is mixed with a larger volume of soil, immobilization of nitrogen will be faster (Sørensen and Jensen, 1995), since both the energy source (carbon) and the nitrogen will be available to a larger number of microorganisms. If the fertilizer is instead concentrated to narrow bands in the soil, microbial access to carbon and nitrogen, and thus nitrogen immobilization, will be more limited. Baitilwake et al. (2012) found greater immobilization and nitrification after soil incorporation than surface application of chicken and cattle manure. However, Delin and Strömberg (2011) did not find any differences in net mineralization of nitrogen between surfaceapplied and incorporated chicken manure and cattle slurry, as long as the manure was kept in concentrated lumps, although when the manure was mixed with a larger soil volume both net immobilization and nitrification increased.

Apart from the potential advantages listed above, band application also means that the fertilizer can be applied at an optimal distance from the crop row. In organic farming, wide row spacing is commonly used to facilitate mechanical weed control (Hiltbrunner et al., 2005). This justifies pellet placement close to the row, instead of broadcasting or placement mid-way between rows, as this can increase nutrient availability for the crop and reduce it for weeds (Rasmussen et al., 1996). This has previously been observed with mineral fertilizers (Rasmussen et al., 1996; Blackshaw, 2005) and animal slurry on corn (Schröder et al., 1997; Rasmussen, 2002; Petersen, 2005; Bittman et al., 2012). In the case of animal slurry, placement close to the seed row has been shown to increase yield in spring cereals (Petersen, 2005) and phosphorus (P) uptake in corn (Schröder et al., 1997; Bittman et al., 2012). Rasmussen et al. (1996) observed 55% less weed biomass and 28% higher yield of spring barley after application of fertilizer close to the crop row compared with broadcasting. In a 4-year experiment, Blackshaw (2005) found that subsurface-banded mineral nitrogen fertilizer applied to spring wheat often led to lower weed nitrogen uptake and biomass and higher grain yield than when nitrogen was surfaceapplied. However, no study performed to date on placement of pelleted organic fertilizers to spring cereals has sought to identify the optimal placement of pellets in relation to seed and quantify the production benefits.

The objective of the present study was to investigate how crop nitrogen (N) uptake, grain yield, and weed density in spring cereals on two different soils are affected by band application of pelleted organic fertilizer at different incorporation depths and distances from the crop row, in comparison with broadcast pellets with very shallow incorporation. The hypotheses are that (i) the nitrogen use efficiency of pelleted fertilizers increases with incorporation depth, but the effect decreases with precipitation, (ii) with row spacing of 25 cm or more, the effect on yield of nitrogen in pelleted fertilizers is higher and the weed pressure lower if the fertilizer is placed less than 6 cm from the seed row than if it is broadcast or placed at greater distance from the row, and (iii) any yield loss caused by 25 cm row spacing instead of the conventional 12.5 cm will be smaller than the yield increase caused by band application with optimized placement. In the experiments, a spring oats crop seeded at 25 cm row distance was fertilized with pelleted MBM, as an example relevant for Swedish organic production.

# MATERIALS AND METHODS

### Experimental Sites

Six microplot field experiments were carried out in spring oats (Avena sativa L.) on silty clay and sandy loam soils (**Table 1**) in Sweden (58◦N, 13◦E) during the growing season of 2014- 2016. The sandy loam was on a farm with regular addition of farmyard manure, whereas on the silty clay soil manure has not been applied for a long time. Mean monthly precipitation and temperature in the period May-August 2014–2016 were recorded at a location <20 km from all study sites (**Table 2**). In 2014, the weather was warm and dry in July, followed by heavy precipitation in August, while 2015 had a cool growing season and 2016 had rather dry weather in May (**Table 2**).

TABLE 1 | Soil characteristics in the topsoil (0–30 cm) at the different experimental sites.


\*Soil organic matter.

\*\*Ammonium lactate-extractable phosphorus (P), where values <4 are considered moderately P deficient and <2 very P deficient.

\*\*\*Ammonium lactate-extractable potassium (K), where values <8 are considered moderately K deficient and <4 very K deficient.

TABLE 2 | Mean monthly air temperature and precipitation at the experimental sites in April–August, 2014–2016, compared with \*mean climate for the period 1960–1990.


### Experimental Design

In randomized block experiments, four blocks and 12 treatments (treatments 1–2, 6–15) were established on silty clay and an additional three treatments (treatments 3–5) on sandy loam (**Table 3**). Apart from two control treatments, one of which received no fertilizer (treatment 1) and one 60 kg N ha−<sup>1</sup> as mineral fertilizer (Axan; 27% N and 4% sulfur (S); Yara Sweden) (treatment 2), all treatments received 60 kg N ha−<sup>1</sup> as pelleted meat bone meal (MBM) (Ekoväx 8-3-5-3; Ekoväx Sweden). The nitrogen in the pelleted MBM was 95% organic, of which 60% was expected to be plant-available within 1– 2 months after application (Delin and Engström, 2010; Delin et al., 2012). The pelleted MBM also contained 23 kg P ha−<sup>1</sup> , 38 kg potassium (K) ha−<sup>1</sup> , and 23 kg S ha−<sup>1</sup> . Placement at different depths and distances from the crop row was compared with shallow incorporation of broadcast pellets. A 25 cm row spacing was used, since it is common practice in organic farming to enable mechanical weeding between crop rows. For comparison, one additional treatment with broadcasting involved the conventional row spacing of 12.5 cm (treatment 13). To test the effects of placement and incorporation under moist conditions, two extra treatments with irrigation, one in plots with broadcasting (treatment 14) and one in plots with 4 cm incorporation and placement in rows (treatment 15) were included in the experiments.

### Seeding and Fertilization

Each microplot was 0.7 m<sup>2</sup> and was seeded and fertilized by hand, with four (or eight in treatment 13) 70 cm long crop rows per plot. During seeding, two 70 cm long iron plates were knocked into the ground to form a trench, into which seeds and/or pellets were poured. The plates were then removed and soil was drawn from the sides to close the trenches. Seeds were placed at 4 cm depth (and thereby together with pellets in treatment 6). The amount of seeds planted in each plot was 550 seeds m−<sup>2</sup> , according to general recommendations. In the broadcasting treatments (12–14), pellets were only gently incorporated by mixing by hand into the upper 1-cm layer of soil. The irrigated treatments (14–15) received around 5 L water per plot, corresponding to 7 mm, directly after sowing and pellet application. To ensure weed pressure, white mustard (Sinapis alba L.) was sown in a diagonal across each microplot in 2014 (14 seeds per plot). This was repeated only on the silty clay soil TABLE 3 | Incorporation depth and distance from the crop row of Axan (NH4NO3) and pelleted meat bone meal (MBM) placement, crop row spacing, and irrigation in different treatments (T) in six experiments on silty clay (C) and sandy loam (S) during 2014–2016.


in 2015 and 2016, since the weed pressure from the naturally occurring weed flora was considered to be high enough on the sandy loam.

### Weed and Crop Sampling

Weed density and biomass and crop yield were measured within a net area of 50 × 50 cm, i.e., 50 cm of the two (or four in treatment 13) middle crop rows. The number of weed plants was counted on two occasions, at stem elongation of the spring oat crop (growth stage (GS) 30) and at panicle emergence (GS 55) (Zadoks et al., 1974). At panicle emergence, the weeds were harvested as close to ground level as possible, dried, and weighed plot-wise. The nitrogen content was analyzed treatment-wise. The oat crop was harvested at ripening (GS 92) by cutting 1-2 cm above the ground, and dried at 60◦C for 24 h. The crop samples were then threshed and grains and straw were weighed separately. The plant material was milled and subsamples were analyzed for water and nitrogen content. Nitrogen analyses were performed with a Leco TruMac CN (LECO Corporation, St. Joseph, MI, USA).

### Data Analysis

Grain yield (15% water content, kg ha−<sup>1</sup> ), nitrogen offtake with grain yield (kg N ha−<sup>1</sup> ), and total aboveground nitrogen (kg N ha−<sup>1</sup> ) in treatments 3-11 had the corresponding values from treatments 1 and 12 deducted, to give the net increase for different placements compared with no fertilization and broadcasting, respectively. These effects, together with weed numbers (no. m−<sup>2</sup> ) and weed biomass (dry matter, g m−<sup>2</sup> ), were statistically analyzed with a two-way ANOVA, including the factors incorporation depth, distance from the crop row and their interaction. Year was treated as a random factor. Experiments on clay soil and sandy soil were analyzed separately. Effects of irrigation were analyzed by one-way ANOVA including treatments 7, 12, 14, and 15, in order to determine whether irrigation increased yield more for broadcasting (treatment 14 compared with 12) than for row incorporation (treatment 15 compared with 7). Effect of crop row spacing was analyzed by comparing treatments 12 and 13. All models were fitted using the general linear model in Minitab 16 Statistical Software (Minitab Inc. 2010).

Nitrogen offtake with grain in treatments 3–15 was compared against the mineral N fertilizer response, in order to calculate the mineral fertilizer equivalent (MFE), i.e., the fraction of total N as



\*Treatment number [Incorporation depth (cm), Distance from crop row (cm), Row spacing (cm)]. BC, broadcast; I, irrigation. Relevant statistics is shown in Figure 1.

available to plants as N applied as ammonium nitrate (Delin et al., 2012; Jensen, 2013).

$$\text{MFE (Tx)} = \frac{\text{N offset (Tx) } - \text{ N offset (T1)}}{\text{N offset (T2) } - \text{ N offset (T1)}} \tag{1}$$

### RESULTS

### Yield Levels and Nitrogen Response

The spring oat crop produced on average comparatively high grain yield in treatment 1 without fertilization (around 4,400 kg ha−<sup>1</sup> ) on the sandy loam, whereas on the silty clay the average grain yield in the unfertilized plots was very low (1,300–2,600 kg ha−<sup>1</sup> ) (**Table 4**). Fertilization with mineral nitrogen fertilizer (treatment 2) resulted in around a 30-40 kg ha−<sup>1</sup> yield increase per additional kg N in most experiments, but only 15 kg ha−<sup>1</sup> per additional kg N on the silty clay in 2014 (calculated from **Table 4**). Yield was very low in all treatments in this experiment (around 2,000 kg ha−<sup>1</sup> ), whereas yield was normal (moderate) in the other two experiments on silty clay (4,500–5,000 kg ha−<sup>1</sup> ) and rather high on the sandy loam (6,000–7,000 kg ha−<sup>1</sup> ). Straw yield also differed between soils, with on average 5,200 kg dry matter (DM) ha−<sup>1</sup> on the sandy loam and 3,000 kg DM ha−<sup>1</sup> on the silty clay (data not shown). Straw yield was linearly correlated with grain yield and of similar magnitude, with a straw/grain ratio of 0.9 on the sandy loam and 1.2 on the silty clay.

### Nitrogen Offtake

Nitrogen offtake with grain yield on the silty clay ranged from on average 26 kg N ha−<sup>1</sup> in the unfertilized treatment to 48 kg N ha−<sup>1</sup> in the treatment with mineral fertilizer. It was 42 kg N ha−<sup>1</sup> in the highest yielding treatment with pelleted MBM (treatment 7). On the sandy loam, nitrogen offtake was on average 62 kg N ha−<sup>1</sup> in the unfertilized treatment, 85 kg N ha−<sup>1</sup> in the treatment with mineral fertilizer, and 89 kg N ha−<sup>1</sup> in the highest yielding treatment with pelleted MBM (treatment 4) (data not shown).

FIGURE 1 | Grain yield increase in treatments with placement of pelleted meat bone meal (MBM) in rows compared with broadcasting, as a function of distance from crop row and incorporation depth, on (left) silty clay and (right) sandy loam (3-year average, error bars indicate standard error).

### Effects of Different Placements on Yield

The effects on yield of band placement of MBM pellets (treatments 3–11) compared with broadcasting (treatment 12) varied depending on placement distance from the crop row and incorporation depth in the soil (**Figure 1**). On both soil types, fertilizer placement 4 cm from the crop row with 4–6 cm incorporation depth gave the highest yield, with on average 800 kg ha−<sup>1</sup> higher yield on clay soil and 1,100 kg higher yield on sandy loam compared with broadcasting (**Figure 1**). This represents an 80–150% yield increase. Some other placement options reduced yield compared with broadcasting, for instance placement together with the seed on loamy sand (**Figure 2**) or shallow incorporation and placement far from the crop row on silty clay (**Figure 1**).

Analysis of the differences in yield increase with the different placement of band-applied pellets compared with broadcast revealed that the interaction between incorporation depth and distance from the crop row was not negligible (p = 0.02 on sandy loam, p = 0.09 on silty clay) and was thus included in the model. On the sandy loam, there were statistically significant differences between different placement distances from the crop row (p < 0.001), with on average 600 kg ha−<sup>1</sup> higher yield and 7 kg N ha−<sup>1</sup> higher nitrogen offtake for 4 cm compared with 12.5 cm distance from the crop row, but not between different incorporation depths (p = 0.07). Treatment 6, where pellets and seeds were placed at the same position in soil, gave the lowest yield effect on the sandy loam soil.

On the silty clay, there were statistically significant differences in crop yield between incorporation depths (p = 0.003) and distances from the crop row (p = 0.04). On this soil, crop yield was on average 450 kg ha−<sup>1</sup> higher and nitrogen offtake on average 4 kg N ha−<sup>1</sup> higher when pellets were incorporated to 4 cm compared with 1 cm depth, and on average 460 kg ha−<sup>1</sup> and 4 kg N ha−<sup>1</sup> higher, respectively, when placed at 4 cm compared with 12.5 cm from the crop row. Combining 4 cm incorporation with placement 4 cm from the crop row gave 1,200 kg ha−<sup>1</sup> higher yield (**Figure 1**) and 11 kg N ha−<sup>1</sup> higher nitrogen offtake than the lowest yielding option (1 cm incorporation, 12 cm from crop row).

### Above-Ground Crop Nitrogen

The differences in above-ground crop nitrogen (in both straw and grain; **Figure 2**) showed a similar pattern to the differences in grain yield (**Figure 1**). On the silty clay, incorporation depth had a significant impact on crop nitrogen (p = 0.01), with on average 5 kg ha−<sup>1</sup> more nitrogen in the crop when pellets were incorporated to 4 cm depth compared with 1 cm. The differences in crop nitrogen depending on pellet distance from the crop row were not statistically significant (p = 0.095), but the trend was similar to that observed for yield, with decreasing nitrogen uptake with increasing distance from the crop row (**Figure 2**). On the sandy loam, the differences were larger (**Figure 2**) and statistically significant for distance from crop row (p = 0.018), but not for incorporation depth (p = 0.062) or interaction (p = 0.12).

### Effects of Irrigation on Fertilizer Placement

Irrigation had no statistically significant effects on yield differences between treatments with placement (treatments 7, 15) compared with broadcasting (12, 14) on either soil type in any year. In the irrigated treatments, placement increased yield by on average 1,300 kg ha−<sup>1</sup> on the sandy loam and 600 kg ha−<sup>1</sup> on the silty clay, which is similar to the yield increase in their unirrigated counterparts (1100 kg ha−<sup>1</sup> and 800 kg ha−<sup>1</sup> , respectively).

### Effects of Row Distance

Yield was on average 300 kg ha−<sup>1</sup> (p = 0.034) higher in the treatment with conventional row spacing (12.5 cm) and broadcast MBM pellets than in the corresponding treatment with double row spacing. The difference varied between years and sites, and was on average larger on the silty clay (440 kg ha−<sup>1</sup> ) than on the sandy loam (120 kg ha−<sup>1</sup> ).

TABLE 5 | Mineral nitrogen fertilizer equivalents (MFE) of meat bone meal (MBM) pellets, calculated from nitrogen offtake (Equation 1) in treatments 3–15 (see Table 3) on sandy loam and silty clay in 2014–2016.


\*Treatment number [Incorporation depth (cm), Distance from crop row (cm), Row distance (cm)]. BC, broadcast; I, irrigation.

TABLE 6 | Weed dry weight (g m−<sup>2</sup> ) in treatments 1–15 (see Table 2) on sandy loam and silty clay in 2014–2016.


\*Treatment number [Incorporation depth (cm), Distance from crop row (cm), Row distance (cm)]. BC, broadcast; I, irrigation.

# Mineral Fertilizer Equivalent

On average for both soils, the MFE for pelleted MBM (**Table 5**) was 79% in treatment 6, i.e. with similar placement of pellets and the mineral fertilizer used for comparison. However, in one experiment (on sandy loam), the MFE values in several treatments were above 100%, indicating that nutrients other than nitrogen probably limited crop yield. In the other five experiments, MFE increased from on average 62% in the treatment with broadcasting to 85% in the highest yielding treatment (placement 4 cm from crop row, 4 cm incorporation depth). Placement of pellets together with the crop seeds was not a good alternative (average MFE 56%).

### Weed Flora

The weed flora on the silty clay soil was dominated by the planted Sinapis alba L. and the naturally occurring weed species Chenopodium album, L., Elymus repens (L.) Gould and Sinapis arvensis L. The weed flora on the sandy loam comprised many species, including Fumaria officinalis L., Viola arvensis Murr., and Myosotis arvensis (L.) Hill.

The number of weeds was approximately twice as high on the sandy loam (300–400 plants m−<sup>2</sup> ) as on the silty clay soil (150–200 plants m−<sup>2</sup> ) at the first count (GS 30). On the sandy loam, the number of weeds declined by on average 20% from the first (GS 30) to the second count (GS 55) and there were no significant differences in the weed decrease depending on pellet incorporation (p = 0.52) or pellet distance from the crop row (p = 0.79). Average weed plant weight (**Table 6**) on the sandy loam was 0.16 g DM plant−<sup>1</sup> , with a tendency for larger weed plants with greater distance from crop rows (p = 0.13). An exception was treatment 6 (seeds and pellets placed together), which had the largest weed weight (0.21 g DM plant−<sup>1</sup> ). On the silty clay, weed numbers increased, from on average 155 plants m−<sup>2</sup> at the first count (GS 30) to 210 plants m−<sup>2</sup> at the second count (GS 55), with no significant differences in the weed increase depending on pellet incorporation (p = 0.53) or pellet distance from the crop row (p = 0.70). On the silty clay, weed plant weight (**Table 6**) was significantly higher (p = 0.001) when pellets were incorporated to 4 cm (0.075 g DM plant−<sup>1</sup> ) compared with 1 cm (0.003 g DM plant−<sup>1</sup> ), with a tendency (p = 0.066) for larger weed plants when pellets were placed 12.5 cm from the crop row.

Average weed biomass at panicle emergence (GS 55) of the oat crop was similar on the two soil types (**Table 5**). On the silty clay, weed biomass was significantly affected by pellet placement, with higher weed biomass with deeper incorporation of the fertilizer (p = 0.045) and greater distance from the crop row (p = 0.049) (**Figure 3**). On the sandy loam, there were no statistically significant effects of placement (p = 0.7), but for the treatments with 1 cm incorporation (treatments 9-11) there was a tendency for higher weed pressure with greater distance from the crop row, whereas no such tendency was observed at deeper incorporation depths.

# DISCUSSION

### Differences Between Soils

The effects of placement differed between the two soils studied, which were chosen to represent a clay and a sandy loam soil. However, the sandy soil had received regular doses of farmyard manure over time, whereas the clay soil had not received any farmyard manure during the previous 50 years. The difference in nitrogen offtake in the unfertilized treatment on these soils indicated that the sandy loam delivered more than twice as

much nitrogen as the silty clay. The sandy loam also had larger crop biomass, on average 10,400 kg DM ha−<sup>1</sup> for all years and treatments, compared with only 5,800 kg DM ha−<sup>1</sup> on the silty clay. This means that competition for water and light between crop and weeds was tougher on the sandy loam, which could explain why the larger amounts of weeds on the sandy loam did not result in higher weed biomass than on the silty clay. The tougher competition with the crop probably also affected differences in weed biomass between treatments, as weed biomass tended to be larger in treatments where yield was lower, such as treatment 6 where yield was suppressed and weed biomass elevated. On the clay soil, both weeds and crop were favored by incorporation of fertilizer, indicating that competition was not important for the outcome. The soil nutrient status of the sandy loam is more typical of organic farms in Sweden, and the results for that soil are therefore more applicable for making fertilizer recommendations for organic grain crops.

# Nitrogen or Other Nutrients

Fertilizer experiments with organic residues containing several nutrients are often designed to study the effect of one element at a time. This is usually achieved by adding excess amounts of the other nutrients, to ensure that they do not limit crop growth in any treatment. This was not done in this experiment, since we wanted to study the total effect of different pellet placements. However, we assumed that the main limiting nutrient would be nitrogen and we therefore chose a nitrogen fertilizer (without P and K) for comparison (treatment 2). However, the high MFE values (>100%) in some experiments (**Table 5**) indicate that other elements may have limited yield in treatment 2. This was especially the case in the experiment on sandy loam in 2015, where treatment 10, with the same pellet placement as the mineral fertilizer in treatment 2, had a MFE value of 160%. That field had a low potassium value (**Table 1**), suggesting there was a potassium fertilization effect from the pellets in addition to the nitrogen effect. The higher MFE value in treatments other than treatment 10 could be partly attributable to better placement of fertilizer in these treatments than in the control (treatment 2).

# Absence of Weed Hoeing

No weed hoeing was conducted at early growth stages in this study, since we wanted to see how weeds were affected by crop fertilization. However, weeds were removed at panicle emergence, so they did not affect subsequent yield. In addition to weed density, weed hoeing could affect pellet incorporation, as pellets placed mid-row at shallow depth could be incorporated into soil, which could potentially affect nutrient availability. Other studies have examined weed survival after hoeing depending on fertilization, with variable results. For example, Melander et al. (2002) obtained higher yield of winter wheat when nitrogen fertilizer was incorporated into soil, but no effect on weeds surviving hoeing. Rasmussen (2002) studied the effect of weed control depending on slurry application strategy to spring cereals and found that both mechanical and chemical weed control were more efficient if the slurry was injected rather than surface-applied. In barley, weed numbers were reduced with slurry injection and no additional weed control measures, which could be explained by earlier crop development with injection.

# Crop Row Spacing

According to Petersen (2005), rapid and high N utilization by the crop and low N uptake by weeds can be achieved by high seed density, short distance between crop row and band-applied slurry, and/or early seeding. The treatment with conventional crop row spacing (13: 12.5 cm) in this study confirmed that it often gave higher yield than the double spacing when fertilization was performed in the same way (treatment 12). However, this difference was much smaller than that caused by fertilizer placement, especially on the sandy loam. If pellets were subsurface-banded to oats with 12.5 cm spacing, distance from the crop row should be less important since it can never be more than 6.25 cm. Band application would probably still be as successful as in 25 cm spacing, as long as pellets are not put together with the seeds. The benefit could be because of more efficient incorporation and lower immobilization (Sørensen and Jensen, 1995; Delin and Strömberg, 2011).

### Incorporation Effects

Incorporation of organic fertilizers into soil is often justified by its reducing effect on ammonia emissions (Webb et al., 2010). However, ammonia emissions are not considered a risk with pelleted fertilizer (Adeli et al., 2012), and incorporation is motivated instead by the assumption that the pellets need close contact with moist soil for the nutrients to be released and accessible to crop roots. We suspected that incorporation would not be needed in wetter conditions, since the pellets would enter moist soil even with surface broadcasting, so we included two treatments with irrigation directly after fertilization. However, we found that the effect of subsurface banding persisted in irrigated soil. The effect of subsurface banding can therefore not be explained by incorporation into moister soil, but rather to a better position for crop roots to reach the nutrients.

### Significance of Results in Relation to Other Published Findings

The results confirm that placement of pelleted MBM benefits yield compared with broadcasting. Similar findings have been made for mineral fertilizers, e.g., a study by Rasmussen et al. (1996) found 28% higher yield in spring barley with placement compared with broadcasting. Distance of placement of mineral fertilizer from the row did not have a consistent effect on final yield in previous investigations (Petersen, 2001), although crop nitrogen uptake was delayed by 0.5 day for every 1 cm increase in distance from the crop row. The magnitude of the delay and whether it makes a difference for crop nutrient uptake probably vary depending on soil properties, climate conditions, and competition for nutrients from weeds and microorganisms. In our experiments, yield was reduced by on average 60 kg ha−<sup>1</sup> per cm increase in distance when comparing yield in treatments with placement 4 and 12.5 cm from crop row (**Figure 1**), but the variation depending on year, site, and incorporation depth was 20–200 kg ha−<sup>1</sup> . Previous studies examining the effects of placement of organic fertilizers to spring cereals have mainly focused on animal slurry. Petersen (2003) reported an increase in crop N recovery in spring barley from 45 to 50 % when using band injection of pig slurry instead of broadcasting. A similar increase in N uptake was observed in the present study, where about 5–15 kg more N was taken up in the crop with optimal placement compared with broadcasting (**Figure 3**), which corresponds to around 10% of the total crop N uptake. Similarly to Rasmussen et al. (1996), we found that the effects on weeds were primarily secondary, i.e., that a well-fertilized crop competed better against weeds. In contrast, Blackshaw (2005), who studied yield of spring wheat and weed density of different weeds, found significant effects on weed biomass in some cases, without any significant effects on crop yield. The effects on weeds are most likely a combination of fertilization and competition with the crop.

# Implementation in Farm Fields

Placement of organic fertilizers with centimeter precision may sound impracticable at farm level. However, with modern technology such as System Cameleon (Gothia Redskap, Fornåsa Sweden), a multifunctional system designed for precision seeding, fertilization, and weeding in organic crop production, this is now a reality. The findings in this study should be useful to guide the technological development of such machinery in the right direction and to assist farmers using such systems. In parallel with our microplot experiments, we performed field experiments with some selected treatments on organic farms using their field equipment. We also performed a few field experiments with a combidrill in experimental plots at a silty clay site. These experiments confirmed that grain yield increases with subsurface banding compared with broadcast, as shown in the microplots. The effect of placement distance from the crop row was more difficult to evaluate in these experiments, with few treatments to compare and with different set-ups in different experiments. However, although not statistically significant, placement 4 cm from the crop row tended to give higher yield than placement midway between rows, while placement under the crop row resulted in earlier crop nitrogen uptake than placement midway between rows. However, in one experiment in a farmer's field, placement midway between rows ultimately resulted in higher yield than placement under the crop row. This could be related to dry weather in that year, in combination with placement under the row negatively affecting moisture conditions for crop roots. Another possible explanation is that the weed hoeing performed by the farmer incorporated and mixed the pellets well into the soil and thus had a beneficial effect on nitrogen release from the pellets placed midway between the rows.

# CONCLUSIONS

The results of this study show that farmers who use equipment for precision placement of pelleted fertilizers can double the grain yield effect from their pelleted MBM compared with using broadcasting and shallow incorporation with a harrow. To achieve this, farmers should aim at placement about 4 cm from the crop row and with at least 4 cm soil incorporation. These effects do not seem to be dependent on moisture conditions. The effect of pellet placement on weed density is small, with a highly competitive crop appearing to be more important for reducing weed plant size than limited weed access to nutrients.

# AUTHOR'S NOTE

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

### AUTHOR CONTRIBUTIONS

AL, SD, and LE designed the study; LE and SD performed the experiments; SD and LE analyzed and/or interpreted the data. SD wrote the manuscript; LE and AL revised the manuscript for important intellectual content. All authors approved the final version of the manuscript and agree to be accountable for the content of the work.

### REFERENCES


### ACKNOWLEDGMENTS

The study was funded by SLU EkoFOrsk, Sweden. The authors wish to thank Per Ståhl, Emil Olsson, and Roland Höckert for valuable input from the farmers' viewpoint when setting up objectives and designing experiments. We also wish to thank the staff at Lanna Research station for providing field sites and practical assistance. Many thanks to Markus Delin, Johanna Wetterlind, Karin Wallin, and May Ibrahim for help with counting and harvesting weeds in a busy period.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Delin, Engström and Lundkvist. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Nitrogen Recovery With Source Separation of Human Urine—Preliminary Results of Its Fertiliser Potential and Use in Agriculture

Eeva-Liisa Viskari <sup>1</sup> \*, Gerbrand Grobler <sup>1</sup> , Kaisa Karimäki <sup>1</sup> , Alexandra Gorbatova<sup>1</sup> , Riikka Vilpas <sup>2</sup> and Suvi Lehtoranta<sup>2</sup>

<sup>1</sup> School of Construction and Environmental Engineering, Tampere University of Applied Sciences, Tampere, Finland, <sup>2</sup> Finnish Environment Institute SYKE, Helsinki, Finland

### Edited by:

Tom Misselbrook, Rothamsted Research (BBSRC), United Kingdom

### Reviewed by:

Paolo Mantovi, Centro Ricerche Produzioni Animali, Italy João Coutinho, University of Trás-os-Montes and Alto Douro, Portugal Harald Menzi, Federal Office for the Environment, Switzerland

> \*Correspondence: Eeva-Liisa Viskari eeva-liisa.viskari@tamk.fi

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 31 January 2018 Accepted: 11 June 2018 Published: 28 June 2018

### Citation:

Viskari E-L, Grobler G, Karimäki K, Gorbatova A, Vilpas R and Lehtoranta S (2018) Nitrogen Recovery With Source Separation of Human Urine—Preliminary Results of Its Fertiliser Potential and Use in Agriculture. Front. Sustain. Food Syst. 2:32. doi: 10.3389/fsufs.2018.00032 The growing demand for food and the increasing costs of cultivation are posing a challenge for agriculture. Diminishing phosphorus reserves, as well as the energy intensive method of producing nitrogen fertilisers are drivers for more intensive reuse of different organic fertilisers, such as manures and excreta. Source separation and fertilisation with human urine can be one option for nutrient reuse. Urine contains all the main nutrients as well as micronutrients in soluble form, but it also contains chemicals, like pharmaceuticals and hormones. The aim of this study was to examine the efficiency and safety of the use of source separated human urine as a fertiliser for barley (Hordeum vulgare). The fertiliser efficiency of source-separated urine was examined in field-scale experiments for the first time in Finland. Two separate cultivation experiments in two fields and barley varieties were conducted. The efficiency of urine as a fertiliser was compared to corresponding amount of mineral fertiliser. No fertiliser was applied to one plot in order to create a reference treatment. The two experiments were conducted using variety Wolmari with 54 kg N ha−<sup>1</sup> and variety Harbinger with 100 kg N ha−<sup>1</sup> . The barley grain and straw yield grown with urine fertiliser was equivalent to the yield in mineral fertilised plots. The growth of barley in both fertiliser treatments was slightly faster, compared to non-fertilised treatment. There were no significant differences between the treatments in terms of protein content of the grain although the results varied in terms of the thousand grain weight (TGW) and germination. The urine analyses indicated that there were no pathogen indicators, nor heavy metal concentrations, exceeding the limit values set by legislation. The main nutrient concentrations (N, P, K) would also meet the requirements for a fertiliser product according to Finnish legislation. Pharmaceuticals and hormones were found from the urine, but apart from progesterone, all of them presented extractable values in soil below the detection limits, and they were not detected in measurable amounts in barley grain at the end of the growing season. These results suggest that source separated urine could be an efficient fertiliser in crop cultivation.

Keywords: urine, fertiliser, barley, source separation, pharmaceuticals, microbiological quality, yield

# INTRODUCTION

Phosphorus is an essential nutrient in crop cultivation. It is estimated, that there are affordable and relatively easily extractable phosphate rock reserves left for only another 50–100 years. Also, the quality of the phosphorus mineral is diminishing. Yet there is an ever increasing demand for phosphorus in agriculture (Cordell et al., 2009, 2011). The production of nitrogen fertilisers—another crucial element in food production - in turn is highly energy intensive. Therefore, in the current need for increasing resource efficiency, we need to look for more intensive nutrient reuse for agricultural purposes. Source separation of human urine offers one option for nutrient reuse from human waste. Urine can be diverted from the solid excreta and it can be used as a liquid fertiliser as such. This diversion enables reuse of waste for agriculture purposes and at the same time protects the natural water bodies from waste pollution and eutrophication (Vinnerås and Jönsson, 2002). Source separated human urine is a nutrient-rich liquid where the main nutrients (N, P, K) occur in water-soluble ionic form and are therefore readily available for plant uptake (Schönning, 2006; Udert et al., 2006). The use of source separated human urine could be one option to complete the demand of nitrogen fertilisation and to close the nutrient cycle.

Source separated urine has been found to be a safe and efficient fertiliser for many crops and vegetables and has been studied in many different contexts since the late 1990s (Kirchmann and Pettersson, 1995; WHO, 2006; Heinonen-Tanski et al., 2007; Mnkeni et al., 2007; Pradhan et al., 2007; Chowdhury and Islam, 2008; Viskari et al., 2009; Pradhan, 2010). However, in Finland and in many other European countries, urine is not accepted as a fertiliser, and there are many institutional constraints in its use.

Source separated urine has the potential to be used as a fertiliser because of its nutrient content, availability and easy application to soils. The main nutrients (nitrogen, phosphorus, potassium and sulphur) occur in water-soluble ionic form and are therefore readily available for plant uptake (Schönning, 2006). The majority of the nutrients are excreted via urine (Schouw et al., 2002; Rose et al., 2015). From nitrogen about 90%, phosphorus 50–65% and potassium, 50–80% is excreted in urine (Lapid, 2008). In addition, urine contains Cl−, Na, Mg, Cu and other organic and inorganic compounds, which can be utilised for plant growth. The amount of nitrogen and phosphorus in urine varies depending on the diet, person and time of the day. For example, nitrogen present in urine can reach a concentration of up to 9 g N L−<sup>1</sup> , the concentration of phosphorus is around 0.7 g P L−<sup>1</sup> (Winker et al., 2009). In comparison with other household waste, urine contains considerably more nutrients than faeces, greywater and biodegradable suspended solids (Vinnerås and Jönsson, 2002; Pradhan, 2010). **Table 1** shows the estimated amounts of nitrogen, phosphorus and BOD produced in urine and faeces per capita annually. The estimated amounts vary, however, depending on the collection and analysis method. For example, Vinnerås and Jönsson (2002) estimated that about 3.7 kg of nitrogen, 0.34 kg of phosphorus and 1.2 kg of potassium per capita/a could be recovered if all urine could TABLE 1 | The estimated amount of nutrients produced in excreta and greywater by one person per year (Weckman, 2005; Udert et al., 2006; Ministry of the Environment Finland, 2017).


be fully separated from other wastewaters. Nitrogen in urine is mostly in the form of urea and/or ammonium (Kirchmann and Pettersson, 1995; Richert Stintzing et al., 2001). Phosphorus (P) and potassium (K) in turn are almost entirely in inorganic, ionic form (Lentner, 1981), which is directly plant-available (Kirchmann and Pettersson, 1995), although inorganic P in soil may be rapidly, and extensively adsorbed to Al, Fe minerals, or precipitated with Ca.

Concerning the hygienic issues, the urine of a healthy person contains only a small amount of pathogens, originating mostly from faecal contamination (Höglund et al., 1998). Storage is an easy method for the urine disinfection in case of any contamination (Höglund, 2001). The current recommendation for urine storage is 6 months at a temperature over 22◦C (Jönssön et al., 2004; Schönning, 2006; WHO, 2006). However, research on the disinfecting effects of urea showed that no E. coli or Salmonella spp. were found after 5 days of storage. Significant reduction in phage was observed after 21 days and no viruses were found after 50 days. The study indicates that the storage time for safe urine reuse is 2 months at a temperature over 20◦C when the nitrogen concentration in the urine is greater than 2 g N L−<sup>1</sup> and pH is over 8.8. (Vinnerås et al., 2003, 2008; Winker et al., 2009).

Concerns about the use of urine as a fertiliser have also arisen. The potential risks of various pharmaceutical and hormone residues as well as other potential micro-pollutants as contaminants are seen as an obstacle. Knowledge of their presence in urine and their behaviour in the soil is still limited. Humans are exposed to different contaminants, like heavy metals, in our environment through skin contact, respiration and diet and the contaminants are excreted in sweat, urine and faeces (e.g., Schouw et al., 2002; Genuis et al., 2011). Compared with animal manure and industrial fertilisers the heavy metal concentrations of urine are lower (Jönssön et al., 2004; Winker et al., 2009). On the other hand, on average, 2/3 of the drugs used by humans are excreted via urine and about 1/3 in the faeces (Lienert et al., 2007). The potential health risks associated with the use of urine and faeces as fertilisers have been extensively studied, and WHO guidelines have been developed for the urine treatment and use (WHO, 2006).

There are existing technologies for nutrient recovery, such as separating dry toilets. There are many factors, however, that are preventing or hindering the nutrient recovery on a larger scale. These are for example public opinion toward the use of urine as a fertiliser, missing storage facilities and logistical chains, as well as legislative barriers (Magid et al., 2006; Lienert and Larsen, 2010).

The aim of this study was to acquire scientific data on the use and efficiency of urine in crop cultivation. For the first time in Finland, field scale experiments using urine as a fertiliser were conducted using barley as a test crop.

# MATERIALS AND METHODS

The feasibility of source separation of urine, management and potential as a fertiliser was examined in this study. The fertiliser efficiency of source-separated urine was tested in two field-scale experiments using barley (Hordeum vulgare, var. Wolmari and Harbinger) as a test plant. The experiments were implemented in cooperation with two farms within 80 kilometres from Tampere (Central Finland). Urine fertilisation was compared with corresponding levels of mineral fertiliser and non-fertilised sites were used as a reference within the experimental fields. The amount of urine spread was defined by farmers, and was based on the field-specific requirements in terms of nitrogen requirement. Other environmental requirements, originating from farming subsidies and environmental regulations were also determining the fertilisation levels in the fields. Experiments were carried out in ways that would cause the least disturbance to the farmers.

### Ethics Review

According to the guidelines of Finnish National Board on Research Integrity and statement from the Academic Ethics Committee of the Tampere Region, the body dealing with ethical statements of Tampere University of Applied Sciences, ethics approval and further consent was not required in this kind of research study. Written and informed consent concerning the implementation of experiments and publication of the results was obtained from the farmers that participated this study. The urine used in the study was collected from thousands of anonymous participants of a festival in August 2015. No oral or written consent were asked from them. Anonymity and discretion, however, was assured at every step in the urine management during this study. From the private household, that also donated urine for the project an oral and informed consent was obtained and the anonymity assured by keeping the detailed information of the donors only at the attention of the authors.

### Urine as Fertiliser

Urine was collected from two different sources. The largest amount, total of 4.8 m<sup>3</sup> of urine, was collected from male urinals of a Weekend Music Festival with about 40,000 participants in Helsinki in August 2015. Another set of urine (1.5 m<sup>3</sup> ) was collected with a separating dry toilet from a private household with a family of four persons. Both of the urine batches were stored over winter in sealed 1 m<sup>3</sup> IBC-containers until used in spring 2016. The storage fulfilled the criteria for safe management and use of human excreta in agriculture (WHO, 2006). The microbial quality, nutrient and element content as well as pharmaceuticals and hormone concentrations were analysed from the urine. Summary of different analysis methods of urine is presented in **Table 2**.

TABLE 2 | Summary of chemical and microbiological analysis methods used in this study.


# Field Experiments

The potential of urine as a fertiliser was studied by comparing it with a corresponding amount of nitrogen in mineral fertiliser and using non-fertilised plots as a reference. The experimental fields were divided into three different plots with the different fertiliser treatments (**Figures 1A,B**). Monitoring was conducted in and samples taken from square shaped sampling plots sized 50<sup>∗</sup> 50 cm in each fertiliser treatment plot. In Field 1 there were seven sampling plots per fertiliser treatment, resulting in 21 sampling plots in total and in Field 2 10 sampling plots, resulting in 30 sampling plots in total. The sampling plots were placed to the different fertiliser treatment sites using systematic placing into lines. The soil physico-chemical characteristics and main nutrient content of the soil in experimental plots were analysed. The analysis methods are summarised in **Table 2** and results are presented in **Table 3**.

There were two different barley varieties, Wolmari and Harbinger, and two nitrogen fertilisation levels (54 kg N ha−<sup>1</sup> in the field 1 and 100 kg N ha−<sup>1</sup> in the field 2) were used in the experiments respectively (**Table 4**). The urine was spread about 2 weeks after sowing to ensure the efficient uptake and use of the nutrients in the beginning of the growth. In field 1, urine was spread using a slurry tanker equipped with a disc injector while in field 2 urine was spread manually by simulating deep injection. The simulation was carried out using a sharpened metal tube attached to a 10 l watering can and by pressing the tip of the tube into the soil while pouring the urine into the

tube. When injecting urine directly into the soil, nitrogen losses are reduced and nutrient uptake enhanced (Johansson, 2000). Mineral fertiliser was amended simultaneously with the seeds. **Table 4** summarises the experimental setup in different fields.

Barley growth and condition was monitored every second week by recording the growth stages of the stems using Zadoks cereal development scale with 99 different stages (e.g., Boys and Geary, 2015), which describes the growth and ripening of the cereal. At the end of the growing season both the barley grains and straw were harvested from the sampling plots and biomass, protein content of the grain, germination and weight of thousand grains (TGW) measured. These results were compared



N = 10/field.

between the different fertiliser treatments. Comparison of the results were also made with the official variety test results made by the Natural Resources Institute Finland (Luke). Official variety tests are implemented cultivating the different crop varieties "using common cultivation methods" and in areas suitable for the cultivation of the crop in question (Laine et al., 2016). This means that fertilisation needed is defined by, for example, the soil quality, desired yield and the level of nitrogen required. The yield results from the 10 sampling plots (0.25 m<sup>2</sup> ) were extrapolated to hectare level. Furthermore, pharmaceutical and hormone concentrations of the soil and barley grains were analysed at the end of the growing season to see the possible accumulation of these micro-pollutants in the soil or yield.

### Statistical Analyses

Descriptive statistical analyses of barley yield and characteristics were carried out using IBM SPSS Statistics software (version 22). Mean values and standard deviations are presented from the yield data. No comparison or statistical tests were made between the fields, because of different locations, barley variety and fertilisation levels used.

### RESULTS

### Urine as Fertiliser

The urine contained relatively high amounts of nitrogen and other macro and micro-nutrients (**Table 5**). Overall average NPK-ratio of all urine collected (20-1.2-4) was almost equivalent to a commercial mineral fertiliser, for example YaraMila Y1 (27-1.3-4). The NPK ratio of urine in Field 1 with 54 kg N ha−<sup>1</sup> was 28-1.6-4.9 and Field 2 with 100 kg N ha−<sup>1</sup> 25-2.3- 7.3 respectively. Stored urine contained no pathogen indicators Salmonella or E. coli and the concentrations of harmful metals were significantly below the limit values given in the Finnish legislation (Ministry of Agriculture and Forestry Finland, 2011).

A total of 55 pharmaceuticals and hormones were analysed and total of 16 drugs, especially anti-inflammatory drugs, were found in the urine (**Table 6**). However, none of the analysed extractable pharmaceuticals were found in soil or barley grains at the end of the growing season.

All extractable hormone concentrations, except progesterone, were below the detection limits of the analysis methods both in the soil and grain sample. Exactly the same amount of progesterone was found from every grain sample, 3 µg kg−<sup>1</sup> DM, regardless of the fertiliser treatment. It was calculated that 3.1 µg



and 195 µg progesterone per m<sup>2</sup> were spread into the test fields 1 and 2 respectively. The detection limit for progesterone was 1 µg kg−<sup>1</sup> DM.

# Barley Growth, Yield, and Quality

The differences in the barley growth with the different fertiliser treatments are shown in **Figure 2**. It demonstrates the late phase of the growing season in the test fields at the end of July 2016. Based on the harvest, urine was found to be as efficient as a mineral fertiliser. The yield with urine was markedly higher compared with the yield without fertilisation (**Figure 2**). The barley yield of variety Wolmari in the 54 kg N ha−<sup>1</sup> treatment was on average 6,200 kg ha−<sup>1</sup> with urine fertiliser, 6,800 kg ha−<sup>1</sup> with mineral fertiliser and 4,500 kg ha−<sup>1</sup> for the nonfertilised treatment. With the variety Harbinger in the 100 kg N ha−<sup>1</sup> treatment, the yield was 7,600, 7,200, and 4,400 kg ha−<sup>1</sup> respectively. For straw yield, the trend was exactly the same as with the grain yield with variety Wolmari. With variety Harbinger the straw yield was higher in mineral fertilisation than urine fertilisation (**Figure 3**). At fertiliser rate 54 kg N ha−<sup>1</sup> , the straw yields were on average 2,800, 3,300, and 1,800 kg ha−<sup>1</sup> in the urine fertilised site, mineral fertilised site and in non-fertilised site. At fertiliser rate 100 kg N ha−<sup>1</sup> , the straw yields were 3,900, 4,900, and 2,400 kg ha−<sup>1</sup> , respectively.

Viskari et al. Nitrogen Recovery With Source Separation



Results are average values from two replicate analyses.

<sup>a</sup>Urine collected from male urinals used in the Weekend-festival in August 2015. <sup>b</sup>Urine collected from a private household with 4-person family using a source separating dry toilet.

\*(Ministry of Agriculture and Forestry Finland, 2011), density ∼1 kg/m<sup>3</sup> .

Thousand-grain weight (TGW) is an indicator of the size of the grain (e.g., Škarpa, 2006). It is a commonly used parameter in describing the quality of yield and varies typically between 20 and 40 g. In **Figure 4**, the results of TGW in the different fertiliser treatments are presented. In the lower nitrogen fertilisation (54 kg N ha−<sup>1</sup> ) TGW of fertilised barley was higher than nonfertilised barley, but in the higher nitrogen fertilisation (100 kg N ha−<sup>1</sup> ) it was the opposite (**Figure 4**).

Fertiliser treatment had no effect on the protein content or germination of the grain at either fertiliser rate or varieties used (**Figures 5**, **6**). The total protein content of the grains with variety Wolmari was 96–98 g kg−<sup>1</sup> in different treatments, i.e., clearly lower than expected, also reflecting the lower fertilisation level. With variety Harbinger the total protein content in this study was 100–105 g kg−<sup>1</sup> in different fertiliser treatments. Both results were lower than the results from official variety tests. For the variety Wolmari, the expected protein content of the grain was 119 g kg−<sup>1</sup> and for Harbinger 117 g kg−<sup>1</sup> (Laine et al., 2016) (**Figure 5**).

At both fertiliser rates and treatments, the Zadoks growth stage (Boys and Geary, 2015) of barley was slightly slower in the non-fertilised sampling plots compared with the urine- and mineral-fertilised plots. In field 1 (54 kg N ha−<sup>1</sup> ) with variety Wolmari the growth without fertiliser was clearly slower at first, but at the ripening stage the barley reached the same growth stage as the barley with urine and mineral fertiliser treatment. In field 2 (100 kg N ha−<sup>1</sup> ) with higher nitrogen levels, the growth stage of barley without fertiliser was slightly behind compared with the fertilised treatments throughout the whole growing season (**Figures 7**, **8**).

### DISCUSSION

### Urine Quality and Characteristics

The urine analysis results of this study show that source separated urine fulfils the criteria of the fertiliser products according to the Finnish legislation (Ministry of Agriculture and Forestry Finland, 2011) in terms of microbiological quality (E. coli and Salmonella) and harmful metal concentrations. All studied parameters were below the limit values defined in the legislation. The results indicate that there is no increased heavy metal input from the use of urine. This is in accordance with earlier findings of Kirchmann et al. (2017) and EEA (2018), which state that the overall heavy metal exposure in the environment has decreased remarkably during the past decades and therefore, the concentrations in wastewaters have also decreased. The results supported also the findings and recommendations in the WHO guidelines (2006).

There is, however, variation in the nutrient concentration depending on the source and the way of collecting the urine. For example, nitrogen concentration of urine can vary from 1 to 9 mg L−<sup>1</sup> , depending on the diet and time of collection (Pradhan, 2010). Based on the nutrient content and the fertiliser efficiency source separated urine meets the criteria for fertiliser products that can be used as such as soil improvers (Ministry of Agriculture and Forestry Finland, 2011). Therefore, in principle, there are no restraints from the quality point of view, to accept source separated urine as a fertiliser, as long as it is correctly stored and managed (WHO, 2006).

Urine contains significant amounts of salts, like sulphate and chloride. In this study, the sulphate concentration was on average about 500 mg L−<sup>1</sup> and chloride concentration about 1,000 mg L−<sup>1</sup> . In long-term fertiliser use, high salt concentrations could be a risk in terms of soil salinization and therefore needs further investigation and long-term field trials.

When urine is stored according to WHO guidelines (2006), the pH of urine increases to a level of between 9 and 10. In this process, urea is hydrolysed to ammonia. Ammonia and other substances in urine are causing an unpleasant odour. Therefore, storage in sealed containers is very important. When using the urine as fertiliser, deep injection is crucial to prevent nitrogen losses into the atmosphere. This also helps preventing odours spreading to the environment. In this study odour was detected


TABLE 6 | Concentrations of pharmaceuticals and hormones found in urine samples used in different field experiments.

Total number of pharmaceuticals and hormones analysed was 55 and the results are means of two replicate analyses.

<sup>a</sup>Urine collected from male urinals used in Weekend-festival in August 2015.

<sup>b</sup>Urine collected from private household with family of four persons using source separating dry toilet.

for only a few minutes after injection to the soil. According to the studies by Johansson, 2000), there is on average about 2–10% atmospheric loss of nitrogen from the field, when urine fertiliser is applied using band spread with trailing hoses. The variation in nitrogen losses to the atmosphere can be high, however, from a few per cent up to 30–40%.

FIGURE 4 | Thousand grain weight (TGW) of barley (±SD) (Hordeum vulgare, var. Wolmari and Harbinger) (kg ha−<sup>1</sup> ) with urine, mineral fertiliser and non-fertilised treatments. N = 30 \*Samples located in clay soil were not included in of analyses, N = 21.

FIGURE 6 | Germination of barley (±SD) (Hordeum vulgare, var. Wolmari and Harbinger) (kg ha−<sup>1</sup> ) with urine, mineral fertiliser and non-fertilised treatments. N = 30. \*Samples located in clay soil were not included in the analyses, N = 21.

# Barley Growth Stages, Yield, and Quality

In the long term field trials made in Denmark (Magid et al., 2007), nitrogen uptake and mineral fertiliser equivalent (MFE) of urine fertilisation to cereal crops, like barley, oats and wheat, was found to be very good and in some years even a better fertiliser than NPK-fertiliser or cattle slurry. Cattle slurry as well as mineral NPK-fertilisers are widely used and accepted fertilisers, while source separated urine is not. Based on the trials urine was found to be a "very reliable fertiliser" (Magid et al., 2007). In this study, results are limited to only one growing season, which is not enough to make long term conclusions. The results indicate, however, a trend, which has been confirmed in other long term cultivation experiments (e.g., Magid et al., 2007).

The barley yield results showed that barley grown with urine fertiliser was equivalent to the mineral fertilised barley. Total grain yield was at the same level with both varieties (Wolmari and Harbinger) and nitrogen fertiliser rates (54 and 100 kg N ha−<sup>1</sup> ). Yields were approximately 60 and 70% higher than the unfertilised treatment for the higher and lower fertiliser application rates, respectively. The trend was exactly same with straw yield with variety Wolmari, where the straw yield was the same in fertilised treatments. With variety Harbinger, straw yield in mineral fertilisation was higher than urine fertilisation. These results are in accordance with the previous findings with grain crops like barley indicating the efficiency of urine as fertiliser (Kirchmann and Pettersson, 1995; Johansson, 2000). Factors affecting the significant differences in both grain and straw yield in different fields are the different fertilisation rates, different variety and slight differences in the soil type. There were no significant differences in the weather conditions in the different fields, since they were located about 50 km apart.

In **Figures 2**, **4**, **5** the official variety test results, which are regularly implemented by Natural Resources Institute Finland, are indicated for comparison to the results of this study (Laine et al., 2016). The total grain yield was clearly higher than the yield indicated in the official variety tests. This might be due to the extrapolation from the small sampling plots to hectare, which magnifies any minor errors in the accuracy of the harvesting. The differences between the treatments, which was the focus in this study, however, were statistically significant. They indicate that urine as a fertiliser is as efficient as mineral fertiliser. The thousand-grain weight (TWG) and the protein content of barley were on the same order of magnitude as the official tests indicated (**Figures 4**, **5**). There was also variation in the results of TGW. The lower TGW in 100 kg N ha−<sup>1</sup> fertilised treatments compared to the non-fertilised treatment might indicate the enhanced growth of total biomass of grain and straw, leaving the grain smaller in size. In the lower nitrogen fertilisation 54 kg N ha−<sup>1</sup> in turn the grains were bigger in both fertilisation treatments compared with the non-fertilised plot, which could indicate allocation of available nitrogen to the growth of the grain.

In addition to the protein content of the grain, the quality of barley yield was also measured by analysing the germination of the seeds. The protein content of variety Wolmari was 9.7– 9.8% in all the different fertiliser treatments. This is about 83% of the official variety test results. This could be due to low nitrogen fertilisation level (54 kg N ha−<sup>1</sup> ) in the test field. With variety Harbinger the total protein content varied between 10 and 10.5% in all the treatments, which is about 91% of the

official variety test results. Fertilisation treatment had no effect on the total protein content of barley in either of the fertilisation levels or barley varieties, not even with non-fertilised treatment. Especially with the variety Wolmari, low protein content might indicate the use of soil nitrogen in the growth phase, leaving less nitrogen to the formation of protein in the grain. There was also no difference in the germination (90–97%) between the different fertiliser treatments and non-fertilised treatment. Previous studies indicate that that nitrogen fertilisation or nitrogen concentration of grain does not affect the germination rate of barley (Ellis and Marshall, 1998).

The fertiliser efficiency of urine depends greatly on the environmental conditions, such as soil type and weather conditions during the growing season. In the experiments made in Sweden in 1996–1998, urine as a barley fertiliser was found almost equally efficient compared to the commercial fertilisers used in the study (Johansson, 2000). The barley yield was 80– 90% of that for the commercial fertilisers. Slightly lower total yield with urine fertilisation was explained by the nitrogen losses to the atmosphere. These losses were 2–10% depending on the year and fertiliser amount used. Urine can be spread without any dilution to cereal crops, which was also done in this study. It is highly recommended to use slurry tankers equipped with either disk or hose injectors to avoid nitrogen losses (Johansson, 2000).

### Pharmaceuticals and Hormones in Urine, Soil, and Yield

Substances that have been orally digested and metabolised are excreted mostly via urine. To some extent, also substances exposed through skin exposure or inhalation are excreted via urine. The substances remain partly unmetabolised. About two thirds of all unmetabolised pharmaceuticals and drugs used are excreted via urine and about one third via faeces (Lienert et al., 2007).

Several different pharmaceuticals and hormones were found in the urine samples. In the urine collected from the festival, in total 16 pharmaceuticals and hormones were found. From the urine collected from the private household, only eight different pharmaceuticals and hormones were found. The range of different pharmaceuticals was greater in the urine collected from the festival. This is obvious, because the number of people using different medication is greater. The groups of pharmaceuticals with the highest concentrations in the urine were antiinflammatory drugs, like ibuprofen, ketoprofen and naproxen, and pain relievers like paracetamol. Other pharmaceutical groups found were antibiotics (sulfamethoxazole, tetracycline, trimetoprim), allergy drugs (methylprednisolone), beta-blockers (bisoprolol, propanolol), anti-depressants (citalopram) and caffeine. The single largest amount of pharmaceutical found was ibuprofen, the concentration being 4,160 µg L−<sup>1</sup> in the urine collected from the festival. The urine also contained on average 852 µg L−<sup>1</sup> of caffeine, which was included in the pharmaceuticals analysed.

Pharmaceuticals in urine can be of concern if any accumulation or other disturbance in soil and plant growth takes place. Previous research studies have shown that plants can uptake via roots certain persistent pharmaceuticals from soils, such as carbamazepine. These pharmaceuticals can accumulate in the roots and foliage of the plants (Winker et al., 2010; Bartha, 2012; Carter et al., 2014), but the amounts have been so small that it has not been considered as a health risk (Winker et al., 2010). Degradation of certain pharmaceuticals have been studied. Many pharmaceuticals, especially antibiotics, are both biodegradable (Winker et al., 2009) and photodegradable (Doll and Frimmel, 2003). For example anti-inflammatory drug naproxen (Topp et al., 2008), and antibiotics triclosane and triclocarbane (Prosser et al., 2014) degrade in soil almost completely and do not accumulate to plants or disturb their growth. Furthermore, several pharmaceuticals have, in fact, been found to degrade quite rapidly (Carter et al., 2014; Song and Guo, 2014), but there are groups of pharmaceuticals which are more persistent in soils. These are for example carbamazepine, diphenhydramine and fluoxetine, which might accumulate and cause risks to the soil environment (Wu et al., 2010). Antibiotics in soils are of special concern, because in the long run, they might cause increased resistance to antibiotics in soil microbes.

The accumulation of pharmaceuticals to plants depends on the characteristics of the substance, especially the biodegradability and adsorption, but also soil characteristics, such as organic matter content and pH (Jjemba, 2002; Song and Guo, 2014). Our knowledge on the fate and degradation of pharmaceuticals in agricultural soils is still limited, especially in the Finnish context.

Since there was no extractable progesterone found in the soil at the end of the growing season, and the concentrations in urine varied greatly in different fields, it is more likely that the question is about the plants' own formation of progesterone. In the literature, there are indications that plants can form progesterone also naturally (Janeczko, 2012; Janeczko et al., 2013). The significance of endogenic progesterone to the plants, however, is unknown.

At the end of the growing season, two replicate mixed soil and barley grain samples were taken from both fields. The traces of extractable pharmaceuticals and hormones was analysed. Apart from the progesterone traces found in the grain (3 µg kg−<sup>1</sup> DM), the concentrations of all extractable pharmaceuticals and hormones remained below the detection limit. This suggests that pharmaceuticals are likely to degrade during the growing season and do not accumulate in the barley, or the concentrations have been below the detection limit of the analysis method. Therefore, there seems to be no risk with pharmaceutical accumulation, which is supported by earlier findings (Topp et al., 2008; Carter et al., 2014; Prosser et al., 2014; Song and Guo, 2014). Our findings, however, are suggestive. There are no long-term trials about the potential accumulation of pharmaceuticals and there might be also other micropollutants, such as microplastics, pesticides etc., to which we are also exposed and which might be excreted via urine. In terms of progesterone, it is possible that there is endogenic production of it in cereal crops. According to Janeczko (2012), plants can produce progesterone without external accumulation and endogenic progesterone production has been found for example from wheat (Janeczko et al., 2013).

Pharmaceuticals in urine can be a limiting factor for the fertiliser use, if they are found in large amounts. Most of the pharmaceuticals do not degrade during storage (Schürmann et al., 2012). Many pharmaceuticals in urine do not accumulate in struvite, which can be precipitated from urine (Schürmann et al., 2012; Kemacheevakul et al., 2014). Also zeolite treatment can be a promising technology in the removal of antibiotics from wastewaters (Malakootian et al., 2016). There are some pharmaceuticals, however, like tetracycline-antibiotics that accumulate particularly in struvite (Kemacheevakul et al., 2012). In this study, tetracycline was also found in urine, the concentration being about 36 µg L−<sup>1</sup> . Tetracycline is one of the antibiotics that are used in treating farm animals and therefore can also be found in soils where manure is used as a fertiliser (Brambilla et al., 2007). Since the biodegradability and fate of different pharmaceuticals vary, there is a need for more research for example of risks of developing antibiotic resistance and disturbance in soil microbiological processes.

Among the public, there are strong opinions for and against the use of urine as a fertiliser. The environmental risks in terms of pharmaceutical accumulation or harmful heavy metal exposure to soil or crops could not be indicated in this study. Furthermore, the hygienic safety and fertiliser efficiency of urine in terms of barley yield and quality was clearly shown. No pathogen indicators were found and the barley yield was equally good compared with a mineral fertiliser and would meet the requirements of Finnish legislation. Therefore, the urine should be accepted as a fertiliser and the use of source separation and fertiliser techniques could be taken into consideration.

# AUTHOR CONTRIBUTIONS

E-LV designed and was in charge of the implementation of the experiments. She also carried out the data analysis and wrote most of the manuscript. GG, KK, and AG were assisting in the practical field work, laboratory analyses and data collection. RV and SL contributed to the data analysis, manuscript writing and discussion.

# FUNDING

This study was financially supported by the Finnish Ministry of Environment, Programme to promote the recycling of nutrients and to improve the status of the Archipelago Sea, during years 2015-2016 (Project code: RAKI-YM167/481/2014-BIOUREA).

# ACKNOWLEDGMENTS

We wish to thank Sipilä and Mikkola farms and Global Dry Toilet Association of Finland for participation, expertise and good cooperation during the study. The linguistic corrections made by Ms. Taru Owston from Tampere University of Applied Sciences are gratefully acknowledged.

### REFERENCES


fertilisers – experiences from Finland," in Third International Dry Toilet Conference, August 12-14 2009, Extended Abstract in Proceedings (Tampere).


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Viskari, Grobler, Karimäki, Gorbatova, Vilpas and Lehtoranta. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Risk Assessment of *E. coli* Survival Up to the Grazing Exclusion Period After Dairy Slurry, Cattle Dung, and Biosolids Application to Grassland

S. M. Ashekuzzaman<sup>1</sup> , Karl Richards <sup>1</sup> , Stephanie Ellis 1,2, Sean Tyrrel <sup>2</sup> , Emma O'Leary <sup>1</sup> , Bryan Griffiths <sup>3</sup> , Karl Ritz 2,4 and Owen Fenton<sup>1</sup> \*

*<sup>1</sup> Environment Research Centre, Johnstown Castle, Teagasc, The Irish Agriculture and Food Development Authority, Wexford, Ireland, <sup>2</sup> School of Water, Energy and Environment, Cranfield University, Cranfield, United Kingdom, <sup>3</sup> Crop and Soil Systems Research Group, Scotland's Rural College, Edinburgh, United Kingdom, <sup>4</sup> School of Biosciences, The University of Nottingham, Nottingham, United Kingdom*

### *Edited by:*

*Tom Misselbrook, Rothamsted Research (BBSRC), United Kingdom*

### *Reviewed by:*

*Chris John Hodgson, Rothamsted Research (BBSRC), United Kingdom David Oliver, University of Stirling, United Kingdom*

*\*Correspondence:*

*Owen Fenton owen.fenton@teagasc.ie*

### *Specialty section:*

*This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems*

> *Received: 08 May 2018 Accepted: 18 June 2018 Published: 10 July 2018*

### *Citation:*

*Ashekuzzaman SM, Richards K, Ellis S, Tyrrel S, O'Leary E, Griffiths B, Ritz K and Fenton O (2018) Risk Assessment of E. coli Survival Up to the Grazing Exclusion Period After Dairy Slurry, Cattle Dung, and Biosolids Application to Grassland. Front. Sustain. Food Syst. 2:34. doi: 10.3389/fsufs.2018.00034* Grassland application of dairy slurry, cattle dung, and biosolids offers an opportunity to recycle valuable nutrients (N, P, and K), which may all introduce pathogens to the soil environment. Herein, a temporal risk assessment of the survival of *Escherichia coli* (*E. coli*) up to 40 days in line with the legislated grazing exclusion time points after application was examined across six scenarios: (1) soil and biosolids mixture, (2) biosolids amended soil, (3) dairy slurry application, (4) cattle dung on pasture, (5) comparison of scenario 2, 3, and 4, and (6) maximum legal vs. excess rate of application for scenario 2 and 3. The risk model input parameters were taken or derived from regressions within the literature and an uncertainty analysis (n = 1,000 trials for each scenario) was conducted. Scenario 1 results showed that *E. coli* survival was higher in the soil/biosolids mixture for higher biosolids portion, resulting in the highest 20 day value of residual *E. coli* concentration (i.e., C20, log<sup>10</sup> CFU g−<sup>1</sup> dw) of 1.0 in 100% biosolids or inoculated soil and the lowest C<sup>20</sup> of 0.098 in 75/25 soil/biosolids ratio, respectively, in comparison to an average initial value of ∼6.4 log<sup>10</sup> CFU g−<sup>1</sup> dw. The *E. coli* survival across scenario 2, 3, and 4 showed that the C<sup>20</sup> value of biosolids (0.57 log<sup>10</sup> CFU g−<sup>1</sup> dw) and dairy slurry (0.74 log<sup>10</sup> CFU ml−<sup>1</sup> ) was 2.9–3.7 times smaller than that of cattle dung (2.12 log<sup>10</sup> CFU g−<sup>1</sup> dw). The C<sup>20</sup> values of biosolids and dairy slurry associated with legal and excess application rates ranged from 1.14 to 1.71 log<sup>10</sup> CFU ha−<sup>1</sup> , which is a significant reduction from the initial concentration range (12.99 to 14.83 log<sup>10</sup> CFU ha−<sup>1</sup> ). The *E. coli* survival in un-amended soil was linear with a very low decay rate resulting in a higher C<sup>20</sup> value than that of biosolids or dairy slurry. The risk assessment and uncertainly analysis showed that the residual concentrations in biosolids/dairy slurry applied soil after 20 days would be 45–57% lower than that of the background soil *E. coli* concentration. This means the current practice of grazing exclusion times is safe to reduce the risk of *E. coli* transmission into the soil environment.

Keywords: biosolids, dairy slurry, *E. coli*, decay, risk assessment, agriculture, soil

# INTRODUCTION

Globally, provision of a circular economy safeguards against volatile fertilizer prices, global diminishing resources (e.g. synthetic fertilizers, fossil fuel) and an increased demand for food (Heffer and Prud'homme, 2013). In the European Union (EU), the Landfill Directive (EC, 1999) promoted a circular economy by targeting an 85% reduction in the disposal of sewage sludge to landfill by 2014 from 1995 levels. Such an ambitious target was aided by the Sewage Sludge Directive (EEC, 1986), which directed a major proportion of sewage sludge to land (Lucid et al., 2013; Fijalkowski et al., 2017). The standard management practice for dairy slurry and manure on dairy farms is land application without any necessary pathogen treatment. In contrast land application of treated sewage sludge (henceforth called "biosolids"), which typically involves pretreatment has variable land application uptake across EU member states ranging from 0% (e.g., Belgium—Brussels and Flanders, Switzerland, and Romania) to >50% (e.g., Norway, Ireland, Spain, UK, France) with an average of 39% being reused in agriculture across the EU (Lucid et al., 2013; Healy et al., 2016a; Fijalkowski et al., 2017). By comparison, about 60% of biosolids in the USA, Canada, and Australia are recycled to agriculture (Tozzoli et al., 2016). The EU figures from 2010 suggest an 81.8% increase in sewage sludge production when compared to 5.5 million tons of dry solids (tds) produced in 1992, and this figure is expected to increase up to 13 million tds by 2020 (EC, 2010; Healy et al., 2017). The positives of land application include a source of nitrogen (N), phosphorus (P), potassium (K), other plant nutrients, and an increase in soil organic matter (Sharma et al., 2017). The negatives can be heavy metal bioaccumulation, runoff losses of nutrient, metal, enteric pathogens and emerging contaminants, and bio-transfer of persistent organic pollutants to the food chain (Healy et al., 2016a,b, 2017; Clarke et al., 2017, 2018; Fijalkowski et al., 2017).

In Ireland, 98% of the biosolids (out of 53,543 tds year−<sup>1</sup> produced) go to land (Irish Water, 2015; Clarke et al., 2018). The application rate is typically determined by pH, metal and nutrient content of the soil, and the nutrient and metal content of the biosolids as per limits recommended in the "Codes of Good Practice for the Use of Biosolids in Agriculture" (Fehily Timoney Company, 1999). The guideline relates to postapplication of biosolids to grassland and restricts the livestock grazing period stating that "cattle should not be turned out onto pasture that has been fertilized with biosolids until 3–6 weeks after the date of application" (Fehily Timoney Company, 1999). There is growing concern on the survival of enteric Escherichia coli (E. coli) in biosolids and associated risk of transferring this fecal indicator organism (FIO) pollutant into the soil environment and subsequently, contamination of crops and nearby water sources, leading to the potential of spread of gastrointestinal disease (Greene et al., 2008; Ellis et al., 2018). The Sewage Sludge Directive 86/287/EC does not specify limits for E. coli counts as a fecal contamination indicator in biosolids, but specifies general land use, harvesting, and grazing limits to provide protection against the risk of infection (Sobrados-Bernardos and Smith, 2012). The revised version of the Sewage Sludge Directive (Working Document 3rd Draft), recommends that the E. coli in the biosolids needs to be less than 1 × 10<sup>3</sup> CFU g−<sup>1</sup> dry weight (dw) and that the sludge must have limited spores of Clostridium perfringens (<3 × 10<sup>3</sup> g <sup>−</sup><sup>1</sup> dw) with an absence of Salmonella. spp in 50 g (wet weight, ww) (EEC, 2000; Healy et al., 2017). This revised working document further states that E. coli concentration in biosolids needs to achieve at least a 2 Log<sup>10</sup> reduction after conventional treatment. Therefore, it is critical to accurately determine the FIO pollution (herein E. coli) risk associated with land application of biosolids to fully understand the potential for environmental loss and consequently, human/animal transmission.

Survival patterns of biosolids-derived E. coli in the environment are complex, and a lack of a standardized approach to E. coli measurement makes quantifying their impact difficult. For example, Avery et al. (2005) spiked treated and untreated biosolids samples with a known concentration of E. coli O157 to quantify the time taken to achieve a decimal reduction. The pathogen response was variable and ranged from 3 to 22 days, depending on sludge properties. Lang and Smith (2007)investigated indigenous E. coli survival in dewatered, mesophilic anaerobically digested (DMAD) biosolids, and in different soil types post DMAD biosolids application. Again, decimal reduction times proved variable, ranging from 100 days when applied to air-dried sandy loam, to 200 days in air-dried silty clay. When field moist soils were used this time decreased to 20 days, demonstrating the importance of water content in regulating survival behavior. Therefore, in order to quantify E. coli risk in a relevant, site-specific manner, it is necessary to incorporate both soil and biosolids characteristics in risk assessment modeling. This has been done previously by conducting soil, biosolids, and dairy slurry incubation studies where E. coli are often spiked to generate a survival response (Vinten et al., 2004; Lang and Smith, 2007; Moynihan et al., 2013). Pathogen decay rate (or death) is then calculated based on decimal reduction times, or a first-order exponential decay model previously described by Vinten et al. (2004), and has been shown to be highly contingent on soil type and biosolids or slurry combinations. Currently the Safe Sludge Matrix provides a legal framework for grazing animals and harvesting crops following land application of biosolids, and stipulates that a time interval of about 20 days (grazing exclusion period, and harvesting interval for grass and forage) and 10 months (harvesting interval for fruit, salads, vegetables, and horticulture) should be enforced to ensure safe practice, respectively (ADAS, 2001). However, further work is required to determine if these regulations are overly stringent, particularly in light of the comparatively larger pathogen concentrations reported for dairy slurries than biosolids. For example, E. coli concentrations ranged from 3 × 10<sup>2</sup> to 6 × 10<sup>4</sup> CFU g−<sup>1</sup> in biosolids (Payment et al., 2001) compared to 7.5 × 10<sup>4</sup> to 2.6 × 10<sup>8</sup> CFU g−<sup>1</sup> in fresh and stored dairy slurry, respectively (Hutchison et al., 2004). Recently, Healy et al. (2017) study pointed out that livestock exclusion times of more than 3 weeks after biosolids application (considering compliant application rates) may be overly strict with respect to the current exclusion criteria recommendation (e.g. 3–6 weeks in Ireland). Therefore, environmental losses of E. coli associated with biosolids application may not be as extensive as previously thought and further comparisons on pathogen risk should form the basis of future research.

The main objective of this study was to assess the risk of E. coli survival as an indication of the risk associated with land spreading biosolids to agricultural soils within the context of legislated grazing exclusion times. Herein, two exclusion time points at 20 and 40 days were considered in line with the exclusion criteria practice in the UK (i.e. Safe Sludge Matrix ∼20 days) and Ireland (i.e. Code of Good Practice for the Use of Biosolids in Agriculture ∼20–40 days). In particular, the objectives of the present study were to: (1) gather empirical data on E. coli concentration, and pathogen decay rate (k) for dairy slurry, cattle dung, and biosolids, and (2) conduct risk assessment modeling and uncertainty analysis of survival of E. coli at different time periods from application of dairy slurry, cattle dung, and biosolids to grassland up to the cattle exclusion time point (i.e. 20 and 40 days).

### MATERIALS AND METHODS

### Empirical Data on *E. coli* Concentration and Decay Rate

The die-off patterns of E. coli in dairy slurry, cattle dung, and biosolids were analyzed from the published peer-reviewed literature to develop an overview of the E. coli concentration and decay rate (k) as presented in **Table 1**. In this case, 12 relevant papers were utilized to generate the data under five categories— (1) un-amended soil, (2) E. coli spiked soil, (3) biosolids, (4) dairy slurry, and (5) cattle dung. These studies were deemed relevant based on the availability or possibility of derivation of initial E. coli concentration and k value. The heterogeneous nature of the above five categorized materials and their diverse treatment conditions like moisture level, seasonality, application dose, and condition were also considered to cover the wide range of data set. Data were obtained from tables or log-linear regression equations where available (Himathongkham et al., 1999; Oliver et al., 2006; Lang and Smith, 2007; Martinez et al., 2013; Hodgson et al., 2016; Roberts et al., 2016); otherwise, data were extracted from digitized figures to derive log-linear regression equation by plotting Log<sup>10</sup> CFU g−<sup>1</sup> dw vs. Time (days) (Avery et al., 2004, 2005; Oliver et al., 2010; Schwarz et al., 2014; Biswas et al., 2018; Ellis et al., 2018). The die-off pattern of pathogens can be described by the first-order kinetics Equation (1), which upon integration gives the linear Equation (2) (Mubiru et al., 2000; Martinez et al., 2013). This natural logarithm based linear Equation (2) was converted to the base 10 logarithm (i.e., Log10) based Equation (3) and compared with a straight line equation (y=mx+c) to get the slope (m) and subsequently, the die-off or decay rate (k) values were obtained using Equation (4) (**Table 1**). The linear Equation (2) can be transformed to an exponential model (Equation 5) to assess the risk of E. coli content in soil after application of different organic residues like dairy slurry, sewage sludge, and cattle dung (Vinten et al., 2004).

$$\frac{d(\text{C})}{dt} = -k\text{C} \tag{1}$$

where C is the E. coli concentration per unit of mass or volume and k is the die-off or decay rate.

$$
\ln \mathbf{C}\_t = \ln \mathbf{C}\_0 - k\mathbf{t} \tag{2}
$$

Here, C<sup>t</sup> is concentration of E. coli at time t in the soil, C<sup>o</sup> is the concentration of E. coli at time zero in the soil, t is fixed time period (e.g. grazing period) (days), k is the die-off function of the E. coli (day−<sup>1</sup> ).

$$\text{Log}\_{10}\text{C}\_{\text{t}} = \text{Log}\_{10}\text{C}\_{\text{o}} - \frac{\text{kt}}{2.303} \tag{3}$$

$$\text{Slope}, \text{m} = -\frac{k}{2.303} \tag{4}$$

$$\mathbf{C}\_{\text{l}} = \mathbf{C}\_{\text{0}} e^{-\mathbf{k}\mathbf{t}} \tag{5}$$

### Risk Assessment and Uncertainty Analysis

In this study, the exponential Equation (5) was used to quantify the concentrations of E. coli in the soil after any time period to be known following land application of the aforementioned organic materials. Traditionally, the burden of E. coli accumulation in soil from livestock feces or land spreading of dairy slurry is calculated by assuming the exponential decay pattern of E. coli survival over time (Oliver et al., 2009, 2010). A risk assessment of the survival of E. coli up to 40 days after application was examined across six scenarios (**Table 2**)—(1) soil and biosolids mixture, (2) biosolids amended soil, (3) dairy slurry application, (4) cattle dung on pasture, (5) comparison of scenario 2, 3, and 4, and (6) maximum legal vs. excess rate of application for scenario 2 and 3. The risk model input parameters i.e., initial E. coli concentration (C0) and decay rate (k) were used from the **Table 1** as presented in **Table 2**.

In scenario 1, the values of C<sup>0</sup> (i.e., concentration of E. coli at day 0) and k were taken as the average for soil to sludge mixture matrix of un-amended soil (Lang and Smith, 2007), 100% soil (E. coli spiked) (Oliver et al., 2006; Ellis et al., 2018), 75% soil to 25% sludge (Ellis et al., 2018), 50% soil to 50% sludge (Ellis et al., 2018), 25% soil to 75% sludge (Ellis et al., 2018), and 100% sludge (Avery et al., 2005; Ellis et al., 2018). In scenario 2, C<sup>0</sup> was considered as the average of biosolids associated E. coli from five different studies and the k value was considered individually from the respective study and also, as an average value of those studies (**Table 1**, **2**). Similar to scenario 2, C<sup>0</sup> and k values (**Table 2**) were assigned to scenario 3 and 4 considering five different studies (as mentioned in **Table 1**) for dairy slurry and cattle dung, respectively. In scenario 5, the average value for C<sup>0</sup> and k was assigned as in scenarios 2, 3, with 4 used to provide a comparison among biosolid, dairy slurry and cattle dung treatments. Scenario 6 was considered to assess the risk of E. coli survival under estimated legal and excess application rate of biosolids and dairy slurry in grassland.

TABLE 1 | Concentration (*E. coli*, Log<sup>10</sup> CFU g−<sup>1</sup> dw) and decay rate (*k*, days−<sup>1</sup> ) for a variety of biosolids, dairy slurry, and cattle dung.


*DMAD, dewatered mesophilic anaerobically digested; ADD, Anaerobically digested dewatered; dw, dry weight.*

*D value indicates the time required for 90% pathogen reduction; [C*0*], initial E. coli concentration;* \**values presented as wet weight basis (Log*<sup>10</sup> *CFU ml*−<sup>1</sup> *) assuming 1 ton* = *1 m*<sup>3</sup> *slurry.*

The estimation of a legal application rate for biosolids and dairy slurry was based on the required P application rate of 40 kg ha−<sup>1</sup> for pasture establishment at a low Morgan's P Index soil (e.g. P Index 2 equivalent to Morgan's P of 3.1–5.0 mg l−<sup>1</sup> ) (Peyton et al., 2016; Teagasc Greenbook, 2016). In general, P is the limiting factor for estimating legal application rate of waste derived organic fertilizers such as biosolids and dairy slurry (Lucid et al., 2013). The legal maximum application rate of biosolids was estimated to be in the range of 3.0 to 5.2 ton ha−<sup>1</sup> by Lucid et al. (2013) based on the P Index of the soil, the legal limits of N, P, and metal concentration of the soil, the dry matter content, and the nutrient and metal concentration of the biosolid amendment. The estimated legal application rate of biosolids and dairy slurry is presented in **Table 3** and these TABLE 2 | Scenario and parameters used for risk assessment modeling and Monte Carlo uncertainty analysis.


\**Values presented as dw basis except for dairy slurry (wet weight basis assuming 1 ton* = *1 m*<sup>3</sup> *slurry).*

TABLE 3 | Biosolids and dairy slurry landspreading rate for risk assessment model and Monte Carlo uncertainty simulation.


*<sup>a</sup>Values presented as dw basis; <sup>b</sup>Values presented as wet weight basis assuming 1 ton* = *1 m*<sup>3</sup> *slurry; <sup>c</sup> (Teagasc Greenbook, 2016); <sup>d</sup>Table 1; <sup>e</sup>5 times higher than the legal application rate.*

values are comparable with those of commonly used application rate in previous studies (e.g. Brennan et al., 2012; Lucid et al., 2013).

In orderto reflect the variability of the model input parameters for a particular soil type, organic material, E. coli concentration (C0) and die**-**off rate (k) across time, we applied a Monte Carlo simulation (run of 1,000 times per scenario) to compute the probability density distributions for the final concentration in the soil. For the analysis we assumed a uniform distribution of C0, k, and time as in **Table 2**.

### RESULTS AND DISCUSSION

### *E. coli* [C0] and *k*

The empirical data on initial concentration (C0) and k values of E. coli are presented in **Table 1**. Results show that these parameters vary widely across each type of material ranging from 0.79– 3.13 (unamended soil), 5.93–6.92 (inoculated soil), 4.25–7.82 (biosolids), 5.86–7.27 (dairy slurry) and 5.36–7.13 (cattle dung) for C<sup>0</sup> (log<sup>10</sup> CFU g−<sup>1</sup> ), and 0.007–0.023 (unamended soil), 0.069–0.131 (inoculated soil), 0.012–0.290 (biosolids), 0.023– 0.230 (dairy slurry), and 0.042–0.071 (cattle dung) for k (day−<sup>1</sup> ) values, respectively. The treatment nature and condition of each type of material is largely heterogeneous (e.g. soil type, soil to biosolids ratio, sludge type, slurry moisture, slurry age, dung condition) across and within the incorporated reference studies, which can reasonably explain such variability for C<sup>0</sup> and k values. However, it was observed that the mean value of both C<sup>0</sup> (log<sup>10</sup> CFU g−<sup>1</sup> ) and k (day−<sup>1</sup> ) when compared among inoculated soil (C<sup>0</sup> = 6.5 ± 0.44, k = 0.092 ± 0.024), biosolids (C<sup>0</sup> = 6.0 ± 0.99, k = 0.115 ± 0.079), dairy slurry (C<sup>0</sup> = 6.2 ± 0.37, k = 0.095 ± 0.064), and cattle dung (C<sup>0</sup> = 6.1 ± 0.52, k = 0.055 ± 0.010) is not statistically different at the 95% significance level as determined by one-way ANOVA [F(3,32) = 0.665, p = 0.579 for C<sup>0</sup> and F(3,32) = 1.477, p = 0.239 for k). This means the empirical range of the C<sup>0</sup> and k values of E. coli for three major organic residue based fertilizers (biosolids, dairy slurry, and cattle dung) as presented in **Table 1** are suitable for risk assessment modeling. The wide data set of C<sup>0</sup> and k values will provide a variability range for the risk assessment and a prediction of uncertainty through the probability distribution.

# *E. coli* Survival Pattern Across Six Scenarios

In scenario 1, the different combinations of soil and biosolids in the incubation experiment produced different k values and therefore different distributions of E. coli concentrations over time in soil i.e. potential losses in runoff. The E. coli survival pattern in 100% inoculated soil and 100% biosolids is similar, and E.coli concentration reduction of ∼5.69 log<sup>10</sup> CFU g−<sup>1</sup> dw was observed leading to the 20 day concentration (C20) of ∼1.0 log<sup>10</sup> CFU g−<sup>1</sup> dw (see **Figure 1A**). The survival is the lowest in the soil to biosolids mixture ratio of 75/25 and after 20 days the concentration was 0.098 compared to 0.282 and 0.509 log<sup>10</sup> CFU g−<sup>1</sup> dw in 50/50 and 25/75 equivalents, respectively. In comparison to the inoculated soil and biosolids or soil/ biosolids mixture, the survival pattern in un-amended soil was linear with a very low decay rate (0.014 day−<sup>1</sup> ) resulting in the highest C<sup>20</sup> concentration of 1.34 log<sup>10</sup> CFU g−<sup>1</sup> dw. After 40 days, the E. coli concentrations (log<sup>10</sup> CFU g−<sup>1</sup> dw) were: 0.166, 0.0015, 0.0126, 0.0409, 0.1436 for the 100% soil (inoculated), 75/25 soil/biosolids, 50/50 soil/biosolids, 25/75 soil/biosolids ratios,

scenario 4, (E) scenario 5, and (F) scenario 6.

and 100% biosolids, respectively, compared to the C<sup>40</sup> value of 1.02 for un-amended soil. These results likely reflect that E. coli populations in un-amended soil are more adaptive than the imported E. coli and can survive as natural soil microflora under favorable soil conditions (e.g. soil texture and structure, pH, moisture, temperature, UV radiation, and nutrient and oxygen availability). For example, E. coli was observed to survive in control soils for more than 9 years, particularly, as becoming naturalized in the low-temperature environments of temperate maritime soils (Brennan et al., 2010a,b).

In scenario 2, the E. coli survival trend in biosolids amended soil was assessed based on the empirical data (**Tables 1**, **2**) from five reference studies as shown in **Figure 1B**. The E. coli concentration (log<sup>10</sup> CFU g−<sup>1</sup> dw) after 20 days was ≤0.57 from an initial value of 6.48 for the average biosolids and three study references (Avery et al., 2005; Roberts et al., 2016; Ellis et al., 2018), except for Schwarz et al. (2014) (C<sup>20</sup> = 1.14) and Lang and Smith (2007)(C<sup>20</sup> = 1.72). The C<sup>40</sup> value ranged from 0.006 to 0.46 log<sup>10</sup> CFU g−<sup>1</sup> dw for all five reference studies.

In scenario 3, E. colisurvival pattern in dairy slurry application associated soil was assessed based on the empirical data (**Tables 1**, **2**) from five reference studies as shown in **Figure 1C**. In this case, the C20, log<sup>10</sup> CFU ml−<sup>1</sup> concentrations were 0.12, 1.47, 1.12, 1.23, 0.90, and 0.74 compared to the initial value of 6.43 from Himathongkham et al. (1999), Avery et al. (2005), Oliver et al.

(2006), Hodgson et al. (2016), and Biswas et al. (2018). The C<sup>40</sup> concentration ranged from 0.002 to 0.34 log<sup>10</sup> CFU ml−<sup>1</sup> for all five reference studies.

In scenario 4, cattle dung associated E. coli survival pattern was assessed based on the input data from five reference studies as shown in **Figure 1D**. In this scenario, the C<sup>20</sup> and C<sup>40</sup> concentrations ranged from 1.82 to 2.67 and 0.54 to 1.16 log<sup>10</sup> CFU g−<sup>1</sup> dw, respectively, compared to the initial value of 6.16 log<sup>10</sup> CFU g−<sup>1</sup> dw for all five reference studies. A comparison of E. coli survival patterns in biosolids, dairy slurry and cattle dung can be seen from scenario 5 (**Figure 1E**). In general, the C<sup>20</sup> value of biosolids (0.57 log<sup>10</sup> CFU g−<sup>1</sup> dw) and dairy slurry (0.74 log<sup>10</sup> CFU ml−<sup>1</sup> ) was 2.9–3.7 times smaller than that of cattle dung (2.12 log<sup>10</sup> CFU g−<sup>1</sup> dw). The C<sup>40</sup> value was < 1.0 log<sup>10</sup> CFU per unit mass or volume for any of this material when compared to the same in un-amended soil. However, the results of actual survival patterns in cattle dung studies under natural field conditions differ from studies that use first-order die-off approximations (Van Kessel et al., 2007; Soupir et al., 2008; Muirhead, 2009; Oliver et al., 2010). The reason for such discrepancies could be the potential of E. coli "re-growth" which were not considered when using first-order decay model. Instead a constant decay rate (k) value was used. In reality, E. coli growth and re-growth phases in deposited dung-pats can be highly interactive with environmental conditions such as: temperature, UV radiation, soil type, and rainfall events (Oliver et al., 2010). For example, the E. coli growth magnitude was observed to vary from 0.5 to 1.5 log<sup>10</sup> CFU g−<sup>1</sup> dw due to different environmental factors (Sinton et al., 2007; Van Kessel et al., 2007; Oliver et al., 2010). This means the estimation of E. coli risk from cattle dung on pasture by single k value based first-order decay model can potentially underestimate the growth potential and provides a conservative indication of fecal indicator organism accumulation over time. The modification of first-order decay equation by incorporating growth factor can improve the model predictability under field conditions. Therefore, the results of the present study represent scenarios without regrowth considerations.

In scenario 6, biosolids and dairy slurry were considered as the most commonly applied organic fertilizer for agricultural landspreading with two estimated application rates (ton ha−<sup>1</sup> ): maximum legal and excess as shown in **Table 3**. The E. coli survival pattern in this case is presented in **Figure 1F**. The C<sup>20</sup> values of biosolids associated with legal and excess application rates are 1.14 and 1.21 log<sup>10</sup> CFU ha−<sup>1</sup> , respectively, in comparison to 1.63 and 1.71 log<sup>10</sup> CFU ha−<sup>1</sup> , respectively, for dairy slurry associated application. The C<sup>40</sup> values in this case were less than ≤0.2 log<sup>10</sup> CFU ha−<sup>1</sup> when compared to C<sup>0</sup> (log<sup>10</sup> CFU ha−<sup>1</sup> ) values of biosolids (12.99–13.69) and dairy slurry (14.13–14.83), respectively (**Figure 1F**).

### Uncertainty and Probability Distributions of *E. coli* Concentration

The uncertainty analysis (**Figure 2**) indicated that soil E. coli concentrations would be at least 3.5 log<sup>10</sup> CFU g−<sup>1</sup> or ml−<sup>1</sup> lower than the C<sup>0</sup> range of 6.2 to 6.5 log<sup>10</sup> CFU g−<sup>1</sup> or ml−<sup>1</sup> in about 75.5% of the time (i.e. Ct≤3 log<sup>10</sup> CFUg−<sup>1</sup> or ml−<sup>1</sup> ) after application of either biosolids or dairy slurry or cattle dung to land (**Figure 2B**). Considering the variability of C<sup>0</sup> and k values due to the material type and study references (scenario 5, **Table 2**), the predicted E. coli concentration at any time can be estimated from y = 6.1262e−0.079x [similar to exponential Equation (5)] as developed from Monte Carlo simulation of 1,000 trials (**Figure 2A**). Accordingly, the C<sup>20</sup> value can be expected as 1.262 log<sup>10</sup> CFU g−<sup>1</sup> or ml−<sup>1</sup> , which is comparatively lower than that of un-amended soil in this study, pointing toward the remaining E. coli after 20 days of application as being soil indigenous E. coli. The Monte Carlo analysis of biosolids (for scenario 2) provides the predictive exponential equation y = 6.3097e−0.112x with a probability distribution of Ct≤3 log<sup>10</sup> CFU g−<sup>1</sup> dw for 82% of the time (Figure S1). Similarly, dairy slurry (scenario 3) and cattle dung (scenario 4) based analysis provide regressions of y = 6.459e−0.123x and y = 6.1179e−0.049x , respectively, with a probability distribution of Ct≤3 log<sup>10</sup> CFU g <sup>−</sup><sup>1</sup> of 83 and 61.5% of the time, respectively (Figures S2, S3). The predicted C<sup>20</sup> (log<sup>10</sup> CFU g−<sup>1</sup> or ml−<sup>1</sup> ) values of biosolids and dairy slurry associated E.coli was 0.672 and 0.552, respectively, while the equivalent for cattle dung was 2.296, indicating a higher risk associated with longer survival of E. coli in cattle dung on pasture. For the estimated legal and excess application rate of biosolids or dairy slurry (scenario 6, Figures S4, S5), the predictive exponential equations developed were y = 13.497e−0.113x and y = 14.169e−0.113x, respectively, with a probability distribution of E. coli concentration remaining ≤3 log<sup>10</sup> CFU ha−<sup>1</sup> , 63% of the time. While the C<sup>20</sup> (log<sup>10</sup> CFU ha−<sup>1</sup> ) concentration for scenario 6 ranged from 1.408 to 1.478, the C<sup>40</sup> value was almost negligible (0.147–0.154 log<sup>10</sup> CFU ha−<sup>1</sup> ).

The outcomes of the uncertainty analyses depended on the distribution of the model variables and the associated parameters of these distributions. In other words, if different distribution parameters had been assumed, different outcomes may have been expected. For the scenarios in this study (**Table 2**) the distributions of the data are based on a range (maxima and minima) of empirical data collected from the literature (**Table 1**). In absence of detailed information on the probability density distributions of these variables, we employed the uniform distribution as the most parsimonious distribution.

### CONCLUSIONS

An empirical database of dairy slurry, cattle dung and biosolids associated E. coli concentration and decay rate (k) was developed to assess the risk of E. coli survival up to a legislated grazing

### REFERENCES


exclusion period. The use of a traditional exponential E. coli decay model and Monte Carlo uncertainty analysis showed that soil E. coli concentrations at 20 days would be at least 3.5 log<sup>10</sup> CFU g−<sup>1</sup> lower than the initial range of 6.2 to 6.5 log<sup>10</sup> CFU g <sup>−</sup><sup>1</sup> or ml−<sup>1</sup> in 75.5% of simulations after application of either biosolids, dairy slurry or cattle dung to land. The predicted C<sup>20</sup> value was 1.262 log<sup>10</sup> CFU g−<sup>1</sup> or ml−<sup>1</sup> , which is lower than that of un-amended soil in this study, indicating that the majority of E. coli 20 days after application would be mainly indigenous soil E. coli. For the estimated legal and excess application rates of biosolids or dairy slurry, the probability distribution of E. coli concentration remained at ≤3 log<sup>10</sup> CFU ha−<sup>1</sup> 63% of the time. The predicted C<sup>20</sup> concentration for the estimated legal to excess application rates was 1.408–1.478 log<sup>10</sup> CFU ha−<sup>1</sup> , while the C<sup>40</sup> equivalent ranged from 0.147 to 0.154 log<sup>10</sup> CFU ha−<sup>1</sup> . This indicates 40 days as safer than 20 days for a grazing exclusion period. However, considering the decay period of E. coli in unamended soil, the 20 day exclusion period seems safe to reduce the risk of E. coli transmission into the soil environment and subsequently, negating the risk of contamination of crops and nearby water sources. The finding of this study supports the current practice of grazing exclusion times in the UK and Ireland.

### AUTHOR CONTRIBUTIONS

SMA, OF, and KR contributed to the study conception and design of the study. SMA developed the empirical database, performed risk assessment modeling and uncertainly analysis with the assistance of OF. SMA wrote the manuscript, OF provided feedback and revised where necessary. SE, ST, and BG reviewed the manuscript and provided comments for improvement. Finally, all authors approved the final version of the manuscript.

### ACKNOWLEDGMENTS

This publication has emanated from research funded by the EU FP7 Environment theme–Grant no. 265269 Marketable sludge derivatives from a highly integrated wastewater treatment plant (END-O-SLUDG).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fsufs. 2018.00034/full#supplementary-material


maritime soils. Appl. Environ. Microbiol. 76, 1449–1455. doi: 10.1128/AEM. 02335-09


following dairy slurry application to land by surface broadcasting and shallow injection. J. Environ. Manage 183, 325–332. doi: 10.1016/j.jenvman.2016.08.047


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer CH and handling Editor declared their shared affiliation.

Copyright © 2018 Ashekuzzaman, Richards, Ellis, Tyrrel, O'Leary, Griffiths, Ritz and Fenton. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Toward Assessing Farm-Based Anaerobic Digestate Public Health Risks: Comparative Investigation With Slurry, Effect of Pasteurization Treatments, and Use of Miniature Bioreactors as Proxies for Pathogen Spiking Trials

### *Edited by:*

*Tom Misselbrook, Rothamsted Research (BBSRC), United Kingdom*

### *Reviewed by:*

*Jaak Truu, University of Tartu, Estonia Clinton D. Church, Pasture Systems and Watershed Management Research Unit (USDA-ARS), United States*

### *\*Correspondence:*

*Florence Abram florence.abram@nuigalway.ie*

### *Specialty section:*

*This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems*

> *Received: 08 May 2018 Accepted: 29 June 2018 Published: 20 July 2018*

### *Citation:*

*Nolan S, Waters NR, Brennan F, Auer A, Fenton O, Richards K, Bolton DJ, Pritchard L, O'Flaherty V and Abram F (2018) Toward Assessing Farm-Based Anaerobic Digestate Public Health Risks: Comparative Investigation With Slurry, Effect of Pasteurization Treatments, and Use of Miniature Bioreactors as Proxies for Pathogen Spiking Trials. Front. Sustain. Food Syst. 2:41. doi: 10.3389/fsufs.2018.00041* Stephen Nolan1,2, Nicholas R. Waters 1,3, Fiona Brennan<sup>2</sup> , Agathe Auer <sup>4</sup> , Owen Fenton<sup>2</sup> , Karl Richards <sup>2</sup> , Declan J. Bolton<sup>5</sup> , Leighton Pritchard<sup>3</sup> , Vincent O'Flaherty <sup>6</sup> and Florence Abram<sup>1</sup> \*

*<sup>1</sup> Functional Environmental Microbiology, School of Natural Sciences, National University of Ireland Galway, Galway, Ireland, <sup>2</sup> Teagasc, Environmental Research Centre, Johnstown Castle, Wexford, Ireland, <sup>3</sup> Information and Computational Sciences, James Hutton Institute, Dundee, Scotland, <sup>4</sup> University College Dublin, Veterinary Science Centre, Dublin, Ireland, <sup>5</sup> Teagasc, Ashtown Food Research Centre, Ashtown, Ireland, <sup>6</sup> Microbial Ecology Laboratory, School of Natural Sciences, National University of Ireland Galway, Galway, Ireland*

Manure and slurry may contain a range of bacterial, viral, and parasitic pathogens and land application of these organic fertilizers typically occurs without prior treatment. *In-situ* treatment through farm-based anaerobic digestion (AD) of such organic fertilizers co-digested with food-production wastes is multi-beneficial due to energy recovery, increased farm incomes and noxious gas reduction. Before risk assessment can be carried out at field scale an investigation of the fate of relevant target pathogens during the actual AD process must be undertaken, requiring the development of practical test systems for evaluation of pathogen survival. The present study examines miniature (50 mL) and laboratory (10 L) scale AD systems. Treatments included slurry co-digested with fats, oils, and grease (FOG) under typical operating and pasteurization conditions used in farm-based AD, in batch-fed miniature and laboratory mesophilic (37◦C) continuously stirred tank reactors. Biogas production, pH, chemical oxygen demand, volatile solids, and ammonia concentration were measured throughout the trial, as were fecal indicator bacteria (FIB) i.e., total coliforms, *Escherichia coli,* and *Enterococcus* species. The miniature and laboratory bioreactors performed similarly in terms of physicochemical parameters and FIB die-off. In the absence of pasteurization, after 28 days, enterococci numbers were below the <1,000 cfu g−<sup>1</sup> threshold required for land application, while *E. coli* was no longer detectable in the digestate. For comparison, FIB survival in slurry was examined and after 60 days of storage, none of the FIB tested was <1,000 cfu g−<sup>1</sup> , suggesting that slurry would not be considered safe for land application if FIB thresholds required for AD digestate were to be applied. Taken together we demonstrate that (i) miniature-scale bioreactors are valid proxies of farm-based AD to carry out targeted pathogen survival studies and (ii) *in situ* AD treatment of slurry prior to land application reduces the level of FIB, independently of pasteurization, which in turn might be indicative of a decreased potential pathogen load to the environment and associated public health risks.

Keywords: anaerobic digestion, fats, oils and grease, fecal indicator bacteria survival, miniature bioreactors, slurry

## INTRODUCTION

Approximately 1.4 billion tons of manure are produced in Europe each year, 80% of which is in the form of slurry, predominantly from cattle (Crowe et al., 2000; Foged et al., 2011). Manure and slurry represents valuable organic fertilizers, but typically contain a broad range of bacterial, viral, and parasitic pathogens (Bicudo and Goyal, 2003; Alam and Zurek, 2006; Ferens and Hovde, 2011). Human and animal pathogens commonly isolated from manure include E. coli O157, Salmonella, Listeria, Campylobacter, Cryptosporidium, Ascaris, Mycobacterium avium subspecies paratuberculosis, and Giardia (Nicholson et al., 2004; Olson et al., 2004; Grewal et al., 2006). These pathogens can be transferred to the environment as bioaerosols during landspreading (Millner, 2009; Dungan, 2010), ingested directly from grass or vegetables (Baloda et al., 2001; Braden and Tauxe, 2013), or may be washed off into connected water bodies, posing a significant threat to human and animal health (Douwes et al., 2003; Gerba and Smith, 2005; Venglovsky et al., 2009). Furthermore, manure is a potent source of noxious and greenhouse gases (GHG), which are released to the atmosphere during storage in slatted tanks and subsequent landspreading (Chadwick et al., 2011). A number of methods for limiting the impact of manure storage and landspreading, both in terms of GHG capture or mitigation and pathogen reduction have been examined, including aeration, and acidification during storage, animal diet manipulation, or alternative landspreading techniques (Nicholson et al., 2004; Franz et al., 2005; Webb et al., 2010). Typically these proposed solutions, however, consider either pathogens or GHG in isolation. Composting, for example, is suggested as an effective solution to reduce pathogens in manure (Ros et al., 2006; Vinnerås, 2007; Mc Carthy et al., 2011; Millner et al., 2014), with scant reference to gaseous N or CH<sup>4</sup> loss to the environment (Rao et al., 2007). Conversely, methods for reduction of ammonia or other GHG losses from manure, such as acidification, rarely consider the fate of pathogens during such treatments (Kai et al., 2008; Petersen et al., 2012). In fact, in this context, the recommended direct incorporation of slurry into soil might lead to increased pathogen survival, as it inevitably reduces UV exposure (Avery et al., 2004; Hutchison et al., 2004).

Rather than tackling pathogen survival or GHG emissions from manure in isolation, a technological solution that addresses both would clearly be preferable. To that end, biogas production as a treatment for manure holds great promise (Monteny et al., 2006). In addition to the obvious benefits of energy recovery, noxious gas and GHG mitigation, farm-based AD could potentially reduce pathogen loads in the environment and the associated public health risks (Olsen and Larsen, 1987; Kearney et al., 1993; Sahlström, 2003; Jiang et al., 2018). Pathogen survival may be significantly impacted, positively or negatively, by a variety of factors. These include: pH, ammonia production, microbial competition, initial pathogen load, operating conditions of farm-based AD plants and addition of co-digestion substrates such as food production waste with varying pathogen risks (Smith et al., 2005; Orzi et al., 2015). Indeed, the AD of slurry alone is hindered by an imbalanced C:N ratio resulting in low potential methane yields of 25–30 m3 ton−<sup>1</sup> (Weiland, 2010). To overcome this limitation, codigestion of slurry with locally sourced organic waste is typically implemented. This in turn helps to balance the C:N ratio and thus improves the relatively low methane yield of slurry alone, whilst taking advantage of its inherent buffering capacity, microbial populations, nutrients, and moisture content (Hamelin et al., 2014; Moset et al., 2017; Neshat et al., 2017).

Congealed fats, oils, and grease (FOG) are a significant problematic food production waste internationally, causing environmental and human health issues when allowed to form "fatbergs" in municipal sewage systems (Wallace et al., 2017). Grease-traps required for licensing in the food-processing industry as well as those in restaurants mitigate the problem, but create large quantities of organic waste, which requires further treatment. The typical biogas yield of FOG (4–8 m<sup>3</sup> kg VS −1 ) dwarfs that of slurry alone (0.148 m<sup>3</sup> kg VS <sup>−</sup><sup>1</sup> ), making co-digestion of FOG with slurry in farm-based AD plants a sustainable treatment option, cheaply increasing methane output (Møller et al., 2004; Weiland, 2010; Long et al., 2012). In Ireland, successful implementation of grease-trap legislation provides a steady supply of organic waste in the form of FOG, which is used as a feedstock in the majority of Irish farm-based AD plants. The co-digestion of slurry with organic waste, however, typically requires some pasteurization treatment to be carried out as stipulated by the legislation. In that context, two pasteurization processes are available in Ireland, set out by (i) the European Union Commission (Directive No. 142/2011) as 60 continuous minutes at 70◦C and (ii) the Irish Department of Agriculture, Food and the Marine as a total of 96 h at 60◦C (DAFM, 2014). Pasteurization can be applied either pre- or post-AD processing with the corresponding digestate quality being assessed using fecal indicator bacteria, typically E. coli and/or enterococci. According to Regulation (EC) No.1069/2009 and Regulation (EU) No. 142/2011, for AD digestate to be deemed safe for landspreading FIB levels must <1,000 cfu g−<sup>1</sup> . As highlighted by Dennehy et al. (2018), further investigations into the effect of AD processing on pathogen loads must be carried out to

determine the need for pasteurization. In addition, in order to meaningfully and accurately carry out risk assessment of digestate landspreading, the determination of the fate of relevant target pathogens during AD processing is necessary. Although previous studies of farm-based AD have reported reductions in target pathogen numbers (Olsen and Larsen, 1987; Kearney et al., 1993; Sahlström, 2003; Dennehy et al., 2018; Jiang et al., 2018), investigations into pathogen survival are typically hampered by difficulties in cultivating sufficient pathogen quantities and the public health concerns associated with spiking large volume bioreactors. Some solutions have been deployed in an effort to overcome this, including the containment of pathogens using sentinel chambers or filters held in steel baskets and submerged into digesters (Gray and Hake, 2004; Wagner et al., 2008). While this may successfully contain the pathogens and thus reduce the associated public health risks, such an experimental set-up greatly limits the interactions of the target pathogens with the surrounding matrix. There is, therefore a crucial need to develop an alternative solution closely mimicking real-life scenarios whereby interactions between pathogens and the AD liquor are not hindered.

Thus the aims of this study were to: (i) propose and validate the use of miniature-scale (50 mL) bioreactors as proxies for 10 L bioreactors; (ii) determine FIB survival under typical operating and pasteurization conditions used in farm-based AD systems; and (iii) assess the suitability of AD as a means of reducing the environmental impact of slurry management.

# MATERIALS AND METHODS

### Feedstock Selection, Collection, and Storage

In order to determine feedstock composition and operating conditions, a characterization of current Irish AD facilities was carried out. All Irish farm-based AD plants currently operate at mesophilic temperatures and process slurry co-digested with food production waste, including FOG (Auer et al., 2016). By visiting these AD facilities and utilizing knowledge gained in Auer et al., the following operation conditions and feedstock composition were determined.

First, a cattle slurry:FOG ratio of 2:1 was used, with a view to replicating full-scale farm-based AD. The FOG was sourced from the Bioenergy and Organic Fertilizer Services (BEOFS) AD plant in Camphill, County Kilkenny, Ireland, collected in a 25 L drum, stored at 4◦C, and mixed thoroughly before use. Cattle slurry for feeding the bioreactors was collected from a dairy farm in County Galway, Ireland in October, 2016. The slatted housing storage tanks were agitated to homogenize the slurry before collection of the sample using a bucket attached to a pole, in accordance with Brennan et al. (2011) and Peyton et al. (2016). Slurry was stored in a 25 L sealed container at 4◦C for 2 days prior to use as feedstock, at which time it was mixed thoroughly. In order to establish levels of farm to farm variation, dairy cattle slurry was collected from two additional farms in County Galway during October 2016. For comparison between digestate and stored slurry, triplicate slurry samples for each farm were stored in a shed at ambient Irish environmental temperatures during October–December, to mimic on-farm storage.

### Inoculum Development

Digestate from the BEOFS full-scale mesophilic continuouslystirred tank reactors (CSTR) co-digesting FOG with slurry was used as the starting inoculum, as it was adapted to the chosen substrate. This inoculum was found, through biomethane potential assays (BMP, data not shown), to be sub-optimal for biogas production. Therefore, augmentation with a mixture of slurry and methanogenic anaerobic granular sludge was deemed necessary to bolster both hydrolysis and methanogenesis. A series of specific methanogenic assays (SMA) were carried out using non-gaseous (acetate, ethanol, propionate, butyrate) and gaseous substrates (H2/CO2) as described by Coates et al. (1996). Based on the SMA results, a 2:1:1 ratio of granular sludge:BEOFS:slurry was selected as the optimum inoculum mixture (**Figure S1**).

## Miniature- and Laboratory-Scale Bioreactors Operation

Three 10-L CSTRs (R1–R3) were operated at 37◦C in batch with a 28-day solid retention time. Prior to operation, the inoculum and starting liquor were adjusted to pH 7 by adding NaHCO3. The organic loading rate for each bioreactor was 30 g VS L−<sup>1</sup> in a 2:1 inoculum to feedstock ratio with a 7 L working volume. Submerged, motor-propelled axial stirrers with large scale paddles were centrally installed in the bioreactor ceilings, with an externally positioned motor, as is typical of agricultural biogas plants (Weiland, 2010). Miniature batch tests (33 mL in 50 mL glass bottles) using identical inoculum and feedstock ratios to the 10 L bioreactors were run simultaneously at 37◦C under shaking conditions in a New Brunswick Scientific Innova◦ 44 incubator and destructively sampled in triplicate at regular intervals (days 0, 7, 14, 21, 28), for comparison. Their contents, as well as samples collected from the 10 L bioreactors were analyzed as described below.

### Analytical Methods

Biogas volume from the 10 L and 50 mL bioreactors was determined using the water displacement method and 10 mL syringes attached with a stopcock, respectively. Methane content of the biogas was analyzed using a Varian gas chromatograph equipped with a flame ionization detector. The carrier gas was nitrogen and the flow rate was 25 mL min−<sup>1</sup> . Analysis of TS and VS was performed gravimetrically according to standard methods (APHA., 2005). Soluble chemical oxygen demand (sCOD) was determined by analyzing the supernatant of centrifuged samples. Total chemical oxygen demand (tCOD) and sCOD analyses were performed according to the Standing Committee of Analysts. (1985). NH<sup>3</sup> concentrations (mg L −1 ) were determined using the HACH AmVer High-Range Ammonia test, available from HACH.

### Pasteurization

In addition to the unpasteurized 50 mL bioreactors used for comparison with the 10 L CSTRs, four pasteurization conditions were examined at the miniature scale to determine the impact on bioreactor performance and FIB survival. At each time point, two pre-AD pasteurization conditions (P1: 60◦C for 96 h; P2: 70◦C for 1 h) were used on the food production waste, and two post-AD pasteurization conditions (P3: 60◦C for 96 h; P4: 70◦C for 1 h) were applied to the digestate. These assays were carried out in triplicate for each time point, totaling 75 miniature-scale assays. Water baths set to the appropriate temperatures were used for pasteurization, and temperature probes were employed to ensure the designated temperature was achieved.

### Fecal Indicator Bacteria Monitoring

In line with the EU Regulation, total coliforms, E. coli, and enterococci numbers were monitored throughout the trial. Most probable numbers (MPN) of total coliforms and Escherichia coli were quantified using IDEXX Colisure with Quanti-Tray/2000 incubated at 35◦C for 24 h. MPN of enterococci were determined using IDEXX Enterolert kit with Quanti-Tray/2000 incubated at 41◦C for 24 h. Slurry and digestate samples were diluted as necessary to fall within the detection range (1 - 2419.6 cfu 100 mL−<sup>1</sup> ) in sterilized phosphate buffered saline (Colisure) and sterilized distilled water (Enterolert).

### Assessing Treatment Effects on FIB Die-Off With Bayesian Hierarchical Modeling

Bayesian hierarchical modeling was used to compare the effects of vessel volume and pasteurization conditions on FIB die-off. Weak Cauchy-distributed priors were used for the pooled parameter estimates of the regression, to allow for outliers (Gelman and Hill, 2006). Stan version 2.17.0 (Carpenter et al., 2017) was used to generate samples from the model using the Rstan interface (Stan Development Team, 2017). The data, model, analysis scripts, and interpretation of the results can be found at https://github.com/nickp60/SI\_Nolan\_etal\_2018. A difference in parameter estimates was considered significant if the 95% confidence intervals were exclusive.

# RESULTS

## Slurry Characterization

The slurry collected from the three farms was tested prior to AD, for initial FIB levels as well as total solids and volatile solids (**Table 1**). TS and VS were consistent across the samples tested, whilst coliforms and E. coli numbers were highest in samples from Farm C. In all cases, enterococci numbers were lower than coliforms and E. coli.

## Miniature- (50 ml) and Laboratory-Scale (10 L) Bioreactor Performance Is Similar

The recorded performance data in the comparative trial displayed similar trends for miniature- and laboratory-scale bioreactors. The pH for both bioreactor scales remained between 7.6 and 8.1 throughout the experiment (**Figure S2**). Volatile solids (VS) degradation was comparable for the 50 mL and 10 L bioreactors with 64 and 61% VS removal, respectively within the first 7 days (**Figure 1A**). Similar trends in ammonia concentration (**Figure 1B**) were also observed across the two scales, with an increase over the first 2 weeks of the trial from 937 to 1,233 mg L −1 in the 10 L bioreactors and from 865 to 1,038 mg L−<sup>1</sup> in the 50 mL bioreactors. This increase likely results from the breakdown of organic compounds. As ammonia concentration has been identified as an important factor in pathogen reduction (Watcharasukarn et al., 2009, the similarity between the scaled bioreactors is of particular relevance.

Soluble and total chemical oxygen demand (sCOD and tCOD) concentrations were also consistent across the two bioreactor scales (**Figures 1C,D**). Soluble COD and tCOD removal primarily occurred within the first 7 days, reaching a maximum of 87–88% by Day 28 for both bioreactor scales (**Figures 1C,D**). The majority of methane production occurred within 14 days, reaching 77.5 and 82% of the total recorded in the 10 L and 50 mL bioreactors within that time frame (**Figure S3A**). Although similar methane production trends were observed at both bioreactor scales, the larger scale bioreactors approached the theoretical yield proposed by Batstone et al. (2002) of 350 mL CH<sup>4</sup> g <sup>−</sup><sup>1</sup> of COD at Day 21 compared to Day 28 for the 50 mL bioreactors (**Figure 1E**). This could partly be attributed to the more thorough mixing occurring in the larger bioreactors.

### Bayesian Hierarchical Modeling

A Bayesian hierarchical model was developed to compare the effects of vessel volume and pasteurization conditions on FIB survival. In short, both the initial effect (from day 0 to day 7) and the latter effect (from day 7 to day 28) of the conditions were considered in relation to the underlying behavior of the data. This piece-wise approach was able to accurately model both the initial perturbation (the addition of feedstock to the inoculum) and the recovery of the system (https://github.com/nickp60/SI\_ Nolan\_etal\_2018).

### Fecal Indicator Bacteria Survival Is Comparable in 50 ml and 10 L Bioreactors

Fecal indicator bacteria levels should be reduced to <1,000 cfu g−<sup>1</sup> for the safe landspreading of digestate (Regulation (EC) No.1069/2009 and Regulation (EU) No. 142/2011). Total coliforms survival showed similar trends in both 50 mL and 10 L bioreactors, with a 3.7 and 4.3 log<sup>10</sup> reduction after 7 days (**Figure 2A**). A similar trend in E coli die-off was also observed in both bioreactor scales (**Figure 2B**). The initial 3.5–4.3 log<sup>10</sup> reductions of both coliforms and E. coli occurring within 7 days (**Figures 2A,B**), followed by relatively stable survival until 21 days suggests the presence of resilient cells with increased ability to survive under mesophilic AD conditions. Although enterococci numbers were slightly above 1,000 cfu g−<sup>1</sup> after 21 days, greater than 3.0 log<sup>10</sup> reduction was observed after 28 days in both bioreactor scales (**Figure 2C)**. The parameter estimates obtained from piece-wise modeling of the FIB die-off data showed well-overlapping confidence intervals, indicating no significant difference between the two bioreactor volumes.

### Pre-pasteurization Impacts Scod Removal and Methane Yield

Two pre-AD (P1: 60◦C for 96 h and P2: 70◦C for 60 min) and two post-AD pasteurization regimes (P3: 60◦C for 96 h

### TABLE 1 | Slurry characterization.


*Mean (n* = *3) slurry pathogen indicator numbers (log*<sup>10</sup> *cfu g*−<sup>1</sup> *) and TS/VS% from 3 cattle farms.*

and P4: 70◦C for 60 min) were tested at the miniature scale. Volatile solids degradation was relatively consistent across all conditions, as was ammonia concentration (**Figures 3A,B**). For both total COD and soluble COD, the rate of removal within the first 7 days was notably higher for AD of feedstock that had been pre-pasteurized at 60◦C for 96 h (89 vs. 74–80% for sCOD, **Figure 3**; and 93 vs. 82–85% for tCOD; data not shown). The impact of P1 on COD removal was observed at the first time point only, as by Day 14, the other conditions displayed similar results (**Figure 3C**). Although the total volume of methane produced for P1 was similar to the other conditions, high levels of COD removal combined with low biogas quality (22–40% CH<sup>4</sup> Day 2, 54–68% CH<sup>4</sup> Day 5; **Figure S3B**) resulted in lower yields of 146 mL CH<sup>4</sup> g COD−<sup>1</sup> by Day 7 (**Figure 3D**),

compared with 227 mL CH<sup>4</sup> g COD−<sup>1</sup> for no pasteurization. Pre- pasteurization at the EU standard (P2) improved methane yield, approaching the maximum theoretical methane yield of 350 mL CH<sup>4</sup> g COD−<sup>1</sup> within 7 days (**Figure 3D**; Batstone et al., 2002). As expected, the two post-AD conditions had no impact on the AD process itself and the results for key performance indicator data recorded for P3 and P4 (**Figure 3**) were comparable to those of the unpasteurized condition presented in **Figure 1**.

## Post-AD Pasteurization Decreases Fecal Indicator Bacteria Survival

Pre-AD pasteurization (P1 and P2) was carried out on the food production waste prior to mixing with slurry and feeding into bioreactors, as is standard practice. This resulted in a reduction in E. coli (1.19–1.33 log10) numbers on Day 0, particularly for P1, but had minimal impact on total coliform numbers compared with no pasteurization (**Figures 2**, **4**). Overall, the effect of pre-pasteurization treatments (P1 and P2) on FIB survival was not statistically significant. Post-AD treatment under Irish and EU transformation parameters (P3 and P4) resulted in lower coliform and E. coli numbers in the digestate, when compared with unpasteurized (**Figures 2**, **4**). When comparing pre-pasteurization with post pasteurization, the post-pasteurized treatments showed significantly lower coliform counts; whilst the other indicators shared similar trends (P1 and P2; **Figure 4**). At all post-AD pasteurization time-points, coliforms and E. coli were below the limit of detection in the majority of replicates, while enterococci numbers were below 1,000 cfu g−<sup>1</sup> within 7 days (**Figure 4**).

# AD Treatment Effectively Reduces Fecal Indicator Bacteria Levels Compared to Stored Slurry

Cattle slurry from three dairy farms was stored in a shed at ambient environmental temperature for 56 days (between 4 and 13◦C in Galway, Ireland). Over the first 7 days of storage there was a 0.32 and 0.36 log<sup>10</sup> reduction in coliforms and E. coli numbers respectively, and a slight increase in enterococci numbers. Hence, within 7 days of AD treatment, the resulting digestate was superior to stored slurry in terms of FIB inactivation (**Figure 5**). It is worth noting that an initial dilution factor of 1–1.5 log<sup>10</sup> is evident in the digestate when compared with unprocessed slurry. This is due to the mixing of slurry with FOG and microbial inoculum prior to AD processing. After 2 months of storage, none of the FIB tested in slurry had dropped below the EU minimum digestate quality standards of 1,000 cfu g−<sup>1</sup> (**Figure 5**).

# DISCUSSION

Systematic examination of the fate of key viral, bacterial and protozoan pathogens in farm-based anaerobic co-digestion of various wastes is hampered by availability of sufficient pathogenic biomass as well as health and safety concerns associated with spiking large-volume bioreactors. This makes the use of larger scale bioreactors for pathogen survival studies impractical. Here, we carried out a comparative trial across two bioreactor scales, of 50 mL and 10 L, in order to assess the potential use of miniaturescale AD bioreactors as proxies for larger scales. Across all the major physicochemical parameters recorded, both bioreactor

scales displayed similar trends. The volatile solids removals obtained in the present study were in line with those reported in the literature (64–67%—Neves et al., 2009; Luste et al., 2012). The majority (61–64%) of the volatile solids degradation at both scales occurred within 7 days (**Figure 1A**), demonstrating the potential for reduced retention time of the substrate in the bioreactors. Reported methane yields vary significantly, depending on feedstock mixtures and ratios, retention time, and temperatures, but a range between 200 and 489 mL g VS−<sup>1</sup> is typical of codigestion containing manure as the primary constituent with food production waste (200–350 mL g VS−1–Neves et al., 2009; 470 mL g VS−1–Creamer et al., 2010; 260 mL g VS−1–Luste et al., 2012; 320–489 mL g VS−<sup>1</sup> Dennehy et al., 2016). The range of 220–488 mL CH<sup>4</sup> g VS−<sup>1</sup> recorded in the present work falls within those previously reported. Here we demonstrate, at 50 mL and 10 L bioreactor scales, that mesophilic AD of slurry co-digested with FOG effectively reduces coliforms and E. coli numbers within 7 days (**Figures 2A,B**). Similarly, whilst examining the effect of varying ratios of pig slurry co-digested with food waste in dry-AD, Jiang et al. (2018) recently reported coliform and E. coli inactivation within 7 days, identifying free VFA concentration as a primary factor in inactivation. Dennehy et al. (2018) found similarly reduced levels of E. coli (1.2–2.2 log<sup>10</sup> cfu g −1 ) in mesophilic CSTR co-digesting pig manure with food waste, although higher total coliform values were reported (4–6 log10). The higher total coliforms reported by Dennehy et al. (2018) may be due to reduced mixing (1 h per day), decreased hydraulic retention time and feeding regime employed (daily feeding vs. batch) when compared to the present study.

Using Bayesian hierarchical modeling provided a flexible framework for assessing the statistical significance of the indicator die-off rates. As the vast majority of change in FIB numbers occurred within the initial 7 days, taking a piecewise approach allowed assessment of both the initial effect of the feedstock addition under the different pasteurization schemes, and also the long-term effect on FIB counts in the system as it stabilized over time. We hope that by releasing both the data and models used to assess the data, such an approach will become a regular tool in assessing bioreactor performance, particularly in relation to pathogen survival.

The results obtained for both bioreactor scales indicate higher enterococci survival in mesophilic anaerobic co-digestion of slurry with FOG, compared with coliforms or E. coli. This observation is in agreement with the previously reported examination of four full-scale Swedish biogas plants, one thermophilic and three mesophilic, co-digesting manure with kitchen, and food-processing waste, where higher numbers of enterococci than coliforms were consistently found in the digestate, despite the use of pre-AD pasteurization in all four plants (Bagge et al., 2005). Furthermore, the enterococci survival results of Bagge et al. (2005) mirror closely those of Dennehy et al. (2018), whereby ∼3 log<sup>10</sup> cfu g−<sup>1</sup> were consistently recorded, using a continuously fed system and three different ratios of pig manure to food waste. Based on these observations, enterococci are recommended as a better indicator for pathogen survival during AD processes (Larsen et al., 1994; Sahlström, 2003).

Numerous studies have examined the impact of pre-AD pasteurization on process performance, typically anticipating improved methane yield caused by preliminary hydrolysis of the feedstock (Luste and Luostarinen, 2010). The corresponding results have however varied widely, ranging from a methane production reduction of up to 34% during the co-digestion of slaughterhouse waste (SHW) with the organic fraction of municipal solid waste (Cuetos et al., 2010) to no significant effect during the AD of SHW (Hejnfelt and Angelidaki, 2009; Ware and Power, 2016), through 14–25% improvements during the co-digestion of SHW and slurry (Paavola et al., 2006; Luste and Luostarinen, 2010). Edström et al. (2003) initially reported a 400% increase in BMPs of pasteurized vs. unpasteurized SHW, although this yield was not achieved in laboratory or pilotscale trials. The variability of these results is likely due to differences in biochemical properties of the feedstocks used, as demonstrated in a study examining the effects of pre-treatment on five different components of SHW (Luste et al., 2009). In the present study, the methane output when FOG was prepasteurized at 70◦C for 1 h was statistically higher than the other conditions in the first 7 days of this trial (**Figure 3D,E**), although Carrere et al. (2016) advise against extrapolating such results to full-scale plants without complex modeling. Although methanogenesis appears to have been impacted differentially by the two pre-AD pasteurization conditions tested, FIB survival was similar for both conditions. Slightly higher FIB numbers were recorded after 28 days in systems processing pre-pasteurized feedstock (P1 and P2; **Figure 4**). This may be indicative of reduced competition for resources, whereby pre-pasteurization reduced the microbial populations in the feedstock, enabling increased FIB survival and/or regrowth of resilient strains or cells.

A number of pasteurization conditions were examined by Coultry et al. (2013) to determine the energy consumption and consequent economic impact on viability of AD plants. Pre-AD pasteurization was demonstrated to be prohibitively expensive; most notably, the energy required to meet the Irish national transformation standard (P1) equates to 4,544% of the digester's output, which is an 80-fold increase in energy consumption when compared with the already prohibitive EU requirement (P2). These numbers are likely to be lower in practice however, as only the imported materials are pasteurized before mixing in with indigenous slurry, reducing the pasteurization treatment efficacy as seen in the FIB survival results for P1 and P2. The energy cost of post-AD pasteurization is mitigated by the mesophilic digestate, but was still found by Coultry et al. (2013) to be substantial, at 30 and 1,893% of the digester's annual energy output for EU (P4) and national standards (P3) respectively. Although some measures could be taken to reduce these costs, such as separation of liquid and solids, they are clearly a substantial burden to the economic viability of bioreactor operation. This burden hinders adoption of farm-based AD and is worth reconsideration in light of the reduction in FIB numbers in unpasteurized trials and the absence of hygienization requirements for unprocessed slurry. The FIB survival rates monitored in the stored slurry are in line with previous studies such as that of Nicholson et al. (2005), who found that E. coli O157, Salmonella and Campylobacter survived for up to 3 months during dairy slurry storage. Similarly, Mycobacterium avium subspecies paratuberculosis has been found to survive beyond 56 days in stored slurry at ambient temperatures (Grewal et al., 2006). Furthermore, survival of pathogens in stored slurry increases with temperatures below 10◦C, such as those typical of winter storage months in north-western European climates (Kudva et al., 1998). In terms of potential pathogen load to the environment, as assessed via the monitoring of FIB levels, we have demonstrated that mesophilic anaerobic co-digestion of slurry with food production waste is superior to simple slurry storage without treatment. Moreover, the slight increase in numbers of enterococci over the first 7 day period of slurry storage (**Figure 5**) highlights the potential risk of pathogens thriving in this environment. Based on these findings, if the EU standard for digestate was applied to slurry (<1,000 cfu g−<sup>1</sup> ), all livestock farms would be required to adopt some form of treatment.

Previous studies have examined the agronomic benefits of anaerobic digestion (AD) of slurry. Benefits include increased homogeneity and decreased viscosity, due to the reduction in volatile solids, resulting in more uniform landspreading (Massé et al., 1997). As detailed by Massé et al. (2011), other studies have demonstrated the added fertilizer value of digestate compared with slurry and mineral fertilizer, resulting from improved plant N uptake and increases in N and P mineralization (Massé et al., 2007; Chantigny et al., 2009). When these agronomic improvements, energy production, waste reduction and mitigation of GHG emissions are considered together with reduced pathogen load to the environment, widespread adoption of AD as a means of slurry amendment prior to landspreading should be encouraged (Clemens et al., 2006). The enterococci survival observed in this study highlights however the scope for future work to improve pathogen inactivation during farm-based AD. Optimization of operational conditions for FIB reduction is currently underway. Future work focusing on landspreading field trials will be necessary to assess further comparative risk from digestate and unprocessed slurry.

### CONCLUSION

In this study we demonstrate that (i) miniature 50 mL bioreactors are valid proxies of farm-based AD to carry out targeted pathogen survival investigations and (ii) in situ AD treatment of slurry prior to land application reduces the level of FIB compared to slurry storage alone, independently of pasteurization, which in turn might be indicative of a decreased potential pathogen load to the environment and associated public health risks. While pathogen indicator die-off was observed, enterococci survival highlights the opportunity for process optimization with a focus on hygienization.

# AUTHOR CONTRIBUTIONS

FA and SN designed the research. SN and AA ran the bioreactors. NW and LP carried out the modeling. SN and FA analyzed the data. SN, FA, NW, and LP wrote the paper with input from VO, FB, OF, KR, and DB.

### FUNDING

This work was carried out as part of the FIRM Project, 14 F847, The comparative public health risks associated with spreading anaerobic digestate, animal manure, and slurry on land: Science, policy and practice, funded by the Irish

### REFERENCES


Department of Agriculture, Food and Marine. NW was funded by a joint studentship by NUI Galway and the James Hutton Institute.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fsufs. 2018.00041/full#supplementary-material

Figure S1 | Comparison of methanogenic activity performance of various ratios of granular sludge:BEOFS digestate:slurry for development of inoculum (*n* = 3). FOG: Fats, oils and grease; PRO: Propionate; BUT: Butyrate; ETH: Ethanol; ACE: Acetate.

Figure S2 | Recorded pH for (A) 10 L and 50 mL bioreactors processing unpasteurized slurry and FOG; (B) 50 mL bioreactors testing pasteurization conditions.

Figure S3 | Mean methane percentages for 10 L and 50 mL (A) and P1-P4 (B) at all timepoints (*n* = 3), with standard deviation error bars. Detailed information about the models used, in addition to the data and analysis script, can be found at http://nickp60.github.io/SI\_Nolan\_etal\_ 2018.


production and process limitations. Proc. Saf. Environ. Prot. 90, 231–245. doi: 10.1016/J.PSEP.2011.10.001


and Finnish national regulations. Water Sci. Technol. 53, 223–231. doi: 10.2166/wst.2006.253


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Nolan, Waters, Brennan, Auer, Fenton, Richards, Bolton, Pritchard, O'Flaherty and Abram. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Pollution of Surface and Ground Water by Sources Related to Agricultural Activities

Nada Sasakova<sup>1</sup> \*, Gabriela Gregova<sup>1</sup> , Daniela Takacova<sup>1</sup> , Jana Mojzisova<sup>1</sup> , Ingrid Papajova<sup>2</sup> , Jan Venglovsky <sup>1</sup> , Tatiana Szaboova<sup>1</sup> and Simona Kovacova<sup>1</sup>

<sup>1</sup> Department of the Environment, Veterinary Legislation and Economy, University of Veterinary Medicine and Pharmacy in Košice, Košice, Slovakia, <sup>2</sup> Parasitological Institute of the Slovak Academy of Sciences, Košice, Slovakia

### Edited by:

Francisco Javier Salazar, Remehue Regional Research Center, Institute of Agricultural Research, Chile

### Reviewed by:

John Hudson Loughrin, United States Department of Agriculture, United States Francisco Jesus Fernandez Morales, Universidad de Castilla-La Mancha, Spain

> \*Correspondence: Nada Sasakova nada.sasakova@uvlf.sk

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 02 March 2018 Accepted: 02 July 2018 Published: 27 July 2018

### Citation:

Sasakova N, Gregova G, Takacova D, Mojzisova J, Papajova I, Venglovsky J, Szaboova T and Kovacova S (2018) Pollution of Surface and Ground Water by Sources Related to Agricultural Activities. Front. Sustain. Food Syst. 2:42. doi: 10.3389/fsufs.2018.00042 The objective of this study was to monitor the quality of ground water supplied to animal farms and 2 villages and of surface water (rivers) in the same area (Košice basin, eastern Slovakia) with the aim to assess contamination of water by potential sources in this area. Samples for physico-chemical and microbiological examination were collected at 12 sampling points (6—surface water; 6—ground water) from May 2014 up to March 2015, covering all four seasons. The examination and evaluation of individual parameters was carried out according to relevant Slovak legislation compatible with EU Drinking water directive. The physico-chemical evaluation focused on parameters that indicate pollution of water resulting from human activities and farming. Microbiological examination included determination of counts of bacteria cultivated at 22◦C and 37◦C (BC22 and BC37), total coliforms, E. coli and fecal streptococci. Ground water intended for mass consumption (farms, villages) is abstracted from wells, collected in storage reservoirs and disinfected before brought to consumers. Some families in the villages use their own wells. Water for individual consumption (individual households) originates directly from individual wells. Examination of potable water used on agricultural farms showed some possibility of contamination of sources by runoff and inappropriate manipulation with excrements. Surface water in in the area close to both farms was polluted with organic substances (CODMn), however they did not exceed the limit set for surface water. At many samplings we detected in surface water presence of total coliforms, E. coli and occasionally also fecal enterococci indicating fecal pollution that could eventually affect ground water in individual wells. Our investigations showed that protection zones of water sources were not always sufficient. There were considerable variations in the quality of surface water during the year but no clear relationship between microbial contamination and seasons was observed. Quality of ground water supplied for mass consumption complied with legislative regulations except for BC 22 (heterotrophic count at 22◦C) in summer and autumn). Water from individual wells contained occasionally presence of total coliforms, E. coli and enterococci and higher heterotrophic counts.

Keywords: ground and surface water, quality, pollution, agriculture, protection zones

# INTRODUCTION

Potable water of good quality is essential for life. Human activities interfere in many ways with natural water cycle and affect the society-water relationship. Constantly increasing human population and its expectations regarding the standard of living increase demands on exploitation of existing resources including water (Chowdhury, 2013).

Different uses of water affect both the quality and the quantity of the water available and the management of water pollution and water resources play an important role at both national and international level. Water remains one of the most poorly managed resources on earth. Division to types of water according to their occurrence reflects only the instantaneous state and location while the real state and its dynamics in nature is not considered. Upon contact with soil, the rain water becomes surface water and after soaking in it may be called ground water. Thus insufficient protection of surface water against contamination with human and animal wastes may cause major water supply problems.

Availability of good quality potable water is affected also by global climate changes that cause shortages and overexploitation in some places and flooding in other places with all related consequences including decreased safety of food and potential disease transfer.

Anthropogenic pressure on the environment leads to decrease in water quality but there is some limit which cannot be exceeded or else global ecological balance will be disturbed.

### Pollution of Water

There are many man-made pollutants that can contaminate water sources. With regard to their origin we recognize two categories of their sources, point and diffuse. Examples of important point sources are industrial premises, towns, agricultural installations, manure storage, and landfills. They can be more easily identified and controlled that diffuse (non-point) sources, such as leaching of nitrates and pesticides into surface and ground water as a result of rainfall, soil infiltration, and surface run off from agricultural land. Such sources cause considerable variations in the contaminant load of water over time (Fawell and Nieuwenhuijsen, 2003).

In addition to division of contaminating sources to point and non-point, we recognize two types of contamination of water: (1) Emergency contamination (single) frequently with immediate catastrophic impact, resulting in death of fish and other water fauna and many serious damages; (2) Long-term contamination manifested by persisting organic pollution. It has a total negative effect on water environment and structure of food supply for water fauna, resulting in absence of some fish species in the affected river zones.

Many infectious diseases of animals and humans are waterborne. The agents of these diseases are transferred by ingestion of water contaminated with human or animal feces that contain pathogenic bacteria, viruses and parasites (protozoa, eggs of parasites). They may survive in water for different periods of time depending on many factors. Monitoring of safety of water sources is based on determination of parameters that indicate pollution caused by sewage, animal excrements, storage of waste, animal manure and artificial fertilizers, and other (Sasakova et al., 2013; Fridrich et al., 2014). An important tool that helps to eliminate pollution of water sources is the Directive 2010/75/EU on integrated prevention of pollution and control that applies to industrial and agricultural installations with large pollution potential. However, this directive generally does not apply to diffuse sources and many smaller point sources.

### Assessment of Water Safety

Assessment of safety of drinking water is carried out on the basis of national standards or international guidelines. The WHO Guidelines for Drinking-Water Quality (1996) serve as a guide for the setting of national regulations and standards for water safety in support of public health. In Slovakia the requirements on water used for watering the animals are set by the Slovak Republic Government Regulation No. 68/2007 Coll<sup>1</sup> ., which amends and supplements the Act No. 322/2003 Coll<sup>2</sup> ., on protection of farm animals. The requirements on the quality of water used for human consumption are determined by the Slovak Republic Government Regulation No. 496/2010 Coll., which complies with the criteria set by European Communities regulations and WHO guidelines. In this Regulation there are included also methods for the control of quality of water used for human consumption.

The limit parameters for surface water are set by Slovak Republic Government Regulation No. 296/2005 Coll<sup>3</sup> ., which stipulates criteria for achieving good water balance.

### Organic and Inorganic Pollutants

The main source of organic pollution of rivers is the organic matter derived from diverse human activities. This involves domestic and industrial sewage, wastes from agriculture and animal production, food processing facilities and other. Many toxic organic compounds are non-biodegradable, or are degraded slowly, so they persist in the ecosystem; some are magnified in the food web; some may cause cancer in humans; others are converted into carcinogens when they react with chlorine used to disinfect water; some affect even kill fish and other aquatic organisms; some are nuisances, giving water and fish an offensive taste or odor. Acidification of inland waters by acidifying compounds of sulfur and nitrogen affects quality of water and causes damage to aquatic ecosystems, especially to fish.

Freshwater eutrophication is another worldwide problem. Eutrophication (excessive growth of phytoplankton and filamentous algae resulting in increased turbidity, production of toxins, diurnal changes in dissolved oxygen) is caused by enrichment of water with nitrogen and phosphorus. Phosphorus emissions arise predominantly from domestic and industrial effluents, but the share of agriculture is not insignificant.

Rivers are recipient for rain water from relevant catchment areas but also of wastewater (treated and untreated) and

<sup>1</sup>Regulation of the government of the SR No. 68/2007 Coll.

<sup>2</sup>Act No. 322/2003 Coll. on protection of farm animals.

<sup>3</sup> Slovak Republic Government Regulation No. 296/2005 Coll., which introduces requirements on the quality and qualitative goals for surface water, as well as the limit indicator values for wastewater and special water contamination.

infiltration from landfills. Removal of some pollutants is very difficult and expensive, therefore prevention of such pollution is preferred. Partial solution of this problem is based on zones of protection of water sources (Sasakova et al., 2014).

The primary pollution of ground water results from substances that naturally occur in ground water and the mineral environment or by all types of point and diffuse sources of pollution. Therefore, ground water also requires protection, regular monitoring and some treatment before it is used for drinking and other domestic uses.

### Microbiological Pollution of Water

Water may be polluted by various pathogens—bacteria, viruses, protozoa, and helminths. According to the WHO, 80% of all diseases in the developing countries results from contaminated water. The major sources of infectious agents are (1) untreated and improperly treated sewage, (2) animal waste in fields and feedlots beside waterways, (3) meat packing and tanning plants that release untreated animal waste into water and (4) some wildlife species, which transmit waterborne diseases. The spectrum of pathogenic and potentially pathogenic microorganisms spread by water is extensive. The most frequent are the causative agents of intestinal diseases (typhoid, paratyphoid, salmonellosis, tuberculosis, brucellosis, tularaemia, leptospirosis, cholera, amoebic dysentery, schisostomiasis). A special group are diseases of viral etiology, such as infectious hepatitis, poliomyelitis, aseptic meningitis, diseases of the respiratory and gastrointestinal tracts. Because water is not examined for the presence of viruses and because general microbiological analysis fails to detect them, they pose a considerable threat for humans and animals.

Determination of microbiological safety of drinking water has traditionally been carried out by monitoring the counts of bacteria that serve as indicators of fecal contamination. They are usually monitored at the entry to the supply system and at certain fixed and randomly located points within the distribution system. Much effort was devoted to finding an ideal indicator microorganism but, at present, no single micro-organism used for this purpose meets satisfactorily all the desired criteria.

Heterotrophic plate counts (bacteria cultivated at 22 and 37◦C) enumerate bacteria that are derived principally from environmental sources. If their levels increase substantially from normal values, there may be cause for concern. Coliform bacteria (CB) and E. coli in drinking water indicate fecal contamination (Horakova et al., 2003) due to insufficient protection of water source, inadequate water treatment, hygiene protection and distribution or secondary contamination. Fecal streptococci or Enterococci (EC) are indicators of fecal contamination and general contamination. They tend to persist longer in the environment than thermotolerant or total coliforms.

According to WHO (2011) Escherichia coli are the only true indicator of fecal contamination; they are exclusively of intestinal origin and are found in feces. Their presence is an indicator of fresh fecal contamination and thus of serious shortcomings in protection of water sources and water safety.

Despite the fact that ground water is filtered when passing through the soil, it is susceptible to microbial contamination, particularly with viruses, and requires periodical checking and should be disinfected when used for mass consumption.

To ensure microbiological safety of potable water various disinfection technologies are used. When using active chlorine for this purpose, existence of a target chlorine residual concentration after a specified contact time serves as a reliable indicator of real-time control of bacteria and viruses (EPA, 2011).

Many research studies indicated that disinfection of water with active chlorine is not the ideal way of ensuring its safety due to development of by-products, particularly when water contains traces of organic substances. Of main concern there is the potential production of trihalomethanes, particularly long exposure of humans to these substances, and formation of chloro- and bromo-benzoquinones, the by-products of the chlorination process (Gunten, 2003).

# Protection of Water

A sufficient quantity of good potable water cannot be ensured without protecting the water sources. Generally, three-fourths of the water used in agriculture, industry, and our homes comes from surface waters, and the rest from groundwater. Although ground waters are less exposed to pollution than surface waters, the consequences of their pollution are longer lasting.

The effort to protect water and watercourses against pollution has a long tradition, but only on a local scale. The growing awareness of possible problems led to a UNECE (United Nations Economic Commission for Europe, 1992) Water Convention on the protection and use of transboundary watercourses and international lakes.

This convention, together with its amendments, is intended to strengthen national measures for the protection and ecologically sound management of transboundary surface waters and ground waters. The Water Convention requires Parties to prevent, control, and reduce transboundary impact, use transboundary waters in a reasonable and equitable way and ensure their sustainable management.

Legislation concerning water pollution is particularly complex. Different statutes may apply to discharges depending on whether they are made into public sewers, the marine environment, and inland, estuarine, and tidal waters. Two possible control regimens are envisaged: individual state directives must specify either a maximum emission standard or quality of water in the recipient after the point of discharge.

# Protection Zones of Ground Water Sources

Source protection zones (SPZs) are the basic measure that allows one to control the risk to groundwater supplies intended for mass consumption from potentially polluting activities and accidental releases of pollutants. The land-surface zoning approach of protection of groundwater against both point and diffuse pollution is hydrogeologically based. It is a policy tool that controls polluting activities activities around water supplies intended for human use.

Three zones are typically defined:



To control diffuse pollution, it is necessary to consider also the nature of the soil cover in the area of potential polluting activities (Adams and Foster, 1992; EPA, 2011).

In Slovakia, the Act No. 29/2005 Coll.<sup>4</sup> on hygiene protection zones of water sources specifies the above mentioned zones of protection as zones of the Ist, IInd, and IIIrd degrees. In the Ist degree zone all activities not related to the operation of water source are banned. Only authorized individuals are allowed to enter this zone. Less strict bans on activities apply to the zones of IInd and IIIrd degree.

### Protection Zones of Surface Water

Pollution of surface water at the site of abstraction of water that is used for drinking after appropriate treatment is also prevented by zones of hygiene protection. This strategy helps to eliminate some pollutants that can be removed with considerable difficulties at higher costs. Several zones of protection (SPZs) are specified. Similar to the zones around ground water sources the zone immediately around the point of withdrawal of water is most strictly protected. The size of individual zones depends on particular situation (population, human activities, geological conditions, etc.).

### lst Degree—Basic Requirements

They apply to the zone around the site of direct water abstraction: 200–300 m upstream, 50 m downstream (or down to the waterworks which raises the river level). This includes the banks to the distance of 15 m from the watercourse. Elimination of all sources of pollution must be ensured. Warning signs, floating buoys, or fences are used. In this zone discharge of wastewaters or sewage, bathing, fishing, storage of crude oil products, supply of biogenic elements, storage of harmful substances and geological prospecting is banned and cemeteries, storage of carcasses, industrial plants producing wastewaters, and animal feedlots must not be built.

### 2nd Degree—Basic Requirements

The zone extends upstream, always up to the watershed (parting, if needed). Protection measures involve regulation of surface flow regimen and prevention of erosion. The following activities are prohibited: discharge of wastewaters, supply of biogenic elements, storage of harmful substances in an inundation area. Industrial plants with harmful wastewaters and animal feedlots must not be put up in the area.

### 3rd Degree—Basic Requirements

The protection of the 3rd degree applies to the entire water catchment area above the site of abstraction. If its part is not protected as the SPZ of the lst or 2nd degree, then the requirements as at 3rd degree apply to this area.

The aim of the study was to monitor changes in the quality of water obtained from ground water sources that was intended for mass consumption (farms, villages) and also of surface water (rivers) in the same area flowing close to animal farms and villages oriented on agricultural production with the aim to identify potential sources of its contamination.

### MATERIALS AND METHODS

Monitoring of quality of ground and surface water in an agricultural area of eastern Slovakia focused on determination of physico-chemical parameters and bacterial counts indicating quality and potential pollution of water sources. Samples of water for examination were collected from May 2014 up to March 2015, to cover all four seasons. The results of ground water were assessed on the basis of the requirements on the quality of water used for human consumption determined by the Slovak Republic Government Regulation No. 496/2010 Coll. The quality of the investigated surface water was evaluated on the basis of the Slovak Republic Government Regulation No. 296/2005 Coll., which stipulates criteria for achieving good water balance.

### Description of the Monitored Area and Sampling Sites

### Monitored Area and Collection of Samples

All sampling sites were located in southeastern part of Košice basin, eastern Slovakia, close to the border with Hungary.

The monitored location is found in geomorphological area Slanské vrchy, 895 above the sea level. The area includes three national natural reservations (NNR), one natural reservation (NR), and one natural landmark (NL). NNR—Velký Milic was ˇ declared a protected area in 1967. It covers an area of about 67.81 hectares with important breeding places of predatory birds and well preserved forest communities on southern igneous rocks of Slanske mountains. NNR—Malý Milic was declared a ˇ protected area in 1950 in order to provide protection to typical primeval Milic oak-beech and beech growth. NNR—Marocká ˇ hola was declared in 1950. It covers an area of 63.76 hectares. It ensures protection to primeval beech forest growth of typical composition and structure on andesite and andesite tuffs. NR— Malá Izra was declared in 1976, covering an area of 0.88 hectares. It ensures protection to rare natural communities of mooralder Slansky mountain forests. This area is covered by marshyalder forest of lowland type located about 700 m above the sea level.

The investigated area is agricultural, with two villages and two animal farms. Both villages and farms are supplied with potable water from ground sources that comply with legislative requirements on potable water intended for mass consumption. In villages there are some families that have their own individual wells. Also some of them keep small number of farm animals.

The investigated area with location of rivers, ground water sources, farms, manure storage, and water sampling sites is

<sup>4</sup>Act No. 29/2005 Coll. which defines details on designation of water management sources, on water protection measures, and on technical treatments within the water management source protection zones.

depicted in **Figure 1**. Village 1 is located about 240 m above sea level and a small river Zidovsky potok, that begins in a mountain ridge about 900 m above sea level, flows through this village where its banks are mostly regulated. It flows to another small river Torocky potok which originates in the same mountain ridge and passes close to Village 1 and Village 2 (185 m above sea level) and next to farm 2. Both rivers are small, but in case of heavy rain or rapid melting of snow they have not sufficient capacity to drain off all water and may overflow. The last heavy flooding occurred in 2010 when water on some streets of Village 1 was more than 0.5 deep.

Village 1 is supplied with potable water from a drilled well, 15 m deep. The water is first pumped to a reservoir of capacity 2 × 150 m<sup>3</sup> and then is brought to consumers in the village by plastic mass distribution system. This water is regularly checked for its chemical and bacteriological quality. Only small number of inhabitants of this village uses water from individual wells the safety of which is not ensured (Fox et al., 2017). Farm 1 located next to this village is oriented on keeping sheep and goats (about 60).

A ground water source supplies potable water to farm 2. It is located 80 m from the manure storage. The water is again pumped into a reservoir of capacity 300 m<sup>3</sup> . The farm keeps about 80 fattening cattle and 20 horses that are used mostly for recreational purposes. There is an unused water well close to the Farm 2, 150 m away from manure storage.

Two liters of water were sampled to chemically clean bottles for physico-chemical evaluation and 1 liter was collected to a sterile bottle for microbiological examination. The samples were processes immediately after returning to a laboratory.

### Physico-Chemical Examination

Chemical examination of surface and ground water included determination of pH, electrical conductivity, dissolved oxygen, chemical oxygen demand (CODMn), chlorides, nitrates, iron, and phosphates. In addition, sum of calcium and magnesium and free chlorine was determined only in potable water and total dissolved solids (TDS) only in surface water.

The pH was determined according to STN ISO 10523<sup>5</sup> by means of a pH-meter HACH and a WATERPROF pH Tester 30. Conductivity was determined by a conductometer WTW InoLab Cond 720 (Germany). Dissolved oxygen was determined electrochemically using an oxygen probe LDO HQ Series Portable Meters supplied by HACH and chemical oxygen demand by oxidation with KMnO<sup>4</sup> according to STN EN ISO 8467<sup>6</sup> . Determination of Ca2<sup>+</sup> and Mg2<sup>+</sup> was carried out by titration according to Horakova et al. (2003), chlorides were determined by titration according to STN ISO 9297<sup>7</sup> by titration and nitrates with ion-selective nitrate electrode WTW (InoLab pH/ION 735P, Germany). Iron was determined by powder HACH Method 8025 Color True at 465 nm. Orthophosphates were determined colorimetrically using HACH DR 2800 analyser and a procedure recommended by HACH.

Potable water was examined for the presence of active chlorine by titration according to STN EN ISO 7393-3<sup>8</sup> by titration. Surface water was examined for the level of total dissolved solids (TDS) by filtration and drying at 105◦C until constant weight, or incineration at 550◦C.

Examination of all parameters was carried out in duplicate.

### Microbiological Examination

Determination of counts of relevant bacteria was carried out in compliance with the Slovak Republic Government Regulations No. 496/2010 Coll.

We determined colony forming units (CFU) of bacteria cultivated at 22◦C (BC22) and 37◦C (BC37) (heterotrophic count) according to STN EN ISO 6222. A pour-plate method was used and the counts of BC22 and BC37 were determined using meat-peptone agar and aerobic incubation at relevant temperature for 24 h.

Coliform bacteria (CB) and E. coli were cultivated according to STN EN ISO 9308-1<sup>9</sup> using Endo agar (HiMedia, India) and incubation for 24 h at 37 and 43◦C, respectively, and the characteristic colonies were counted. In the absence of colonies, the incubation was prolonged for additional 24 h. According to the respective regulation, lactose fermentation test was performed for confirmation of coliform bacteria.

Determination of counts of fecal enterococci (FE) was carried out according to STN EN ISO 7899-2. It consisted of filtering 100 ml or 10 ml of water sample (for water intended for mass consumption or individual consumption, respectively) through a membrane filter (filter of pore size 0.45µm). The filter was then placed onto a solid selective medium containing sodium azide (to suppress growth of Gram-negative bacteria) and colorless

5 STN ISO 10523 (2010) Water quality. Determination of pH. 2,3,5-trifenyltetrazolium chloride, which is reduced by intestinal enterococci to red formazan.

All samples were examined in duplicate.

### RESULTS

### Results of Physico-Chemical Examination

Results of physico-chemical examination are presented in **Tables 1**, **2**.

Levels of all investigated parameters were within the limits specified by relevant regulation except for N-NO<sup>−</sup> 3 (**Figure 2**), which was exceeded in most of the samples.

Examination of ground water showed that some limits specified by SR Government Regulation No. 496/2010 Coll. was exceeded while the levels of pH, conductivity, dissolved oxygen, CODMn and active chlorine (residual concentration after disinfection of water intended for mass consumption) corresponded with the requirements on potable water.

Results of microbiological examination of monitored waters are presented in **Tables 3**, **4**.

The levels of nitrates in samples of surface and ground water determined in samples collected in individual seasons are presented in **Figures 2**, **3**.

TABLE 1 | Results of physico-chemical examination of surface water during the monitored year and legislative limits of parameters according SR Government Regulations No. 296/2005.


TDS, total dissolved solids; CODMn, chemical oxygen demand.

TABLE 2 | Results of physico-chemical examination of ground water monitored during 1 year period and legislative limits of parameters according SR Government Regulations No. 496/2010.


CODMn, chemical oxygen demand.

<sup>6</sup> STN EN ISO 8467 (2000) Water quality. Determination of permanganate index (in Slovak).

<sup>7</sup> STN ISO 9297 (2000) Water quality. Determination of chloride. Silver nitrate titration with chromate indicator (Mohr's method) (in Slovak).

<sup>8</sup> STN EN ISO 7393-3 (1990) Water quality. Determination of free chlorine and total chlorine. Part 3: Iodometric titration method for the determination of total chlorine. (in Slovak).

<sup>9</sup> STN EN ISO 9308-1 (1990) Water quality. Detection and enumeration of coliform organisms, thermotolerant coliform organisms and presumptive Escherichia coli. Membrane filtration method (in Slovak).

According to WHO (2008), neither E. coli nor coliform bacteria can be detected in any 100 ml sample (WHO, 1996, STN EN ISO 9308-1:90).

The Water Quality (2000) stipulates that fecal enterococci must not be detected in any 100 ml sample of water.

### DISCUSSION

Protection of water sources from pollution that can ensure availability of potable water of good quality is an essential requirement for sustainable development. Surface waters are polluted by point sources, such as agricultural or industrial installations, or via overland flow from rain or snowmelt. Subsequently, by transport through the soil profile, pollutants can reach groundwater and, according to their character, can have very serious consequences.

### Results of Physico-Chemical Examination

The physico-chemical properties of water, particularly pH, temperature, the presence of organic matter, level of dissolved oxygen, electric conductivity, turbidity, content of NH3, metals, and other chemical components, affect the quality of drinking water and some of them may exert direct effect on the health of consumers (Pitter, 2009). In addition, these parameters can affect survival of potential disease agents and the effectiveness of disinfection (Block, 2001).

pH is a measure of acidity or basicity of water. It is influenced by biological processes that occur in water. N-substances, Psubstances and chlorides serve as indicators of fecal pollution but some of these substances may have also serious health effects. Chemical oxygen demand (COD) is an important water quality parameter. Higher COD levels in surface water mean a greater amount of oxidisable organic material, which will reduce dissolved oxygen (DO) levels. A reduction in DO can lead to anaerobic conditions, which is deleterious to higher aquatic life forms. Presence of COD in ground water indicates the risk of development of by-products (trihalomethanes) in water disinfected with active chlorine. The by-products (BPs) of chlorine disinfectants can affect the health of consumers of the disinfected water or induce in them various responses. Their extent depends on a number of factors such as the period of action, concentration, and frequency of exposure (Gunten, 2003).

One of the most common groundwater contaminants in rural areas is nitrate. Nitrate in groundwater originates primarily from fertilizers, septic systems and manure storage or spreading operations. Nitrate compounds are soluble and the nitrate ion is not retained in soil. Nitrate is thus the nitrogen species most exposed to loss by leaching. Excess levels in drinking water are particularly serious for infants as their immature digestive system allows the reduction of nitrate to nitrite leading to methemoglobinemia. The levels in the range of 100–200 mg/l nitrate-N (443 to 885 NO<sup>−</sup> 3 ) start affecting the health of general population, but the effect on any given person depends on many factors. However, they are not considered indicators of possible presence of other more serious residential or agricultural contaminants, such as bacteria or pesticides (McCasland et al., 1998).

Our monitoring showed that levels of all chemical parameters determined in samples of surface water were below the limits specified by the relevant regulation in all seasons except for concentrations of N-NO<sup>−</sup> <sup>3</sup> **—**which exceeded the legislative limits at all sampling sites except for the site 2e (Torocky potok) (**Figure 2**).

Of chemical parameters determined in ground water increased levels were observed only for chlorides, nitrates, and phosphates. There were considerable differences between quality of water intended for mass supply (1a, 1b,1c, 1e) and water from private wells (1d, 1f). The level of chlorides was


TABLE 3 | Results of microbiological examination of surface water collected in individual seasons.

CB, coliform bacteria (total coliforms); FE, fecal enterococci; BC22, bacteria cultivated at 22◦C; BC37, bacteria cultivated at 37◦C; CFU, colony forming units. The bold values are exceeded values in comparison with legislative limits.

exceeded only in sample 1d (private well in Village 1, used by a family keeping some farm animals—the well is located 15 m from manure storage) in each season (212.7, 210.1, 212.6, 198.6 mg/l). Higher concentration of chlorides in regions with a low content of salts indicates organic pollution of water. In such cases ammonia, nitrites and nitrates are also increased. Phosphates were exceeded also in this sample (1d), 2 times (1.4 mg/l**—**summer; 2.1 mg/l—winter).

Examination of samples of water from an unused well (1f), located at a distance of 150 m from manure storage and close to farm 2 showed that nitrate level exceeded the legislative limit in all seasons. Other chemical parameters complied with the regulations but microbiological examination indicated considerable bacteriological pollution of water in this well as the levels of all examined groups of bacteria highly exceeded the legislative limits.

Phosphorus is a common constituent of agricultural fertilizers, manure, and organic wastes in sewage and industrial effluent. Phosphates in ground water can also originate from P-deposits.

The level of nitrates was exceeded only in private wells but still was not as high as to cause serious problems in adults. However, it was unsuitable for infants. Ground water in source 1e showed somewhat higher level of nitrates in comparison with source 1a (Ground water reservoir in Village 1 supplying water to the village and Farm 1).

The original WHO recommendations for the use of chlorine as a disinfectant stipulated a minimum free chlorine concentration of 0.5 mg/l after 30 min contact time (EPA, 2011). This level was not exceeded in any sample (ground water, surface water).

### Results of Microbiological Examination

Runoff is the key mechanism of pathogen transport to surface waters. During a rain event, the partitioning of flow between surface runoff and infiltration through the soil depends upon a number of factors. Storm intensity and duration, soil hydraulic characteristics (e.g., permeability, antecedent moisture, and temperature), land slope, and soil cover have all been shown to


TABLE 4 | Results of microbiological examination of ground water collected in individual seasons.

CB, coliform bacteria (total coliforms); FE, fecal enterococci; BC22, bacteria cultivated at 22◦C; BC37, bacteria cultivated at 37◦C. The limits for sources intended for individual consumption—max. 30 persons or capacity—are as follows: \*0 CFU/10 ml; \*\*500 CFU/1 ml; \*\*\*100 CFU/1 ml. The bold values are exceeded values in comparison with legislative limits.

influence runoff and therefore pathogen transport (Rosen, 2000; USEPA, 2002). If rainfall intensity exceeds the capacity of the soil to infiltrate water, overland flow occurs, and microorganisms can be carried rapidly in surface runoff (Tyrrel and Quinton, 2003; Unc and Goss, 2003).

To be available for transport in runoff, pathogens are released from the manure, most of them remain associated with the fecal deposit. Their amount depends upon a number of factors such as the manure itself, loading of pathogens in the manure, the pathogen types and survival characteristics, and the age and source of the manure. Aging can greatly reduce the amount of microorganisms that leach out of the manure (NRCS/USDA, 2012).

Transport through the soil profile and in ground water involves an extremely complex interplay of physical and chemical processes that depend upon the size and surface properties of the microorganism, the composition, mineral surface properties, and texture of the soil or aquifer material, hydraulic conditions and other.

Pathogen survival in water also depends on many factors including water quality (e.g., turbidity, dissolved oxygen, pH, organic matter content) and environmental conditions (i.e., temperature, predation by zooplankton). Exposure to UV light is a key factor in bacterial, viral, and protozoan die-off in surface waters (Rosen, 2000; Cotruvo et al., 2004; Fong and Lipp, 2005). An aquifer environment also protects pathogens against UV exposure and facilitates their survival in ground water.

Our microbiological examination concentrated on indicators of fecal contamination that is associated with transfer of many disease agents.

While BC22 reflect general contamination of water, the BC37 is more important parameter as it indicates contamination with microflora of warm-blooded animals.

For surface water there is a limit only for KM 22 (5000 CFU/1 ml), which was not exceeded in any sample. The limits for BC22 (200 CFU/1 ml) and BC37 (50 CFU/1 ml) in ground water intended for mass consumption are logically much stricter

(Slovak Republic Government Order No. 496/2010 Coll10). The BC22 counts were exceeded in source 1a (Ground water reservoir in Village 1, supplying water to the village and Farm 1) in summer and autumn and in individual sources, where less strict requirements were applied, highest counts were detected also summer but also in winter. The counts of BC37 were exceeded in source 1a in autumn and were much exceeded in individual sources practically in all seasons.

Determination of bacteriological safety of water intended for drinking have been associated for long time on detection of total coliform bacteria (CB). It was designed to detect potential, not an actual health hazard.

The limit set for CB in surface water (**Table 3**) was exceeded several times in all seasons at all sampling sites but most frequently in autumn and winter that can be related partially to application of manure. Limit for CB counts in potable water was exceeded only once in water for mass consumption (1 b in summer—ground water intended for mass supply—Farm 1 next to Village 1). In the same sampling site there was detected also 1 E. coli colony. It is possible that some contamination occurred before or during the sampling (from water tap).

Samples of ground water intended for mass consumption were free of indicators of fecal pollution except for sample from source 1b in summer. This water is periodically checked for its quality and disinfected before distribution to consumers. However, some risk of fecal pollution exists for individual wells.

Microbiological examination of water from the unused well (1f), located 150 m from manure storage, showed increased counts of total coliforms and E. coli exceeding legislative limits (0 KTJ/100 ml) in all seasons. During the summer the fecal enterococci counts increased (34 CFU/100 ml) which indicated potential contamination of water by manure storage. At the same time also the remaining microbiological parameters (BC22, BC37) exceeded the legislative limits. In case of the future potential use of this well as a source of drinking water it is necessary to increase the distance of potential contamination sources. In addition, maintenance of well surroundings, regular cleaning and disinfection and regular monitoring of water quality are the routine activities that should be carried out to protect the consumers of water.

The only source of surface water in which E. coli counts did not exceed the legislative limit was 2f (Zidovsky potok—upriver from Farm 1).

At the sampling site of surface water 2e (Torocky potok downriver from Farm 2) neither concentrations of nitrates nor of other determined chemical parameters were exceeded but all microbiological examination revealed highly increased counts of all determined bacterial groups in warmer seasons with the exception of winter sampling. This sampling site was located downriver from both villages and the increased bacterial counts were related to anthropogenic and agricultural activities.

Determination of E. coli and fecal enterococci (FE) counts indicated some fecal contamination of surface water practically at all sampling sites. There were considerable variations with regard to seasons. As expected, the lowest microbial contamination was detected at the sampling site 2f (Zidovsky potok—upriver from Farm 1), situated upriver from both farms.

### CONCLUSION

Our examinations showed relatively good quality of surface water with respect to determined physico-chemical parameters, even at sampling points where some pollution from point sources was expected. However, not all parameters reflecting quality of surface water were determined. Also more frequent sampling is required to support fully such conclusion. Some

<sup>10</sup>Slovak Republic Government Order No. 496/2010 Coll. defining requirements for water intended for human consumption and quality control of water intended for human consumption.

fecal contamination of surface water was detected practically at all sampling sites. The best microbial quality was observed at sampling site 2f (Zidovsky potok—upriver from Farm 1).

Chemical examination of the quality of ground water intended for mass consumption indicated that the required level of some parameters was exceeded (chlorides, phosphates nitrates). Bacteriological safety of this water is ensured by regular monitoring and disinfection. The contamination of individual sources of drinking water was not very high and could be improved by their cleaning and disinfection. However, removal of sources of potential contamination would appear as the best choice.

Availability of water of good quality is essential for preservation of health of humans and animals. This can be ensured by regular monitoring and protection of water sources against point and diffuse pollution particularly that related to spreading of diseases. Importance of water protection zones was confirmed also by our results as surface water exposed to the

### REFERENCES


lowest potential contamination showed best microbial quality. Ground water intended for mass is regularly controlled and disinfected and thus presents low risk to consumers. Individual water wells require higher attention as their safety is less frequently checked and little controlled.

### AUTHOR CONTRIBUTIONS

NS, GG, JM, and JV preparation of manuscript. DT preparation of manuscript, technical support. IP chemical examination of samples. TS laboratory works. SK collection of samples.

### ACKNOWLEDGMENTS

The study was supported by the project VEGA 2/0125/17 and the Slovak Ministry of Culture and Education Grant Agency No. 003UVLF-4/2016.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Sasakova, Gregova, Takacova, Mojzisova, Papajova, Venglovsky, Szaboova and Kovacova. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Improvements in the Quality of Agricultural Soils Following Organic Material Additions Depend on Both the Quantity and Quality of the Materials Applied

Anne Bhogal <sup>1</sup> \*, Fiona A. Nicholson<sup>1</sup> , Alison Rollett <sup>1</sup> , Matt Taylor <sup>2</sup> , Audrey Litterick <sup>3</sup> , Mark J. Whittingham<sup>4</sup> and John R. Williams <sup>5</sup>

<sup>1</sup> ADAS Gleadthorpe, Mansfield, United Kingdom, <sup>2</sup> Grieve Strategic Ltd., Shipston on Stour, United Kingdom, <sup>3</sup> Earthcare Technical Ltd., Waterlooville, United Kingdom, <sup>4</sup> School of Biology, Newcastle University, Newcastle upon Tyne, United Kingdom, <sup>5</sup> ADAS Boxworth, Cambridge, United Kingdom

### Edited by:

Francisco Javier Salazar, Remehue Regional Research Center, Institute of Agricultural Research, Chile

### Reviewed by:

Peter Sørensen, Aarhus University, Denmark João Coutinho, University of Trás-os-Montes and Alto Douro, Portugal

> \*Correspondence: Anne Bhogal anne.bhogal@adas.co.uk

### Specialty section:

This article was submitted to Waste Management in Agroecosystems, a section of the journal Frontiers in Sustainable Food Systems

> Received: 05 December 2017 Accepted: 26 March 2018 Published: 19 April 2018

### Citation:

Bhogal A, Nicholson FA, Rollett A, Taylor M, Litterick A, Whittingham MJ and Williams JR (2018) Improvements in the Quality of Agricultural Soils Following Organic Material Additions Depend on Both the Quantity and Quality of the Materials Applied. Front. Sustain. Food Syst. 2:9. doi: 10.3389/fsufs.2018.00009 It is widely recognized that the application of organic materials is one of the most effective ways of increasing soil organic carbon (SOC) levels and improving soil quality, but do all forms of organic matter input have the same impact on soil properties A network of seven experimental sites investigated the effects on soil quality of annual applications over a minimum of 3 years of compost and food-based digestate in comparison with farmyard manure (FYM) and livestock slurry. Two of the sites were existing experimental platforms which had previously benefitted from applications of FYM, livestock slurry and green compost allowing the effects of longer-term applications (6–17 years) on soil properties to be quantified. The application of all organic materials increased soil nutrient supply (total nitrogen, extractable phosphorus, potassium, and magnesium) within a short timescale (<3 years), whereas SOC contents were only increased following the long-term (9 years or more) application of bulky organic materials (compost and FYM). SOC increases were associated with improvements in soil biological (microbial biomass) and physical properties (reduced bulk density), although the level of improvement was dependent on the quality of the organic material applied (as determined by its lignin content, an indicator of resistance to decomposition). Applications of low dry matter content materials (digestates and livestock slurries) had a limited capacity to improve soil biological and physical functioning, due to their low organic matter loading.

Keywords: organic materials, soil organic carbon, soil quality, digestate, compost

# INTRODUCTION

The application of organic materials to agricultural soils is a widely recommended practice not only as a source of essential plant nutrients which can provide savings in inorganic fertilizer use (Defra, 2010), but also as a means of increasing soil organic carbon (SOC) levels with associated improvements in soil biological and physical functioning (Bhogal et al., 2009). Indeed, the benefit of a range of organic material applications (livestock manures, composts, biosolids, etc.,) for SOC and soil quality has been widely documented and reviewed (e.g., Edmeades, 2003; Johnston et al., 2009). Studies have been conducted to evaluate the potential of organic materials as nutrient sources (Schröder et al., 2005) and soil conditioners (Diacono and Montemurro, 2010), as well as a means to sequester carbon (C) in the mitigation of climate change (Powlson et al., 2012).

Developing nutrient management and circular economy policies (e.g., European Union, 2015) have led to increasing amounts of organic materials being directed away from landfill and beneficially recycled to agricultural land, to complete natural nutrient and C cycles. Consequently, the utilization of a wide range of "alternative" organic materials to the more traditional livestock manures is being actively encouraged within agricultural systems. Notably the anaerobic digestion of sourcesegregated food waste is an area of significant growth in the UK, with around 5 Mt of the 7 Mt of food waste currently sent to landfill each year predicted to be available for digestion by 2020 (DECC and Defra, 2011). In order to provide confidence in the use of these materials within agricultural systems, it is important to demonstrate their long-term effects on soil and food quality as well as developing sustainable nutrient management practices by minimizing environmental losses (Bhogal et al., 2016; Nicholson et al., 2017).

The long-term impact of digestates on soil properties is a largely unexplored area of research, particularly where food waste is the feedstock (Nkoa, 2014), but can lessons be learnt from the application of other, similar organic materials such as livestock slurries? Do all organic carbon (OC) inputs have the same impact on soil properties regardless of the source? For example, changes in the total SOC pool have been shown to depend on the amount of organic material (C) applied, but not on the type of material (e.g., Rasmussen et al., 1980) whereas there is conflicting evidence on the response of other soil properties to the type of material applied. For example, Bhogal et al. (2009) observed that repeated OC inputs (for at least 7 years) in the form of livestock manures led to improvements in soil physical properties (bulk density, porosity, and available water capacity) whereas OC additions in the form of crop residues (straw) did not. By contrast, Peltre et al. (2015) measured changes in specific draft force that were related to the amount of OC applied and its effect on bulk density rather than due to differences in the type of organic material (including compost, livestock manures and sewage sludge applied annually over 11 years).

A recent research programme evaluating the use of digestate and compost derived from food waste in UK agriculture has provided a good opportunity to explore some of these issues ("DC-Agri;" www.wrap.org.uk/dc-agri). At its inception (in 2010) there were less than five commercially operated anaerobic digestion (AD) plants processing food waste in the UK, compared to over 100 plants currently in operation (2016), returning c.1.4 Mt digestate to agricultural soils on an annual basis (WRAP, 2014). This compares to the annual return of c.93 Mt of farm manures (Nicholson et al., 2008), 1.1 Mt of biosolids (Water, 2010) and 1.3 Mt of compost (c. 20% as green/food compost; WRAP, 2013). The experimental programme evaluated the soil quality implications of repeated applications of food-based digestate and green/food compost and included comparator materials more commonly applied to agricultural soils (livestock manures, green compost). This paper uses the results from this work to determine whether the soil quality benefits of organic materials depend on the amount (number of applications × OC content) or on the form of OC applied.

## MATERIALS AND METHODS

### Experimental Sites and Treatments

In autumn 2010, a network of seven sites was established on a range of soil types and across different agroclimatic zones in Great Britain, namely: Aberdeen and Ayr in Scotland, Devizes, Faringdon, Harper Adams, and Terrington in England, and Lampeter in Wales (**Table 1**). Two of the sites were permanent (cut) grassland and five were in an arable rotation. At each site, 18 experimental plots (60–160 m<sup>2</sup> in size, depending on the


TABLE 1 | Experimental site details.

a sl, sandy loam; scl, sandy clay loam; zcl, silty clay loam; c, clay; cl, clay loam.

<sup>b</sup>Measured on the control treatment at the beginning of the experiment.

<sup>d</sup>Crops grown each harvest year: SB, spring barley; WB, winter barley; WOSR, winter oilseed rape; WW, winter wheat; G, grassland; POT, potatoes; Lin, Linseed; WC, whole crop oats/peas.

<sup>e</sup>Existing experimental platforms, see text and Bhogal et al. (2009) for details.

<sup>c</sup>30 year average annual rainfall.

site) were laid out in a randomized block design comprising three replicates of six treatments (five organic materials and a control receiving inorganic fertilizer only; **Table 2)**. Crops were grown according to best farm practice using commercially recommended seed rates, with crop protection products applied according to good agricultural practice to control weeds, pests and diseases, with the aim of growing healthy and productive crops. At the arable sites, annual cultivations (by plow/disc ≤25 cm deep) occurred in opposite directions each year to minimize soil movement between the plots. The grass sites were cut twice during each season for silage.

In the first cropping year, organic materials were applied in autumn 2010 at Ayr and Terrington, and at the other sites (Aberdeen, Devizes, Faringdon, Harper Adams, and Lampeter) once ground conditions and Nitrate Vulnerable Zone (NVZ) regulations allowed in spring 2011. Organic material applications were repeated in autumn 2011 at Aberdeen, Devizes, Faringdon, Lampeter, and Terrington and in spring 2012 at Ayr and Harper Adams, with a final application in autumn 2012 at all seven sites. Cattle farmyard manure (FYM) and slurries were used at all sites (from sources local to each site), except Terrington where pig manures (FYM and slurry) were used. The sites at Harper Adams and Terrington were existing experimental platforms and had previously benefitted from applications of FYM and livestock slurry for 16–17 years prior to 2010 and green compost for 6 years (Bhogal et al., 2009, 2011), with Harper Adams also previously receiving food-based digestate applications for 3 years. At Aberdeen, green compost had also been previously applied for 1 year (Litterick, 2009). All the digestate and composts used at the experimental sites were certified quality products according to the PAS Assurance Schemes (PAS 100 for composts and PAS 110 for digestates; BSI, 2011, 2014).

All the high-dry matter (solid) organic materials were applied at a target rate of c.250 kg nitrogen (N)/ha, and the low dry matter (liquid) materials were applied at rates between 120 and 250 kg N/ha, depending on the volume (restricted to ≤80 m<sup>3</sup> /ha) and based on analysis provided by the supplier. Supplementary inorganic fertilizer N was applied to all treatments to balance crop N supply with that recommended for the crop, in line with the control treatment, using MANNER-NPK predictions of organic material crop available N supply (Nicholson et al., 2013). Phosphate (P2O5), potash (K2O) and sulfur (SO3) were applied at a single rate to all treatments, based on the requirements of the untreated control (Defra, 2010; SAC, 2010). Mineral fertilizer

TABLE 2 | Experimental treatments.


applications aimed to ensure (as far as was practically possible) that no major nutrient limited crop growth, and that crop yields and residue returns were the same on all treatments (i.e., the only difference in OC inputs was from the applied organic material treatments).

Triplicate samples of each organic material were taken at spreading every year and analyzed as detailed in **Table 3**. In some years there were slight discrepancies between the total N analysis provided by the organic material supplier (which was used to calculate application rates) and that determined at the time of application, which, on some occasions, led to N loadings in excess of the target rate (**Table 3**). Over all the sites, the average annual N loading was close to the target ranges at c.160–250 kg/ha/yr total N (**Table 3**). FYM supplied the most P, K and S, compost the most total N (although over 95% of this was in slowly available organic forms), and the food-based digestate and livestock slurry supplied the most readily available N (RAN; **Table 3**).

### Soil Quality Measurements

In spring 2013, c.6 months following the final application of treatments, a range of topsoil (0–15 cm) chemical, biological, and physical properties were measured at each site (**Table 4**). This involved taking c.5 kg topsoil from each plot for determination of soil chemical properties, microbial biomass, respiration and potentially mineralisable N (PMN), three replicate intact soil cores (0–5 cm depth) per plot for determination of soil bulk density, porosity and water held at field capacity, and a representative 500 g/plot topsoil sample for determination of water held at 2 and 15 bar. Shear strength (10 vanes/plot) and penetration resistance (10 penetrometer readings/plot) were determined in the field.

### Data Analysis

At each site, the effect of the different organic material treatments on soil quality was evaluated using conventional analysis of variance (ANOVA) and comparison of P-values; post-hoc testing was undertaken to evaluate which treatment means were different from each other using a Duncan's multiple range test (using Genstat version 12; VSN International Ltd., 2010).

Multi-predictor models (using R statistical software; R Core Team, 2014) were then used to establish whether differences observed at the individual sites were consistent across all sites or whether the responses differed with site (i.e., soil type and climatic conditions), land use (grass/arable) and prior history (i.e., whether the sites had a previous history of repeated organic material additions, as at Harper Adams and Terrington; Aberdeen was not included as a site with a prior history as green compost had only been applied for 1 year). Both generalized linear mixed models (GLMMs) and general linear models (GLMs) were used, with experimental site included as a random effect in the former and as a fixed effect in the latter, with all models nested. The importance of individual predictors within the models (i.e., site, land-use and prior history) was assessed by comparing Akaike's Information Criteria (AIC) values (i.e., a lower AIC value indicated a better fit), with an improvement in AIC of greater than six indicating a substantially improved



<sup>a</sup>n, number of sites and seasons (mean of samples taken from seven sites in each of three seasons for most organic materials); three replicate samples were taken at each site in each season.

b kg/t fw, kilograms/ton fresh weight; % dm, percent dry matter.

<sup>c</sup>RAN, Readily available nitrogen (i.e., ammonium-N + nitrate-N).

fit (i.e., a meaningful difference) and values of between two and six a minor improvement (Richards, 2005, 2008). This approach is well-recognized (e.g., Verzani, 2014), with the results from the models supported by observed patterns in the data. Most of the multi-predictor models assumed a normal distribution, with responses transformed to normality as required (e.g., by log transforming data). Model fits were assessed using standard diagnostics such as Quantile-Quantile (QQ) plots. Due to poorly distributed data, SOC was analyzed using the non-parametric Kruskal–Wallis Test, a one-way test of response variable vs. treatment, without controlling for site.

### RESULTS

### Organic Carbon Loading Rates

Total OC loadings from the organic materials applied over the 3 year experimental period, together with OC loadings from historic applications (at Harper Adams and Terrington), are summarized in **Table 5**. Over the 3 year experimental programme the green compost and FYM treatments supplied similar amounts of OC (c.9 t/ha OC), the green/food compost c.7 t/ha OC, livestock slurry c.5 t/ha OC and food-based digestate c.1 t/ha OC, with differences between the sites reflecting the different sources and hence composition of the organic materials. Harper Adams and Terrington were existing experimental platforms and had benefitted from historic applications of FYM, slurry, and green compost, and at Harper Adams from food-based digestate applications. The recent and historic organic material applications extended the range of OC loadings from <1 t/ha OC up to c.50 t/ha OC.

### Effect of Organic Material Additions on Soil Quality

There was a substantial improvement in model fit when site was included in the multi-predictor model (AIC values improved by >6 in many cases) for nearly all of the measured parameters (**Table 6**), which was not surprising given the range of soil types and agroclimatic locations. However, land-use (grass vs. arable) and prior history (i.e., whether there was previous history of organic material additions) also had an impact on the response of some parameters to the treatments applied (**Table 6**). As the underlying baseline soil conditions varied across the sites, the results have been presented as a percentage difference from the control treatments in order to normalize the data and identify the overall direction of change in soil properties as a result of the organic material additions across all the sites. Error bars have therefore not been shown on the charts, as these would be misleading given they would be due to variations in both site and treatment. For the full (individual site) experimental results<sup>1</sup>

### Soil Organic Carbon

Topsoil SOC contents were highly variable across the sites, despite this, the Kruskal–Wallis test indicated a significant

<sup>1</sup>http://www.wrap.org.uk/content/digestate-and-compost-agriculture-dc-agrireports.

Bhogal et al. Organic Additions and Soil Quality

treatment effect (P < 0.001; **Table 6**). Inspection of the individual site ANOVAs revealed that treatment effects were only evident at the two sites with a prior history of green compost, FYM and livestock slurry applications (i.e., Harper Adams and Terrington). Here, there was clear evidence that repeated applications of bulky organic materials for 9 years or more increased SOC (**Figure 1**), with both green compost and FYM resulting in a





Terrington (supplying up to 26 t/ha OC; **Table 5**), the 5–10%

increase in SOC was not statistically significant (P > 0.05). Although the 9 years of green compost applications supplied only half the OC (c.30 t/ha) that had been supplied by the almost 20 years of FYM applications (50–60 t/ha OC), it resulted in a comparable increase in total SOC levels. Retention of the OC (i.e., the increase in SOC relative to the control treatment expressed as a percentage of the total C loading) following the green compost additions (20–24%) was therefore almost double that from FYM (12%), which suggested the green compost was more resistant to decomposition. This was supported by the lignin composition of the applied materials, with the green compost containing c.70% lignin compared to c.55% in the FYM (**Table 3**).

### Soil Microbial Biomass

The multipredictor modeling suggested that there was no treatment effect on microbial biomass C, but a strong treatment × site interaction and prior history effect on biomass N (**Table 6**). The determination of microbial biomass involves analysis of the dissolved organic C and N content of a soil sample before and after fumigation, with the before-and-after difference equating to the microbial biomass (Brookes et al., 1985). Either the C or N content can be used as a measure of the size of the soil microbial population, with C contents typically larger, but more variable than N (with a microbial C:N ratio ranging between 4 and 8). This variability most likely explains the absence of any overall treatment or prior history effect on microbial biomass C,


Results are for the 3 year DC-Agri experimental programme (2010–2013), plus historic applications where applicable (breakdown between the recent and historic applications given in parenthesis).

<sup>a</sup>Green compost was applied at two rates in 2009 at Aberdeen (Litterick, 2009); supplying c.1.5 t/ha OC to the green compost treatment and c.3.2 t/ha OC to the green/food compost treatment.

<sup>b</sup>At Harper Adams, cattle FYM and slurry were applied annually for 16 years prior to DC-Agri, supplying c.51 and c.21 t/ha OC, respectively (Bhogal et al., 2009, 2011); green compost was introduced in 2004 and applied for 6 years, supplying c.19 t/ha OC; food-based digestate was applied for 3 years, supplying c.2.4 t/ha OC (Charles Murray, pers. comm). <sup>c</sup>At Terrington, pig FYM and slurry were applied annually for 17 years prior to DC-Agri, supplying c.36 and c.9 t/ha OC, respectively (Bhogal et al., 2009, 2011); green compost was introduced in 2004 and applied for 6 years, supplying c.18t/ha OC.

TABLE 6 | Summary of multipredictor modeling results performed on data from all sites<sup>a</sup> .


<sup>a</sup>Models tested for the effect of site, treatment, treatment × site interaction (i.e., is the effect of treatment the same at all sites?), grass/arable (i.e., is the effect of treatment different at grass sites compared to arable sites?); prior history (i.e., is the effect of treatment different at sites with a prior history of organic material additions?). \*\*\*Strong evidence of an effect (AIC improved by > 6); \*Weak evidence of an effect (AIC improved by 2–6); N, no evidence of an effect (AIC values similar i.e., <2 difference); nd, not determined as not possible to fit model due to lack of transformation to normality or other sensible error structure not appropriate); Parameters were untransformed unless indicated (ln = logged) and were fitted with a Gaussian Error Structure. Note that the main effects of "Site" or "Treatment" were assessed without the higher order interaction term included in the model.

<sup>b</sup>There was evidence of the effect of site on respiration rates, but no effect of any of the other factors testing in the modeling exercise. There were no differences in AWC that could be explained by the models. This confirmed the single site analyses, so AWC has not been included in this table.

c \*\*\*P < 0.001 in the Kruskal–Wallis Test.

despite this being evident in the microbial biomass N results. It is also surprising that there was no effect of land use (grass/arable) on both microbial biomass C and N, although there were only two grassland sites compared to five arable sites in the dataset. Individual site ANOVAs confirmed the effect of prior history, with increases in both biomass C and N (P < 0.01) only observed at the two sites (Harper Adams and Terrington) with a prior history of green compost, FYM and livestock slurry additions (**Figures 2**, **3**). Although the repeated compost and FYM additions had the same effect on the total SOC pool (a 25% increase), the FYM had a proportionally greater effect on the soil microbial biomass increasing it by 50–60% compared to a c.20% increase in biomass following repeated green compost additions (**Figures 2**, **3**). Despite these differences in soil microbial biomass, topsoil respiration rates did not differ significantly (P > 0.05) between the treatments (**Table 6**).

### Soil Nutrient Supply

As expected, the application of all the organic materials increased soil nutrient supply (**Table 7**) with improvements in topsoil total N, extractable P, extractable Mg (P < 0.05 at four of the seven sites for each of these nutrients) and extractable K (P < 0.05 at six of the seven sites). Overall, there was little difference in the response between grass and arable sites, and sites with a prior history of organic material additions (**Table 6**), which suggested that the addition of organic materials had improved soil nutrient status over a relatively short time-frame (within 3 years). The greatest increases in topsoil nutrient status were following FYM applications, which increased topsoil total N by on average 10%, extractable P by c.35%, extractable K by c.80% and extractable Mg by c.20%. Moreover, the capacity of soils to retain and exchange nutrient cations was also improved, as measured by the cation exchange capacity (CEC), with significant treatment effects measured at



Results are expressed as a percentage difference from the control treatment averaged across all seven sites.

two of the seven sites (Harper Adams and Lampeter), and weak evidence of an effect across all sites (**Table 6**). Here, the application of bulky organic materials (green compost and FYM) resulted in the greatest increases (6–7% increase relative to the control).

The organic material additions had a significant effect on topsoil pH at four of the seven experimental sites, namely, the two grassland sites and those with a prior history of organic material additions. This was confirmed by the modeling results which showed strong evidence of a difference between grass and arable sites and weak evidence of a prior history effect (**Table 6**). At the grassland sites, pH tended to increase by 0.3– 0.5 pH units where all organic materials had been applied, most likely a reflection of the pH (and neutralizing value) of the organic materials. The only exception was on the foodbased digestate treatment at Lampeter where pH decreased by 0.2 units, which was probably a reflection of the local soil conditions (e.g., buffering capacity and moisture content) in combination with the acidifying effect of the nitrification process as the ammonium-N within the digestate was converted to nitrate-N. At the two arable sites with a prior history, the pH was increased by 0.3–0.5 units on the long-term FYM and livestock slurry treatments (but not the green compost; P < 0.05), again most likely reflecting the pH of the applied materials, but only apparent where these materials had been applied for 20 years.

Topsoil potentially mineralisable N (PMN), a biological measure of the soils capacity to supply N through the mineralization of soil organic N reserves, also increased following FYM and livestock slurry additions at three of the seven experimental sites (Devizes, Harper Adams, and Terrington, P < 0.05), with the multipredictor modeling results showing a strong treatment affect which was similar across all sites. There was no improvement in model fit by comparing grass and arable sites or sites with a prior history (**Table 6**). However, two of the three sites with significant treatment effects were those with a prior history of organic material applications, with the relative increase in PMN (compared to the fertilizer only control) most marked where FYM, livestock slurry and green compost had been applied for 9+ years (**Figure 4**). Again, differences were proportionally greater where FYM had been applied forc.20 years (>100% increase) compared to green compost additions over 9 years (c.60% increase), despite similar total SOC and total N contents.

### Soil Physical Properties

There was a marked improvement in the multipredictor model fits for the variation in topsoil bulk density across the sites due to treatment (i.e., AIC improved by >6), with grassland sites responding differently to arable sites, and a weak improvement in the model fit due to prior history (AIC improved by 2–6; **Table 6**). At the arable sites, the application of bulky organic materials (i.e., FYM and green compost) for 9 or more years resulted in lower BD and consequently higher porosity; these treatments also had the highest SOC contents (**Figure 1**). Unlike the changes in SOC, the decrease in BD was greater following repeated addition of FYM (c.8% decrease relative to the control) compared to green compost (c.5% decrease relative to the control), despite the similar total SOC contents (**Figure 5**). This was similar to the pattern observed for both microbial biomass and PMN (**Figures 2**–**4**).

The topsoil bulk density at the grassland sites responded differently to the applied treatments compared with the arable sites (**Table 6**). Grassland soils generally have inherently lower bulk density than arable soils (largely due to higher SOC contents). At both of the grassland sites there were small (c.5%) decreases in bulk density following the application of compost and FYM for 3 years, which were statistically significant at Ayr (P < 0.05) and marginal (P = 0.06) at Lampeter. However bulk density increased (and porosity decreased), where organic materials with a low dry matter content (i.e., food-based digestate and livestock slurry) had been applied (**Figure 6**).

Results from the modeling exercise also revealed a weak effect of treatment on topsoil shear strength (a measure of the force required to work the soil), but no differences between grass and arable sites and no effect of prior history (**Table 6**). Looking in more detail at the individual site analyses, shear strength decreased at Harper Adams following the application of green compost, FYM and livestock slurry for 9 or more years (P < 0.05), with no treatment effects observed at the sites where materials had only been applied for 3 years. Again the decrease in shear strength at Harper Adams was greater on the longterm FYM treatment (c.20% decrease relative to the control) compared to the long-term green compost and livestock slurry treatments (c.10% decrease). By contrast, topsoil shear strength (and penetration resistance) at the Ayr grassland site increased following 3 years of food-based digestate (P < 0.05), with no effect of the other organic material treatments.

The decreases in soil bulk density (and increases in porosity) did not, however, lead to statistically significant increases in the total available water capacity (AWC) at any of the sites. However, the multipredictor modeling suggested a significant effect of treatment on the volumetric water content held at field capacity (0.05 bar), 2 bar and 15 bar and the easily available water capacity (EAWC—water held between field capacity and 2 bar pressure), with the model fits markedly improved by taking into account land use (i.e., grass/arable; **Table 6**). Inspection of the individual site analyses revealed an increase in the volumetric water content held at field capacity, 2 and 15 bar where FYM and to a lesser extent, green compost had been applied at the Lampeter grassland site (P < 0.05), but a decrease in EAWC where food-based digestate had been applied at Ayr (P < 0.05).

FIGURE 4 | Change in potentially mineralisable nitrogen (PMN) following the repeated addition of organic materials for 3 and 9–20 years. Results are expressed as a percentage difference from the control treatment averaged over two sites with a prior history of green compost, FYM, and slurry additions, five sites with 3 years of green compost, FYM, and slurry additions and seven sites with 3 years of food-based digestate and green/food compost additions.

These differences were most likely due to changes in BD (which was used to calculate the volumetric moisture content). There were no treatment effects on the volumetric moisture contents at the arable sites (P values ranged from 0.13 to 0.95, except at Aberdeen where P = 0.06 for the 2 bar measurement).

### DISCUSSION

The results from this multi-site field study have provided further evidence of the beneficial effects to soil quality and health of recycling different organic materials to agricultural land. Some soil properties such as nutrient status (N, P, K, Mg), responded to all organic material additions (both solid and liquid) within a short timescale (<3 years), but other properties, such as total SOC, microbial biomass, and selected soil physical properties only changed to a statistically significant extent after multiple applications (9 or more years) of bulky organic materials (compost and FYM).

Given the central role of SOC in driving soil processes and properties (Kibblewhite et al., 2008), sustainable soil management is very much about managing SOC (Newell Price et al., 2015). Comparable increases in SOC were observed for both 9 years of green compost additions and 20 years of FYM additions. The capacity of a soil to hold OC is finite, such that after a change in management practice SOC may increase (e.g., after the introduction of regular organic material additions) or decrease (e.g., after plowing out long-term grass) toward an equilibrium (after 100 years or more) that is characteristic of the

soil type, land use and climate (Powlson et al., 2012). Annual rates of SOC accumulation (or depletion) therefore change over time and gradually decline to zero as the new equilibrium is approached. Typically, c.50% of the SOC accumulation achieved after 100 years of introducing a management change, occurs within the first 20 years (Powlson et al., 2012). It is possible that the rate of SOC accumulation on the long-term FYM treatment at Harper Adams and Terrington was entering this slower phase. However, the retention of OC supplied by the green compost was almost double that of the FYM, suggesting that the OC in green compost was in a more stable form, due to loss of labile C during the composting process. The higher lignin content of the green compost (c.70%) compared to the FYM (c.55%) supports this conclusion. The greater stability of the OC supplied by the green compost additions therefore enabled a more rapid buildup of SOC over a shorter timeframe. Retention of OC from the FYM was c.12%, which is identical to that reported by Maillard and Angers (2014) in a global meta-analysis of long-term field experiments with animal manures. Retention of compost OC (at 20–25%) was almost double that of FYM, although not as great as that reported by Bhogal et al. (2010) from four UK studies where green compost had been applied for 5–8 years and OC retention was over 40% (±8%). Given the interest in exploring potential land management strategies for increasing soil carbon storage in the mitigation of climate change, these OC retention coefficients are useful for improving national GHG inventory methodologies (Maillard and Angers, 2014) and demonstrate the value of green compost for increasing soil carbon storage.

A key result, however, was that whilst green compost was found to be a good source of stable organic C able to build-up SOC pools over a relatively short time-frame, it did not produce the same level of improvement in associated soil biological and physical functioning as a similar increase in SOC produced by FYM applications (albeit over a longer time period and with a higher OC loading—another 4–6 years of experimentation would be required in order to establish whether a similar green compost OC loading could achieve the same level of improvement in soil biological and physical functioning as achieved on the long-term FYM treatment). As a more readily decomposable C source, the SOC increase induced by the FYM applications was able to support a bigger microbial population than that produced by the green compost additions. Importantly, this also led to a proportionally greater improvement in soil physical functioning (BD, porosity, and workability) and provides field evidence of the influence of the microbial community in the development of soil structure which has hitherto predominantly been concluded from laboratory studies (e.g., Watts et al., 2001, 2005). Moreover, it demonstrates the importance of the quality (particularly the C composition) of organic inputs, as well as the quantity, in influencing soil quality. Indeed, in a laboratory study Watts et al. (2005) clearly demonstrated the involvement of the soil microbial community in soil aggregation, with the incorporation of grass residues (with a low C:N ratio) resulting in greater micro-aggregation than straw incorporation, and no aggregation occurring where charcoal (with a C:N ratio of 600) was incorporated.

The extent of decomposition of organic matter that is added to soil is one of the important factors that define the "quality" of the amendment. Composting increases the proportion of aromatic structures (Chefetz et al., 1996), which will influence the composition of the resultant soil organic matter (Spaccini et al., 2009) and the soil biological community it supports (e.g., Ros et al., 2006). For example, Lucas et al. (2014), in a laboratory incubation study, observed that the formation of large macroaggregates was highest in soils amended with vetch, followed by livestock manure, with green compost either having no effect or reducing aggregate formation relative to a non-amended control. This was related to the amount of microbially available C supplied by the different materials and the composition of soil microbial community they supported, with the vetch and manure providing greater amounts of labile C and a shift toward fungal-dominated soil microbial communities. In a 9 year field experiment, Annabi et al. (2011) observed that the addition of three different composts (municipal solid waste, sewage sludge/green waste, and a biowaste compost) had a larger positive effect on aggregate stability than FYM. Compost derived from municipal solid waste was the most efficient in improving aggregate stability in the first 6 years of the experiment due to a larger labile organic C content stimulating soil microbial activity. However, after the first 6 years, the two other, more stable composts, became more efficient, linked to a greater increase in soil organic C contents.

The "quality" of the organic materials applied not only affects the soil microbial community response, but has also has implications for the whole of the soil food web. For example, Leroy et al. (2008) measured significant differences in earthworm populations following repeated organic material additions (four applications over c.2 years), with FYM and cattle slurry having the highest earthworm abundance, compost intermediate, and an un-amended control having the lowest earthworm numbers. Stroud et al. (2016) measured a similar effect when comparing 3 years of FYM and compost additions, with the compost having no impact on the abundance of Lumbricus terrestris earthworms, whereas FYM increased abundance by c.38%. Rates of OC addition were the same for all treatments in these experiments, with the different earthworm abundance attributed to differences in the nutritional value of the organic materials.

Application of organic materials with a low dry matter content (digestate and livestock slurries) produced few measurable changes in soil properties in the short-term. Indeed, these materials are typically applied to recycle nutrients to the soil/crop system and reduce the need for manufactured fertilizer, rather than to improve SOC levels and overall soil quality. In fact, Coban et al. (2015) observed that the application of digestate (derived from livestock manures) caused a priming effect resulting in the mineralization of native soil organic matter and concluded that intensive and repeated application of such materials "should be avoided" due to the potential to decrease SOC. However, SOC levels were not (P > 0.05) affected by the short-term (<3 years) application of digestate and livestock slurry in this study. Moreover, in the long-term (i.e., up to 20 years) repeated livestock slurry additions increased SOC and soil biological and physical functioning, although not to the same level as comparable applications of FYM. It is therefore possible that repeated digestate applications over a similar timeframe could lead to similar improvements. Nkoa (2014) reviewed evidence from a number of studies which suggested that in the majority of cases, the short-term effects of digestate application resulted in an improvement in soil quality (microbial biomass, N and P contents), with one study reporting a reduction in bulk density and increase in soil moisture retention (Garg et al., 2005).

At the grassland sites, compost and FYM additions decreased bulk density, but there was evidence of soil compaction (i.e., increased bulk density) where digestate and livestock slurry had been applied for 3 years. Soil compaction is often observed where livestock slurries have been applied due to heavy trafficking by the tanker during application, particularly if conducted under wet conditions. However on almost all occasions, all the organic materials (including the livestock slurries and digestates) were applied by hand, so it is unlikely that soil compaction occurred as a result. It is possible that the volume, viscosity and conductivity of the liquids applied may have caused partial break-down (slaking) of the surface soil aggregates, leading to a decrease in porosity and increase in bulk density. However, this has not been widely reported as a problem with slurry applications to grassland and further experimentation would be required in order to elucidate the reasons behind the observed increases in bulk density.

# CONCLUSIONS

The results provide robust evidence of the soil quality benefits of recycling organic materials to agricultural land. However, the level and nature of benefit depends on both the quantity (carbon loading) and quality (decomposability) of the organic material applied. Most organic materials are valuable sources of plant nutrients, enabling a reduction in manufactured fertilizer use. However, significant improvements in soil biological and physical functioning appear to be dependent on supplying sufficient OC that is biologically available (e.g., materials with a low C:N and concentration of aromatic compounds). In this study, this was achieved through repeat addition of FYM. Where more rapid increases in SOC are required, to increase soil carbon storage for example, then materials which are more resistant to decomposition, such as composts, would be more beneficial.

# AUTHOR CONTRIBUTIONS

AB led the research project from which this paper was written, supported by FN. AR undertook much of the data analysis together with MW who performed the multi-predictor modeling. AL managed the Scottish experimental sites. JW and MT oversaw the whole DC-Agri experimental programme which this work contributed to.

# ACKNOWLEDGMENTS

This work was commissioned by WRAP and Zero Waste Scotland and funded by Defra, Scottish, and Welsh Government<sup>2</sup> . The authors wish to acknowledge the large number of people at ADAS, Harper Adams University and SRUC who have made an invaluable contribution to this project.

<sup>2</sup>www.wrap.org.uk/dc-agri.

Diacono, M., and Montemurro, F. (2010). Long-term effects of organic amendments on soil fertility. a review. Agron. Sustain. Dev. 30, 401–422. doi: 10.1051/agro/2009040

Defra (2010). The Fertiliser Manual (RB209). Norwich: The Stationary Office. DECC and Defra (2011). Anaerobic Digestion Strategy and Action Plan. Defra, London. Available online at: http://www.defra.gov.uk/publications/files/


of exogenous organic matter Pedobiologia 52, 139–150. doi: 10.1111/ejss. 12025


REFERENCES

doi: 10.1016/0038-0717(89)90117-X

382–389. doi: 10.1016/j.agee.2011.07.005

276–286. doi: 10.1111/j.1365-2389.2008.01105.x

doi: 10.1111/j.1365-2389.2010.01319.x

doi: 10.1016/0038-0717(85)90144-0

J. Environ. Qual. 25, 776–785.

doi: 10.1016/j.soilbio.2015.03.023

anaerobic-digestion-strat-action-plan.pdf

pas-110-producing-quality-anaerobic-digestate

uk/Prc790.HTM

Anderson, T. H., and Domsch, K. H. (1989). Ratios of microbial biomass carbon to total organic carbon in arable soils. Soil Biol. Biochem. 21, 471–479.

Annabi, M., Le Bissonnais, Y., Le Villio-Poitrenaud, M., and Houot, S. (2011). Improvement of soil aggregate stability by repeated applications of organic amendments to a cultivated silty loam soil. Agric. Ecosyst. Environ. 144,

Bhogal, A., Chambers, B. J., Whitmore, A. P., and Powlson, D. S. (2010). Effects of Reduced Tillage Practices and Organic Material Additions on the Carbon Content of Arable Soils. Final report to Defra for project SP0561. Available online at: randd.defra.gov.uk/Document.aspx?Document=SP0561\_6892\_FRP.doc Bhogal, A., Nicholson, F. A., and Chambers, B. J. (2009). Organic carbon additions: effects on soil bio-physical and physico-chemical properties. Eur. J. Soil Sci. 60,

Bhogal, A., Nicholson, F. A., Young, I., Sturrock, C., Whitmore, A. P., and Chambers, B. J. (2011). Effects of recent and accumulated livestock manure carbon additions on soil fertility and quality. Eur. J. Soil Sci. 62, 174–181.

Bhogal, A., Nicholson, F., Taylor, M., Rollett, A., and Williams, J. (2016). Crop available nitrogen supply from food-based digestate. Int. Fertil. Soc. Proc. 790, 1–15. Available online at: http://fertiliser-society.org/proceedings/

Brookes, P. C., Landman, A., Pruden, G., and Jenkinson, D. S. (1985). Chloroform fumigation and the release of soil nitrogen: a rapid direct extraction method to measure microbial biomass nitrogen in soil. Soil Biol. Biochem. 17, 837–842.

BSI (2011). PAS 100:2011 Specification for Composted Materials. Available online at: http://www.wrap.org.uk/content/bsi-pas-100-compost-specification-1 BSI (2014). PAS 110:2014 Specification for Whole Digestate, Separated Liquor and Separated Fibre Derived from the Anaerobic Digestion of source Segregated Biodegradable Materials. Available online at: www.wrap.org.uk/content/bsi-

Chefetz, B., Hatcher, P. G., Hadar, Y., and Chen, Y. (1996). Chemical and biological characterization of organic matter during composting of municipal solid waste.

Coban, H., Miltner, A., Elling, F. J., Hinrichs, K. U., and Kastner, M. (2015). The contribution of biogas residues to soil organic matter formation and CO<sup>2</sup> emissions in an arable soil. Soil Biol. Biochem. 86, 108–115.


Wu, J., Joergensen, R. G., Pommerening, B., Chaussod, R., and Brookes, P. C. (1990). Measurement of soil microbial biomass C by fumigationextraction – an automated procedure. Soil Biol. Biochem. 22, 1167–1169. doi: 10.1016/0038-0717(90)90046-3

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Bhogal, Nicholson, Rollett, Taylor, Litterick, Whittingham and Williams. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.