# CANOPIES IN AQUATIC ECOSYSTEMS: INTEGRATING FORM, FUNCTION, AND BIOPHYSICAL PROCESSES

EDITED BY : Virginia B. Pasour, Brian L. White, Marco Ghisalberti, Matthew Philip Adams, Matthew H. Long, Matthew A. Reidenbach, Uri Shavit and Julia E. Samson PUBLISHED IN : Frontiers in Marine Science

#### 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 978-2-88963-340-1 DOI 10.3389/978-2-88963-340-1

#### 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

# CANOPIES IN AQUATIC ECOSYSTEMS: INTEGRATING FORM, FUNCTION, AND BIOPHYSICAL PROCESSES

Topic Editors:

Virginia B. Pasour, Army Research Office, United States Brian L. White, University of North Carolina at Chapel Hill, United States Marco Ghisalberti, University of Western Australia, Australia Matthew Philip Adams, University of Queensland, Australia Matthew H. Long, Woods Hole Oceanographic Institution, United States Matthew A. Reidenbach, University of Virginia, United States Uri Shavit, Technion Israel Institute of Technology, Israel Julia E. Samson, Consultant, Amsterdam, Netherlands

Citation: Pasour, V. B., White, B. L., Ghisalberti, M., Adams, M. P., Long, M. H., Reidenbach, M. A., Shavit, U., Samson, J. E., eds. (2020). Canopies in Aquatic Ecosystems: Integrating Form, Function, and Biophysical Processes. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-340-1

# Table of Contents

*04 Editorial: Canopies in Aquatic Ecosystems: Integrating Form, Function, and Biophysical Processes*

Julia E. Samson, Marco Ghisalberti, Matthew Philip Adams, Matthew A. Reidenbach, Matthew H. Long, Uri Shavit and Virginia B. Pasour


Arnold van Rooijen, Ryan Lowe, Marco Ghisalberti, Mario Conde-Frias and Liming Tan

*64 Canopy-Mediated Hydrodynamics Contributes to Greater Allelic Richness in Seeds Produced Higher in Meadows of the Coastal Eelgrass* Zostera marina

Elizabeth Follett, Cynthia G. Hays and Heidi Nepf

*78 Bridging the Separation Between Studies of the Biophysics of Natural and Built Marine Canopies*

Craig Stevens and David Plew


Mark S. Fonseca, James W. Fourqurean and M. A. R. Koehl

# Editorial: Canopies in Aquatic Ecosystems: Integrating Form, Function, and Biophysical Processes

Julia E. Samson<sup>1</sup> \*, Marco Ghisalberti <sup>2</sup> , Matthew Philip Adams <sup>3</sup> , Matthew A. Reidenbach<sup>4</sup> , Matthew H. Long<sup>5</sup> , Uri Shavit <sup>6</sup> and Virginia B. Pasour <sup>7</sup>

*<sup>1</sup> Department of Biology, University of North Carolina at Chapel Hill, Chapel Hill, NC, United States, <sup>2</sup> Oceans Graduate School, University of Western Australia, Perth, WA, Australia, <sup>3</sup> School of Earth and Environmental Sciences, School of Biological Sciences, and School of Chemical Engineering, University of Queensland, Brisbane, QLD, Australia, <sup>4</sup> Department of Environmental Sciences, University of Virginia, Charlottesville, VA, United States, <sup>5</sup> Department of Marine Chemistry & Geochemistry, Woods Hole Oceanographic Institution, Woods Hole, MA, United States, <sup>6</sup> Department of Civil and Environmental Engineering, Technion Israel Institute of Technology, Haifa, Israel, <sup>7</sup> Army Research Office, Durham, NC, United States*

Keywords: fluid dynamics, ecosystem engineering, coral, algae, canopy, mass transport, light availability, nutrient cycling

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

#### **Canopies in Aquatic Ecosystems: Integrating Form, Function, and Biophysical Processes**

This Research Topic presents new research investigating the coupling between physical (fluid dynamics, mass transport, and light availability) and biological (nutrient cycling, particle transport, ecosystem structure, and biodiversity) processes in aquatic canopies. The starting point for this topic was the observation that our notion of "canopy" in the aquatic sciences, in contrast to that of our terrestrially-focused colleagues, remains underdeveloped. Forest canopy studies have been considered a new field of science (Nadkarni et al., 2011) and the concept of forest canopy research is clearly documented in the literature (Barker and Pinard, 2001; Nadkarni, 2001; Lowman, 2009); we have not found similar mentions of the canopy concept in aquatic studies. Over the past decade, however, there has been an increase in the number of studies on underwater canopies, as well as a shift toward more multidisciplinary studies that consider more than just the physical impacts of the canopy's presence (Ackerman, 2007; Nepf et al., 2007; O'Brien et al., 2014).

Through this Research Topic, we provide a platform to explore the various physical and ecological impacts of aquatic canopies on the broader environment. We considered a fairly broad definition of canopy and did not restrict the concept to macroscale algae and corals. Any biological or physical entity displaying canopy-like characteristics (notably resistance to flow in the water column) is of interest for understanding canopy impacts. Additionally, we acknowledge that underwater canopies are not usually static structures but display dynamic behavior and can change over time and space.

An important goal of this Research Topic was to start integrating different (methodological) approaches and discipline-specific viewpoints to develop a more holistic view of how canopies shape their ecosystems. Oftentimes, studies have focused on a single aspect of the canopy, creating a one-dimensional view of its function in a given ecosystem, for example as a flow regulator (Nepf and Vivoni, 2000; Ghisalberti and Nepf, 2009) or as a photosynthetic structure (Binzer et al., 2006). Understanding the strong and inherent coupling between a canopy's physical and biological impacts, however, would provide much more insight into the importance and function of canopies in aquatic ecosystems.

The manuscripts we received were diverse in the topics they treated as well as their aims and approaches. Several papers in our collection investigated canopies from a mechanistic point of view, looking at the effects of canopy structure on flow and the resulting ecosystem impacts.

#### Edited and reviewed by:

*Angel Borja, Technological Center Expert in Marine and Food Innovation (AZTI), Spain*

> \*Correspondence: *Julia E. Samson juliasamson@mac.com*

#### Specialty section:

*This article was submitted to Marine Ecosystem Ecology, a section of the journal Frontiers in Marine Science*

Received: *28 September 2019* Accepted: *29 October 2019* Published: *15 November 2019*

#### Citation:

*Samson JE, Ghisalberti M, Adams MP, Reidenbach MA, Long MH, Shavit U and Pasour VB (2019) Editorial: Canopies in Aquatic Ecosystems: Integrating Form, Function, and Biophysical Processes. Front. Mar. Sci. 6:697. doi: 10.3389/fmars.2019.00697* Starting at the sub-meter scale, van Rooijen et al.'s work on predicting drag forces in canopies offers a detailed understanding of canopy-flow interactions. The authors provide a robust tool to quantify canopy flow resistance across a range of canopy types (emergent or submerged, rigid or flexible). Their model will prove to be useful in further studies requiring accurate drag quantification in canopy environments, for example when studying reduced in-canopy flow environments or measuring the impact of the canopy on sedimentation.

At the canopy level, Fonseca et al. considered the interactions between canopy-forming organisms and their environment (both biotic and abiotic factors). They examined the importance of shoot flexibility and shoot density in seagrass beds that are exposed to flow and how these parameters influence hydrodynamics, turbulence, sedimentation, and light penetration within the seagrass bed.

Moving up from the seagrass bed to the meadow scale, Reidenbach and Thomas show that seagrass canopies exert significant control over both wave height and hydrodynamic conditions at the sediment-water interface. Their findings suggest that the role of seagrass canopies in sedimentation and the (re-)suspension of sediment particles in the water column is not confined to the seagrass bed, but extends beyond and above it, impacting the ecosystem more generally.

A couple of papers in the Research Topic looked directly at the functional effects a canopy can have at the ecosystem level. An important impact of the canopy that was highlighted is the influence of algal canopies on local recruitment. As shown by Umanzor et al., low canopy densities favor the recruitment of more seaweeds whereas high canopy densities displayed a higher abundance of microphytobenthic (benthic diatoms and cyanobacteria) recruits. They conclude that smallscale biophysical interactions linked to seaweed morphologies and densities can have profound effects on the recruitment and settlement of new primary producers. These interactions are often overlooked but can have significant consequences on the dynamics of the overall ecosystem.

Shifting ecosystem dynamics have been observed in the seagrass beds of the Chesapeake Bay, where one seagrass species, Zostera marina, is being replaced in some locations by Ruppia maritima. French and Moore investigated how seagrass species, biomass, and density affected invertebrate communities and sediment properties. They found correlations between seagrass species and sediment coarseness, shoot density and invertebrate biodiversity, and between seagrass biomass and both invertebrate biodiversity and abundance. Although seagrass species might not directly influence which invertebrate species are found, changes in sediment coarseness and seagrass biomass could well-affect the fauna abundance as well as the physical conditions under which they thrive.

Other papers highlighted the importance of going beyond the existing boundaries between research communities. In their perspective article, Stevens and Plew call for more connection and exchange between biophysicists focusing on natural marine canopies and those concentrating on "built" canopies (i.e., suspended aquaculture canopies) commonly used in (shell)fish farms. Though their purposes might differ (answering ecological vs. economic questions), both groups of researchers would benefit from learning more about each other's approaches and insights.

Taking a higher-level view, Folkard's comprehensive review provides guidelines for future exploration (including a request for physicists and ecologists to move toward each other in terms of methodology, reminding us of Stevens and Plew's call for more connection between research communities) and urges researchers to make the leap to the landscape-scale. Putting biophysical processes happening in aquatic canopies back in their landscape-wide context is crucial to support and inform management and conservation efforts since most of them take place at this scale.

In fact, a few papers in this collection have already taken up this call. Follett et al.'s contribution shows how seagrass bed parameters such as shoot density affect local hydrodynamics, which in turn affect pollen dispersion in the bed, and thus genetic variation in offspring (seeds) based on location height within the canopy. This paper also illustrates how modeling and experimental/field approaches can complement each other and lead to a more robust understanding of canopy systems.

Finally, on the largest scale, Ørberg et al. investigated the role of canopy-forming algae in the subarctic intertidal. Ascophyllum seaweeds were shown to facilitate higher species richness and recolonization by increasing habitat surface and complexity and modifying environmental stressors such as extreme temperature or desiccation. In the context of climate change, Ascophyllum nodosum's distribution range is expected to shift northwards, thus promoting the northward colonization of intertidal fauna in the Arctic.

From microscale hydrodynamic forces affecting sedimentation to allelic variation, invertebrate biodiversity, and the colonization of new habitats, the many impacts of aquatic canopies on the broader environment constitute a burgeoning area of research. Our understanding of what these canopies are, how they function, and how they influence entire ecosystems is rapidly expanding. Cross-disciplinary initiatives, including those presented in this topic, will continue feeding this momentum and lead us to new and important insights.

# AUTHOR CONTRIBUTIONS

JS, MG, MA, MR, ML, and VP contributed input for the editorial. JS wrote the editorial. MG and MA gave extensive feedback on the draft. MG, MA, MR, ML, US, and VP gave feedback on the finalized version.

# FUNDING

MA acknowledges funding support from Australian Research Council (ARC) Linkage Grant LP160100496 and the National Environmental Science Programme (NESP) Tropical Water Quality Hub. Funding to MR provided by the National Science Foundation (DEB-1237733 and DEB-1832221) and by a NSF CAREER grant (OCE-1151314). ML was supported by NSF OCE grant 1633951.

# REFERENCES


**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 Samson, Ghisalberti, Adams, Reidenbach, Long, Shavit and Pasour. 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.

# Intertidal Seaweeds Modulate a Contrasting Response in Understory Seaweed and Microphytobenthic Early Recruitment

#### Schery Umanzor 1†, Lydia Ladah<sup>1</sup> \* and José A. Zertuche-González <sup>2</sup>

<sup>1</sup> Department of Biological Oceanography, CICESE, Ensenada, Mexico, <sup>2</sup> Departamento de Oceanografía Biológica, Instituto de Investigaciones Oceanológicas, Universidad Autónoma de Baja California, Ensenada, Mexico

#### Edited by:

Marco Ghisalberti, University of Western Australia, Australia

#### Reviewed by:

Anna R. Armitage, Texas A&M University at Galveston, United States David Ian Taylor, Cawthron Institute, New Zealand

> \*Correspondence: Lydia Ladah lladah@cicese.mx

#### †Present Address:

Schery Umanzor, Department of Ecology and Evolutionary Biology, University of Connecticut, Stamford, CT, United States

#### Specialty section:

This article was submitted to Marine Ecosystem Ecology, a section of the journal Frontiers in Marine Science

Received: 05 February 2018 Accepted: 03 August 2018 Published: 03 September 2018

#### Citation:

Umanzor S, Ladah L and Zertuche-González JA (2018) Intertidal Seaweeds Modulate a Contrasting Response in Understory Seaweed and Microphytobenthic Early Recruitment. Front. Mar. Sci. 5:296. doi: 10.3389/fmars.2018.00296 Recruitment is a fundamental step upon which all subsequent interactions within a community occur. We explored how the attenuation of physical conditions by seaweed plots comprised of either Chondracanthus canaliculatus, Pyropia perforata, Sylvetia compressa or a mixed aggregation, at varying densities (average 1,199, 816, and 408 in. m−<sup>2</sup> ), affected recruitment of seaweeds and microphytobenthic organisms in the understory, and if physical factors modulate their abundance and distribution. We outplanted macroscopic seaweeds in the intertidal and measured changes in understory irradiance, particle retention, and bulk water flow. Both factors influenced physical conditions below the canopy. However, only canopy density had a significant effect on recruitment. The low-density canopy treatments had a greater abundance of seaweed recruits, with the opposite found for microphytobenthic organisms. The recruitment processes of seaweeds and microphytobenthic organisms, however, appeared to be independent of each other and were not due to competition. We conclude that it is crucial to consider microscale biological interactions, which are rarely addressed when assessing recruitment processes of benthic primary producers.

Keywords: bioengineers, rocky intertidal, seaweed spores, sporophytes, understory settlement

# INTRODUCTION

Abiotic factors can affect both the distribution and abundance of organisms, thus modulating community structure (Crain and Bertness, 2006). In a recent contribution, Umanzor et al. (2017) explored how different seaweed aggregations influenced the abundance and distribution of understory microphytobenthic (MPB) organisms (benthic diatoms and cyanobacteria) on an exposed rocky intertidal. Results showed that the settlement of microphytobenthic organisms was modulated by the interaction between the species composition and the density of the seaweed aggregations, with branched morphologies at higher densities having higher particle retention and greater abundance of MPB organisms underneath their canopies. Authors also reported recruitment (here defined as early post- settlement sensu Vadas et al., 1992) of seaweeds, although there was no further analysis on recruitment patterns of seaweed spores across the treatments. Therefore, in this contribution, we repeated the experiment assessing the effect that macroscopic seaweeds as ecosystem engineers had on the recruitment of the microscopic stages of seaweeds (spores, gametophytes, and early sporophytes). We then evaluated the factors that could be affecting the patterns of distribution and abundance of seaweed recruits underneath manipulated canopies in the intertidal compared to MPB settlement.

Positive interactions and stress attenuation can play an important role in determining the establishment and survival of seaweed recruits (Bertness et al., 1999; Choi and Norton, 2005; Bennett and Wernberg, 2014). Overall, recruitment is the key process upon which all subsequent interactions within a community will occur. As such, variations in successful recruitment events can substantially influence the dynamics of adult populations (Woodin, 1991; Vadas et al., 1992). Evidence shows that there are multiple factors, both inherent and external to the species, that can influence successful settlement and recruitment of early stages. Inherent factors can include the number of propagules produced, growth rates and size of settling cells, germination and spore viability, and even the strength of adhesion by early propagules (Vadas et al., 1992).

On the other hand, external factors such as particle movement resulting in sedimentation, siltation, scour or increased turbidity might affect early settlers by either preventing or enhancing their survival. On the Great Barrier Reef, for example, increased sedimentation has significantly decreased the rates of recruitment, survival, growth, and regeneration of Sargassum sp. (Umar et al., 1998). Also, in a laboratory experiment, Watanabe et al. (2016) found that the adhesion rate of spores and the gametophyte survival and growth rates of Eisenia sp. (now Ecklonia) declined noticeably with increasing sedimentation rates. Particle movement resulting in complete burial or sand scour can, however, also have a positive outcome for some algal species such as Rhodomela, Penicillus, and Halimeda, allowing their colonization of areas where other species would not thrive (Hurd et al., 2014).

Moreover, substrate properties such as its topography and stability can also greatly enhance or reduce successful seaweed recruitment. In a controlled experiment, Callow et al. (2002) tested how varying microtopographies affected the settlement of Enteromorpha sp. spores. They found that lower profile topographies significantly reduced the abundance of settled spores. Contrarily, Schumacher et al. (2007) found that smooth surfaces enhanced spore settlement of Ulva sp. In fact, Linskens (1966) reported that algae propagules will either settle on smooth or rugose surfaces, depending on the species. Coupled with the substrate properties, water motion is another factor that has long been studied as critical in influencing settlement and survival of seaweed propagules (Vadas et al., 1992). Seaweed zygotes use a range of adhesive mechanisms for attachment, allowing them to either thrive in low or high wave-energy environments. Through field and laboratory experiments, Taylor and Schiel (2003) demonstrated that the "stickability" of Durvillaea antarctica zygotes allowed the species to attach immediately and firmly to surfaces exposed to different wave regimes, resulting in high rates of survival when compared to zygotes of Hormosira banksii and Cystophora torulosa.

For intertidal seaweeds, desiccation stress due to aerial exposure can also cause increased mortality of early stages. Brawley and Johnson (1991) showed that without the protection against water loss provided by parental canopies, a large percentage of seaweed early settlers would inevitably die. However, pre-existing canopies can also represent a stressful biotic force, which can also potentially influence recruitment success. Canopies can cause shading, sweep propagules away or prevent settlement entirely, outcompete them for space and nutrients, or cause chemical interferences (Sousa, 1979; McCourt, 1984; Brawley and Johnson, 1991), that in the short term will trigger high mortality rates of potential new settlers. Evidence for canopy inhibition of early-post settlement or recruitment has been recorded in succession and reproductive ecology studies in the intertidal zone, with overall recruitment increasing due to the attrition of canopy cover (Sousa, 1979; Robertson, 1987).

Together, seaweed aggregations and microphytobenthic biofilms can interact directly or indirectly by modifying biophysical parameters (Fong et al., 1993; Hardison et al., 2013), and thus a influence the recruitment of associated organisms. For example, Hardison et al. (2013) measured the independent and interactive effect that both the MPB and benthic macroalgae can have on the quality and quantity of sediment organic matter (SOM). They concluded that while the MPB increased the SOM liability, benthic macroalgae tend to decrease it. They also found that both groups influenced bacterial buildup that could have a further effect on hypoxia events, sulfide accumulation, mineralization or denitrification of shallow water systems. Bacterial build up can also be important in determining the abundance and distribution of a variety of organisms, as for many benthic invertebrates, larval settlement occurs in response to bacterial cues (Freckelton et al., 2017).

Despite the many contributions related to seaweed recruitment in the intertidal zone, few studies have simultaneously explored the recruitment and development of seaweeds and the MPB. In part, this could be attributed to the difficulty in obtaining in situ measurements from organisms of such small size, but also could be due to the complexity of characterizing the microenvironment they inhabit. However, because seaweed recruits and MPB colonize similar areas, we expect the abundance of their early stages to be limited or enhanced by the same physical factors. Consequently, we tested the following hypotheses: (1) seaweed recruitment shows a similar abundance and distribution underneath intertidal canopies of different species and densities and (2) seaweed recruitment follows the same pattern of distribution as the microphytobenthos underneath seaweed canopies. We constructed experimental quadrats consisting of Pyropia perforata (Agardh, 1883), Silvetia compressa (Agardh, 1848), Chondracanthus canaliculatus (Harvey, 1840), and a mixed assemblage comprised of the former three, at three densities. We then determined if seaweed recruitment was correlated to the attenuation of bulk water flow, particle transport, and irradiance driven by the experimental canopies.

### MATERIALS AND METHODS

#### Study Site

Experiments were conducted for a 15-day period on a rocky shore in Baja California (31◦ 51′ 41.6′′ N and 116◦ 39′ 58.1′′ W) during spring (2016) when the selected seaweeds (C. canaliculatus, Silvetia compressa, and P. perforata) were abundant. These species were selected because they are among the most common species in local intertidal sites, often forming dense beds or patches. This area has a semidiurnal tidal cycle with two low tides and two high tides of different heights per day (Umanzor

TABLE 1 | Plaster bar erosion, irradiance, particle retention, and abundance of microphytobenthic organisms and microscopic stages of seaweeds based on species composition, density, and their interaction, using a two-factor crossed ANOVA.

fragments of the selected seaweeds.


et al., 2017). The experimental quadrats were fixed in the midintertidal zone, where seaweeds are generally abundant, but in a clear section allowing us to minimize effects caused by natural stands of seaweed. All quadrats remained submerged during high tides and the less extreme low tides.

#### Targeted Seaweeds

Chondracanthus canaliculatus is a corticated red alga, abundant from the mid to low rocky intertidal zones in Baja California. Although the species is perennial, it is particularly abundant during spring and summer when growth peaks (García-Lepe et al., 1997). The fucoid Silvetia compressa is a leathery brown alga, abundant in the upper and mid intertidal. It is a perennial species, which produces eggs ranging from 80 to 100µm, showing peak recruitment periods throughout the year (Johnson and Brawley, 1998). Finally, P. perforata is a foliose red algae, abundant in the upper and mid intertidal, with peak abundance during spring and summer (Zertuche-González et al., 2000). In Baja California, the macroscopic stages of P. perforata can persist throughout the year. Both red algae can produce spores larger than 20µM (Knight and Nelson, 1999; Maggs and Callow, 2003; Avila et al., 2011). At the time of the experiment, at least 50% of all blades were reproductive.

#### Experimental Design

To determine the effect that different canopies had on the recruitment of seaweeds compared to recruitment of MPB cells, we installed experimental quadrats in the intertidal zone. Quadrats consisted of monocultures of the selected species or a mixed tri-culture (including all three species) at three densities each. We assembled each quadrat using seaweed fragments of approximately 10 cm in lenght collected from the intertidal zone. Densities were assigned based on the maximum density of these species per square meter on a nearby rocky shore, quantified in situ before sample collection. In addition, control treatments with no algae were included. Quadrats were checked daily, and seaweeds showing any damage were immediately replaced by fresh ones to maintain the appropriate density for each treatment.

As each treatment included three replicates, the design comprised 48 quadrats total (**Figure 1**), following an orthogonal approach (Underwood, 1997). Quadrats consisted of 30 x 30 cm steel frames covered with plastic netting to which seaweeds and data collection devices were attached. Quadrats were individually secured to the intertidal using 2.5 kg lead weights. Settlement slides, particle collectors, light meters, and plaster cylinders were fixed underneath the canopy in each quadrat.

#### Physical Variable Sampling

Relative measurements of light levels were obtained using light meters (ONSETTM computer corp., Ma, USA) attached below the canopies and programmed to record every 15 min for a 7-day period. We calculated particle retention below the canopies by dry weight differences of two synthetic fiber pads (25 × 75 mm, initial weight 1.362 ± 0.003 g) per quadrat. Pads were collected after a 48 h period underneath the canopies. After collection, fiber pads were oven dried at 70◦C for 60 h and then weighed three times (Sartorius, Germany ± 0.0001 g) to obtain the average weight per day per quadrat. A proxy measure of the relative bulk water flow underneath the canopies was acquired using the dissolution of plaster. Two cylindrical plaster bars (1 × 8 cm, initial weight 10.422 ± 0.005) were installed per quadrat and subsequently removed after 48 h. The bars were then oven dried at 70◦C for 72 h before weighing them three times to obtain an average per quadrat. The difference in dry weight before and after deployment allows a relative estimate of bulk water flow based on the dissolution of plaster in a given area over a standardized time when compared to a control with no water motion (Komatsu and Kawai, 1992).

# Microscopic Seaweed and Microphytobenthic Recruitment

To assess recruitment by seaweed microscopic stages and the microphytobenthos, a transparent polycarbonate slide (25 × 75 × 3 mm) was fixed underneath every canopy treatment and collected after a 15-day period. After collection, slides were placed individually in Petri dishes containing filtered (1µM) seawater and immediately fixed with Lugol's solution (1%) for direct cell counting at 400x with an inverted microscope (Zeiss AxioObserver, Germany). We divided each slide into 10 equally sized sections from which a photograph was taken. We used all photographs for recruitment quantification and identification. For seaweed recruitment, we considered spores, gametophytes, and early sporophytes, regardless of size. When possible, we further classified them as red, brown or green. For the MPB, we only considered cells bigger than 20µm because we could not photograph smaller cells with enough detail to ensure their correct identification.

# Data Analysis

Density (high, medium, low, and control) and species composition (S. compressa, C. canaliculatus, P. perforata, and mixed culture) were considered categorical and independent factors. Natural log transformations were conducted as required to satisfy the assumptions (Underwood, 1997). Normality (Shapiro-Wilk test), independence of variables (Durvin-Watson test) and homogeneity of variances (Cochran's test) were confirmed per factor and level.

The iterative effect of the two categorical factors on bulk water flow, particle retention, irradiance, and seaweed and MPB recruitment was used in an ANOVA by least mean squares at an alpha value of 0.05. Post-hoc (Tukey test) comparisons were conducted where differences were found. Also, simple and multiple regressions were performed to identify which environmental factor or combined factors resulted in significant


predictors of the abundance of both groups and to determine if any correlation existed between them. Outputs and raw data of the variables measured in this manuscript are available through the figshare repository (http://figshare.com), doi: 10.6084/m9. figshare.5797137.

#### RESULTS

Overall, the differences in the ability of the canopies to attenuate intertidal physical conditions seemed to influence the abundance of seaweed recruitment underneath the canopies directly. In general, higher abundances of seaweed recruits occurred underneath canopy treatments with the least attenuated physical conditions, whereas the MPB showed the lowest abundance under these conditions.

Algal species and canopy density influenced the measured physical parameters below the canopies. There was no significant effect of species composition on water bulk flow. There was, however, a significant effect of density on water flow (p < 0.001; **Table 1**). Plaster bars underneath the high and medium density treatments had significantly lower dissolution than the low density and control treatments (Tukey p < 0.001), suggesting there was significantly less bulk water flow underneath the canopies of higher densities (**Table 2**).

Furthermore, there was no significant interaction effect between species and density on the attenuation of irradiance below the canopies, yet there was an effect driven separately by each factor (p < 0.001; **Table 1**). The species Silvetia compressa and overall the high and medium density treatments attenuated irradiance the most (Tukey p < 0.05; **Table 2**). There was also a significant interaction between species composition and density on particle retention below the canopy (p < 0.05 **Table 1**). Synthetic fiber pads below the S. compressa and C. canaliculatus canopies at high densities retained more particles than other treatments (Tukey p < 0.001; **Table 2**).

On the other hand, only density treatments showed a significant effect on the abundance of seaweed and MPB recruitment underneath the canopies (p < 0.001, **Table 1**). However, both groups had contrasting distributions. Diatoms were abundant underneath higher density treatments (**Figure 2**), while seaweed recruits were abundant underneath lower density treatments (**Figure 3**).

Early sporophytes of brown seaweeds were the most abundant seaweed microscopic stage found and often grew close to one another. Red algal spores were second in abundance and showed a more isolated distribution, with few to no other seaweeds or microphytobenthic organisms settled next to them (**Figure 4**). Conversely, benthic diatoms were the dominant organism within the MPB with Cocconeis spp. representing 89% of the settlers and often forming biofilm mats. Fewer representatives of other benthic diatoms, such as Navicula sp. (10%), Climacosphenia sp. (<1%), and cyanobacteria, such as Chroococcus sp. (<1%), were present at lower abundances (**Figure 5**).

Surprisingly, none of the physical factors used in the regression analyses explained the variability in the abundance of seaweed recruits (**Table 3**) (r <sup>2</sup> = 0.047, p = NS). However,

FIGURE 2 | Abundance of microphytobenthic organisms settled underneath the experimental canopies. Mean values ± one standard error.

the regression analyses did show that particle retention by the seaweed canopy aggregations explained 64% of the variability in the abundance of MPB cells in the understory (n = 48, p < 0.001).

#### DISCUSSION

Overall, our results indicate that the attenuation of physical factors driven by macroscopic seaweeds influences the abundance and distribution of seaweed and MPB recruits differently. We found a greater abundance of MPB cells underneath the most attenuated canopies and a greater abundance of seaweed settlers underneath the least attenuated canopies. In contrast to the MPB, particle retention did not appear to affect seaweed recruitment significantly. Moreover,

FIGURE 5 | Microphytobenthic organisms settled underneath the canopies. (A) Cocconeis sp., (B) Navicula sp., (C) Chroococcus sp., and (D) Climacosphenia sp.

no relationship with the physical factors measured herein explained the abundance of seaweed recruits. We neither find any relationship between the distribution patterns of MPB cells and seaweed recruits. The distribution and abundance patterns might suggest competition for space, light or nutrients, due to their apparently inverse, but not significantly related distribution, however no relationship was found to evidence this. Huang and Boney (1984) experimentally demonstrated that although MPB cells can outcompete juvenile brown and red seaweeds, both groups could also coexist with no competition between them, which seems to be the case in our study.

At least 50% of the blades in our experimental quadrats were fertile during the study period, coinciding with the reproductive periods described for these species in the region (Pacheco-Ruiz et al., 1989; García-Lepe et al., 1997; Johnson and Brawley, 1998; Zertuche-González et al., 2000). Many of the germlings or spores measured in this study could have

TABLE 3 | Regression summary for the recruitment of seaweed microscopic stages (p = NS) and microphytobenthic cells (r <sup>2</sup> = 0.64, p < 0.001) and underneath the canopy.


resulted from self-seeding in our quadrats, which might be related to the relatively high abundance of brown and red seaweed microscopic stages underneath the canopies. The nonmotile eggs of the fucoid seaweeds are within the largest reproductive cells among seaweeds. Although spores released from our outplants might sink relatively faster into the water than the MPB (Okuda and Neushul, 1981), thus facilitating colonization of available substrate, many could have been lost to mortality (Santelices, 1990; Hurd et al., 2014) or could have been impeded from settling by the high canopy densities of our treatments. In fact, a number of studies have reported settlement of recruits to be inhibited by previously established seaweeds; either due to the physical barrier thalli represent (McCourt, 1984), a sweeping effect (Fletcher and Callow, 1992; Johnson and Brawley, 1998), or by particular traits of the canopies (Brawley and Johnson, 1991). C. canaliculatus, for example, can form dense mats that can trap enough sediment which prevents subsequent settlement by algal spores (Sousa, 1979; Sousa et al., 1981). Moreover, studies show that seaweed recruits also respond to the chemical properties of the microenvironment. For instance, flagellated spores can detect and respond to a variety of inorganic and organic nutrients, swimming toward or away from microhabitats with nutrients at concentrations that enhance or inhibit growth (Amsler et al., 1992; Maggs and Callow, 2003; Hurd et al., 2014). It is possible that inhibitory cues by our experimental canopies could have deterred green motile spores from settling, therefore explaining their relatively low abundance in our experimental treatments.

Early colonizers play a key role in ecology because they can modify the physicochemical conditions of their environment. In turn, they can furthermodulate settlement of other organisms (Dobretsov, 2008; Orvain et al., 2015). In fact, benthic diatoms are among the first algae to settle on new substrate, and it is suggested that they may affect the settlement of later successional organisms (Amsler et al., 1992). Research shows that together, the MPB and bacteria can form biofilms that can influence the adhesion of seaweed spores by modifying the chemical characteristics of the settlement surface (Callow and Callow, 2006; Ma et al., 2010). This effect, however, is not always consistent, as the strength of adhesion depends on the species of seaweeds tested, their stage of development and the physicochemical properties of the substrate (Mieszkin et al., 2013). Although this study was repeated a year later, similarly to Umanzor et al. (2017), we found that diatoms of the genus Cocconeis almost solely colonized all understories. This result provides further evidence of the relevance of particle retention for the settlement of benthic organisms, as particle transport and deposition significantly influence their establishment, survival, and development (Eckman and Duggins, 1991; Morrow and Carpenter, 2008). Even though we only considered diatoms and cyanobacteria within these films, bacteria were also most likely abundant.

It is possible that MPB organisms were already attached to the seaweed canopies as epiphytes, and that the transfer to the seabed occurred directly underneath the high-density treatments. In contrast to seaweed propagules, the MPB can grow exponentially and cover a large area in a relatively short time, from a few founding cells (Blanchard et al., 2001), facilitating rapid colonization of new substrate. Such developmental and colonization attributes could have resulted in the differences in the abundance and distribution of the MPB and seaweed recruits measured underneath the canopies. It is also possible that grazers could have influenced the recruitment process of both the MPB and seaweed microscopic stages. Experimental studies show that grazers and filter feeders can ingest and digest algal spores, affecting the pattern of dispersal and settlement on rocky shores (Buschmann and Santelices, 1897; Eckman and Duggins, 1991). Also, Coleman et al. (2006) showed that limpet grazers affected diversity and biomass of intertidal seaweeds across a latitudinal gradient. However, in this study, we did not witness any grazers on our settlement slides. Nonetheless, our experiment did not consider any exclusion caging, therefore, selective feeding underneath the treatment canopies could have potentially influenced the outcome of the experiment, though we do believe this to be unlikely.

We conclude that although both seaweed microscopic stages and MPB cells inhabit similar microhabitats, microscale processes appear to affect their recruitment in different ways. As shown here, the presence of macroscopic seaweeds can greatly influence the settlement and distribution of a variety of benthic organisms. Microscale changes promoted by seaweeds with different morphologies and at different densities can have profound effects on the settlement of associated organisms and these effects could trigger changes at a larger scale that should not be underestimated. Although intertidal seaweed aggregations seem to modulate a contrasting response in understory microphytobenthic and seaweed recruitment, more experimental work considering longer periods and control environments is required to determine the reciprocal effects that the MPB and seaweed microscopic stages might have on each other. Determining the identity of the recruits and if there is an effect by bacterial films and chemical cues is also recommended. Complex interactions in intertidal dynamics, the small size of the target organisms, and the characteristics of the microenvironment they inhabit all make it difficult to obtain in situ measurements of the potential feedback loops between seaweeds and microphytobenthic communities. Nonetheless, the interactions between seaweed aggregations and other benthic microorganisms need further focus as the ecological effects driven by changes in the recruitment of primary producers can have significant further consequences on the dynamics of the overall ecosystem.

#### AUTHOR CONTRIBUTIONS

SU was responsible for data collection and processing. LL contributed to the experimental design and writing of the manuscript. JZ-G contributed with intellectual inputs and data processing.

#### REFERENCES


#### FUNDING

This project was supported by CONACYT under Grant 221662 to LL and a scholarship from CONACYT (CVU 576942) to the first author. The Coastal Complexity Crew: Towards a paradigm shift for the near-shore ocean by exploring the biophysical complexity of spatial-temporal scales in coastal productivity.

#### ACKNOWLEDGMENTS

We would like to thank Dr. David Siqueiros and Alberto Galvez for their feedback in identifying and photographing the microorganisms. We also appreciate the contributions by J. Manuel Guzman and the volunteers who helped to set up the experiment in such a challenging intertidal.


microphytobenthic settlement. J. Phycol. 53, 1060–1071. doi: 10.1111/jpy. 12565


**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 Umanzor, Ladah and Zertuche-González. 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.

# Canopy-Forming Macroalgae Facilitate Recolonization of Sub-Arctic Intertidal Fauna and Reduce Temperature Extremes

Sarah B. Ørberg1,2 \*, Dorte Krause-Jensen1,2, Kim N. Mouritsen<sup>3</sup> , Birgit Olesen<sup>3</sup> , Núria Marbà<sup>4</sup> , Martin H. Larsen<sup>5</sup> , Martin E. Blicher <sup>6</sup> and Mikael K. Sejr 1,2

<sup>1</sup> Department of Bioscience, Aarhus University, Silkeborg, Denmark, <sup>2</sup> Arctic Research Centre, Aarhus University, Aarhus, Denmark, <sup>3</sup> Department of Bioscience, Aarhus University, Aarhus, Denmark, <sup>4</sup> Global Change Research Group, IMEDEA (CSIC-UIB), Institut Mediterrani d'Estudis Avançats, Esporles, Spain, <sup>5</sup> Danish Centre for Wild Salmon, Randers, Denmark, <sup>6</sup> Greenland Climate Research Centre, Greenland Institute of Natural Resources, Nuuk, Greenland

#### Edited by:

Matthew H. Long, Woods Hole Oceanographic Institution, United States

#### Reviewed by:

Stein Fredriksen, University of Oslo, Norway Jan Marcin Weslawski, Institute of Oceanology (PAN), Poland

> \*Correspondence: Sarah B. Ørberg sao@bios.au.dk

#### Specialty section:

This article was submitted to Marine Ecosystem Ecology, a section of the journal Frontiers in Marine Science

Received: 25 June 2018 Accepted: 28 August 2018 Published: 19 September 2018

#### Citation:

Ørberg SB, Krause-Jensen D, Mouritsen KN, Olesen B, Marbà N, Larsen MH, Blicher ME and Sejr MK (2018) Canopy-Forming Macroalgae Facilitate Recolonization of Sub-Arctic Intertidal Fauna and Reduce Temperature Extremes. Front. Mar. Sci. 5:332. doi: 10.3389/fmars.2018.00332 Ice can be an important structuring factor physically removing intertidal flora and fauna. At high latitudes in particular, the removal of canopy-forming algae by ice scour may be important as their canopy may serve to modify the extreme environment for marine organisms at low tide. We simulated the effect of ice scouring by manipulating the biomass of the canopy-forming algae Ascophyllum nodosum in a sub-Arctic fjord ["Full canopy," "Reduced canopy," "Bare (start)," "Bare (annual)"]. Over a three-year period, we quantified key physical parameters and the recolonization of flora and fauna to test the hypothesis that A. nodosum and rock rugosity facilitate recolonization of sub-Arctic intertidal fauna and that potential facilitation could rely on an ability of A. nodosum canopy to modify air temperature and ice scour. Finally, we estimated the recovery period of A. nodosum canopy height to pre-disturbance levels based on estimated early growth rates. We found that A. nodosum canopy facilitated higher species richness and recolonization of dominating faunal species (Littorina saxatilis, Littorina obtusata, Mytilus edulis, and Semibalanus balanoides), and also significantly reduced the high temperatures in summer and raised the low temperatures in winter. The abundance of M. edulis and A. nodosum recolonization increased significantly with rock rugosity and the recovery of A. nodosum canopy height was estimated to a minimum of 15 years. We conclude that algal canopy and rock rugosity play key roles in structuring sub-Arctic intertidal communities, likely by modifying environmental stress such as extreme temperature, desiccation, and by increasing the settling surface and the habitat complexity. As the distribution of canopy-forming algae is expected to shift northward, they may act as a key habitat facilitating a northward colonization of intertidal fauna in the Arctic. We highlight the importance of considering scales relevant to biological communities when predicting impacts of climate change on distributional patterns and community structure in the Arctic intertidal.

Keywords: biotic interactions, physical disturbance, rocky intertidal, community recovery, recruitment, Greenland

# INTRODUCTION

In intertidal ecosystems, air temperature, exerting a major control on biological processes, can be modified by a number of factors acting at scales that are relevant to biological communities (Helmuth, 1998; Helmuth et al., 2010). For example, sea ice modifies air temperature directly (Scrosati and Eckersley, 2007) and ice scouring may indirectly influence air temperature through the removal of canopy-forming algae (Gutt, 2001; Petzold et al., 2014). Algal canopies may also insulate organisms from extreme temperatures in the high intertidal as typically seen in temperate regions (Beermann et al., 2013; Watt and Scrosati, 2013a) and, thereby, influence community structure locally (Crowe et al., 2013). Ice, either in the form of sea ice or glacial ice, is a characteristic feature of high latitude coastal systems such as those found in Greenland, where export of glacial ice into the coastal ocean is increasing (Howat et al., 2007). In the Godthåbsfjord, West Greenland, for example, the loss rate of glacial ice has doubled within a decade, likely increasing the output of icebergs and thereby the risk of ice scouring in benthic communities (Motyka et al., 2017).

Here, we aim to understand the interplay of biotic and abiotic factors in structuring sub-Arctic rocky intertidal communities that can also improve predictions for climate change-induced range shifts (Gilman et al., 2010; HilleRisLambers et al., 2013). Several studies have shown the impact of canopy-forming algae on the understory community and patterns of recolonization as they alter the physical environment (Dayton, 1971; Hawkins, 1983; Jenkins et al., 1999a, 2004; Cervin et al., 2004). However, these studies are mostly restricted to the temperate intertidal as we found only one example from the sub-Arctic intertidal, mainly focusing on biotic factors (Ingólfsson and Hawkins, 2008).

The literature reports differential responses of intertidal organisms to canopy cover, also depending on the environmental stress level (McCook and Chapman, 1991; Bertness et al., 1999; Broitman et al., 2009; Crowe et al., 2013; Watt and Scrosati, 2013b). For instance, algal canopy cover enhances the survival of newly-settled barnacles only in the high intertidal zone (Dayton, 1971; Hawkins, 1983; Jenkins et al., 1999b). Moreover, species richness and diversity increase with algal canopy cover in the high and mid intertidal zone, again underlining the importance of the bioengineering effects of a canopy mainly in stressful environments (Watt and Scrosati, 2013a,b). Most likely canopies create an interplay of negative and positive interspecific interactions (Jenkins et al., 1999b; Beermann et al., 2013). As an example, barnacle recruitment may be negatively affected by whiplashes from algal fronds, but positively affected by lowered water loss and buffering of temperature, together resulting in a neutral effect of algal canopy cover on barnacles in the mid- and high intertidal (Beermann et al., 2013).

In a highly stressful environment, such as the sub-Arctic intertidal zone, the positive effects of algal canopy likely exceed the negative as suggested by the stress gradient hypothesis (Bertness and Callaway, 1994). However, we lack field studies from the sub-Arctic intertidal to support this hypothesis. In particular, the ability of algal canopies to buffer extreme air temperatures may be important in shaping high latitude intertidal communities. Variation in air temperature is a key stressor for intertidal organisms, impacting a range of biochemical and physiological processes (Helmuth, 1998; Denny and Harley, 2006). Water loss and thereby the risk of desiccation is also affected by air temperature (Helmuth, 1998), and even a few degrees temperature change can markedly impact mortality rates in the intertidal, especially for newlysettled organisms (Foster, 1971b). Ice scouring is another key stressor for intertidal organisms, and crevices in the rocky shore may, like canopy-forming algae, offer microhabitats, that shield organisms from destruction by ice scouring as well as other physical stressors (Foster, 1971b; McCook and Chapman, 1991; Walters and Wethey, 1996; Helmuth et al., 2010).

Sub-Arctic communities are considered to be shaped by largescale climate variables and physical exposure, but clearly there is a potential for small scale variation induced by canopyforming algae and rock roughness that may greatly affect the local physical regime, supporting community recovery after a disturbing event such as ice scouring. Therefore, the ability and speed of recovery of algal canopies may greatly affect the recovery process of the intertidal faunal community after ice scouring and potentially limit their northern distribution range.

Kobbefjord is a sheltered Greenlandic fjord in the sub-Arctic region, i.e., immediately south of the Arctic Circle. However, according to the AMAP definition, Kobbefjord is considered to be in the Arctic. We chose this study area as parts of this rocky intertidal are characterized by high biomass of the longlived fucoid canopy-forming alga Ascophyllum nodosum (Olsen et al., 2010), and the level of mechanical stress from sea ice is considered low. Yet, patches of the community may be in a recovering state after mechanical stress caused by scouring sea ice that form seasonally in the area. Ascophyllum nodosum has a wide geographical distribution extending to 69.7◦N on the coast of Greenland (Lüning, 1990) and the growth rate of the Greenland populations respond positively to a warming climate (Marbà et al., 2017).

Here, we present a first attempt at disentangling the multiple factors that influence small-scale variation in physical regimes experienced by sub-Arctic intertidal organisms. First, we test the hypothesis that A. nodosum canopy facilitates the recolonization of sub-Arctic intertidal fauna. We do so by quantifying faunal recolonization rates at different manipulated levels of canopy cover over a 3-year period. Secondly, we measure the temperature and ice scouring intensity experienced by the intertidal organisms at different levels of algal canopy cover. Thirdly, we consider the physical properties of the rock as a settlement surface and microhabitat during recolonization. Finally, we quantify the early growth rates of A. nodosum recruits, and attempt to estimate the recovery period to pre-disturbance canopy height after dislodgement by mechanical disturbance, such as ice scouring. The recovery period of A. nodosum canopy height is expected to be rather slow due to the colder climate as growth rates are lowered at low temperatures (Steen and Rueness, 2004; Keser et al., 2005).

quadrats from each treatment by August 2014 (the end of experimental period) [1: Full canopy, 2: Reduced canopy, 3: Bare (start), 4: Bare (annual) (Table 1)].

# MATERIALS AND METHODS

#### Study Area

The study was conducted in the sub-Arctic Kobbefjord, a branch of the Godthåbsfjord system in south-west Greenland (64◦ 08N, 51◦ 23W) (**Figure 1A**). The shoreline is largely dominated by bedrock, and the mountains surrounding Godthåbsfjord and Kobbefjord are dominated by granites and granitoid gneiss (Mosbech et al., 2000; Nutman and Friend, 2009). The fjord is 17 km long and 0.8–2 km wide with a maximum depth of 150 m. It is influenced by daily tidal amplitudes of 1–5 m (Richter et al., 2011) and sea surface temperatures ranging from −1 to 9 ◦C (Versteegh et al., 2012). Air temperature ranges from a minimum of −25◦C in winter to a maximum of 20◦C in summer, measured in Nuuk (Blicher et al., 2013). From April to October, the fjord receives freshwater run-off from several rivers in the innermost part of the fjord, resulting in a salinity gradient in the surface water. From December to May, sea ice usually covers the inner part of the fjord (Mikkelsen et al., 2008). This results in a system characterized by large seasonal variation in key physical parameters such as light, temperature, salinity, and mechanical stress (ice scouring).

# Field Experiment

The experiment was conducted in the inner part of the fjord at a rocky intertidal area covered by canopy-forming fucoid algae (predominantly A. nodosum with occasional presence of Fucus vesiculosus) and spanned a 3-year period from August 2011 to August 2014. Additional quantification of algal recolonization was conducted in August 2016. We used an experimental design with five replicate sites located along 200 m of the shoreline having similar overall vertical rock slope, similar compass direction (all S-SW facing) and evenly developed A. nodosum canopy. At each of the five replicate sites, four experimental treatments were established in 25 × 25 cm quadrats (**Table 1**, **Figures 1B,C**) ["Full canopy," "Reduced canopy," "Bare (start)" and "Bare (annual)"] and the horizontal sequence of the four treatments was fully randomized within each replicate site. All quadrats were laid out just below the mean tidal level (determined during a full tidal cycle). The slope of the rock within the resulting 20 quadrats varied between 5 and 30◦ . In "Full canopy," macroalgae were left untouched whereas fauna was removed from both canopy and rock face at the initiation of the experiment (August 2011). In "Reduced canopy," macroalgae were cut to a height of 15 cm to imitate a moderate impact of mechanical disturbance from ice scouring still allowing macroalgae to recover (Gendron et al., 2017), and fauna was removed from both the remaining canopy and the rock face at the initiation of the experiment (August 2011). In "Bare (start)," the entire quadrat was cleared at the initiation of the experiment (August 2011) for all macroalgae including their holdfasts and all fauna to imitate maximum ice scouring impact.



In "Bare (annual)," the quadrat was cleared annually in a similar way (i.e., August 2011, 2012, and 2013) in order to estimate the variation in annual settling and hence the recolonization potential. We used a metal brush for the clearing of rock surfaces and ensured that all depressions and crevices were thoroughly cleared for organisms. Macroalgae were cleared by hand for macroscopic invertebrate fauna, by working through the canopy, algal individual by individual. In a buffer zone of approximately 10 cm surrounding each quadrat, macroalgae were scraped from the rock and the canopies of algae further away were cut up to levels that prevent them from overlaying the quadrats. In order to quantify and analyze the algal and faunal community at the start of the experiment (August, 2011), referred to as the "Pre-experimental" community, all the organisms cleared from "Bare (start)" quadrats were collected and subsequently counted according to species or taxa, weighed (drained wet weight after being kept in wire mesh sieves) and measured at their maximum dimension (e.g., shell length of mussels, carapace diameter of barnacles, and height of macroalgae). The average minimum age of A. nodosum "Pre-experimental" canopy was evaluated for all individuals longer than 10 cm by counting the number of air bladders (vesicles) on the longest axis, assuming one bladder is formed annually (Åberg, 1996). This method renders a minimum age since it does not account for the age of the shoot before production the first bladder and also does not account for possible breakage of shoots. All values for minimum age, weights and lengths are given as mean (x ± SE).

In August 2012, 2013, and 2014 all fauna and macroalgae in "Bare (start)" and "Bare (annual)" were counted and measured at their maximum dimension (e.g., shell length of mussels, carapace diameter of barnacles, and height of macroalgae) to account for inter-annual settling and to estimate the early growth rate of A. nodosum recruits. At the termination of the experiment in August 2014, all quadrats were harvested for both macroalgae and fauna using the same method as in "Bare (start)" August 2011. Retrieved organisms were identified by taxa, counted, weighed, and their maximum dimensions were measured. Macroalgae without bladders were counted as recruits. The vast majority of recruits was below 10 cm length, and as bladders (used to age adult shoots) typically occurred only in individuals >10 cm length, this length limit coarsely separated individuals, here defined as recruits and adults. For each quadrat we quantified the faunal species richness (S) as the number of species and/or taxa present. Finally, in August 2016, 2 years after the final harvest, we quantified the number and length of A. nodosum individuals recruited into all 20 quadrats, adding to the estimate of early growth rates after maximum disturbance.

Between August 2011 and 2014, we quantified a range of physical variables in selected quadrats to characterize the habitat and potential differences between "Bare (start)" and "Full canopy" treatments. The temperature was logged every 1.5 h by sensors (Thermochron iButtons <sup>R</sup> ) placed at site 4. The sensors were placed inside spherical brass housing, to protect them from ice scour and attached to a rock surface cleared of macroalgae and below A. nodosum canopy, respectively. To verify extreme temperatures measured in the intertidal, we compared with air temperature data from a nearby climate station, obtained from the Greenland Ecosystem Monitoring database (GEM). The overall extent of sea ice during the three winters of 2011–2014 was evaluated from photos taken automatically by a camera mounted on the mountain above the experimental area. The photos were taken daily in the period from January to May each year as part of the GEM monitoring program. Additionally, the ice scouring intensity was quantified at each of the five replicate sites by the degree of bending of steel screws inserted into the rock. The screws were standard commercial stainless steel screws having a height of 45 mm, a head diameter of 8 mm and a shaft diameter just below the head of 4 mm. Two screws protruding 2 cm from the rock surface were placed above each quadrat during the winters 2011–12, and 2013–14 (i.e., a total of 80 screws), and the maximum angle of bending (0–90◦ ) of the two screws from each quadrat was used as a proxy for ice scouring intensity. To measure the roughness of the rock surface, i.e., rugosity, within the quadrats when cleared for algae and fauna, we used a profile gauge tool that captured the surface profile of the rock, which was then photographed for later image analysis using the "measure" tool in ImageJ. The ratio of the true surface profile to the linear surface profile gave an estimate of substrate rugosity (Luckhurst and Luckhurst, 1978; Zawada et al., 2010). The unit of analysis was the mean rugosity across the two diagonal profiles in each sample quadrat.

# Statistical Analysis

#### Status of Main Treatments at the End of the Study Period

We compared the macroalgal biomass in each of the main treatments ["Full canopy," "Reduced canopy," "Bare (start)"] at the end of the study period (August 2014) with the "Preexperimental" biomass using Two-sample t-tests. Similarly, macroalgal adult (> 10 cm) and recruit (< 10 cm) densities were compared to the "Pre-experimental" densities.

#### Algal Canopy and Faunal Recolonization

To assess the effect of the main treatments [three level factor; "Full canopy," "Reduced canopy," "Bare (start)"] on biomass of each species and faunal species richness at the end of the study period (2014), we performed one-way ANOVAs with biomass or faunal species richness as response variable and treatment as dependent variable. Biomass data was log-transformed (adding 1 before the transformation) to improve normality and homogeneity of variance assessed visually by Q-Q plots and box plots. To assess the effect of treatment on the density of each species, we performed one-way ANOVAs (GLM) assuming density to follow a Poisson distribution. Before performing the ANOVAs, we confirmed that there was no significant interaction between replicate sites and treatment. Least square means post hoc analyses were performed to test the pairwise difference between treatments. Finally, in order to account for probable inter-annual variation in faunal recolonization, a oneway ANOVA was conducted to compare faunal density between years in the "Bare (annual)" treatment.

#### The Effect of Canopy-Forming Algae on Temperature and Ice Scouring

For the comparison of extreme temperatures measured in "Full canopy" and "Bare (start)," expected to reflect air temperatures during low tide, the 5th and 95th percentiles of temperature measurements were calculated for each month, year and the entire 3-year period. Means of the percentiles in each month, each year and the entire 3-year period between the "Full canopy" and "Bare (start)" were compared using a two-sample t-test.

To assess the variation in ice scouring intensity across replicate sites between years, the relationship between the maximum degree of the bending of screws from 2011 to 2012 and 2013 to 2014 was examined by linear regression. Then, to assess the effect of the main treatments on ice scouring, we performed a one-way ANOVA with the maximum degree of the bending of screws (2013–14) as the response variable. Similarly, a one-way ANOVA was performed to assess whether rock rugosity differed between the main treatments. Assumptions of normality and homogeneity of variance were assessed visually by Q-Q plots and box plots.

#### Rock Rugosity and Recolonization

In the following analysis, we treated the 15 quadrats in the main treatments from 2014 as independent data points since rock rugosity did not differ between treatments (**Table 4C**). Relationships between faunal recolonization (densities or biomasses) and rock rugosity were assessed with linear regression and multiple linear regression (MLR). Similarly, the relationship between A. nodosum recruitment density and rock rugosity was assessed by linear regression. All above analyses were performed using SAS statistical software 9.4 (SAS Institute Inc. Cary, NC, USA).

#### Recovery Period and Early Growth Rate of Ascophyllum nodosum

The early growth rate of A. nodosum recruited during the study period was quantified based on yearly length measurements of recruits in "Bare (annual)" and "Bare (start)." In total, 30 quadrat measurements (10 quadrat measures × 3 years) were used for the cohort analysis. Since we expected recruit lengths in "Bare (start)" to reflect multiple age groups of 2–3 years after experiment start, we applied Hartigan's diptest (Hartigan, 1985), testing the presence of multiple age groups with the R (R Core Team, 2017) package diptest (Maechler, 2016). By comparing the maximum difference between the observed distribution and a unimodal distribution, which minimizes this difference, Hartigan's diptest statistically tests the null hypothesis of unimodality. If p < 0.05, the alternative hypothesis of bi- or multimodality is accepted, i.e., the presence of multiple age groups. Subsequently, the Expectation-Maximization (EM) algorithm (Redner and Walker, 1984) was used to estimate the mean length of recruits within each age group with the R package mixtools (Benaglia et al., 2009). This algorithm uses the maximum likelihood method to find the value of each peak in a multimodal distribution. Here, we looked for three peaks, because we expected new recruits each year in "Bare (start)," i.e., three age groups.

Finally, the growth rate of A. nodosum was calculated with linear regression based on the estimated mean lengths within each age group of individuals recruited in the period August 2011 - August 2014. Additional quantification of algal recruit lengths was conducted in August 2016, i.e., 2 years after the final harvest of all quadrats, providing 20 additional estimates similar to the treatment "Bare (start)." This almost doubled the sample area (30 quadrat measurements + 20 quadrat measurements = 50 quadrat measurements), strengthening the overall estimate of early growth rates of algal recruits in inner Kobbefjord. We used the estimated early growth rates for A. nodosum recruits to evaluate the time it takes from initial settling to the recovery of the "Pre-experimental" canopy mean length in Kobbefjord. This was based on the growth pattern for A. nodosum found by Viana et al. (2014), showing linear growth during the first 2 years and, thereafter, exponential growth until a certain length set by abrasion. We calculated the length of A. nodosum after 2 years of linear growth with our early growth rate, and inserted this (as b) in a natural exponential function, which we solved for x (years):

$$\boldsymbol{\gamma} = \boldsymbol{b} \cdot \boldsymbol{e}^{(\boldsymbol{a} \cdot \boldsymbol{x})},$$

where y is the mean length of "Pre-experimental" canopy, a is the adult relative growth rate of A. nodosum from inner Kobbefjord, calculated as the mean linear growth rate (4.92 cm yr−<sup>1</sup> , Marbà et al., 2017) relative to the intermediate shoot length after starting exponential growth (i.e., length after 2 years + mature length)/2. Finally, we added the first 2 years to achieve the estimate of years it takes to reach the mean length of "Pre-experimental" canopy.

#### RESULTS

#### Canopy-Forming Algae in Inner Kobbefjord

Based on the pre-experimental monitoring in August 2011 ["Bare (start)"], the mid intertidal zone of the inner part of Kobbefjord was dominated by the canopy-forming fucoid, A. nodosum, with the occasional presence of F. vesiculosus. The average length of the A. nodosum canopy (> 10 cm) was 46.9 ± 6.0 cm and the minimum age was 5.9 ± 1.3 years for the "Pre-experimental" community (n = 5).

**Figure 2** shows the combined biomass and densities of A. nodosum and F. vesiculosus in each treatment, noting that A. nodosum represented over 99% of the total observed biomass and adult density, and over 80% of the observed recruit density. The macroalgal biomass in the "Pre-experimental" community was on average 31 kg m−<sup>2</sup> , which was not significantly different from the macroalgal biomass found in either "Full canopy" and

of algal recruits and adults, i.e., for A. nodosum and F. vesiculosus combined, in the main treatments "Full canopy," "Reduced canopy," and "Bare (start)" in Kobbefjord at the end of the experiment (August 2014). "Pre-experimental" was a baseline measurement based on the material cleared from "Bare (start)" at the start of the experiment (August 2011). \*Annotates a significant difference of p < 0.05 for each treatment compared to "Pre-experimental" by a two-sample t-test. N = 5 for all mean values. Error bars are SE.

"Reduced canopy" in August 2014, three years after experiment start (**Table 2**, **Figure 2**). In contrast, the "Bare (start)" treatment, which was fully cleared at the beginning of the experiment, revealed minimal recovery in macroalgal biomass compared to "Pre-experimental," supporting only 0.11 kg m−<sup>2</sup> by August 2014 (**Table 2**, **Figure 2**). In addition, the density of macroalgal adults (>10 cm) in "Bare (start)" showed minimal recovery by August 2014 compared to "Pre-experimental," while "Full canopy," "Reduced canopy" and "Pre-experimental" displayed a similar level of adult density (**Table 2**, **Figure 2**). However, as the adult density in "Pre-experimental" was slightly higher than in "Full canopy" (**Table 2**, **Figure 2**), this was not reflected in the biomass. The density of recruits (<10 cm) at the end of the experiment tended to be higher in all treatments than in "Preexperimental" (**Figure 2**), yet differences were not significant (**Table 2**).

#### Algal Canopy and Faunal Recolonization

In total, we observed 11 faunal taxa in the study area (**Table 3**). Four species, S. balanoides, M. edulis, L. obtusata and L. saxatilis, contributed with 98.5% of the faunal biomass (wet weight). Three years after removal, the recovery of total faunal biomass relative to "Pre-experimental" was 26.1% for "Full canopy," 28.5% for "Reduced canopy" and 1.4% for "Bare (start)." The recovery of faunal density was considerably faster than biomass for the dominant taxa, being 45.5% for "Full canopy," 53.8% for "Reduced canopy" and 4.4% for "Bare (start)." Total faunal biomass nor density differed significantly from the "Preexperimental" total faunal biomass or density.

TABLE 2 | Summary of two-sample t-tests comparing "Pre-experimental" macroalgal biomass, adult and recruit density to each treatment.


\*p < 0.05, \*\*\*p < 0.001.

TABLE 3 | Faunal taxa of "Pre-experimental" community in the mid intertidal zone of inner Kobbefjord, SW Greenland, August 2011.

The combined biomass (WW) of dominating fauna is given.

In terms of biomass, L. obtusata, L. saxatilis and M. edulis were affected by canopy cover (**Figure 3A**, **Table 4A**), with L. obtusata showing a significantly higher biomass in both "Full canopy" and "Reduced canopy" compared to "Bare" (post-hoc LS test, Bare-Full canopy: p = 0.007, Bare-Reduced canopy: p = 0.004) and L. saxatilis showing a significantly higher biomass in "Reduced canopy" compared to both "Bare" and "Full canopy" (post-hoc LS test, Reduced canopy-Bare: p = 0.001, Reduced canopy-Full canopy: p = 0.042). Mytilus edulis displayed a significantly higher biomass in both "Full canopy" and "Reduced canopy" compared to "Bare" (post-hoc LS test, Bare-Full canopy: p < 0.001, Bare-Reduced canopy: p < 0.001). The density of all four species were affected by canopy cover (**Figure 3B**, **Table 4A**). Littorina obtusata displayed a significantly higher density in both "Full canopy" and "Reduced canopy" compared to "Bare" (post hoc LS test, Bare-Full canopy: p = 0.007, Bare-Reduced canopy: p = 0.004) and L. saxatilis showed significantly higher density in "Reduced canopy" compared to "Bare" and "Full canopy" (post-hoc LS test, Reduced canopy-Bare: p < 0.001, Reduced

of p < 0.05 by a post-hoc test between main treatments for each species, i.e., excluding "Pre-experimental." N = 5 for all mean values. Error bars are SE.

canopy-Full canopy: p = 0.003) (**Figure 3B**). Mytilus edulis displayed a higher density in both "Full canopy" and "Reduced canopy" compared to "Bare" (post-hoc LS test, Bare-Full canopy: p < 0.001, Bare-Reduced canopy: p < 0.001) (**Figure 3B**) and S. balanoides displayed a higher density in "Reduced canopy" compared to "Bare" (post-hoc LS test, Reduced canopy-Bare: p = 0.004) (**Figure 3B**). Moreover, faunal species richness was significantly affected by canopy cover (**Table 4B**), displaying a higher species richness in "Full canopy" and "Reduced canopy" compared to "Bare" (post-hoc LS test, Bare-Full canopy: p = 0.008, Bare-Reduced canopy: p = 0.026). The inter-annual recolonization of M. edulis density, based on measurements in "Bare (annual)," was not significantly different between years [ANOVA, F(2, 12) = 0.45, p = 0.65]. However, S. balanoides density was twice as high in 2012 as in the following 2 years [ANOVA, F(2, 12) = 5.39, p = 0.021].

### Modifying Properties of the Canopy-Forming Algae

Sea ice presence varied between years, with just 2 months of sea ice cover in the inner fjord in 2013, compared to 4–5 months TABLE 4 | Summary statistics of one-way ANOVAs (A) assessing the effect of treatment ["Full canopy," "Reduced canopy" and "Bare (start)"] on biomass and density of dominating intertidal faunal species in Kobbefjord August 2014.


(B) Assessing effect of treatment ["Full canopy," "Reduced canopy" and "Bare (start)"] on faunal species richness in Kobbefjord August 2014. (C) Assessing effect of treatment ["Full canopy," "Reduced canopy" and "Bare (start)"] on rock rugosity and ice scour intensity. \*\*p < 0.01, \*\*\*p < 0.001.

of sea ice cover in 2012 and 2014 (**Figure 4**). We were unable to quantify the ice scouring intensity at site 1 and 2 in 2011– 2012 due to loss of screws. However, the maximum value of the two screws in each quadrat, showed a similar pattern across sites in 2011-2012 and 2013-2014 [LR, R <sup>2</sup> = 0.30, F(1, 13) = 5.556, p = 0.035], indicating that the variation in ice scouring intensity was consistent across sites between years. Therefore, we used the more complete dataset from 2013–2014 as a proxy for ice scouring intensity in the inner Kobbefjord. The maximum ice scouring intensity in terms of screw bending ranged between 0 and 75◦ across all quadrats, and maximum ice scouring intensity (2013–2014) did not differ significantly between treatments (**Table 4C**).

The running mean temperature measured in "Full canopy" and "Bare (start)," along with the running mean air temperature (AT) measured in Nuuk city from August 2011 to August 2014 are illustrated in **Figure 4**. Here, we show a slight difference between the mean temperatures in "Full canopy" compared to "Bare (start)." However, in order to uncover and compare the extreme temperatures measured in "Full canopy" and "Bare (start)," we extracted the 5th and 95th percentiles (**Table 5**, see **Supplementary Table 1** for monthly comparisons). We expect these to reflect air temperatures during low tide since the 5th percentiles in winter and the 95th percentiles in summer are below and above the typical range of sea surface temperatures in this area (Versteegh et al., 2012). The means of the percentiles across the 3-year period differed significantly between "Full canopy" and "Bare (start)," suggesting that algal cover buffers the temperature variations (**Table 5**). Testing the percentiles of years separately also showed that the extreme temperatures of

from August 2011 through August 2014 (data not available for July 14th–Aug 24th 2012). Air temperature (AT) with running mean (336 observations) recorded in Nuuk by the Greenland Ecosystem Monitoring program in the same period. Hatched bars represent periods of sea ice cover in inner Kobbefjord based on automatized photographing (data only available for January through May).

TABLE 5 | 5th (extreme low) and 95th (extreme high) percentiles of yearly temperatures (◦C) in treatment "Full canopy" and "Bare (start)" in inner Kobbefjord mid intertidal from August 2011 to August 2014.


Monthly percentiles are shown in Supplementary Table 1. \*Marks level of p-value from two-sample t-test contrasting "Bare (start)" and "Full canopy" mean percentiles. \*\*\*p < 0.001.

2011–2012 and 2013–2014 differed significantly between the two treatments, while this was not the case for 2012–2013 (**Table 5**). Testing the percentiles of each month separately showed that mainly summer and winter months differed between treatments, especially so for colder months (see **Supplementary Table 1**). In some summer months, canopy cover reduced the 5% highest temperatures by a maximum of 5.5◦C and during winter months the 5% lowest temperatures were raised by a maximum of 0.5◦C, again with canopy-forming algae reducing temperature extremes.

### Rock Rugosity and Recolonization

Rock rugosity did not differ significantly between the three main treatments (**Table 4C**). Therefore, we chose to combine all 15 quadrats in a linear regression to test the relation between recruitment and rugosity. As opposed to all other invertebrate taxa, the total density of M. edulis showed a positive linear relationship with rock rugosity [LR, R <sup>2</sup> = 0.551, F(1, 13) = 18.21, p < 0.001] [see **Supplementary Table 2** in Supplementary Material for full (M)LR details]. With a multiple linear regression, we show that both canopy biomass and rock rugosity influenced the total M. edulis density positively [MLR, R <sup>2</sup> = 0.726, F(2, 12) = 19.53, p < 0.001]. When corrected statistically for canopy biomass, rock rugosity explained 62% of the variation in mussel density (R <sup>2</sup> = 0.62, p < 0.001) and when corrected statistically for rock rugosity, canopy biomass explained 44% (R <sup>2</sup> = 0.44, p = 0.010). In addition, A. nodosum recruit density showed a positive linear relationship with rock rugosity [LR, R <sup>2</sup> = 0.419, F(1, 13) = 11.11, p = 0.005].

#### Recovery Period of Ascophyllum nodosum

We estimated the early growth rate of A. nodosum recruits in the period August 2011–August 2014 to be 1.4 cm yr−<sup>1</sup> . This was based on Hartigan's diptest (HD) that showed the distribution of recruit lengths to be multimodal, representing three age groups (HD, n = 76, D = 0.06, p = 0.042), for which we estimated mean lengths with the Expectation-Maximization algorithm. The addition of the algal recruit length data from 20 quadrats, cleared in August 2014 and quantified in August 2016, increased the sample size 9 times. This strengthened analysis (HD, n = 673, D = 0.04, p < 0.001) identified a similar early growth rate of 2.0 cm yr−<sup>1</sup> . Based on the strengthened estimate of early growth rate, resulting in 4 cm long shoots after 2 years of linear growth, and assuming an exponential growth pattern thereafter with an estimated growth rate of 0.19 % yr−<sup>1</sup> [4.92 cm yr−<sup>1</sup> /(4 + 46.9 cm)/2], we estimated that it would take ca 15 years (including the first 2 years of linear growth) for A. nodosum recruits to reach the mean length of the "Pre-experimental" canopy in inner Kobbefjord.

# DISCUSSION

# Macroalgal Canopy Cover Facilitates Faunal Recolonization

The recolonization of M. edulis, S. balanoides and the two snails L. obtusata and L. saxatilis displayed a positive response to macroalgal canopy cover (predominately A. nodosum) at the mid intertidal level in inner Kobbefjord. These findings support the hypothesis that algal canopy has an overall facilitating effect on faunal recolonization in a sub-Arctic intertidal. This is opposed to findings from a temperate intertidal, where an overall neutral response in barnacle recruitment to A. nodosum canopy cover was observed at both mid- and high intertidal levels (Beermann et al., 2013). Other studies find positive biotic responses to algal canopy cover confined to the high intertidal (Dayton, 1971; Hawkins, 1983; Jenkins et al., 1999b). In relation to the stress gradient hypothesis, we have already suggested that in a stressful environment like the sub-Arctic, the positive effects of canopy cover in modifying the environment are likely to overrule the negative effects such as space competition and whiplashes from algal fronds. To further verify this hypothesis, we explored the effect of algal canopy in modifying the key stressors in the sub-Arctic intertidal, i.e., extreme air temperature and sea ice scouring.

# Macroalgal Canopy Cover Modifies Extreme Temperatures

While earlier studies from Greenland have reported marked diurnal and seasonal temperature variation in the sub-Arctic intertidal (Høgslund et al., 2014), our results take a step further and document that the A. nodosum canopy created significant small-scale spatial variation in temperature by acting as an insulating cover. The macroalgal cover in inner Kobbefjord reduced the highest 5% temperatures in summer by up to 5.5◦C and raised the lowest 5% in winter by up to 0.5◦C (see **Supplementary Table 1**). Thus, A. nodosum canopy buffers extreme air temperatures experienced by organisms in a sub-Arctic rocky intertidal. This pattern is consistent with studies from the temperate intertidal where the daily maximum temperature, measured over a month, was lowered by the canopy cover in the mid- and high intertidal zone (Beermann et al., 2013; Watt and Scrosati, 2013a). Therefore, we suggest that the ability of algal canopy to buffer extreme temperatures, likely contributes to the overall positive biotic response to canopy cover found at this study site. In contrast, the canopy-forming algae had no effect on ice scouring intensity, indicating that the canopy did not markedly protect the understory community from the mechanical disturbance of sea ice scouring, which is consistent with previous studies describing A. nodosum's low capacity to withstand ice scouring (Åberg, 1992).

The extreme temperatures measured in this study were within the mean lower and upper thermal limits of the dominating intertidal faunal species, L. saxatilis, L. obtusata, S. balanoides, and M. edulis, documented in Scotland (−16.4– 35◦C, Davenport and Davenport, 2005). However, the current measures of thermal lethal limits in southern populations may not be the same as for sub-Arctic populations. As an example, M. edulis populations in Canada (Bourget, 1983) displayed lower, mean lower lethal limits than the populations in Scotland (Davenport and Davenport, 2005). Moreover, mortality rates at increasing temperatures may differ depending on life stage (Foster, 1971a; Bourget, 1983) and most studies include only adult life stages.

In inner Kobbefjord mid intertidal, the minimum 5th percentile temperature was −6.0◦C and the maximum 95th percentile was 22◦C (see **Supplementary Table 1** in Supplementary Material). These temperatures are expected to increase in the future, based on a warming trend in mean annual air temperatures in west Greenland (Thyrring et al., 2017). In a nearby climate station, the air temperatures have increased by 0.13 ± 0.02◦C per year over the last three decades together with an increase in the duration of annual ice-free cover along the Greenland west coast (Marbà et al., 2017). Thus, in future scenarios, the ability of algal canopy to buffer temperature extremes may become more important in preventing increased mortality in understory faunal communities, both with regard to warmer air temperatures in summer and colder air temperatures in winter, as the annual sea ice cover decreases. For example, it was evident that sea ice was buffering cold air temperatures during winter at the studied site, as we observed a marked difference between air temperatures (AT) and temperatures measured in the intertidal during winter compared to the other seasons. During the winter 2013, with only 2 months of sea ice cover in Kobbefjord, the temperatures from the intertidal also showed greater variation than in the other two winters with 5 months of sea ice cover (2012, 2014, **Figure 4**).

Multiple other physico-chemical stress factors, not included in our study, such as desiccation, water flow, wave action, salinity and pH, may also be altered by canopy-forming algae, contributing to the overall positive biotic response to canopy cover (Helmuth, 1998; Jenkins et al., 1999a; Beermann et al., 2013; Wahl et al., 2018).

### Rock Rugosity Facilitates Recolonization

The coupling between rock roughness and the colonization of intertidal species has been noted before (Guidetti et al., 2004; Skinner and Coutinho, 2005; Chase et al., 2016). In the studied area, M. edulis is, in general, confined to cracks and crevices in the intertidal zone (Blicher et al., 2013) and here, we show that the total recolonization of M. edulis across the main treatments was positively related to both rock rugosity and macroalgal biomass. Moreover, A. nodosum canopy may contribute to an additional level of rugosity, in that holdfasts attached to the rock, as well as the canopies above, form multiple crevices for settlement. This type of rugosity, habitat complexity and increased surface area for colonization may contribute to the overall positive effect that the A. nodosum canopy has on faunal recolonization. In addition, we found that recolonization of A. nodosum was positively related to rock rugosity, suggesting that rock roughness is important for the settlement and early development of germlings, likely explained by an increase in settling surface and protection from ice scouring and other physical stressors in the small crevices (Heaven and Scrosati, 2008; Gerwing et al., 2015; Musetta-Lambert et al., 2015).

#### Recovery Time of the Intertidal Community

In a sub-Arctic intertidal, where mechanical disturbance such as ice scouring may dislodge the entire community, we show that, for both fauna and macroalgae, the subsequent recovery in density is faster than in biomass. Thus, while the organisms recolonize relatively fast after maximum disturbance, it takes longer to recover the biomass of the intertidal community. On the other hand, when A. nodosum individuals prevail and the canopy is simply sliced by ice scouring, the algal biomass fully recovers after 3 years, and the recolonization rates resemble those found in an untouched canopy.

The early growth rate of newly-settled A. nodosum in the studied intertidal community averaged 1.4–2 cm yr−<sup>1</sup> . Comparably, the growth rate of adult A. nodosum measured in the same macroalgal community is 4.92 cm yr−<sup>1</sup> (Marbà et al., 2017). Early growth rates of A. nodosum recruits (≤2 years old) from the Isle of Man and south west Spain are reported at 7.2 cm yr−<sup>1</sup> and 9.6 cm yr−<sup>1</sup> , respectively (Cervin et al., 2005; Viana et al., 2014). However, in Maine, USA, Keser and Larson (1984) reported an early growth rate of 1.6 cm yr−<sup>1</sup> , which is similar to what we found. Based on existing growth patterns for this species (Viana et al., 2014), but modified for the local conditions, we estimated that A. nodosum in inner Kobbefjord would reach its full length, averaging 46.9 cm, ca 15 years after removal by mechanical disturbance. Correspondingly, we found the macroalgal biomass to be very low, even after 3 years of recovery. The estimated recovery period, in height, is higher than the minimum mean age (6 years) of unbroken branches of A. nodosum found in this study, bearing in mind that holdfasts of A. nodosum have been suggested to be 50–70 years old, and the community has been referred to as "marine trees" (Olsen et al., 2010), displaying similar patterns of slow population growth as terrestrial trees (Capdevila et al., 2016). However, it is close to the minimum age of the longest individuals, assessed as the presence of 13 full years of consecutive growth, of A. nodosum individuals also observed in Kobbefjord (Figure 3 in Marbà et al., 2017). A long recovery period of A. nodosum height corresponds to findings in north temperate regions (Printz, 1956; Cervin et al., 2005; Ingólfsson and Hawkins, 2008). However, they largely assign a delay in A. nodosum biomass recovery to the competition from other fucoid species. In inner Kobbefjord, algal recruits on bare rock were dominated by A. nodosum, with a few F. vesiculosus, suggesting that recovery of A. nodosum in inner Kobbefjord was not delayed by interspecific competition. Rather, the slow recovery of A. nodosum canopy height may be largely due to the constraints of the abiotic environment. Whereas our study assessed the recovery time of canopy height, a complete study of recovery of canopy biomass should also include quantification of population dynamics in terms of recruitment and mortality. Recruitment and mortality rates in the studied A. nodosum population, assessed through the Greenland Ecosystem Monitoring program, showed that the vast majority of the biomass was quite persistent with loss rates of individuals of only 1.6–6.2% per year, whereas the pool of new shoots add markedly to the population dynamics, resulting in overall recruitment rates of 6–111% and mortality rates of 1.6– 60% for the period 2012–2015. However, the overall population growth ensured a biomass increase of the population of 1– 50% yr−<sup>1</sup> during the assessment period 2012–15, which, by far, outweighed the biomass of lost individuals (Juul-Pedersen et al., 2015).

Ascophyllum nodosum mainly dominates in wave-sheltered areas (Scrosati and Heaven, 2008), while other fucoid species, such as F. vesiculosus and F. distichus/F. evanescens takes over at more exposed sites as well as above the northern distributional range of A. nodosum. Like A. nodosum, we expect these other fucoid species to display a similar buffering effect on extreme temperatures and facilitate the colonization of intertidal species. However, buffering capacity most likely depends on the level of canopy biomass, which was high in the studied area.

Based on future projections on climate change, the distributional ranges of both A. nodosum and F. vesiculosus are expected to shift northward (Jueterbock et al., 2013). A northward shift in other high latitude intertidal macroalgal communities have already been documented (Weslawski et al., 2010), and a recent study found that warming tends to enhance the growth of A. nodosum at its northern distribution limit (Marbà et al., 2017). In comparison, the southern distribution limits of dominant intertidal fauna, M. edulis and S. balanoides, are also moving northward (Jones et al., 2010, 2012), and based on our results, algal canopies may be a key habitat in facilitating the northward colonization and abundance of these intertidal species at their northern distribution limit (Gilman et al., 2010).

# CONCLUSION

This study emphasizes the role of A. nodosum canopy and rocky rugosity in structuring high latitude intertidal communities. We show that intertidal canopy-forming macroalgae have an overall facilitating effect on faunal species richness and recolonization, likely by reducing temperature variations and increasing settling surfaces and habitat complexity. We show that the canopy modifies the small-scale variation in temperature experienced by intertidal fauna, but fails to reduce ice scouring at the rock surface, underlining the complexity in how a physical stressor can vary depending on scale. The facilitating properties of algal canopies are important to consider when predicting future distributional patterns of high latitude intertidal fauna. To fully comprehend the community-structuring role of algal canopies, we need to study whether and how they modify other key stressors, such as salinity, pH, water loss and flow, including different intertidal levels. Finally, we estimated that the full establishment and recovery period of the intertidal community may take at least 15 years while biomass and height of A. nodosum canopy builds up, which is important to consider when predicting latitudinal range shifts in the sub-Arctic and Arctic intertidal community.

#### DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher. The air temperature data from Nuuk was obtained from Greenland Ecosystem Monitoring program (GEM). Requests to access these datasets should be directed to http://data.g-e-m.dk/. The long time series of photos documenting the sea ice in Kobbefjord was obtained from the Andreas Westergaard-Nielsen and request to access these photos should be directed to Andreas Westergaard-Nielsen, (awn@ign.ku.dk).

#### AUTHOR CONTRIBUTIONS

All authors have approved the submission version. MS, DK-J, KM, and BO conceived and planned the experiment. DK-J, KM, BO, NM, SØ, MB, and ML conducted the field work. Data analysis was led by SØ with input from DK-J, MS, KM, and BO. SØ produced all figures and wrote the first draft with contributions and comments from all the authors.

#### REFERENCES


#### ACKNOWLEDGMENTS

This project was funded by the Danish Environmental Protection Agency's Arctic initiative (DANCEA) and by the Arctic Research Centre at Aarhus University. We are thankful to Anissa Merzouk for helping with the fieldwork sampling, and we are thankful to Jakob Thyrring and Søren Larsen for support on data analysis. We thank Andreas Westergaard-Nielsen for providing photos of the sea ice cover in Kobbefjord, and we thank the Greenland Ecosystem Monitoring program (www.G-E-M.dk) for providing weather data from Nuuk city. We are also grateful to the Greenland Institute of Natural Resources (GINR), Nuuk, Greenland, and their staff for providing excellent working conditions.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00332/full#supplementary-material


intertidal macroalgae Ascophyllum nodosum near the northern distribution limit. Ambio 46, 119–131. doi: 10.1007/s13280-016-0873-7


edulis: Archives of glacier meltwater in Greenland? Biogeosciences 9, 5231–5241. doi: 10.5194/bg-9-5231-2012


**Conflict of Interest Statement:** ML is employed by company Danish Center for Wild Salmon.

The remaining 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 Ørberg, Krause-Jensen, Mouritsen, Olesen, Marbà, Larsen, Blicher and Sejr. 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.

# Influence of the Seagrass, Zostera marina, on Wave Attenuation and Bed Shear Stress Within a Shallow Coastal Bay

#### Matthew A. Reidenbach\* and Emily L. Thomas

Department of Environmental Sciences, University of Virginia, Charlottesville, VA, United States

#### Edited by:

François G. Schmitt, Laboratoire d'Océanologie et de Géosciences-UMR 8187 LOG, Centre National de la Recherche Scientifique (CNRS), France

#### Reviewed by:

Alejandro Orfila, Instituto Mediterráneo de Estudios Avanzados (IMEDEA), Spain Ian Townend, University of Southampton, United Kingdom

> \*Correspondence: Matthew A. Reidenbach reidenbach@virginia.edu

#### Specialty section:

This article was submitted to Coastal Ocean Processes, a section of the journal Frontiers in Marine Science

Received: 29 March 2018 Accepted: 09 October 2018 Published: 26 October 2018

#### Citation:

Reidenbach MA and Thomas EL (2018) Influence of the Seagrass, Zostera marina, on Wave Attenuation and Bed Shear Stress Within a Shallow Coastal Bay. Front. Mar. Sci. 5:397. doi: 10.3389/fmars.2018.00397 Local effects of flow interaction with seagrass structure modify meadow scale hydrodynamics, resulting in lower current velocities and wave heights within a seagrass meadow. This attenuation promotes the deposition of suspended sediment, increasing the light available locally to benthic organisms. To elucidate the relationship between small-scale hydrodynamics that occur at the sea floor and the meadow scale effects of seagrass, high resolution velocity profiles were recorded adjacent to the sediment-water interface within a Zostera marina seagrass meadow in South Bay, Virginia. Additionally, instrumentation was deployed across the meadow to seasonally monitor corresponding changes in wave height across the seagrass meadow. Results show that wave height was reduced by 25–49% compared to an adjacent bare site, and by 13–38% compared to an analytical model of wave attenuation over an unvegetated seafloor with the same bathymetry. The greatest attenuation of wave height occurred during the spring and summer when seagrass biomass was greatest, while the lowest attenuation occurred in winter, corresponding to periods of minimal seagrass biomass. Significant wave height attenuation coefficients, αw, calculated for the meadow ranged from α<sup>w</sup> = 0.49 in spring to 0.19 during winter, but were highly dependent on wave conditions, with greater α<sup>w</sup> for larger wave heights and longer period waves. Within the seagrass meadow during summer, the highest measured bed shear stress was τbed = 0.034 ± 0.022 Pa, which occurred during peak wave conditions. This suggests that during high biomass conditions, the bed shear stress rarely exceeds the critical bed shear, τcrit = 0.04 Pa necessary to initiate sediment resuspension. This is in contrast to the bare site which showed elevated values of τbed above the critical threshold across all seasons. These findings suggest the seagrass meadow does exert significant control over both wave heights and the hydrodynamic conditions at the sediment-water interface, and this control is due to the attenuation of wave motion by drag induced from the seagrass over the expanse of the meadow.

Keywords: seagrass, waves, attenuation, turbulence, shear, canopy flow

# INTRODUCTION

Coastal waters in which seagrasses exist often experience complex hydrodynamics containing both wind-driven wave motions and tide-driven currents. Seagrass beds have been shown to attenuate both oscillatory flows (Fonseca and Cahalan, 1992; Luhar et al., 2017) and current velocities (Koch and Gust, 1999; Lacy and Wyllie-Echeverria, 2011), impacting the flow structure above and within the canopy. Seagrasses are typically light limited, so their range and productivity is restricted by both the depth and clarity of the water column (Carr et al., 2016). Seagrasses reduce suspension of sediment due to the stabilizing effect of root structure and flow modification caused by the plant structure (Gacia and Duarte, 2001; Hansen and Reidenbach, 2013), creating a positive feedback loop that increases the light available at the sea floor and promotes primary productivity (Carr et al., 2010). Therefore, understanding flow within a combined wave-current boundary layer is critical to estimating local bed shear stress, which influences sediment suspension and light attenuation, which ultimately alters seagrass productivity and the ecosystem services provided by a healthy seagrass meadow.

Flow conditions adjacent to the sediment-water interface have the greatest implication for sediment suspension. Sediment grains are mobilized when the bed shear stress acting on a grain surpasses the critical shear stress necessary to initiate the grain's motion. Shear stress is a tangential force imparted to the sea floor by velocity gradients found in the boundary layer of the overlying water. As a result, the bottom boundary layer in shallow coastal waters is a combination of a thin wave boundary layer superimposed over a well-developed current boundary layer that may be orders of magnitude thicker (Grant and Madsen, 1979). This interaction between waves and currents is nonlinear, and the shear stresses created at the seafloor can be dramatically different than those expected under either condition independently (Jing and Ridd, 1996). For waves generated within fetch-limited shallow coastal bays, the magnitude of wave orbital motions decrease with depth and the amount of wave energy that reaches the seafloor depends on the wave height, wave period, and water depth (Chen et al., 2007; Fagherazzi and Wiberg, 2009) as well as the structure and density of vegetation (Zeller et al., 2014; Weitzman et al., 2015; Hansen and Reidenbach, 2017).

Drag is fundamentally a process of removing energy from a flow. The bulk drag coefficient (CD) is a dimensionless measure of the resistance of an object to mean flow (Denny, 1988). This drag coefficient can be calculated in pure current environments (Rippeth et al., 2002) as well as in environments with waves and currents (Bricker et al., 2005), however is defined as a measure of momentum attenuation, and is not a measure of wave height reduction. In coastal environments, C<sup>D</sup> is often estimated as the amount of energy dissipation caused by flow interaction with the seafloor, and historically has been computed from estimates of the bed shear velocity, u<sup>∗</sup> , through quantification of the mean velocity profile, turbulent Reynolds stresses, or rate of turbulent kinetic energy dissipation (e.g., Reidenbach et al., 2006). Due to the presence of vegetation, this drag can be due to interactions with the seafloor or with the seagrass blades themselves, therefore integrated drag through the water column must be taken in to account (Nepf, 1999). The relative magnitude of C<sup>D</sup> can be impacted by the incident wave and current forcing, the characteristics of the seagrass canopy, and morphology and flexibility of the individual blades (Houser et al., 2015). C<sup>D</sup> for a given canopy can range up to an order of magnitude or more due to variations in Reynolds number of the mean flow and Keuglan-Carpenter number of the oscillatory flow (Bradley and Houser, 2009; Zeller et al., 2014).

Although C<sup>D</sup> is an important measure of the integrated drag by the canopy of the flow, often it is advantageous to determine canopy effects solely on the wave climate through a separate wave attenuation parameter. Wave attenuation, αw, is typically calculated using the change in wave height or wave velocities measured at two or more locations along the direction of wave propagation. This method was first explored using wave height decay by Bretschneider (1954) and revised by Bretschneider (1958) for wind waves in shallow water with impermeable sediment, and has been applied to a variety of settings (Mazda et al., 1997; Quartel et al., 2007). Lowe et al. (2007) also calculated wave attenuation by comparing wave velocity spectra above and within a canopy. Using a similar method, Paul and Amos (2011) determined wave attenuation by comparing differences in wave spectra horizontally across a meadow. These methods are different in that the Bretschneider (1954) approach only considers energy dissipation across the mean of the highest 1/3 of the waves (the significant wave height), whereas Lowe et al. (2007) calculates the total energy dissipation across the entire wave spectra. However, it is the largest waves that are often most relevant to coastal erosion, so although only calculating energy loss at the significant wave frequency is a less complete measurement of attenuation, it may be a more practical metric for use in numerical models and to understand the dominant physical processes causing sediment resuspension, etc.

Over the past few decades, numerous field and laboratory studies have been performed to determine the effects of vegetation on wave attenuation. In a field study in a Florida bay, Bradley and Houser (2009) measured changes in wave height across a seagrass meadow using a linear array of pressure sensors. Wave height decayed at an exponential rate once within the meadow, with a total decrease of 30% across the 39 m transect. Dissipation was not uniform across all frequencies of wave motion; seagrasses acted as a low-pass filter for wave attenuation by selectively removing high frequency motions due to the slower oscillations of the seagrass blades. In flume studies, Fonseca and Cahalan (1992) compared attenuation of wave energy by four different seagrass species including Zostera marina over a 1 m section in a flume, and found an average wave energy reduction of 40% per meter of seagrass when the blade length was approximately equal to the water depth. The relation between leaf length and water depth was found to be the most important parameter for all four species studied. In another flume study, Paul et al. (2012) determined that seagrasses were less effective at attenuating wave energy in combined wave-current environments compared to environments with waves alone. Paul et al. (2012) suggest that the reduction in attenuation in wavecurrent environments is a combination of decreased frontal area as seagrass blades are flattened by the current and a reduction in the amount of wave energy transferred to blade motion as individual blade are prevented from swaying with wave orbital motion in the presence of a strong current.

Within a shallow coastal bay in Virginia United States, Hansen and Reidenbach (2012) compared flow between vegetated sites of varying density at points above and below the seagrass canopy to assess the effect of seagrass on wave orbital motion, turbulence and sediment suspension. The seagrass canopy was found to reduce flow velocity, turbulence, and wave orbital velocities compared to measurements above the canopy. Significant wave heights were also observed to be lower compared to an adjacent unvegetated site. Hansen and Reidenbach (2013) continued this work with a seasonal comparison of hydrodynamics between a bare and vegetated site, and found that wave heights were reduced at the vegetated site in all seasons; however, wave orbital motions were still able to penetrate vertically through the canopy and induce bottom shear stress. However, no direct measurements of wave-current interactions within the fluid layer adjacent to the seafloor have yet been made within a seagrass meadow.

To determine the effects of a seagrass meadow on wave attenuation and wave-current interactions at the sediment-water interface, the same Z. marina meadow in South Bay, Virginia, was monitored seasonally across a 20-month time period. The two specific questions that are addressed are:


### MATERIALS AND METHODS

#### Study Area

Research was conducted in South Bay, a shallow coastal lagoon located within the Virginia Coast Reserve Long Term Ecological Research site (VCR LTER) on the Delmarva Peninsula, Virginia (**Figure 1**). The VCR LTER consists of a system of barrier islands, lagoons, and salt marshes which are strongly influenced by semidiurnal tidal variations. Mean water depth in South Bay is 1.0 m, with an average tidal amplitude of 0.6 m, with winds primarily from the SSE-SSW and N-NE (Fagherazzi and Wiberg, 2009). Waves in the bay are primarily wind driven, as the barrier island system blocks larger ocean swell from propagating further inland.

The dominant vegetation in South Bay is the seagrass (eelgrass) Z. marina. Seagrasses, once a dominant feature in the coastal bays of the Delmarva Peninsula, disappeared in the 1930's due to a combination of a pandemic wasting disease and the 1933 Chesapeake-Potomac hurricane. South Bay was a site of eelgrass restoration that was reseeded with Z. marina from 2001– 2005. As of summer 2010, the reseeded patches in South Bay had coalesced and grown to a meadow encompassing >6 km<sup>2</sup> (Orth et al., 2012). Three sites were chosen for the study within the restored seagrass meadow, named North, Center, and South sites, which were in an array spaced approximately 1 km apart (**Figure 1**). The fourth site was in a bare (unvegetated) portion of the bay adjacent to the meadow, approximately 0.5 km west of the Center site. Mean water depths at the four sites were North (1.21 m), Center (1.00 m), South (1.16 m), and bare (0.94 m). The fetch length at each site varied depending on wind angle due to the location of the marsh and barrier island system. South Bay is long and narrow, oriented with the long axis in a north to south direction, where the north and south fetch lengths were the longest (7.5 km), while east and west fetch lengths were typically between 1 and 2 km.

#### Data Collection

Hydrodynamics were measured at the four sites (North, Center, South, and bare) and a seasonal sampling regime was implemented between October 2011 and July 2013 to assess the impact of morphological changes of the seagrass on the meadow scale hydrodynamics. Wave motions were recorded simultaneously at each site using wave gauges (Richard Branker Research© TWR-1050), which were programmed to measure pressure in bursts of 1024 samples at 4 Hz every 10 min. These 1,024 samples were averaged to compute mean tidal height every 10 min. Significant wave height, the mean wave height of the highest third of waves, was determined from pressure data using spectral analysis (Wiberg and Sherwood, 2008). Wave gauge pressure data was corrected for atmospheric pressure fluctuations using measurements recorded at the nearby Wachapreague National Oceanographic and Atmospheric Administration (NOAA) meteorological station. Meteorological and wave data collected for this study is publicly available on the Virginia Coast Reserve, Long Term Ecological Research Data Repository (Thomas and Reidenbach, 2015; Porter et al., 2018).

Adjacent to the sediment-water interface, high resolution velocity profiles were recorded using a Nortek© Vectrino II acoustic Doppler profiler. Data were collected during July 2012 and July 2013 at the center seagrass site, and during July 2012, October 2012, and July 2013 at the bare site. The Vectrino profiler recorded velocity data over a 30 mm profile at 1 mm resolution at a rate of 20 Hz. Doppler pulses are emitted from a center transmitter and received by four passive transducers angled 30◦ toward the center, resulting in a sample volume 40–70 mm from the center transmitter. This high spatial and temporal resolution allowed for the resolution of wave and turbulence motions. The Vectrino profiler was positioned to record velocity measurements directly adjacent to the sediment-water interface to capture water motions associated with wave-current boundary layer development. To ensure measurements adjacent to the sediment were included in the data set, several bins were positioned below the interface, and the signal to noise ratio (SNR) was used to indicate the location of the bottom. If necessary, seagrass was removed from directly beneath the probe at the vegetated site to prevent blade interference. Velocity profiles were first rotated into the direction of the mean flow, using an average of the rotation required for the top three points. Each raw data record was then analyzed to see if the SNR and percent correlation

averaged across the 10-min record exceeded 20 dB and 80%, respectively, the recommended limits for velocity measurements. If these conditions were met, the raw data were used to compute velocity, wave, and turbulence statistics. In general, velocity data within approximately 5 mm from the seafloor did not meet these conditions, and therefore were not used in the analysis.

#### Seagrass Morphology and Biomass

Physical characteristics of the Z. marina bed (shoot density, blade length and width) were measured during each deployment at the Center site, which was considered representative of the established meadow area (**Table 1**). Density was determined in situ via 0.25 m<sup>2</sup> shoot counts, and shoots were brought to the laboratory to measure length and width. Lengths and widths were recorded for the longest three blades per shoot. The meadow was most dense with the longest and widest blades during the summer for both years sampled, with values that peaked during July 2013 with a density of 411 ± 33 shoots m−<sup>2</sup> and were lowest during January 2012 at 100 ± 36 shoots m−<sup>2</sup> . Summertime aboveground seagrass biomass has been measured in the South Bay seagrass meadow annually since 2009 (McGlathery, 2017), and was 107 ± 19 g m−<sup>2</sup> in 2012 and 173 ± 52 g m−<sup>2</sup> in 2013.

### Wave Attenuation

The effects of the seagrass meadow on wave development were made in three different ways. The first was by directly comparing the significant wave height at the Center site within the seagrass meadow to significant wave height measured simultaneously at TABLE 1 | Seasonal morphometric data summarizing characteristics of Z. marina meadow in South Bay, Virginia across five sampling periods from April 2012 through July 2013.


Density was determined as number of shoots per 0.0625 m<sup>2</sup> (25 cm by 25 cm) quadrat. Values presented below are mean ±1 SD. In each season, 10 randomly placed quadrats were measured.

the bare site. The second was through calculation of a wave attenuation coefficient (αw) for attenuation of significant wave heights as waves propagated across the meadow, while the third was comparisons of wave heights at specific locations within the seagrass meadow to predicted wave heights using the commonly applied Young and Verhagen (1996) wave height model.

Wave data were first filtered to include only sampling periods during which the wind direction was nearly parallel to a line between the sites and significant wave heights were >3 cm. Because waves in South Bay are wind driven, wind direction can be used as an approximation for the direction of wave propagation. A range of 45◦ (22.5◦ to either side of the direction between two sites) was used in the filter to provide adequate data for analysis of attenuation. After preliminary analysis, significantly lower wave heights were measured at the South site compared to the other two sites during periods when winds were blowing from the south. The South site has limited fetch for south winds with a distance of approximately 1 to 1.5 km. As a result of these short fetch lengths, waves are not fully developed when they reach the South site, and the processes of wave growth due to wind energy and decay due to seagrass induced drag compete in a way that cannot be easily predicted. Due to this limitation, only data for time periods of north winds (blowing north to south) was used in the computation of wave attenuation.

A wave attenuation coefficient, αw, was calculated using the formulation from Bretschneider (1954) for waves propagating across an impermeable bottom with constant depth and no refraction:

$$H\_2 = \left. H\_1 \left[ \frac{f H\_1 \phi\_f \Delta x}{K\_s T^4} + 1 \right]^{-1} \right. \tag{1}$$

where H<sup>1</sup> and H<sup>2</sup> are the initial and final wave heights, f is the friction factor, 1x is the horizontal distance the wave traversed, T is the wave period, K<sup>s</sup> is the shoaling factor, and φ<sup>f</sup> is the bottom friction dissipation function [s<sup>4</sup> m−<sup>2</sup> ], defined as:

$$\phi\_{\emptyset} = \frac{64\pi^3}{3g^2} \left[ \frac{K\_s}{\sinh\left(\frac{2\pi h}{L}\right)} \right]^3 \tag{2}$$

where L is the wavelength and h is the water depth. Bretschneider (1954) describe f as a dimensionless "calibrated friction factor" which describes the reduction in wave height caused by bottom characteristics. This friction factor, f, can be rewritten as a measure of wave induced drag (Mazda et al., 1997), αw, where Equation 1 then becomes:

$$\frac{H\_2}{H\_1} = \frac{1}{1 + \frac{\pi^5 K\_i^2}{\sqrt{2}g^2 T^4} \* \alpha\_w H\_1 \Delta \pi \left(\sinh \frac{2\pi h}{L}\right)^{-3}}\tag{3}$$

This method estimates flow resistance across the entire height of the plant structure as bottom friction. Equation 3 can be simplified for South Bay by setting the shoaling factor (Ks) equal to 1 because South Bay is roughly a constant depth (with mean water depths ranging between 1.0 and 1.2 m) and does not typically induce wave shoaling. Solving for αw:

$$\alpha\_{\mathcal{W}} = \frac{g^2 \sqrt{2} T^4}{\pi^5} \ast \frac{\left(\sinh\frac{2\pi h}{L}\right)^3}{H\_1 \Delta x} \ast \left(\frac{H\_1}{H\_2} - 1\right) \tag{4}$$

Mazda et al. (1997) called this α<sup>w</sup> parameter a coefficient of drag (CD), however in mixed wave-current dominated systems, it is not a true measure of integrated drag on the flow, but only the effects of the frictional resistance on waves. We therefore use α<sup>w</sup> to indicate this parameter is solely a function of wave activity.

A wave height model, created by Young and Verhagen (1996), was applied to predict wave heights within South Bay assuming no seagrass is present. The Young and Verhagen (1996) model (abbreviated as YV) uses semi-empirical relationships that relate fetch length, wind velocity, and water depth to significant wave height and peak wave period. This commonly applied fetch-limited, finite-depth model was calibrated for fine-grained, cohesive mud within a 1–2 m depth shallow lake, and therefore is a good approximation for the sediment and physical conditions in South Bay in the absence of vegetation. In addition, this model has been previously validated for use in the coastal bay system of the VCR (Mariotti and Fagherazzi, 2013b; Mcloughlin et al., 2015). The significant wave H<sup>s</sup> and peak wave period T<sup>p</sup> were computed as:

$$\frac{gH\_s}{\left(U\_{wind}\right)^2} = 0.2413 \left\{ \tanh A\_1 \tanh \left[\frac{B\_1}{\tanh A\_1} \right] \right\}^{0.87} \tag{5}$$

$$\frac{gT\_p}{U\_{wind}} = 7.518 \left\{ \tanh A\_2 \tanh \left[ \frac{B\_2}{\tanh A\_2} \right] \right\}^{0.37} \tag{6}$$

with A<sup>1</sup> = 0.493 gh/ [Uwind] 2 0.75 , B<sup>1</sup> = 3.13 × 10−<sup>3</sup> gχ/ [Uwind] 2 0.57 , A<sup>2</sup> = 0.331 gh/ [Uwind] 2 1.01, and <sup>B</sup><sup>2</sup> <sup>=</sup> 5.215 <sup>×</sup> <sup>10</sup>−<sup>4</sup> gχ/ [Uwind] 2 0.73, where h is the depth, χ is the fetch, and Uwind is the reference wind speed. Fetch was determined from aerial photographs of the VCR LTER bays, wind velocity and direction were determined from the Wachapreague NOAA meteorological station, and water depth was determined from the wave gages following the atmospheric pressure correction. The model output, assuming no seagrass was present, was then compared to field data collected within the seagrass site to determine the effects of the seagrass meadow on wave attenuation. Model output was also compared to measured wave heights at the bare site.

Observed waves with significant wave heights below 3 cm were filtered from the data set to reduce contamination from small turbulent fluctuations that could artificially skew attenuation values. Small values below this cutoff may be the result of random non-wave motion, and are unlikely to be influenced by bottom drag (Wiberg and Sherwood, 2008). Model outputs were compared to observations collected in South Bay, and the percent difference between the modeled and measured values was calculated as follows:

$$Percent\text{ Difference} = \frac{H\_{\text{s, model}} - H\_{\text{s, observed}}}{\left(\frac{(H\_{\text{s, model}} + H\_{\text{s, observed}})}{2}\right)} \ast 100\tag{7}$$

Positive values indicate that the modeled wave heights predicted over an unvegetated seafloor were greater than the in situ observations of wave heights measured in the presence of seagrass. Modeled data was also computed and compared to observed wave heights measured at the bare site. In addition, Equation 7 was used to quantify the percent difference in observed wave heights measured simultaneously at the seagrass and bare sites to determine in situ values of wave height reduction.

#### Bed Shear and Friction Velocity

High resolution velocity measurements using a Nortek© Vectrino II acoustic Doppler profiler were used to quantify bed shear and friction velocities at the bare and seagrass sites. In combined wave-current flow, both waves and currents can induce shear at the sediment-water interface. Bed shear is often estimated in current-dominated regimes by the magnitude of turbulent energy within the overlying boundary layer or through quantification of the logarithmic profile of the mean flow (Reidenbach et al., 2006). However, the presence of vegetation often disrupts the development of a logarithmic profile within the mean flow and alters the shear at the sediment-water interface (Hansen and Reidenbach 2013). In addition, wave orbitals induce large variances in the horizontal and vertical velocity components, which mask velocity variances due to turbulence, making it necessary to separate the wave and turbulence parts of the signal. For wave-current flows, the components of the total instantaneous velocity u (as well as for v and w, not shown) can be separated via Reynolds decomposition according to:

$$
\mu = \overline{\mu} + \tilde{\mu} + \mu' \tag{8}
$$

where u is the mean current velocity, u˜ is the instantaneous wave orbital velocity, and u ′ is the instantaneous turbulent velocity. Current, wave and turbulence perturbations were separated using a spectral decomposition technique called the Phase method (Bricker and Monismith, 2007), which converts the temporal velocity record into frequency space via Fourier transformation and isolates the wave stress using the phase lag between horizontal and vertical components of the velocity signal. Further details of the Phase method can be found in Bricker and Monismith (2007) and Stocking et al. (2016).

Bed shear stress τbed was calculated using the formulation of Wiberg and Smith (1983), defined as:

$$
\pi\_{bed} = \sqrt{\mathfrak{r}\_c^2 + \mathfrak{r}\_w^2} \tag{9}
$$

where τ<sup>c</sup> is bed shear due only to currents and τ<sup>w</sup> is bed shear due only to wave orbital motions. The current-induced bed shear stress is estimated using near-bed turbulent kinetic energy (TKE) outside the wave boundary layer, according to Soulsby and Dyer (1981):

$$
\sigma\_{\mathfrak{c}} = 0.19 \,\rho TKE^2 \tag{10}
$$

where ρ is the density of seawater and:

$$TKE = \frac{1}{2} (\overline{\mu'^2} + \overline{\nu'^2} + \overline{\nu'^2}) \tag{11}$$

The wave-induced contribution is parameterized using the nearbottom wave orbital velocity, uorb, computed by quantifying motions only within the wave band of the velocity spectra measured at z= 0.02 m above the seafloor, and a wave friction factor f<sup>w</sup> (Jonsson, 1967):

$$
\pi\_{\text{wave}} = \frac{1}{2} f\_{\text{w}} \rho u\_{\text{orb}}^2 \qquad f\_{\text{w}} = 0.04 \left[ \frac{u\_{\text{orb}} T}{2 \pi k\_b} \right]^{-0.25} \tag{12}
$$

where ρ is the fluid density, and k<sup>b</sup> is the characteristic roughness length (Fagherazzi and Wiberg, 2009). Lawson et al. (2007) estimated k<sup>b</sup> as 3D84, where D<sup>84</sup> is the sediment grain diameter (D) at which 84% of the sample is finer than D. D<sup>84</sup> measured at the seagrass site ranged between 151 ± 1µm and 158 ± 3µm, while the unvegetated site ranged from 152 ± 4µm to 162 ± 2µm.

#### RESULTS

Wind speed for each deployment is shown in **Figure 2A**. Seasonal averages ranged from 2.1 m s−<sup>1</sup> in July 2012 to 4.8 m s−<sup>1</sup> in July 2013; however, the deployments captured enough variation that all seasonal means were within one standard deviation. The maximum 10-min averaged wind speed was 27.8 m s−<sup>1</sup> . No seasonal trends were apparent in the data; although the second and third highest averages both occurred in April, possibly indicating stormier weather and the potential for larger wave development in the spring.

The seasonal average depth, measured at the Center site ranged between 0.93 m in January 2013 and 1.28 m in October 2011, with a standard deviation of 0.46 and 0.42 m, respectively (**Figure 2B**). A semi-diurnal tide drives the variation in water depth in South Bay, so the standard deviation is affected by the tidal range. At the bare site, the mean depth ranged from 0.82 m in January 2013 to 1.19 m in October 2012. The mean depths at the bare and seagrass sites were within one standard deviation for all seasons, with a maximum difference of 0.11 m in the mean seasonal values. Seasonal differences in mean water depth are a result of weather systems; including wind or barometric pressure variations which may force more water toward or away from the shoreline. Seasonally averaged mean water velocities ranged from 3.5 to 7.1 cm s−<sup>1</sup> at the seagrass site, with significantly higher mean seasonal velocities (ANOVA, p < 0.01) at the bare site of 11.9–25.9 cm s−<sup>1</sup> .

#### Wave Characteristics

Waves in South Bay are primarily wind driven, with seasonally averaged significant wave heights ranging from 0.02 m within the seagrass bed to 0.09 m at the bare site (**Figure 3A**). Seasonally averaged significant wave periods were between 1.6 and 2.4 s (**Figure 3B**). The distribution of significant wave height was highly skewed, with intermittent, large waves occurring during storm events. Consequently, the maximum 10-min significant wave heights for each season were much higher than the average value, with the largest H<sup>s</sup> across all deployments reaching 0.55 m. In addition, significant wave height varied considerably as a function of wind direction due to altered fetch lengths.

Average H<sup>s</sup> was statistically lower within the seagrass meadow compared to the bare site during all seasons (ANOVA, p < 0.01). As predicted by the seasonal mean wind values, the high wind speeds in April 2013 and July 2013 correspond to the largest mean significant wave heights at the bare site, however the mean significant wave height at the vegetated site follows a trend more aligned with the seasonal shoot density distribution, with maximum values when densities were low. Average significant wave period, T<sup>s</sup> , was slightly higher at the vegetated site across all seasons except during October 2012, although the difference was not statistically significant. Controls on significant wave

height include; fetch length, wind speed, water depth, and wave attenuation, which varies as a property of meadow structure. The fetch length, wind speed, and water depth were all statistically similar at the bare and seagrass sites, and therefore the difference in apparent wave properties is primarily due to the presence of seagrass.

#### Wave Attenuation

Plots of significant wave height at the bare vs. vegetated site are displayed across four consecutive seasons (**Figure 4**). Data points that fall below the 1:1 line indicate wave attenuation at the seagrass site, and the slope of the regression line represents the degree to which wave height is attenuated by the seagrass bed (adapted from Koch 2006). Overall, significant wave height at the seagrass site was 51% of significant wave height at the bare site in the summer, 59% in the fall, 75% in the winter, and 53% in the spring. Mean water depths for these time periods are shown in **Figure 2B**, which range from 0.82 to 1.18 m across seasons. Although there is seasonal variability in the mean depths due to variability in tidal and atmospheric forcings, within each season mean water depths at the seagrass and bare sites are very similar. Data from the summer and fall 2012 show greater variation and therefore lower R 2 -values (R 2 = 0.50 and 0.60, respectively), compared to winter and spring

2013 (R <sup>2</sup> = 0.75 and 0.70, respectively). This may be a result of the changes in seagrass cover during the sampling period due to growth or senescence of the seagrass. In the winter, the bed topography of the less dense meadow matches more uniformly with the surrounding unvegetated region, causing a line of best fit that better matches the 1:1 line of no attenuation (R <sup>2</sup> = 0.75, slope = 0.75).

#### Wave Attenuation Coefficients

Attenuation coefficients for South Bay averaged across seasonal deployments varied from α<sup>w</sup> = 0.19 to 0.49 (**Table 2**). The largest seasonally averaged wave attenuation coefficients occurred in spring, while the smallest occurred in winter when seagrass density was lowest. However, there is large variability in attenuation coefficients that are dependent upon the wave environment. Plots of wave frequency vs. α<sup>w</sup> show a trend of higher attenuation coefficients for lower frequency waves during each season (**Figures 5A,B** shown for summer 2012 and winter 2013, respectively). Linear wave theory suggests that waves will attenuate prior to reaching the sea floor when the wave frequency (f) is greater than p g/(4πh) (Wiberg and Sherwood, 2008). For the water depths typical in South Bay (1–2 m), this relationship suggests that waves with a frequency >0.63–0.88 s−<sup>1</sup> , depending on the tidal conditions, will not interact with the seafloor. This is consistent with the decrease in attenuation coefficients observed as wave frequencies increase, approaching α<sup>w</sup> = 0 as frequencies surpass 0.75 s−<sup>1</sup> (**Figures 5A,B**). Negative α<sup>w</sup> values for large wave frequencies, indicate waves that were likely not interacting with the seafloor, and therefore represent waves that were still increasing in wave height due to wind energy input. Since winds are continually adding energy to the waves as they propagate across the shallow bay system, the actual wave energy dissipated

FIGURE 4 | Significant wave height at the center vegetated site compared to significant wave height at the nearby bare site over four seasons (A) summer 2012, (B) fall 2012, (C) winter 2013, and (D) spring 2013. The slope of the regression line represents the degree of wave height attenuation between these two sites, with shallower slopes indicating greater attenuation. The regression line was fit to pass through the origin. Black line represents 1:1 slope line, indicating a case with no attenuation between the bare and vegetated sites.

TABLE 2 | Seasonally averaged mean wave attenuation coefficients (± standard error) for north winds.


The years indicate when data were collected and averaged. N is number of 10-min averaged values included in the seasonal average.

by the seafloor and seagrass canopy is likely greater than these wave attenuation coefficients suggest.

Wave attenuation increased with water depth (**Figures 5C,D** for summer 2012 and winter 2013, respectively), which was consistent across all seasons. Negative α<sup>w</sup> values for small wave heights, similar to described above for small frequency waves, were likely not interacting with the bottom, and therefore represent waves that were still growing in wave height. Larger significant wave heights generally resulted in increased wave attenuation coefficients (**Figure 6**), however the scatter in these data were greater than those of wave frequency vs. αw. Overall, there is a strong correlation between wave period and water depth in the seagrass meadow, as well as significant wave height and water depth (**Figure 7**). In general, these findings suggest that larger waves with longer wave periods can form when the water is deeper, and because of this wave orbitals penetrate deeper in to the water column and interact with the seagrass and seafloor to a greater extent. Although not directly computed, the wavelength, L, can be calculated according to linear wave theory for intermediate

FIGURE 5 | Wave frequency vs. wave attenuation coefficient (αw) for north winds during (A) summer 2012 and (B) winter 2013. Raw data are presented in green points. Data were averaged in 0.1 Hz bins and averages were plotted along with the raw data as gray diamonds. Water depth vs. wave attenuation coefficient (αw) for north winds during (C) summer 2012 and (D) winter 2013. Raw data are presented as purple points. Depth data were averaged in 0.01 m bins and plotted along with the raw data as gray diamonds. Means that had fewer than 10 data points in a bin were excluded from the plots. Error bars on the averaged data show standard error.

waves as: L = L<sup>∞</sup> q tanh( <sup>2</sup>π<sup>h</sup> L∞ ), where L<sup>∞</sup> = g 2π T 2 . This relationship suggests that with increases in wave period, waves with longer wavelengths also form in South Bay, and therefore increased wave attenuation occurs with larger wavelengths.

# Comparison to Young and Verhagen Wave Height Model

An empirical model developed by Young and Verhagen (1996) was utilized to predict significant wave height in South Bay as a function of wind speed, water depth, and fetch length, assuming an unvegetated seafloor composed of fine-grained sediments. Model accuracy was verified by comparing model output values of wave height to the measured values at the bare site. Although there is inherent variability in model performance over short timescales when comparing modeled vs. observed wave heights at the bare site (**Figures 8A,B**), when seasonally averaged the model both over- and under- predicted wave heights within ±15% of observed values (**Table 3**). This suggests reasonably good agreement and calibration of the model to wave dynamics in South Bay given uncertainty in water depth and bathymetry across the bare site, changes in wind direction that impact fetch length, and the likely presence of some small amounts vegetation at the bare site. Young and Verhagen (1996) do not report a grain size for the sediment used to calibrate their model, but describe the bottom as "relatively fine grained but cohesive mud" in which "the bed is not mobile and ripples do not develop." This description is consistent with the fine-grained sediment in South Bay. Utilizing the same wind speed, water depth, and fetch length as that measured in South Bay, the difference between the model output and the recorded significant wave height was considered

FIGURE 6 | Significant wave height (Hs) vs. wave attenuation coefficient (αw) for north winds across (A) April 2012, (B) July 2012, (C) October 2012, (D) January 2013. Wave height data were averaged in 0.01 m bins and plotted along with the raw data as gray diamonds. Bins that had fewer than 10 data points were excluded. Error bars indicate standard error.

to represent the wave height attenuation caused by seagrass compared to a theoretical unvegetated site at the same physical location (**Figures 8C,D**). This comparison showed wave height attenuation caused by the seagrass of up to 44.8% at the Center site during summer, with minimum wave height attenuation of 12.3% during winter (**Table 3**). Seasonally averaged attenuation estimates from in situ measurements, model comparisons, and attenuation coefficients are summarized in **Table 4**.

#### Bed Shear and Friction Velocity

Near bed velocity profiles were collected over a range of flow conditions in nine separate deployments at the bare and center seagrass site. Measurements within the seagrass bed were performed during summer 2012 and 2013 months when the meadow was near peak biomass, with blade densities of 347 ± 73 shoots/m<sup>2</sup> and 411 ± 33 shoots/m<sup>2</sup> , respectively. Bare site measurements were conducted during fall 2012 and summer 2013. Significant wave height averages at the bare site ranged from 0 to 0.11 m, and H<sup>s</sup> values between 0 m and 0.08 m at the seagrass site. These wave heights were relatively small, but encompassed the average H<sup>s</sup> range of 0.02 to 0.09 m observed during seasonal deployments in South Bay. Within each deployment, data was collected over approximately 3–4 h. However, to determine the effects of hydrodynamics on near bed boundary layer flows, the deployments with greatest and least

measured bed shear stress were selected from both the bare and vegetated site for further analysis (**Table 5**).

At the bare site, the time period with the lowest bed shear stress, τbed = 0.019 ± 0.008 Pa, was recorded in fall 2012. This time period coincided with essentially no wave activity, with mean velocities across the 1 h deployment of 3.9 cm s−<sup>1</sup> at 0.5 cm above the bed and 4.8 cm s−<sup>1</sup> at 2.5 cm above the bed. The time period of highest τbed = 0.07 ± 0.034 Pa occurred at peak wave conditions with Hs= 0.11 m. The critical shear stress for sediment resuspension in South Bay was previously found to be τcrit = 0.04 Pa with a D<sup>50</sup> ≈ 63µm and sorting coefficient <sup>√</sup> D84/D<sup>50</sup> ≈ 2 (Lawson et al., 2007; Fagherazzi and Wiberg, 2009), suggesting that sediment resuspension was actively occurring during higher wave conditions. In contrast, at the seagrass site, the lowest measured τbed = 0.004 ± 0.003 Pa, while the highest was τbed = 0.034 ± 0.022 Pa which occurred during peak wave conditions of Hs=0.08 m, suggesting that at no time during these summertime velocity measurements did the bed shear stress exceed τcrit to initiate sediment resuspension. These findings suggest the seagrass meadow does exert significant control over the hydrodynamic conditions at the sediment-water interface, and this control is due to the attenuation of wave motion by drag induced from the seagrass over the expanse of the meadow.

#### DISCUSSION

This work presents several estimations of wave attenuation induced by a temperate Z. marina meadow across a range of seagrass density and morphologic variations. Wave height attenuation, both comparing a vegetated site to a bare site and comparing measured values to model predictions, showed large seasonal variations with the largest reduction occurring during spring and summer when the meadow was near peak density, and the lowest reduction during the winter (**Table 4**). Direct comparisons of wave heights showed reductions between 25 and 49% in the seagrass bed compared to an adjacent bare site. Results using the Young and Verhagen (1996) model showed similar seasonal trends, although model output consistently estimated less attenuation within the seagrass bed than direct in situ measurements would suggest. Seasonal estimates of wave heights predicted over the bare site using the YV model ranged between 12.7% lower and 14.7% greater than in situ wave heights (**Table 3**), with modeled significant wave heights being on average 5% greater than those measured in situ when averaged over annual timescales.

As expected, large variability in wave height attenuation was found within the seagrass canopy due to the seasonal variation in seagrass biomass, changes in water depth due to tidal fluctuations, and high variability in fetch length in response to changes in wind direction. Seagrass morphology varied from a minimum blade length of 19 ± 4 cm, blade width of 0.21 ± 0.04 cm, and shoot density of 100 ± 36 blades m−<sup>2</sup> in winter, to peak blade length of 53 ± 8 cm, blade width of 0.41 ± 0.03 cm, and shoot density of 411 ± 33 blades m−<sup>2</sup> in summer. Increased seagrass blade size and shoot density correlated directly with increased wave attenuation across the seagrass meadow.

The distance to the top of the seagrass bed, where waves will begin to feel drag, also decreases in the summer when the canopy is established with longer blades. For both the unvegetated and seagrass sites, wave theory predicts that waves with a wave frequency, f > p g/(4πh) will be attenuated before reaching the bottom (Wiberg and Sherwood, 2008). For the range of depths found within the seagrass meadow and bare site, oscillatory motion will generally not reach the seafloor for wave frequencies ≥1 Hz. Although high frequency waves may interact with the vegetation, and experience drag even when the wave motions are too small to reach the sea floor, our results (**Figure 5**) suggest that wave attenuation when wave frequencies ≥1 Hz is minor. The maximum α<sup>w</sup> value occurs at a wave frequency of approximately 0.55 Hz for every season, corresponding to a wave period of 1.8 s. The averaged α<sup>w</sup> value for this wave period is as high as 1.3 during summer, and drops to 0.7 during winter. This suggests that variability in the seagrass biomass with season is a dominant contributor to wave attenuation. Unlike wavelength and wave period, which provides a lengthand time-scale of a wave and results in a clear relationship for when waves will interact with the sea floor, there is no

TABLE 3 | Percent difference between observed significant wave heights and significant wave heights predicted by the Young and Verhagen (1996) model at each site for southward blowing winds, averaged across each season.

Center seagrass site during April 2013. (D) Time series of percent attenuation between measured and modeled wave height at the seagrass site.


theoretical relationship between wave height and when a wave of a certain height will experience bottom drag. However, there are theoretical limits to wave steepness, which define a maximum wave height for any given wave period (Dean and Dalrymple, 1991).

The relationship between water depth and wave attenuation coefficient first appears counterintuitive, showing an increase in α<sup>w</sup> at greater depths (**Figure 5**). Since waves in South Bay are locally generated, wave development is controlled by a combination of water depth, wind speed, and fetch length (Young and Verhagen, 1996). The shallow water depth essentially limits the development of large waves, causing overall smaller significant wave heights and wave periods at low tides, with relatively larger waves formed during high tide or during storm events when storm surge increases water depths. During these periods of increased water depths, longer period waves are formed which can then interact with the seafloor and seagrass, causing greater attenuation coefficients. Although attenuation coefficients were estimated only for significant wave height and period, there is a range of wave heights and frequencies formed at any given time within coastal bay system. Results of Bradley and Houser (2009) found that seagrasses serve as a low-pass filter, where higher frequencies in the spectra tend to be more attenuated. Their results suggest that the rate of energy dissipation is not uniform over a range of wave frequencies, and waves at higher frequencies are attenuated, but waves at lower frequencies are less affected by the seagrass.

Velocity measurements adjacent to the sediment-water interface showed substantial reduction in bed shear stresses within the seagrass meadow compared to the bare site. Within the bare site, τbed ranged from 0.019 to 0.07, while in the seagrass bed τbed ranged from 0.004 to 0.034 Pa. The critical stress threshold to initiate sediment resuspension was found to be τbed = 0.04 Pa (Lawson et al., 2007), suggesting that during summer, the seagrass canopy limits sediment from being suspended, although these measurements occurred over a limited time frame. The relative reduction of wave orbital motion caused by the canopy can be estimated using the ratio of wave orbital excursion length (A) to blade spacing (S), where A = uorb/ω, ω = 2π/T, and S as the characteristic spacing between shoots (Lowe et al., 2005). For A/S >1 orbital attenuation within the canopy is expected to be significant. Utilizing blade density and average blade width, blade spacing of the seagrass bed ranges from S = 8.2 cm in winter to 4.5 cm in summer, giving approximate estimates of A/S = 0.1 in winter to 0.2 in summer, assuming wave periods of T = 1.5 s and <sup>u</sup>orb <sup>=</sup> 2.9 cm s−<sup>1</sup> (**Table 5**). This indicates that orbital motions are not significantly altered due to direct interaction with the seagrass blades and oscillatory water motion is able to effectively penetrate the seagrass canopy, even though tidallydriven flows may be damped. This finding is similar to laboratory measurements within a model Z. marina meadow (Luhar et al., 2010), where it was found that unidirectional flows were reduced within the meadow but in-canopy orbital velocities were not significantly altered.

Complex bathymetry within the coastal bay and barrier island system provided a wide variety of fetch lengths to consider. Due to this variability, only north winds within a 45◦ wind range around true north were analyzed. The Young and Verhagen


Multiple seasons were sampled, however values from different years are averaged together to determine representative seasonal statistics. For consistency, both in situ and YV model average attenuation are computed at the Center seagrass site. α<sup>w</sup> is listed as mean ± standard error.

(1996) model predicted wave heights greater than those measured within the seagrass meadow during all seasons, indicating that actual wave heights were reduced compared to estimates of wave heights that would occur across an unvegetated South Bay. Although of comparable magnitude, direct in situ measurements of reductions in significant wave heights compared to the bare site were greater than wave height reductions compared to estimates from YV model predictions (**Table 3**). This deviation between in situ and modeled wave heights was consistent across seasons. Wave heights measured within the seagrass bed during winter were closest to YV model results, with an average reduction of 13%. This also corresponded with the lowest measured meadow density was lowest.

Overall, the YV model predicted wave heights that were within ±15% of those directly measured at the bare site in South Bay. Variations in model output compared to direct in situ measurements of wave height could be the result of multiple influences, including the coarseness of fetch distances input to the model, discrepancies in the magnitude of winds that occurred over South Bay from measured winds, or due to variations in sediment properties. Mariotti and Fagherazzi (2013a) found that the Young and Verhagen (1996) model under-predicted wave heights in a study in Willapa Bay, Washington. They attributed this to an underestimation of fetch-limited wave growth to the smooth mud substrate in the bay, which they hypothesize induced less drag on the flow than the sediment used in the formulation of the YV model. They found that a low friction coefficient for the sediment was necessary to match the observed wave dynamics. This highlights the importance of drag on wave attenuation, as even small differences in sediment characteristics can have a measurable effect on wave height, even in the absence of vegetation. Young and Verhagen (1996) do not report a grain size for the sediment used to calibrate the model, but describe the bottom as fine grained but cohesive mud. While the sediment in South Bay is fine-grained (Hansen and Reidenbach, 2012), it may not be as cohesive, as it was easily resuspended. This discrepancy could result in higher bottom friction in South Bay, and would explain wave height over-prediction by the model.

# CONCLUSION

A seagrass canopy within a shallow coastal bay was found to have a large impact on wave development, which varied seasonally in response to seagrass morphology and density. Although the largest wave attenuation compared to an adjacent

TABLE 5 | Wave properties and environmental conditions for Vectrino profiler deployments with lowest and highest wave activity at the bare and vegetated site.


Temperature, water depth, Hs, Hmax , and T<sup>s</sup> data were obtained from wave gages, while velocities and τbed (±1 standard deviation) were obtained from the Vectrino profiler.

Reidenbach and Thomas Wave Attenuation in Seagrass

bare site occurred in summer when seagrass biomass was greatest, the largest wave attenuation coefficient, measured as wave decay as waves propagated across the seagrass canopy, occurred in spring, with α<sup>w</sup> = 0.49. This discrepancy is likely due to the large impact that physical conditions, including water depth and wind magnitude, has on wave development and attenuation and signifies that a single mean attenuation coefficient should be used with caution. α<sup>w</sup> is highly variable with respect to water depth, wind speed, and seagrass biomass. During summer, Hansen and Reidenbach (2012) found a 45– 70% reduction in wave height between a seagrass meadow and an unvegetated region in South Bay, with sites ranging from a seagrass density of 150 ± 80 shoots m−<sup>2</sup> to 570 ± 70 shoots m−<sup>2</sup> . The average reduction in wave height compared to a bare site of ∼50% for this study falls into this range, with summer densities of 347 ± 73 and 411 ± 33 shoots m−<sup>2</sup> . Comparing a wide range of wave conditions, bed shear stresses measured at the sediment water interface were statistically lower at the vegetated compared to the unvegetated site, even though scaling of wave motions in the presence of seagrass suggest that wave

#### REFERENCES


orbitals should be locally unaffected by the seagrass blades, with waves being able to penetrate through the canopy. Since substantial net attenuation of wave energy as waves propagate across the meadow was found, this suggests that the integrated meadow-scale fluid drag controls the apparent decrease in bed shear within seagrass meadows compared to an unvegetated seafloor.

#### AUTHOR CONTRIBUTIONS

MR conceptualized the study. ET performed the experiments. MR and ET analyzed the data and wrote the manuscript.

#### ACKNOWLEDGMENTS

We thank A. Schwarzschild, C. Buck, and D. Boyd for field assistance. This research was funded by the National Science Foundation (NSF-DEB 1237733) to the Virginia Coast Reserve Long Term Ecological Research program and by NSF-OCE 1151314 to MR.


**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 Reidenbach and Thomas. 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.

# Canopy Functions of R. maritima and Z. marina in the Chesapeake Bay

Emily French\* † and Kenneth Moore

Virginia Institute of Marine Science, College of William and Mary, Gloucester Point, VA, United States

Shoots in seagrass beds form canopies: structurally complex habitats that provide refuge for fauna and trap sediment particles by dampening water movement. Unfortunately, seagrasses are faced with continuing negative impacts to survival, including climate change and poor water quality. In areas where several seagrass species coexist, changing conditions may influence composition of beds so one species is favored over another. Two species found worldwide, Zostera marina and Ruppia maritima, are undergoing this shift: as Z. marina dies back, in some locations it is replaced by R. maritima, a smaller-form seagrass with shorter, thinner shoots. This process is occurring in Virginia, United States in the southern Chesapeake Bay, at intermediate depths where the species co-occur. Although changes in seagrass species abundance have previously been documented, few studies have measured the resulting effects on ecosystem functioning. We evaluated three sites to determine whether canopies of the two species displayed similar small epifaunal invertebrate animal assemblages and sediment properties, and found that Z. marina beds exhibited a greater amount of fine surface sediment than those of R. maritima, but found no effect of seagrass species on invertebrate assemblages. Epifaunal invertebrates were, however, more abundant and speciose with greater biomass, and more abundant with greater shoot density. This study provides baseline information from one summer for areas where the two seagrass species coexist. Although more research is needed, this study suggests in mixed beds, decline of Z. marina could result in coarsening of sediment, but dense R. maritima canopies could harbor similar small invertebrate assemblages.

Keywords: seagrass, eelgrass, Chesapeake, Chesapeake Bay, climate change, water quality, epifauna, widgeongrass

### INTRODUCTION

Seagrass canopies provide habitat for marine life, including small crustacean and gastropod invertebrates that are part of the diet of nearshore fishes (Orth et al., 1984; Valentine and Duffy, 2006). Seagrasses also attenuate waves and reduce currents, causing fine particles to be deposited and retained within beds, creating a positive feedback loop in which water clarity is improved for seagrass growth (Ward et al., 1984; Hansen and Reidenbach, 2013). Seagrasses are currently in decline due to decreased water quality and climate change (Cardoso et al., 2004; Orth et al., 2006), and beds in many areas could shift to being dominated by macroalgae or other seagrass species better able to cope with changing or degraded conditions (Armitage et al., 2011; van Tussenbroek et al., 2014). When a shift occurs between morphologically similar species, ecosystem functioning may not change (Christiaen et al., 2016), however, many co-occurring seagrasses have differing morphologies and life cycles. Several regions in the

#### Edited by:

Susana Carvalho, King Abdullah University of Science and Technology, Saudi Arabia

#### Reviewed by:

Juan Moreira Da Rocha, Universidad Autónoma de Madrid, Spain Marina Dolbeth, Centro Interdisciplinar de Investigação Marinha e Ambiental (CIIMAR), Portugal

\*Correspondence:

Emily French emily.d.french@gmail.com

#### †Present address:

Emily French, Oak Ridge Institute for Science and Education, Environmental Protection Agency, Washington, DC, United States

#### Specialty section:

This article was submitted to Marine Ecosystem Ecology, a section of the journal Frontiers in Marine Science

Received: 01 September 2018 Accepted: 15 November 2018 Published: 03 December 2018

#### Citation:

French E and Moore K (2018) Canopy Functions of R. maritima and Z. marina in the Chesapeake Bay. Front. Mar. Sci. 5:461. doi: 10.3389/fmars.2018.00461

**45**

United States have experienced seagrass habitats transitioning from larger-form, spatially stable species to more opportunistic, smaller-form species (Fourqurean et al., 1995; Johnson et al., 2003; Bologna et al., 2007; Cho et al., 2009; Lopez-Calderon et al., 2010; Micheli et al., 2014).

In Chesapeake Bay, two seagrasses co-occur: Zostera marina and Ruppia maritima. Z. marina has taller canopies, wider straplike leaves and spatially stable populations (Orth and Moore, 1988; Moore et al., 2014); comparatively, R. maritima has shorter canopies, smaller leaf areas, and unpredictable annual abundances (Kantrud, 1991; Orth et al., 2016). These species coexist in other regions of the mid-Atlantic United States, and in California and Maine (Orth and Moore, 1988; Kantrud, 1991; Johnson et al., 2003). Z. marina in Chesapeake Bay is in decline and its recovery problematic (Moore et al., 2012; Lefcheck et al., 2017).

Seagrass canopy structure may play a role in epifaunal community composition and sediment deposition (Orth et al., 1984; Fonseca and Callahan, 1992), but these effects are incompletely understood. In this study we compared surface sediment characteristics and small invertebrate assemblages in canopies of each seagrass to determine whether a shift from Z. marina to R. maritima could result in changes in seagrass bed features. We hypothesized that Z. marina would have greater canopy biomass than R. maritima over the summer sampling took place, and that Z. marina areas would contain more fine sediment and organic matter and provide refuge for more abundant and speciose invertebrates.

#### MATERIALS AND METHODS

#### Study Locations

Three sites in the lower Chesapeake Bay (**Supplementary Figure S1**) with monotypic stands of Ruppia maritima and Zostera marina nearby to one another (50–300 m) in similar water depths (50 ± 10 cm MLLW) were sampled. The sites were physically diverse: one is located within an embayment with surrounding salt marsh, while the others are more exposed to prevailing winds and tidal currents, although all possess similar salinity and turbidity measurements<sup>1</sup> . Sites were within areas where an increase in R. maritima and decrease in Z. marina populations has occurred in recent years (Moore et al., 2014; Richardson et al., 2018).

#### Sample Collection Overview

Seagrass, sediment and invertebrate animal samples were taken in each of the habitat types (monospecific stands of Z. marina or R. maritima) during 2 months: June and August of 2013. Sampling timing was adapted from the NERRS (National Estuarine Research Reserve System) Vegetation Monitoring Protocols, which specify sampling within 2–3 weeks of peak biomass for seagrasses present (Moore, 2013). Here Z. marina increases in density during the late spring and senesces though mid-summer and into fall (Orth and Moore, 1986; Moore and Jarvis, 2008) and R. maritima has a similar trajectory, although it tends to peak in the fall (Moore et al., 2000). Replicate sampling areas were haphazardly chosen by throwing a meter-square quadrat within 10 m of a pole marking habitat type, samples were taken within the quadrat. HOBO temperature loggers (Onset Computer Corporation, Bourne, MA, United States) were anchored between the two habitat types at each site, where they recorded temperature every 15 min between 1 June and 26 August 2013.

# Seagrass Biomass Sampling and Processing

Seagrass samples were taken with a 12 cm diameter acrylic core, sieved in the field to remove sediment, and transported back on ice to the laboratory where aboveground material was separated from roots and dried at 65◦C until reaching a constant weight.

### Sediment Sampling and Processing

Samples for grain size and sediment organic matter (SOM) were taken using a 7 cm diameter acrylic core, and transported back to the lab on ice, where the top (0–2 cm) layer of sediment was processed by homogenizing samples, separating sand from the sample, and using a pipetting method to determine fine (silt and clay) fractions (modification of Plumb, 1981). Silt and clay fractions were combined for a measurement of fine sediment, while the remaining sand represented coarse sediment. SOM was determined via the loss-on-ignition method (Erftemeijer and Koch, 2001) by drying samples at 65◦C oven, then combusting at 500◦C for 5 h.

## Epifaunal Sampling and Processing

Samples were collected using a Virnstein grab (Virnstein and Howard, 1987), which closes over a 400 cm<sup>2</sup> area of sediment surface, collecting seagrass shoots, any overlying macroalgae, and epifauna without collecting sediment. During processing, epifauna were separated from seagrass shoots and macroalgae, shoots were counted, animals were sieved using 0.5 mm mesh, and all invertebrates were identified to species or genus levels. Seagrass and macroalgae was dried at 65◦C until reaching a constant weight.

#### Data Analyses

Seagrass core samples were used to compare mean biomass between Z. marina and R. maritima habitats and mean biomass between the 2 months sampled. Separate mixed models were used to test for differences in fine sediment content, sediment organic matter content, invertebrate richness, and invertebrate abundance between Z. marina and R. maritima habitats and between the 2 months sampled. The habitat and month during which a sample was taken were used as fixed factors, and site where a sample was taken was used as a random factor in the models. Richness and abundance of epifauna were normalized to biomass from the grab samples from which they came: both to seagrass biomass and total biomass including macroalgae. A oneway ANOVA and Tukey's HSD test were used to determine the differences between daily means of temperature at the three sites. All data were inspected for normality and homogeneity of variance and transformed if assumptions were not met.

<sup>1</sup>http://www.vecos.org

Pearson's correlation was used to test relationships between seagrass biomass, total biomass including macroalgae, or shoot density within grab samples and the richness or abundance of invertebrates (6 comparisons). All analyses were considered significant at the p < 0.05 level and were performed in the R programming language (R Core Team, 2015).

#### RESULTS

#### Seagrass Biomass and Site Conditions

Water temperatures were higher in August than June, and mean temperatures were higher overall at the embayment site over the summer (p < 0.001 and p < 0.001, **Supplementary Figure S2**). Across sites, Z. marina had greater biomass than R. maritima during the months sampled (**Supplementary Table S1**), although at two sites in August, R. maritima biomass was greater. The greatest dieback of both R. maritima and Z. marina occurred at the embayment site, where seagrass biomass fell 90% between the beginning and end of the growing season.

#### Sediment Characterization

Zostera marina areas contained more fine sediment (silt and clay) than R. maritima areas (p = 0.027, **Figure 1**): Z. marina had an average of 86.8% sand and 13.2% fine sediment, compared to R. maritima with 89.4% sand and 10.6% fine sediment. The embayment site lost fine sediment from June to August, contrary to the other two sites, where the percentage of fine sediment increased. Mean organic content was greater, although not significantly so, in Z. marina sediments than R. maritima sediments in June, and means were similar in August (**Figure 1**). Mean organic content of sediments was greater in August than in June.

error at each habitat during the sampling periods. Asterisk indicates mean values are significantly different.

### Invertebrate Abundance and Diversity

Thirteen species of small invertebrates (<3 cm) were found. There was no effect of seagrass species on abundance of invertebrates. There was a significant effect of month (p = 0.001); between June and August, mean abundance decreased from 165.6 to 28 individuals per sample. There was no effect of seagrass species on richness of invertebrates, though there was a significant effect of month (p = 0.001); a mean richness of 5.5 species per sample in June fell to 3 in August (**Supplementary Table S2**).

Greater seagrass biomass was strongly associated with both greater richness and abundance of invertebrates (**Table 1**). Several of the grabs included the macroalgal species Gracilaria vermiculophylla, where present, it was added to seagrass biomass for total biomass present, which resulted in stronger relationships. Higher invertebrate abundances were associated with higher shoot densities (**Table 1**), but the relationship was not significant between shoot density and invertebrate richness.

# DISCUSSION

This study demonstrated that mean biomass of Z. marina across sites was greater than the biomass of R. maritima, as we hypothesized, during both June and August of the summer we sampled when Z. marina typically dies back. Biomass decline of Z. marina is typical between June and August, however, the decline demonstrated in R. maritima between June and August is somewhat uncharacteristic (Wetzel and Penhale, 1983; Orth and Moore, 1988; Kantrud, 1991; Moore et al., 2014). It is possible that R. maritima only declined in the areas sampled in this study, intermediate depths where the seagrasses cooccur, and not the shallow depths at which R. maritima is typically more abundant (Kantrud, 1991). When examining site data individually, the embayment site stands out: it lost the most biomass of the three sites between June and August, and had the highest temperatures over the summer, making it likely that high temperatures caused seagrass dieback.

Greater fine sediment content in Z. marina habitats compared to R. maritima suggests that Z. marina may possess a more enhanced capability to trap sediment than R. maritima. Because greater wave attenuation occurs when shoots occupy more of the water column (Fonseca and Callahan, 1992), the taller Z. marina canopy may trap more fine material than R. maritima.

TABLE 1 | Pearson's correlation matrix of relationships between invertebrate abundance and richness with biomass and shoot density.


Bold numbers indicate significant relationships.

Measuring sediment trapping by weighing the accumulation of fine vs. coarse sediments has limitations, thus the use of sediment traps would add valuable information to subsequent studies. However, fine-grained particles are more easily suspended than coarser particles (Ward, 1985); therefore, the increase of fine sediment from June to August in Z. marina beds is likely indicative of either settlement or retainment influenced by plant structure. The ubiquity of organic matter, including seagrass blade senescence and allochthonous sources, in the late summer could have contributed to the similar organic enrichment of sediments seen in August (Oreska et al., 2017).

Lack of differences in richness between seagrass species for small invertebrates (isopods, amphipods, decapods, and gastropods) could be attributable to their low richness in the lower Chesapeake Bay in general, and lack of differences in abundances could be attributable to the relative proximity of the habitat types in this study, or the mobility of invertebrates (Parker et al., 2001; Valentine and Duffy, 2006). Invertebrate richness and abundance were, however, greater in samples with higher shoot density, seagrass biomass, and total biomass including macroalgae. This intuitive relationship has been demonstrated in other studies (Orth et al., 1984 and sources therein; Douglass et al., 2010), and the present study supported this concept. Higher shoot density did not always correspond to higher biomass; most samples showed R. maritima had higher shoot density than Z. marina, while having less biomass. This suggests dense stands of R. maritima could harbor abundant epifauna. Abundance and richness of invertebrates decreased from June to August, this is likely due to the decrease in canopy biomass across all sites and both seagrasses.

This study highlights the need for further research in seagrass habitats transitioning from larger-form, stable species to smaller-form, opportunistic species. Recent research on seagrass populations in Chesapeake Bay show that Z. marina abundance is on a downward trajectory aligning in part with a warming climate (Lefcheck et al., 2017; Richardson et al., 2018), and the potential for long-term persistence of R. maritima populations warrants further research before concluding its canopies may provide ecosystem benefits comparable to that of Z. marina.

# AUTHOR CONTRIBUTIONS

This research was completed when EF was a graduate student at the Virginia Institute of Marine Science, School of Marine

### REFERENCES


Science, College of William and Mary and KM was her graduate advisor and mentor.

# FUNDING

This study received financial support from the Chesapeake Bay National Estuarine Research Reserve and from the Virginia Institute of Marine Science, College of William and Mary (contribution number 3790).

# ACKNOWLEDGMENTS

This manuscript was greatly improved through conversations with Andrew Johnson, Ashley Smyth, and Jonathan Lefcheck. We thank Erin Shields and Tavis Sparrer for assistance during field work and sample processing. We also thank the two reviewers whose comments significantly improved the manuscript.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00461/full#supplementary-material

FIGURE S1 | The three locations where the study took place in the lower Chesapeake Bay, VA, United States. Mobjack is the "embayment site" referred to in the methods section, and the two fringing marsh wave-exposed sites are Goodwin and Poquoson.

FIGURE S2 | Daily mean water temperatures at the three sites during the sampling events and throughout the summer. Mobjack is the "embayment site" referred to in the methods section, and the two fringing marsh wave-exposed sites are Goodwin and Poquoson.

TABLE S1 | Mean biomass of seagrass canopies ± standard error during the growing season. Mobjack is the "embayment site" referred to in the methods section, and the two fringing marsh wave-exposed sites are Goodwin and Poquoson.

TABLE S2 | Small epifaunal invertebrates (isopods, amphipods, gastropods and decapods) found in grab samples. Values are means (n = 3) ± standard error. G, M, and P correspond to the site names Goodwin, Mobjack and Poquoson. Mobjack is the "embayment site" referred to in the methods section, and the two fringing marsh wave-exposed sites are Goodwin and Poquoson.

eutrophication and implications for recovery. J. Exp. Mar. Biol. Ecol. 302, 233–248. doi: 10.1016/j.jembe.2003.10.014


and abiotic control of community structure. Limnol. Oceanogr. 55, 1499–1520. doi: 10.4319/lo.2010.55.4.1499


**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 French and Moore. 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.

# Predicting Current-Induced Drag in Emergent and Submerged Aquatic Vegetation Canopies

Arnold van Rooijen1,2,3,4 \*, Ryan Lowe1,2,3,5, Marco Ghisalberti 2,5,6, Mario Conde-Frias 2,3,5 and Liming Tan<sup>7</sup>

<sup>1</sup> School of Earth Sciences, University of Western Australia, Perth, WA, Australia, <sup>2</sup> UWA Oceans Institute, University of Western Australia, Perth, WA, Australia, <sup>3</sup> ARC Centre of Excellence for Coral Reef Studies, University of Western Australia, Perth, WA, Australia, <sup>4</sup> Unit of Marine and Coastal Systems, Deltares, Delft, Netherlands, <sup>5</sup> Oceans Graduate School, University of Western Australia, Perth, WA, Australia, <sup>6</sup> Department of Infrastructure Engineering, University of Melbourne, Melbourne, VIC, Australia, <sup>7</sup> Institute of Port, Coastal and Nearshore Engineering, Ocean College, Zhejiang University, Zhoushan, China

#### Edited by:

Roshanka Ranasinghe, IHE Delft Institute for Water Education, Netherlands

#### Reviewed by:

Tomohiro Suzuki, Flanders Hydraulics Research, Belgium Vanesa Magar, Centro de Investigación Científica y de Educación Superior de Ensenada (CICESE), Mexico

\*Correspondence:

Arnold van Rooijen arnold.vanrooijen@ research.uwa.edu.au

#### Specialty section:

This article was submitted to Coastal Ocean Processes, a section of the journal Frontiers in Marine Science

Received: 27 August 2018 Accepted: 08 November 2018 Published: 04 December 2018

#### Citation:

van Rooijen A, Lowe R, Ghisalberti M, Conde-Frias M and Tan L (2018) Predicting Current-Induced Drag in Emergent and Submerged Aquatic Vegetation Canopies. Front. Mar. Sci. 5:449. doi: 10.3389/fmars.2018.00449 Canopies formed by aquatic vegetation, such as mangroves, seagrass, and kelp, play a crucial role in altering the local hydrodynamics in rivers, estuaries, and coastal regions, and thereby influence a range of morphodynamic and biophysical processes. Prediction of the influence of canopies on these hydrodynamic processes requires a fundamental understanding of canopy drag, which varies significantly with both flow conditions and canopy properties (such as density and submergence). Although our knowledge on canopy drag has increased significantly in recent decades, a conclusive, physics-based description for canopy drag that can be applied to both emergent and submerged canopies is currently lacking. Here, we extend a new theoretical canopy drag model (that employs the velocity between canopy elements as the reference velocity) to submerged aquatic canopies. The model is validated for the first time with direct measurements of drag forces exerted by canopies across broad ranges of flow conditions and canopy density and submergence. The skill and broader applicability of the model are further assessed using a comprehensive set of existing experimental data, covering a broad range of natural conditions (including flexible vegetation). The resulting model provides a simple tool to estimate canopy drag forces, which govern hydraulic resistance, sediment transport, and biophysical processes within aquatic ecosystems.

Keywords: ecohydraulics, vegetated flows, flow-plant interaction, drag model, drag coefficient

# INTRODUCTION

It is widely recognized that aquatic vegetation, such as seagrass, reeds, kelp, and mangroves, greatly influences hydrodynamic processes within rivers, estuaries, and coastal regions (e.g., Nepf, 2012). The drag exerted by emergent and submerged vegetation impacts the local hydrodynamics, morphodynamics, and ecology over a range of spatial scales (Koch et al., 2007). The canopies formed by vegetation can affect the local flow environment at the smallest scale (i.e., the plant scale, mm to cm) to the larger-scale (>1 km) flows that occur across benthic ecosystems. Canopy drag forces contribute to reducing flow velocities within canopies (López and García, 2001) and enhancing local turbulence (Nepf and Vivoni, 2000). In areas with significant wave action, such as in coastal regions and large lakes, the rate of work done by canopy drag forces also results in van Rooijen et al. Predicting Current-Induced Canopy Drag

wave energy attenuation (e.g., Fonseca and Cahalan, 1992). The flow reduction induced by canopy drag can, in turn, influence a number of morphodynamic and biophysical processes (Koch et al., 2007). For example, canopies can modify local bed shear stresses (James et al., 2004), thereby affecting sediment transport, deposition (Hendriks et al., 2008, 2010) and resuspension (Widdows et al., 2008). Similarly, canopy drag also indirectly influences other particle dynamics, affecting pollination (Ackerman, 1995), establishment of seedlings (Balke et al., 2013), and recruitment and settlement of larvae, spores, and fauna (Kenyon et al., 1999). The effect of the reduced incanopy flow on the diffusive boundary layer around plant leaves (Koch et al., 2007) also governs nutrient uptake (Morris et al., 2008) and can influence the growth of epiphytes (Cornelisen and Thomas, 2002). Under strong flow conditions, the drag forces exerted on canopy elements can result in their physical removal from the seabed (Duarte, 2002; Edmaier et al., 2011). Globally, aquatic ecosystems are under increasing pressure from anthropogenic and climate change impacts (Duarte, 2002), and it is crucial we increase our understanding of canopy drag as it directly influences many important biophysical processes in aquatic environments.

To be able to quantify the influence of aquatic canopies on the local hydrodynamics, a comprehensive understanding of the mechanics governing canopy drag is required. Given the diversity of plant morphologies in natural environments, individual plants are often schematized as uniform, rigid cylinders to establish a general knowledge framework for the processes governing drag (see review by Vargas-Luna et al., 2016). The drag force per unit length of a cylinder in isolation is given by:

$$f\_d = \frac{1}{2} \rho d\_c \mathbf{C}\_d \mathbf{U}\_{ref}^2 \tag{1}$$

where ρ is the water density, d<sup>c</sup> is the cylinder diameter, C<sup>d</sup> is the drag coefficient, and Uref is a reference flow velocity (which, in the case of an isolated cylinder, is equal to the upstream velocity). For reference, a **Notation table** specifying all variables is provided at the end of this manuscript. Predicting the drag coefficient for a cylinder in isolation is historically well-established, and it can be robustly predicted as:

$$C\_d = 1 + 10Re^{-2/3} \tag{2}$$

(White, 1991), where the Reynolds number is defined as Re = Uref dc/ν, with ν is the kinematic viscosity. For real-world application, considering plants rather than cylinders, temporal fluctuations in the drag force (due to turbulence) and vertical variation of the drag are often of less interest than the mean drag force, which governs the range of biophysical processes described earlier. As the plant biomass and flow velocity may vary significantly over the height of the plant, the total mean drag force on the plant is usually defined as:

$$\overline{F\_d} = \frac{1}{2} \rho \overline{\int\_{z=0}^{h\_r} d\_\nu \mathbf{C}\_d \mathbf{U}\_{ref}{}^2 dz} \tag{3}$$

where the drag force is integrated over the vertical dimension (z) and averaged over time (denoted by the overbar), h<sup>v</sup> is the vegetation (cylinder) height (with z = 0 at the bed), and d<sup>v</sup> is the vegetation stem (cylinder) diameter.

In the case of a single plant, the upstream velocity is usually weakly vertically-varying over most of the water column and the depth-averaged velocity is an obvious choice for the reference velocity (Uref ) needed to estimate the drag force in Equation (3). However, in the case of multiple plants forming a canopy, the flow throughout the canopy (and therefore the "upstream" velocity for each plant) is spatially non-uniform. It is thus unclear which actual velocity governs drag and could be used as the appropriate reference velocity. In emergent canopies (denoted hereafter with the superscript "em"), previous studies have chosen the reference velocity to be either: (1) the bulk velocity (i.e., U em <sup>b</sup> <sup>=</sup> <sup>Q</sup>/Wh, where <sup>Q</sup> is the flow discharge, <sup>W</sup> is the channel width and h is water depth) (e.g., Wu et al., 1999) or, more commonly, (2) the pore velocity (U em <sup>p</sup> <sup>=</sup> <sup>U</sup> em b / 1 − λ<sup>p</sup> , where λ<sup>p</sup> is the canopy density that is equivalent to the canopy element plan area per unit bed area) (e.g., Tanino and Nepf, 2008), representing the spatially-averaged velocity inside the fluid spaces within a canopy. However, through Large Eddy Simulation, Etminan et al. (2017) found that the "constricted cross-section velocity," the average velocity in the constriction between adjacent canopy elements, is the velocity scale that actually governs wake pressure and thus canopy drag. The relationship between the pore velocity and the constricted crosssection velocity (U em c ) is dependent on the arrangement of canopy elements, and is obtained through conservation of mass [i.e., U em c 1 − dv Sv,<sup>l</sup> = U em p 1 − λ<sup>p</sup> ]. Here, Sv,<sup>l</sup> is the lateral spacing between adjacent elements at the same streamwise (x) location, and can only be strictly defined for regular arrays (such as linear or staggered arrangements). This relationship between the constricted cross-section velocity and the pore velocity can be written as a function of the canopy density:

$$U\_c^{em} = \frac{1 - \lambda\_p}{1 - \sqrt{4\frac{\lambda\_p}{\beta\pi}}} U\_p^{em} \tag{4}$$

(Stone and Shen, 2002; Etminan et al., 2017). In Equation (4), β represents the ratio between Sv,<sup>l</sup> and the distance between two rows of canopy elements in the streamwise direction (Sv,s). For random arrays, as can be found in nature, the constricted cross-section velocity can be computed from the bulk velocity:

$$U\_c^{em} = \frac{1}{1 - d\_\nu \sqrt{N\_\nu}} U\_b^{em} = \frac{1 - \frac{\pi}{4} N\_\nu d\_\nu^2}{1 - d\_\nu \sqrt{N\_\nu}} U\_p^{em} \tag{5}$$

where N<sup>v</sup> is the total number of plants per unit area. Note that this will result in a canopy-average value of U em c , and local values may vary significantly.

In the case of submerged canopies, the shear layer present at the top of the canopy results in strong vertical variations in the spatially-averaged flow, further complicating canopy drag predictions. In many cases, the reference velocity used to predict the drag in submerged canopies is based on the bulk velocity (U sub <sup>b</sup> <sup>=</sup> <sup>Q</sup>/Wh, where the superscript "sub" refers to a velocity scale used for submerged canopies) (Wu et al., 1999; López and García, 2001). However, this approach does not account for the attenuation of flow within the canopy that will significantly influence canopy drag. An exception is the study of Liu and Zeng (2017) who proposed a more representative in-canopy flow velocity that accounts for vertical variation in the spatiallyaveraged flow. However, their approach did not account for the local (horizontal) spatial variation in the mean flow inside the canopy.

In emergent canopies, experimental measurements of drag coefficients have most commonly been obtained by measuring the surface slope and assuming a force balance of canopy drag and hydraulic gradient (Liu et al., 2008; Tanino and Nepf, 2008). The drag force of an individual plant within the canopy is then given by:

$$\overline{F\_d} = -\left(1 - \lambda\_\rho\right) \rho \lg \frac{d\eta}{d\kappa} h\_\nu N\_\nu^{-1} \tag{6}$$

where g is the gravitational acceleration, and η is the (measured) water surface elevation. By combining Equations (3) and (6), the drag coefficient can be obtained when assuming a depthuniform velocity profile. For emergent canopies, this is relatively straight-forward, although the choice of reference velocity may greatly affect the calculated C<sup>d</sup> values (Etminan et al., 2017). A large range of empirical relations have been established to relate canopy drag coefficients to plant shape, the flow regime (i.e., Reynolds number) and canopy properties (e.g., density). The drag coefficient is generally found to decrease exponentially with increasing Reynolds number (e.g., Liu and Zeng, 2017), following a similar trend to the isolated cylinder case (Equation 2). In terms of canopy geometry, some studies have found that the drag coefficient decreases with increasing canopy density (e.g., Nepf, 1999), while many others obtained conflicting results (e.g., Wu et al., 1999; Tanino and Nepf, 2008; Wang et al., 2014). Relatively few studies have directly measured the forces on canopy elements using force sensors either mounted at the top (e.g., Kothyari et al., 2009) or at the base of a canopy element (e.g., Schoneboom et al., 2010).

Furthermore, the majority of studies have focused on emergent canopies, such that there are still significant knowledge gaps in predicting the drag of submerged canopies. This is largely due to the more complex vertical flow structure within submerged canopies. The in-canopy flow velocity is often significantly lower than the freestream velocity and, as for emergent canopies, horizontal variation in the flow field are expected to play a significant role in canopy drag. Even with accurate measurements of submerged canopy drag forces, it is still unclear how to predict the constricted cross-section velocity within a submerged canopy when velocity measurements are lacking. The main reason for this is that the in-canopy flow velocity is dependent on the drag itself (Lowe et al., 2005), so that C<sup>d</sup> is a function of Uref , and vice versa.

This paper aims to reduce the uncertainty in canopy drag estimation through direct measurements of the drag force in aquatic vegetation canopies subject to unidirectional flow. The experimental program includes both emergent and submerged canopies with varying densities, and a range of hydrodynamic conditions covering a broad range of natural conditions that can be found in aquatic systems. In addition, a theoretical canopy drag model for emergent canopies is extended to submerged canopies and validated for the first time using direct force measurements, and then more broadly assessed using a compilation of data reported in previous studies.

#### CANOPY DRAG MODEL

For both emergent and submerged canopies, the mean drag force exerted on a single plant or canopy element is governed by Equation (3). For emergent canopies, the mean horizontal flow velocity is often assumed to be depth-uniform. For submerged canopies, the horizontal flow profile can be approximated as a two-layer flow with depth-uniform velocities both above and inside the canopy (e.g., Lowe et al., 2005; Liu et al., 2008) (see **Figure 1** for a definition sketch and relevant velocity definitions).

#### Emergent Canopies

For emergent canopies, Etminan et al. (2017) proposed to use the theory of drag for isolated cylinders (i.e., Equation 2) as the basis to compute the drag coefficients associated with emergent canopy. Their model employs the constricted cross-section velocity (U em c ) as the reference velocity (Uref) to determine the drag coefficient through the Reynolds number (Equation 2) and to compute the drag force (Equation 3), and was validated through Large Eddy Simulation (Etminan et al., 2017).

#### Submerged Canopies

For a given in-canopy flow, one can hypothesize that an analogous method to emergent canopies can be applied to submerged canopies, i.e., the in-canopy constricted cross-section velocity (U sub c ) can be computed using Equation (4) or (5). However, as discussed in section Introduction, the estimation of U sub p is not straight-forward due to the vertical variation in the mean velocity profile (**Figure 1**); the magnitude of the in-canopy velocity both governs, and depends on, the canopy drag. Here, we propose to use a canopy flow model to predict the in-canopy pore velocity (U sub p ) based on the (undisturbed) above-canopy flow velocity (U∞). This model takes the form:

$$U\_p^{sub} = U\_\infty \sqrt{L\_d/L\_s} \tag{7}$$

depth (h), and the canopy properties: height (hv), stem diameter (dv), and canopy density (λp, Nv). For submerged canopies, an initial value of C<sup>d</sup> = 1 (to (Lowe et al., 2005). In Equation (7), L<sup>d</sup> is the drag length scale, given by

$$L\_d = \frac{2h\_\nu \left(1 - \lambda\_\mathcal{P}\right)}{C\_d \lambda\_f},\tag{8}$$

(Lowe et al., 2005; Ghisalberti, 2009), and represents the flow resistance of the canopy. λ<sup>f</sup> is the canopy element frontal area per unit bed area (=hvdvNv). L<sup>s</sup> is the shear length scale, given by

$$L\_s = \frac{2h\_v}{C\_f} \tag{9}$$

(Lowe et al., 2005) (where C<sup>f</sup> is a friction coefficient), which parameterizes the magnitude of the shear stress at the top of the canopy. This shear stress is generated by the velocity difference between the flow within and above the canopy. If velocity measurements are available, the friction coefficient can be estimated based on the peak in the Reynolds stress profile near the top of the canopy (z ≈ hv):

$$C\_f = 2\frac{\mu\_\*^2}{U\_\infty^2} = 2\frac{-\overline{\mu'\mu'}\_{z\simeq h\_\nu}}{U\_\infty^2} \tag{10}$$

(Lowe et al., 2005), where u<sup>∗</sup> is the friction velocity and u ′ and w ′ are the horizontal and vertical turbulent velocity fluctuations, respectively. Data from a wide range of canopies indicates that u∗ U∞ tends to be consistently O(0.1), which corresponds to C<sup>f</sup> = O(0.01) (e.g., Harman and Finnigan, 2007; Lowe et al., 2008; Luhar et al., 2010; Moltchanov et al., 2011; Weitzman et al., 2015). Therefore, for a given canopy geometry and above-canopy flow velocity (U∞), the in-canopy pore velocity U sub p can be estimated from Equations (7–10). Subsequently, the constricted cross-section velocity inside a submerged canopy can be obtained through Equations (4) or (5), and is used as the reference velocity (Uref ) to calculate the drag coefficient through the Reynolds number (Equation 2) and to compute the drag force (Equation 3).

In summary, the model that is proposed here relies on information on above-canopy flow velocity (U∞) or bulk velocity (for emergent canopies), the local water depth (h), and the canopy properties: height (hv), stem diameter (dv), and canopy density (λp, Nv). It includes one empirical parameter (namely, C<sup>f</sup> ) in the case of a submerged canopy. It is important to emphasize that given the drag coefficient C<sup>d</sup> is also needed in Equation (8) to predict the in-canopy flow (hence the drag forces and in-canopy flow are inherently coupled), for submerged canopies the model involves an iterative process. A flow diagram summarizing the model is provided in **Figure 2**. In the following sections, the model is validated using newly obtained velocity and drag force data, as well as a large dataset covering a broad range in canopy geometries and flow conditions obtained from literature.

#### EXPERIMENTAL METHODS

Experiments were carried out in a 20-m-long, 0.6-m-wide, and 0.6-m-deep recirculating flume using emergent (**Table 1**)

calculate Ld) is suggested.

TABLE 1 | Experimental emergent vegetation conditions: canopy density (λp), canopy height (hv), water depth (h), flow rate (Q), bulk velocity (U em b ), pore velocity (U em p ), constricted cross-section velocity (U em c ), measured velocity averaged over the dowel height (U em <sup>m</sup> ) and the measured time-averaged drag force acting on a single cylinder (Fd).


TABLE 2 | Experimental submerged vegetation conditions: canopy density (λp), canopy height (hv), water depth (h), flow rate (Q), bulk velocity (U sub b ), pore velocity (U sub p ), constricted cross-section velocity (U sub c ), measured in-canopy velocity averaged over the canopy/dowel height (U sub <sup>m</sup> ) and the measured time-averaged drag force acting on a single cylinder (Fd).


and submerged (**Table 2**) model vegetation. To accommodate the drag force sensor, a 10-cm-high false bottom was placed over a length of 10 m. Model canopies were constructed using perforated PVC sheets and two sets of 6.4-mm-diameter dowels with heights of 30 cm (emergent) or 9 cm (submerged). Dowels were distributed in a staggered arrangement over the entire width of the flume. The dowel diameter used in this study has been used previously in numerous studies to represent a generic aquatic vegetation canopy (e.g., Nepf, 1999) and was originally based on actual observed stem diameters of cordgrass (Spartina alterniflora, see Zavistoski, 1994). An important design parameter for experimental studies with canopies is the canopy

length (Lv). Lowe et al. (2005) found a canopy flow adjustment length (x0) of 3–5 times the drag length scale (Ld) in their experiments. Hence, to ensure fully-developed canopy flow, it was required that L<sup>v</sup> >> x<sup>0</sup> resulting in L<sup>v</sup> ranging between 2.4 m (λ<sup>p</sup> = 0.1) and 3.6 m (λ<sup>p</sup> = 0.025).

The drag force exerted on a representative aluminum dowel (canopy element) was measured using a load cell with 2 N capacity (Uxcell, Hong Kong) connected to a load cell amplifier (RW-ST01A, SMOWO, China). The load cell was mounted vertically onto the underside of the false bottom in the flume, ensuring the bottom end of the load cell was fixed but allowing the upper end to move slightly with the bending moment (MY) generated by the drag force acting on the dowel (**Figure 3**). Data was obtained from the load cell using a National Instruments data acquisition system (NI-DAQ PCI-6009) and LabVIEW software. This experimental setup relies on the linear relationship between drag force and the instrument voltage output. To confirm the load cell's linearity, the load cell was placed at the edge of a table and the voltage output recorded for cases with both no weight and a weight of (approximately) 1.9 N. The (linear) calibration coefficient was derived by calculating the ratio between the change in voltage output and the change in applied weight. The linear response was subsequently verified using 9 (smaller) weights ranging from 0.01 to 1.2 N (R <sup>2</sup> > 0.99). Prior to each individual experimental run, the load cell was re-calibrated using a set of three known weights ranging between 0 and ∼0.3 N.

For emergent canopies, the water level was measured using a point gauge at three locations both upstream and downstream of the canopy. The water level gradient (dη/dx) was then obtained by averaging the water level in time at the upstream and downstream locations and dividing by the canopy length. To calculate the flow rate, velocity measurements were obtained several meters upstream of the canopy using a Nortek Vectrino II Acoustic Doppler Velocity (ADV) profiler, resulting in 3-cmtall velocity profiles with 1 mm resolution. The vertical position of the ADV was varied to obtain a full velocity profile extending from the bottom to ∼5 cm below the water surface. In a similar manner, the velocity profile in and above the canopy was obtained for the submerged cases, and extended from the base of the canopy up to ∼5 cm below the water surface. The ADV was positioned within the constricted cross-section in between two canopy elements at a lateral distance of ∼0.25Sv,<sup>l</sup> from one of the elements. This was based on the modeling of Etminan et al. (2017), who found that the velocity at this point in a staggered canopy was similar to the constricted cross-section value. Experimental runs were repeated several times to obtain the full velocity profile over depth within and above the canopy. Both the load cell and ADV were placed at a distance of ∼2/3 of the canopy length downstream from the leading edge, which is at least 10 times the drag length scale (Ld) for all cases.

The experimental program included a range of canopy densities (λ<sup>p</sup> = 0.025, 0.05, and 0.10), canopy submergence ratios (h/h<sup>v</sup> = 1, 2, and 3, where h is the water depth at still water and h<sup>v</sup> is the canopy height) and flow rates (**Tables 1**, **2**). The upstream flow velocity ranged between ∼0.05 and 0.35 m/s, which in combination with the range in canopy density and

FIGURE 3 | Schematic view of the load cell, which was attached to a single aluminum dowel and placed under the false bed. The drag force (Fx) due to the flow acting on the dowel translates into a moment (My) around the base of the load cell.

submergence ratio covers a broad range of conditions that can be found in aquatic canopies. The drag force and velocity data were processed and the drag coefficient was subsequently computed using Equation (3).

Following Taylor (1997), measurement uncertainties were propagated, with an estimated velocity uncertainty of 0.1 cm/s and drag force uncertainty of 0.4 mN. For the model-data comparison the model skill was quantified using scatter index (SCI), and the relative bias. The scatter index is a relative measure of the scatter between computed (xc) and measured data (xm) and is computed by normalizing the root-mean-square error ( q (x<sup>c</sup> − xm) 2 ) with the maximum of the root-mean-squarevalue of the data (<sup>q</sup> (xm) 2 ) and the absolute value of the mean of the data (|xm|). The relative bias is a relative measure of the bias or mean error ((x<sup>c</sup> − xm)) and is normalized in the same way as the scatter index.

To date only relatively few studies have used load cells to measure canopy drag, hence a comparison is made between the drag force measured directly using the current methodology (Fd,direct) and the drag force obtained through an indirect measuring method commonly used in previous studies [Fd,momentum, from Equation (6)]. The indirect estimate (Fd,momentum) for the emergent cases (E05 and E10, see **Table 1**) shows the same trend (R <sup>2</sup> = 0.99) as the drag force directly measured with the load cell (Fd,direct, see **Figure 4**). Although measured drag forces with magnitudes above 0.01 N are very similar for both methods (up to 8% difference), for drag forces <0.01 N the discrepancy between both methods increases (with an average 22% difference). For these low flow cases the percentage uncertainty associated with the measured water level gradient increases (with the water level dropping only ∼3 mm over the length of the canopy) leading to larger errors. Given

the high instrument linearity, the force sensor is able to provide more accurate measurements for these cases and is therefore preferred.

#### RESULTS AND DISCUSSION

#### Measurement of Drag Coefficients Isolated Cylinder

Although the focus in this study is on assessing canopy drag, a limited number of experiments were conducted with isolated emergent (**Table 1**) and submerged (**Table 2**) cylinders. The isolated cylinder drag coefficients were then compared to theory (Equation 2) to gain confidence in the experimental methodology (particularly the drag force data obtained from the load cell). For the emergent case, there is excellent agreement between the directly-measured drag on an isolated cylinder and Equation (2) (**Figure 5**, squares). For the submerged case (with same height as the submerged canopy), the value of C<sup>d</sup> derived from the measured drag force and measured in-canopy velocity (averaged over the cylinder height) is consistent with isolated cylinder theory (**Figure 5**, triangles). In other words, despite the single vertical cylinder occupying a fraction of the water column in a boundary layer flow, its forces can be predicted by Equation (2) originally developed for a cylinder in a uniform crossflow.

#### Emergent Canopies

As discussed in the Introduction, for emergent canopies, both the bulk velocity and pore velocity are often used as the reference

velocity in Equation (3) to relate a given flow condition to the canopy drag force through a drag coefficient (i.e., Cd,<sup>b</sup> and Cd,<sup>p</sup> , respectively). Here, when using both the bulk velocity (U em b ) and pore velocity (U em p ) are used as the drag reference velocity, there are large discrepancies with values for isolated cylinders (Equation 2), similar to results reported in other studies (e.g., Liu and Zeng, 2017). For the highest density canopies (λ<sup>p</sup> =0.1), there is an exponential decrease in the drag coefficient with Reynolds number using both the bulk velocity U em b (**Figure 6A**, squares) and pore velocity U em p (**Figure 6B**). For the 5% density emergent canopies, the drag coefficient shows a slight decrease with Re using both reference velocities. When considering both U em b and U em p , the drag coefficient appears to take an approximately constant value at high Re (i.e., Re >1,000), consistent with other studies (e.g., Tang et al., 2014). Given that the pore velocity accounts for the volume of water being occupied by the canopy, Cd,<sup>p</sup> is always smaller than Cd,<sup>b</sup> , but still deviates substantially from isolated cylinder values. To account for these discrepancies, previous studies have arrived at highly empirical Cd,<sup>p</sup> -Re<sup>p</sup> relationships that are parameterized as a function of canopy density (Tanino and Nepf, 2008) and (sometimes) stem diameter (Sonnenwald et al., 2018).

The direct experimental measurements support the canopy drag model proposed by Etminan et al. (2017)–when the constricted cross-section velocity (U em c ) is used as the reference velocity, calculated drag coefficients closely match the isolated cylinder values (**Figure 7**, squares). Therefore, while the drag coefficients derived using the bulk and pore velocities exhibit significant scatter (**Figure 6**), the use of U em c in the drag coefficient (Cd,<sup>c</sup> ) calculations serves to collapse the data onto the isolated cylinder curve. For the range of Reynolds numbers investigated

(Re<sup>c</sup> = 380–1,680), the drag coefficients for the emergent cases show relatively little scatter and approaches a canonical isolated cylinder value of Cd,<sup>c</sup> ≈ 1.

#### Submerged Canopies

The velocity exhibits more vertical variation in submerged canopies than in emergent canopies due the drag discontinuity and resulting shear layer present at the top of the canopy. When using the bulk velocity U sub b as the reference velocity to derive Re<sup>b</sup> and Cd,<sup>b</sup> , relatively low drag coefficients (that substantially deviate from the isolated cylinder values) are obtained (**Figure 6A**, circles and triangles). This approach neglects the effect of canopy drag on reducing the in-canopy velocity, which is significant at higher canopy densities. Hence, we use measured U sub c obtained approximately within the constricted cross-section area and derive the associated drag coefficients. For the submerged canopy cases, the measured values of Cd,<sup>c</sup> (i.e., evaluated using the constricted cross-section velocity U sub c ) generally follows the isolated cylinder theory curve (**Figure 7**). There is more scatter at Re<sup>c</sup> < 500, which can be attributed to the greater uncertainty associated with measuring flow and forces at such low Reynolds numbers. Therefore, analogous to the emergent canopy observations in **Figure 6**, where C<sup>d</sup> evaluated using bulk and pore velocities deviates markedly from isolated cylinder theory, these results indicate that the constricted cross-section velocity U sub c is the optimal reference velocity for evaluation of drag of a submerged canopy (**Figure 7**).

#### Canopy Drag Model Assessment Emergent Canopies

The canopy drag model for emergent canopies, based on Equations (2)–(4) using the computed constricted cross-section velocity U em c (see section Canopy Drag Model), was used to predict the drag force on a single canopy element in all experimental cases (**Table 1**). These predictions were then compared to the time-averaged drag force measured by the force sensor (**Figure 8**, squares). Using only the bulk flow velocity, which was derived from the known flow rate, and canopy geometry as model input, the canopy drag forces are accurately predicted over the full range of experimental cases. The results provide direct experimental validation of the finding of Etminan et al. (2017) that the constricted cross-section velocity U em c is the most appropriate reference velocity to parameterize canopy drag.

#### Submerged Canopies

To assess the ability of the model to predict the drag of submerged canopies, we first compared the predicted in-canopy velocities with the experimental measurements. Specifically, we compared the measured time-averaged constricted cross-section velocity integrated over the canopy height (U sub <sup>m</sup> ) with predicted U sub c values, which generally reveals good agreement (**Figure 9**). The model (with above-canopy velocity and canopy geometry as input) is subsequently applied to calculate the drag force for all submerged canopy cases (**Table 2**). Canopy friction coefficient values were derived for each case through Reynolds stress profiles (Equation 10), resulting in a range of C<sup>f</sup> values between 0.01 and 0.04. However, due to the experimental setup in this study, that used a downward facing ADV, the velocity measurement was limited to measuring only ∼6 cm below the water surface. The above-canopy velocity is therefore likely underestimated, particularly in the h/h<sup>v</sup> = 2 cases, and actual C<sup>f</sup> values are expected to be lower. Due to this uncertainty, here we opt to use a conventional value of 0.01 for all experimental cases (see section Canopy Drag Model). Compared to emergent canopies, there is greater scatter in the relationship between measured and predicted forces (particularly at low Re), but overall there is still relatively good agreement (**Figure 8**). Given the complexity

velocity (U

sub c

) in submerged canopies.

involved with submerged canopies (including the uncertainty involved with measurements under low Re), and the range in submergence ratio and density investigated, the model error averages about 11% (SCI = 0.114), and suggests that the whole model outlined in **Figure 2** can serve as a useful tool to obtain robust estimates of canopy drag forces.

## Model Application to Other Submerged Canopy Data Sets

To date, the canopy drag model has been validated for emergent canopies (Etminan et al., 2017) (albeit using only numerical simulations); here we have provided direct experimental validation for both emergent and submerged canopies for the first time. Nevertheless, the experiments only covered a relatively small range of possible canopy geometry and flow conditions. To further assess the validity of the model across a large range of flow conditions and canopy properties (e.g., density, submergence ratio, flexibility), the model was tested against a large number of existing datasets. Experiments were limited to those with submerged canopies in which the energy slope was reported.

#### Rigid Vegetation

For rigid vegetation, experiments that employed either staggered (as in the present study), linear or random arrangements were selected here (**Table 3**). Although on the individual canopy element scale, the velocity distribution may vary significantly among these arrangements, we hypothesize that the model can still be used to estimate bulk canopy drag. Indeed, the

TABLE 3 | Overview of experimental studies on drag in submerged rigid canopies from which data was obtained.


All experiments either used staggered (stag.), linear (lin.) or random (rand.) canopy setups.

Tables 3, 4, respectively).


TABLE 4 | Overview of experimental studies on drag in submerged flexible canopies from which data was obtained.

work of Etminan et al. (2017) suggested that in the case of randomly-distributed canopy elements, the constricted crosssection velocity can still be considered as the velocity scale governing canopy drag, indicating that the model can be applied here without modification. Hence, using provided values of flow rate and canopy properties, the bulk drag was computed and compared with the measured drag (**Figure 10A**). Using a constant C<sup>f</sup> value of 0.01 (as before), the model shows a similar trend as the measurements (R <sup>2</sup> = 0.74) with reasonably low bias and scatter (rel. bias = 0.02, SCI = 0.45). It should be emphasized that the main uncertainty is likely to be attributed to the schematization of relatively complex three-dimensional canopy hydrodynamics into a relatively simple (two-layer) model.

#### Flexible Elements

Although most studies so far have represented vegetation canopies using rigid elements, aquatic vegetation in natural systems is often flexible (e.g., seagrasses, kelp), adapting its shape and thereby frontal area in response to the flow. Hence, there is now increased experimentation with flexible mimics in hydraulic experiments (e.g., Abdolahpour et al., 2017). The canopy drag model presented in this study does not explicitly account for flexibility, but it is hypothesized that it could still be used as a tool provided the deflected vegetation height (i.e., the height of the vegetation under stationary flow condition) rather than the actual length of the element is used. Hence, data was obtained from three studies that investigated drag in submerged flexible canopies and reported the deflected canopy height (**Table 4**). From these studies, both Dunn et al. (1996) and Järvelä (2003) observed swaying motions of their flexible plants/(cylindrical) elements, resulting in a time-varying deflected canopy height. Okamoto and Nezu (2010) reported both swaying and the more organized monami-type motions (Ackerman and Okubo, 1993) in their experiments. Here, we use the time-averaged deflected canopy height as input for the model. Furthermore, for the experiments by Okamoto and Nezu (2010) we use the width of the flexible strip as a proxy for d<sup>v</sup> given that it is equivalent to the frontal area.

For the flexible canopies, the model is able to predict the bulk drag relatively well (**Figure 10B**, R <sup>2</sup> = 0.69, rel. bias = 0.04, SCI = 0.53). This is surprising to some extent, as the complexity associated with flexible elements is only accounted for to some extent by the (deflected) plant height. Both the measurements by Järvelä (2003) and Okamoto and Nezu (2010) are consistently underpredicted, which may be related to the plant geometries that involved flat strips and real plants, respectively. Dunn et al. (1996), on the other hand, used flexible cylinders in their experiments which provide more similarity with rigid cylinders, and may therefore better be represented by model.

Overall, with limited information (above-canopy velocity derived from flow rate, canopy properties) the relatively simple canopy drag model is able to provide reasonably accurate estimates of the bulk canopy drag for both rigid and flexible vegetation canopies. Given the fact that the model performs well over such a broad range of hydrodynamic conditions (U<sup>b</sup> = 1.5– 85 cm/s, Re = 65–5,430) and canopies (h/h<sup>v</sup> = 1.1–5, λ<sup>p</sup> = 0.08– 11.9%, both rigid and flexible vegetation), and is based on theory rather than empirical relations, it is thus expected the model can robustly predict hydraulic resistance of aquatic canopies, including in field setting with natural vegetation (e.g., where stem diameters are often of order 0.1–1 cm and current velocities of order 0.05–0.5 m/s, which translates to Re ranging between 50 and 5,000).

#### SUMMARY AND CONCLUSIONS

In this study we present new direct experimental measurements of canopy drag forces using emergent and submerged canopies with a broad range of flow conditions and canopy properties (i.e., density and submergence ratio). Drag coefficients were derived using direct measurements of the drag force on a dowel within the canopy. We found that if the constricted cross-section velocity is used as the reference velocity, the drag coefficient of both emergent and submerged canopies is equal to that of an isolated cylinder. Comparison between canopy drag model predictions and current and existing experimental data shows that the model is able to robustly and accurately predict canopy drag across the field range of flow conditions and canopy characteristics, including flexible canopies. The model can thus be used to predict drag forces in emergent and submerged canopies and is considered a simple and practical tool for estimating the hydraulic resistance of aquatic canopies.

### DATA AVAILABILITY STATEMENT

The datasets analyzed for this study can be obtained by sending a written request to the corresponding author at arnold.vanrooijen@research.uwa.edu.au.

### AUTHOR CONTRIBUTIONS

AvR designed the experiments, acquired the data, performed the data analysis, and wrote the manuscript. RL and MG provided support in the experimental design, data interpretation and the development of the manuscript. MC-F and LT provided support in data acquisition during the experiments. All authors revised the draft manuscript and approved the final version for submission.

#### FUNDING

This work was supported by an Australian Research Discovery Project Grant (DP170100802) to RL and MG.

#### REFERENCES


#### ACKNOWLEDGMENTS

This research was undertaken by AvR as part of a Ph.D. at The University of Western Australia who was funded by a University Postgraduate Award for International Students (UPAIS) and an Australian Government Research Training Program (RTP) Scholarship. LT was supported by the China Scholarship Council (grant number: 201706320274). The authors would like to thank Mr. Carlin Bowyer, Ms. Dianne King, and Mr. Brad Rose for assistance in the lab, Dr. Guido Wager for providing support with the force sensor, Dr. Vahid Etminan for valuable discussions, and the reviewers for their constructive feedback that helped to improve 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.

Copyright © 2018 van Rooijen, Lowe, Ghisalberti, Conde-Frias and Tan. 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.

# NOTATION

C<sup>d</sup> = drag coefficient for an isolated cylinder Cd,c = canopy drag coefficient based on constricted cross-section velocity Cd,b = canopy drag coefficient based on bulk velocity Cd,m = canopy drag coefficient based on measured velocity Cd,p = canopy drag coefficient based on pore velocity C<sup>f</sup> = canopy friction coefficient d<sup>c</sup> = cylinder diameter d<sup>v</sup> = plant / canopy element diameter f<sup>d</sup> = drag force per unit length of a cylinder F<sup>d</sup> = plant / canopy element total drag force g = gravitational acceleration h = water depth h<sup>v</sup> = plant / canopy height hvd = deflected plant / canopy height L<sup>v</sup> = canopy length L<sup>d</sup> = canopy drag length scale L<sup>s</sup> = canopy shear length scale N<sup>v</sup> = number of plants / canopy elements per unit bed area Q = flow rate / discharge Re = Reynolds number for an isolated cylinder Re<sup>c</sup> = Reynolds number (canopy) based on constricted crosssection velocity Re<sup>b</sup> = Reynolds number (canopy) based on bulk velocity Re<sup>m</sup> = Reynolds number (canopy) based on measured velocity Re<sup>p</sup> = Reynolds number (canopy) based on pore velocity Sv,l = lateral distance between two canopy elements at the same streamwise (x) location Sv,s = streamwise distance between two canopy element rows u ′ = turbulent velocity fluctuation in x direction u∗= friction velocity based on canopy shear stress U<sup>∞</sup> = free stream flow velocity U em b , U sub <sup>b</sup> <sup>=</sup> bulk velocity for emergent or submerged canopy resp. U em c , U sub <sup>c</sup> <sup>=</sup> constricted cross-section velocity for emergent or submerged canopy resp. U em <sup>m</sup> , U sub <sup>m</sup> <sup>=</sup> measured depth-averaged velocity for emergent or submerged canopy resp. U em p , U sub <sup>p</sup> <sup>=</sup> pore velocity for emergent or submerged canopy resp. Uref = reference velocity w ′ = turbulent velocity fluctuation in z direction W = channel width x = streamwise direction x<sup>0</sup> = canopy flow adjustment length z = vertical elevation measured from bottom β = ratio between lateral and streamwise canopy spacing (Sv,l/Sv,s) η = water surface elevation λ<sup>f</sup> = canopy (frontal) density / element frontal area per unit bed area λp= canopy (plan) density / element plan area per unit bed area ρ = water density ν = kinematic viscosity

# Canopy-Mediated Hydrodynamics Contributes to Greater Allelic Richness in Seeds Produced Higher in Meadows of the Coastal Eelgrass Zostera marina

#### Elizabeth Follett <sup>1</sup> \*, Cynthia G. Hays <sup>2</sup> and Heidi Nepf <sup>1</sup>

*<sup>1</sup> Department of Civil and Environmental Engineering, Massachusetts Institute of Technology, Cambridge, MA, United States, <sup>2</sup> Department of Biology, Keene State College, Keene, NH, United States*

#### Edited by:

*Matthew Philip Adams, The University of Queensland, Australia*

#### Reviewed by:

*Maike Paul, Technische Universitat Braunschweig, Germany Maryam Abdolahpour, University of Western Australia, Australia*

> \*Correspondence: *Elizabeth Follett emf@alum.mit.edu*

#### Specialty section:

*This article was submitted to Marine Ecosystem Ecology, a section of the journal Frontiers in Marine Science*

Received: *31 August 2018* Accepted: *11 January 2019* Published: *08 February 2019*

#### Citation:

*Follett E, Hays CG and Nepf H (2019) Canopy-Mediated Hydrodynamics Contributes to Greater Allelic Richness in Seeds Produced Higher in Meadows of the Coastal Eelgrass Zostera marina. Front. Mar. Sci. 6:8. doi: 10.3389/fmars.2019.00008* Seagrass meadows, which mediate ocean acidity and turbidity, sequester carbon, and increase biodiversity by providing shelter for larvae and small fish, are among the fastest disappearing ecosystems worldwide. Seagrasses are ecosystem engineers, creating distinct regions of enhanced and diminished flow and turbulent mixing, dependent upon canopy physical parameters, such as canopy density and blade morphology, which in turn impact the transport of pollen, sediment, and nutrients. The health and resilience of seagrass meadows increase with intraspecies genetic diversity, which depends on successful sexual reproduction and the transport of pollen particles between reproductive shoots, which in turn depends on the hydrodynamic conditions created by the meadow. This paper explored the transport of pollen grains in seagrass meadows using a random walk model. The model was parameterized with profiles of mean velocity and eddy diffusivity derived as functions of shoot density, canopy height, canopy shear velocity, canopy drag coefficient, and blade width, and validated with experimental measurements of a tracer plume evolving in a submerged model canopy. Model results showed that release at the top of the canopy led to significantly greater dispersal than release within the canopy, which was consistent with observed patterns of genetic diversity in *Zostera marina* seeds collected from coastal Massachusetts meadows. Specifically, seeds produced from upper inflorescences had greater allelic richness than seeds from lower inflorescences on the same reproductive shoot, and were the product of a greater number of fathers, reflecting the greater in-canopy pollen movement farther from the bed. Pollen grains modeled with a realistic elongated shape experienced significantly higher rates of capture by the canopy relative to spherical grains of the same volume. The effect of submergence depth (the ratio of water depth to canopy height) on pollen dispersal depended on the nature of the surface boundary: when pollen reflected off the water surface, the mean travel distance before pollen capture decreased with decreasing submergence depth. In contrast, when pollen accumulated at the water surface, surface transport increased pollen dispersal distances, especially at low submergence depths.

Keywords: seagrass, genetic diversity, pollen, particle transport, allelic richness

# 1. INTRODUCTION

Seagrasses are a foundation species in nearshore coastal ecosystems, forming highly productive meadows that provide structured habitat for many fish, invertebrate and bird species (Bruno and Bertness, 2001; Williams and Heck, 2001; Barbier et al., 2011). In addition to ecosystem services related to productivity and biodiversity (Unsworth et al., 2010), dense seagrass meadows also reduce bed shear stress, which increases sediment retention, and in some situations promote carbon sequestration (Duarte et al., 2013). Unfortunately, the area covered by seagrass meadows continues to decline, with an annual decrease of 7% since 1990 (Waycott et al., 2009; van Katwijk et al., 2016). A recent meta-analysis of meadow restoration found that only 37% of restoration efforts were successful, highlighting the importance of supporting existing meadows before collapse occurs (van Katwijk et al., 2016). Seagrasses are rhizomatous; like terrestrial grasses, they reproduce both clonally and sexually (i.e., from seeds). Seagrass pollination typically occurs without the use of an animal vector, with pollen release, transport, and capture all occurring within the water column (Ackerman, 1995; Kendrick et al., 2012). Sexual reproduction is critical for the establishment of new meadows and for genetic diversity, which improves meadow resilience (Hughes and Stachowicz, 2004; Hughes et al., 2008; Kendrick et al., 2012). Genetically diverse, outcrossed seeds are produced by pollen that travels beyond the area dominated by its parent clone and is captured by a mature flower of a separate genetic individual. Pollen that is captured on reproductive structures of its parent plant or vegetative blades may fail to produce viable seeds, while pollen that escapes the canopy and is transported downstream beyond the extent of the canopy will have a lower likelihood of successful pollination. Because hydrodynamic transport is a crucial link in this process, pollination depends on the canopy physical parameters that influence the velocity profile within and above the meadow, such as the canopy frontal area density, canopy height, and submergence depth (the ratio of water depth to canopy height).

The drag produced by seagrass blades attenuates flow within the canopy, creating a zone of low velocity close to the bed, with higher velocity above the canopy (Ackerman and Okubo, 1993; Ghisalberti and Nepf, 2002; Nepf, 2012). The magnitude of near-bed flow reduction depends on the canopy frontal area density (a) and the canopy height (h). Seagrass blades are flexible, and reconfigure (bend) in response to current, with meadow height decreasing with increasing current speed. The mean meadow height can be predicted from meadow morphology and current speed (see Equation 4 in Luhar and

a submerged canopy. The canopy height is *h*. The time-mean velocity at the bed and far above the canopy are denoted *u*1 and *u*2, respectively, with <sup>1</sup>*<sup>u</sup>* <sup>=</sup> *<sup>u</sup>*<sup>2</sup> <sup>−</sup> *<sup>u</sup>*1. The mixing layer extends between *<sup>z</sup>*<sup>1</sup> and *<sup>z</sup>*<sup>2</sup> (*tml* <sup>=</sup> *<sup>z</sup>*<sup>2</sup> <sup>−</sup> *<sup>z</sup>*<sup>1</sup> ), with endpoints defined by *utml* − *u*<sup>1</sup> /1*u* = 0.01 and *u*<sup>2</sup> − *utml* /1*u* = 0.01, *with utml* = *u*<sup>1</sup> + *u*<sup>2</sup> /2.

Nepf, 2013). When the canopy has sufficient non-dimensional density (ah > 0.1, ah = frontal area per bed area), the velocity profile resembles a mixing layer, with a hyperbolic tangent profile present between the vertical positions z<sup>1</sup> and z2, defined by the canopy density (**Figure 1**, Ghisalberti and Nepf, 2002). In the temperate seagrass Z. marina, or eelgrass, the blade density falls within this high-density regime (ah = 0.1–3, based on data provided in Dennison and Alberte, 1982; Chandler et al., 1996; Moore, 2004; McKone, 2009; Lacy and Wyllie-Echeverria, 2011; Infantes et al., 2012; Hansen and Reidenbach, 2013; Reidenbach and Thomas, 2018). As the shoot density increases, the vertical extent of the mixing layer region inside the canopy is reduced (increasing z1, **Figure 1**) and the thickness of the mixing layer (tml <sup>=</sup> <sup>z</sup><sup>2</sup> <sup>−</sup> <sup>z</sup><sup>1</sup> ) is also reduced. In the upper canopy (z > z<sup>1</sup> in **Figure 1**), the turbulence is dominated by canopy-scale vortices, which form in the shear layer at the top of the canopy. The canopy-scale vortices significantly enhance vertical turbulent diffusion in the upper canopy. For dense canopies, the canopy-scale vortices do not penetrate to the bed, and in the lower canopy (z < z1, **Figure 1**) the turbulence is dominated by much smaller blade-generated vortices, which results in a much smaller vertical turbulent diffusion (Luhar et al., 2008; Nepf, 2012).

Seagrass pollen has a filamentous shape, with Z. marina pollen length L<sup>p</sup> = 3 to 5 mm and diameter B<sup>p</sup> = 7.5e − 6 m (Ackerman, 1997). The filamentous shape enhances the capture efficiency relative to spheroid pollen (de Cock, 1980; Ackerman, 2002). Z. marina is monoecious and selfcompatible, producing spadices with both male and female flowers that develop acropetally along each reproductive stalk (Ackerman, 2006). Flowering shoots are slightly taller than vegetative shoots (Ackerman, 2002). Within each spathe, flower development is protogynous, which prevents self-fertilization at the scale of the inflorescence, although geitonogamy may occur across flowering shoots that are part of the same clone (e.g., Reusch, 2001; Rhode and Duffy, 2004). Successful seed production requires pollen movement among spathes and flowering shoots, and the movement of pollen is expected to be highly dependent on the location of release within the canopy, and the canopy- and landscape-scale flow structure (Kendrick et al., 2012; Follett et al., 2016).

In this paper, we numerically simulated the trajectory of individual pollen grains to examine factors that impact pollen transport and capture in a seagrass canopy, focusing on the dominant seagrass species in the northern hemisphere, Z. marina, or eelgrass. Previous work has considered the dispersal of floating seagrass seeds and propagules, including rafted seeds of Z. marina, which are spread via large-scale coastal currents (Kallstrom et al., 2008; McMahon et al., 2014; Grech et al., 2016). Here we consider transport at the meadow scale, including the vertical turbulent diffusion that carries pollen out of and into a submerged meadow. The simulation was used to explore the impact of submergence depth and pollen release height. The effect of pollen shape was also investigated, building upon previous studies of the role of filiform pollen (Ackerman, 1995, 2000). Finally, the transport trends were related to observed genetic diversity in seeds produced from paired inflorescences located at higher and lower vertical positions on the same Z. marina reproductive shoot. Recruitment from seeds is increasingly recognized as important for seagrass population dynamics as well as genetic connectivity (Kendrick et al., 2012); thus a more detailed consideration of the hydrodynamic influences on pollen transport can improve our understanding of the processes underlying successful sexual reproduction in seagrasses, and in doing so enhance future investigations of meadow genetic diversity, clonal structure and demography.

# 2. METHODS

This study explored the dispersion of negatively buoyant particles (representing pollen) using a random displacement model (RDM) to simulate the trajectory of individual pollen grains. Within the model, the velocity profile u(z) and eddy diffusivity profile Kz(z) were predicted from canopy physical properties, including the canopy height, shoot density, blade width, and shear velocity at the top of the canopy, which is equal to the square root of the Reynolds stress at the top of the canopy, u<sup>∗</sup> = q −u ′ w ′ <sup>h</sup> (Follett et al., 2016). Experimental observations of velocity and turbulent diffusivity in a series of rigid and flexible submerged model canopies (Ghisalberti and Nepf, 2002, 2004, 2005, 2006; Lacy and Wyllie-Echeverria, 2011) were used to validate the modeled profiles of time-mean velocity and eddy diffusivity.

# 2.1. Vertical Profiles of Velocity and Eddy Diffusivity Used to Parameterize RDM

The model used coordinates (x, y, z) aligned with the current, in the cross-stream direction, and normal to the bed, respectively (**Figure 1**). The velocity vector −→<sup>u</sup> <sup>=</sup> (u, <sup>v</sup>, <sup>w</sup>) corresponded to the coordinates x, y, z , respectively. An overbar denotes a time-average. The model only considered dense canopies (ah & 0.1), representative of many Z. marina meadows (e.g., Dennison and Alberte, 1982; Chandler et al., 1996; Moore, 2004; McKone, 2009; Lacy and Wyllie-Echeverria, 2011; Infantes et al., 2012; Hansen and Reidenbach, 2013; Reidenbach and Thomas, 2018). For a dense canopy, the velocity profile within and above the canopy resembles a mixing layer, with a hyperbolic tangent profile centered slightly above the top of the canopy (Ghisalberti and Nepf, 2002):

$$\overline{u}\left(z\right) = 0.5\Delta u \tanh\left(\frac{z - \left(h + 0.5\theta\right)}{2\theta}\right) + 0.5\left(u\_1 + u\_2\right) \tag{1}$$

Here, u<sup>1</sup> and u<sup>2</sup> are the velocities at the limits of the mixing layer, z<sup>1</sup> and z2, respectively, and 1u = u<sup>2</sup> − u<sup>1</sup> (**Figure 1**). The momentum thickness of the mixing layer, θ, is related to the physical thickness of the mixing layer tml = z<sup>2</sup> − z<sup>1</sup> = 7.1θ (Ghisalberti and Nepf, 2002). As summarized below, the empirical constants included in the mixing-layer profile (Equation 1) were estimated using measurements over nine rigid and six flexible canopies, with different frontal area density and depth-averaged velocity, the details of which are given in Ghisalberti and Nepf (2002, 2004, 2005, 2006). The height of the flexible canopies was defined as the maximum blade height as determined from video footage (Ghisalberti and Nepf, 2006). First, the velocity difference across the mixing layer (1u = u<sup>2</sup> − u1) is correlated with shear velocity at the top of the canopy, u<sup>∗</sup> Specifically, 1u = u∗ 0.14 for rigid canopies (Ghisalberti and Nepf, 2005), and 1u = u∗ 0.11 for flexible canopies (based on data presented in **Table 1**, Ghisalberti and Nepf, 2006). Second, the lower endpoint of the mixing layer, z1/h, increases with canopy density, with <sup>z</sup><sup>1</sup> h dependent on the canopy drag parameter, CDah, with C<sup>D</sup> the canopy drag coefficient. Specifically, <sup>z</sup><sup>1</sup> <sup>h</sup> <sup>=</sup> 0.39 <sup>−</sup> 0.078 <sup>C</sup>Dah for rigid canopies, and <sup>z</sup><sup>1</sup> <sup>h</sup> <sup>=</sup> 0.63 <sup>−</sup> 0.29 <sup>C</sup>Dah for flexible canopies, based on data presented in **Table 1**, Ghisalberti and Nepf (2006). Third, the extent of the mixing layer region tml = h−z<sup>1</sup> 0.38 . For submergence depths less than <sup>H</sup>/<sup>h</sup> <sup>=</sup> 2, the mixing layer is restricted by the water surface, and the penetration of the mixing layer into the canopy is limited to the distance H−h from the top of the canopy (Nepf and Vivoni, 2000), so that for H/h ≤ 2, the mixing layer extends from z<sup>1</sup> = h − (H − h) to the water surface (z<sup>2</sup> = H). Fourth, the velocity in the lower canopy, u1, is related to the velocity at the top of the meadow, uh. Specifically, the ratio <sup>u</sup><sup>1</sup> uh decreases with non-dimensional meadow density, ah. Specifically, <sup>u</sup><sup>1</sup> uh = 0.38 <sup>C</sup>Dah−0.26 for rigid canopies; and u1 uh = 0.21 <sup>C</sup>Dah−0.45 for flexible canopies (Ghisalberti and Nepf, 2006). Finally, the velocity difference across the mixing layer (1u), normalized by the velocity at the top of the canopy uh, is a function of the frontal area per canopy volume a and the stem diameter d ( 1u uh = 6.8 ad0.42 , Ghisalberti and Nepf, 2004). In order to construct velocity profiles using Equation (1), z<sup>1</sup> was first found using CDah. Next, u<sup>∗</sup> or 1u was chosen to define a

*z/h* = 1

*z/h* > *z*2*h*

TABLE 1 | Equations describing the curves of (A) mean longitudinal velocity and (B) eddy diffusivity, which require the canopy shear velocity, canopy height, drag coefficient, frontal area, and blade width.


#### (B) Kz profile:

*K<sup>z</sup>* = (0.17 ± 0.08) *uL<sup>v</sup> z/h* < *z*1/*h K<sup>h</sup>* = 0.0961*u h* − *z*<sup>1</sup> *K<sup>z</sup>* = 0.0391*u h* − *z*<sup>1</sup> Using these relations, *K<sup>z</sup>* = *f*(*u*∗, *h*, *CD*, *a*, *Lv*)

(C) canopy parameters:


*Physical parameters (C) of the canopies considered in this paper (Ghisalberti and Nepf, 2005, Lacy and Wyllie-Echeverria, 2011) are presented.*

specific flow condition, and uh, u1, and u<sup>2</sup> were found using the fitted relations described above.

The vertical profile of eddy diffusivity is divided into two regions. In the lower canopy (z < z1), the eddy diffusivity scales with the size of the vegetation elements, Lv, which here is the seagrass blade width. For conditions that produce turbulence in the blade wakes ( <sup>u</sup>1L<sup>v</sup> <sup>ν</sup> > 100, with ν the kinematic viscosity of water), the vertical eddy diffusivity is K<sup>z</sup> = (0.17 ± 0.08) uL<sup>v</sup> (Lightbody and Nepf, 2006). Within the mixing layer (h ≥ z ≥ z1), canopy-scale vortices elevate Kz. The eddy diffusivity peaks at z = h and scales with the mixing layer thickness, tml, and velocity difference, 1u (K<sup>h</sup> = 0.0321utml, Ghisalberti and Nepf, 2005). When the submergence depth is low (H/h . 5), the eddy diffusivity has a constant value (K<sup>z</sup> = 0.0131utml) above the mixing layer (z > z2) (Ghisalberti and Nepf, 2005). The eddy diffusivity was assumed to follow a linear function between the regions of constant diffusivity (z<sup>1</sup> > z > z2) and z = h. An example of a K<sup>z</sup> profile is included in **Figure 1**. Lateral transport was assumed to be dominated by blade-scale turbulence, so that the lateral diffusivity was assumed to be equal to K<sup>y</sup> = (0.17 ± 0.08) uLv. For validation of the method, the vertical profiles of velocity and diffusivity predicted from the above relationships across a range of canopy densities are compared to measured velocity and eddy diffusivity from experiments reported in Ghisalberti and Nepf (2005, **Figure 2**). On average, the predicted profiles agreed with the measurements to within 10%.

#### 2.2. Random Displacement Model (RDM) for Particle Transport

Within the random displacement model (RDM), individual particles originated at a specific source height within the meadow (0 < z < h) and were tracked until they deposited on the canopy, settled to the bed, or left the model domain. The size of the model domain in the directions (x, y, z) was (200h, 20h, H). In each time-step (1t) the position of the particle (xp, yp, zp) advanced longitudinally with the mean velocity u(z) and vertically due to both settling (ws) and turbulent eddy velocity (w ′ ). The equations used to model the particle position were (Wilson and Sawford, 1996):

$$\mathbf{x}\_{\rho,i+1} = \mathbf{x}\_{\rho,i} + \overline{u}\left(\mathbf{z}\_{\rho,i}\right)\Delta t \tag{2}$$

$$\mathcal{Y}\_{\mathcal{P}^{j,i+1}} = \mathcal{Y}\_{\mathcal{P}^j} + \mathcal{R}\_{\mathcal{Y}} \sqrt{2\mathcal{K}\_{\mathcal{Y}}\left(\mathbf{z}\_{\mathcal{P}^j}\right)\Delta t} \tag{3}$$

$$z\_{p,i+1} = z\_{p,i} + \left(\frac{dK\_z\left(z\_{p,i}\right)}{dz} - w\_s\right)\Delta t + R\_z\sqrt{2K\_z\left(z\_{p,i}\right)\Delta t} \text{ (4)}$$

Nepf (2005), and the predicted profiles are shown with solid lines. The colors denote different experimental case. In alphabetical order of experimental run, *u*<sup>∗</sup> = 0.45, 0.74, 0.53, 1.33, 0.84, 0.50, 1.57, 0.86, 0.55 cm/s, *C<sup>D</sup>* = 0.81, 0.77, 0.85, 0.67, 0.71, 0.82, 0.61, 0.66, 0.79, and *a* = 0.025, 0.034, 0.034, 0.040, 0.040, 0.040, 0.080, 0.080, 0.080 cm−<sup>1</sup> . For all cases the canopy height was *h* = 13.9 cm and the plant width was *L<sup>v</sup>* = 0.64 cm.

The last term in (3) and (4) represents transport by turbulent velocity v ′ <sup>=</sup> <sup>R</sup><sup>y</sup> p 2Ky/ √ t and w ′ <sup>=</sup> <sup>R</sup><sup>z</sup> √ 2Kz/ √ t, respectively, with Ry, R<sup>z</sup> random numbers drawn from a normal distribution with mean 0 and standard deviation 1. The vertical transport (Equation 4) also included a drift correction, or pseudo-velocity, associated with the vertical variation in diffusivity (dKz/dz). The pseudo-velocity term prevented the artificial accumulation of particles in regions of low diffusivity (Durbin, 1983; Wilson and Yee, 2007). The formulations for u (z), K<sup>y</sup> (z) and Kz(z) were described in the section Vertical Profiles of Velocity and Eddy Diffusivity used to Parameterize RDM. The particle position was saved at every time-step. The constant model time-step, 1t, was constrained so that the vertical particle excursion within each time-step was much smaller than the scale of vertical gradients in the diffusivity and velocity (Israelsson et al., 2006). Within the canopy, both the velocity and diffusivity varied over length scales of approximately 0.1h. Within each time-step, after a particle was moved, the position was assessed to determine if the particle had settled to the bed (zp,<sup>i</sup> = 0), reached the water surface (zp,<sup>i</sup> = H), traveled beyond the model domain (xp,<sup>i</sup> > 200h, <sup>y</sup>p,<sup>i</sup> <sup>&</sup>gt; <sup>20</sup>h), or deposited to the canopy.

Z. marina pollen contains a sticky coating, increasing adherence to surfaces contacted by the pollen grains (de Cock, 1980). Based on this, the canopy capture model used for this study assumed that all pollen particles contacting vegetation surfaces were captured and ceased further transport. Canopy deposition was described as the probability of capture of one particle in one time-step, found from the probability that the particle was located upstream of the projected area of a blade, multiplied by the capture efficiency, or the fraction of particles

removed from the volume of water passing the projected area of the blade (Palmer et al., 2004). The probability that the particle was upstream of the projected area of a blade was equal to the blade width L<sup>v</sup> = 0.0055 m (McKone, 2009) divided by the average center-to-center spacing between blades (1S) (**Figure 3A**). In addition, successful canopy capture required that the particle be located close enough to the blade so that it could impact the blade within one time-step, which was equal to the longitudinal particle excursion in one time-step (1xp,<sup>i</sup> = u zp,<sup>i</sup> t) divided by the average longitudinal spacing between blades (1S). The capture efficiency (η) was based on the measured efficiency for particle impaction on rigid cylinders coated with Vaseline (Palmer et al., 2004), assuming that the blade width could be substituted for the cylinder diameter:

$$
\eta = 0.224 Re\_c^{0.718} R\_p^{2.08} \tag{5}
$$

with the collector Reynolds number Re<sup>c</sup> = uLv ν . The particle ratio R<sup>p</sup> = dp Lv is the ratio of the particle diameter (dp) to the blade width, so that Equation (5) indicates that the capture efficiency increases with increasing particle diameter. The impact of pollen grain shape was explored by simulating both filamentous and spherical pollen grains with the same settling velocity and volume. The filamentous pollen grains were given the following attributes based on Ackerman (2002): w<sup>s</sup> = 3.5e − 6 m/s, L<sup>p</sup> = 4e−3 m, B<sup>p</sup> = 7.5e−6 m. The spherical pollen grains were given the same settling velocity as the filamentous grains and diameter d<sup>p</sup> = 3.5e−5 m. To adapt Equation (5) for the filamentous pollen, an effective pollen diameter was determined in the following way. The pollen filaments rotate continuously (Ackerman, 1997), with the pollen having equal probability of assuming any rotation angle (γ , φ) from the (x, y, z) axes, such that all possible pollen endpoints were evenly distributed over the surface of a sphere. Within each time-step, the effective pollen diameter was assumed to be the projected length of the pollen on the y-axis, so that a pollen grain with its long axis aligned with the y-axis would have an effective diameter equal to L<sup>p</sup> and a pollen grain with its long axis aligned with the x-axis would have an effective diameter equal to the diameter of the pollen, Bp. At each time step, a new particle rotation angle was chosen for each particle from a uniform distribution of all possible pollen positions. Specifically, j and k were chosen from a uniform distribution on (0, 1). Then

$$
\gamma \,= 2\pi j \,\tag{6}
$$

$$\phi = \cos^{-1}(2k - 1) \tag{7}$$

The effective pollen diameter was equal to the Cartesian transformation of the pollen position from spherical coordinates, d<sup>p</sup> = Lpsinγ sin φ (Edwards and Penney, 2002). Using this model, a pollen grain perpendicular to the y-axis would have diameter 0. In order to accommodate the finite width of real pollen grains, the minimum effective diameter was set equal to Bp.

Multiplying the capture efficiency by the probability that the particle is in the plane of the blade and close enough to contact the blade in one time-step, the probability of particle impaction on a blade surface in one time-step is:

$$dp\_{\mathbf{x}} = \eta \frac{L\_{\mathbf{v}}}{\Delta S^2} \Delta \mathbf{x}\_{\mathbf{p},i} = \eta a\_{\mathbf{x}} \Delta \mathbf{x}\_{\mathbf{p},i} \tag{8}$$

with (a<sup>x</sup> ) the component of the one-sided blade area projected in the x direction. Similarly, the probability of particle impaction on horizontal surfaces is dp<sup>z</sup> = min(0, ηaz1zp,i), with the impact of gravitational settling and turbulent transport included in the vertical particle excursion 1zp,<sup>i</sup> (Equation 2). Particle impaction on lateral surfaces was neglected due to the relatively smaller blade thickness (1 mm).

We used equation 4 in Luhar and Nepf (2011) to describe the current-induced changes in seagrass blade position. To find the component of the blade area projected in the x and z directions at each vertical position, the curved blade posture was broken into linear segments, such that the blade surface was converted to a series of vectors (**Figure 3B**). The projected area in the x direction, Px(z), was found by multiplying the magnitude of the vector segment by the cosine of the angle (cos θ) between the vertical vector (f) and the blade position (g(z)) found by dividing the dot product of f and g(z) by the magnitudes of both vectors (cos θ (z) = f ·g(z) |f||g(z)| , **Figure 3B**). Similarly, the projected area in the z direction, Pz(z), was found by multiplying the magnitude of the vector segment by the sine of the angle (sin (θ)) between the blade position and the vertical vector. The component of the blade area projected in the x and z directions was found by multiplying the onesided blade area (a) by Px(z) and Pz(z), respectively. The probability of particle deposition was the sum of the probability of particle impaction on both the vertical and horizontal surfaces dp = dp<sup>x</sup> + dp<sup>z</sup> . To determine if the particle deposited to the canopy during the time-step, a random number, R<sup>c</sup> , was chosen from a uniform distribution between 0 and 1. If R<sup>c</sup> was less than or equal to the probability of deposition, the particle deposited to the canopy.

The RDM model was validated using measured profiles of dye concentration along a rigid model canopy (case I in Ghisalberti and Nepf, 2004). The dye was released at x/h = 0 and z/h = 1 and measured at six downstream longitudinal locations (x/h = 1.4, 3.9, 6.6, 10.8, 18.0, 27.3). To represent the neutrally buoyant tracer, the RDM particles were given neutral buoyancy (w<sup>s</sup> = 0), retention on canopy elements was omitted, and the bed and water surface (z = 0, H) were treated as no-flux boundaries. In order to calculate simulated concentration and mass flux, a vertical column of interrogation boxes (0.02 m long × 0.01 m high) was defined, centered at the longitudinal measurement points. Particles were continuously released until a steady-state particle concentration was established in each box. The concentration, C(z), was found by dividing the steady-state number of particles in each box by the box volume. The mass flux was found by integrating the flux at each box over the water depth, Q<sup>x</sup> = R u (z) C (z) dz.

The validated transport model was used to explore the influence of several factors on pollen transport, including pollen release height within the canopy, pollen shape, interaction with the water surface, submergence depth ratio (H/h), and depthaverage velocity. The flexible meadow parameterizations were used in order to represent the reconfiguration of real seagrass. The selected range of submergence, H/h = 5, 4, 3, 2, 1.5, represented the range observed for Z. marina canopies at our field sites near Nahant and Gloucester, MA, where H/h = 1.5 to 6.25 at the shallow and deep canopy edges. In the field, pollen grains have sometimes been observed to collect at the water surface, forming a connected floating layer that restricts future transport (de Cock, 1980), while in other instances pollen collected at the surface returned to the water column due to wave and wind action (Ducker et al., 1978; Ruckelshaus, 1996; Ackerman, 1997). These two conditions were represented in the simulation by different surface boundary conditions. On the one hand, the return of pollen from the water surface into the water column was represented as a no-flux, or reflecting, boundary. On the other hand, the collection of pollen on the water surface was represented as an absorbing (perfect sink) boundary condition. For all cases, the pollen was released at the top of the canopy (z/h = 1).

#### 2.3. Genetic Diversity of Seeds Produced on Paired Upper and Lower Inflorescences

Flowering shoots were selected haphazardly, at least 5 meters apart at a depth of approximately 3 m MLLW, from the interior of two Z. marina meadows in northern Massachusetts (Dorothy Cove, Nahant, and Niles Beach, Gloucester). Canopy height (h) ranged from 85 to 135 cm, so that H/h > 2 for all samples. From each flowering shoot (n = 7; 3 from Nahant and 4 from Gloucester), we collected a leaf sample and 6–10 developing seeds from each of two spathes, one from the highest rhipidia and one from the lowest, and the location of those spathes were recorded as the distance above the substrate of the midpoint of the inflorescence. Maternal leaf tissue was frozen at −20◦C until DNA extraction; seeds were frozen at −80◦C. DNA was extracted from leaves by grinding each sample with a Retsch mixer mill MM400 and using the Omega Bio-Tek E.N.Z.A. <sup>R</sup> Plant DNA Kit. Individual seeds were ground by hand in microfuge tubes with micro-pestles and DNA was extracted with Chelex 100 (Bio-Rad); samples were incubated at 55◦C for 8–24 h, gently mixed, heated to 98◦C for 10 min, vortexed, centrifuged at 14,000 rpm for 5 min, and the supernatant stored at −20◦C until PCR.

Each seed and maternal shoot was genotyped using 8 microsatellite loci developed for Z. marina and known to be polymorphic for these populations, multiplexed in 3 11 µl PCR reactions (**Table 1**). Each reaction consisted of 1 µl DNA template, 5 µl 2X Type-It multiplex master mix (Qiagen), and 0.25 µl of each 10 µM primer. PCR cycling conditions included initial activation/denaturation at 95◦C for 5 min, followed by 28 cycles of 95◦C for 30 s, 60◦C for 90 s, and 72◦C for 30 s, and final extension at 60◦C for 30 min. PCR products were separated on a 3730xl Genetic Analyzer (Applied Biosystems) and fragment analysis was performed using GeneMarker version 2.6 (SoftGenetics).

To compare seed diversity across samples that differed in size (n = 6 to 10 seeds per spathe), we calculated allelic richness (AE) rarified to the smallest sample size (i.e., 12 alleles) using the program HP-RARE (Kalinowski, 2005). The program GERUD 2.0 (Jones, 2005) was used to determine the number of unique fathers contributing to each seed set. GERUD uses an exclusion approach to determine the minimum number of sires necessary to produce a given progeny array when the mother's genotype is known, as in our data set. If a progeny array was too complex to solve using all 8 loci, we removed the least informative locus until a solution could be found (Jones, 2005); pairs of inflorescences from the same maternal shoot were always analyzed with the same set of loci. We divided the minimum number of unique fathers by the number of seeds genotyped to standardize comparisons across inflorescences with different numbers of seeds and used paired t-tests to compare the diversity of seeds produced on high vs. low inflorescences on each flowering shoot.

# 3. RESULTS

#### 3.1. Validation of RDM Model

The simulation was validated against vertical profiles of dye concentration measured along a rigid model canopy (case I in Ghisalberti and Nepf, 2004). To mimic the real tracer, the model particles were assumed to be neutrally buoyant (w<sup>s</sup> = 0). The measured and modeled concentration fields displayed similar vertical dispersion for x/h ≥ 3.9, where the measured and modeled observations agreed to within 7% (**Figure 4**). However, closer to the source, x/H = 1.4, the modeled concentration (**Figure 4**, open squares) exhibited greater vertical dispersion than the measured concentration (**Figure 4**, solid circles), as the modeled concentration spread over a larger vertical distance with a reduced peak magnitude relative to the measured concentration. The initially smaller rate of dispersion for the real tracer can be explained by the presence of multiple eddy scales within the real system. The vertical distribution of real tracer was initially influenced by turbulence scales smaller than the canopy-scale vortices, and thus experienced slower diffusion. The canopy-scale vortices did not contribute to the diffusion of the real tracer until the real tracer cloud grew to a size comparable to the canopy-scale vortices. In contrast, the simulation applied a constant diffusivity based on the canopy-scale vortices, so

that the tracer was effectively influenced by the largest scale of turbulence from the point of release, and thus initially dispersed more rapidly than the real tracer.

### 3.2. Validation of Velocity Model

The velocity model was also validated with field measurements in a Z. marina meadow, as reported in Lacy and Wyllie-Echeverria (2011, site ED3). The velocity profile was measured at 3 time points over a tidal cycle, resulting in different depthaverage velocity and canopy height h = 0.84, 0.78, 0.61 m and 1U = 0.032, 0.050, 0.072 m/s (**Figure 5**, open circles, triangles, and squares). The frontal area per canopy volume a = 2 m−<sup>1</sup> was computed for vertical blades from measurements of blade dimensions and shoot density, with blade width L<sup>v</sup> = 1 cm. The frontal area per canopy volume was assumed to be constant over the reconfigured canopy height h. As a canopy composed of flexible blades reconfigures, there is decrease in drag predominantly because the pronated component of the blades contribute little to drag. Luhar and Nepf (2011) demonstrated that the drag imparted by a flexible blade can be represented as an equivalent rigid blade of shorter length, called the effective blade length, le. Since the effective blade is rigid, its imparted drag is described by the drag coefficient for a rigid blade, C<sup>D</sup> = 1.95 (Luhar and Nepf, 2011). We assumed that the deflected meadow height (h) was a reasonable approximation for the effective blade length, but acknowledge that this may produce a slight over prediction, based on the comparison shown in Figure 2 in Luhar and Nepf (2013). The canopy was assumed to be deeply submerged (H/h > 2) based on the reported site elevation of −2 m MLLW. The modeled velocity profile (**Figure 5**, solid, dashed, and dash-dot lines) agreed with the measurements to within a maximum difference of 3.8%.

The likelihood of outcrossing vs. selfing will depend on the relationship between mean clone size (i.e., the area covered by a given genotype) and the pollen dispersal kernel, both of which will vary among seagrass meadows. However, all else being equal, increased within-meadow pollen dispersal will increase opportunity for pollen to encounter non-self-flowers. Therefore, we are interested in what attributes of the meadow impact the maximum distance traveled before canopy capture. The validated model was used to explore the impact of different flow and morphology parameters on pollen transport. The effect of submergence depth, depth-average velocity, pollen shape, and pollen release height were explored, with canopy parameters a = 2 m−<sup>1</sup> , h = 1 m, L<sup>v</sup> = 1 cm, C<sup>D</sup> = 1.95 remaining constant for all simulations.

#### 3.3. Influence of Submergence Depth H/h on Pollen Transport

The simulation was used to understand the impact of submergence depth on long-distance pollen transport. Filamentous particles were released at zrel/h = 1 for submergence depths H/h = 1.5, 2, 3, 4, 5 in a canopy with u<sup>∗</sup> = 0.01 m/s. As discussed in the methods, two surface boundary conditions were considered, particles rebounding off the water surface (**Figure 6**) and particles captured by the water surface (**Figure 7**). For all submergence depths and for both surface boundary conditions, the number of particles in the water column, N, most of which were above the canopy, decreased with the same dependence on distance, N ∼ x −1/2 (**Figures 6**, **7**). This dependence arose because the canopy was a strong sink for pollen, so that the in-water concentration of particles was close to zero within the canopy. Then, the loss of particles to the canopy, ∂N/∂x, depends on the vertical turbulent flux of

FIGURE 6 | Number of particles remaining in the water column with distance from the release point *xrel*/*h* = 0 for 10,000 filamentous particles released at *<sup>z</sup>rel*/*<sup>h</sup>* <sup>=</sup> 1 in a canopy (*u*<sup>∗</sup> <sup>=</sup> 0.01 m/s, *<sup>L</sup><sup>v</sup>* <sup>=</sup> 1 cm, *<sup>a</sup>* <sup>=</sup> 2 m−<sup>1</sup> , *C<sup>D</sup>* = 1.95, *h* = 1 m) with varied submergence depth, *H*/*h* = 1.5, 2, 3, 4, 5, respectively, represented by open triangles, squares, diamonds, circles, and asterisks. Particles were evaluated at 10 points spaced equally in log space, (*x*/*h* = 0.05, 0.13, 0.32, 0.80, 2.0, 5.0, 13, 32, 80, 200). Particles reaching the water surface (*z*/*h* = *H*/*h*) were reflected off of the surface boundary, remaining available for transport within the water column and the possibility of capture on canopy elements. A line illustrating a power law with decay constant −1/2 is plotted for reference (black dashed line).

particles toward the canopy, which in turn depends on the vertical gradient of particle concentration, ∂C/∂z, above the canopy, i.e., ∂N/∂x ∼ ∂C/∂z. The remaining particles, N, are distributed above the canopy over the vertical length-scale for diffusion, δ<sup>z</sup> ∼ x −1/2 , such that the mean concentration above the canopy is Cmean ∼ N/δz. Since the concentration is close to zero in the canopy, ∂C/∂z ∼ Cmean/δ<sup>z</sup> ∼ N/δ<sup>2</sup> <sup>z</sup> <sup>∼</sup> Nx−<sup>1</sup> . Since ∂N/∂x ∼ ∂C/∂z, we can also write ∂N/∂x ∼ Nx−<sup>1</sup> , which has the solution N ∼ x −1/2 .

The particle number (N) deviated from the x <sup>−</sup>1/<sup>2</sup> dependence when the particles began to interact with the water surface. For the reflecting boundary condition (**Figure 6**), the loss of particles to the canopy was accelerated, because the reflected particles were forced back toward the canopy. For example, for the smallest submergence depth, H/h = 1.5 (triangles in **Figure 6**), this transition occurred at x/h = 5, at which point a significant fraction of the particles (26%) had interacted with the surface. For the absorbing boundary condition (**Figure 7**), the transition in particle loss occurred at the same x-positions (i.e., x/h = 5 for H/h = 1.5), but once the particles were captured by the water surface, capture by the canopy ceased, so that the number of particles in the water column stopped decreasing. Particles trapped at the surface could remain in the water column until wave breaking or wind-generated turbulence knocked them free of the surface layer, making them available again for canopy capture. Although wavebreaking is not included in the model, this implied that capture at the water surface is a mechanism that enhanced long-distance dispersal.

Note that the comparisons made in **Figures 6**, **7** assume the same depth-average velocity for all submergence ratios, whereas in the field, water depth and current speed are typically correlated through the tidal cycle. Therefore, the maximum transport of pollen in the field may not occur at high tide (maximum submergence ratio), but at some intermediate point in the tide with higher velocity. Nevertheless, the general trend of greater pollen transport at greater water depths (greater H/h) would be valid when comparing a meadow across a depth gradient.

# 3.4. Influence of Velocity Variation

constant −1/2 is plotted for reference (black dashed line).

The impact of variation in the depth-average velocity was considered by varying the canopy shear velocity between u<sup>∗</sup> = 0.001 and 0.02 m/s, based on the range of u∗ found in four Z. marina meadows (Grizzle et al., 1996; Lacy and Wyllie-Echeverria, 2011). Filamentous particles were released in a canopy with H/h = 5, zrel/h = 1, and a reflecting surface boundary condition. Increasing u∗ was associated with an increase in depth-averaged velocity, U, which spanned 0.8 to 17 cm/s, respectively, for the u∗ values given above. The vertical diffusivity within the mixing layer was linearly proportional to 1u and thus u∗, and for the range of u∗ values spanned Kz,<sup>h</sup> = 4 to 80 cm<sup>2</sup> /s2 , respectively. Because the velocity has higher values above the canopy and lower values within the canopy (**Figure 5**), particles that initially escaped the meadow and traveled for some distance above the meadow, where capture could not occur, had the greatest travel distances. The average distance traveled divided by the time spent above the canopy

(z/h ≥1) was 0.97, 4.8, 19 cm/s, which was close to the average longitudinal velocity above the canopy for each u∗ condition ( <sup>u</sup>|H−<sup>h</sup> <sup>=</sup> 0.99, 4.9, 20 cm/s), indicating that longitudinal transport predominantly occurred when particles were located above the canopy.

Particles released in canopies with higher depth-averaged velocity had only slightly greater travel distances. This is shown by a slightly higher number of particles remaining in the water column for larger values of u∗ at each distance from the release (**Figure 8**). Similarly, increasing the depth-average velocity from 0.8 to 17 cm/s increased the number of particles reaching the end of the domain only slightly, from 86 to 146 (**Figure 8**). That is, increasing U by a factor of about 20 only increased the number of particles leaving the domain by a factor of about 2. The weak dependence of travel distance on depth-average velocity can be explained by the linear dependence of K<sup>z</sup> on the velocity shear (1u) and thus on the depth-average velocity (U). Specifically, the ratio Kz/U was roughly a constant. Therefore, although increasing U allowed particles above the meadow to advect more quickly away from the source, a comparable increase in K<sup>z</sup> enhanced particle flux into the meadow, so that the spatial rate of particle capture by the meadow had little dependence on depth-average velocity.

### 3.5. Influence of Pollen Shape

Pollen shape significantly influenced the maximum distances traveled by pollen. Using the same meadow and release conditions (H/h = 5, zrel/h = 1, reflecting surface boundary, u<sup>∗</sup> = 0.01 m/s), the transport of filamentous pollen (L<sup>p</sup> = 4e − 3 m, B<sup>p</sup> = 7.5e − 6 m) was compared to that of spherical

pollen grains of the same volume and settling velocity, but diameter d<sup>p</sup> = 3.5e − 5 m (**Figure 9**). The filamentous pollen had significantly higher canopy capture rates, resulting in more of the filamentous pollen remaining closer to its release point. Specifically, for the filamentous pollen over 98% of the pollen was captured by the canopy before the end of the modeled domain, and no particles reached the bed (z = 0). In addition, the filamentous pollen remaining in the water column decays following the x <sup>−</sup>1/<sup>2</sup> dependence, which is consistent with the canopy behaving as a nearly perfect sink (see discussion in section Influence of Submergence Depth H/h on Pollen Transport). In contrast, the spherical pollen was not as efficiently captured by the canopy, and as a result the spatial decline in water column pollen was much slower and specifically did not follow the x −1/2 dependence. Further, only 54% of the spherical pollen grains were captured by the canopy within the modeling domain, and 46% of particles traveled beyond the end of the modeling domain.

plotted for reference (black dashed line).

# 3.6. Influence of Pollen Release Height, zrel/h

Vertical diffusivity was elevated by canopy-scale vortices within the canopy mixing layer, z<sup>1</sup> ≤ z ≤ z2. The importance of this region of elevated diffusivity to pollen transport was explored by varying the pollen release height. The lowest release point (zrel/h = 0.25, **Figure 10**) was below the mixing layer (zrel/h < z1/h = 0.49) in a region of reduced diffusivity dominated by blade-scale turbulence. The middle release point (zrel/h = 0.75, **Figure 10**) was within the region affected by both blade-scale and canopy-scale turbulence, which significantly elevated the vertical diffusivity relative to the lower canopy region (**Figure 3**). The highest release point was zrel/h = 1, reflecting the location of the highest inflorescences observed within coastal Massachusetts Z. marina canopies [Dorothy Cove, Nahant, MA; Niles Beach, Gloucester, MA; Woods Hole, MA (Ackerman, 1986)]. For the lowest release point (zrel/h = 0.25), significant canopy capture (63%) occurred close to the release point (x/h < 0.05). Because the vertical diffusivity was low in this region and the release point was far from the top of the canopy, no pollen escaped the canopy from this release point. Specifically, none of the particles traveled more than x/h = 0.43, or roughly 0.43 m. Further, the maximum vertical position achieved by pollen released near the bed was only z/h = 0.29, indicating that pollen released below the mixing layer remained below the mixing layer. The fraction of particles remaining in the water column followed an exponential decay curve, reflecting first-order decay due to the distributed sink formed by the canopy elements. The capture of pollen over such small length-scales makes self-pollination likely. In contrast, at the highest release point (zrel/h = 1), the pollen experienced both higher vertical diffusivity and proximity to the canopy top, both of which facilitated the particles being quickly carried above the canopy, allowing them to avoid canopy capture and experience higher current speeds. Together, this produced much greater travel distances. Specifically, 9% of the pollen released at <sup>z</sup>rel <sup>h</sup> <sup>=</sup> 1 traveled beyond <sup>x</sup>/<sup>h</sup> <sup>=</sup> 5 before canopy capture occurred, with 1.2% (120 particles) reaching the longitudinal extent of the model domain (L = 200h). The tendency to travel greater distances before capture reduces the probability of selfpollination and enhances the probability of genetically diverse, outcrossed seeds. The particles released within the upper canopy region initially followed an exponential decay curve due to the distributed canopy deposition sink, similar to the decay of the particles released at zrel/h = 0.25 (**Figure 10**, open squares and diamonds). However, some particles were able to travel above z = h, after which particle deposition followed the pattern of diffusion to a boundary sink, similar to the release at zrel/h = 1, with the number of particles decreasing with x −1/2 .

#### 3.7. Genetic Diversity of Seeds Produced on Paired Upper and Lower Inflorescences

All eight loci used were polymorphic, displaying 3-7 alleles with a mean of 5.75 alleles locus−<sup>1</sup> . We obtained complete multi-locus genotypes for 125 seeds from seven maternal half-sibling families. The majority of seeds showed Mendelian evidence of outcrossing (i.e., non-maternal alleles at one or more loci), with only 15 of 125 seeds (12%) possessing genotypes consistent with geitonogamous selfing; twice as many of these were found on lower inflorescences than on upper ones (10 vs. 5). Pooling data from both field sites and using a paired approach, we found that both measures of seed diversity tested were greater for inflorescences closer to the top of the canopy. Allelic richness (mean number of alleles per loci) was significantly greater for seeds produced in spadices close to the top of the canopy than those produced lower on the shoot (mean and standard deviation 2.87 ± 0.24 vs. 2.60 ± 0.41; one-tailed paired t-test, p = 0.013), and the difference in allelic richness between sibling seed cohorts tended to increase with increasing linear distance between spadices (**Figure 11**). Likewise, the minimum number of unique fathers was greater for

0.13, 0.32, 0.80, 2.0, 5.0, 13, 32, 80, 200). A line illustrating a power law with decay constant −1/2 is plotted for reference (black dashed line).

seed families collected from upper spadices than from lower ones (3.52 vs. 2.93 fathers/10 seeds; one-tailed paired t-test, p = 0.052).

### 4. DISCUSSION

Although vegetative growth historically has been assumed to be the primary mode of reproduction in seagrasses, multiple lines of evidence suggest that recruitment from seed is both more common and more important than previously believed (Kendrick et al., 2012). For example, significant variation in genetic and genotypic diversity has been documented within and among natural seagrass beds (Williams and Orth, 1998; Olsen et al., 2004; Hughes and Stachowicz, 2009), and disturbed seagrass meadows often recover at a rate faster than could occur with vegetative propagation alone (Orth et al., 2006). Further, there is increasing evidence that the genetic diversity and relatedness of seeds influences resulting seedling performance, which highlights the importance of a diverse pollen supply. The fitness costs of self-pollination in Z. marina vary among populations, but experimentally selfed inflorescences often show significantly lower seed set (Ruckelshaus, 1995; Reusch, 2001; Rhode and Duffy, 2004) and reduced seed germination compared to outcrossed inflorescences (Ruckelshaus, 1995; Billingham et al., 2007). The positive effect of seed diversity can be seen beyond the dichotomy of selfed vs. outcrossed: increased allelic richness in seeds has been shown to increase both primary and secondary production in the field (Reynolds et al., 2012), and genetically diverse, non-sibling seeds outperformed

sibling seeds in a greenhouse mesocosm experiment (Randall Hughes et al., 2016). Thus, the physical processes that affect pollen delivery in the nearshore environment impact both seed quantity and quality, and this can have important ramifications for population dynamics and genetic structure in eelgrass meadows.

The transport distances achieved by the simulated pollen in our model reflected the vertical structure of turbulence within the canopy. Specifically, longitudinal transport over distances greater than x/h = 2 was only possible for pollen released within the region of the upper canopy influenced by canopy-scale vortices (z > z1). The elevated vertical diffusivity within this region carried pollen above the canopy where it experienced higher current speeds and avoided canopy capture until additional longitudinal transport had occurred (**Figure 10**). Observations in real eelgrass canopies suggest that Z. marina inflorescences are located in the vertical zone influenced by canopy scale vortices, suggesting that reproductive stalk resources were directed toward release points that can benefit from higher vertical diffusivity. For example, at our field site, the lowest spadices ranged from z/h = 0.36 to 0.64 in meadows for which the corresponding z1/h = 0.31 to 0.53 was estimated based on the non-dimensional meadow density, <sup>z</sup><sup>1</sup> <sup>h</sup> <sup>=</sup> 0.63 <sup>−</sup> 0.29 <sup>C</sup>Dah . That is, the spadices were all located within the zone influenced by canopy-scale vortices, z > z1. Similarly, the vertical extent of inflorescences in a Z. marina canopy reported by Ackerman (1986) extended from z <sup>h</sup> <sup>=</sup> 0.3 to 1, which corresponds with <sup>z</sup>/<sup>h</sup> <sup>&</sup>gt; <sup>z</sup>1/<sup>h</sup> <sup>=</sup> 0.31, estimated from the non-dimensional meadow density (ah). Over the range of reported dimensionless Z. marina meadow densities (ah = 0.1–3, based on data provided in Dennison and Alberte, 1982; Chandler et al., 1996; Moore, 2004; McKone, 2009; Lacy and Wyllie-Echeverria, 2011; Infantes et al., 2012; Hansen and Reidenbach, 2013; Reidenbach and Thomas, 2018), z1/h = 0 to 0.6. This suggests that in some sparse canopies (ah / 0.5), the canopy-scale vortices penetrate to the bed (z<sup>1</sup> = 0) and spadices over the entire meadow would experience elevated diffusivity. However, for dense canopies, inflorescences located

below z1would experience reduced diffusivity, resulting in much shorter distances for pollen transport.

The effect of vertical flow structure observed in the model can also be seen in our empirical data on seed diversity. Seeds collected from lower inflorescences had reduced allelic richness relative to seeds produced from higher inflorescences on the same flowering shoot (**Figure 11**), consistent with the significantly shorter travel distances for pollen released lower in the canopy in the simulation study. For particles released in the lower canopy, the maximum travel distance predicted was 0.43 m, vs. 5 m and higher for the upper canopy. This difference in travel distance implies that pollen released in the lower canopy is more quickly captured by the canopy and thus has a higher probability of self-pollination. Observed selfing rates in our genotyped seeds were low, as has been found in other eelgrass populations with polyclonal patches (Ruckelshaus, 1995; Reusch et al., 2000; Furman et al., 2015). However, more putatively selfed seeds were found on lower infloresences than upper ones, and upper inflorescences showed greater paternal diversity, suggesting greater access to diverse pollen sources higher in the canopy.

Overall, the simulated pollen travel distances exhibited a wide range, from essentially zero to distances that extended beyond the end of the model domain (x > 200h = 200 m) indicating that both short- and long-distance fatherhood is possible. Furman et al. (2015) similarly found genetic evidence of a range of father distances through genotyped single seeds from the tallest rhipidium on flowering shoots in a dynamic and patchy eelgrass meadow. Fathers were identified, with pollination found to occur from fathers located 0.57–74 m upstream. The lack of viable parents at distances shorter than 0.57 m was attributed to developmental synchrony of neighboring reproductive shoots. The measured pollen dispersal distance was limited by the study size, indicating the potential for pollen transport over further distances, similar to the fate of the modeled pollen grains that reached the longitudinal extent of the domain. Approximately 63% of modeled pollen particles would have been captured by the canopy within the first 0.57 m after release, suggesting that developmental synchrony would aid in the prevention of selfing over short distances. The model results showed a decrease in particles present in the water column with x −1/2 . Because the probability of pollination is directly related to particle density, the probability of pollination would also depend on x −1/2 , favoring long-distance fatherhood relative to a linear dependence. However, this dependence was also shown to depend on the proximity of the water surface (**Figures 6**, **7**). Once particles begin to interact with the water surface, which occurred closer to the source when submergence was lower (smaller H/h), then the capture to the canopy was either accelerated (when the water surface reflected the pollen) or shut-off (when the water surface absorbed the pollen).

Global declines in seagrass distributions may be accompanied by shifting depth distributions (e.g., loss in shallow regions due to high temperature episodes, loss at depth due to eutrophication and resulting light reduction). Modeled variation in submergence depth and pollen release height explored how such changes may impact pollen delivery. Particles that reflected off the water surface boundary displayed reduced longitudinal transport relative to the most deeply submerged canopy (H/h = 5) which was not affected by the water surface within the model domain, L = 200h. For the two most shallow cases considered, H/h = 1.5, 2, all particles were captured within the model domain (x/h = 32, 80). In these cases, long-distance fatherhood would be reduced relative to more deeply submerged canopies due to the influence of the surface boundary, which could weaken observed outcrossing rates in shallow canopies relative to more deeply submerged canopies. However, because the surface only affected particle transport after the vertical extent of the plume reached z = H, pollen traveling shorter distances would not be influenced by the surface boundary.

Z. marina produces elongated pollen with length comparable to blade width (L<sup>p</sup> = 3–5 mm, Ackerman, 2002; L<sup>v</sup> = 5.5 mm, McKone, 2009), with a pollen:ovule ration of approximately 10,000:1 (Ackerman, 2002).The high ratio of pollen length to blade width increased the capture efficiency (Equation 5) of the filamentous pollen grains relative to spherical grains of the same settling velocity. Previous observations (Ackerman, 1997) indicate that the elongated pollen shape increases capture efficiency on flowers as the pollen threads tumble through flowergenerated velocity gradients toward the center of the flower, increasing the incidence of capture on stigmas. High canopy capture rates were reflected in the significantly reduced transport distance for filamentous pollen grains, compared to spherical grains (**Figure 9**). In all scenarios simulated in this study, > 91% of the filamentous pollen was captured by the canopy within the model domain (**Figure 9**). The high rate of canopy capture and resulting short travel distances are consistent with previous observations suggesting that the elongated pollen is not optimized for long distance transport (Ackerman, 1986; Zipperle et al., 2011; Kendrick et al., 2012). The elongated dimension of the filamentous pollen increased the canopy capture rate, promoting successful capture by reproductive stalks, while maintaining a small settling velocity relative to canopy shear stress, <sup>w</sup><sup>s</sup> u∗ = 0.0002 − 0.004. Particles assigned the same settling velocity and an equivalent spherical diameter that conserved particle mass and density (section Influence of Pollen Shape) displayed reduced canopy deposition, with only 27% of spherical particles becoming captured by the canopy before the end of the modeling domain, L = 200h. Pollen with these characteristics would be expected to experience reduced pollination efficiency within a meadow (L < 200h), because over half the pollen released would exit the longitudinal extent of the meadow before becoming captured by the canopy. In contrast, larger particles with a higher settling velocity, as is seen in some terrestrial species such as ragweed ( <sup>w</sup><sup>s</sup> u∗ = 0.04, Boehm et al., 2008) Lycopodium (clubmoss) spores ( <sup>w</sup><sup>s</sup> u∗ = 0.04, Pan et al., 2014) could experience reduced longitudinal transport following periods of slack water as some particles could settle out of the water column. Recall that the canopy capture model used in this study was modified from a model based on observed capture of spherical particles on cylindrical collectors (Palmer et al., 2004). To improve the canopy capture model, future experiments should be done with filamentous pollen and flexible blade collectors. It is possible

that canopy capture on vegetative blades may be reduced due to interaction of the pollen coating and vegetative blade surface, as some sources note that pollen grains may slide off of vegetative blades under some conditions (Ackerman, 2002). These additional aspects should be considered in future research in order to improve characterizations of pollen transport.

To conclude, in this study, field samples of seeds on paired inflorescences and numerical simulations of pollen transport together demonstrated the connection between reduced incanopy transport of pollen particles and reduced seed diversity. Pollen released in the lower canopy had significantly reduced longitudinal transport compared to pollen released in the upper canopy, which traveled for extended periods above the canopy. Consistent with this, seeds from lower inflorescences had reduced allelic richness relative to seeds produced from higher inflorescences. The transport distance of modeled pollen grains increased with increasing submergence depth, owing to the greater availability of flow space above the meadow in which pollen could travel without the likelihood of becoming captured by the canopy. Finally, paired simulations of filamentous and spherical pollen grains showed that filamentous grains were more rapidly captured by the blades, resulting in significantly shorter travel distances for most particles, consistent with conclusions

#### REFERENCES


drawn from previous field studies that filamentous pollen favors shorter dispersal distances (Ackerman, 1986). The results in this study highlight the importance of the physical processes of advection, diffusion, and interception in determining the travel distance of pollen grains and canopy genetic diversity.

#### AUTHOR CONTRIBUTIONS

EF and HN designed the model and analyzed the modeled data. EF implemented the model simulations. HN supervised model development. CH planned and carried out the genetic analyses. All authors discussed the results and contributed to the final manuscript.

#### FUNDING

This material is based on work supported by the National Science Foundation under Grant No. AGS-1005480 to HN; molecular analyses were conducted with support for STEM education from the USNH system to CH. Any opinions, findings, or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.


its possible effects on larval mussel settlement. J. Exp. Mar. Biol. Ecol. 206, 165–177.


accumulation and removal in aquatic systems. Limnol. Oceanogr. 49, 75–85. doi: 10.4319/lo.2004.49.1.0076


**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 © 2019 Follett, Hays and Nepf. 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.

# Bridging the Separation Between Studies of the Biophysics of Natural and Built Marine Canopies

Craig Stevens1,2 \* and David Plew<sup>3</sup>

<sup>1</sup> National Institute of Water and Atmospheric Research, Wellington, New Zealand, <sup>2</sup> Department of Physics, The University of Auckland, Auckland, New Zealand, <sup>3</sup> National Institute of Water and Atmospheric Research, Christchurch, New Zealand

We argue that there is a separation between studies of the biophysics of natural and "built" marine canopies. Here, by "built" we specifically refer to floating, suspended aquaculture canopies. These structures, combining support infrastructure and crop, exhibit several unique features relative to natural marine canopies, in that they take a particular species, suspend them in spatially structured, mono-cultured arrangement and then induce a systematic harvest cycle. This is in contrast to natural canopies that are irregular and variable in form, have natural recruitment and growth, and sustain some level of biodiversity and more exposed to climate extremes. We synthesize published work to identify the points of difference and similarity with natural canopy studies. This perspective article identifies four main themes relating to (i) key scales, (ii) structural configuration, (iii) connections between biology and physics and (iv) connecting natural and built canopy science. Despite clear differences between natural and built canopies, they have more in common than not and we suggest that both sub-fields would benefit from better connection across the divide.

#### Edited by:

Marco Ghisalberti, The University of Western Australia, Australia

#### Reviewed by:

Rodrigo Riera, Catholic University of the Most Holy Conception, Chile Andrew William Mackay Pomeroy, The University of Western Australia, Australia

### \*Correspondence:

Craig Stevens craig.stevens@niwa.co.nz

#### Specialty section:

This article was submitted to Marine Ecosystem Ecology, a section of the journal Frontiers in Marine Science

Received: 20 November 2018 Accepted: 08 April 2019 Published: 24 April 2019

#### Citation:

Stevens C and Plew D (2019) Bridging the Separation Between Studies of the Biophysics of Natural and Built Marine Canopies. Front. Mar. Sci. 6:217. doi: 10.3389/fmars.2019.00217 Keywords: canopies, stratification, mixing, biophysics, flow-structure interactions

# INTRODUCTION

In the marine environment, natural canopies (kelp beds, seagrass meadows etc.) provide significant ecological value (Seitz et al., 2013). At the same time, "built" canopies provided by aquaculture installations have economic value (Troell et al., 2014). We argue that, in part because of these different high-level drivers, there is a separation in the marine canopy biophysics literature whereby it is rare to see studies make the connection between natural marine canopies and built aquaculture canopies. In this Perspective article we consider the nature of the two classes of canopy, explore key comparative themes and then make suggestions for how to build interconnections between the two sub-fields.

Natural canopies are irregular and variable in form, have natural recruitment and growth, and sustain some level of biodiversity. Studies on natural canopies tend to have objectives that are ecosystem function-centric, documenting ecosystem services like maintaining habitats, promoting biodiversity and sustaining pathways of nutrients and matter (e.g., Duarte, 2000). Aquaculture canopies are constructed, regular structures with a controlled mono-culture life-cycle (recruitment and harvesting) and an associated diverse transient fouling community and associated motile

fauna (Stevens et al., 2008). Restored natural canopies form an intersectional category (Fonseca and Fisher, 1986). Aquaculturebased studies have focused primarily on (i) how various natural processes affect crop production and (ii) the role of cultured stocks in controlling some coastal ecosystems; largely as a result of how their feeding activity interacts directly and indirectly with energy flow and nutrient cycling. Ecological studies on aquaculture systems have much in common with studies on natural canopies, except that they tend to have at least one anthropogenic objective (i.e., what is the local carrying capacity? e.g., Byron et al., 2011).

Despite these overarching differences, there is little to distinguish the two classes of canopies from hydrodynamic and even biophysical perspectives. Considering flow distortion, turbulence, waves and nutrient transport; natural and built marine canopies have more in common than not – so why are their respective literatures so distinct? This perspective article addresses several fundamental questions in the context similarities and differences between built and natural canopies. First, is the separation real? Assuming it is, what are the key differences in canopy-induced processes for determining feeding and scales of food depletion? Are the differing details of the canopy structural configuration important? Finally, in the two systems do biological aspects like productivity and physiology connect differently with physical processes like transport and turbulence?

### DOES THE SEPARATION EVEN EXIST AND IS THERE BENEFIT TO CLOSING IT?

A consideration of selected papers from the reference list supports the argument that a separation exists as it is rare to see an aquaculture hydrodynamics paper referenced in a natural canopy study, although the reverse is somewhat more common. Built canopies have key attributes that differ from natural canopies (Stevens et al., 2008; Klebert et al., 2013), prime amongst these is structural regularity (**Figure 1**). For example, a shellfish longline tethers a sequence of buoys at the water surface and then suspends the shellfish crop beneath. Multiple long-lines are laid out at regular intervals in parallel, occasionally with gaps of open water between lines to improve flushing and/or navigation, to form a "farm." Similarly, caged fish structures moor a sequence of cages side by side along with their operating infrastructure. The suspended cages or crop extend vertically anywhere from a few meters to nearly on the sea bed. The result, whatever the arrangement, is that a population finds itself constrained mid-water column, some distance from shore with its success depending on adequate food supply and waste removal. The literature on built canopies is clearly useful, in itself, for applied questions relating to crop production and environmental impact. However, some understanding can be transferred in either direction (see **Table 1**) when considering related processes in natural canopies. In other words, what discoveries from the built literature can aid in understanding natural canopies – and vice versa?

(submerged used here as an example) canopies suggesting that while there are similarities in generalities (flow deformation, wakes, shear layers, etc.) there are substantial differences in the details like mooring/holdfast, buoyancy and nutrient supply. The decaying wave orbits in both panels indicate the approximate direct influence of surface waves.

# WHAT ARE THE KEY SCALES FOR DETERMINING "FOOTPRINTS"?

When considering either canopy system, one can identify several key scales. The outer boundary conditions are set by (1) the farfield "embayment and inner continental shelf " scale which relates to the regional oceanography and meteorology. Depending on the scale of the canopy, this can reasonably be carried out in the absence of consideration of the effect of the canopy. If, however, the canopy is extensive it may well affect the wider circulation (Plew, 2011b). (2) The canopy/farm-scale is that at which crop feeding and effect is strongly influenced by the canopy and so representation of the details of the canopy is required. Focusing on the immediate region around the canopy provides information about the conditions that the population must live with. This highlights a common challenge at the canopy scale trying to separate the "canopy effect" from background variability. (3) The crop/organism scale is where the feeding takes place (**Figure 1**). This is likely influenced by the canopy-scale modulation of far-field drivers and needs detailed representation of the canopy and flow variability including waves.

At the regional to large scale, while models can have a vertical dimension (Wu et al., 2017), if the canopy is represented at all, it is typical to represent canopy drag as an enhanced bottom friction in 2D models (Grant and Bacher, 2001; Plew, 2011a,b; Shields et al., 2017). This approximation precludes detailed interaction between canopy and water column stratification (Plew et al., 2006) which can influence where food comes from as well as the fate of impacted water. This remains a challenge for both sub-disciplines.

Focus at the scale of the plant or crop forming the canopy brings the benefit that this is the scale at which biophysical interactions (photosynthesis, feeding, nutrient absorption, etc.) take place (e.g., Ackerman and Nishizaki, 2004). It is instructive to see approaches to determining useful scaling for natural

TABLE 1 | Comparison of key questions for built and natural canopy systems.


canopy physical effects, like the laboratory experiments of Rosman et al. (2010), consider different physical representations for different scales. In a similar way, numerical models of built canopies move from highly resolved representations at fine scale (Delaux et al., 2011) to coarse representation at wider scales (Plew, 2011a; O'Donncha et al., 2015; Zhao et al., 2017) but the need to represent the canopy in a meaningful way remains. The same decisions need to be made when representing natural canopies (e.g., Mullarney and Henderson, 2018; van Rooijen et al., 2018).

The boundaries of any canopy are critical to quantifying fluxes of material in and out of the canopy region – and are often a key point of difference between built and natural canopies (**Figure 1**). Consequently, flow instability due to velocity shear is a key flux in both natural (Ghisalberti and Nepf, 2006) and built (Plew et al., 2006) systems. A suspended canopy can potentially sustain such boundaries both above and below the canopy. Adequately determining the mixing at the boundaries of the canopy is fundamental to understanding the canopy biomechanics. This is borne out when examining canopy retention times at larger scales in both natural and built canopies where any attempt at a nutrient budget requires estimation of boundary fluxes (e.g., Nepf et al., 2007; Venayagamoorthy et al., 2011; Lin et al., 2016). In addition, the leading and training edges form critical points whereby accelerated flow around the canopy (Tseung et al., 2016) or the recirculation in the trailing edge (Liu et al., 2017) perform a key mixing role in the near-field.

#### ARE THE DETAILS OF THE STRUCTURAL CONFIGURATION IMPORTANT?

It is important to consider if the details of the structural configuration have a bearing on the canopy effects – or if mixing or larger-scale variability dominate flows and transfer rates. Certainly, it has been long-established that substrate heterogeneity affects marine species (e.g., Menge, 1976). Shellfish long-line farming structures have a particular set of scales associated with their configuration (number of droppers, spacing, canopy size etc.). But this is not the only such arrangement to be found in built marine systems. Rafted shellfish structures tend to have quite dense aggregations of vertical lines beneath each raft (Newell and Richardson, 2014) but the rafts themselves are relatively well spaced to allow navigation and avoid line entanglement (Blanco et al., 1996). Fish cage installations take this to an extreme where several cages (more like flexible pens) are usually side by side and then the next installation is some distance away. A further complication with fish cages is they confine a free-swimming crop – sometimes in the tens of thousands per cage (He et al., 2018).

The ultimate cross-fertilization in canopy science is the improved understanding of the interaction of different types of canopies (O'Donncha et al., 2017). Abiological canopies are relevant in the discussion of built vs. natural canopies as they provide examples of flow evolution in the absence of any biological process. Marine renewable energy extraction arrays, while not common, are planned and in development in many places and offshore wind farms are already common (Broström, 2008). These would introduce what could be considered very large, sparse canopies. Wave energy devices, in particular, have much in common with early floating breakwaters which in turn were involved in some of the earliest canopy hydrodynamics studies (Elwany et al., 1995; Seymour, 1996). Non-canopy forming aquaculture such as on-bottom bivalve cultivation (e.g., Petersen et al., 2013; Saurel et al., 2013) provides extreme examples of structural configuration effects.

It is common for built canopies to be comprised of regular elements of crop (**Figure 1**) so that there are gaps in the canopy distribution affecting flow and in-out of canopy exchange (Stevens and Petersen, 2011). High resolution numerical

simulations found flow behavior very sensitive to flow-canopy orientation (Delaux et al., 2011). Mixing rates will affect the influence of any heterogeneity (Abdolahpour et al., 2017). Similarly, the role of gaps in canopy-flow exchange is a wellstudied aspect of terrestrial canopies (e.g., Bohrer et al., 2008). Despite this, there has been less examination of this facet in the natural canopy literature (except e.g., Rosman et al., 2007; Kregting et al., 2011; El Allaoui et al., 2015; Hamed et al., 2017). Related to this point, most built canopies (aquaculture) are relatively sparse compared to natural canopies, and their dimensions commonly insufficient for a fully developed canopy flow to occur (the transition length is a function of the density of the canopy, Tseung et al., 2016). It is also important to recognize that the structural regularity actually makes some of the built canopy-scale observations possible. For example, it would not be possible to tow an undulator through most natural macroalgal canopies, whereas it is possible to work in the channels between shellfish rafts and long lines (Cranford et al., 2014).

An example of where built canopy understanding can benefit from natural canopy work is the role of surface waves (**Figure 1**). The macroalgal literature continues to make substantial advances in understanding of the response of individuals and canopies to wave forcing (e.g., Denny et al., 1997; Mullarney and Henderson, 2018). While some open-ocean studies exist (Plew et al., 2005; Gentry et al., 2017), aquaculture typically takes place in sheltered waters, where it is implicit that the farms will only be exposed to fetch-limited (i.e., short wavelength) waves. The influence of a surface wave decays exponentially with depth to zero, at half the wavelength (which is short). A consequence is that in aquaculture studies it is unusual to consider the influence of waves upon shellfish growth and behavior. However, it is clear that, as for a buoyant macroalgal frond, suspended canopy or raft shellfish culture structures directly link the crop to the surface wave field through the buoyancy of the supporting elements (Stevens et al., 2007). Therefore, even short wavelength waves can induce vertical velocities at depth in built canopies. This results in a relative velocity between crop and water – potentially with a strong vertical component. Again, relative motion in suspended natural canopies has been explored (Stevens et al., 2001; Lowe et al., 2005), along with waves associated with the canopy itself (monami – Ghisalberti and Nepf, 2009). So, it is not remarkable that this could affect shellfish and even caged fish and is a topic where the built canopy science stands to gain from the experiences of the macroalgal biophysics literature.

# ARE THE CONNECTIONS BETWEEN BIOLOGICAL AND PHYSICAL PROCESSES UNDERSTOOD?

There are a range of connections between biology and physics – most of which will vary depending on canopy organism (e.g., Riisgård et al., 2011) and the in-canopy flow conditions. At the largest scale, the changing climate is central to natural canopy biophysical studies (e.g., Arias-Ortiz et al., 2018). Phenomenon like marine heat waves are impacting on both kinds of canopy and the associated ecosystems (Wernberg et al., 2016), but only the built canopy systems are in a position to rapidly adjust (Weatherdon et al., 2016).

Often these connections at the large scale, and the canopy differences, cascade to incorporate smaller scale interactions. For example, mussels generate feeding flows associated with their siphons (Riisgård et al., 2011), algae rely on diffusive processes (Hurd, 2000) and caged fish generate canopy-scale flows through their motility (Gansel et al., 2014; Johansson et al., 2014). The flow variability in the crop near-field suggests that any instantaneous snapshot of food depletion from a single point would fail to identify the effect of the feeding. This highlights the need when talking about scale to consider temporal scales also. For example, a "nutrient halo" whose scale is determined by time-averaging background turbulence and feeding rate as identified by Nielsen et al. (2016). Working at this crop-scale in natural (O'Riordan et al., 1995) and built (Plew et al., 2009) shellfish applications has a direct analogy with biophysics at the canopy-forming individual scale.

In a recent review of the magnitude and spatial extent of food depletion by bivalve aquaculture, Cranford (2018) emphasized the important role of the built canopy in limiting food depletion at all spatial scales. Canopy-induced flow reduction directly affects the degree of food depletion around individual mussel droppers as a result of the mussels re-filtering more of the same water instead of the feeding zone being sufficiently replenished by advection. This canopy effect at the scale of individual mussels ultimately limits the degree of depletion at canopyand ecosystem-scales. Canopy-induced flow reduction has been shown to result in food depletion levels that are substantially lower than predicted by models that do not account for this physical effect (Cranford et al., 2014; Nielsen et al., 2016). Similar effects are seen in natural canopy studies (Boyle et al., 2004) but it would appear to be more difficult to achieve the depletion vs. distance measurements in a natural canopy as the crop elements are not so consistently spatially separated from one another.

Turbulence is a key quantity that influences transfer rates of material and energy both within, at the boundaries of, and in the background flow. One overarching question in all canopy mechanics is – "does the canopy generate more turbulence?" The answer is seemingly obvious – yes, due to the wakes of all the structural material in the water column. However, with the drag of the canopy and the slow-down of flow at the embayment scale (Plew et al., 2005), the system has the potential to produce less drag and wake at the individual element scale (Cranford et al., 2014). This is a demonstration of the multi-scale nature of flow-canopy interaction and seen equally in natural canopies (Ackerman and Okubo, 1993). Getting this balance right is fundamental to understanding mass fluxes in canopies (e.g., Pilditch et al., 2001; Strohmeier et al., 2005; Gaylord et al., 2007). Furthermore, the nature of the canopy edges has some bearing of how the wake develops as a more porous canopy with a relatively diffuse boundary will likely have a different wake than the abrupt end of a built canopy (**Figure 1**). The limited range of scales in built canopies lends itself to being more readily empirically quantified – the results of which are useful to both sub-disciplines.

# CLOSING THE SEPARATION

fmars-06-00217 April 20, 2019 Time: 18:52 # 5

Despite the various differences (**Figure 1** and **Table 1**) in canopy structure between built and natural systems, they have much in common. Furthermore, we suggest that studies of a particular approach (field observation, numerical simulations, physical model) have more in common with each other, regardless of being built or natural canopies, compared to the differences between approaches – which are substantial. It is not uncommon to see that studies designed to explore natural canopy behavior, especially where the approach involves physical or numerical modeling often appears as if it would have at least as relevant to built canopies as it does with the target natural canopy (e.g., Rosman et al., 2007; Qiao et al., 2016). Conversely, because they are constructed from a limited set of scales, built canopy studies seek to reduce, as much as is possible, the level of complexity when exploring flow-canopy interactions.

It is clear that the flow of information should be bidirectional. While debatable, we suggest that the natural canopy studies approach a problem with a more open perspective (e.g., Gaylord et al., 2007; Fram et al., 2008) than aquaculture studies which might have some focused targets (e.g., Maar et al., 2010; Newell and Richardson, 2014). This is in-part, we believe, due to the differing overarching drivers (economic vs. ecosystem) as well as the only partially overlapping research communities.

This perspective article is a call to better connect between the two fields of endeavor – so how do we do this? Certainly, it requires a better awareness of the various literature threads from both sides and cross-over of application of results. A common language should be possible relating in-water mixing processes to the canopy and substrate and the associated feedbacks

### REFERENCES


(Nepf, 1999). It is likely that the research communities do not overlap sufficiently, so targeted combined special sessions at conferences would be one avenue of fostering better integration. A valuable result of such sessions would be an integrated review article. Ultimately, studies that seamlessly integrate across the separation to get at the best representation of bio-mechanics stand to generate the greatest advances.

#### AUTHOR CONTRIBUTIONS

CS and DP conceived the ideas. CS wrote the manuscript.

# FUNDING

This work was supported by the National Institute for Water and Atmospheric Research (NIWA) Aquaculture Sustainability Program, the Danish Council for Strategic Research, research project "Production of Mussels: Mitigation and Feed for Husbandry (MuMiHus)" under grant agreement no. 09-066983 and the New Zealand Sustainable Seas National Science Challenge Project 4.2.2 Stressor Footprints.

#### ACKNOWLEDGMENTS

We would like to thank Pernille Nielsen, Peter Cranford, Jens K. Petersen, Barb Hayden, Joe O'Callaghan, and Niall Broekhuizen for support and advice. We would also like to thank two reviewers and the editor for their input and suggestions.


giant kelp, Macrocystis pyrifera. Limnol. Oceanogr. 53, 1589–1603. doi: 10.4319/ lo.2008.53.4.1589


mussel farm in the eutrophic Skive Fjord. Denmark. Aquac. Environ. Interact. 8, 311–329. doi: 10.3354/aei00175



**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 © 2019 Stevens and Plew. 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.

# Biophysical Interactions in Fragmented Marine Canopies: Fundamental Processes, Consequences, and Upscaling

#### Andrew M. Folkard\*

Lancaster Environment Centre, Lancaster University, Lancaster, United Kingdom

Spatial fragmentation is a near-ubiquitous characteristic of marine canopies. Biophysical interactions with fragmented canopies are multi-faceted and have many significant implications at multiple scales. The aims of this paper are to review research on biophysical interactions in fragmented marine canopies, identify current gaps in knowledge and understanding, and propose ways forward. The review starts at the patch/gap scale and focuses initially on hydrodynamic interactions. It then considers the consequences of these interactions for particulate and dissolved material, and distributions of canopy-associated organisms. Finally, it addresses issues of upscaling to landscape-scale and ways in which this research can be applied to marine landscape management. Work on a broad range of canopy types is considered, including micro-algal biofilms and turf algae; macro-algae, seagrasses and coral reefs; saltmarsh vegetation and mangroves. Although the focus is on marine canopies, insights from studies of fragmented canopies in other contexts are drawn on where relevant. These include freshwater environments and terrestrial forests, grasslands, crop canopies, and urban areas. Specific areas requiring greater attention are highlighted. As a result of this meta-analysis, the following recommendations are made for further research. A lack of basic data is identified across all canopy types regarding the formation, fate and spatial and temporal characteristics of canopy patches, gaps, and spatial structure. Studies of hydrodynamics with fragmented canopies would benefit from shifting focus toward more non-uniform, realistic configurations, while ecological research in this area would benefit from a move toward configurations that are more controlled and tractable for quantitative modeling. More comparative studies across canopy types would enable understanding of their biophysical interactions and their consequences to be more fully tested and developed. A greater incorporation of chemical aspects of canopy systems into work that has hitherto focused on biophysical interactions would also be pertinent. Upscaling of patch and gap-scale phenomena to landscape-scale is identified as a crucial topic, since it is at the latter scale that management efforts are most readily carried out. Overall, an approach that balances hydrodynamics, marine canopy ecology, spatial analysis of landscapes, biogeochemistry, and socio-environmental interactions is recommended.

#### Edited by:

Marco Ghisalberti, The University of Western Australia, Australia

#### Reviewed by:

Mariana Mayer Pinto, University of New South Wales, Australia H. Nepf, Massachusetts Audubon Society, United States

> \*Correspondence: Andrew M. Folkard a.folkard@lancaster.ac.uk

#### Specialty section:

This article was submitted to Marine Ecosystem Ecology, a section of the journal Frontiers in Marine Science

Received: 13 August 2018 Accepted: 13 May 2019 Published: 07 June 2019

#### Citation:

Folkard AM (2019) Biophysical Interactions in Fragmented Marine Canopies: Fundamental Processes, Consequences, and Upscaling. Front. Mar. Sci. 6:279. doi: 10.3389/fmars.2019.00279

Keywords: marine canopies, biophysical interactions, hydrodynamics, fragmentation, upscaling, patches, gaps

# INTRODUCTION

fmars-06-00279 June 6, 2019 Time: 18:18 # 2

## Aims, Methods, and Structure

Spatial fragmentation – which can range from apparently random distributions to strongly-ordered patterning – is a near-ubiquitous characteristic of biotic and biogenic marine canopies (Thomson et al., 2012; Folkard and Bouma, 2016). It is found in both newly-developing and damaged canopies, as well as in established canopies in equilibrium with their surroundings. It can indicate healthy, resilient ecosystem functioning (Pringle and Tarnita, 2017) or stress and increasing – often anthropogenic – pressures (Fraschetti et al., 2012). The effects of interactions between hydrodynamic processes and fragmented canopies are complex, and dependent on a variety of biotic and abiotic factors. They are of fundamental importance to the structure, functioning and services of canopy ecosystems, since hydrodynamic processes are primary causes of (mechanical) stresses and facilitations (e.g., nutrient supply) to canopies and the ecosystems they support (e.g., Folkard, 2016). The hydrodynamics of fragmented canopies are more spatially heterogeneous than the hydrodynamics of homogeneous canopies – because of the spatial heterogeneity of the canopies themselves – and this leads to heterogeneity in their stresses and facilitations, e.g., spatial variations in sheltering (e.g., Folkard, 2005) and nutrient supply (e.g., Morris et al., 2008). Fragmentation of canopies also leads to their becoming more vulnerable to external pressures (Gera et al., 2013). Understanding the large-scale impacts these effects have is therefore important for managing the many coastal areas where marine canopies are found, especially those affected by anthropogenic stresses and climate change (El Allaoui et al., 2016).

The term 'marine canopy' can cover a wide range of types of bed cover. These have been studied with different emphases, according to the priorities and drivers of the research communities working on them. This paper brings together research on spatial canopy fragmentation and its interactions with physical processes (primarily hydrodynamics, but also sediment transport and other phenomena driven by hydrodynamics) from this wide range of canopy types. From this collation, it creates a structured synthesis of work in this field. From this, it compares approaches and progress in work focused on different canopy types to determine what can be learnt about each one, and to identify ways of developing more universal understanding. To perform this collation of literature, searches were carried out using Web of Science<sup>1</sup> . Searches were made for articles that included in their title, abstract or keywords: words beginning with "patch," "gap," "fragment," or "heterogen"; one or more canopy-type name (seagrass, seaweed, macrophyte, macroalgae, kelp, mangrove, saltmarsh, alga, biofilm, coral) or the word "canopy" or "canopies"; and one or more word referring to hydrodynamic processes (flow, current, wave, turbulence, hydrodynamics). For each word, wildcard asterisks were added to the end so that plurals and other related words would be found (e.g., turbulen<sup>∗</sup> was used to pick up turbulent, turbulence, turbulently etc.). This gave several thousand results, which were then filtered by date (the last 10 years – since 2008, as the initial search was completed in 2018) and for papers before that date, by number of citations (selecting only those papers with > 50 citations). The cutoffs for date and citation number were chosen semi-arbitrarily to reduce the number of papers to a sufficiently large, but manageable amount. The resulting sample of several hundred papers were then filtered by subjective analysis of their titles and, in cases where this did not produce a clear decision, by reading their abstracts. The resulting list was checked to ensure that it included key papers from the author's own knowledge of the literature. This resulted in a final set of approximately 300 papers on which this review was constructed, although not all of them survived the drafting process to appear in the final manuscript. Subsequently, smaller, more focused searches were used to identify papers on specific topics (approaches to habitat fragmentation in landscape ecology and flow-canopy interactions in terrestrial contexts) to fill gaps that arose in earlier drafts of the review. A small number (∼10) papers recommended by reviewers have also been added to the final form of the manuscript.

The remainder of the paper is structured as follows. I begin, in the next section, by justifying my use of patches and gaps as the 'unit elements' of canopy fragmentation, and identifying the types of canopies to be covered. The following sections then lay out the structured synthesis of research across canopy types mentioned above. This starts from fundamental processes and works up to their consequences at landscape-scale for restoration and management. Thus, firstly, the causes, formation and evolution of individual canopy patches and gaps are reviewed. I then focus on the hydrodynamics of individual patches and gaps. This is followed by a section reviewing current knowledge of the consequences of these hydrodynamics-patch/gap interactions in terms of the transport and deposition of particulates and solutes, and the distribution of organisms. I then consider the upscaling of these consequences, firstly via studies of interactions between multiple patches and gaps, and then from a landscapescale perspective. I then consider application of this work to marine canopy landscape restoration and management. Finally, I take a comparative overview of all this work, identify key research questions and recommend possible ways forward. Although the focus here is on marine benthic canopies, throughout, insights from studies of fragmented canopies heterogeneity in other contexts will be drawn upon where they provide relevant insights. These include freshwater lentic and lotic environments and terrestrial environments, including forest, grassland and crop canopies and urban areas.

# FRAMING: PATCHES, GAPS, AND CANOPY TYPES

# Characterization of Canopy Fragmentation in Terms of Patches and Gaps

The landscape-scale spatial structure of the sea floor in the coastal zone is often made up of patches of biota of various different sizes within a larger abiotic matrix (Robbins and Bell, 1994). The

<sup>1</sup>https://clarivate.com/products/web-of-science/

inverse situation is also possible, in which the matrix made up of an approximately continuous biotic or biogenic canopy punctuated by gaps of relatively bare substrate or reduced canopy density (El Allaoui et al., 2016). Thus, these landscapes can be characterized spatially in terms of patches and gaps. There are many metrics associated with this perspective – for example, patch or gap size and density, edge length density (the mean length of canopy edge per unit area) – that have become widely used, particularly via earlier versions of the widely-used landscape ecology software package FRAGSTATS (McGarigal and Marks, 1995). As the field of landscape ecology has developed, alternative perspectives, in which fragmented landscapes are characterized by gradients, have been promoted (e.g., McGarigal and Cushman, 2005; McGarigal et al., 2009). More recently, there have been calls to re-think the conceptual foundations of landscape ecology, in light of better understanding of the non-linearity of relationships that determine the effects of fragmentation on ecosystems (e.g., Didham et al., 2012; Villard and Metzger, 2014; Liao et al., 2017). Thus, in some ways, research into canopy fragmentation has moved significantly beyond its characterization in terms of patches and gaps. Nevertheless, there are also many situations, where canopies are commonly organized into clearly delineated patches (e.g., salt marsh pioneer zones), or extended cover with clearly delineated gaps (e.g., seagrass meadows, mangroves). Therefore, the patch/gap conceptualization remains important. It also provides an idealized "unit element" of fragmentation for modeling studies of interactions between hydrodynamics and other physical processes, and fragmented canopies. Since this review starts from a focus on canopy-hydrodynamics interactions, therefore, it uses the patch/gap conceptualization.

This raises the question of how patches and gaps are defined. In many situations, this will be straightforward – as noted above, canopy patches and gaps are clearly delineated in many marine contexts. However, at times their edges are indistinct, and at others it is not clear whether a canopy distribution is a single but morphologically-complex patch, or a mosaic of several individual patches. For further discussion on this issue, the reader is directed toward Schoelynck et al. (2018). Hereinafter, I will assume that patch and gap edges are clearly defined.

### Canopy Typology

In a marine context, a range of benthic canopies may be identified. They can be distinguished from each other in many ways. Given that this review starts with canopy-hydrodynamics interactions, they are distinguished here in terms of the main parameters that govern these interactions, namely: (i) their height compared to the surrounding substrate and water depth; and (ii) the rigidity or flexibility of the canopy-forming organisms. The choice of what to include and exclude from this typology is, inevitably, somewhat subjective. Because the intention of the paper is comparative, the choice of canopy types is deliberately broad. Moreover, what constitutes a patch of canopy, and what is deemed the surrounding matrix depends on the scale of interest. Thus, for example, when studying seagrass meadows, the matrix of "bare substrate" surrounding patches of seagrass is typically covered by lower-growing organisms. But if the lower-growing organisms are the community of interest, interest will focus on a smaller spatial scale, at which their vertical structure becomes relevant (e.g., Salta et al., 2013), and at which they can therefore be considered to be canopies. Hence, the smallest canopies included in this comparison are "micro-canopies" including micro-algal biofilms and turf algae. The second type of canopy is arguably the "classical" canopy, consisting of a permeable region whose elements rise significantly above the substrate. These include macro-algae beds, seagrass canopies and coral reefs. In the cases of macro-algae and seagrasses, these are made up of highly flexible elements, which pronate in the direction of flow, and oscillate back and forth in response to wave-forcing. The third category consists of pioneer saltmarsh vegetation but also includes pneumatophore roots of mangrove genera such as Avicennia and Sonneratia. These canopies are also permeable, but their elements tend to be more rigid than macro-algae and seagrasses. Moreover, they are found in inter-tidal regions, so that sometimes they fill the full water column depth and emerge above the water surface, while at others they may be fully submerged. Finally, mangrove genera with stilt roots such as Rhizophora form relatively rigid sub-aqueous canopies that are always emergent. Many of the areas of research reviewed herein are dominated by studies of one of these canopy categories, whereas others have been investigated in the context of several. Moreover, some of these types of canopy form patches or gaps more than others. For example, saltmarsh plants such as Spartina anglica typically form clearly defined patches in the pioneer zone, and canopy gaps are relatively common occurrences in seagrass meadows and mangroves. On the other hand, whilst coral reefs are highly spatially heterogeneous, they are not typically organized into clearly distinguished patches and gaps (Lowe and Falter, 2015). As a result, while there is an extensive literature on coral reef hydrodynamics (for reviews, see Monismith, 2007; Lowe and Falter, 2015), there is very little on the hydrodynamics of coral patches or gaps.

# PATCHES AND GAPS: FORMATION AND CAUSES

#### Patches

Canopy patches are often found in areas that have been newly-colonized by the canopy-forming organism. The patches can grow from propagules or by clonal growth in clonal organisms such as seagrasses. Propagules can be dispersed widely, hence the creation of isolated patches. This dispersal can occur over very long distances, notably in seagrasses (Grech et al., 2016). and mangroves (Harwell and Orth, 2002).

Once they establish, the spatial structure of canopy patches is controlled by their growth strategy, which differs between species. For example, dense plant species can cause flow to deflect and accelerate around themselves, which militates against patch expansion because new shoots cannot survive in these faster flows. As a result, they tend to expand by releasing plant fragments which establish on bare substrate away from the parent patches and grow into new patches. On the other hand,

sparse species, which alter the flow less strongly, can grow more easily via expansion of existing patches (Verschoren, 2017). The distribution of patches is often affected by geology, topography and shelter from hydrodynamics stresses (e.g., energetic waves or strong tidal currents), although the underlying factor is usually access to sufficient nutrients and light (Koch, 2001). For example, Parnell (2015) found patches of giant kelp (Macrocystis pyrifera) and elk kelp (Pelagophycus porra) occurred preferentially on topographic highs, and Rinde et al. (2014) found that patches of the kelp Laminaria hyperborea established mainly on ridges. The higher elevation thus afforded to the kelp provides better access to water column nutrients and light. In contrast, Di Carlo et al. (2005) found that in shallow, well-lit waters, seagrass establishment was densest in topographic lows, because of the deeper layer of nutrient-rich sediment there. Similarly, Nardin et al. (2016) found that mangroves in a rapidly prograding area of the Mekong delta expanded as continuous coverage in areas of high sediment availability, but as sparse patches in areas of lower sediment supply. In both these cases, greater access to substrate nutrients is the determining factor. Patchy colonization may also be associated with preferential growth in areas where hydrodynamic stresses are lower, which may themselves be patchily distributed. For example, Francoeur et al. (1998) found that microform bed clusters provided refugia that allowed spatially-patchy establishment of periphyton in fast flowing rivers.

Canopy patchiness may also occur in post-disturbance areas, where canopies are damaged but recovering. The spatial pattern that this results in is affected by multiple factors. Underwood (1998) found that in an inter-tidal community of the macroalgae Hormosira banksia, post-storm patchiness was primarily caused by variations in the amount of damage sustained during the storm. Such recovery may also be subject to Allee effects (the weakening of individual plants' reproductive success with decreasing population density) leading to the canopy's accelerated decline, for example due to pollen limitation in sparse communities of intertidal Zostera noltii. This can impair their recovery even after environmental conditions have improved from those that caused their decline (van Tussenbroek et al., 2016).

As well as being indicative of newly-colonizing and post-disturbance canopies, patchiness can be a consequence of communities' spatial self-organization (Rietkerk and van de Koppel, 2008). This tends to result in pattern formation (Pringle and Tarnita, 2017), which can be complex and not just uniformly-spaced or -sized patches or stripes (van Wesenbeeck et al., 2008; van de Koppel et al., 2012). For example, it may be manifested as frequency distributions of patch size or separation distance which follow power law functions, i.e., where the probability that a patch's size (or distance from its nearest neighbor patch) is greater than some value s is proportional to s−<sup>β</sup> , where β is a constant (Schoelynck et al., 2012). Spatial self-organization is defined as a process whereby large-scale ordered patterns emerge from disordered initial conditions due to small-scale interactions between organisms (Liu et al., 2014). It is generally considered to be due to scale-dependent feedbacks (Rietkerk and van de Koppel, 2008) – processes whereby organism interactions change from positive to negative as spatial scale is varied. For example, pioneer salt marsh vegetation shoots grow together in patches for mutual protection and sediment and nutrient retention. But this causes flow to accelerate around them, causing increased substrate scouring which militates against shoots establishing within a certain distance of an established patch. Thus, near-field positive effects combined with far-field negative effects create patches (van Wesenbeeck et al., 2008). In other cases, the positive effect of nutrient retention by patches is counterbalanced by reducing light availability for each shoot as patches get larger (Gera et al., 2013).

#### Gaps

Homogeneous canopies in which gaps may form tend to occur in regions where conditions are relatively uniform over wide areas, i.e., in the sub-tidal or in mangrove forests, rather than in transitional areas such as the inter-tidal, where canopies are often patchier. The main causes of gap formation include edaphic variations; damage or removal; burial and grazing. Edaphic variations result in canopy gaps where substrate is relatively poor or thin. Damage or removal can be due to natural forces – typically storms, but also mechanisms such as ice scouring (Cervin et al., 2004) – or anthropogenic causes, such as boat anchors, propellers, or infrastructure installation (e.g., Serrano et al., 2016). Likewise, burial can be caused by natural movements of bed material, usually during storms (Bell et al., 1999), or human activity such as deposition of dredging spoil. Herbivorous grazing can be relatively subtle, creating small scale mosaics of herbivore preferred and avoided patches, promoting plant biodiversity and resilience (Weerman et al., 2011; Howison et al., 2017), or more complete, leading to invasion by other species (e.g., Davies et al., 2007) or the creation of bioturbation pits (Yager et al., 1993). Grazing appears to be more important in some types of canopy (e.g., seagrass meadows, Townsend and Fonseca, 1998; Gera et al., 2013; periphyton biofilms, Gresens and Lowe, 1994; Holomuzki et al., 2006) than in others (e.g., macro-algae beds), although it may have a role in the persistence of gaps in the latter (Thomson et al., 2012). In some contexts, gap formation can be more complex. For example, in Argentinian salt marshes, Escapa et al. (2015) found salt pans formed gaps in Sarcocornia-dominated areas. They deduced that this was due to Sarcocornia growing in dense patches and providing shelter for crabs (Neohelice granulata). The crabs construct burrows, causing the Sarcocornia patch centers to die off. The patches then lose elevation relative to the surrounding marsh, and salt pans form within the depressions. Thus, the gaps are formed by interactions between biotic and abiotic processes.

Data on the extent, number and size of gaps in marine canopies is patchy. Bell et al. (1999)reported that gaps constituted 2.4–5.7% of a monospecific meadow of the seagrass Halodule wrightii. Gaps in algal canopies have been found to be larger at exposed sites than on sheltered coastlines, suggesting that a combination of exposure to external stresses and internal ecosystem context – e.g., canopy composition and grazing intensity – is fundamental to gap characteristics (Wernberg and Connell, 2008; Gera et al., 2013). Gap age appears to vary greatly. For example, Bell et al. (ibid.) found that most gaps in the

Halodule wrightii meadow they studied persisted for less than 6 months, whereas Thomson et al. (2012) inferred, from the ages of invertebrates within them, that gaps in the macroalgae canopies they studied had persisted for decades.

As is the case for patches, gaps are often in a dynamic state of evolution. Both seagrass (Halodule wrightii, Bell et al., 1999) and macrolagae (Ascophyllum nodosum, Cervin et al., 2004) canopy gaps have been observed to recover by re-growth or recruitment of the same species as the surrounding meadows. Jimenez-Ramos et al. (2017) noted that gaps and sparse areas of canopies have less self-shading, which should boost growth in them, homogenizing the canopy. Similarly, in a study of flow through seagrass meadow simulations made up of sparser and denser regions in various configurations, Adhitya et al. (2014) found that the flow tends to favor supply of water column resources to sparse areas, which may lead to homogeneity. In other cases, however, gaps are filled by other species. This can be caused by these invasive species being faster growing (e.g., Cervin et al., 2005) or having other advantageous species characteristics. For example, Vogt et al. (2012) observed that the regeneration of open patches after hurricanes in mangroves was dominated by flood-tolerant Rhizophora, which outcompeted the faster-growing, but less flood-tolerant pioneer Laguncularia racemosa), while Voerman et al. (2017) found that the ability of the invasive species Caulerpa filiformis to respond better to disturbance than native species allows it to outcompete them. Gap-filling mechanisms may also be affected by the clarity of gaps. For example, Wernberg and Connell (2008) found fucalean algae to be recruited into complete clearings in a macroalgae canopy, whereas turf algae cover was more prominent in less complete clearings. The relative competitiveness of re-growth of meadow immediately surrounding a gap versus invasion into the gap from outside the meadow is likely affected by gap geometry and the permeability of its surrounding meadow. This determines whether flow arriving from through the canopy (favoring the surrounding meadow) or over the canopy (favoring invasion from outside) dominates within the gap (Folkard, 2016).

# HYDRODYNAMICS OF CANOPY PATCHES AND GAPS

Much work has been carried out in recent decades on the hydrodynamics of marine canopy patches, gaps, and their boundaries. Most of this work has been concerned with canopies of seagrass, saltmarsh plants or mangroves. These canopies are in many ways analogous to forest, crop or urban canopies in terrestrial settings, and much understanding has been developed by comparison of results from studies of aquatic and terrestrial canopies.

Significant contributions in this area have been made via analytical, numerical and physical modeling. The last of these has largely been carried out in laboratory flumes (or in wind tunnels for terrestrial canopies), often using artificial simulants – wooden dowels for rigid plants and plastic strips for flexible plants (see Thomas et al., 2014, for further discussion). These approaches are deliberately reductionist, the intention being to strip away complexity and focus on quantitative, mechanistic understanding of specific biophysical interactions, with the subsequent aim of inferring their wider implications and thus moving toward greater understanding in a "bottom–up" way. They have also tended to focus on uni-directional currents, rather than on waves. There has, however, been significant recent progress in understanding wave interactions with homogeneous canopies through studies of simulated canopies representing a range of flexible and rigid vegetation (Pujol and Nepf, 2012; Pujol et al., 2013a,b) and canopies of seagrass (e.g., Zhang et al., 2018), saltmarsh vegetation (Moeller et al., 2014; Maza et al., 2015) and coral reefs (Lowe and Falter, 2015). Moreover, El Allaoui et al. (2015, 2016) have analyzed wave interactions with canopy gaps aligned both parallel and perpendicular to the wave direction.

Measurements of flow velocity – the essential form of hydrodynamic data – are analyzed by separating them into a mean flow field – the long-term average velocity at each location – and a turbulent field – the time-varying aspect of the velocity that remains when the mean velocity is subtracted from the time series of measured velocity. In biophysical terms, the impact of hydrodynamics on canopies can be divided into positive processes (supply of nutrients, dispersion of propagules etc.) and negative processes (e.g., physical stresses, which may lead to damage or uprooting). The mean flow and turbulence fields play different roles in these processes. Canopy organisms respond by adjusting their physiology and growth strategies, which in turn alter their impacts on the hydrodynamics. Thus, there arises a non-linear interaction between hydrodynamics and canopy biology. This rest of this section reviews recent work aimed at understanding these interactions at the single patch or gap scale. Historically, research in this area focused first on simpler, more idealized two-dimensional configurations, and then proceeded to more complex and realistic three-dimensional configurations. Therefore, the following exposition adopts the same distinction.

# Two-Dimensional Patch and Gap Simulations

The generic configuration under consideration here consists of a steady flow encounters a patch of a permeable canopy of obstacles with a uniform upstream edge. At the upstream edge of the patch, the flow adjusts to the presence of the canopy. If the canopy has non-negligible height and is permeable, it does this partly by flowing at an accelerated speed over the patch (the "overflow") and partly by flowing at a slower rate through the patch (the "throughflow"). If the canopy is flexible, the flow causes it to pronate, thus the canopy also adjusts to the hydrodynamics. Belcher et al. (2003) provided an idealized model of this situation and identified three stages of adjustment. In the first, pressure due to canopy drag decelerates the flow in an impact region upstream of the canopy. The second region is an adjustment region that extends a distance L<sup>C</sup> into the canopy. Here, flow decelerates until there is a local balance between downward turbulent transport of momentum from the overflow and removal of momentum by canopy drag forces. L<sup>C</sup> is inversely proportional to the canopy density (the frontal area of canopy elements per unit bed area), so the extent of edge effects varies

with canopy structure (Peterson et al., 2004). L<sup>C</sup> may be longer than the patch itself, so the flow doesn't reach equilibrium within the patch. Indeed, in highly fragmented landscapes, the flow may always be under the influence of canopy edges (Dupont et al., 2011). The third region is where the flow reaches equilibrium with the canopy. Here, the flow structure within the canopy depends on canopy density variations: in general, flow will be faster at heights at which the canopy structure is sparser. Depending on the height and density of the canopy, the flow above the canopy in this region may resemble either a mixing layer (denser canopies) or a boundary layer (sparser canopies) (Sukhodolova and Sukhodolov, 2012).

For patches of other types of canopy, the adjustment will be similar. Micro-canopies, in which through flow will be very small in comparison with the overflow, can be idealized for hydrodynamic purposes as changes in bed roughness with no significant change in bed height. In these conditions, the overflow will adjust to the different roughness characteristics of the patch compared to the surrounding substrate (Chamorro and Porte-Agel, 2009). This increases both the generation of turbulence and the bed shear stress. The former is generally advantageous to canopy organisms, since it increases the supply of nutrients through vertical turbulent diffusion. The latter is generally deleterious, since it increases mechanical stress, which can lead to physical damage or removal. For emergent canopies, only the throughflow will occur. Its mean flow and turbulent characteristics are determined by the size, spacing and frontal area density of the canopy elements (Nepf, 1999, 2012; James et al., 2004).

In terms of the turbulent flow field, enhanced turbulent energy is found close to leading canopy patch edges (Folkard, 2005). Similar effects are seen in wind fields at rural-urban transitions and upwind forest edges (Cheng and Porte-Agel, 2016). In emergent aquatic canopies, such as mangroves, this may be partly due to wave breaking, but also to turbulence generation in canopy element wakes (Norris et al., 2017). As the flow develops into the patch, the turbulence evolves. The sharp gradient in flow speed between the patch and its surroundings creates strong shear layers, in which coherent turbulent structures are generated (Siniscalchi et al., 2012). In a terrestrial context, Dupont and Brunet (2009) found that Kelvin-Helmholtz instabilities develop at the top of forest canopies, where there is a quasi-discontinuous change in the drag conditions between the region within the forest and the clear air above. As they move along the canopy, they roll over, then form transverse vortices. Secondary instabilities then destabilize these vortices, and by nine canopy heights downstream, they have become complex coherent structures. Submerged aquatic canopies have similar effects on the flow, but there are differences because of the finite depth of the water above the canopy. In aquatic contexts, the growth of the coherent structures stops when the production of the turbulent kinetic energy that feeds them in the shear layer at the top of the canopy is balanced by dissipation of that energy within the canopy (Ghisalberti and Nepf, 2004). These structures, and thus influence of the canopy overflow, penetrate a significant depth into the canopy, vertically dividing the canopy into an upper region dominated by the overflow and a lower region dominated by throughflow (Nepf and Vivoni, 2000). These two layers often have very different flushing timescales, which can lead to their ecology and water quality also being different (Nepf and Ghisalberti, 2008).

Downstream of canopy patches in two-dimensional configurations, where the canopy edge geometry and flow velocity allow, recirculation zones form immediately behind the canopy which are similar to those found downstream of impermeable obstacles with a backward-facing step configuration (Detto et al., 2008). Whether this occurs or not, the shear layer at the top of the canopy extends downstream of the patch, forming a free shear layer "wake." In the wake, the turbulence increases first as the shear layer grows (the "near-wake"), then decreases as it decays (the "far-wake") (Folkard, 2005). At the same time, a new boundary layer starts to form above the bed downstream of the patch. As this grows, it comes to dominate the wake, which decays downstream. In the transition region from the canopy edge to the point where the wake is negligible and the bed boundary layer completely re-established, the flow structure is dynamic and multi-layered (Folkard and Bouma, 2016). This region often extends far downstream of the patch: Markfort et al. (2010) found that wind adjusting to a lake surface downstream of a tree canopy had reduced surface shear stress up to 50 canopy heights downwind of the transition, and in wind tunnel experiments, Markfort et al. (2014) found that mean turbulent quantities required at least 100 canopy heights to adjust to the new surface.

Less work has been done on the hydrodynamic influence of canopy gaps than on patches. Folkard (2011) compared flow in submerged canopy gaps to Morris's (1955) characterization of skimming flow, wake interference flow and isolated roughness flow, expanding the typology to five categories by separating wake interference flow into recirculation flow, boundary layer recovery, and canopy throughflow. He found that the type of flow that occurred could best be predicted using a Reynolds number based on overflow speed and gap depth, and the gap aspect ratio (i.e., the ratio of the gap length to its height). A Froude number based on the same speed and length scales was found to predict bed shear stress in the gaps well. Extending this work, Adhitya et al. (personal communication) found that longer leaves, lower shoot densities, deeper water and narrower gaps all led to dominance of throughflow over overflow in determining conditions in canopy gaps. In a study of wave interactions with canopies, Lowe et al. (2005) found that short wave orbital velocity is not significantly diminished in canopies compared to bare substrate, in contrast to canopies' significant attenuation of current velocities. As a result, Luhar et al. (2008) suggest that fragmented meadows are more likely to persist in current-dominated environments, because of the enhanced current feedback within canopy gaps, than in wave-dominated environments, where there will be a tendency toward homogeneity because of this lack of feedback. El Allaoui et al. (2015, 2016) reported flume experiments in which waves interacted with gaps aligned perpendicular and parallel to the wave direction, simulating sagittal channels that form perpendicular to coastlines in seagrass canopies due to

currents transporting waters mixed near the shoreline seaward. They found that, for both types of gap, wave velocity increased over the gap compared to the canopy and that denser canopies attenuated both wave velocity and turbulent kinetic energy within adjacent gaps, compared to sparser canopies. Modeling based on these results showed that, for the same total gap area, canopies with large gaps cause more mixing than canopies with small gaps.

## Three-Dimensional Patch and Gap Simulations

As computer power and physical modeling facilities have developed, more hydrodynamic studies of three-dimensional patches have been carried out, although to date, there appear to have been no three-dimensional studies of canopy gap hydrodynamics. Most commonly, 3D patch experiments have used idealized, circular patches made up of uniform elements. The flow is diverted around their sides, as well as flowing over and through them. Horseshoe or necklace vortices form around the upstream patch edge (Chang and Constantinescu, 2015; Chang et al., 2017). The flow accelerates as it moves around the patch, the lateral distance from the patch where maximum flow occurs increasing with patch size (Vandenbruwaene et al., 2011). As at the top of the canopy, there is a strong velocity gradient across the lateral patch edges, causing coherent horizontal vortices to form (Yan et al., 2016). These enhance lateral transport across the patch edges (Zong and Nepf, 2011).

Downstream of three-dimensional patches, the wake structure is complicated in comparison to the two-dimensional case by the convergence of the flow around the patch with the overflow and throughflow. For dense patches where overflow dominates throughflow, the patch width-height ratio determines the orientation of wake vortices. If the height is less than the width, vortices form in the vertical plane within a few patch heights downstream of the patch. If the height is greater than the width, horizontal vortices form closest to the patch and control velocity recovery within the wake (Liu et al., 2018). In cases where the throughflow is significant compared to the overflow, there are two peaks of turbulent intensity behind a circular patch. The first is directly behind the patch and related to the wakes of the individual patch elements. The second is further downstream and related to the patch-scale wake (Chen et al., 2012; Chang and Constantinescu, 2015). As patches narrow, the horizontal shear layer becomes more important and there is a mix of horizontal and vertical shear layers, so wake recovery is slower (Chen et al., 2013).

# CONSEQUENCES OF HYDRODYNAMIC INTERACTIONS WITH CANOPY PATCHES AND GAPS

#### Mineral and Organic Particulates

An important consequence of the hydrodynamic influences of canopy patches is their effects on sediment resuspension, transport and deposition. This involves highly non-linear interactions, since each element of the hydrodynamics-canopysediment triad influences the others. Enhanced sediment deposition creates new substrate, which provides nutrients and anchoring, encouraging enhanced canopy growth, and this positive feedback maintains spatial correlation between canopy and substrate distributions (Baattrup-Pedersen and Riis, 1999). Sediment resuspension reduces light levels, which reduce canopy growth rates, leading to sparser canopies, enabling further resuspension (Adams et al., 2016). Resuspension can occur because of enhanced turbulence or enhanced mean flow – thus sediment may be resuspended within patches even if mean flow speeds are below the threshold of sediment motion, due to stem wake turbulence (Lefebvre et al., 2010). Conversely, sediment deposition within patches may only be enhanced in the absence of stem wake turbulence (Liu and Nepf, 2016). Patches generally have two sources of sediment – from upstream and laterally. The relative contributions of each determine the spatial pattern of in-patch deposition (Zong and Nepf, 2011). Where advection from upstream dominates, net deposition initially increases as flow decelerates on entering a patch, then decreases as suspended sediment concentration decreases (Zong and Nepf, 2010), so there is a point of maximum sedimentation at some distance into a patch.

Sedimentation downstream of circular patches varies depending on the rigidity and density of the patch elements. For rigid elements, the patch throughflow shifts the patch-scale wake downstream, so there is a region of relatively stagnant flow and thus enhanced deposition immediately downstream of the patch (Chen et al., 2012). For flexible elements, the flow adjustment is more three-dimensional, and turbulence is enhanced immediately downstream of the patch so deposition is reduced there (Ortiz et al., 2013). In sparse patches of rigid elements, sediment is scoured from within the patch and deposited closer downstream than that from denser patches, because the latter divert flow more (Follett and Nepf, 2012). The spatial pattern of sediment deposition around and downstream of a circular patch of model vegetation varies primarily with the ratio of shear velocity to critical shear velocity. If this is < 0.7, there is high deposition in both the wake and adjacent zones. If it is 0.7–3, deposition is high in the wake only. If it is > 3, deposition is low everywhere. The deposition pattern correlates better with shear velocity than with settling velocity, implying that the patterns are driven by resuspension, not deposition (Shi et al., 2016). Again, there have been very few similar studies of the sedimentary consequences of canopy gap hydrodynamics, although Folkard (2011) provides some speculative inferences from a purely hydrodynamic flume study.

These effects in mineral particles are important, since they provide nutrients to canopy organisms and scale-up to affect landscape-scale geomorphology. Of greater importance ecologically, the influence on hydrodynamics of canopy patches and gaps also affects organic particles – including food, waste material, reproductive propagules and plankton. In transport, organic particles behave physically in many ways like mineral sediments. However, the timing of release of organic particles is governed by organism biology, and their deposition is governed by their varying buoyancy and morphology (Gurnell, 2007).

Therefore, the behavior of organic particles is more complex than that of mineral particles, so these complexities need to be taken into account in their modeling.

### Solutes

Because of their influence on fluxes and budgets of solutes, aquatic canopies are important in determining the biogeochemistry of water bodies (Bal et al., 2013) and canopy fragmentation can have a significant effect on this. Canopy organisms can take up nutrients from the water column or, if they are rooted, from the substrate. Canopies of plants can also provide substrate for epiphytic biofilms, which also take up nutrients (Levi et al., 2015). Spatial patterns of in-canopy flow are highly correlated with solute uptake rates, which are enhanced by up to 20% at the leading edges of canopies (Morris et al., 2008; Bal et al., 2013). The hydrodynamic effects of seagrass canopy leading edges also drives nutrient exchange between the water column and the substrate; this is caused by pressure gradients arising from flow deceleration (Adhitya et al., 2016). Canopy patches often concentrate and store dissolved nutrients (Schoelynck et al., 2012). Tussocks of wetland sedges efficiently retain biogenic silica, giving them a competitive advantage (Opdekamp et al., 2012). In the Okavango delta, aquatic macrophytes accumulate and concentrate organic matter in sediments below patches, allowing high productivity in an otherwise oligotrophic environment (Schoelynck et al., 2017). Liu et al. (2017) also found this 'soil island' effect around isolated and clustered tamarisk (Tamarix chinensis Lour.) in a coastal wetland. However, the effect size was less for clustered tamarisks than for isolated ones, implying that the effect will be weakened by vegetation restoration or natural expansion. Solute retention can vary within patches: Hemminga et al. (1998) found that growth at Spartina anglica patch edges was dependent on nutrients in the local substrate, whereas in the (older) patch centers, this material had been depleted in previous years, and growth depended on nutrients bound to allochthonous organic particles.

# Distribution of Canopy-Dwelling Organisms

Patch and gap interiors and edges can be very different environments, and strong gradients can exist in both environmental and canopy parameters across patches and gaps. This can alter ecological interactions even within single patches or gaps (Mota et al., 2015). In biological terms, differences between edges and interiors are found in the properties of the canopy-forming organism itself (e.g., Brun et al., 2003); in faunal abundances (e.g., Barbera-Cebrian et al., 2002; Bologna and Heck, 2002; Efird and Konar, 2014); in the levels of thermal (Jurgens and Gaylord, 2016) and mechanical (Folkard, 2005) stresses experienced by organisms; and in terms of sediment quality (e.g., Alves et al., 2017). Local diversity and distribution of benthic fauna is intimately associated with canopy type and distribution (e.g., Begin et al., 2004; Bouma et al., 2009). Macrophyte structural complexity plays an important role in determining differences in macroinvertebrate distribution between canopies of different species (O'Hare and Murphy, 1999). This is likely driven by differences in hydrodynamic stress attenuation and food availability rather than structural complexity per se (Bell et al., 2013). Canopies of macroalgae with greater structural complexity also promote spatial and temporal patchiness of microphytobenthos, with potential significant effects on the overall productivity of ecosystems (Umanzor et al., 2017). Fragmentation of seagrass canopies also alters their interactions with filter feeders. Within canopy patches, filter feeders' food supply is reduced, strongly restricting their growth (Reusch and Williams, 1999), but in the gaps between canopies, they can find greater protection from hydrodynamic forces and higher resource availability (Gonzalez-Ortiz et al., 2014). Thus the fragmentation allows the seagrass and filter feeders to co-exist compatibly.

# UPSCALING TO LANDSCAPE-SCALE

An important aim of studies of fragmented canopies is to be able to quantify total or average parameter values at the whole-landscape scale. Whilst these require knowledge of inputs and characteristics at that scale, they also require understanding of structure and processes at smaller scales, i.e., patch and gap-scale. In part, this is born of necessity, since field measurements are generally made at patch and gap scales for logistical reasons, so landscape-scale measurements tend to have to be derived from their upscaling. Moreover, variability at patch and gap scale within fragmented canopies often has important effects on landscape-scale structures and processes, and this has given rise to many different approaches to upscaling (e.g., Bou-Zeid et al., 2004; Chesson et al., 2005; Denny and Gaylord, 2010; Nikora, 2010).

If parameters of interest scaled linearly with spatial scale, upscaling would be trivial – it would simply comprise of adding up the contributions of each patch or gap-scale area to give a total for a whole landscape. However, this is not the case. Most, if not all, parameters of interest in aquatic canopy ecosystems scale non-linearly with spatial scale (Chesson et al., 2005). In addition, emergent forms and processes often arise at larger scales that are not apparent at the individual patch of gap scale, due to the spatial distribution of patches and gaps. For example, the spatial density and distribution of patch or gap edges plays a large part in governing landscape-scale flow structure (Dupont et al., 2011; Folkard and Bouma, 2016), and in saltmarshes, regions of relatively dense vegetation deflect flow into more sparsely-vegetated regions, where drainage channels form. Thus, landscape-scale drainage rates are determined in part by the spatial distribution of vegetation patches (Temmerman et al., 2007; Vandenbruwaene et al., 2013). These issues provide the main challenges in upscaling of patch and gap-scale phenomena to enable derivation of landscape-scale parameter values.

These challenges are addressed in two general ways. Empirically, correlations can be sought between variations in metrics describing the patch/gap-scale structure of fragmented

canopy landscapes and variations in landscape-scale total or average parameter values. While this approach can provide evidence of these cross-scale relationships, they do little to provide causal, mechanistic insights to them. The alternative approach is to create spatially-distributed models into which the smaller scale processes are explicitly incorporated but which operate over domains covering entire landscapes. These can then be explored to elicit mechanistic understanding of the cross-scale relationships. They can also be used to infer correlative relationships between patch/gap-scale causes and landscape-scale effects of the type described above that are underpinned by that mechanistic understanding (e.g., Luhar and Nepf, 2013; Larsen et al., 2017). However, such models require detailed understanding, not only of the nature of processes at the smaller scale, but also of ways in which these interact with each other as spatial complexity and scale are increased. Therefore, there is a need for studies of these interactions and the ways in which they influence landscape-scale phenomena, as well as for development of robust and broadly-applicable techniques for their upscaling.

An ecological perspective on the problem of non-linearity in upscaling can be illustrated by the example of trying to estimate the growth rate of a canopy from knowledge of the percentage of algal cover. Because increased algal cover enhances algal growth rate due to mutual protection effects at the patch scale, applying the patch-scale relationship to calculate growth rate from percentage cover at landscape-scale will not give an accurate value. To address this type of problem, scale transition theory (Chesson et al., 2005; Benedetti-Cecchi et al., 2012; Chesson, 2012) quantifies the non-linear scale-dependence of interactions between parameters in terms of the variances and co-variances of their patch/gap-scale values across whole landscapes. Larsen et al. (2017) provide an illustration of how the problems of non-linearity and spatial distribution effects in upscaling are addressed in a hydrological context in a study of the flow through the vegetated ridge-and-slough landscape of the Florida Everglades. Following approaches that have been used previously in the groundwater literature (Cushman et al., 2002; Farmer, 2002), they calculate the landscape-scale average flow resistance as a non-linear spatial average of small scale roughness, using an approach based on the ergodic hypothesis (Lumley and Panofsky, 1964). They conceptualize the landscape as binary – being made up purely of patches and gaps of 'matrix' between them. The landscape-scale average of a parameter (flow resistance in the case of Larsen et al., 2017), Hland, is then calculated from values of the same parameter for the patches, hp, and the gaps, hg, and the fractional cover of patches, p, across the whole landscape as

$$H\_{\rm land} = \|\rho h\_{\rm p}^{\alpha} + (1 - p)h\_{\rm g}^{\alpha}\|^{1/\alpha} \tag{1}$$

The non-linearity and dependence on spatial distribution are incorporated in the exponent ω, which is calculated by fitting the data produced by repeat runs of a numerical model of the landscape, based on long-term field observations (see Larsen et al., 2017, for further details). They then, via this model, explore the dependence of ω on changes in various metrics describing the heterogeneity of the landscape.

In order to be able to test and develop upscaling approaches such as scale transition theory and non-linear spatial averaging, understanding is needed of ways in which all aspects of canopy ecosystems interact as spatial complexity and scale increase. The remainder of this section identifies progress that has been made in understanding these interactions, firstly via studies of interactions between two or more patches or gaps. It then covers interactions between patch-scale and landscape-scale processes, and finally identifies some landscape-scale consequences of these interactions in fragmented canopies. Because work in this area to date is at a relatively early stage of development, the coverage is necessarily illustrative, rather than comprehensive.

## Hydrodynamically-Mediated Interactions Between Canopy Patches

Taking a bottom–up approach, the first stage in understanding how processes at patch/gap-scale scale up is consideration of the interactions between two patches or gaps. In an aquatic context, hydrodynamics is usually the dominant mediator in these interactions. For example, where a downstream patch is located in the hydrodynamic wake of an upstream patch, the wake's enhanced turbulence will alter the conditions in the downstream patch (Folkard, 2005). This can affect its nutrient uptake rate, due to changes in both the mean flow speed and the levels of turbulence (Cornacchia, 2018). Other interactions will involve hydrodynamically-mediated sediment processes. For example, when the two side-by-side patches are far apart, their wake interactions are weak, and each has its own region of sediment deposition behind it. If the transverse distance between them reduces, their wakes will start to interact and a depositional region will form further downstream where their wakes merge (Meire et al., 2014). This encourages formation of a new vegetation patch, which will slow the flow between the patches and allow them to merge (de Lima et al., 2015; Liu et al., 2018). Over time, this process may lead to continuous vegetation coverage (Kondziolka and Nepf, 2014). Because of the differences in their interactions with flow, these morphological feedbacks will be different for rigid and flexible vegetation (Ortiz et al., 2013). This leads to a different set of outcomes when patches of different species interact. For example, Cornacchia et al. (2019) found that when a patch of a vegetation species with a taller, denser canopy (Callitriche) was located upstream of a patch of a shorter, sparser species (Groenlandia), it generated a turbulent wake that enhanced nutrient uptake by the Groenlandia. At the same time, the uptake rate of the Callitriche benefited from being exposed to the higher mean velocity of the upstream flow, as its canopy was too dense for turbulence to penetrate.

# Influence of Canopy Patch Interactions at Landscape-Scale

Interactions between canopy patches, hydrodynamics and sediment processes of the type described in the previous section can lead to landscape-scale structure in the spatial distribution of canopies. For example, in saltmarshes, regions of relatively dense vegetation can deflect flow into more sparsely-vegetated

regions, leading to preferential formation of drainage channels in the latter. As a result, the saltmarshes evolve with some regions characterized by dense vegetation, and others by drainage creek networks (Temmerman et al., 2007; Vandenbruwaene et al., 2013). Other examples of feedbacks between vegetation canopies, flow and sediment processes governing the evolution of landscapes have been found by Larsen et al. (2007) and Larsen and Harvey (2010, 2011) in studies of the ridge-and-slough patterning of vegetation distributions in the Florida Everglades.

In forming these spatial structures, ecological traits of the canopy species are often important. For example, rate of growth is an important determinant of canopy patches' contributions to landscape dynamics (Bertoldi et al., 2011). Where patches grow fast, they are more resilient due to their ability to recover from disturbance more quickly. Slower growing species also tend to decline more slowly, so are resistant to degradation (O'Brien et al., 2018). These differences will lead to differences in the evolution of canopy-hydrodynamics-sediment interactions and thus differences in spatial structure. Variations in establishment strategies amongst canopy-forming species (e.g., clonal extension, ruderal gap filling), variable spatial and temporal patterns of disturbance (van Hulzen et al., 2006), and the extent to which they change their physical structure and biomechanical properties over their growth-senescence cycles (Kleeberg et al., 2010) will also significantly modify their landscape-forming function. In general, canopies species' role in landscape dynamics is to act as ecosystem engineers (Jones et al., 1994) – they trap and stabilize sediments, organic matter and propagules of other species, modify local sediment and morphology, and drive development of landforms and habitats (Gurnell, 2007; Gurnell et al., 2012). Aboveground biomass modifies flow and retains sediment, while below ground biomass affects the hydraulics and mechanical properties of the substrate. Thus, their effects change as above and below ground biomass fractions change in response to climatic and hydrodynamic forcing (Gurnell, 2014).

The upscaled consequences of patch-scale interactions will also often interact with larger scale processes. Generally, the large-scale processes determine the overall extent of the fragmented canopy and can shape and orient the patches and gaps in the landscape, while small-scale interactions generate the patch/gap-scale structure (van de Koppel et al., 2012). For example, Fonseca et al. (2008) found that the spatial organization of Halophila decipiens (Caribbean seagrass) in an open ocean setting subject to hurricane damage was dictated first by large scale dispersal of propagules (over 100s of meters) then, within a growing season, by clonal organization of individual seagrass patches. The large-scale controls can include anthropogenic disturbance: in the Wadden Sea in NW Europe, the landscape-scale consequences of increased human disturbance of sediment (e.g., dredging of navigation channels and ports) has interfered with biological controls of sediment dynamics and have shifted the inter-tidal zone from a state of internal regulation (by the ecosystems within the zone) and spatial heterogeneity to external regulation (by anthropogenic impacts originating outside the zone) and spatially homogeneity (Eriksson et al., 2010).

# Consequences of Patch-Scale Interactions at Landscape-Scale

Understanding of the landscape-scale consequences of interactions between canopy patches, hydrodynamics and hydrodynamically-mediated sediment processes is mainly focused on those particular elements of the ecosystem (i.e., the patches, hydrodynamics and sediment themselves). However, there have also been some studies of their influences on some of other physical, chemical and biological aspects of fragmented canopy ecosystems. For example, they have an important influence on solute diffusion coefficients and residence times (Nepf et al., 1997; Nepf, 1999). These can vary by an order of magnitude across fragmented canopies because of the great difference in flow speed between canopy throughflow, and flow over and around patches (Lightbody et al., 2008). This can affect the canopies themselves, for example by varying their exposure to pollution or their access to dissolved nutrients and gasses (Lara et al., 2012).

Another major landscape-scale ecological consequences of interactions between hydrodynamics and fragmented canopies is their effect on habitat diversity. However, the nature and direction of these effects (i.e., whether they increase or decrease habitat diversity, or leave it unchanged whilst changing the mox of habitats) remains unclear. The heterogeneous flow conditions created by fragmented canopies create a highly diverse mosaic of habitats (Sukhodolov and Sukhodolova, 2010; Verschoren, 2017). According to the long-established patch dynamics concept, high levels of spatial habitat variability imply high levels of species richness (Townsend, 1989). However, this is not always well-supported by data (Resh et al., 1994). In a meta-analysis of seagrass research, neither literature review nor field measurements suggested that habitat fragmentation has any consistent effect on fauna, and there was little evidence of fragmentation sensitivity in any taxonomic group (Bell et al., 2001). Lefcheck et al. (2016) found that abundance, species richness, Simpson and functional diversity and composition of faunal communities were invariant to fragmentation in experimental eelgrass landscapes. They concluded that this is likely a consequence of the fauna's rapid life histories and high mobility. In other studies, however, such relationships have been found: Matias et al. (2015) found that higher habitat complexity in fragmented macro-algae canopies promoted species colonization, so the higher the level of fragmentation, the more species were present. Thus, the relationship between hydrodynamic interactions with fragmented canopies and species richness of communities inhabiting those canopies requires further investigation.

# APPLICATIONS TO MARINE CANOPY LANDSCAPE MANAGEMENT AND RESTORATION

As noted by Bell et al. (1997), there is a powerful mutualistic relationship between the practice of landscape restoration and the science of landscape ecology. Restoration can provide

experimental spatial distributions and opportunities for experiments over large spatial scales. Landscape ecology can provide insights into selecting reference sites and establishing restoration project goals, and appropriate spatial configurations to aim for. The same kind of mutualistic relationship is also evident between the practice of river restoration and the science of eco-hydromorphology and eco-hydraulics, which focus on interactions between ecology, catchment hydrology, river channel hydraulics and channel and floodplain geomorphology (e.g., Vaughan et al., 2009). Similar mutually beneficial links exist between the practices of coastal zone restoration, protection and management, and the science of biophysical (and chemical) interactions in coastal marine ecosystems – many of which have canopy organisms as keystone species. Some examples of ways in which understanding of canopy spatial distribution and fragmentation can be utilized in coastal zone management are given below.

Analysis of canopies' spatial distributions can be used effectively to describe the impacts of multiple human stressors in marine environments (Tamburello et al., 2012). van der Heide et al. (2010) found consistent responses of spatially self-organized patterns in seagrass meadows to changing abiotic conditions, and suggested that this could lead to the use of self-organized spatial patterns as stress indicators in these meadows. Even in inter-tidal diatom micro-canopies, spatial patterns can provide important clues about level of degradation of ecosystem (Weerman et al., 2012). However, interpretation of these patterns requires detailed knowledge of the nature of underlying feedbacks, including hydrodynamically-mediated feedbacks, as the patterns differ markedly between ecosystems. Of potentially greater value, understanding of the spatial distribution of canopy fragmentation can be used to predict the development of canopies and identify those that are at risk of catastrophic decline (Rietkerk et al., 2004). For example, the shape of Spartina anglica patches has been found to indicate the long-term development of salt-marsh pioneer zones, although the outcome is conditional on large scale morphodynamics and sediment grain size (Balke et al., 2012). Fraschetti et al. (2012) suggested that increasing spatial heterogeneity of both intertidal and subtidal assemblages probably represents an early warning of increasing human pressure in marine protected areas. In a model of seagrass meadow spatial patterning, Ruiz-Reynés et al. (2017) found that a transition to patches of vegetation arranged in approximately hexagonal formations indicates that the meadow is close to a tipping point where further increase in mortality may lead to catastrophic loss of the meadow.

Fragmentation of canopies at landscape-scales may also be used as a bio-indicator of loss of abundance amongst canopy-using organisms. For example, properties of fragmented landscapes at 10–100 m scales have been found to be effective indicators of nekton distributions, with lower nekton abundances correlating with higher degrees of canopy fragmentation and loss of habitat connectivity (Baillie et al., 2015; Favre-Bac et al., 2017).

Better understanding of the hydrodynamics of canopies at patch/gap and landscape-scales may also help attempts to re-establish or restore marine canopies. Attempts to re-seed and re-turf seagrass canopies have been made in marine environments but have had limited success (van Katwijk and Hermus, 2000; van Katwijk et al., 2009). Natural re-establishment of macrophyte patches has been somewhat more successful in streams (Larned et al., 2006). In these environments, the main bottleneck for re-colonization is the initial establishment of attached roots in the sediment from propagules or seedlings (Riis, 2008), therefore understanding of the hydrodynamic conditions that facilitate this process for different canopy types in different contexts would be valuable. Once they have established, patches of plant canopy are able to create interactions with the flow, leading to positive feedback that causes enhanced sediment deposition and allows the patches to expand (Sand-Jensen, 1998).

# COMPARISON ACROSS CANOPY TYPES AND PROPOSED DIRECTIONS FOR RESEARCH

Having provided a structured synthesis of work on biophysical interactions with a wide range of canopy types in the preceding sections, this final section compares the approaches taken and progress made in work focused on different canopy types, and draws out what can be learnt about each one, and how we can develop more universal understanding of these interactions, their consequences and how they can be harnessed for management purposes.

In terms of patches and gaps themselves there appears to be a lack of basic data across all canopy types regarding how commonly they occur; their size, shape and orientation distributions; and how long they typically persist. This sort of information is important, as it allows models predicting their evolution and consequences to be developed on the basis of realistic data. Comparative studies of the modes of formation, maintenance and destruction of patches and gaps across different canopy types might also help to elucidate the relative importance of different factors (environmental gradients, biotic and abiotic stresses and facilitations, catastrophic events) for each one.

The study of fundamental hydrodynamic interactions with canopies has largely been carried out via laboratory flume or basin studies, numerical modeling and field experiments. These have typically used more-or-less idealized hydrodynamics (uniform flows or wave fields) and canopies in simplified twoor three-dimensional configurations. Often, the elements of the canopies have been idealized using simulants, which are uniformly rigid (e.g., wooden dowels) or flexible (e.g., plastic strips). These simulations have tended to be based on the essential biomechanical and morphological properties of seagrasses, saltmarsh vegetation, and mangroves. From this, a relatively detailed and thorough understanding of the hydrodynamics of these types of canopies has been built up. The review carried out suggests that less work has been done on the fundamental hydrodynamics of lower growing organisms – biofilms, turf algae etc. Moreover, the strongly reductionist, idealizing approach taken in this work to date suggests that moves toward greater realism in these experiments is needed. For example, this

would include studies of the hydrodynamics of patches or gaps with boundaries that are not quasi-discontinuous, which vary in height and density, and which have elements with variable morphology. Studies incorporating a wider range of configurations of patches and gaps – for example cases in which patch-gap edges are aligned at intermediate angles to the direction of the oncoming current or waves (rather than being parallel or perpendicular to it, as has almost universally been the case hitherto), or where edges are not either straight or circular – and less uniform hydrodynamic conditions might also elucidate non-linear interactions between variations in canopy and hydrodynamic characteristics.

Studies of the effects of hydrodynamic interactions with canopy patches and gaps on particulates, solutes and canopydwelling organisms appear from this review to have focused mainly on canopies of seagrasses, saltmarsh vegetation, macroalgae, and mangroves. This suggests that there is a need for further work in this area on low-growing canopies, where the focus hitherto appears to have been more biological (e.g., on patterns of grazing) than hydrodynamic. In general, as with studies of the hydrodynamic interactions with canopies, the physical aspects of this topic appear to have been studied largely through idealized configurations, whereas the biological aspects have been mainly studied through field measurements of in situ ecosystems. Moving the former toward more complex, realistic settings, and the latter toward more controlled focused conditions will help to bring understanding of biophysical interactions in these contexts from biological and physical perspectives closer together.

Arguably the most consequential motivation for studying these interactions is a desire to be able to predict how marine canopy landscapes will be affected by our actions, and how they can help us via their ecosystem services. This ability would enable us to guide our actions and harnessing of those services. This implies that landscape-scale is the scale at which the insights delivered by research may be applied most usefully. This is at odds with the fact that the most common scale for measurement and modeling – due to logistical and technological limitations – is the patch/gap scale. Therefore, upscaling from patch/gap scale to landscape-scale is arguably the most important current problem in fragmented marine canopy research. Although significant progress has been made in this area in recent years there still remains much to be done. Further studies are needed into the mechanics of interactions between multiple canopy patches and gaps at all scales – from interactions between two patches, through studies of patch mosaics (Schoelynck et al., 2018) and fragmented canopies with more complex spatial distributions, to whole-landscape scales. These need to take into account the roles of a wide range of different variables, including those related to hydrodynamics (waves, currents, turbulent mixing), sediment (erosion, resuspension, transport, deposition), and other physical variables such as light levels (e.g., Koch, 2001; Adams et al., 2016) and water temperature. From a chemical perspective, they need to include concentrations of nutrients, dissolved gasses, pollutants and a wide range of biogenic chemicals, as well as their flux rates, both in terms of physical movement between the substrate, water column, biota and atmosphere, and in terms of chemical changes, for example from dissolved to particulate form, or organic to inorganic form. From a biological perspective, they need to include rates of primary production, bulk biomass, species diversity and richness, and metrics of ecosystem structure, functioning and services. Clearly, no single study or model could incorporate all of these variables. They are listed here to emphasize the importance of considering the full range of factors that may be at play in determining the dynamics of fragmented marine canopies.

In summary, a number of general ways of progressing the science driving our ability to manage ecosystems and landscapes characterized by fragmented marine canopies approaches can be identified. Firstly, closer collaboration is required between researchers carrying out work aimed at improving our understanding of the fundamental processes of biophysical interactions with fragmented canopies, practitioners of landscape management and restoration, and policymakers concerned with coastal environments. Within the research community, traditionally reductionist, laboratory and numerical model-based hydrodynamics research would benefit from a move toward studying more non-uniform, varied and realistic configurations, and traditionally holistic, field-based ecological research would benefit from a move toward studying more controlled, idealized and quantitatively-modelable configurations. Moreover, a greater appreciation of the importance of chemical aspects of the systems studied needs to be incorporated into the current biophysical approach. Further development of techniques for upscaling understanding and predictions of bio-chemo-physical interactions at the patch/gap scale to the landscape-scale in the context of spatially complex canopies is required. All of these would benefit greatly from a globally distributed experiment approach (Borer et al., 2014) with a clear shared direction and aims. Finally, in attempting to interweave the fields of hydrodynamics, marine canopy ecology and spatial analysis of landscapes, whilst incorporating biogeochemistry and socio-environmental interactions, an approach that balances these disciplines, rather than viewing one as subordinately serving the other, would be the best way forward.

# AUTHOR CONTRIBUTIONS

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

# ACKNOWLEDGMENTS

I am grateful to MG and the editorial team for this special issue for the opportunity to contribute this review article. I am also grateful to the two reviewers whose comments and suggestions helped greatly with the completion of this manuscript.

### REFERENCES



M. R. Moss (Cambridge: Cambridge University Press), 112–119. doi: 10.1017/ cbo9780511614415.013




**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 © 2019 Folkard. 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.

# Effect of Seagrass on Current Speed: Importance of Flexibility vs. Shoot Density

#### Mark S. Fonseca<sup>1</sup> \*, James W. Fourqurean<sup>2</sup> and M. A. R. Koehl<sup>3</sup>

<sup>1</sup> CSA Ocean Sciences Inc., Stuart, FL, United States, <sup>2</sup> Department of Biological Sciences, Center for Coastal Oceans Research, Florida International University, Miami, FL, United States, <sup>3</sup> Department of Integrative Biology, University of California, Berkeley, Berkeley, CA, United States

#### Edited by:

Virginia B. Pasour, Army Research Office, United States

#### Reviewed by:

Lucy Gwen Gillis, Leibniz Centre for Tropical Marine Research (LG), Germany Jennifer Joan Verduin, Murdoch University, Australia Free Espinosa Torre, University of Seville, Spain

> \*Correspondence: Mark S. Fonseca mfonseca@conshelf.com

#### Specialty section:

This article was submitted to Marine Ecosystem Ecology, a section of the journal Frontiers in Marine Science

Received: 06 September 2018 Accepted: 17 June 2019 Published: 09 July 2019

#### Citation:

Fonseca MS, Fourqurean JW and Koehl MAR (2019) Effect of Seagrass on Current Speed: Importance of Flexibility vs. Shoot Density. Front. Mar. Sci. 6:376. doi: 10.3389/fmars.2019.00376 Water flow through seagrass beds transports nutrients, affects sediment stability and chemistry, and imposes hydrodynamic forces on shoots that alter canopy configuration. Past studies done under diverse conditions yielded conflicting results about the effects of shoot density on flow through seagrass bed canopies. We used eelgrass, Zostera marina, to study how the density of flexible shoots affect the hydrodynamics of seagrass beds in unidirectional water flow. By exposing randomly-arranged shoots of uniform length to current velocities controlled in a flume, the effects of shoot density and distance downstream from the bed edge could be determined without confounding factors. Comparison of velocity profiles within beds to those upstream of beds showed that flow was slower in the beds. However, shoot density, downstream distance, and current velocity did not affect the percent reduction in flow velocity in a bed. Turbulence enhances mixing of substances carried in the water. Here, turbulence intensity (index of the importance of turbulent velocity fluctuations relative to average current velocity) was lower when ambient flow was faster, but was not affected by shoot density or downstream position, Drag (hydrodynamic force on a shoot that bends it over in the flow direction) provides another measure of how the canopy affects flow experienced by a shoot. Drag was not affected by current velocity, shoot density, or downstream position in the bed. Gaps between shoots can enhance light and flow penetration into the canopy, but when shoots are bent over by flow, they can cover gaps. Faster ambient currents caused greater gap closure, which at each current speed was greater for high shoot densities. Thus, canopy gap closure did not correlate with percent flow reduction in grass beds or with drag on individual shoots, both of which were independent of shoot density and ambient current velocity. Since changing shoot density does not affect the flow in a grass bed exposed to a given ambient current, our results are inconsistent with the hypothesis that the high shoot densities observed in grass beds in habitats exposed to rapid flow are due to a direct, adaptive response of the grass to the flow environment.

Keywords: seagrass, current speed, shoot density, flexibility, canopy, gaps

# INTRODUCTION

fmars-06-00376 July 5, 2019 Time: 15:16 # 2

Studies of water flow through seagrass beds have been carried out under different conditions (e.g., unidirectional currents vs. waves, field sites vs. laboratory flumes, various species vs. physical mimics or mathematical models), and thus have yielded conflicting results about the effects of shoot density on flow within and above seagrass beds. Even though focusing on the numerous studies that have examined the behavior of unidirectional water flow in and around seagrass canopies (e.g., Fonseca et al., 1982, 1983, 2007; Fonseca and Fisher, 1986; Fonseca and Kenworthy, 1987; Gambi et al., 1990; Ackerman and Okubo, 1993; Worcester, 1995; Koch and Gust, 1999; Verduin and Backhaus, 2000; Abdelrhman, 2003; Newell and Koch, 2004; Peterson et al., 2004; Fonseca and Koehl, 2006; Lacy and Wyllie-Echeverria, 2011; Lei and Nepf, 2016) consistently reveals that seagrass beds reduce flow velocity within the canopy, the effect of (seagrass<sup>1</sup> ) shoot density on this velocity reduction is not clear.

# Effects of Shoot Density on Water Flow in Seagrass Beds

The discrepancies between various empirical studies of shoot density effects on flow may be due to differences between studies in the range of densities compared, the flow conditions, the position within a seagrass bed that the flow was measured, and the morphology of the seagrass. For example, Lacy and Wyllie-Echeverria (2011) found for eelgrass (Zostera marina) that flow speed was attenuated more by beds with higher shoot densities. However, their shoot densities were low (44–63 shoots m−<sup>2</sup> ) and the plants they studied were from water depths of 3– 4 m and thus had longer leaves (∼1.5 m leaf length) than do Z. marina from shallower sites (e.g., plants used in this study taken from seagrass beds of <2 m water depth; Fonseca et al., 2002). Likewise, Worcester (1995) considered the lower end of the shoot density range in an open, natural system and under very low ambient current speeds. She found that in a natural setting with current speeds of only 0.05 m s−<sup>1</sup> and shoot density (with very large plants) of <200 shoots m−<sup>2</sup> , found there was still significant reduction in currents within as opposed to above the canopy, whereas the turbulence of the flow was unaffected. In contrast Adhitya et al. (2014) studied seagrass beds with higher shoot densities (400 and 1100 shoots m−<sup>2</sup> ) and found that shoot density affected the flow dynamics at a very low ambient current speed (0.10 m s−<sup>1</sup> ), which is so slow that it does not fully deflect a seagrass canopy (Fonseca and Kenworthy, 1987). However, Adhitya et al. (2014) also found the spatial organization of seagrass shoots at the meadow scale was more important than shoot density within a patch in explaining hydrodynamic effects of seagrass beds. Paul and Gillis (2015) also found an effect of shoot density on flow reduction by beds of Z. noltii. They visually compared velocity profiles and concluded that canopy bending, and waving did not affect the flow-reduction capacity of their grass beds. In contrast, Fonseca et al. (1982, 2007), Fonseca and Fisher (1986), and Gambi et al. (1990) found little influence of shoot density on the degree of current speed reduction within beds of Z. marina. Interestingly, while Gambi et al. (1990) found no statistically significant difference in shear velocity or turbulence intensity in seagrass beds as a function of shoot density, they concluded "More replicates in each position probably would reveal statistically significant differences between densities." Their assumption that shoot density affects flow was compelling, and their paper has been interpreted by others as having demonstrated that shoot density affects flow (e.g., Koch et al., 2006).

Various mathematical models have explored whether shoot density affects water flow within grass beds. Dijkstra and Uittenbogaard (2010) simulated the effect of shoot density and flexibility and found that turbulence intensity was decreased as shoot density increased (up to 10,000 shoots per m−<sup>2</sup> ), and that the drag force on stiff shoots was greater than on more flexible shoots. Their model suggests that both high shoot density and increased flexibility could be mechanisms that enable seagrass to withstand rapid ambient currents. Abdelrhman (2003) used shoot density as a factor influencing in-canopy flow in a model and found that high shoot density reduced transport through the canopy by increasing the tortuosity of the paths of water moving around seagrass shoots. However, Abdelrhman's model did not clarify the hydrodynamic conditions under which these different shoot densities may occur in the field. The model of Newell and Koch (2004) assumed critical shoot-density thresholds for wave attenuation by seagrass beds, and the model of Koch et al. (2006) also assumed that higher shoot density produced lower in-bed flow velocities. Thus, shoot density effects have become codified in models and perception, in spite of the conflicting empirical results. One way to resolve this issue would be to measure the hydrodynamic effects of shoot density at known positions in beds of actual plants of one species (rather than physical mimics) when they are exposed to the same unidirectional flow conditions, and to compare the effects of shoot densities using conventional statistics that directly consider the variability of each effect.

# Effects of Seagrass Flexibility on Canopy Structure and Water Flow

Seagrass shoots, particularly those of strap-leaved species, are extremely flexible (Fonseca and Koehl, 2006; Fonseca et al., 2007), as anyone who has handled these plants can attest. Flexible seagrasses are bent over by the drag force imposed by flowing water (Bouma et al., 2005), while the buoyancy of grass blades resists their downward bending (Luhar and Nepf, 2011), as has also been demonstrated for macroalgae (e.g., Stewart, 2004). The magnitude of drag on a sessile organism depends on the projected area of the organism normal to the flow direction, thus bending over in flowing water is a mechanism of reducing drag (e.g., Charters et al., 1969; Koehl, 1976, 1977, 1986; Denny et al., 1985; Denny, 1988; Koehl and Alberte, 1988; Carrington, 1990; Abdelrhman, 2003). Increasing the ambient flow velocity encountered by a sessile organism causes higher drag. Therefore, the drag on a shoot within a grass bed is a measure of how the other plants in the canopy affect the flow experienced by that shoot.

<sup>1</sup>Hereafter "shoot density" is in reference to seagrass unless otherwise stated.

When a group of flexible seagrass shoots bend in response to flowing water, they can form a compact, interwoven structure that intensifies the reduction of water velocity within the compressed canopy and simultaneously redirects the flow over the canopy (Fonseca et al., 1982; Gambi et al., 1990), thereby shifting high shear stresses from the sediment surface to the top of the canopy (Fonseca et al., 1982). Such a reduction in flow-induced shear on the substratum should enhance both rhizosphere stability and carbon accumulation on the sediment (sensu Kenworthy et al., 1982).

Flexible and compressible grass canopies pose a challenge to understanding the effects of shoot density on flow. The relationship between shoot density and water-speed reduction in canopies composed of stiff macrophytes (e.g., marsh grass communities composed of upright shoots with relatively high flexural stiffness) has been determined via experiments using arrays of rigid rods in a flume (e.g., Nepf, 1999). In contrast, the degree of bending and canopy compression of flexible plants like seagrasses can depend on ambient current velocity and presumably on shoot density as well. Therefore, the effects of shoot density on flow through flexible canopies should be determined for the same sets of grass arrays at a range of different ambient flow speeds.

Fonseca et al. (2007) suggested that the arrangement of shoots in a seagrass bed might affect both water flow and light transmission into the bed. Seagrass shoots sometimes are arranged in rows (Fonseca et al., 2007) that can create anisotropic gaps in the canopy, and random arrangements of shoots can also create gaps in grass beds. Canopy gaps in terrestrial systems have long been recognized as local regions of higher light intensity and as initiation points of disturbance propagation (e.g., Sprugel, 1976; Veblen, 1985; Iwasa et al., 1991), and similar effects have been postulated for seagrasses (Fonseca et al., 1983; Valentine et al., 1994). Folkard (2011) studied how gaps within an array of flexible vegetation introduce flow into the canopy and described the likely role of gaps on various depositional process. The effects of shoot density on the formation or closure of gaps in seagrass canopies as these flexible plants are bent over by ambient water flow should be determined for different current speeds, and the effects of these gaps on flow velocities and turbulence should be measured. This would provide useful information to assess various mechanisms that have been proposed for how shoot density in a seagrass bed might be changed in response to ambient water flow.

# Does Ambient Current Speed Affect Shoot Density in Seagrass Beds?

It has long been suggested that faster ambient water flow might lead to an increase in shoot density in seagrass beds (e.g., Conover, 1964; Short, 1975; Peterson et al., 2004). Subsequent work showing that shoot density of Z. marina was positively correlated with current speed (Kenworthy et al., 1982) was consistent with this prediction. However, a number of other factors are known to affect shoot density. For example, in light-limited environments, shoot density varies positively with light intensity (Short et al., 1995; Ruiz and Romero, 2001; Krause-Jensen et al., 2003). Furthermore, increasing sediment organic content or nutrient availability can lead to increases in shoot density (e.g., Koch, 2001; Holmer et al., 2008; Howard et al., 2016; Ceccherelli et al., 2018). These factors are in turn affected by the deformation of shoots by hydrodynamic drag, the formation or closure of gaps in the canopy in flowing water, and the transport of dissolved nutrients and sediment by water moving around and through a grass canopy. Therefore, determining whether current speed drives a shoot density response in seagrasses is complicated by the dynamic canopy architecture of most seagrasses in response to water motion.

# Objectives of This Study

The goal of this study was to determine whether shoot density in a seagrass bed exposed to unidirectional ambient water flow affects the hydrodynamics and canopy architecture of the bed. We avoided the contradictory results of earlier studies conducted under diverse conditions by focusing on one species of seagrass (Z. marina) with very flexible shoots exposed to unidirectional currents of different velocities that we controlled in a laboratory flume. For all shoot density and flow treatments, we held shoot size constant and measured water flow, forces on shoots, and gaps in the canopy at the same defined positions in the grass bed. The specific questions that we addressed using this system were:


Answers to these questions can be used to address the issue of whether differences in seagrass shoot density observed in the field represent a direct adaptive response to ambient flow regime. For example, if seagrass shoots in a bed of a given density experience the same hydrodynamic forces and gap structure irrespective of downstream distance in the bed, and if those forces and gaps are not affected by shoot density, then the speed of the ambient unidirectional current flow should not be considered a direct driver of shoot density. Consequently, other factors (e.g., nutrients, light) could alone explain the large range in seagrass shoot densities often seen for Z. marina over small geographic distances.

# MATERIALS AND METHODS

fmars-06-00376 July 5, 2019 Time: 15:16 # 4

#### Seagrass Used in the Flume

Healthy (i.e., intact, minimally epiphytized) shoots of Zostera marina were collected near Beaufort, NC, in July from two sites described by Fonseca and Bell (1998): Middle Marsh [site mmnl1], and southern Core Sound [site hih2]). Shoots were kept in flowing seawater in tanks (40 l) at ambient temperature (∼24◦C) and salinity (∼34) under flood lamps (Philips 250 W model: K250PARFL) that exposed them to ∼250 µE m−<sup>2</sup> s −1 for 24 h per day to stimulate photosynthesis and stabilize flexural stiffness.

A seagrass shoot attached to the substratum and exposed to unidirectional water flow parallel to the substratum is bent by the hydrodynamic drag like a cantilevered beam is bent by an applied force. The flexural stiffness (EI) of a beam, its resistance to being bent by a force, depends on the stiffness (E, elastic modulus) of the material from which it is made, and on the distribution of material around its axis of bending (I, second moment of area of a beam cross-section) (e.g., Wainwright et al., 1976). Because one focus of our study was the role that seagrass flexibility plays in affecting gaps in seagrass canopies and thus flow through the canopies, we measured the flexural stiffness of grass from both field collection sites. Using methods described in Fonseca and Koehl (2006), the EI of both leaf and sheath samples from randomly selected seagrass shoots from each site were measured by being bent like a cantilevered beam by a point load, where:

$$EI = (F \times L^3) / 8\\$ \tag{1}$$

where, F is the force (N) applied at a point along the cantilever at distance L (m) from the attached end of the cantilever, and δ is the linear deflection distance (m) in the direction of force application of the point on the beam where the force was applied. Deflection was always δ < 0.10 L so that small-deflection beam equation (Equation 1) would apply. Differences among the plants from the two field sites were tested by one-way analysis of variance (ANOVA) after natural log +1 transformation to determine if there were differences in flexural stiffness among plant sources that could influence subsequent trials.

#### Flume and Experimental Seagrass Beds

The seawater flume (8 m long × 1 m wide × 0.75 m deep) used for this study is described in Fonseca and Koehl (2006) and Fonseca et al. (2007). Consistent with past studies in this flume, the constructed seagrass bed was 0.25 m wide by 1.0 m long, and the water depth was 0.3 m. Shoots were held in place in holes in a clear acrylic plate (0.25 × 1.0 m) that was fit into a recessed portion of flume floor to be flush with the main floor of the flume. Random shoot arrangements were used to avoid the flow artifacts measured in beds of shoots arranged in unnatural, evenly-spaced rows (see Fonseca et al., 2007). Random arrangements of shoots were created in the plate by overlaying it with a clear plastic grid (25 × 100 cm) on which each point where lines intersected was numbered (1–2500). A random-number generator (produced by a HP-11c calculator) was used to select the points where shoots should be placed for the highest shoot density desired (1246 shoots m−<sup>2</sup> ). Holes (6.25 mm diameter) were drilled through the plate at these points. The shoot of a Z. marina was wedged through each hole so that the upper surface of the plate intersected the shoot at the same position as the sediment surface intersected the shoot in the field. The shoots were held in natural upright positions and the orientation of the shoots with respect to the direction of water flow in the flume was arbitrary. Shoots were kept moist when in air and were not exposed to air for more than 3 min between the time collected and when installed in the flume.

Seagrass-flow interactions were tested for seven different shoot densities (1246, 1046, 846, 646, 448, and 246 shoots m−<sup>2</sup> ) that were chosen to approximate the normal range of densities of monotypic stands of Z. marina in the Beaufort, NC, area (natural observed range of 185–1133 shoots m−<sup>2</sup> ; Fonseca and Bell, 1998). As described above, shoots were randomly arranged at the highest shoot density. After the flow trials were run at the highest density, some shoots were selected using the random-number generator to be removed to achieve the next lower density, and so on.

# Drag Measurements

The hydrodynamic force pushing a shoot downstream (drag) was measured for shoots in grass beds of different densities. For each shoot density tested, a seagrass shoot was arbitrarily selected from those collected in the field and this single shoot was attached to a force transducer as described in Fonseca and Koehl (2006) and Fonseca et al. (2007). The force transducer and attached shoot were placed in the grass bed at the flume midline, first at the position 0.50 m downstream from the leading edge of the grass bed, and then at the position 1.00 m downstream from the leading edge. At each position, drag force was measured to the nearest 0.01 N at a sampling rate of 5 Hz for a period of 5 min, and the mean force was calculated. Such drag measurements were repeated 9 times at each position for each shoot density and freestream current velocity.

### Flow Measurements

Three freestream water speeds representative of currents measured just upstream of beds of Z. marina at sites near Beaufort, NC (Fonseca and Bell, 1998), were tested in the flume: 21, 32, and 63 cm s−<sup>1</sup> . Following protocols developed in Fonseca and Koehl (2006) and Fonseca et al. (2007), a Marsh-McBirney Model 523 electromagnetic current flow meter was used to record horizontal water velocities at the midline of the flume at heights of 2, 5, 8, 11, 14, 17, 20, and 23 cm above the substratum. Water velocities were recorded for 30 s at 5 Hz at each height. Examples of the velocity profiles measured in this way are shown in **Figure 1**. Velocity profiles were measured at three sampling locations along the midline of the flume: over bare substratum at a position 25 cm upstream of the leading edge of the constructed grass bed, and in the grass bed at distances of 50 and 100 cm downstream from the leading-edge.

The height of the seagrass canopy above the substratum was measured through a glass panel in the side of the flume. Canopy heights were measured at both the 50 and 100 cm positions downstream of the leading edge of the grass bed for every shoot

density and water speed tested. A ruler was set against the glass panel and the time averaged height of the deflected canopy visually estimated to the nearest cm. The canopy heights were used to determine which positions above the substratum in each velocity profile were within or above the canopy. In every case, the heights at which flow was measured that were within the canopy were 2, 5, 8, and in some lower velocities, 11–13 cm above the substratum, while all the other heights at which flow was measured were above the canopy.

For each shoot density and freestream current speed tested, the reduction in flow velocity within the grass bed canopy was determined at the positions 50 and 100 cm downstream from the leading edge of the bed. The mean flow velocity at each height within the canopy was determined for each treatment, as was the mean velocity at the same height at the position 25 cm upstream from the canopy. The percent reduction in flow within the canopy was then calculated for each height at each downstream position as described by Fonseca and Koehl (2006) and Fonseca et al. (2007). The mean % reduction in velocity for all the heights within the canopy was then determined for each downstream position (50 and 100 cm) in the canopy.

Turbulent eddies stir the water, thereby enhancing mixing of substances carried by the water and reducing their local depletion or accumulation in the water around organisms. Turbulence intensity is a dimensionless number that is a measure of the importance of velocity fluctuations (due to turbulence) relative

reduction of canopy height at higher ambient current velocities (compare B,D,F).

to the average current velocity. Arrays of plants can decrease turbulence intensity of ambient flow by damping out fluid motion, or can increase turbulence intensity by shedding vortices or fluttering as the water flows past them. Comparison of the turbulence intensity of the ambient water flow outside of arrays of macrophytes with that of the flow inside the canopies has long been used to assess the effects of those canopies on the level of turbulence in the water (e.g., Anderson and Charters, 1982; Koehl and Alberte, 1988; Gambi et al., 1990; Worcester, 1995). So that we could compare our results to those of such earlier studies, we used our measurements of flow velocity as a function of time at each height (h) within the canopy to calculate the turbulence intensity at that height (Ih), as defined in Gambi et al. (1990):

$$I\_h = (rmsU\_h/\mathfrak{u}\_h) \times 100\tag{2}$$

where rmsU<sup>h</sup> is the root mean square of the velocity at height h (in our study velocities were measured at 0.02 s intervals for a duration of 30 s, and rmsU<sup>h</sup> was computed using the RMSE function in Sas Institute Inc, 2002), and û<sup>h</sup> is the mean of the velocities measured over that 30 s period at height h. We calculated the I<sup>h</sup> 's for each of the heights at the positions in a grass bed that were 0.50 and 1.00 m downstream from the leading edge of the bed.

#### Canopy Deflection and Gap Area

For each experimental grass bed tested, the lengths of 10 arbitrarily selected shoots were measured to the nearest millimeter to determine mean grass length. For each current


TABLE 2 | Drag force on shoots as a function of shoot density, distance downstream into the canopy, and freestream current speed.


speed tested, the ratio of canopy height to mean shoot length was used as a measure of canopy deflection.

When shoots are randomly arranged in a seagrass bed, there can be gaps in the canopy. However, when flexible grass canopies are bent over in flowing water, the blades can cover gaps in the canopy and separate the slower flow within the canopy from the faster flow above it. We evaluated the gaps in the test grass beds at each shoot density and freestream current velocity by taking digital images of the entire test bed from above the flume. A downward-facing digital camera was positioned at a fixed point above the middle of the grass bed while the current was flowing and the entire bed was captured in a single digital image. Point Count software (Kohler and Gill, 2006) was utilized to randomly select 100 points in each image. The number of points on seagrass (as opposed to gaps, i.e., flume floor visible between regions of grass) grass were tallied for each photograph and used to calculate the percent canopy gap closure (PCG) as the percent of points falling on seagrass.

### Data Analysis

At each ambient freestream flow velocity (21, 32, or 63 cm s−<sup>1</sup> ) and position in the grass bed (0.50 or 1.00 m downstream from the leading edge of the bed), we tested whether shoot density or percent canopy closure affected three dependent variables: (1) percent reduction in current velocity within the canopy, (2) turbulence intensity within the canopy, and (3) drag force exerted on a shoot. For each flow velocity and shoot density, we also tested whether downstream position in the grass bed affected these three dependent variables.

One-way ANOVAs were performed after log transformation to satisfy assumptions of normality in order to test the effects of current speed, shoot density, PCG, and distance into the test canopy on percent flow reduction, turbulence intensity and force on individual shoots. Because of flow continuity, there can be no expectation of independence among measurements in the flume, thus freeing us to use this statistic as a tool to determine the change in velocities specifically as the result of non-independent factors (Fonseca and Koehl, 2006). Because we were interested in flow conditions within the entire canopy, our statistical tests of percent flow reduction and turbulence intensity utilized means of these parameters at each discrete elevation within the canopy as replicates. Percentages were arcsine transformed prior to application of ANOVA. Differences among means were determined by Duncan's multiple range test [DMRT] to guard conservatively against Type I error.

# RESULTS

#### Shoot Characteristics

Average leaf length from five randomly selected Z. marina shoots was 103.8 mm (s.d. 40.7), sheath length was 25.6 mm (s.d. 6.2), and leaf and sheath width were 3.2 mm (s.d. 0.45). There was no significant difference in shoot total length (sheath + leaf) between shoots collected from the two field sites (ANOVA, df = 1, F = 0.03, p = 0.861).

The flexural stiffness (EI) of sheaths of Z. marina shoots (mean = 1.89 × 10−<sup>6</sup> N m−2, s.d. = 1.13 × 10−<sup>6</sup> , n = 12 shoots) was approximately seven times higher than the EI of individual leaves (mean = 2.84 × 10−<sup>7</sup> N m−<sup>2</sup> , s.d. = 1.99 × 10−<sup>7</sup> , n = 24 leaves, each from a different shoot) This difference in stiffness was significant (ANOVA, df = 1, F = 47.51, p < 0.0001). Flexural stiffnesses of both leaves and sheaths were not significantly different between the two field sites (**Table 1**), indicating that all plants used in this study should have responded similarly to water flow.

#### Force on Shoots

There were no significant difference in the force exerted on shoots across current speeds as a function of shoot density (1246, 1046, 846, 646, 448, and 246 shoots m−<sup>2</sup> ), nor was there a suggestion of a trend (**Table 2**). There was no significant difference in drag force exerted on shoots as a function either of distance into the canopy or of current speed (**Table 2**).

#### Percent Reduction in Current Speed

There were no significant differences in the effect of shoot density on percent current speed reduction among distances downstream into the canopy if all current speeds were pooled (p > 0.05) (**Figures 2A,B**). Similarly, there was no effect of shoot density on percent current speed reduction for any of the three current velocities tested (**Figures 3A– C**). There was some indication that the lowest shoot density (246 shoots m−<sup>2</sup> ) was tending toward a diminished capacity to reduce current velocity at the 32 and 63 cm s −1 speeds, but the variability of percent current velocity

reduction within the canopy environment resulted in a non-significant difference.

#### Turbulence Intensity

There were no significant differences in the effect of shoot density on turbulence intensity among distances downstream into the canopy (p > 0.05; **Figures 4A,B**). Similarly, there was no consistent effect of shoot density on turbulence intensity for any of the three current velocities tested (**Figures 5A–C**). At the mid-range current speed (32 cm s−<sup>1</sup> ), some shoot densities were significantly different from others, but there was no systematic (with shoot density) pattern to these differences (**Figure 5B**). The faster the current speed, the lower the turbulence intensity within the canopy (**Figures 5A–C**).

#### Canopy Gap Closure

Percent canopy gap closure (PCG; shown as a fractional change) increased with shoot density irrespective of current regime (**Figure 6A**); however, there was no significant difference in PCG among the five lowest shoot densities and limited overlap in nonsignificant differences among the highest shoot densities, despite increased shoot bending and canopy compression (**Figure 1**, compare **Figures 1B,D,F**). There was a stepwise increase in canopy closure with both shoot density and increased current speed, with similar proportions of PCG occurring among current speeds (**Figure 6B**). Current speed resulted in significant (p < 0.05) differences in PCG only among the highest and lowest speeds tested (mean PCG at 21 cm s−<sup>1</sup> was 20.3% [n = 6; s.d. 10.4], at 32 cm s−<sup>1</sup> was 32% [n = 6; s.d. 15.9], and at 63 cm s−<sup>1</sup> was 41% [n = 6; s.d. 23.9]).

# DISCUSSION

Canopies of aquatic plants play important roles in marine and freshwater ecosystems, such as providing food and habitat for animals, and stabilizing and altering the chemistry of sediments. The performance of these functions by canopies of seagrasses or macroalgae depends on their interaction with the ambient water flow. We have focused on seagrass beds exposed to unidirectional water currents to examine the roles of shoot density and flexibility in affecting the hydrodynamics of submerged aquatic plant canopies.

#### Shoot Density

Many studies have revealed that the vertical velocity profile and turbulence structure of a unidirectional water current are altered upon encounter with a seagrass bed, where a zone of rapid flow above the canopy and a zone of slower flow within the bed develop (Fonseca et al., 1982, 1983, 2007; Fonseca and Kenworthy, 1987; Gambi et al., 1990; Ackerman and Okubo, 1993; Worcester, 1995; Koch and Gust, 1999; Verduin and Backhaus, 2000; Abdelrhman, 2003; Peterson et al., 2004; Chen et al., 2013; **Figure 1**). If a seagrass bed is composed of flexible shoots, they are bent over by flowing water and overlap, thereby forming a compressed canopy. The roles of shoot density and of canopy deflection in determining the degree to which water flow is slowed within and re-directed above a bed of flexible seagrasses has been unclear. Our study of flow through beds of very flexible Zostera marina plants showed that shoot density had little influence on flow reduction and turbulence intensity at each downstream position in the bed and at each current velocity that we tested. Although our data hinted that the capacity of a grass bed to reduce flow velocity might be lower for the sparsest shoot densities we tested when exposed to the most rapid ambient currents we imposed, this result was not significant. Our data do not show the dramatic reductions in flow within a canopy as shoot density increases that were predicted by Abdelrhman (2003), who assumed that the simple displacement of water by shoots would limit water movement within the canopy. Instead, our data suggest that the effect of shoot density

on velocity reduction was largely eclipsed by other changes in the canopy architecture due to flow-induced bending of flexible shoots and the resulting canopy compression. Furthermore, our data suggests that the sedimentation that occurs in the slowed flow in a grass bed, and thus carbon deposition and sediment organic content, would be facilitated across much of the range of Z. marina shoot densities observed in nature.

### Shoot Flexibility

The inconsistent results of different studies of the effects of shoot density on the hydrodynamics of seagrass beds exposed to unidirectional currents can be explained in part by differences in the flexibility of the shoots or physical models composing the canopy. For example, Chen et al. (2007) adopted Nepf's (1999) approach of using arrays of rigid cylinders to evaluate the effects of shoot density on flow through and around submerged plant canopies and found that increasing shoot density decreased water velocities in the canopy. In contrast, we found that shoot density had little effect on flow reduction in canopies of flexible seagrass. Dijkstra and Uittenbogaard (2010), who used both mathematical modeling and flume experiments with plastic mimics of shoots of different stiffness to determine water flow in canopies, discovered that shoot stiffness had a bigger effect on flow reduction in the canopy than did shoot density.

When a flexible organism attached to the substratum is bent over by ambient water flow, its projected area normal to the flow direction is reduced and the drag it experiences is lower than drag on a stiff organism of the same size and shape. This drag-reducing function of flexibility has been measured for macroalgae and sessile aquatic animals (e.g., Koehl and Wainwright, 1977; Koehl, 1984, 1986; Koehl et al., 2001), and for seagrass shoots and physical models (e.g., Bouma et al., 2005; Fonseca et al., 2007; Dijkstra and Uittenbogaard, 2010). Such drag reduction enhances the ability of a flexible organism to avoid being ripped off the substratum by ambient water motion. Another advantage of flexibility occurs when groups of attached organisms form canopies. Fonseca and Fisher (1986) found that canopies composed of species of seagrass that had flexible shoots compressed more in flowing water, and thereby provided greater protection from sediment erosion, than did canopies of species with stiffer shoots. Because faster flow causes greater deflection and streamlining of flexible shoots, and thus more canopy compression, it is not surprising that the drag we measured on flexible Z. marina shoots within grass beds was independent of ambient current velocity.

## Canopy Configuration and Gap Closure

Gaps between shoots in a canopy can enhance light and flow penetration into the canopy, but when flexible shoots are bent over by flowing water, they can cover gaps. Not unexpectedly, we found that gap closure was greater when canopies of flexible Z. marina were exposed to fast ambient currents than when they experienced slower flow. Furthermore, at each current speed, gap closure was more pronounced in canopies with high shoot densities than in those with low densities. However, canopy gap closure did not correlate with the percent velocity reduction in grass beds or with drag on individual shoots, both of which were independent of ambient current velocity and shoot density. A possible explanation is that the effects of canopy closure on the flow above and through a flexible seagrass canopy were manifested at very low levels of gap closure (∼20% PCG) and at low ambient current speeds (e.g., Fonseca and Kenworthy, 1987). Our result that downstream position had no effect on flow reduction, turbulence intensity, drag on a shoot, and gap closure suggests that the flow structure in a flexible canopy develops quickly with distance as water moves into the canopy, as has been seen in other studies (Fonseca et al., 1982, 2007; Fonseca and Fisher, 1986; Gambi et al., 1990; Abdelrhman, 2003).

The lack of an effect of shoot density on velocity reduction and turbulence intensity in seagrass canopies could help explain why experiments in which shoot density was manipulated on a local patch scale in one hydrodynamic environment did not alter sediment characteristics in seagrass beds (Armitage and Fourqurean, 2016; Howard et al., 2016). This could explain the lack of a relationship between sediment stores of organic carbon and seagrass density that has been observed on a regional scale (Campbell et al., 2015).

# Does Ambient Current Speed Affect Shoot Density in Seagrass Beds?

High shoot densities have been observed in grass beds in habitats exposed to rapid water flow (e.g., Conover, 1964; Kenworthy et al., 1982; Fonseca and Bell, 1998). We found that changing the density of flexible shoots does not affect the flow in a grass bed exposed to a given ambient current. Therefore, our results are inconsistent with the hypothesis that high shoot density is due to a direct, adaptive response of the grass to rapid water flow.

Rather than current speed providing some signal that stimulates an increase in shoot density, water flow might affect shoot spacing indirectly via a variety of mechanisms. For example, the accumulation of sediment and its content of nutrients and organic matter all are functions of the depositional environment, and all are drivers of seagrass shoot density

(Holmer et al., 2008; Howard et al., 2016; Ceccherelli et al., 2018). In very oligotrophic environments, high nutrient supply increases shoot density (e.g., Fourqurean et al., 1995; Agawin et al., 1996). Similarly, in such resource-limited habitats where nutrients are taken up through seagrass leaves rather than the roots, high flow velocities increase nutrient delivery by making the diffusion boundary layer around the leaves thinner (e.g., Koch, 2001). In contrast, at eutrophic sites where sediment anoxia induces plant stress, slow flow regimes can result in reduced shoot density (e.g., Borum et al., 2005; Holmer et al., 2008; Brodersen et al., 2015). Feedback between a number of interacting environmental factors can also affect the light reaching seagrass, and thus their growth patterns, as described by van der Heide et al. (2011). For example, in habitats exposed to rapid water flow, seagrass canopies trap suspended particles, thereby improving water clarity and enhancing light. In contrast,

in estuaries with slow water flow, eutrophication reduces the light reaching the seagrass.

Our data are consistent with the view that the high shoot densities of seagrass canopies in habitats exposed to rapid water flow can be due to the effects of water motion on other factors, such as light and nutrients, that affect seagrass growth.

#### AUTHOR CONTRIBUTIONS

MF led the research and oversaw the experimental process and drafted the original manuscript. MK reviewed and revised the manuscript with regard to flow dynamics and interaction with the flexible seagrass shoots. JF reviewed and revised the manuscript with regard to environmental feedbacks and alternative mechanisms influencing seagrass shoot density.

#### REFERENCES


#### FUNDING

Support was provided early and in part by the NOAA, Beaufort Laboratory (to MF) and later by CSA Ocean Sciences Inc. (to MF) and by National Science Foundation Grant IOS-1655318 (to MK). This is contribution #123 from the Center for Coastal Oceans Research in the Institute of Water and Environment at Florida International University.

#### ACKNOWLEDGMENTS

Thanks to L. Rose for providing technical support of the study at the NOAA, Beaufort Laboratory. Early review of this work by the late E. M. Koch and subsequently, G. Piniak, A. Uhrin and three reviewers greatly improved the manuscript.

and suspended sediment transport. Est. Coasts 30, 296–310. doi: 10.1007/ bf02700172



**Conflict of Interest Statement:** MF was employed by NOAA during the collection of the data and by CSA Ocean Sciences Inc. during the analysis and writing of the manuscript.

The remaining 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 Fonseca, Fourqurean and Koehl. 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.