# MULTISCALE APPROACH TO ASSESS FOREST VULNERABILITY

EDITED BY : Giovanna Battipaglia, Veronica De Micco and Andreas Rigling PUBLISHED IN : Frontiers in Plant 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 978-2-88963-860-4 DOI 10.3389/978-2-88963-860-4

#### 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

# MULTISCALE APPROACH TO ASSESS FOREST VULNERABILITY

Topic Editors:

Giovanna Battipaglia, University of Campania Luigi Vanvitelli, Italy Veronica De Micco, University of Naples Federico II, Italy Andreas Rigling, Swiss Federal Institute for Forest, Snow and Landscape Research (WSL), Switzerland

Citation: Battipaglia, G., De Micco, V., Rigling, A., eds. (2020). Multiscale Approach to Assess Forest Vulnerability. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-860-4

# Table of Contents


Mahmuda Islam, Mizanur Rahman and Achim Bräuning

*59 Changes in Spatiotemporal Patterns of 20th Century Spruce Budworm Outbreaks in Eastern Canadian Boreal Forests*

Lionel Navarro, Hubert Morin, Yves Bergeron and Miguel Montoro Girona

*74 Early-Warning Signals of Individual Tree Mortality Based on Annual Radial Growth*

Maxime Cailleret, Vasilis Dakos, Steven Jansen, Elisabeth M. R. Robert, Tuomas Aakala, Mariano M. Amoroso, Joe A. Antos, Christof Bigler, Harald Bugmann, Marco Caccianaga, Jesus-Julio Camarero, Paolo Cherubini, Marie R. Coyea, Katarina Čufar, Adrian J. Das, Hendrik Davi, Guillermo Gea-Izquierdo, Sten Gillner, Laurel J. Haavik, Henrik Hartmann, Ana-Maria Hereş, Kevin R. Hultine, Pavel Janda, Jeffrey M. Kane, Viachelsav I. Kharuk, Thomas Kitzberger, Tamir Klein, Tom Levanic, Juan-Carlos Linares, Fabio Lombardi, Harri Mäkinen, Ilona Mészáros, Juha M. Metsaranta, Walter Oberhuber, Andreas Papadopoulos, Any Mary Petritan, Brigitte Rohner, Gabriel Sangüesa-Barreda, Jeremy M. Smith, Amanda B. Stan, Dejan B. Stojanovic, Maria-Laura Suarez, Miroslav Svoboda, Volodymyr Trotsiuk, Ricardo Villalba, Alana R. Westwood, Peter H. Wyckoff and Jordi Martínez-Vilalta

*88 Projecting Tree Species Composition Changes of European Forests for 2061–2090 Under RCP 4.5 and RCP 8.5 Scenarios* Allan Buras and Annette Menzel

*101 River Regulation Causes Rapid Changes in Relationships Between Floodplain Oak Growth and Environmental Variables* Maksym Netsvetov, Yulia Prokopuk, Radosław Puchałka, Marcin Koprowski, Marcin Klisz and Maksym Romenskyy

*112 Limitations at the Limit? Diminishing of Genetic Effects in Norway Spruce Provenance Trials*

Marcin Klisz, Allan Buras, Ute Sass-Klaassen, Radosław Puchałka, Marcin Koprowski and Joanna Ukalska


Maria Holmberg, Tuula Aalto, Anu Akujärvi, Ali Nadir Arslan, Irina Bergström, Kristin Böttcher, Ismo Lahtinen, Annikki Mäkelä, Tiina Markkanen, Francesco Minunno, Mikko Peltoniemi, Katri Rankinen, Petteri Vihervaara and Martin Forsius

	- Hannes Seidel, Michael Matiu and Annette Menzel

# Editorial: Multiscale Approach to Assess Forest Vulnerability

#### Giovanna Battipaglia<sup>1</sup> \*, Andreas Rigling2,3 and Veronica De Micco<sup>4</sup>

<sup>1</sup> Department of Environmental, Biological and Pharmaceutical Sciences and Technologies University of Campania L. Vanvitelli, Caserta, Italy, <sup>2</sup> Swiss Federal Research Institute WSL, Birmensdorf, Switzerland, <sup>3</sup> Institute of Terrestrial Ecosystems, ETH Zurich, Zurich, Switzerland, <sup>4</sup> Department of Agricultural Sciences, University of Naples Federico II, Naples, Italy

#### Keywords: climate change, forest vulnerability, forest dynamics, forest growth, drought mortality

**Editorial on the Research Topic**

#### **Multiscale Approach to Assess Forest Vulnerability**

#### Edited by:

Sergio Rossi, Université du Québec à Chicoutimi, Canada

#### Reviewed by:

Miguel Montoro Girona, Université du Québec en Abitibi Témiscamingue, Canada

#### \*Correspondence:

Giovanna Battipaglia giovanna.battipaglia@unicampania.it

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 06 March 2020 Accepted: 11 May 2020 Published: 05 June 2020

#### Citation:

Battipaglia G, Rigling A and De Micco V (2020) Editorial: Multiscale Approach to Assess Forest Vulnerability. Front. Plant Sci. 11:744. doi: 10.3389/fpls.2020.00744 In recent decades, forest, vulnerability to climate change is rapidly increasing worldwide; forest dieback episodes have been recorded in all biomes (Allen et al., 2010; Adams et al., 2017). In forest ecosystems a high probability of intensified occurrence of extreme events, such as heat waves, droughts, fires, flooding or pest outbreaks is expected (Seidl et al., 2011; FAO, 2018).

Forest vulnerability can be defined as the degree to which a forest ecosystem is susceptible to adverse effects of climate change. Thus, vulnerability is a function of the climate variation which the forest is exposed to (exposure), its sensitivity, and its adaptive capacity to respond to the potential impacts (**Figure 1A**).

Several cases of widespread dieback and increased mortality rates have been described for different tree and shrub species, revealing the high vulnerability of some forest ecosystems, manifested as a loss in tree vigor, growth decline and sometimes tree death (Montoro Girona et al., 2019). The increasing forest vulnerability and occurrence of dieback cases, extended to a larger scale, can have the potential to rapidly alter forest functioning and respective ecosystem services, with important implications on the carbon-water balance, plant community composition, and tree population dynamics (Rewald et al., 2020). Thus, assessing increased forest vulnerability is a challenging issue and this Research Topic (RT) aimed to contribute to close research gaps in identifying vulnerable species and ecosystems, understanding factors triggering forest vulnerability, estimating the adaptive capacity of forest ecosystems to climate change, and identifying possible effects of forestry practices on forest vulnerability.

The 17 articles collected in this RT underlined how the assessment of vulnerability is a complex matter, leading to a common agreement on the key role of interdisciplinary research and networking to fasten the understanding of the phenomena and related factors affecting tree mortality as well as to develop sustainable forest management practices under climate change.

The majority of papers deal with forest vulnerability to extreme drought. Different species- and provenance specific-impacts, site conditions and stand compositions are considered as factors

affectingf plant growth and forest mortality. Wang et al. investigated the vulnerability of Quilian junipers, a widespread species of the Northeastern Tibetan Plateau, showing negative growth trends when facing extreme drought events. The authors underlined that the oldest populations are the most vulnerable, characterized by the lowest resistance values, the narrowest annual rings, and the highest proportion of missing rings during the drought years.

The importance o understanding and selecting the more resilient provenances of Scots pine to alleviate pressures of climate change on forest ecosystems, was investigated by Seidel et al. The authors demonstrated that southern provenances of Scots pine were better adapted to drought conditions than northern ones, showing a less severe drought response and exhibiting morphological characteristics associated with drought resistance. Klisz et al. studied provenance-specific climate sensitivity of Norway spruce growing in the natural range and at the climatic margin of species distribution. Results showed that the marginality of site has to be considered when evaluating climate sensitivity, since the provenance growing at the margin of distribution was much more affected by drought than the one growing in the center of the distribution.

Not only species, but also site conditions and stand structure, influenced the responses to drought of pure and mixed forests of Mediterranean forests. Indeed, Zalloni et al. demonstrated that the impact of drought on Quercus ilex growing in a pure or mixed stand with Pinus pinea, in Southern Italy, is strongly related to the characteristics of the stands. In a mixed stand, where Q. ilex trees are young and stand density is high, thinning of P. pinea, is advisable to limit inter-specific competition for resources and to promote Q. ilex wood growth. On the contrary, when stand density is lower, promoting the co-existence of Q. ilex and P. pinea could facilitate complementarity in resource use. In the Mediterranean forest, considered particularly vulnerable to climate change (FAO, 2018), Gazol et al. showed a potential compositional shift in two large Spanish forests as a consequence of recent warming trends and the severe droughts, including that occurring in 2012, which caused die-off in both forests. The authors findings warned against the local extinction of some tree populations near their southernmost distribution limits as in the P. sylvestris forest studied which may become more vulnerable to forecasted aridification trends.

Moving to tropical ecosystems, a large dendrometer dataset was analyzed (Raffelsbauer et al.) to understand growth recovery after dry events in Ecuador. Even if the precipitations are abundant, dry spells occur regularly during so-called "Veranillo del Niño" (VdN) periods in October-November. Analyzing two broadleaved species, the authors showed that higher frequency of drought might increase inter-species competition and species-specific mortality and might finally alter the species composition of the ecosystem. A second study carried out in the tropics was based on wood traits (Islam et al.) and covered different tree species. The authors demonstrated different adjustment strategies among the species, probably due to the differences in evolutionary trends in xylem traits.

The importance of large databases in assessing general growth patterns of species and mortality rate is highlighted in the studies of Etzold et al. and Cailleret et al. Indeed, Etzold et al. studying 276 permanent plots across Switzerland, assessed mortality rates of five dominant species throughout the last century (1898–2013) and examined factors driving mortality change. Surprisingly the authors found that mortality rates increased only slightly over the last ∼120 years, which could be mainly related to changes in stand structure. This suggests that Swiss forests have been resilient to recent climate change so far and that changes in species composition might occur more gradually. A large tree-ring database was used by Cailleret et al. to explore forest mortality and to provide a robust method for estimating early-warning signals of tree mortality. The authors reported a decrease in growth rates, an increase in inter-annual growth variability and a decrease in growth synchrony in gymnosperm species, that could be used as powerful predictors of mortality risk, but not necessarily so for angiosperms. Their results are in agreement with several studies where species distributional range and forest composition changes are predicted (Sykes and Prentice, 1996; Rigling et al., 2013; Fekete et al., 2017; Scherrer et al., 2017).

Buras and Menzel projected tree species composition changes, according to future climate scenarios. Their results indicate significant changes in European forest species composition: species richness decline in the Mediterranean and Central European lowlands, while increasing diversity in Scandinavian and Central European high-elevation forests.

The impact of climate change on the ecosystem services (ES) of boreal forests are evaluated by Holmberg et al., applying two dynamic ecosystem models. Even if they found both, beneficial and detrimental consequences, the main output of their paper was the high range of uncertainty in future provision of boreal forest ES.

Apart from climate impacts, especially drought, other important factors, including fire, flood or pest outbreaks, can trigger forest vulnerability, but their impact is less studied. According to the Web of ScienceTM database (Thomson Reuters) (**Figure 1B**), considering the last 20 years (1999–2019) the majority of studies are related to drought and models, while only a small percentage is related to the other non-climatic factors probably because of the difficulty to disentangle the effects of climate and biotic and abiotic factors on forest dieback (Niccoli et al.). A part from models, this percent distribution of papers among subjects is reflected in this RT as well, with only one study dealing with the effects of fire on forest growth, (Niccoli et al.) one paper on the effects of floodplain on forest productivity, (Netsvetov et al.) two studies analyzing the relationships between pest outbreaks and tree mortality (Navarro et al.; Rossi et al.) and two contributions reporting on the effects of silvicultural practices in reducing forest vulnerability (Girona et al.; Thyroff et al.). All those papers presented novel and interesting results in their topics. Indeed, Niccoli et al. used a multidisciplinary approach to clarify the ecophysiological processes allowing Pinus pinaster to survive after a severe wildfire, while Netsvetov et al. demonstrated that trees growing in areas exposed to urban development are the most susceptible to downside effects of river regulation. The paper of Navarro et al. was the first evidence of a northward shift of the distribution area of spruce budworm, as well as the first spatio-temporal reconstruction at the landscape level in Quebec, highlighting the complexity of outbreak dynamics and, at the same time, the necessity of this approach for assessing boreal forest adaptation and modification to future climate change. Rossi et al. studied how the spruce budworm outbreak affected the responses of Picea mariana to a subsequent thinning. Their dendrochronological analyses underlined that the effect of thinning on tree-ring width of Picea mariana was independent of the growth reduction that trees had experienced during the outbreak.

To cover the importance of new silvicultural practices in reducing forest vulnerability, Girona et al. presented a regeneration study in Picea mariana stands. Their findings confirmed, for the first time, that experimental shelterwood and seed-tree harvesting followed by scarification allow the establishment of an abundant black spruce regeneration in North American boreal forests and can be a viable silvicultural alternative to clear-cutting. Finally, the study of Thyroff et al. demonstrated the high plasticity of Quercus virginiana to resource availability (such as light) and the importance of applying more than one silvicultural treatment to reduce the need for costly competing vegetation control.

Taken together, the 17 contributions to this RT cover many aspects of timely research on forest vulnerability to different environmental stressors. Notwithstanding the numerous claims on the need to apply multidisciplinary and multi-scale approaches to evaluate forest ecosystem condition, to assess vulnerability to environmental stress and adopt countermeasures against mortality, there is still a lack of integration among disciplines, data fusion capacities and of collaborative actions at regional and global scales (Battipaglia et al., 2014; De Micco et al., 2019).

Therefore, as an output of this RT, many outstanding research gaps were evidenced which highlighted the following issues:



To fill these gaps, making synergies among disciplines to speed-up the process of knowledge achievement, a networking effort is needed as also highlighted recently in other contexts. Networking also across disciplines and communities would be highly welcome to achieve an in-depth understanding of forest functioning through the whole soil-plant-atmosphere pathway allowing for predictions on forest destiny and ES under various climate change scenarios.

#### REFERENCES


### AUTHOR CONTRIBUTIONS

GB, AR, and VD: conceptualization. GB: writing original draft. All authors: writing—review & editing.

# ACKNOWLEDGMENTS

GB would like to thanks the PRIN 2017 Project The Italian TREETALKER NETWORK: continuous large-scale monitoring of tree functional traits and vulnerabilities to climate change.

interactions and physiological functioning above and below ground. Front. Plant Sci. 11:173. doi: 10.3389/fpls.2020.00173


**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 © 2020 Battipaglia, Rigling and De Micco. 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.

# Does a Spruce Budworm Outbreak Affect the Growth Response of Black Spruce to a Subsequent Thinning?

Sergio Rossi1,2 \*, Pierre-Yves Plourde<sup>1</sup> and Cornelia Krause<sup>1</sup>

<sup>1</sup> Département des Sciences Fondamentales, Université du Québec à Chicoutimi, Chicoutimi, QC, Canada, <sup>2</sup> Key Laboratory of Vegetation Restoration and Management of Degraded Ecosystems, Guangdong Provincial Key Laboratory of Applied Botany, South China Botanical Garden, Chinese Academy of Sciences, Guangzhou, China

In Canada, new forestry practices involving the natural dynamics of tree growth and regeneration are proposed for integrating forest management with biodiversity. In particular, the current spruce budworm [Choristoneura fumiferana (Clemens)] outbreak in northeastern North America is forcing natural resource managers to clarify the potential interactions between natural disturbances and commercial thinning. The aim of this study was to investigate if the spruce budworm outbreak of the 1970s affected the responses of black spruce [Picea mariana (Mill.) B.S.P.] to a subsequent thinning. Stem growth was reconstructed by measuring and cross-dating chronologies of tree-ring width of 1290 adult trees from 34 control and thinned stands within an area of 11,000 km<sup>2</sup> in the boreal forest of the Saguenay-Lac-Saint-Jean region (QC, Canada). The treatment consisted of a low thinning performed during 1995–1999 that removed 25–35% of the basal area. Segmented models were applied to the tree-ring chronologies to define the growth pattern during the outbreak and thinning periods within a time window of 8 years, representing the average duration of the effects of defoliation on growth. Trees showed abrupt growth decreases during the outbreak, with the tree-ring index showing minimum values in 1977–1979. The tree-ring index had a flat trend before thinning, while it increased for 6–10 years after thinning. The growth pattern during the outbreak period was characterized by a reduction, mainly in trees with larger tree rings, while slow-growing trees showed less sensitivity to the disturbance. Thinning produced a significant increase in tree growth. No relationship was found between the effects of spruce budworm outbreaks in trees and the changes in growth pattern after thinning. If the timespan between the two disturbances exceeds 7 years, partial cutting can be applied independently of the growth reductions that had occurred during the outbreak. When applied in black spruce stands with high annual radial growth, thinning is expected to optimize the volume growth of the residual trees.

Keywords: boreal forest, Choristoneura fumiferana, dendroecology, disturbance, growth rate, growth release, Picea mariana, sylviculture

# INTRODUCTION

In North America, the cyclical stand renewal of several boreal species is closely related to two main disturbances: fire and insect outbreaks, the latter frequently being more important than the former (Gauthier et al., 2009). The larvae of spruce budworm [Choristoneura fumiferana (Clemens)], a native defoliator, feed on young and, more rarely, old needles consuming the photosynthetic tissues

#### Edited by:

Giovanna Battipaglia, Università degli Studi della Campania Luigi Vanvitelli, Italy

#### Reviewed by:

Martina Pollastrini, Università degli Studi di Firenze, Italy Miguel Montoro Girona, Swedish University of Agricultural Sciences, Sweden

> \*Correspondence: Sergio Rossi sergio.rossi@uqac.ca

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 10 May 2018 Accepted: 29 June 2018 Published: 24 July 2018

#### Citation:

Rossi S, Plourde P-Y and Krause C (2018) Does a Spruce Budworm Outbreak Affect the Growth Response of Black Spruce to a Subsequent Thinning? Front. Plant Sci. 9:1061. doi: 10.3389/fpls.2018.01061

**9**

in spring and early summer (Gauthier et al., 2009). During outbreaks, the repeated defoliations dramatically deplete the reserves of trees, causing growth reductions and mortality of single trees or whole stands in a few years (Bouchard and Pothier, 2010). Although balsam fir [Abies balsamea (L.) Mill.] and white spruce [Picea glauca (Moench) Voss] are the main host species, recent studies have demonstrated the occurrence of distinct growth reductions in black spruce [Picea mariana (Mill.) B.S.P.] during the last four spruce budworm outbreaks (Gauthier et al., 2009; Tremblay et al., 2011). Because of its wide transcontinental distribution across the boreal forest of North America, black spruce is a keystone species of huge ecological and economic importance. Thus, all effects of natural disturbances on its growth can be spread over vast areas and have relevant consequences for the forest industry in northern regions.

The population of spruce budworm oscillates consistently over time, generating recurrent outbreaks (Montoro Girona et al., 2018). Jardon et al. (2003) demonstrated that the population increases in density at sub-continental scale every 25–38 years. The last outbreak occurred during 1970–1985 affecting a huge area with mature stands (Gauthier et al., 2009). The population of spruce budworm is now once again intensely rising. Within 4 years, the current outbreak has affected more than 4.2 M ha in Quebec (MFFP, 2017).

In Canada, the recent ecosystem-based management is attempting to modify conventional forestry with new practices closely adapted to the various regional differences and lower environmental impacts. This approach to silviculture is mainly based on the best available ecological knowledge to establish the most suitable harvesting strategies and targets (Gauthier et al., 2009). The application of these practices aims as far as possible to integrate management with biodiversity and the natural dynamics of growth and regeneration of the forest (Haeussler and Kneeshaw, 2003). For example, a number of partial cuttings have been developed to stimulate the growth of residual trees and advance their harvesting by simulating the forest gaps produced by the mortality following a spruce budworm attack (Zhang et al., 2006; Gauthier et al., 2009; Montoro Girona et al., 2016; Lemay et al., 2018).

Thinning is frequently used in even-aged stands of the boreal forest (Park and Wilson, 2007; Pelletier and Pitt, 2008). In ecosystems dominated by black spruce, canopy opening allows trees to receive more light for accomplishing their photosynthetic processes, warms up the soil, and increases plant nutrition by reactivating the cycling of nutrients accumulated in the thick organic layers (Pothier and Prevost, 2002; Vincent et al., 2009). However, because of the superficial distribution of the root system in black spruce, thinning could make stands more susceptible to wind, thus increasing the risk of windthrow (Cremer et al., 1982; Ruel et al., 2003; Gardiner et al., 2008). This expensive practice is worthwhile only if the reduction in stand density and competition allows the growth in diameter of the remaining trees to be substantially stimulated (Nyland, 2002; Vincent et al., 2009). Although encouraging results have been demonstrated for black spruce production (Vincent et al., 2009; Soucy et al., 2012; Pamerleau-Couture et al., 2015; Montoro Girona et al., 2017), some questions on the long-term effects of thinning on stand resistance and resilience remain unanswered (Gauthier et al., 2009; Bauce and Fuentealba, 2013). In particular, the current spruce budworm outbreak occurring in Eastern Canada is forcing forest scientists to urgently clarify the interactions between natural disturbances and silvicultural practices. The aim of this study was to investigate if the spruce budworm outbreak of the 1970s affected the responses of black spruce to a subsequent thinning realized in 1990s. To attain this objective, we reconstructed the stem growth by measuring and cross-dating tree-ring width in trees from 34 control and thinned stands.

# MATERIALS AND METHODS

#### Study Area

The study was conducted within an area of 1,1 M ha in the coniferous boreal forest of the Saguenay-Lac-Saint-Jean region (QC, Canada) belonging to the black spruce-feather moss domain. The region has a gently rolling topography with hills reaching 700 m a.s.l. on thick and undifferentiated glacial till deposits. The climate is boreal with cold winters and cool summers. The mean monthly temperatures can reach 17◦C during July, but drop as far as −20◦C in January. The mean annual temperature is approximately 2◦C, with a May–September temperature of 13◦C, and May–September rainfall of 401 mm. Winters are long, with daily mean temperatures around or below zero for a period of 180 days or more (Rossi et al., 2011).

# Stand Selection and Experimental Design

Thirty-four mature stands were selected (**Figure 1**). The stands, of fire origin, were even-aged and had a comparable species composition strongly dominated by black spruce, in association

with balsam fir [Abies balsamea (L.) Mill.], Jack pine (Pinus banksiana Lamb.), white spruce [Picea glauca (Moench) Voss], white birch (Betula papyrifera Marsh.), and trembling aspen (Populus tremuloides Michx.). Twenty-three stands underwent a low thinning treatment during 1995–1999 (Supplementary Table S1). The thinning removed 25–35% of the stand basal area and was performed from extraction trails opened to establish an access network to the stands (Tremblay and Laflèche, 2012).

#### Data Collection

In each stand, 20–84 dominant or co-dominant trees were selected (Supplementary Table S1). The large variability in the number of sampled trees was due to a preliminary analysis to assess the more suitable sample size, which was estimated to 35 trees (Vincent et al., 2009). In some cases, chronologies could not be cross-dated, which reduced the sample size to a value lower than 35. The trees had a height of between 9 and 18 m and a diameter at breast height (DBH) up to 20 cm. The decayed inner part of several trees prevented an exact evaluation of their age. On average, the minimal age of the trees was estimated to range between 56 and 126 years.

One increment core was collected with a Pressler borer 30 cm above the root collar of each tree and processed according to the standard methods in dendrochronology (Stokes and Smiley, 1968). The cores were collected close to the root collar to precisely estimate tree age and deepen as much as possible the tree-ring series. The samples were prepared according to the standard method in dendrochronology (Tremblay et al., 2011), and measured to the nearest 0.01 mm using either WinDENDRO (Regent Instruments Inc., Canada) or a manual Henson micrometer (LINTABTM, Rinntech, Heidelberg, Germany). The ring-width series were corrected by visual cross-dating that was confirmed using the COFECHA computer program (Holmes, 1983). Each time series was standardized by applying a spline of 60 years to remove the long-term fluctuation. Autocorrelation in the data was not removed as the cumulative effect of defoliation was considered important for the analysis.

#### Identification of the Growth Patterns

A segmented model was applied to the tree-ring chronologies to define the growth pattern during the outbreak and thinning periods. The model E(x) consisted of two linear segments according to the following system:

$$E(\mathbf{x}) = \begin{cases} a + b\mathbf{x}, & \mathbf{x} < X \\ c + d\mathbf{x}, & \mathbf{x} \ge X \end{cases}$$

Where x represents the time axis, a, b, c, and d are the coefficients of the model, and X is the reference year for the outbreak [1978 (Krause et al., 2012)] or thinning (reported in **Table 1**). A condition was imposed that the two segments of the model were connected at their extremity in X:

$$a + bX = c + dX$$

Thus, the coefficients were related to the following relationship:

$$X = \frac{c - a}{b - d}$$

TABLE 1 | Effects of thinning (TH), growth rate (GR), and concavity during the spruce budworm outbreak (CO) on the growth pattern after thinning.


The growth patterns tested by GLM and binary logit models are represented by the slope of the regression calculated for the four years after thinning and by the proportion of convex growth calculated for four years before and four years after thinning, respectively.

The model was fitted to each tree-ring series and for the period of outbreak and thinning using the NLIN procedure in SAS 9.4 (SAS Institute Inc., Cary, NC, United States). A time window of 8 years (four before and four after outbreak or thinning) was used for the fitting, which was considered to represent the average duration of the effects of the outbreak on black spruce growth (Tremblay et al., 2011). The model for the outbreak period was fitted on standardized data after removing the long-term fluctuations assessed using a spline of 60 years. The segmented model identified two growth patterns according to the slopes b and d of the segments. A concave or convex pattern resulted if b > d or b < d, respectively, corresponding to a negative or positive growth after the event, as shown in the examples of **Figure 2**.

#### Statistical Analyses

fpls-09-01061 July 21, 2018 Time: 15:45 # 4

The trees were separated in two groups (slow- and fastgrowing) of similar size according to their growth rates calculated as the average tree-ring width measured during the 6 years preceding the outbreak (1965–1970). The two resulting groups represented the trees with a growth rate lower and higher than the median, respectively. The effect of the growth pattern during the outbreak and growth rate on the slope of the regressions after thinning were assessed with a Generalized Linear Model (GLM). In addition, the binary responses to thinning, represented by the concave or convex growth patterns, were compared with the growth pattern during the outbreak and between the two growth rates using a binary logit model in SAS 9.4 (SAS Institute Inc., Cary, NC, United States).

#### RESULTS

#### Tree-Ring Chronologies

Both slow- and fast-growing trees showed an abrupt decrease in tree-ring width during the period of spruce budworm outbreak (**Figure 3**). The tree-ring index dropped dramatically in 1976, with the minimum values observed in 1977–1979. After that period, growth remained low for 2–3 years in slow-growing trees. A gradual recovery was observed in fast-growing trees. Control and thinned stands exhibited the same growth patterns, and seemed to respond similarly to the outbreak.

During the 10 years before thinning, the tree-ring index had a flat trend, except for the fast-growing trees of control stands, which showed a more irregular pattern (**Figure 4**). Gradual increases in tree-ring width followed by a plateau were observed for 6–10 years after thinning. However, fast-growing trees of control stands increased their growth only 7 years after thinning.

#### Regression Slopes

The segmented models analyzed specifically the growth patterns occurring during the period close to the outbreak and thinning events and only in part reflected the results observed in the chronologies, which covered the average tree-ring width over a wider time window. Before 1978, i.e., the selected outbreak year,

regressions had a decreasing pattern, showed by the negative values of their slopes, −0.8 × 10−<sup>2</sup> and −9.4 × 10−<sup>2</sup> on average for slow- and fast-growing trees, respectively (**Figure 5**). After the outbreak, regressions exhibited both negative and positive patterns except for slow-growing trees, which still had negative trends. Control and thinned stands showed similar growth patterns. Control and thinned stands had similar slopes before thinning, with values close to or below zero. After thinning, significant differences in slope were detected between treatments, with thinned stands having more positive regressions and slopes with higher absolute values than control stands (**Table 1**). A greater dispersion in the slope of the regressions occurred in fast-growing than in slow-growing trees, as also detected by significant results of GLM. No effect of the growth pattern occurring during the spruce budworm outbreak was observed in the slope of the regression after thinning (GLM, p > 0.05).

The proportion of negative and positive slopes of the regressions confirmed the previous results but clarified the main trends of the growth patterns (**Figure 6**). Both before and after the outbreak, negative slopes were more frequent than positive ones except for fast-growing trees after the outbreak. Negative slopes were observed before thinning, but greater proportions of positive slopes occurred afterward. Thinned stands had a higher percentage of positive slopes than control stands, but this difference was more marked for fast-growing trees.

#### Growth Patterns

The growth pattern during the outbreak period was markedly convex in fast-growing trees (75 and 80% in thinned and control stands, respectively), while the slow-growing trees in control stands exhibited 67% of concave growth patterns (**Figure 7**). Convex growth patterns were observed during thinning in both control and thinned stands, although the proportion of convex growth patterns significantly increased in both slowand fast-growing trees of thinned stands (GLM, p < 0.0001, **Table 1**). No significant difference was observed between slowand fast-growing trees (GLM, p > 0.05). Similarly, GLM detected no effect of the growth patterns occurring during the spruce budworm outbreak on the concavity during thinning (GLM, p > 0.05).

FIGURE 5 | Slope of the regressions calculated during the four years before and after spruce budworm outbreak and thinning. Box plots represent the median, drawn as a horizontal solid line, with upper and lower quartiles and whiskers achieving the 10th and 90th percentiles.

# DISCUSSION

#### Thinning and Spruce Budworm Outbreak

Development and implementation of management strategies based on natural forest dynamics require the understanding of the ecosystem processes and their interactions with human activities (Gauthier et al., 2009). The effects of spruce budworm outbreak on stand dynamics in relationship with commercial thinning were previously approached by focussing on forest practices for reducing the damage produced by a successive outbreak of defoliators (Fuentealba and Bauce, 2012; Bauce and Fuentealba, 2013). A spruce budworm outbreak is currently producing dramatic growth reductions in trees that could also result in a diffuse mortality in several forested regions

of northeastern North-America (Bouchard and Pothier, 2010). Such a mortality could also affect composition and structure of the forest stands on wide landscape scale. There is thus an urgent need to assess how the surviving stands respond to the silvicultural practices applied during the successive endemic period. In order to do this and for the first time to our knowledge, this study investigated the influence of a spruce budworm outbreak on the growth of trees submitted to a thinning treatment. Our analysis used tree-ring chronologies from 34 black spruce stands in the boreal forest of Quebec, Canada. The results indicated that the effect of thinning on tree-ring width of black spruce was independent of the growth reduction that trees had experienced during the outbreak. On the other hand, the growth rate of individual trees measured before the last spruce budworm outbreak was linked to the reduction in growth during the defoliation period, as well as the growth recovery following commercial thinning. Our findings detected less sensitivity to the disturbance in terms of tree-ring width reduction of slow-growing trees. This meta-analysis used data from a number of sites from a wide forest region, involving an important diversity in terms of stand density, composition, soil, and severity of defoliation. Consequently, we attempted to extract the general and common information contained in the growth time series despite the variability expected from such a heterogeneous dataset.

The timings of the treatment could explain the lack of interaction between outbreak and thinning. The treatment was performed between 1995 and 1999, 17–21 years after 1978, the reference date estimated by our chronologies as the peak of the outbreak period. Defoliated trees need some years to recover and we assume that a complete recovery is attained when individuals reach a growth rate similar to that exhibited before the outbreak (Morin, 1998). On average, our measurements estimated that a sustained growth was attained after 2–7 years, long before the thinning was done. During an outbreak, the root system is strongly affected, with abrupt growth reductions, absence of root growth, and mortality (Krause and Morin, 1995, 1999). The development of adventive roots is often observed to increase the efficiency of water uptake (Krause, 2006). Similarly, the photosynthetic needle biomass returns to a higher level and photosynthesis rate increases (Piene and MacLean, 1999). The allocation of carbohydrates to the cambium allows tree-rings to be formed as large as before the defoliation period or even larger (Pothier et al., 2005).

#### Growth Reductions During Spruce Budworm Outbreak

Despite the marked growth reduction in the tree-ring chronologies during the outbreak, the growth response to the outbreak was not homogeneous among trees, as observed previously (Montoro Girona et al., 2017). Although most trees showed a clear contraction in growth, in some individuals the reduction in tree-ring width was marginal or lacking, confirming the literature (Vincent et al., 2009; Krause et al., 2012). Black spruce can be subjected to defoliation, but it is not the main host species of spruce budworm (Tremblay et al., 2011). It is likely that the different timings of larval emergence and bud break of black spruce prevent a continuous feeding by spruce budworm during the endemic periods and in part reduce the severity of the attacks during the outbreak (Hennigar et al., 2008; Colford-Gilks et al., 2012). Fuentealba and Bauce (2012) found a slower larval development on black spruce compared to the main host tree species. Black spruce is also known to develop smaller leaf biomass than the main host balsam fir (Nealis and Régnière, 2004). Pothier et al. (2012) observed that the growth reductions following moderate to severe defoliations were produced earlier in balsam fir than in black spruce. Moreover, the trees analyzed in this study were obviously only those that survived the outbreak, and which probably suffered less damage or better tolerated the effects of defoliation. The well-known capacity of adaptation and tolerance of black spruce to natural perturbations is also demonstrated by the low mortality after a spruce budworm outbreak (Colford-Gilks et al., 2012). This capacity of adaptation, in addition to the heterogeneous severity of the defoliation across stands could to some extent explain the variability observed in the growth reductions among trees.

The age of the trees plays an important role in the sensitivity of the stand to defoliation by spruce budworm, with mortality following spruce budworm outbreak increasing with stand age. Young individuals exhibit higher resistance than older trees to the damage caused by severe defoliations (Rossi et al., 2009a), because of their fast metabolism and greater capacity for enhancing photosynthesis (Boege, 2005; Mencuccini et al., 2005). According to our estimates, when the outbreak occurred in the studied sites, trees were between 30 and 40 years old, with the exception of two stands. This is only a minimum age calculated on complete cores. However, black spruce regenerates after fire within a few years, creating even-aged stands (Rossi et al., 2013). As a consequence, it is likely that trees in the same stand have the same or a similar

age, thus, our estimate could be considered to approximate the real stand age.

#### Growth Response to Thinning

Commercial thinning represent a practice alternative to the other traditional loggings with a lower impact on soil and canopy and are proposed as a suitable solution for maintain a heterogeneous pattern within and among stands (Gauthier et al., 2009). The effect of thinning on tree-ring width was expected, as previous studies have demonstrated the positive influence of the reduction in stand density and competition in improving black spruce growth (Soucy et al., 2012; Pamerleau-Couture et al., 2015; Lemay et al., 2017). Individuals with thinner tree-rings (called slowgrowing trees), had a smaller and more homogeneous response to the treatment in terms of growth rate than trees with higher annual radial growth (fast-growing). Vincent et al. (2009) found that the response to thinning was related to stem diameter as well as competition, with smaller individuals having the highest growth increments after treatment. This was explained by the fact that small or suppressed individuals were more advantaged by the reduction in competition than dominant trees. However, Vincent et al. (2009) calculated a partial R <sup>2</sup> of 0.042 for stem diameter, which was then supposed to affect only a minimal fraction of the growth response in trees. Thus, other, more specific approaches were proposed to disentangle and take into account the individual responses in growth within a stand (Montoro Girona et al., 2017).

The diameter was not related to the growth response after thinning in our analysis. On the other hand, growth rate did not correspond exactly to tree size. Growth rate changes with tree age: the classical growth trend is characterized by larger ring width close to the pith followed by a decrease in radial growth with age. This pattern is often not respected in forests with periodic spruce budworm outbreaks (Morin, 1998). The forest opening caused by tree mortality leads to gaps in the natural forest and the radial growth pattern can vary greatly between individuals (Rossi et al., 2009b; Montoro Girona et al., 2017). As spruce budworm has a cyclic recurrence every 30– 35 years, it is possible that the defoliation periods in the 1950s had already determined a separation of trees into the two growth classes (slow- and fast-growing). Overall, the growth rate in black spruce before a thinning event seems to be closely connected to the successive growth release of trees. The problem of the individual variability and homogeneity in growth within the stands still remain partially unknown and require further and deeper investigations (Pamerleau-Couture et al., 2015; Montoro Girona et al., 2017).

# REFERENCES


#### CONCLUSION

In this study, we measured and cross-dated tree-ring width in black spruce stands to answer the question if the spruce budworm outbreak of the 1970s affected the growth responses to the commercial thinning realized in 1990s. No relationship was found between spruce budworm outbreaks and changes in growth pattern after commercial thinning. If the timespan between the two disturbances is sufficient, more than 7 years, partial cutting can be applied without affecting the success of the growth release. However, the growth release after thinning seems to be related to the growth rate of trees, with the higher increases being concentrated in fast-growing individuals. Based on the results of this study, strategies of forest management should select black spruce stands with relatively high annual radial growth for thinning in order to optimize the volume growth of the residual trees.

## AUTHOR CONTRIBUTIONS

CK designed the study. P-YP performed the field and lab work. SR analyzed the data and wrote the first draft. CK and P-YP prepared the final version.

# FUNDING

This work was funded by grants from Consortium de Recherche sur la Forêt Boréale Commerciale and Fonds de Recherche sur la Nature et les Technologies du Québec (grant numbers: 2004-FB-101620, 2007-FO-118063, and 2009-FS-128240).

# ACKNOWLEDGMENTS

The authors thank M. Blackburn, M. Boulianne, A. Castro, C.-A. Déry Bouchard, P. Émond, É. Lapointe, S. Lapointe, A. Lemay, B. Lusczcynski, É. Pamerleau-Couture, G. Savard, and M. Vincent for technical support and A. Garside for editing the English text.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018.01061/ full#supplementary-material



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

The reviewer MG declared a past co-authorship with one of the authors SR to the handling Editor.

Copyright © 2018 Rossi, Plourde and Krause. 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.

# Conifer Regeneration After Experimental Shelterwood and Seed-Tree Treatments in Boreal Forests: Finding Silvicultural Alternatives

Miguel Montoro Girona1,2 \*, Jean-Martin Lussier <sup>3</sup> , Hubert Morin<sup>2</sup> and Nelson Thiffault <sup>3</sup>

<sup>1</sup> Ecology Restoration Group, Department of Wildlife, Fish and Environmental Studies, Swedish University of Agricultural Sciences (SLU), Umeå, Sweden, <sup>2</sup> Département des Sciences Fondamentales, Université du Québec à Chicoutimi, Saguenay, QC, Canada, <sup>3</sup> Canadian Wood Fibre Centre, Natural Resources Canada, Ottawa, QC, Canada

#### Edited by:

Giovanna Battipaglia, Università degli Studi della Campania "Luigi Vanvitelli" Caserta, Italy

#### Reviewed by:

Ugo Chiavetta, Consiglio per la Ricerca in Agricoltura E L'analisi dell'economia Agraria (CREA), Italy Carlo Urbinati, Università Politecnica delle Marche, Italy

\*Correspondence:

Miguel Montoro Girona miguel.montoro.girona@slu.se

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 26 May 2018 Accepted: 17 July 2018 Published: 17 August 2018

#### Citation:

Montoro Girona M, Lussier J-M, Morin H and Thiffault N (2018) Conifer Regeneration After Experimental Shelterwood and Seed-Tree Treatments in Boreal Forests: Finding Silvicultural Alternatives. Front. Plant Sci. 9:1145. doi: 10.3389/fpls.2018.01145 Forest regeneration is a key element in achieving sustainable forest management. Partial harvest methods have been used extensively in temperate broadleaf and mixedwood ecosystems to promote regeneration on poorly stocked sites and to maintain forest composition and productivity. However, their effectiveness in promoting conifer establishment has yet to be demonstrated in unmanaged boreal forests, especially those dominated by black spruce (Picea mariana (Mill.) BSP) where constraints for regeneration differ from those found in more meridional regions. We aimed to evaluate conifer seedling density and dimensions, 10 years after the onset of a gradient of silvicultural treatments varying in harvesting intensities, and to identify the critical factors driving the regeneration process. Study blocks of even-aged black spruce stands in the eastern Canadian boreal forest were submitted to three variants of shelterwood harvesting: a seed-tree harvest, a clear-cut and an untreated control. Shelterwood and seed-tree harvesting were combined with spot scarification to promote regeneration. Shelterwood and seed-tree harvesting produced a density of conifer regeneration sufficient to maintain forest productivity, but they did not promote seedling growth. Black spruce was the predominant species in terms of regeneration density, with proportions 3–5× higher than that for balsam fir (Abies balsamea (L.) Mill.). Ten years after treatment, seed-origin black spruce seedlings were abundant in skidding trails, while layers dominated the residual strips. Balsam fir density was not influenced by treatment nor by tree position relative to skidding trails. Balsam fir and black spruce had different responses to treatment in terms of height and diameter, the former exhibiting a better growth performance and larger diameter in the residual strips. Spot scarification created micro-sites that had a significant impact on the regeneration process. Overall, our results support that shelterwood and seed-tree harvesting combined with scarification enable adequate regeneration in black spruce stands, confirming these treatments as viable silvicultural alternatives to clear-cutting when required by sustainable forest management objectives.

Keywords: balsam fir, black spruce, ecosystem-based management, even-aged stands, partial cutting, seedling, shade-tolerant species, sustainable forest management

# INTRODUCTION

Seedling establishment is of crucial importance to the persistence, productivity and resilience of forest ecosystems; adequate regeneration is critical in the sustainable management of the boreal biome (Prévost, 1996; Gauthier et al., 2009). While clearcutting is the most widely-used harvesting method in North-America, it can lead to highly fragmented landscapes, declines in habitat diversity and losses of productivity (Groot et al., 2005; Fischer and Lindenmayer, 2007; Rosenvald and Lõhmus, 2008; Puettmann et al., 2015). Diversifying forestry practices and adapting silvicultural treatments are also necessary due to the pressures of climate change on forest ecosystems (Lindenmayer et al., 2012; Fedrowitz et al., 2014; Hof et al., 2017; Montoro Girona et al., 2018). The shelterwood and seed-tree systems are silvicultural alternatives to clear-cutting that can potentially address these concerns (Kern et al., 2017).

The shelterwood system favors the establishment and growth of regeneration through a uniform opening of the canopy, while limiting the growth of competing vegetation (Nyland, 2016). This system could be appropriate to ensure the regeneration of boreal conifers, maximize wood production and maintain biodiversity due to the high retention levels of forest stands (Vanha-Majamaa et al., 2007; Gauthier et al., 2009; Montoro Girona et al., 2017). Seed-tree harvesting, on the other hand, is a variant of clear-cutting that maintains only 5–30 seed-trees/ha, either in small groups or as dispersed individuals (Nyland, 2016). The remaining trees are chosen to provide sufficient seed sources after harvesting; the remaining cover is low, thus enabling light to reach the soil surface.

The shelterwood and seed-tree systems are potential options in current forest management strategies in Eastern Canada for two reasons: first, these systems may conciliate commercial harvest of timber and the maintenance of the key ecological attributes of mature stands, most important in ecosystem management strategies (Gauthier et al., 2009). Second, many jurisdictions rely on natural regeneration for the sustainable management of the boreal forest. Shelterwood or seed-tree systems can ensure and enhance regeneration, in particular in young and dense stands that often have marginal regeneration relative to current stocking standards (Oliver and Larson, 1996).

While the growth (Pamerleau-Couture et al., 2015; Montoro Girona et al., 2017) and mortality (Ruel et al., 2003; Anyomi and Ruel, 2015) of residual trees as well as the response of vegetation (Kneeshaw et al., 2002; Man et al., 2008) to partial harvesting have been documented for North American boreal forests, the factors contributing to the regeneration success remain unclear for most species under partial harvesting variants, including shelterwood and seed-tree cuts. In Eastern Canada, regeneration responses to shelterwood treatments have only been studied in silvicultural trials that lacked replication of treatments (Hatcher, 1961), were established in small, non-operational experimental designs (Raymond et al., 2000; Zarnovican et al., 2001) or were undertaken in mixedwood and deciduous stands (Tubbs, 1969; Metzger and Tubbs, 1971; Boivin, 1977). Moreover, whereas the regeneration of white spruce (Picea glauca [Moench] Voss) and balsam fir (Abies balsamea (L.) Mill.) has been documented in these contexts (Man and Lieffers, 1999; Beguin et al., 2009; Prévost and Gauthier, 2013), the establishment of black spruce (Picea mariana (Mill.) BSP) following partial harvesting and seed-tree cuts remains largely understudied (Kolabinski, 1991; Prévost, 1997) despite its ecological and economic importance (Giroud et al., 2016).

Ecological factors, such as soil characteristics and light availability, exert a marked influence on the establishment and growth of seedlings (Thiffault et al., 2015). Black spruce is a shade-tolerant species that mostly regenerates by layering from mature trees (>80%) in the absence of fire (Viereck and Johnston, 1990). However, seed-origin seedlings also contribute in maintaining productive stands (Lussier et al., 1992); and their establishment is highly dependent on the characteristics of the germination bed. For example, exposed mineral soil favors sexual regeneration success for this species (Jeglum, 1984; Prévost, 1997). Mechanical soil preparation through scarification following harvesting can improve seedbed receptivity, as does wildfire under natural disturbance dynamics (Raymond et al., 2000; Zarnovican et al., 2001; Hille and Den Ouden, 2004). Removal of the forest canopy, even partially, affects light levels in the understory, with consequences on the availability of other resources (Canham et al., 1990; Lieffers et al., 1999; Coates, 2000; Raymond et al., 2006). However, no studies have yet documented the effects of the modified light regime following mechanized shelterwood or seed-tree harvesting on the regeneration success in black spruce– dominated stands.

Hence, we aimed to evaluate 10 years of regeneration for evenaged natural black spruce stands in the boreal forest of Eastern Canada that were subjected to three experimental variants of mechanized shelterwood, seed-tree and clear-cut silvicultural systems. Our working hypotheses were that (i) the combined effect of partial shading from residual trees and scarification allows an adequate regeneration density in the short- to midterm for conifers, resulting in shelterwood variants and seed-tree methods that have a higher seedling density than clear-cutting when the former is combined with scarification; and (ii) seedling size is greater in seed-tree and clear-cut harvests relative to the shelterwood treatments, due to the high harvest intensity that increases light availability in the understory.

#### MATERIALS AND METHODS

#### Study Area

We conducted this study in even-aged, naturally established black spruce stands located in the Monts-Valin and North Shore regions of Quebec, Canada. The study areas lie within two bioclimatic regions, namely the balsam fir–white birch (Betula papyrifera Marsh.) and the black spruce–feathermosses bioclimatic domains (Saucier et al., 2009; **Figure 1**). The climate is subhumid subpolar, with a short vegetation season of 140 days (Rossi et al., 2011). Annual mean temperature ranges from −2 to 1.5◦C and average annual precipitation ranges from 950 to 1,350 mm (Robitaille and Saucier, 1998). Surface deposits consist primarily of thick glacial tills, and rocky outcrops are frequent

at the top of steep slopes (Robitaille and Saucier, 1998). The predominant soil type is humo-ferric podzols.

#### Silvicultural Treatments

Five harvesting treatments were performed in 2003 and compared in a fully replicated experimental design: ministrip harvesting (MS), distant selection cutting (DS), close selection cutting (CS), seed-tree harvesting (ST), and clearcutting (CC) with protection of advanced regeneration. The first three treatments are variants of a uniform shelterwood system (Montoro Girona et al., 2016). The main differences between the studied treatments were the spatial distribution of skidding trails and characteristics of the residual strips (**Table 1**). Harvest intensity, expressed in terms of proportion of basal area removed during the first harvest, was 50% for each shelterwood variant, 75% in ST and 100% for CC. MS consisted of a succession of 5 m-wide cut strips, with 5-m-wide residual strips. ST had wider cut strips (15 m) than MS, with 5-m-wide intact residual strips. In the case of CS and DS, trails were established at 20- and 30 m intervals, respectively, and the stand was partially harvested on each side of the trails, at a maximum distance of 5 m from the trail edge. DS had short secondary trails, perpendicular to the main operational trails. Each was separated by 10 m. In 2004, we designed and applied various patterns of soil scarification in each treatment (except in CC) within 2 m<sup>2</sup> rectangular plots (**Figure 2**). Scarification was performed using a 10-ton excavator equipped with a 1 m<sup>3</sup> bucket.

#### Experimental Design

We set up the experiment as a factorial design with completely randomized blocks. Six blocks were sampled, each one including six experimental units of 3 ha each, corresponding to one replicate of each silvicultural treatment and one untreated control plot (**Figure 2**). Two stand types were selected: three blocks were established in dense, relatively young forests (80–100 years, average density of 2,600 trees/ha), characterized by a low level of pre-established regeneration (average density of 2,600 trees/ha), and three blocks in open and relatively old forests (120–150 years), characterized by a high level of pre-established regenerated (average density of 1,500 trees/ha). In all cases, black spruce was the dominant species, accounting for at least 90% of the stand basal area (**Table S1**). Within each block, we situated the experimental units in areas that were relatively homogeneous in terms of species composition and stand density. We installed a permanent rectangular (10 × 60 m) sampling plot in the center of each experimental unit, perpendicular to the main skidding trails (**Figure 3a**). Two transects, each comprising 21 circular micro-plots (4 m<sup>2</sup> ), were established parallel to the sampling plots to study the regeneration response to the treatments (42 micro-plots × 6 treatments × 6 blocks = 1512 micro-plots). The sampling design was such that it covered the within-stand spatial heterogeneity of each harvesting treatment (**Figure 3b**). Overall, the experimental factors included the combinations of stand types (younger, older stands) and the silvicultural treatments (MS, DS, CS, ST, CC, Control).

#### Regeneration Assessment

Measurements were taken 1 year before cutting (b.c.) as well as 1 and 10 years after cutting (a.c.). We performed two inventories to study seedling density and growth. First, all seedlings in each micro-plot were counted and classified by species and height class (0–4.9, 5–29.9, 30–99 cm, and >1 m). Second, in each micro-plot, we selected one dominant seedling to evaluate the regeneration response 10 years after treatment. For each selected individual, we noted height, diameter, age (whorl count), origin (sexual or vegetative), rooting substrate (woody debris, mineral soil, dead wood and vegetation cover by stratum) and moss species found at the base. Other ancillary data were also collected at

#### TABLE 1 | Characteristics of the silvicultural treatments.


<sup>a</sup>Corresponds to the variability in the intervention as a consequence of secondary trails.

<sup>b</sup>Standard error is shown in parentheses.

the seedling level: micro-plot disturbance as the percentage of soil surface affected (in four classes: 0–25% undisturbed, 25–50% moderate, 50–75% high, and 75–100% very high) and the type of disturbance (rut, mound, scarification, windthrow or intact forest floor). The spatial position of micro-plots relative to the harvesting trails was also noted (strip, edge, and trail). Edge surface was considered as the area within 1.25 m on each side of the trails. We calculated seedling mortality due to the treatment as the difference in the number of seedlings b.c. and 1 year a.c. at the micro-plot level. Solar radiation was measured b.c. and 10 years a.c. as the percentage transmittance of photosynthetically active radiation (PAR; 400–700 nm) by leaving a quantum sensor and a data logger at 1-m above the ground, while another sensor was positioned in fully open conditions to measure the incident PAR as a control (Lieffers et al., 1999; Paquette et al., 2007). Two measurements of solar radiation were performed and averaged for each micro-plot (one parallel and one orthogonal to microplot orientation).

### Statistical Analyses

We conducted an analysis of variance (ANOVA) to evaluate treatment effects on the density of seedling regeneration by species, 10 years a.c. using the MIXED procedure of SAS 9.2 (SAS Institute, Inc., Cary, NC, USA). The model included blocks as a random effect, and stand type, treatment and their pairwise interactions as fixed effects. Natural logarithmic transformation of density values was used to satisfy the assumptions of normality and homogeneity of variance. We used the SLICE statement of the MIXED procedure to partition analyses of the LS-means in the case of significant interactions (p < 0.05). The same model was used to evaluate treatment effects on stocking and seedling size variables (height, diameter and age). Stocking was defined as the proportion of regeneration plots with a least one living seedling. The observed stocking was compared with the expected stocking assuming a random dispersion of seedlings in the stand. We assumed that the distribution frequency of the number of seedling per 4 m<sup>2</sup> plot followed a Poisson distribution.

As such, the expected probability of having at least one seedling per plot, considering the average density of seedlings, was equal to:

$$P\left(n>0\right) = 1 - e^{-average\ density}$$

Principal component analyses (PCA) were conducted to elucidate the relationships between the most influential factors on regeneration density and size, for both black spruce and balsam fir. We ran PCA with the FACTOR procedure of SAS 9.2 (SAS Institute, Inc., Cary, NC, USA) using all the variables collected at the micro-plot level.

#### RESULTS

# Density, Stocking, Age and Size of Seedlings

Black spruce was the dominant regenerating species a.c., with seedling densities that were three to five times higher than that of balsam fir, 1 and 10 years a.c., respectively. Black spruce regeneration density 10 years a.c. was significantly affected by treatments in interaction with stand types (**Figure 4**). Younger stands exhibited seedling densities that were at least 3 times higher a.c. compared to b.c., whereas older stands showed only slightly higher values 10 years a.c. than b.c. (**Table 2**). In both stand types, black spruce seedling density increased significantly in the scarified shelterwood and ST plots compared to control and clear-cut plots (**Figure 4**; **Table 3**). MS was the most effective treatment to promote black spruce regeneration in both stand types. The treatment × stand type interaction was caused by significant differences in CC. In younger stands, CC showed the lowest density of seedlings 10 years a.c. (even with the control), while in older stands, the values were similar to the other studied treatments, with the exception of MS (**Figure 4**). For balsam fir, stand type and treatments did not significantly influence seedling density (**Table 3**; **Figure 4**). However, some trends were observed, as balsam fir density was two times higher in older stands than in younger stands a.c. Shelterwood and ST followed by scarification induced 35% mortality of pre-established regeneration at 1-year a.c.

Initial black spruce stocking values were 52 and 84% for younger and older stands, respectively, with lower values for balsam fir (21 and 38%, respectively, **Table S2**). Stocking, 10 years after the shelterwood and ST treatments combined with scarification, ranged from 71 to 94%. In younger stands, black spruce stocking a.c. was, in most cases, 40% higher than b.c., whereas in older stands, stocking was 10% higher a.c. than b.c. For balsam fir, we observed similar stocking levels a.c. and b.c. In all cases, the observed stocking was lower than the expected stocking. Hardwood stocking 10 years a.c. was 10% higher than b.c. levels for both stand types (**Table S1**). CC and ST had the highest stocking values, at 56 and 71%, respectively, while the shelterwood variants with scarification showed lower values (43–56%).

The growth characteristics of seedlings showed that balsam fir and black spruce had different responses to treatments in terms of height and diameter, 10 years a.c. (**Figure 5**). In that regard, both species were influenced by the treatment effect; although in the case of black spruce, regeneration significantly responded to stand type as well (**Tables 4**, **5**). For black spruce, mean height ranged from 30 to 86 cm and diameter varied from 4 to 21 mm. Black spruce had higher values for height and diameter (1.4–2.8× higher) in older stands than in younger stands for shelterwoods with scarification and control plots, while responses to stand types were similar in CC and ST with scarification (**Table 4**). Age of black spruce seedlings ranged from 4 to 6 years, with no difference among treatments (p = 0.85). However, seedling age was significantly higher in older stands than in younger stands (p < 0.02). Balsam fir seedlings ranged from 34 to 197 cm in height and 9–50 mm in diameter (**Figure 5**). Treatments had a significant effect on balsam fir height and diameter, with higher values for treated plots than the controls; however, we detected no significant differences between stand types and their interaction (**Table 5**). Seedling age was higher for balsam fir (from 6 to 9 years), and it was similar between stand types and treatments, although being slightly higher in older stands. We detected no significant relationship between available insolation and seedling density or size for both species (**Figure S1**).

Regeneration distribution in the various height classes differed among stand types, treatments and species, and in most cases showed unimodal distributions (**Figure S2**). The height of black spruce seedlings was less variable among stand types and treatments than the height of balsam fir seedlings. Most black spruce seedlings 10 years a.c. were found in the 5– 29 cm height class, with the exception of CC and ST, in which regeneration seedlings were taller (30–99 cm). Overall, seedlings were smaller in shelterwood variants than in the control. The height distribution of balsam fir differed, especially in older stands, where we observed two dominant height classes (5–29 cm; 30–99 cm). Ten years after the experimental shelterwood cutting and for both stand types, most of balsam fir seedlings were between 5 and 29 cm (as in the control plots). However, DS had a greater variability in height and taller seedlings in older stands than in younger stands. For CC (in both stand types) and ST (in older stands only), the majority of balsam fir seedlings were taller than those under


#### TABLE 3 | Analysis of variance (ANOVA) results for seedling density 10 years after cutting, by seedling species.


The analysis assumed a mixed model in which the fixed effects were the two stand types and the five cutting treatments plus a control. MS, mini-strip; CS, close selection; DS, distant selection; ST, seed-tree and CC, clear-cutting. Only orthogonal contrasts are shown. Significant values are shown in bold (p < 0.05).

shelterwood and control conditions, ranging from 30 to 99 cm in height.

matter) or vegetative saplings on residual strips (moss cover and undisturbed micro-plots) (**Figure 6**).

### Multivariate Analyses of the Regeneration Process

Black spruce accounted for 85% of the measured seedlings. In shelterwoods and ST (having scarification after the treatment), seedlings originated mostly from sexual reproduction, although differences were detected between younger (proportion of sexual/asexual: 51–76%) and older stands (29–63%). Layering was the main origin of seedlings in the control and CC plots, accounting for 88–99% and 66–86% of regeneration, respectively. Seedlings originating from sexual reproduction were mainly located in trails, whereas vegetative layers were mostly found in the residual strips. Polytrichium sp. was the dominant moss species observed in trails, while residual strips were mostly covered by Pleurozium sp. and Hylocomium sp. Trails were characterized by a high disturbance level; most of their surface (50–100%) was dominated by mineral soil exposed from the scarification treatments, as well as mounds, ruts and woody debris. The soil in the residual strips was mostly undisturbed (intact forest floor).

PCA showed that position relative to the trails influenced the regeneration process for both black spruce and balsam fir. The analyses identified a common group of micro-plot factors related to either sexual seedlings on trails (rounds, mounds and high disturbance level), edges (windthrow and woody

For black spruce regeneration density, the first two axes of the PCA explained 51% of the variance (**Figure 6**, top-left). Scarification, position and ruts correlated strongly with the regeneration density of this species. PCA revealed that black spruce seedling density was highly correlated with scarification. The first two axes of the PCA explained 54% of balsam fir regeneration density. However, correlations were lower than for black spruce, as indicated by a short vector for this variable on the first axis and no correlation with the second axis (**Figure 6**, bottom-left). Balsam fir regeneration was mainly located on intact residual strips.

PCA, using seedling height, diameter and micro-plot characteristics, explained about 43% of the variation for black spruce and balsam fir seedlings and revealed differences between species (**Figure 6**, top-right and bottom-right, respectively). Black spruce seedlings were separated into two groups, one associated with a vegetative origin in residual strips in older stands, the other associated with a regeneration having a sexual origin in the trails of younger stands. Vegetative regeneration was associated with tall seedlings, large diameters and undisturbed seed-beds covered with moss. Sexual regeneration was associated with harvest trails having an exposed mineral soil and Pleurozium sp. Most pre-established balsam fir seedlings were in the residual strips, characterized by undisturbed micro-plots, although post-cutting seedlings were also identified in the trails, especially on mounds having a high amount of woody debris.

# DISCUSSION

Understanding and quantifying regeneration responses after harvesting in general, and partial cutting in particular, is essential to identify optimal forest management strategies that can support ecosystem-based management objectives (Messier et al., 1999). Our study confirms, for the first time, that experimental shelterwood and seed-tree harvesting followed by scarification allow the establishment of an abundant black spruce regeneration in North American boreal forests. In combination with soil scarification, maintaining residual trees in shelterwoods and seed-tree treatments improved regeneration density and stocking compared to control plots, particularly in younger and denser stands. Mini-strip harvesting followed by scarification resulted in the highest seedling density among the treatments, in conjunction with the highest density of scarified micro-plots (a result of the highest proportion of skidding trails). However, major differences were observed between the prescribed and the obtained density of scarified spots, the resulting density being 12–49% lower than expected. Residual trees and dense patches of advanced regeneration prevented the creation of more scarified micro-sites. Control plots and clear-cuts were not scarified and had the lowest level of black spruce regeneration densities; values were in line with those observed in previous studies conducted in boreal ecosystems (Harvey and Brais, 2002; MacDonald and Thompson, 2003). Clear-cutting induced little damage to advanced regeneration, as it was performed following careful logging practices, limiting soil disturbance, which in turn restricted regeneration establishment (Greene et al., 1999). The lack of soil disturbance is not the only factor that can explain the low densities found in clearcut plots. The limited range of seed dispersal also played a TABLE 4 | Analysis of variance (ANOVA) results for black spruce seedling characteristics.


The analysis assumed a mixed model in which the fixed effects were the two stand types, five cutting treatments plus a control and their interaction. MS, mini-strip; CS, close selection; DS, distant selection; ST, seed-tree; and CC, clear-cutting. Only significant orthogonal contrasts are shown. Significant values are shown in bold (p < 0.05).

TABLE 5 | Analysis of variance (ANOVA) results for the seedling characteristics of balsam fir.


The analysis assumed a mixed model in which the fixed effects are the two stand types, five cutting treatments plus a control and its interaction. MS, mini-strip; CS, close selection; DS, distant selection; ST, seed-tree; and CC, clear-cutting. Only significant orthogonal contrasts are shown. Significant values are shown in bold (p < 0.05).

role. Prévost (1997) observed that stocking decreases linearly with distance from seed-tree groups, up to 50 m in adjacent stands (2.5–3× farther than stand height). Thus, shelterwood variants and seed-tree harvesting combined with scarification were likely more effective than clear-cutting in promoting sexual regeneration because of the reduced distance from seed sources, increased soil humidity, decreased maximum air temperature and minimized frost occurrences and severity (Man and Lieffers, 1999; MacDonald and Thompson, 2003). Soil scarification may have similar effects in clear-cuts than in alternative treatments, but only in a limited distance from seed trees from the uncut forest border. The treatments have been designed to promote natural regeneration. As seed dispersion in black spruce stands depends on the distance of residual forests, seedling establishment in clear-cuts will be dependent on harvesting block size. Otherwise, we would expect only marginal establishment of regeneration in this case. We also hypothesize that spot scarification reduced the direct competition for light by feathermosses, which have a similar height to black spruce germinants.

Balsam fir characterized the secondary regeneration species, 10 years after partial cutting, although the species was more abundant than in the original stand (which was >95% black spruce). This is an important issue as after silvicultural intervention in black spruce stands, species composition can be altered and balsam fir can become the dominant species in these stands. This phenomenon increases the vulnerability of stands to spruce budworm, the most important defoliator in eastern Canadian forests (Maclean, 1996; Montoro Girona et al., 2018). Based on our results, this change in species composition after treatment was not enough to significantly increase stand vulnerability. Density of balsam fir was not influenced by stand type, treatment nor spatial position. This suggests that the pretreatment light and soil conditions may not have been factors limiting the establishment of balsam fir, but rather it was the density and distribution of seed-trees in these spruce-dominated

FIGURE 6 | Principal component analysis of micro-plot variables, seedling density and seedling size for black spruce and balsam fir, respectively. Each point represents a micro-plot. The proportion of the explained variance is indicated for each axis.

stands. However, balsam fir density after CC in older stands was similar to that reported by Harvey and Brais (2002), whereas in younger stands, density was two times lower. Shelterwood and ST systems modified the light regime in both space and time (by harvesting operations and the windthrow that followed), but it did not favor the establishment of shade intolerant competitors (i.e., deciduous species).

The growth of regenerated black spruce is slow in comparison with other conifer species, such as jack pine (Pinus banksiana Lamb.), tamarack [Larix laricina (Du Roi) K. Koch] (Thiffault et al., 2004), and balsam fir (Doucet and Boily, 1995; Messier et al., 1999). Ten years after treatment, black spruce height was similar to that reported in other studies (Harvey and Brais, 2002; Thiffault et al., 2004; Renard et al., 2016). Mean annual growth was 8.3 cm·yr−<sup>1</sup> in shelterwoods and 9.5 cm·yr−<sup>1</sup> in CC and ST. These growth rates were higher than those observed by Harvey and Brais (2002) (6.1 cm·yr−<sup>1</sup> ), but lower than the 15 cm·yr−<sup>1</sup> reported by Boily and Doucet (1993), 7–8 years a.c. The dominant seedling height classes 10 years a.c. corresponded to those that Riopel et al. (2011) observed at 5 years a.c. This was likely due to the stand not being even-aged, and the regeneration being mainly composed of pre-established layers in the latter study. CC and ST followed by scarification showed the best growth performance for black spruce and balsam fir. This is in agreement with MacDonald and Thompson (2003) who noted that the height and diameter of planted conifers increased with harvest intensity in a boreal mixedwood forest. We observed that shelterwood with scarification was not the most efficient treatment to promote seedling growth 10 years a.c. Growth performance increased with harvest intensity; differences were more than 2 cm·yr−<sup>1</sup> between ST-CC and experimental shelterwoods.

For many species, natural regeneration by seed depends on the receptivity of germination beds (Galipeau et al., 1997; Hille and Den Ouden, 2004). Even-aged black spruce stands are derived mostly from seeds that germinate after fire (Greene et al., 1999; Gagnon and Morin, 2001), whereas advanced regeneration is largely dominated by layers. Our results show that all treatments that removed the soil organic layer produced a higher seedling density than those maintaining a high proportion of undisturbed forest floor (Hille and Den Ouden (2004). Scarification, in combination with micro-plot position relative to the intact strips and trails, favored black spruce germination by exposing the mineral soil and providing lateral shadow from residual strips (Messier et al., 1999). Black spruce regeneration was mostly concentrated in the trails, where the removal of the organic layer likely increased water availability and decreased early competition from moss. Our results showed that scarification was essential for achieving the satisfactory establishment of black spruce regeneration (Prévost (1996), an effect also observed with other species (Nilsson et al., 2002; Hille and Den Ouden, 2004).

Spatial position played an important role in the distribution of species and the type of regeneration. Black spruce seedlings located in the scarified trails were mostly of sexual origin, while those found in the residual strips were mostly of asexual layering. Trails had a black spruce density six times greater than for residual strips without any soil preparation, a result matching observations in white spruce stands (Solarik et al. (2010). Without proper site preparation, regeneration in trails can be lower than in residual strips (Riopel et al., 2011). Balsam fir, on the other hand, has larger seeds than black spruce and can successfully grow roots and survive in undisturbed humus layers (Greene and Johnson, 1998). Hence, balsam fir regeneration was mostly located in the residual strips (Harvey and Brais, 2002; Riopel et al., 2011). Black spruce and balsam fir also have different regeneration responses in terms of size and density (**Figure 5**). Under shade conditions, suppressed balsam fir modify their crown architecture and favor a lateral expansion at the expense of vertical growth, hence producing high survival rates in residual strips (Messier et al., 1999).

Growth and survival of seedlings under shady conditions involves the complex interaction between the plant and resources, such as light, nutrients and water availability (Messier et al., 1999). Several studies have demonstrated than black spruce layers surviving in the understory for more than 100 years is common (Morin and Gagnon, 1991), and that the age of balsam fir saplings in the understory has been substantially underestimated as this species can have up to 40 missing rings (Morin and Laprise, 1997). Consequently, residual strip regeneration was the most challenging to correlate with environmental changes induced by silvicultural treatments for both conifer species (**Figure 6**).

Black spruce and balsam fir regeneration was not related to insolation levels during the first 10 years a.c. This can be explained by the intermediate levels of harvest intensity applied in our study, levels not severe enough to promote high light levels. However, we expect that light availability, as influenced by competing vegetation (i.e., deciduous species) will influence seedling growth and survival in the coming years. Long-term monitoring will be necessary to verify the impacts of insolation and deciduous species' competition on seedling growth.

Organic matter, sphagnum moss and mineral soil are the ideal seed-beds for establishing black spruce as they promote high seedling survival and density (Zasada et al., 1992; Duchesne and Sirois, 1995; Prévost, 1997; Raymond et al., 2000). Black spruce seedlings were mostly found in concave micro-sites created by scarification treatments, similar to the observations of Filion and Morin (1996). Depressions in the soil may favor higher seedling densities as germinants can benefit from runoff water pooling in the depressions. Furthermore, seeds can be washed down into and accumulate in depressions by heavy rains. However, the concave micro-topography of scarified plots can also reduce seedling survival by favoring excessive water accumulation and anaerobic conditions when there is poor drainage. We did not specifically assess seedling mortality resulting from flooding, predation or harvesting operations. Further investigation of the shelterwood and seed-tree systems in these ecosystems should take this into account (Frisque et al., 1978; Côté et al., 2003). Nevertheless, our findings indicate that seedling mortality a.c. was low overall, hence enabling the establishment of an abundant regeneration layer.

Previous research has shown that in black spruce stands the shelterwood system results in a significant growth response of residual trees and low mortality due to post-cutting windthrow (Montoro Girona et al., 2016, 2017). Here, we show that shelterwoods and ST harvesting followed by scarification result in an acceptable stocking and proportion of black spruce regeneration. Regeneration standards usually require that postharvest stocking be equal to or greater than the stocking of the harvested stand (based on 4 m<sup>2</sup> plots). In our study, the estimated stocking of the original stands ranged from 0.61–0.69 for the young and dense forests, and from 0.35 to 0.55 for the old and open ones. After 10 years, almost all treatments, including clear-cutting, resulted in a stocking equal or greater than 0.90 for both stand types. Hence, the shelterwood system followed by scarification did not increase the abundance of regeneration, compared to CC. However, the shelterwood treatments and scarification had a significant impact on the composition of the regeneration layer, as it increased the proportion of black spruce over balsam fir. The increased black spruce/balsam fir ratio is desirable, as it positively affects stand resilience to natural disturbances, such as fire and spruce budworm outbreaks, and preserves the economic value of future harvests.

Mini-strip shelterwood was the most efficient treatment for promoting black spruce regeneration. This harvesting variant had the highest proportion of trail surface per hectare and was the least expensive to implement due to the lack of tree selection. Our results demonstrate the importance of combining soil disturbance with partial cutting to create adequate seed-beds for black spruce. We expect that comparable results could be achieved at an even lower cost using disk trenching rather than spot scarification. Therefore, to evaluate these new silvicultural treatments in context of sustainable forest management objectives future research will be essential to determine the implantation costs and the biodiversity implications.

#### CONCLUSIONS

Ensuring regeneration for adequate density and growth of conifers is one of the most challenging issues of boreal forest management. Our study provides a better understanding of the regeneration process in black spruce–dominated stands. We demonstrated that the experimental shelterwood and seed-tree systems followed by scarification are effective treatments for promoting regeneration in spruce-moss forest ecosystems. Black spruce regeneration was favored over balsam fir regeneration. The highest seedling densities were observed in the experimental shelterwood and seed-tree treatments. Soil disturbances were a key factor in the establishment success of black spruce, and insolation did not influence seedling density and growth, 10 years after cutting. Shelterwood and seed-tree systems followed by scarification enable an adequate regeneration in black spruce stands, confirming these treatments as viable silvicultural alternatives to clear-cutting when required by sustainable forest management objectives.

#### AUTHOR CONTRIBUTIONS

MM and NT: conceptualization; MM: data curation and fieldwork; MM: formal analysis; MM, J-ML, NT, and HM: investigation; MM, J-ML, NT, and HM: methodology; MM: project administration; HM: resources; HM: supervision; MM, NT, and J-ML: validation; MM: visualization and edition; MM: writing–original draft; MM, NT, J-ML, and HM: writing–review: HM, J-ML, and MM: funding.

### FUNDING

This project was funded by the Fonds de Recherche du Québec – Nature et Technologies (FQRNT), the Programme

#### REFERENCES


de mise en valeur des ressources forestières (MFFPQ), the Canadian Wood Fibre Centre of the Canadian Forest Service (Natural Resources Canada) and the Forest Complexity Modeling Program of the Centre for Forest Research.

#### ACKNOWLEDGMENTS

We thank R. Gagnon, E. Dussault-Chouinard, G. Grosbois, J. P. Girard, and A. Lemay for fieldwork assistance and D. Walsh for statistical advice and validation. We give special thanks to L. De Grandpré, A. Leduc, S. Rossi, C. Krause, A. Hof, and J. Hjältén for their suggestions on an earlier version of this manuscript. We also thank C. Gosselin and F. Marchand for their support. This manuscript is part of the Ph.D. thesis of MM (Montoro Girona, 2017).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018. 01145/full#supplementary-material

Figure S1 | Relationships between insolation and density of black spruce seedlings, 10 years after mini-strip shelterwood harvesting across a transect of plots, by stand type. PAR represents the percent transmittance of photosynthetically active radiation.

Figure S2 | Seedling and sapling height-class distributions, 10 years after treatment by stand type and species; numbers from 1 to 4 represent each height class: (1) 0–4.9 cm; (2) 5–29.9 cm; (3) 30–99 cm; (4) >1 m.

Table S1 | Initial stand characteristics by stand type for each study block. Data were obtained in permanent rectangular (10 × 60 m) sampling plots established in the center of the experimental units. Sampling covered the spatial heterogeneity of each silvicultural treatment (trails, edge and residual strip). Measurements were taken in 2002, one year before cutting (b.c.), on trees having a diameter at 1.3 m (DBH) of ≥9 cm for all tree species (n = 3,739). Stocking is the one expected from a random dispersion of trees, based on a Poisson distribution and the average tree density in 4 m<sup>2</sup> plots.

Table S2 | Stocking of main conifers species and hardwood regeneration, 10 years after cutting (mean ± standard error). Expected values assume the random distribution of seedlings and a density distribution following a Poisson distribution. Paper birch (Betula papyrifera Marsh) and aspen (Populus tremuloides Michx) were grouped as intolerant hardwoods.


**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 Montoro Girona, Lussier, Morin and Thiffault. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Post-drought Resilience After Forest Die-Off: Shifts in Regeneration, Composition, Growth and Productivity

Antonio Gazol<sup>1</sup> , J. Julio Camarero<sup>1</sup> \*, Gabriel Sangüesa-Barreda1,2 and Sergio M. Vicente-Serrano<sup>1</sup>

1 Instituto Pirenaico de Ecología (IPE-CSIC), Zaragoza, Spain, <sup>2</sup> Departamento de Ciencias Agroforestales, EU de Ingenierías Agrarias, Universidad de Valladolid, Soria, Spain

#### Edited by:

Giovanna Battipaglia, Università degli Studi della Campania "Luigi Vanvitelli" Caserta, Italy

#### Reviewed by:

Martina Pollastrini, Università degli Studi di Firenze, Italy Peter Prislan, Slovenian Forestry Institute, Slovenia Angela Luisa Prendin, Università degli Studi di Padova, Italy

> \*Correspondence: J. Julio Camarero jjcamarero@ipe.csic.es

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 25 June 2018 Accepted: 03 October 2018 Published: 25 October 2018

#### Citation:

Gazol A, Camarero JJ, Sangüesa-Barreda G and Vicente-Serrano SM (2018) Post-drought Resilience After Forest Die-Off: Shifts in Regeneration, Composition, Growth and Productivity. Front. Plant Sci. 9:1546. doi: 10.3389/fpls.2018.01546 A better understanding on the consequences of drought on forests can be reached by paying special attention to their resilience capacity, i.e., the ability to return to a state similar to pre-drought conditions. Nevertheless, extreme droughts may surpass the threshold for the resilience capacity triggering die-off causing multiple changes at varying spatial and temporal scales and affecting diverse processes (tree growth and regeneration, ecosystem productivity). Combining several methodological tools allows reaching a comprehensive characterization of post-drought forest resilience. We evaluated the changes in the abundance, regeneration capacity (seedling abundance), and radial growth (annual tree rings) of the main tree species. We also assessed if drought-induced reductions in growth and regeneration of the dominant tree species scale-up to drops in vegetation productivity by using the Normalized Difference Vegetation Index (NDVI). We studied two conifer forests located in north-eastern Spain which displayed drought-induced die-off during the last decades: a Scots pine (Pinus sylvestris) forest under continental Mediterranean conditions and a Silver fir (Abies alba) forest under more temperate conditions. We found a strong negative impact of a recent severe drought (2012) on Scots pine growth, whereas the coexisting Juniperus thurifera showed positive trends in basal area increment (0.02 ± 0.003 cm<sup>2</sup> yr−<sup>1</sup> ). No Scots pine recruitment was observed in sites with intense die-off, but J. thurifera and Quercus ilex recruited. The 2012 drought event translated into a strong NDVI reduction (32% lower than the 1982–2014 average). In Silver fir we found a negative impact of the 2012 drought on short-term radial growth, whilst long-term growth of Silver fir and the coexisting Fagus sylvatica showed positive trends. Growth rates were higher in F. sylvatica (0.04 ± 0.003 cm<sup>2</sup> yr−<sup>1</sup> ) than in A. alba (0.02 ± 0.004 cm<sup>2</sup> yr−<sup>1</sup> ). These two species recruited beneath declining and non-declining Silver fir trees. The 2012 drought translated into a strong NDVI reduction which lasted until 2013. The results presented here suggest two different post-drought vegetation pathways. In the Scots pine forest, the higher growth and recruitment rates of J. thurifera correspond to a vegetation shift where Scots pine is being replaced by the drought-tolerant juniper. Conversely, in the Silver fir forest there is an increase of F. sylvatica growth and abundance but no local extinction of the Silver fir. Further research is required to monitor the evolution of these forests in the forthcoming years to illustrate the cumulative impacts of drought on successional dynamics.

Keywords: dendroecology, drought stress, Normalized Difference Vegetation Index, resilience, tree recruitment

# INTRODUCTION

fpls-09-01546 October 23, 2018 Time: 14:26 # 2

A significant increase in air temperature has been recorder from the early 1980s to the 2010s over Southern Europe, particularly affecting the Mediterranean area (IPCC, 2014; Spinoni et al., 2017). These drying trends have been more marked from winter to summer, i.e., encompassing most of the growing season of the affected species in spring and thus reducing forest productivity and tree vitality (Ciais et al., 2005). In drought-prone areas from the Mediterranean region, some forests are showing dieoff events and increased mortality rates as a result of warmer conditions, increased vapor pressure deficit, and more frequent or longer dry spells (Sarris et al., 2007; Linares and Camarero, 2012; Sánchez-Salguero et al., 2012; Camarero et al., 2015). Such chronic and rapid climate-driven disturbances may lead to no-analog situations causing forest disequilibrium (Anderegg et al., 2012), which requires efforts to assess post-drought resilience capacity (Gazol et al., 2018b), and also to better predict die-off impacts on forest dynamics and associated ecosystem services (Anderegg et al., 2016; McDowell et al., 2018). Thus, it is necessary to identify forest states preceding and following drought-driven die-off and to characterize the potentially nonlinear transitions separating those states (Cobb et al., 2017).

Drought-triggered die-off usually induces higher tree mortality rates which can trigger vegetation shifts and accelerate successional dynamics (Breshears et al., 2005; Suarez and Kitzberger, 2008; Suarez and Sasal, 2012). Two extreme scenarios after drought-induced mortality can be expected: (1) the persistence of the previously dominant tree species through the survival of most of its adults or their regeneration from the seed and/or seedling bank (Suarez and Lloret, 2018); or (2) the replacement of less drought-resistant species by more resistant species including changes in the functional types prevailing at the community (e.g., shifts from a forest to a scrubland or from gymnosperms to angiosperms; see Gazol et al., 2017c). In a recent review, almost half of the evaluated case studies showed no change in the dominant vegetation type indicating a high post-drought resilience of forests affected by die-off (Martínez-Vilalta and Lloret, 2016). Therefore, this lack of widespread drought-induced forests shifts can be explained by demographic compensation which improves resilience and allows forest structure and composition to be maintained (Lloret et al., 2012). For example, growth decline and increased tree mortality after a drought may be compensated by enhanced recruitment and growth improvement of survivors due to the release of competition for limiting resources such as light and water (Madrigal-González et al., 2017). Alternatively, droughtrelated die-off may be a too patchy or diffuse disturbance as compared with other more severe or widespread natural (e.g., fire, landslides) and human disturbances so as to trigger vegetation shifts (Anderegg et al., 2016).

Past legacies can also influence how die-off in droughtinduced forests is displayed. First, recurrent or more severe droughts could produce a cumulative effect by weakening trees, reducing their growth rates, rising mortality rates and diminishing forest post-drought resilience capacity (Bigler et al., 2006; Peltier et al., 2016; Gazol et al., 2018a). Second, in the case of historically managed ecosystems as Mediterranean forests, die-off could be predisposed by past use leading to a negative selection of slow-growing trees more vulnerable to recent droughts (Camarero et al., 2011; Sangüesa-Barreda et al., 2015). These historical influences require retrospective approaches and prospective monitoring of changes in composition, growth and mortality through time before and after forest die-off (e.g., Suarez and Lloret, 2018). In this sense, annual-growth rings provide a mid-term time perspective on how trees grow and respond to drought and satellite images could provide a picture of how a proxy of productivity and vegetation activity such as the Normalized Difference Vegetation Index (NDVI) responds to drought and potential changes in forest structure through time (Gazol et al., 2018b). Lastly, the combination of several methodological approaches to evaluate shifts in forest composition after drought-induced dieback further supports the novelty and highlights the importance of this study with respect to previous papers (e.g., Camarero et al., 2015). In this sense, multiple methodological tools may provide comprehensive characterizations of post-drought forest resilience. These complementary measures can help answering if drought-induced reductions in growth and regeneration of the dominant tree species scale-up to drops in vegetation productivity.

Here we evaluate the responses of different forest components to an extreme drought in two conifer forests subjected to contrasting climate conditions and displaying declining growth trends during the last decades. In particular, we study how forest productivity, tree growth and regeneration changed in the short (3 years) and long terms (30 years). The short-term response was evaluated after an extreme drought that affected most of the Iberian Peninsula in 2012 and triggered forest die-off. The longterm trend was evaluated by considering tree growth (annualgrowth rings) and ecosystem productivity (NDVI) over the last three decades (1980-present). We compared these responses in two forests located in north-eastern Spain, a Scots pine (Pinus sylvestris) forest subjected to continental and dry conditions and a Silver fir (Abies alba) forest subjected to temperate and wetter conditions. These two species displayed marked growth declines as a consequence of the occurrence of warming trends and related droughts during the last decades (Camarero et al., 2015). We aim to disentangle whether warming trends together with severe drought induced a change in forest composition, growth and productivity by accelerating a replacement of drought-vulnerable with drought-tolerant species.

### MATERIALS AND METHODS

fpls-09-01546 October 23, 2018 Time: 14:26 # 3

#### Study Sites and Tree Species

We studied a Scots pine (P. sylvestris L.) and a Silver fir (A. alba Mill.) forest, both located in Aragón, north-eastern Spain (**Table 1**). These two forests are situated near the southernmost distribution limit of both species in Europe. The Silver fir forest is located in the Spanish central Pyrenees (Paco Ezpela, Ansó) where temperate and wet conditions prevail (see **Table 1**). Here soils are basic (pH = 6.8), loamy Cambisols with a high proportion of lime (41%) and clay (13%; Gazol et al., 2018a). Despite this forest present signs that intense logging activity was undertaken in the past such as the presence of stumps and wood trails (Sangüesa-Barreda et al., 2015), most of these management activities ceased in the 1970s (Camarero et al., 2015). The Scots pine forest is located in the Iberian System (El Carrascal, Corbalán) and it is subjected to a continental Mediterranean climate. It presents acid (pH = 7.3) Cambisol soils with a relatively high proportion (76%) of sand (Gazol et al., 2018a). This forest presents no evident signs of human management during the last 50 years (Camarero et al., 2015).

Several investigations performed in these forests suggest that growth decline in the Silver fir forest started after the 1986 drought that affected the Pyrenees, whereas growth decline in the Scots pine forest started in the early 1980s as a consequence of a climate shift toward warmer and dryer conditions (Camarero et al., 2018). Moreover, the two study forests showed a marked growth decline and increased mortality as a consequence of the severe winter-spring drought that occurred in 2011–2012 (Trigo et al., 2013) and affected most Spain (Camarero et al., 2015). For many trees, tree growth decline resulted in rapid death (see more details in **Table 1**).

## Evaluation of Post-drought Tree Regeneration and Recruitment

In 2012, a total of 38 dominant trees were randomly selected and permanently marked with a balanced number of declining and non-declining individuals in each forest (Camarero et al., 2015). We measured the size (Dbh, diameter at breast height measured at 1.3 m; height) and estimated the crown transparency (defoliation degree) using binoculars of each tree (Dobbertin, 2005). Tree crown transparency, tree vigor and mortality were monitored in 2012 and annually from 2015 to 2017. Declining and non-declining individuals were defined as those showing crown transparency above or below the 50% threshold, respectively.

To account for the effects of local neighborhood on tree performance (Zambrano et al., 2017) the neighborhood (basal area) of each of the sampled trees was characterized in 2015. We measured the Dbh and identified the species of all stems higher than 1.3 m and located within a circular plot of radius 7.6 m centered on the sampled focal tree. A similar radius was used in the two forests to avoid changes due to vegetation structure and composition. These values were used to calculate the basal area and the density of the main species of woody plants in the community below declining or recently dead and non-declining trees. We also quantified the regeneration capacity of the different woody species by counting the number of recruits (seedlings and saplings with height < 0.25 m) located in the forest understory. Particularly, we quantified the presence of recruits from different species in 100 squares (size 25 cm × 25 cm) randomly placed in a representative area of the forest floor. These representative areas were selected to reflect the general conditions of the forest across an area of 0.1 ha, i.e., the number of squares was proportionally located according to the area with or without die-off signs (dead or declining trees showing high needle shedding). The evaluation of tree regeneration was repeated yearly from 2015 to 2017 in the same area of the two forests.

### Growth Data: Dendrochronological Methods

In order to quantify the radial-growth patterns and trends of the two dominant conifers and the most important co-dominant species, we sampled several trees using dendrochronological methods. All the sampled individuals were apparently healthy trees, with low crown defoliation, to avoid potential bias in growth as a consequence of the decline of the dominant tree species. Specifically, we sampled P. sylvestris (n = 35 trees) and Juniperus thurifera (n = 15 trees) in the Scots pine forest and A. alba (n = 36 trees) and Fagus sylvatica (n = 49 trees) in the Silver fir forest. The low number of sampled J. thurifera individuals was due to the low abundance of mature individuals of this species in the Scots pine forest (**Table 2**). Two increment cores were taken from each tree at 1.3 m and perpendicularly to the maximum slope using Pressler increment borers in late 2015.

These cores were mounted and air dried in the laboratory. After that, the cores were sanded with progressively finer sandpaper until tree rings were visually recognizable. Ring widths were measured at 0.01 mm resolution using the LINTAB measuring device (Rinntech, Heidelberg, Germany). Visual cross-dating was performed and checked using the program COFECHA (Holmes, 1983). Tree ring widths were transformed to basal area increment (BAIt) using the following formula:

$$\text{BAIt} = \left(\text{R}\_t^2 - \text{R}\_t - \text{}^2\right)^2$$

where R<sup>t</sup> is the radius of the ring formation year and Rt−<sup>1</sup> is the radius of the year preceding the ring formation. We also calculated the relative basal area increment as the ratio between the observed BAI divided by maximum BAI of each species to compare post-drought growth responses between species.

#### Remote-Sensing Data: NDVI

To quantify the temporal variability in aboveground forest gross-primary productivity in each forest we used 1.1 km<sup>2</sup> spatial resolution bi-weekly series of the Normalized Difference Vegetation Index (NDVI) over the period 1982–2014. NDVI was obtained from daily NOAA-AVHRR images from 1981 to 2015.

TABLE 1 | Characteristics of the studied Silver fir and Scots pine forests showing die-off and mortality after the 2012 drought.


Values are means ± SE. § Climate data for the Silver fir and Scots pine forests were taken from the Jaca (42◦ 34<sup>0</sup> N, 0◦ 33<sup>0</sup> W, 818 m, located at 30 km from the study forest) and Teruel (40◦ 21<sup>0</sup> N, 1◦ 06<sup>0</sup> W, 915 m, located at 15 km from the study forest) meteorological stations, respectively. #Based on the annual monitoring of vigor proxies (e.g., crown transparency, bud and needle production) and tree death in 35 adult trees of the main species at each site.

TABLE 2 | Values of basal area (in m<sup>2</sup> ha−<sup>1</sup> ; means ± SE) of the most abundant tree species for neighborhoods in declining or recently dead trees (die-off) or nondeclining trees (no die-off).


The measurements were taken in 2015 (3 years after the 2012 drought). The last column shows the comparison between variables based on t-tests with their corresponding P-values. Significant t-values (P < 0.05) are indicated with bold characters.

These images were subjected to a complete calibration, quality control, geometric matching, cloud removal and topographic correction (Hantson and Chuvieco, 2011). After the calculation of the NDVI, the daily images were aggregated into semimonthly composites by means of the Maximum Value Composite (MVC) technique. The quality and accuracy of this dataset was verified, given its strong temporal coherence in comparison to other available datasets at lower spatial resolution (Martín-Hernández et al., 2017). In this study we selected the two 1.1 km<sup>2</sup> pixels that corresponded to the two study forests.

#### Statistical Analyses

We compared the neighborhood below declining and nondeclining trees in each forest. Particularly, we tested whether some of the most abundant woody plant species were more or less abundant below declining and non-declining trees. A t-test was applied to study whether there were significant differences in the abundance of woody plants below declining and non-declining trees.

To compare tree regeneration below declining and nondeclining trees, we used the Cramér's V statistic (Agresti, 2013). This statistic is the square root of χ <sup>2</sup> divided by sample size (n)

times m, which is the smaller of r (number of rows − 1) or k (number of columns − 1):

$$V = \sqrt{\frac{\chi^2}{n \times m}}$$

Cramér's V ranges from 0 to 1 corresponding to dissociation (no association) or complete association between the compared variables, respectively (Agresti, 2013).

We assessed the short- and long-term growth trends of the dominant tree species in each forest (P. sylvestris and A. alba) and the two main co-dominant tree species (J. thurifera and F. sylvatica, respectively). In the short-term, we studied whether the radial growth (BAI) 3 years after the 2012 drought (period 2013–2015) differed from that observed during the drought and 3 years before (period 2009–2011). We selected the 3-year period based on previous studies showing that it adequately captured the post-drought response (Gazol et al., 2017a,b, 2018b). Comparisons of growth between these periods were performed using ANOVA followed by Tukey-HSD tests when significant differences in growth between the pre-drought and post-drought periods was observed.

In the long-term, we studied the BAI trends of the main tree species (P. sylvestris, J. thurifera, A. alba, and F. sylvatica) for the common period 1982–2015. We fitted linear mixed-effect models using BAI as response variable (Zuur et al., 2010). We run a separate model for each tree species (P. sylvestris, J. thurifera, F. sylvatica, and A. alba). To represent the linear growth trend, we included calendar year as a predictor variable; a positive effect of year indicates growth enhancement whereas a negative effect indicates growth decrease. Tree radial growth depends on tree ontogeny and Dbh (Gazol et al., 2017b). Therefore, we included the estimated tree age and the Dbh reconstructed back in time as covariates in the model. Tree age at 1.3 m was estimated by counting the number of rings in the oldest core. In cores without pith, pith-offset estimations were used to calculate the number of missing rings by fitting a geometric pith locator. Since the BAI of each tree represents repeated measures across an individual, we include tree identity as random factor. To control for the potential temporal autocorrelation, we also included a first-order autocorrelation structure (AR1). To linearize growth measures and achieve normality assumptions, we log transformed [log(x + 1)] BAI prior to the analyses (Zar, 2010).

To identify the set of covariates that better explained the observed BAI trends, we applied a multi-model inference approach based on information theory (Burnham and Anderson, 2002). We ranked all potential models according to the secondorder Akaike information criterion (AICc) and selected as the best model that showing the lowest AICc value and largest Akaike weight (which represents the relative probability that the selected model is the best one). The different competing models were ranked according to the 1AICc (AICc differences between the model selected and the rest of remaining models).

To compare the growth trends between species, we fitted linear mixed-effect models using BAI as response variable (Zuur et al., 2010) and including calendar year, tree age and dbh and species identity as explanatory variables. To account for different growth trajectories between species, we included an interaction between calendar year and species identity. We also considered potential interactions between tree age and species identity and Dbh and tree species. A multi-model inference approach based on information theory (Burnham and Anderson, 2002), was used to select the most parsimonious model (lowest AICc value and largest Akaike weight). When significant interactions between calendar year and species identity were found, we performed least-squares means based on and Tukey HSD tests to assess the differences between tree species (Bretz et al., 2010).

In order to detect changes in aboveground productivity along time in each forest, we modeled NDVI for the period 1982– 2014 using Generalized Additive Models (Wood, 2006). These are flexible semiparametric models that allow modeling the response variable as different smooth functions of a set of explanatory covariates. Clearly, the NDVI bi-weekly data contain a strong year cycle, but it is less clear whether it presents long term trends. Thus, we modeled how NDVI varies within and between years along the entire study period. To represent variations in NDVI biweekly data within a year, we used a cyclic cubic regression spline since a cyclic smooth is recommended for repeated patterns (Wood, 2006). To model the long-term variation in NDVI along the entire study period we used thin plate splines. A first-order correlation structure (AR1) was included to account for the temporal autocorrelation in NDVI.

All statistical analyses were performed in the R environment (R Core Team, 2017). The 'dplR' package was used to convert tree-ring width series into BAI (Bunn et al., 2016). The lme function of the 'nlme' package was used to fit the linear mixedeffects models (Pinheiro et al., 2014) and the 'MuMIn' package was used to perform the multi-model selection (Barton, 2012). The gam function of the 'mgcv' package was used to fit the generalized additive models (Wood, 2006).

## RESULTS

#### Mortality, Forest Structure and Post-drought Tree Recruitment

Mortality from 2012 to 2017 was higher and increased more in the Scots pine forest (mean ± SE = 52.0 ± 4.2%; mean rate, 22.0% yr−<sup>1</sup> ) than in the Silver fir forest (64.3 ± 19.3%; mean rate, 4.7% yr−<sup>1</sup> ; **Table 1**). We found a high abundance of P. sylvestris and A. alba individuals in the neighborhood of the focal P. sylvestris and A. alba trees sampled in each forest, respectively (**Table 2**). This means that the neighborhood of either declining or non-declining trees is mainly composed by conspecifics in both forests. Nevertheless, we found differences in the abundance of other species in the neighborhood of declining and non-declining individuals. In the P. sylvestris forest, Quercus ilex was significantly more abundant below declining than nondeclining trees. Similarly, F. sylvatica was more abundant below declining than below non-declining A. alba trees (**Figures 1A,B** and **Table 2**). In the P. sylvestris forest, J. thurifera was abundant trees.

fpls-09-01546 October 23, 2018 Time: 14:26 # 6

(**Figures 1C,D**), but its basal area did not differ between declining and non-declining trees.

Regarding tree recruitment, Cramér's V values indicated that regeneration below declining and non-declining trees differed in 2015 but tended to be more similar in 2017, particularly in the A. alba forest (**Figure 2** and **Table 3**). Overall, species recruited below declining and non-declining trees differed more in the P. sylvestris (mainly in 2016) than in the A. alba forest. The species most abundantly recruiting in the P. sylvestris forest were Q. ilex, J. thurifera, and Q. faginea, whilst Hedera helix, A. alba, and F. sylvatica were the main recruiting species in the A. alba forest (**Figure 2**). In 2015, no A. alba recruit was found below declining conspecific trees, whereas we did not observe any P. sylvestris recruit below conspecific declining trees neither in 2015 nor in 2016 (**Supplementary Figure S1**).

#### Radial Growth Patterns at Long- to Short-Term Scales

The radial growth (BAI) of the main tree species presented a marked decline as a consequence of the 2012 drought (**Figures 3**, **4**). All species showed marked growth reductions in response to the severe 2005 and 2012 droughts (**Figure 3**). A marked growth decline as a consequence of the 2012 drought was significant in all species excepting F. sylvatica. Long-term growth trends differed significantly between species (**Figure 3** and **Table 4**). In the P. sylvestris forest, the model for P. sylvestris BAI explained 56% of the variation in its growth trends and did not include the effect of year, indicating no clear growth trends. Conversely, the model for J. thurifera presented a significant positive trend (0.02 ± 0.003 cm<sup>2</sup> yr−<sup>1</sup> ) indicated by the significant effect of year (**Table 4**). This model also included a significant effect of tree dbh and accounted for 61% of the variation in J. thurifera BAI. In the A. alba forest, A. alba and F. sylvatica presented significant positive trends (significant influence of year, **Table 4**) but with a stepper slope in the case of F. sylvatica (0.04 ± 0.003 cm<sup>2</sup> yr−<sup>1</sup> ) than in A. alba (0.02 ± 0.004 cm<sup>2</sup> yr−<sup>1</sup> ). The models accounted for 49 and 76% of the variation in A. alba and F. sylvatica BAI, respectively. In all models excluding the F. sylvatica model, tree age exerted a negative influence on BAI trends, whereas tree Dbh was significantly and positively related to BAI trends for all species (**Table 4**). These results were consistent with those obtained when considering the four species in a single model (**Supplementary Figure S2**). The model showed the existence of significant effect of tree Dbh, tree age and calendar year and their interactions with tree species on BAI. This model, which accounted for 81% of the variation in BAI, indicated the existence of significantly different growth trajectories between species (**Supplementary Table S2**). The post hoc analyses showed that P. sylvestris presented growth rates lower than the rest of species. The growth rates of A. alba and J. thurifera did not differ between them. F. sylvatica presented the greatest growth rates.

The short-term responses of relative BAI to the 2012 drought showed different patterns for the growth of each species (**Figure 4**). In the Scots pine forest, P. sylvestris growth was significantly higher before (mean BAI was 504.8 ± 42.9 mm<sup>2</sup> ) than during (146.0 ± 16.4 mm<sup>2</sup> ) and after (314.1 ± 40.9 mm<sup>2</sup> ) the drought, and the post-drought growth was also higher than the growth during the drought year. The J. thurifera growth was also significantly higher after (172.8 ± 38.1 mm<sup>2</sup> ) than during (92.6 ± 13.8 mm<sup>2</sup> ) the drought and no differences in growth were detected before (135.7 ± 23.3 mm<sup>2</sup> ) and after the drought event. In the A. alba forest, A. alba growth was significantly lower during (947.0 ± 99.7 mm<sup>2</sup> ) and after the drought (1037.6 ± 125.6 mm<sup>2</sup> ) than before it (1246.2 ± 131.8 mm<sup>2</sup> ). No differences between F. sylvatica growth before (294.1 ± 50.8 mm<sup>2</sup> ), during (276.1 ± 48.5 mm<sup>2</sup> ) and after the drought (333.0 ± 50.8 mm<sup>2</sup> ) were observed.

#### Changes in Gross Primary Production as Reflected by the NDVI

In the P. sylvestris forest, the NDVI pattern was noisier than in the A. alba forest which showed a more marked seasonal pattern (**Figure 5** and **Supplementary Table S1**). The 2012 drought triggered a strong decrease in NDVI in the P. sylvestris forest, where predicted NDVI values were 22% higher than observed NDVI values (**Figure 5**), which were also 32% lower than the observed NDVI values for the entire period. In the A. alba forest, no clear differences between predicted and observed NDVI values were observed. Longterm NDVI trends suggest the existence of a lagged decline in

#### NDVI during 2013–2014 in the Silver fir forest (**Supplementary Figure S3**).

TABLE 3 | Cramér's V tests comparing the seedlings and saplings density of the main woody species recruited in sites with or without declining or recently dead trees due to drought-induced die-off.


Comparisons were made three (2015) to five (2017) years after die-off and tree mortality of the two dominant tree species in each forest (Silver fir, Scots pine) started in 2012. Cramér's V ranges from 0 to 1 corresponding to lack of or complete association between variables, respectively. In all cases P-values are lower than 0.0001 for the corresponding χ 2 statistics.

# DISCUSSION

We found evidences showing a potential compositional shift in the two studied forests as a consequence of recent warming trends and the severe droughts, including that in 2012, which caused dieoff in both forests. We support this argument on the following facts: (i) recruitment of P. sylvestris and A. alba was low, null in the case of P. sylvestris, below declining trees during two or one of the three monitoring years; (ii) the neighborhood of declining individuals showed an increase in the abundance of other tree species; and (iii) the growth of P. sylvestris and A. alba decreased as a consequence of drought, at least in the short term, or were less steep than that of coexisting species (J. thurifera and F. sylvatica, respectively). Based on these findings, we expect that changes in canopy species composition can be observed at local scales in the studied forests during the forthcoming decades. Whether these changes can induce further changes in forest productivity remains an unresolved issue, but we observed a clear reduction

indicate the average BAI for the period 1982–2015. Vertical gray lines highlight the 2005 and 2012 droughts.

in NDVI in the P. sylvestris forest which showed the highest mortality rates (**Table 1**). In this sense, this forest represents an excellent setting to monitor post-drought compositional changes due to high tree diversity, the climatic marginality of this P. sylvestris population and a relatively low tree coverage which makes it sensitive to temperature and evapotranspiration increases (Vicente-Serrano et al., 2013).

The resilience of a forest to an extreme drought event mainly depends on the capacity of trees to resist or recover after such severe episode of water shortage (Gazol et al., 2018b). Alternatively, forest resilience is contingent on the capacity of other tree species to become more abundant by recruiting in suitable microsites with favorable soil and microclimatic conditions (Lloret et al., 2012; Hodgson et al., 2015; Redmond et al., 2018). Thus, tree regeneration capacity is the main driver of forest resilience capacity when drought results in the destruction of the vast majority of overstory trees of the dominant tree species in mixed stands (Redmond et al., 2015; Martínez-Vilalta and Lloret, 2016; Redmond et al., 2018; Suarez and Lloret, 2018). In the P. sylvestris forest, we found an extremely low presence of P. sylvestris seedlings particularly in the neighborhood of declining and dead individuals during three consecutive years after the severe drought (**Figure 2**). Tree recruitment mainly depends on seed availability, germination capacity and postdrought survival of recruits, and it is well known that drought and climate warming can strongly suppress cone production, reduce seed viability and increase recruit mortality in several conifer species (e.g., Redmond et al., 2012, 2018). In this sense, Galiano et al. (2010) and Vilà-Cabrera et al. (2013) observed a decline of P. sylvestris regeneration in some Spanish forests which they attributed to a failure in reproductive success and cone production. The lower seedling density of P. sylvestris as compared to the rest of co-occurring species, suggest a greater capacity of the rest of species for seedling establishment at least in declining stands subjected to recent dry spells. In this sense, P. sylvestris was among the most drought-vulnerable species of that forest according to its poor regeneration. These results contrast with the A. alba forest in which the seedling density of this species did not differ between declining and non-declining sites.

In the P. sylvestris forest, we found that in the shortterm the radial growth of P. sylvestris and J. thurifera responded negatively to drought. However, while J. thurifera presented a growth enhancement after drought, P. sylvestris

calculated as observed BAI divided by maximum BAI of each species. The relative basal area increment observed before (2009–2011), during (2012) and after the 2012 drought (2013–2015) event in the (A) Scots pine and (B) Silver fir forests is shown. Different colors represent different species (black and gray). Lines represent mean BAI and the shaded areas the standard errors. Letters indicate the existence of significant differences between growing periods according to Tukey-HSD tests.

showed similar growth rates than during drought (**Figure 3**). Consistently, J. thurifera displayed strongly positive growth trends whereas P. sylvestris showed no clear trends. Together with the strong P. sylvestris mortality observed in this forest (Camarero et al., 2015; Gazol et al., 2018a), these results point to the local extinction of P. sylvestris in this site as a consequence of warming trends and recurrent drought during decades. J. thurifera is a species perfectly adapted to the Mediterranean continental conditions that prevail in that area which allows this species displaying a great ability to couple either with warm and dry stressing conditions in summer or cold stress in winter (Gimeno et al., 2012; Camarero et al., 2014). Several studies suggest that Fagaceae broadleaf species may replace pine species after disturbances (e.g., Alfaro-Reyna et al., 2018). In this sense, we found a greater abundance of Q. ilex below declining P. sylvestris individuals than below non-declining individuals, and Q. ilex seedlings were fairly abundant below declining and non-declining P. sylvestris individuals (**Figure 2**). Thus, both species, J. thurifera and Q. ilex, might become dominant if further die-off and mortality affect nearby P. sylvestris populations due to warmer and drier conditions.

In the A. alba forest, long-term positive growth-trends were found in both, the dominant species and the co-dominant F. sylvatica, despite with a stepper slope in the case of beech (**Figures 3** and **Supplementary Figure S2**). In the short term, A. alba had a significantly lower growth during and after the drought than before the drought, whereas no changes in growth in response to drought were found in F. sylvatica (**Figure 4**). Many A. alba forests in the Pyrenees displayed declining growth trends and die-off after the 1986 severe drought (Camarero et al., 2011; Gazol et al., 2015, 2018a; Sangüesa-Barreda et al., 2015), and in many of these forests declining growth trends have been exacerbated by successive droughts in 2005 and 2012 (Camarero et al., 2018). This is the case of the studied forest, which presents negative growth trends in several A. alba individuals and enhanced mortality during the last decades (Camarero et al., 2015, 2018). Conversely, no signs of decline and high mortality have been observed in F. sylvatica trees at the same forest (personal observation). In this forest we found a greater growth resilience capacity of F. sylvatica and higher long-term growth trends. Moreover, we also observed that F. sylvatica trees were more common below declining than non-declining A. alba trees, despite no clear patterns in beech regeneration were found. Thus, these results suggest that F. sylvatica is favored over Silver fir in the studied forest after the 2012 drought and the subsequent A. alba die-off. Such forecasted replacement agrees with the expected successional dynamics (replacement of the conifer by the angiosperm species; cf. Bond, 1989) which have been probably accelerated by drought.

It must be also commented that the studied forests did not include the whole forested area dominated by P. sylvestris and A. alba in each region. Therefore, we are dealing with local die-off events causing patchy mortality patterns and leading to vegetation shifts in particularly unfavorable sites (Anderegg et al., 2016). Those sites often present the worst conditions for the performance of the dominant tree species such as high aridity, negative selection of slow-growing trees due to past logging and soils with a low ability to infiltrate and hold water (Camarero et al., 2011, 2015; Gazol et al., 2018a). Consequently, the described effects of drought-triggered die-off on tree composition, growth and productivity mainly occur at small to mid-spatial scales (1–100 ha), which allow the affected species growing and surviving in nearby sites with more favorable conditions for their performance thus assuring their regional persistence.

There is a general congruence between tree growth response to drought and primary productivity variation at larger scales (Gazol et al., 2018b). However, whether the observed drying trends in the dominant tree species in the forest scale-up to vegetation productivity at larger scales remains an unanswered question. Semi-monthly NDVI data showed a more seasonal and predictable temporal pattern in the A. alba forest than in the P. sylvestris forest, probably because of the prevailing climatic conditions: temperate-wet vs. continental and dry Mediterranean conditions (Vicente-Serrano et al., 2013). The two sites showed a marked NDVI reduction as a consequence


TABLE 4 | Results of the linear mixed-effect models selected to study radial-growth trends of the four-tree species (P. sylvestris, J. thurifera, A. alba, and F. sylvatica).

Radial growth was quantified as log<sup>10</sup> (BAI+1) where BAI is the basal area increment. The t-statistic associated to each variable included in the model is shown. Positive signs indicate a positive relationship whereas negative signs indicate a negative relationship. The ∆AICc, Akaike-weight, and pseudo-R<sup>2</sup> (due to fixed and due to fixed and partial effects, respectively) of each selected model is also provided. Significance of the t statistic is indicated by: ∗∗p < 0.01.

of the 2012 drought which in the case of the A. alba forest lasted until the following year. This is probably due to the wellknown dependency of A. alba growth on the climatic conditions of the previous year, particularly on the late-summer water balance (Camarero et al., 2011). Unfortunately, the evaluated NDVI series end in 2015 (Martín-Hernández et al., 2017), making difficult to study the consequences of the 2012 drought due to the short post-drought period. Thus, further analyses on longer NDVI series might be required to understand this issue.

In a recent review of vegetation changes associated to drought-induced forest decline, Martínez-Vilalta and Lloret (2016) found evidence of vegetation shifts in only 8 out of 35 cases. From a demographic point of view, drought-induced vegetation shifts should be characterized by the mortality of at least one dominant species and enhanced growth and recruitment of other potential dominant species in the forest (Lloret et al., 2012; Martínez-Vilalta and Lloret, 2016). The results obtained in the P. sylvestris forest support this criterion indicated by: (i) the high mortality rates of the dominant species (P. sylvestris); (ii) a growth enhancement of the co-dominant species (J. thurifera); and (iii) a much greater regeneration success of the co-dominant as compared to the formerly dominant species. Nevertheless, additional time is required to certainly establish that the observed changes are a truly drought-induced vegetation shift as there is a certain period of time in which the former vegetation still has a chance to return to the previous state which could be regarded as community or population inertia (Hughes et al., 2013). However, the results presented here concur with our previous findings indicating that the 2012 drought induced drastic population and community changes in the P. sylvestris forest (Camarero et al., 2015). These changes are related to shifts at the ecosystem level as reflected the NDVI data (Gazol et al., 2018b), and in sight of the results presented here may lead to vegetation changes toward communities or populations dominated by more drought-tolerant species or individuals. Our findings illustrate and warn against the local extinction of some tree populations near their southernmost distribution limits as in the P. sylvestris forest studied which may become more vulnerable to forecasted aridification trends (Sánchez-Salguero et al., 2017). This is a relevant and urgent concern for researchers and managers which demands mitigation measures such as thinning or considering replacing droughtvulnerable species or individuals by more drought-tolerant species.

#### AUTHOR CONTRIBUTIONS

fpls-09-01546 October 23, 2018 Time: 14:26 # 11

AG and JC conceived the study. AG, JC, and GS-B designed the field study and collected the data. SV-S provided the NDVI data. AG and JC conducted the analyses. SV-S was in charge of the NDVI data processing. AG wrote the manuscript. All authors contributed to revising the paper.

## REFERENCES


## FUNDING

We acknowledge funding by the Spanish Ministry of Economy "Fundiver" project (CGL2015-69186-C2-1-R). GS-B was supported by Spanish Ministry of Economy, Industry and Competitiveness Postdoctoral grant (FJCI 2016-30121, FEDER funds). We acknowledge support of the publication fee by the CSIC Open Access Publication Support Initiative through its Unit of Information Resources for Research (URICI).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018.01546/ full#supplementary-material


regime shifts. Trends Ecol. Evol. 28, 149–155. doi: 10.1016/j.tree.2012. 08.022


**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 Gazol, Camarero, Sangüesa-Barreda and Vicente-Serrano. 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.

# Long-Term Hydraulic Adjustment of Three Tropical Moist Forest Tree Species to Changing Climate

Mahmuda Islam1,2 \*, Mizanur Rahman1,2 and Achim Bräuning<sup>1</sup>

<sup>1</sup> Department of Geography and Geosciences, Institute of Geography, Friedrich-Alexander University Erlangen-Nuremberg, Erlangen, Germany, <sup>2</sup> Department of Forestry and Environmental Science, Shahjalal University of Science and Technology, Sylhet, Bangladesh

Xylem hydraulic adjustment to global climatic changes was reported from temperate, boreal, and Mediterranean tree species. Yet, the long-term hydraulic adjustment in tropical tree species has not been studied so far. Here we developed the first standard chronologies of three hydraulic trait variables for three South Asian moist forest tree species to analyze their long-term hydraulic responses to changing climate. Based on wood anatomical measurements, we calculated Hagen–Poiseuille hydraulically weighted vessel diameter (DH), potential specific hydraulic conductivity (KS), and vulnerability index (VX) and developed standard chronologies of these variables for Chukrasia tabularis, Toona ciliata, and Lagerstroemia speciosa which are different in their xylem structure, wood density, shade tolerance, growth rates, and habitat preferences. Bootstrap correlation analysis revealed that vapor pressure deficit (VPD) strongly positively influenced the xylem water transport capacity in C. tabularis, whereas T. ciliata was affected by both temperature and precipitation. The hydraulic conductivity of L. speciosa was mainly affected by temperature. Different adjustment strategies were observed among the species, probably due to the differences in life history strategies and xylem properties. No positive relationship of conductivity and radial growth was found, but a trade-off between hydraulic safety and efficiency was observed in all studied species.

Keywords: Bangladesh, hydraulic conductivity, tropical forests, climate change, hydraulic safety, xylem anatomy, safety-efficiency trade-off

# INTRODUCTION

The xylem in woody plants serves multiple function such as water and nutrient transport, mechanical support, maintaining stomatal opening and carbon uptake (Carlquist, 2001; Sperry et al., 2003). However, safe and efficient water transport from root to leaf is considered as one of the vital functions (Brodribb and Hill, 1999; Sperry et al., 2003) which determines plant performance, especially during stressful environmental conditions (Tyree and Sperry, 1989; Larcher, 2003; Fonti et al., 2010). Global environmental changes such as rising temperature, changes in temporal precipitation patterns, frequent, and extreme drought events (IPCC, 2014) may have substantial impact on plant growth, phenology, physiology, and other developmental processes (Ceulemans et al., 1997; Crous et al., 2011; Nitschke et al., 2017; He et al., 2018). In order to cope with changing environment particularly related to water availability, vascular plants may adjust their hydraulic architecture so that they can maximize water transportation and can minimize the risk of hydraulic

#### Edited by:

Veronica De Micco, Università degli Studi di Napoli Federico II, Italy

#### Reviewed by:

Ignacio García-González, Universidade de Santiago de Compostela, Spain Ze-Xin Fan, Xishuangbanna Tropical Botanical Garden (CAS), China

#### \*Correspondence: Mahmuda Islam mahmuda.islam@fau.de

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 12 July 2018 Accepted: 13 November 2018 Published: 04 December 2018

#### Citation:

Islam M, Rahman M and Bräuning A (2018) Long-Term Hydraulic Adjustment of Three Tropical Moist Forest Tree Species to Changing Climate. Front. Plant Sci. 9:1761. doi: 10.3389/fpls.2018.01761 failure (Sperry et al., 2002; Fonti et al., 2010; Martinez-Vilalta et al., 2014). For example, plants produce a small number of large size vessels in response to drought (Hacke et al., 2006; Sperry et al., 2006; Wheeler, 2011; Haworth et al., 2017). This phenotypic plasticity allows plants efficient transportation of water from the root to the leaf (Zimmermann, 1983) since the water flow along a vessel is proportional to the fourth power of its radius according to the Hagen-Poiseuille Law (Tyree and Zimmermann, 2002; Sperry et al., 2006; Steppe and Lemeur, 2007; McCulloh et al., 2010). Likewise, as a resopnse to avoid hydraulic failure due to cavitation, vessel frequency is increased so that a higher number of smaller vessels remain functional in terms of cavitation (Sperry et al., 2008; Zanne et al., 2010; Venturas et al., 2017; Pérez-de-Lis et al., 2018).

The hydraulic adjustment mechanism has been investigated across tree populations, species and at different spatial scales (Borghetti et al., 2017). However, less attention has been paid to get insight into the long-term temporal plasticity particularly in tropical forest trees although tropical forests are supposed to be more vulnerable to climate change as expected before (Clark et al., 2003; Doughty and Goulden, 2009). Studying temporal plasticity in xylem hydraulic structure in response to environmental changes requires developing chronologies of hydraulic traits from long-term wood anatomical time series. Particularly, the necessity to develop time series of hydraulic traits has been emphasized to study climate change adaptability both in temperate (Bryukhanova and Fonti, 2013; Oladi et al., 2014; Rita et al., 2015; Schuldt et al., 2016; Noyer et al., 2017; Granda et al., 2018; Pérez-de-Lis et al., 2018) and Mediterranean forests (Corcuera et al., 2004a, 2006; Gea-Izquierdo et al., 2012; Rita et al., 2016; Castagneri et al., 2017; Martínez-Sancho et al., 2017). So far, no hydraulic conductivity (KS) time series were developed for tropical forest trees. A recent global scale metaanalysis (Borghetti et al., 2017) on the temporal plasticity in xylem hydraulic traits revealed that almost all existing studies were done in high latitude regions (36–52◦ N), with no studies in tropical and sub-tropical regions (−30◦ S to 30◦ N). This points to a clear research gap concerning hydraulic adjustment of tree species in tropical regions.

Tropical broadleaf species are understudied in terms of longterm wood anatomical climate adaptations. Tectona grandis was frequently used to develop vessel chronologies (Pumijumnong and Park, 1999; Bhattacharyya et al., 2007; Venegas-González et al., 2015; Sinha et al., 2017). Besides, some mangrove species such as Heretiera fomes and Rhizophora mucronata were studied to develop wood anatomical time series (Verheyden et al., 2005; Chowdhury et al., 2008). In a preliminary study, we investigated wood anatomical features in 27 tree species from a moist tropical forest in Bangladesh and showed their dendrochronological potential (Islam et al., 2018a). In a further step, we established vessel chronologies of diffuse-porous and intermediate shade tolerant Chukrasia tabularis from a Bangladeshi moist tropical forest and showed that the vessel features are highly sensitive to climate (Islam et al., 2018b). Some other studies have shown the potential of developing wood anatomical time series in moist tropical forests (Ohashi et al., 2011, 2014). Nevertheless, time series of hydraulic conductivity of any of the tropical tree species have not been measured so far. Variations of hydraulic conductivity along a water flow path from root to shoot were studied in an Indonesian tropical forest (Kotowska et al., 2015). Also, radial variations of tree hydraulic properties were studied in a Southeast Asian tropical forest (Rungwattana and Hietz, 2017). However, annually resolved time series of hydraulic traits in tropical tree species that can be related to environmental variable are yet to be developed. Moreover, moist tropical forests are rich in species and functional diversity (Slik et al., 2015). It is reported that different functional groups may show different hydraulic adjustment strategy in response to changing environment (Poorter et al., 2010). Several studies suggested that shade intolerant species have higher hydraulic efficiency than shade tolerant species (Markesteijn et al., 2011a,b; Hoeber et al., 2014; Hietz et al., 2017; Rungwattana and Hietz, 2017). Similarly, ring-porous species are likely to be more vulnarable to cavitation than diffuse porous species (Sperry et al., 1994; Taneda and Sperry, 2008; Ogasa et al., 2014).

Interspecific differences in xylem hydraulic architecture (vessel features) may reflect differences in the way of hydraulic adjustment in response to environmental variability (Fonti et al., 2010), providing valuable information about the plasticity a particular species shows under changing environmental conditions. It is therefore important to test the range of hydrosystem adjustment in species of different functional types. In the present study, we developed long-term chronologies of three hydraulic trait variables for three tropical moist forest trees which differ in their xylem anatomy (diffuse porous, semi ringporous and ring-porous), shade tolerance (shade tolerant, shade intolerant, and partial shade tolerant), wood density, growth rates, and habitat preferences. We aimed at testing the following hypotheses:


# MATERIALS AND METHODS

## Study Area and Species

The study was carried out within the Rema-Kalenga forest in the north eastern part of Bangladesh (24◦ 06′−24◦ 14′N, and 91◦ 36′−91◦ 39′E). An area of 1,795 ha in Rema-Kalenga forest was declared as Wildlife Sanctuary which is bounded on the east and south by Tripura state of India. The study sites consist of hilly areas of around 100 m elevation and lowlying valleys. Soils of the forest vary from clay loam on the relatively level ground or valleys to sandy loam on the hills (Hassan, 1994). The climate of the study area is characterized as tropical humid monsoonal climate (Holdridge, 1967). According to the climate data recorded at the nearest climate station (Sreemangal, Bangladesh), the mean annual temperature and total annual precipitation over the period 1950–2015 are 24.8◦C and 2,363 mm, respectively (**Figure 1**). A distinct seasonality is present, with a dry season spanning from November to February (monthly rainfall <100 mm) (**Figure 1**). May-August is the main monsoon season, March–April is the pre-monsoon and September-October is the late-monsoon season in our study region. A significant increasing trend was observed in mean, minimum, and maximum temperatures, whereas precipitation did not show any clear trend over 1950–2015 (**Figure 2**).

The studied forests are recognized for the ecological services they provide and are dominated by deciduous and evergreen trees including Chukrasia tabularis, Toona ciliata, Syzygium grandis, Terminalia bellirica, Lagerstroemia speciosa, Dillenia pentagyna, Dipterocarpus turbinatus, Tectona grandis, as well as various species of Ficus and Albizia. We selected three species (Chukrasia tabularis, Toona ciliata, and Lagerstroemia speciosa) based on their wide distribution across the tropics, their functional differences, and the characteristics of growth-ring boundaries (Islam et al., 2018a). C. tabularis comprises diffuse porous wood anatomy with distinct growth-ring boundaries delineated by marginal parenchyma bands. Xylem anatomy of T. ciliata is characterized by semi ring-porous vessel distribution, with distinct growth-ring boundaries defined by marginal parenchyma and large earlywood vessels. L. speciosa is characterized by ring-porous wood anatomy with distinct growth-ring boundaries detected by large early wood vessels following a marginal parenchyma band. All three study species were found to form annual growth rings (Bhattacharyya and Yadav, 1989; Heinrich et al., 2009; Vlam et al., 2014; Rahman et al., 2018). A detailed description of the selected species is given in **Table 1**.

# Wood Sample Collection, Preparation, and Anatomical Measurements

We extracted increment cores from 105 trees at breast height (1.3 m) by using a 5.0 mm diameter increment borer (Vantaa, Finland). In the field, plastic holders were used to store the cores immediately after extraction. Cores were air dried for 24 h to avoid any attack by fungi. In the laboratory, increment cores were mounted on wooden holders and sanded with increasingly finer grain paper up to 2,000 grit to make wood anatomical features clearly visible (Stokes and Smiley, 1968). A high-pressure water blast was used to remove wood dust and tyloses inside the vessels. White chalk was applied on the surface to fill the vessels, thus improving the contrast from the bulk tissues. Ringwidth (RW) was measured directly on the core surface by using a Lintab 6 measuring system (Rinntech, Heidelberg, Germany) with a precision of 0.01 mm. Ring-width series were visually crossdated using TSAP-Win (Rinntech, Heidelberg, Germany) and statistically evaluated by t-test and Gleichläufigkeit values (GLK, sign test; Eckstein and Bauch, 1969). Crossdating quality was finally checked using COFECHA (Grissino-Mayer, 2001). A total of 70 trees were successfully crossdated.

#### TABLE 1 | Characteristics of studied species.


<sup>a</sup>SA, South Asia; EA, East Asia; SEA, South-East Asia; A, Australia; Af, Africa.

<sup>b</sup>Deciduous: tree leafless for more than 4 weeks.

<sup>c</sup>P, Pioneer; ST, Shade tolerant; PST, Partial shade tolerant.

<sup>d</sup>MP, Marginal parenchyma band (terminal or initial); P, Porosity; FWT, Thick-walled latewood fibers.

<sup>e</sup>F = Furniture; C, Construction; V, Veneer and plywood; B, Boat making.

\* (Williams et al., 2008; Orwa et al., 2009).

Wood anatomical preparation and measurements are laborious and time consuming. Hence, a subset of 9 cross-dated trees from each species (total 27 trees) was finally selected for wood anatomical measurements. We took digital images of the cross sectional surfaces of the cores (5,184 × 3,456 pixels) by using a digital microscope Zeiss Smartzoom 5 (Carl Zeiss Microscopy GmbH 2014, Jena, Germany) for xylem anatomical measurement (**Figure 3**).To simplify the measurements, Adobe Photoshop was used to further increase the contrast between the vessels and other wood elements and manually edit unclear cases of vessel lumen area due to preparation artifacts. Before the measurement, the image was calibrated from a scale bar of known length in the image. On each image, tree-ring boundaries were identified with their year of formation and an analysis region was created in each tree ring by closing the regions delineated by the ring boundary paths in WinCELL. All vessels (vessel area: 0.002–20.0 mm<sup>2</sup> , other features smaller or larger were discarded) were measured for each dated tree ring using an image–analysis software WinCELL 2012a (Regent Instruments Inc., Québec, Canada) which is specifically designed for wood cell analysis. We measured four vessel parameters including the number of vessels (NV), mean vessel tangential diameter (MVTD), mean vessel radial diameter (MVRD), and mean vessel area (MVA). NV, MVTD, and MVRD were then used to calculate vessel density (VD) and hydraulic traits such as the Hagen–Poiseuille hydraulically weighted vessel diameter (DH), potential specific hydraulic conductivity (KS), and vulnerability index (VX).

#### Xylem Hydraulic Traits Calculations

Since most of the vessels in the studied tree species are not circular but mostly elliptical in shape, radial and tangential diameters (Oladi et al., 2014) of each vessel were considered as major (ai) and minor (bi) axis. The Hagen–Poiseuille hydraulically weighted D<sup>H</sup> was computed by applying the following equations (Lewis and Boose, 1995; Tyree and Zimmermann, 2002; Steppe and Lemeur, 2007)

$$D\_H = \sqrt[4]{\frac{1}{n} \sum\_{i=1}^n \frac{2a^2b^2}{a\_i^2 + b\_i^2}}$$

Where n is the number of measured vessels, a and b (µm) are the two axes of the vessels

The potential specific hydraulic conductivity is the total estimated conductivity of a group of vessels assuming that their flow is governed by the Hagen–Poiseuille equation. Potential specific hydraulic conductivity K<sup>S</sup> (kg m−<sup>1</sup> s <sup>−</sup><sup>1</sup> Mpa−<sup>1</sup> ) was calculated using the Hagen–Poiseuille's law (Tyree and Zimmermann, 2002) as follows:

$$K\_{\mathbb{S}} = (\frac{\pi \,\rho}{128 \,\eta}) \ast D\_{H}{}^{4} \ast V D\_{}^{4}$$

Where η is the viscosity of water (1.002 × 10−<sup>9</sup> MPa s−<sup>1</sup> ), ρ the density of water at 20◦C (998.21 kg m<sup>3</sup> ), VD (mm−<sup>2</sup> ) is the ratio of NV per analyzed area (mm<sup>2</sup> ), and D<sup>H</sup> is the Hagen–Poiseuille hydraulically weighted D<sup>H</sup> (m).

The vulnerability index indicates a rough valuation about the trees sensitivity to the risks of cavitation. An increase in the V<sup>X</sup> shows higher potential susceptibility of the hydraulic system to damages (Tyree and Zimmermann, 2002). The V<sup>X</sup> was calculated as follows (Carlquist, 1977; Bauerle et al., 2011; Aref et al., 2013)

$$V\_X = \frac{D\_H}{VD}$$

where VD is vessel density and D<sup>H</sup> is the Hagen–Poiseuille hydraulically weighted DH.

#### Xylem Hydraulic Traits Time Series

For investigating the inter-annual variability of xylem hydraulic traits, we developed chronologies of DH, KS, and V<sup>X</sup> for the three studied species following standard dendrochronological procedures. We observed an increasing trend in hydraulic traits' time series (**Figure S2**). It has been shown that D<sup>H</sup> systematically increases with plant size, including plant height and correlated stem diameter in tropical woody species (Olson and Rosell, 2013) as well as on a global scale (Olson et al., 2014). To avoid any bias of increasing vessel size with increasing stem diameter, we removed age or size related trends from each hydraulic trait time series before chronology development. We used a cubic smoothing spline function with a 50% frequency response at 10 years for detrending, because we aimed at removing multidecadal variability from our rather short (<100 years) time series, and to retain high-frequency signals related to inter-annual climate variability. Detrended time series were then obtained by dividing the observed values by the fitted values (Briffa and Jones, 1990). The detrended time series of nine trees of each of the three species were averaged by a bi-weight robust

features of three South Asian tropical moist forest tree species. White triangles indicate growth-ring boundaries. Hollow arrows indicate growth direction. SV, solitary vessels; GV, grouped vessels; WR, wood rays; MP, marginal parenchyma; LF, latewood fibers; AP, Apotracheal parenchyma; PP, Paratracheal parenchyma. Scale bars are 200µm in the thin sections and 1 mm in the cross sectional wood surface images.

mean to build the chronologies. The chronologies were build using the "chron" function in the Dendrochronology Program Library in R (dplR) (Bunn, 2008, 2010) within the R statistical programming environment (R Development Core Team, 2016). The robustness of the chronologies were assessed using standard statistics commonly used in dendrochronology including the mean inter-series correlation between trees (rbar.bt) (Briffa and Jones, 1990), the expressed population signal (EPS) (Wigley et al., 1984; Buras, 2017) and mean sensitivity (MS) (Briffa and Jones, 1990).

Intra-annual variability of wood anatomical parameters is also a research focus particularly in temperate regions. Such investigations require splitting of a ring into several sections. However, the number of the generally rather large vessels is very low in our studied species particularly in T. ciliata and L. speciosa (**Figure 3**). Splitting a ring into different sections might further reduce the vessel number per section due to splitting across some vessels. Particularly in narrow rings, sectioning is not feasible due to the very low number or even complete absence of vessels in some sections which does not provide adequate replication for statistical analysis. Hence, we refrained from intra-annual analysis of vessel features and instead considered the complete annual rings.

#### Dendroclimatic and Statistical Analysis

Bootstrap correlation analysis was performed to assess the influence of climate variables (temperatures, Tmean, Tmin, and Tmax; precipitation; Palmer Drought Severity index, PDSI; and Vapor Pressure Deficit, VPD) on xylem hydraulic traits. The bootstrap procedure generated 1,000 bootstrapped samples from the original hydraulic trait indices to test the significance of correlation coefficients and the stability of the error estimates more precisely (Guiot, 1991). We used the "bootRes" package in R to compute Pearson correlation coefficients between the hydraulic traits and each of the climatic parameters. A 20 months window from May of the previous year to current year December was used to correlate hydraulic conductivity time series with monthly climate data. Previous year's climate was included because current year's tree-ring features might be influenced by previous year climate through carbon carryover effects which is evident from the auto-correlation in the developed chronologies (0.09–0.56) (Fritts, 1976). One way analysis of variance (ANOVA) was performed to test the differences in hydraulic traits (DH, KS, and VX) among the three species. Requirements for normal distribution and statistical methods were checked. In order to assess the relations of hydraulic traits (DH, KS, and VX) with tree radial growth (RW) and various vessel features (MVA), we used simple linear regression analysis. To investigate the relationships between the hydraulically weighted D<sup>H</sup> and VD we also used regression equation fitting exponential curve.

Temperature, precipitation, and humidity data were obtained from the Bangladesh Meteorological Department (BMD). PDSI data were collected from the KNMI climate explorer (Royal Netherlands Meteorological Institute) (https://climexp. knmi.nl/start.cgi). We calculated vapor pressure deficit (VPD) from the station data following the equations: VPD = [(100 - RH)/100]<sup>∗</sup> SVP, where SVP = 0.610exp<sup>∗</sup> [(17.27∗T)/(T+237.3)], T = temperature, RH = relative humidity (Murray, 1967). All analyses were performed using different statistical packages within the R statistical programming environment (R Development Core Team, 2016)

#### RESULTS

#### Interspecies Variation in Xylem Hydraulic Traits

Xylem hydraulic traits showed significant differences between species (**Figure 4**). Highest variations was observed in the hydraulic vulnarability to cavitaiton (VX) [F(2, 243) = 582.51, p < 0.001], followed by the hydraulically weighted vessel diamter (DH) [F(2, 243) = 294.67, p < 0.001]. The lowest but still strongly significant variation was observed in potential hydraulic conductivity (KS) [F(2, 243) = 45.008, p < 0.001]. Intermediate shade tolerant and diffuse porous C. tabularis exhibited the lowest DH, KS, and V<sup>X</sup> in comparison to the ring posours T. ciliata and L. speciosa. However, semi ring-porous and shade intolerant T. ciliata had higher DH, KS, and V<sup>X</sup> than shade tolerant and ring-porous L. speciosa (**Figure 4**). Differences in DH, KS, and V<sup>X</sup> showed a consistent pattern among the species.

#### Xylem Hydraulic Traits Chronologies

We developed three hydroulic trait chronologies for all the three species (**Figure 5**, **Figure S1**). The statistical parameters indicating the signal strength of the chronologies are described in **Table 2**. The raw tree ring series used to develop standard chronologies are also shown in **Figure S2**. Among the three

diameter (DH), Potential specific hydraulic conductivity (KS) and vulnerability index (VX) of three South Asian tropical moist forest tree species. Different letters indicate significant difference (p < 0.01) among the species.

chronologies within a species, specific hydraulic conductivity (KS) showed the highest common signal in all species as indicated by mean inter series correlation (rbar.bt) that varied between 0.17 and 0.22 among the species. The subsequent EPS was also highest in K<sup>S</sup> among the three hydraulic traits with highest EPS in Toona ciliata (0.72). The lowest rbar.bt and EPS were observed in V<sup>X</sup> chronology in all species. However, inter annual variation indicated by MS was high in all chronologies except DH. In all species, highest MS was found in KS. Based on the chronlogy statistics we progressed with K<sup>S</sup> for further dendroclimatic analysis because K<sup>S</sup> chronologies in all species showed the highest chronology signal which were consistent with values reported for wood anatomical time series (García-González et al., 2016). The K<sup>s</sup> chronologies also provide indirect evidence of vulnerability

to cavitation since hydraulic efficiency is directly related to hydraulic vulnerability (Markesteijn et al., 2011b).

### Xylem Hydraulic Response to Climate Variability

Among the climate variables, VPD had the dominant influence on the variability of hydraulic conductivity (KS) in C. tabularis (**Figure 6**). Both current year and previous year monsoon and post monsoon season VPD had a positive impact on K<sup>S</sup> whereas late spring and summer VPD negatively affected K<sup>S</sup> in C. tabularis (highest correlation with current year September, r = 0.61, p < 0.001). K<sup>S</sup> was significantly positively correlated with late monsoon season (September) mean and maximum temperatures (Tmean, Tmax) and early monsoon season (May) precipitation (**Figure 6**). Previous year May temperatures also positively influenced K<sup>S</sup> whereas temperature in the later growing season (November) negatively affected K<sup>S</sup> in C. tabularis.

In T.ciliata, K<sup>S</sup> was positively related to both day and night time temperatures (Tmean, Tmin, and Tmax) and precipitation during the late monsoon season (**Figure 6**). In contrast, previous year temperatures during the main monsoon and late monsoon season were negatively related to KS, with a strong negative correlation with previous year August Tmean (r = −0.39, p < 0.01) and previous year June and July Tmin (r = −0.49, p < 0.001). Previous year late monsoon precipitation positively affected KS.

K<sup>S</sup> in L. speciosa was strongly and positively affected by previous year winter temperature mean temperature (December; r = 0.35, p < 0.01; **Figure 6**). Similarly, late spring and early summer temperatures (April) (Tmean, Tmin, and Tmax) significantly positively influenced KS. Mean annual maximum temperature (Tmax) was also positively related to KS. Nonetheless, Precipitation had no significant influence on the K<sup>S</sup> in L. speciosa. We found no significant correlation of K<sup>S</sup> with PDSI in any of the studied species.

## Interrelationship Between Hydraulic Traits and Radial Growth

A strong linear and inverse relationship was found between radial growth and hydraulically weighted diameter (DH) in T. ciliata (R <sup>2</sup> = 0.19, p < 0.01) and L. speciosa (R <sup>2</sup> = 0.29, p < 0.01). TABLE 2 | Chronology characteristics of three hydraulic traits for three South Asian moist tropical forest tree species.


Chronology lengths of C. tabularis, T. cliliata and L. speciosa cover 85 years (1930–2015), 81 years (1934–2015), and 91 years (1924–2015), respectively. Mean series lengths (year) of C. tabularis, T. cliliata, and L. speciosa chronologies are 73, 69, and 80 years, respectively. For the description of abbreveations of hydraulic trait variables see Figure 4. \*AC1, 1st order autocorrelation; GLK, Gleichläufigkeit (sign test); rbar.bt, Mean inter-series correlation between trees.

However, the relationship was not significant in C. tabularis (**Figure 7**). The potential hydraulic conductivity (KS) was linearly positively related to radial growth in all three species, with the highest variance explained in T. ciliata (26%) followed by L. speciosa (23%).

K<sup>S</sup> was strongly positively connected to MVA in all the three species, explaining 85–94% of the variations. Likewise, hydraulic vulnerability strongly increased with MVA explaining the highest variation by L. speciosa (62%) (**Figure 8**). MVA was negatively associated with radial growth in two of the three species except C. tabularis (**Figure S3**). A strong negative relationship was found between radial growth and VD in all the studied species, with highest variation (56%) explained in L. speciosa (**Figure S3**). As we expected, hydraulic conductivity was positively related to hydraulic vulnerability in all species (p < 0.001) (**Figure S4**).

#### Xylem Adjustment: Safety vs. Efficiency

We observed the commonly expected trade-off between hydraulically weighted D<sup>H</sup> and VD. VD increased when D<sup>H</sup> decreased, leading to decreased vulnerability to hydraulic failure due to cavitation. This inverse relation was stronger when analyzed on the tree community level (R <sup>2</sup> = 0.37, p < 0.001) than on the species level (trees within a species) (**Figure 9**). Ring-porous T. ciliata and L. speciosa exhibited a stronger trade-off between D<sup>H</sup> and VD than diffuse porous C. tabularis. The variation explained by the inverse relationship of D<sup>H</sup> and VD ranged between 5 and 21% among the three species, with T. ciliata explaining the highest variation.

#### DISCUSSION

## Long Term Hydraulic Adjustment to Climate Variability

We compared standard chronologies of specific hydraulic conductivity (KS) with climate variables to evaluate the long-term hydraulic adjustment to climate variability. The robustness of the K<sup>S</sup> chronologies for dendroclimatic analysis was reflected in the chronology statistics because Ks displayed the highest values in all the statistical parameters among the three chronologies in all three species. Although the common signal was rather low and did not pass the recommended threshold for the EPS statistic of 0.85 (**Table 2**), the values were within the normal range commonly found in wood anatomical time series (García-González et al., 2016). Our chronology statistics are consistent with hydraulic conductivity (KS) time series of other ring-porous (Corcuera et al., 2006; Campelo et al., 2010; Gea-Izquierdo et al., 2012; Pérez-de-Lis et al., 2018) and diffuse-porous (Corcuera et al., 2004b; Oladi et al., 2014; Rita et al., 2015; Schuldt et al., 2016; Noyer et al., 2017) broadleaf trees. Despite of the low common signal, the chronologies exhibited strong relationships with climate variables. Our findings are in line with other studies which reported low common signals but a high correlation with climate in a wide range of species and forest types (Pumijumnong and Park, 1999; García-González and Eckstein, 2003; Fonti et al., 2007; Campelo et al., 2010; González-González et al., 2013; Pritzkow et al., 2016; Martínez-Sancho et al., 2017). Despite of species specific climate responses, we observed some common climate influence on all three species in spring and summer (pre-monsoon) season.

Overall, hydraulic conductivity (KS) of all species positively responded to current year temperatures, precipitation and VPD except the inverse relation of K<sup>S</sup> in C. tabularis and April-May VPD. VPD in association with mean and minimum temperatures exerted dominant control over the xylem water transportation capacity in C. tabularis, whereas T. ciliata was affected by both temperature and precipitation. Temperature had the dominant effect on the hydraulic conductivity (KS) of L. speciosa. Higher temperature increases VPD, leading to increased evapotranspiration. In order to response to high evaporative demand, trees maximize their water transport efficiency. The underlying mechanism of this increased hydraulic conductivity (KS) lies in the fact that high temperature impairs the cell differentiation processes, particularly during latewood formation. Occasionally, during the post-monsoon season, latewood formation is completely ceased in response to high evaporative demand, which results in a ring with a small number of large size earlywood vessels. Consequently, the MVA increased as an adjustment to support efficient water transport. However, previous year temperatures had a negative influence on hydraulic conductivity (KS). Higher temperatures limited tree growth in the previous year, facilitating storage of carbohydrate reserves which create favorable conditions for tree growth of the following year (Fritts, 1976). This results in a reduced conductivity in the following year since small size latewood vessels reduce mean hydraulically weighted diameter and MVA. The positive influence of precipitation on conductivity is most likely caused by growth reduction due to high soil moisture content, particularly in the main monsoon months and in the post monsoon season (Rahman et al., 2018). Both C. tabularis and T. ciliata are sensitive to water saturation since they prefer well-drained soil conditions (Orwa et al., 2009).

Vulnerability to cavitation is an important hydraulic trait which determines the competitiveness of a particular species

under climatic stress. Due to a lower common signal in V<sup>X</sup> chronologies we were unable to compare them with climate variables. However, we observed that higher hydraulic conductivity (KS) increased hydraulic vulnerability in all species (**Figure S4**). Consistent with our results many studies have found the positive relationship between hydraulic conductivity and vulnerability (Markesteijn et al., 2011a; Oladi et al., 2014; Pérezde-Lis et al., 2018). It is also noted that trunk level hydraulic conductivity may be confounded by whole tree architecture such as leaf area to sapwood area and may vary along a height gradient from root to shoot (Kotowska et al., 2015). Considering these factors however, requires huge time and budget and we hope to take them in to account in our next project.

### Inter Specific Variations in Phenotypic Plasticity

Plants may show differential growth and wood anatomical adjustment to stressful environmental conditions. The adjustment mechanism may vary depending on tree functional types and their life history strategies (Poorter et al., 2010; Markesteijn et al., 2011a). In our study, semi ring-porous and ring-porous T. ciliata and L. speciosa showed similar responses to water stress, maintaining an inverse relation of radial growth with hydraulically weighted diameter (DH) and MVA (**Figure 7** and **Figure S3**). By the way of increasing DH, ring-porous trees increase the efficiency of water transportation as reflected in increased potential specific hydraulic conductivity (KS). When radial growth decreases during the late growing season, more carbohydrate reserves may be available for earlywood formation in the following year (Lacointe, 2000; Richburg, 2005). Simultaneously, vessel frequency increased to keep a higher number of smaller vessels hydraulically functional and thus to avoid drought induced cavitation (Sperry et al., 2008; Zanne et al., 2010; Venturas et al., 2017; Pérez-de-Lis et al., 2018). Our results are in line with studies on ring-porous tree species in both temperate (Gea-Izquierdo et al., 2012; Pérez-de-Lis et al., 2018)

intervals.

and Mediterranean forests (Corcuera et al., 2006; Gea-Izquierdo et al., 2012; Rita et al., 2016; Castagneri et al., 2017; Martínez-Sancho et al., 2017). Some studies relating hydraulic traits with life history strategies in tropical regions also yielded similar adjustment mechanisms as reported here (Poorter et al., 2010; Zanne et al., 2010; Markesteijn et al., 2011a; Fan et al., 2012).

In contrast, diffuse porous C. tabularis did not show an increase in water transport efficiency since D<sup>H</sup> and MVA did not significantly increase in water stress conditions, although vessel frequency increased to reduce the risk of hydraulic failure (**Figure 7** and **Figure S3**). The facts that C. tabularis maintains a constant D<sup>H</sup> and MVA under water stress and shows a strong linear positive relation of MVA to V<sup>X</sup> (**Figure 8**), suggest that the species is less vulnerable to cavitation risk than the ringporous species we studied. Hence, the species specific differences in DH, KS, and V<sup>X</sup> (**Figure 4**) clearly support differences in adjustment strategies between ring-porous and diffuse porous trees in our study. In line with our results, diffuse porous Fagus sylvatica and Fagus orientalis showed similar plasticity in order to avoid hydraulic failure due to water stress in temperate forests (Pourtahmasi et al., 2011; Rita et al., 2015; Schuldt et al., 2016; Noyer et al., 2017). Ring-porous and shade tolerant L. speciosa was found to be more effective to prevent hydraulic failure under drought conditions than the shade intolerant T. ciliata. The above findings suggest that ring-porous species are more efficient in water transport than the diffuse porous trees, whereas diffuse porous trees are less vulnerable to cavitation under water stress, as it is clearly reflected in mean K<sup>S</sup> and V<sup>X</sup> indices of our studied species (**Figure 4**).

intervals.

# Does Higher Hydraulic Efficiency Favor Tree Radial Growth?

Efficient water transport is usually expected to favor tree radial growth (Zanne et al., 2010). Numerous studies confirmed the positive association between hydraulic conductivity and tree radial growth across major biomes in the world (Fichot et al., 2009; Zanne et al., 2009; Poorter et al., 2010; Markesteijn et al., 2011a; Fan et al., 2012; Gleason et al., 2012; Hoeber et al., 2014; Rungwattana and Hietz, 2017). Very recently, Hietz et al. (2017) analyzed a larger data set of 325 species from a Panamanian rainforest and revealed that hydraulic conductivity is the best predictor of biomass growth rates. The probable mechanism underlying this positive relationship is attributed to higher leaf level photosynthetic rate due to efficient water transport (Meinzer et al., 2008; Choat et al., 2012). Higher hydraulic efficiency results in plant tissues comprised of higher ratio of vessel lumen area which has lower construction and maintenance costs, which may also explain the positive alignment of hydraulic conductivity to radial growth (Gleason et al., 2016).

Interestingly, we observed a negative relationship between hydraulic conductivity and tree radial growth in all species, with L. speciosa showing the strongest negative relation (**Figure 7**). However, our results are consistent with findings from many other ring-porous (Gea-Izquierdo et al., 2012; Rita et al., 2016) and diffuse porous species (Oladi et al., 2014; Rita et al., 2015). Our studied species responded to the drought stress conditions (e.g., higher temperature, higher evaporative demand, and higher VPD) by slowing down the cell differentiation process (reduced radial growth), and maintaining plasticity in hydraulic architecture such as increased vessel frequency as a safety mechanism against cavitation. Under stress, carbohydrate accumulation is highly favored rather than investing in biomass growth, which leads to reduced radial growth but following year tree radial growth might be stimulated by the accumulated carbohydrate reserve (Lacointe, 2000; Richburg, 2005). Nonetheless, we cautiously interpret the differences in the conductivity-growth relationships observed in our study from that in other studies which mainly focused on conductivity over the whole lifespan rather than to yearto-year variations. Our findings and the evidence of other studies, however, suggest that the widely expected conductivityproductivity positive relationship may not be consistent and calls for further studies.

# Xylem Safety vs. Efficiency: How Consistent at Tree and Species Level Under Humid Environment?

Trees respond to stress by drought or higher VPD by producing a higher number of small vessels to reduce the risk of cavitation (hydraulic safety) and reduce maximum vessel diameters to maintain the efficiency of water transpiration (hydraulic efficiency) (Sperry et al., 2008; Hacke et al., 2017; Haworth et al., 2017; Venturas et al., 2017; Pérez-de-Lis et al., 2018). However, proper knowledge whether this safetyefficiency trade-off occurs only within one individual in response to short-term variations to drought, or whether this trade-off is consistent between trees of the same species or even between individuals of different species is particularly important in the context of long-term global environmental change adaptability. We analyzed hydraulically weighted D<sup>H</sup> and vessel frequency data derived from a total of 1,429 tree rings of three species over the past nine decades. Our analysis confirmed the well-known trade-off between hydraulic conductivity and safety both in trees within and between species for a long-term (**Figure 9**). The species-specific hydraulic efficiency-safety relationships fitted exponential functions which reached lower R 2 values than a function fitted through the community of studied trees of all species.

Consistent with our results, many studies confirmed this trade-off across temperate (Zanne et al., 2010), Mediterranean (LoGullo and Salleo, 1988; Tognetti et al., 1999; Martínez-Sancho et al., 2017), riparian (Pockman and Sperry, 2000), and tropical tree species (Poorter et al., 2010; Fan et al., 2017; Hietz et al., 2017), although long-term analyses are still absent for the tropics. Many studies found a week or even no trade-off between hydraulic conductivity and hydraulic safety (Maherali et al., 2004; Westoby and Wright, 2006; Gleason et al., 2016; Schuldt et al., 2016) which implies that this relationship is variable. By analyzing a large data set, Gleason et al. (2016) found that many woody species showed low safety and low efficiency, but no acceptable causes were attributed after they have analyzed this trade-off in terms of both species level traits and climate variables at different spatial scales. Gleason et al. (2016) (and references therein) also stated that micro anatomical features such as pit membrane structure or conduit length could be relevant to determine xylem safety although there features were not captured in vessel anatomy parameters measured in cross-sections. Nevertheless, species and climate specific long-term adjustment mechanisms as we investigated in this study can provide important insight into how a particular tree species and its populations will respond to environmental stress in a particular ecosystem.

# CONCLUSIONS

Overall, we observed differential responses and adjustment mechanisms of hydraulic behavior to climatic stress varying with tree life history strategies and wood traits. The hydraulic adjustment however could not avoid reduced tree growth in any of the three species we studied despite of their functional differences. Growing conditions in our study region are expected to getting worse due to ever increasing temperature and higher frequency of drought and other extreme events. It is however not clear to what extent tropical trees will be able to adjust their hydraulic system under future global changes. To get more insight in to the possible future hydraulic response of tropical trees, such studies should be extended to additional species and sites across the tropics. Future research should also consider intra-annual variations of hydraulic traits which might allow improving our understanding of the seasonality and associated mechanisms driving the hydraulic behavior of tropical trees.

# AUTHOR CONTRIBUTIONS

MI and AB designed the study. MI and MR conducted field work and prepared samples for wood anatomical measurements. MI performed wood anatomical measurements and analyzed data. MI, MR, and AB interpreted the results and wrote the manuscript.

# FUNDING

The research was supported by the German Academic Exchange Service (DAAD) and the University Grant Commission, Bangladesh.

# REFERENCES


Buras, A. (2017). A comment on the expressed population signal. Dendrochronologia 44, 130–132. doi: 10.1016/j.dendro.2017.03.005


## ACKNOWLEDGMENTS

We sincerely acknowledge the support of the Forest Department (FD) of Bangladesh during the field work. Special thanks to Bishwajit Gosswami for his valuable support during data processing. We thank Iris Burchardt for her technical supports during sample preparation and measurement of wood anatomical features.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018. 01761/full#supplementary-material

Heidelberg: Springer Berlin Heidelberg) 1–7. doi: 10.1007/978-3-662-04 578-7\_1


change. Landsc. Urban Plan. 167, 275–287. doi: 10.1016/j.landurbplan.2017. 06.012


species: the role of extreme climatic events. Front. Plant Sci. 7:1126. doi: 10.3389/fpls.2016.01126


and their relation to wood density and potential conductivity. Am. J. Bot. 97, 207–215. doi: 10.3732/ajb.0900178

Zimmermann, M. H. (1983). Xytem Structure and the Ascent of Sap. Heidelberg: Springer-Verlag.

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

The handling Editor declared a past co-authorship with one of the authors AB.

Copyright © 2018 Islam, Rahman and Bräuning. 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.

# Changes in Spatiotemporal Patterns of 20th Century Spruce Budworm Outbreaks in Eastern Canadian Boreal Forests

#### Lionel Navarro<sup>1</sup> , Hubert Morin<sup>1</sup> , Yves Bergeron<sup>2</sup> and Miguel Montoro Girona1,3 \*

<sup>1</sup> Département des Sciences Fondamentales, Université du Québec à Chicoutimi, Saguenay, QC, Canada, <sup>2</sup> Chaire Industrielle CRSNG-UQAT-UQAM En Aménagement Forestier Durable, Institut de Recherche sur les Forêts, Université du Québec en Abitibi-Témiscamingue, Rouyn-Noranda, QC, Canada, <sup>3</sup> Ecology Restoration Group, Department of Wildlife, Fish and Environmental Studies, Swedish University of Agricultural Sciences (SLU), Umeå, Sweden

#### Edited by:

Giovanna Battipaglia, Università degli Studi della Campania "Luigi Vanvitelli" Caserta, Italy

#### Reviewed by:

Peter Prislan, Slovenian Forestry Institute, Slovenia Minhui He, Northwest Institute of Eco-Environment and Resources (CAS), China

#### \*Correspondence:

Miguel Montoro Girona miguel.montoro.girona@slu.se; miguel.montoro1@uqac.ca

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 08 October 2018 Accepted: 07 December 2018 Published: 21 December 2018

#### Citation:

Navarro L, Morin H, Bergeron Y and Girona MM (2018) Changes in Spatiotemporal Patterns of 20th Century Spruce Budworm Outbreaks in Eastern Canadian Boreal Forests. Front. Plant Sci. 9:1905. doi: 10.3389/fpls.2018.01905 In scenarios of future climate change, there is a projectedincrease in the occurrence and severity of natural disturbances inboreal forests. Spruce budworm (Choristoneura fumiferana)(SBW) is the main defoliator of conifer trees in the North American boreal forests affecting large areas and causing marked losses of timber supplies. However, the impact and the spatiotemporal patterns of SBW dynamics at the landscape scale over the last century remain poorly known. This is particularly true for northern regions dominated by spruce species. The main goal of this study is to reconstruct SBW outbreaks during the 20th century at the landscape scale and to evaluate changes in the associated spatiotemporal patterns in terms of distribution area, frequency, and severity. We rely on a dendroecological approach from sites within the eastern Canadian boreal forest and draw from a large dataset of almost 4,000 trees across a study area of nearly 800,000 km<sup>2</sup> . Interpolation and analyses of hotspots determined reductions in tree growth related to insect outbreak periods and identified the spatiotemporal patterns of SBW activity over the last century. The use of an Ordinary Least Squares model including regional temperature and precipitation anomalies allows us to assess the impact of climate variables on growth reductions and to compensate for the lack of non-host trees in northern regions. We identified three insect outbreaks having different spatiotemporal patterns, duration, and severity. The first (1905–1930) affected up to 40% of the studied trees, initially synchronizing from local infestations and then migrating to northern stands. The second outbreak (1935–1965) was the longest and the least severe with only up to 30% of trees affected by SBW activity. The third event (1968–1988) was the shortest, yet it was also the most severe and extensive, affecting nearly up to 50% of trees and 70% of the study area. This most recent event was identified for the first time at the limit of the commercial forest illustrating a northward shift of the SBW distribution area during the

**59**

20th century. Overall, this research confirms that insect outbreaks are a complex and dynamic ecological phenomena, which makes the understanding of natural disturbance cycles at multiple scales a major priority especially in the context of future regional climate change.

Keywords: black spruce, climate change, dendroecology, GIS, insect outbreaks, landscape ecology, natural disturbances, sustainable forest management

## INTRODUCTION

The boreal forest is the second-largest terrestrial biome in the world, covering 14 million km<sup>2</sup> . It forms a circumpolar forest belt (Burton et al., 2003) that represents about 25% of the world's forests (Dunn et al., 2007). At present, two thirds of this surface is managed for wood production, and this proportion accounts for 37% of the global wood supply (Gauthier et al., 2015). However, an increasing number of studies predict marked consequences of climate change on boreal ecosystems through modifying the dynamics of natural disturbances at different scales and increasing the frequency and severity of events such as wildfires or insect outbreaks (Overpeck et al., 1990; Dale et al., 2001; Millar et al., 2007; Seidl et al., 2014, 2017; Alifa et al., 2017). Thus, improving our understanding of the variability of natural disturbance cycles at multiple scales will be a major challenge in the mitigation and adaptation of boreal forests and their management to climate change.

Natural disturbance regimes determine the dynamics, structure, and composition of forests by altering ecosystem functioning (Anyomi et al., 2016; Montoro Girona et al., 2018b). Insect outbreaks are a key disturbance to consider in forestry planning due to the important economic and ecological implications of these events (Sturtevant et al., 2015). Insect outbreaks affect timber supplies and have a marked impact on overall forest productivity and dynamics. Among all the major insect pests, spruce budworm [Choristoneura fumiferana (Clemens)] (SBW) is the most important defoliator of conifer trees in North American boreal forests (Hardy et al., 1983; Morin and Laprise, 1990). In Canada, more than 90% of spruce and fir forests are affected cyclically by SBW outbreaks, and more than 50% of the annual loss of volume caused by insect damage is attributed to SBW-related defoliation (Natural Resources Canada, 1994). Whereas the consequences of defoliation remain relatively moderate in the western Canada, in the eastern portions of Canada, SBW is responsible for significant losses for the forest industry through high tree mortality and a loss of forest productivity (MacLean, 2016).

SBW outbreaks are complex phenomena influenced by multiple factors that include the affected tree species, the specific ecoregion, and regional climate conditions (MacLean, 2016). Although insect outbreaks play an important role in forest dynamics, most studies involving SBW focus on the relationship with its primary host, balsam fir [Abies balsamea (L.) Mill.]. Mortality occurs in fir stands after 4 years of severe defoliation and outbreaks affect a very high proportion of trees (MacLean, 1980; Bergeron et al., 1995). For secondary hosts, such as black spruce [Picea mariana (Mill.) BSP], the damage (and death) of tree tops and branches is often accompanied by reductions in growth of up to 75% (MacLean, 1984; Nealis and Régnière, 2004). In black spruce, the resistance to defoliation is the result of a phenological asynchrony between the insect and its host (Volney and Fleming, 2000; Pureswaran et al., 2015). As the buds of black spruce burst 14 days later than those of balsam fir, the former is protected from severe SBW-related damage (Nealis and Régnière, 2004) since SBW emergence is synchronized to balsam fir bud burst. Indeed, although SBW can reach high latitudes corresponding to the distribution area of balsam fir (Harvey, 1985; Payette, 1993; Levasseur, 2000), its impact on the black spruce domain is lower, especially in situations where a cold summer prevents eggs from hatching, disrupting the annual cycle of the insect (Nealis and Régnière, 2009). Thus, it is expected that epidemic cycles should be more difficult to identify in the spruce–moss domain, the ecoregion that supports most of the timber industry in eastern Canada given its extent and the excellent wood properties of black spruce (Saucier, 1998; Zhang and Koubaa, 2008).

The reconstruction of insect outbreak cycles at the landscape scale is a major challenge as aerial surveys of defoliation – conducted annually since the 1960s – cover only one outbreak in the last century and are concentrated mostly in the balsam fir area. Dendroecological approaches are a reliable alternative for studying natural and anthropic disturbances in forest ecosystems at a fine resolution (Boulanger et al., 2012; Montoro Girona et al., 2016, 2017). Tree rings provide indirect measurements of insect activity, through the identification of years of growth reduction related to insect outbreaks, thereby allowing the reconstruction of SBW cycles at multiple scales (Morin and Laprise, 1990; Morin et al., 1993; Krause, 1997; Jardon, 2001; Boulanger and Arseneault, 2004; Boulanger et al., 2012). In regard to the spatial extent of SBW outbreaks, some studies have attempted to produce a portrait of past events via both modeling from aerial survey datasets (Fleming and Candau, 1998; Gray and Mackinnon, 2006) or tree-ring analysis (Jardon, 2001). However, the spatiotemporal changes of outbreak dynamics at the landscape scale over the last century in North American boreal forests remains poorly known.

The main goal of this study is to reconstruct the SBW outbreaks during the 20th century at the landscape scale and to evaluate changes in the spatiotemporal patterns in terms of distribution area and severity. For this, we use dendroecological data collected from the eastern Canadian boreal forest. We hypothesize that the spatial pattern will be similar from one outbreak to another with some variation in terms of intensity

and expansion. We expect the last outbreak of the 20th century to have a greater expansion in the spruce domain. This would give credence to the hypothesis of a northward shift in the distribution of SBW in Quebec over the last cycles. However, we expect to observe a time lag in the emergence of the epidemic in the north, as well as a lag in growth reductions; both lags occur due to the lower susceptibility of black spruce stands. To improve the understanding of spatial temporal patterns of SBW activity, historical climatic data were used to examine the influence of precipitation and temperature on outbreaks periods.

# MATERIALS AND METHODS

#### Study Area

The study area is located in the boreal zone of Quebec (Canada) and lies within an area 45.5–53◦N and 58–79◦W, thereby covering nearly 800,000 km<sup>2</sup> (**Figure 1**). This research involves a gradient of stand structures and ecoregions from closed, dense forests in the fir and spruce–moss domain to the south, to the more open and fragmented forests in the spruce–lichen domain to the north. The study area crosses the northern limit of the commercial forest, separating managed forests to the south from unmanaged ones to the north. The coniferous forest landscape is dominated by pure black spruce stands in the north and mixed forests of white spruce, fir, and broad-leaved trees in the south. Regional climate is subpolar humid with a growing season of ≥170 days in the fir domain to a cold subpolar subhumid climate having a much shorter growing season (≤100 days) in the spruce domain (Rossi et al., 2011). The eastern portion of the study area has a greater annual precipitation (950–1350 mm), and the fire return interval is longer in this portion at 270 to >500 years (Cyr et al., 2007).

# Data Compilation and Experimental Design

We undertook a data collection strategy to obtain the maximum amount of dendroecological data available for the years

FIGURE 1 | Location of study sites in Quebec (Canada). The different colors correspond to the various original datasets.

1900–1990 from the study area. This database incorporated sites from eight projects undertaken at the University of Quebec in Chicoutimi over the last 20 years, two datasets from the Canadian Forest Service, and one project from the University of Quebec in Abitibi-Témiscamingue (**Table 1**). The dataset was complemented by an important field survey by Natural Resources Canada, undertaken between 2005 and 2010 in the northern portion of the study area (northern limit project). In this survey, more than 800 sampling plots (400 m<sup>2</sup> each) were sampled. In each plot, wood disks were collected from seven dominant living trees and three dominant living saplings.

Due to the size of our study area and the diversity of datasets sources, the original dataset was filtered to delete sites having a low number of samples (<8) and to keep only black spruce having an age of >100 years. The age criteria was established to guaranty that tree samples were able to register multiple insect outbreaks during the 20th century providing long chronologies across the study area. Based on the scale of this metanalysis and to maximize the number of trees per location, we aggregated some sites from the same ecoregions if they were close enough (≤20 km). This dataset is original and is valuable due to the size of the study area, the high number of trees used in this study, and the inclusion of new chronologies at the limit of the commercial forest in remote areas that are not accessible by road (**Table 1** and **Figure 1**).

#### Dendroecological Data

We selected trees based on dominant species criteria to ensure that the samples were representative of the study stands. All sites were composed exclusively of black spruce, with the exception of Jardon et al. (2003) where the samples were composed of white spruce (**Table 1**). For this study, a total of 3837 samples were used. The samples were prepared, measured, and analyzed based on standard dendroecological protocols (Krause and Morin, 1995). Breast height collected wood disks were air-dried and sanded before tree rings were measured with a WinDendroTM system (Guay et al., 1992) or a manual Henson micrometer having an accuracy of 0.01 mm.

TABLE 1 | Datasets compiled for this study.

Measurements of tree rings covered the entire life of the sampled tree, and the ring patterns were cross-dated using COFECHA (Holmes, 1983). We applied a double detrending method having a 50-year window spline and a negative exponential using ARSTAN (Holmes et al., 1986). Detrending reduced the effects of tree age, genetic growth potential, microsite and stand history, as well as minimizing the effect of climate allowing trees of different growth rates to be compared (Fritts, 1971). Autocorrelation in standardized time series was not removed for the sake of uniformity with similar studies (e.g., Krause, 1997; Boulanger and Arseneault, 2004; Tremblay et al., 2011; Boulanger et al., 2012). The detrended chronologies were averaged to produce a mean standardized chronology for each stand.

For the purpose of extracting the climatic signals in host series, most studies use a host–non-host correcting method using the OUTBREAK program (Holmes and Swetnam, 1996). In our case the lack of non-host species in northern latitudes (black spruce domain) make their use at large scales challenging. In order to overcome this issue, we used modeled climatic data (see below). An epidemic period was defined as a growth reduction (≥1.28 SD on the mean standardized chronologies) of at least five consecutive years allowing 1 year of growth release (Jardon et al., 2003). The severity of insect outbreaks was defined by the proportion of trees at each site that presented such a pattern of growth reduction. This dendroecological approach was used on previous research on black spruce stands (Tremblay et al., 2011; Boulanger et al., 2012, Rossi et al., 2018).

#### Data Analysis

To establish the patterns of SBW activity, spatial data related to insect outbreaks were interpolated based on the percentage of trees affected using an inverse distance weighted interpolation (Childs, 2004) with GIS analysis techniques from the function "Spatial Statistic" extension of ArcGIS 10.3 (ESRI Inc., 2017). Only the epidemic years are shown. The complete chronology can be found in the **Supplementary Material** (**Supplementary Figure S1**).


#### Cluster and Hotspot Analysis

To evaluate the spatial synchrony of insect outbreaks, hotspot and coldspot analyses estimated spatial clustering among the study sites affected or not affected by SBW outbreaks in eastern Canada based on the Getis-Ord local statistic using a fixed distance band estimated as:

$$G\_i^\* = \frac{\sum\_{j=1}^n \boldsymbol{w}\_{i,j} \, \mathbf{x}\_j - \bar{\mathbf{X}}(\sum\_{j=1}^n \boldsymbol{w}\_{i,j})}{\mathrm{S}\sqrt{\frac{n\sum\_{j=1}^n \boldsymbol{w}\_{2,j} - (\sum\_{j=1}^n \boldsymbol{w}\_{i,j})}{n-1}}} \tag{1}$$

$$\bar{X} = \frac{n\sum\_{j=1}^{n} x\_j}{n} \tag{2}$$

$$S = \sqrt{\frac{\sum\_{j=1}^{n} \lambda\_j^2}{n} - \left(\bar{X}\right)^2} \tag{3}$$

where x<sup>j</sup> corresponds to the percentage of trees affected for site j, wi,<sup>j</sup> is the spatial weight between feature i and j, and n is equal to the total number of sites.

Hotspot/coldspot fields were recognized based on statistically significant levels (i.e., 0.1, 0.05, or 0.01); these fell into hotspots, where high values were intermingled with high values and coldspots where low values were intermingled with low values. Areas where high values were surrounded by lower values and where low values were surrounded by higher values were considered as non-significant clusters. To summarize the overloading of maps, these outputs were also presented using Hovmöller diagrams (Persson, 2017). This tool is effective for displaying large amounts of data. It is a technique that is used frequently in atmospheric sciences (Du and Rotunno, 2018). This diagram represents the longitude (or latitude) versus time with the value of the dataset represented through color or shading. RasterVis and LevelPlot R packages were used to plot Gi∗Z-scores and p-values.

#### Climate Model

To improve the interpretation of patterns in the dendrochronological data, we used a climatic model provided by the Climatic Research Unit at the University of East Anglia (CRU TS 3.10). This model is based on an updated gridded climate dataset across the global land areas (excluding Antarctica). The data available for our study area was provided by the Canadian Historical Temperature Database (Vincent and Gullett, 1999). The dataset is composed of monthly precipitation and mean temperature observations on a 0.5 degree latitude/longitude grid over the entire 20th century. Anomalies (positive and negative mean deviations) were estimated using the mean values for each cell for the period with best coverage (1961–1990) (Jones et al., 2012). These anomalies were averaged for each season and used as explanatory variables to compute Ordinary Least Squares linear regression in order to model the relationship between climate variables and the percentage of affected trees (dependant variable). In the absence of non-host chronologies at the landscape scale, this procedure allows us to better assess the proportion of the variability relative to climatic factors versus the unexplained variation (residuals) which can be attributed to SBW outbreaks. Spatial autocorrelation (Moran's Index) was conducted on standard deviations of the residuals to analyze its clustering level.

### RESULTS

The percentage of affected trees over the entire study area revealed three main SBW outbreak periods in eastern Canadian forests over the last century (**Figure 2**). Each insect outbreak differed in terms of duration and severity. The first outbreak occurred between 1905 and 1930, and nearly 40% of the studied trees were affected by SBW activity at the epidemic's peak (1914). The second outbreak was the longest infestation, lasting from 1935 to 1965, although it had the lowest severity level with only 30% of trees being affected during the peak (around 1950). The third outbreak from 1968 to 1988, was the shortest, yet it was the most severe affecting nearly 50% of the studied trees in 1977 (**Figure 2**).

Hotspot and cluster analyses revealed changes in the spatiotemporal patterns of SBW dynamics, as well as the impacts across the eastern Canadian boreal forest during these three periods of budworm outbreak over the last century (**Figures 3, 4**). From 1905 to 1910, no hotspots were recorded. This pattern indicates no synchronization at the landscape scale, although we detected a pattern of locally affected sites from the southwest to the east with a moderate percentage of affected trees (**Figure 4** and **Supplementary Figure S1a**). A significant hotspot, composed of multiple sites was registered in the southwestern portions of the study area in 1911. This infestation reached a maximum affected area of ≈280,000 km<sup>2</sup> in 1914, and then fell to a lower impact phase in 1921 (**Figure 4** and **Supplementary Figure S1b**). This outbreak affected primarily white spruce sampled within the fir domain (Jardon, 2001). SBW activity demonstrated a temporal delay when the area north of the 50th parallel was affected at a later date (1920–1930), with a lower proportion of affected trees and a lower clustering level compared to previous events to the south of our study area (**Figures 3**, **4**). A similar pattern was observed for the second insect outbreak; various sites recorded moderate to severe outbreaks at the local scale prior to the onset in 1944. Four spots had persistently

high percentages of affected trees from the early 1940s (the Abitibi region, southwest of Lake Saint-Jean, the Upper North Shore, and Lake Mistassini). There was a cluster of sites in the southwestern region (1944) that formed a hotspot followed by an eastward expansion of moderate to severe SBW impact until 1957. This outbreak reached an affected area of ≈170,000 km<sup>2</sup> in 1950 (**Figure 4**). The third outbreak had a widespread impact on stands of the fir domain from west to east across our study area, corresponding to 80% of the forest area (≈550,000 km<sup>2</sup> ). This insect outbreak period was first recorded in the northern portion of the spruce–moss domain (especially in sites close to the limit of the commercial forest) in 1970. Analysis of the hotspots

revealed a significant cluster of high values in the northern forest for a short period (1970–1971). A persistent pattern having a high percentage of affected trees was identified in the southwest in 1973, followed by an increasing eastward distribution until 1978, finally ending in a retraction to its original position from 1978 to 1982 (**Figures 3**, **4**).

Based on severity (number of affected trees) and duration (number of years), we determined the differences in SBW outbreak intensity between the northern and southern portions of the eastern Canadian forests over the last century (**Figure 5**). In the northern stands, tree-ring chronologies registered a higher number of years of weaker outbreaks than in the southern sites. However, severe SBW outbreaks were rare at the northern limit of the commercial forest during the 20th century (**Figure 5**). The southern portion of the study area was characterized by shorter, more severe, and more synchronized periods of SBW infestation.

Climate regressions had relatively high R 2 values for each outbreak period, ranged from 0.56 to 0.70 (**Table 2**). However, the Koenker and the Jarque-Bera statistics indicated non-stationarity and heteroscedasticity. Thus, the relationship between climate and outbreak periods was spatially inconsistent and changed with explanatory variable magnitudes. Indeed, even if each explanatory variable was significantly correlated to the frequency of affected trees, and even if the Variance Inflation Factors (VIF) indicated no redundancy among explanatory variables (<7.5), the model was improperly specified. The Moran's Index between 0.36 and 0.47 indicated significant tive clustering of the regression residuals. Thus, a key variable was missing and the model was mis specified suggesting that a key variable is missing and that the model is misspecified. Therefore, As a matter of fact, the spatial patterns of the regression residuals was is similar to the one of the measured percentage of trees affected and the higher values are consistently underestimated,

#### TABLE 2 | Ordinary Least Squares model for each outbreak period.


#### Second outbreak: 1945–1957



Third outbreak: 1970–1982

The variables included were selected among spring, summer, autumn and winter precipitation and temperature in order to best fit the model (Variance Inflation Factor −VIF < 7.5 and significant coefficient). Values in bold correspond to significant results (P < 0.05), SE correspond to standard error.

in the southwest and southeast portion of the study area for the first and the second outbreak periods (**Figures 6c1,c2**) and in the southwest and the northeast for the third one (**Figure 6c3**).

#### DISCUSSION

Under most climate change scenarios, disturbance regimes are likely to be most pronounced within the boreal biome (Seidl et al., 2017). As a consequence, much research has been aimed on improving our understanding of fire (Cyr et al., 2016; Drobyshev et al., 2017; Portier et al., 2018), insect outbreaks (Boulanger et al., 2016; Boulanger et al., 2017b; Montoro Girona et al., 2018b), and windthrow (Anyomi et al., 2017; Saad et al., 2017) in the North American boreal forest. In the recent years, the eastern Canadian boreal forest has been experiencing a SBW outbreak, during which some of the most productive forest areas have been severely damaged (e.g., along the North Shore region), thereby with implications at the ecological (forest dynamics) and economic level (financial losses) due to the large extent of affected forest. As the frequency and severity of disturbances are expected to increase under future climate change scenarios, understanding the impact of SBW outbreaks in the past becomes essential for adapting to the uncertainties of climate change. In this study, we provide for the first time a landscape reconstruction of the spatiotemporal pattern of SBW dynamics over the last century across a vast study area of almost 1 million km<sup>2</sup> in the eastern

(C) Residuals of the OLS model.

Canadian boreal forest. We also reveal the first evidence of the presence of endemic populations of SBW north of the 50th parallel.

This study demonstrated that SBW outbreaks have a major impact on forest ecosystems in terms of growth reduction that influences tree survival, regeneration, and succession (MacLean, 2016). Dendroecological series across the entire study area have identified three main periods of elevated SBW activity (Morin and Laprise, 1990; Jardon et al., 2003; Boulanger et al., 2012). Contrarily to our preliminary expectation, the first hypotheses was rejected, because each insect outbreak was manifested by a different spatiotemporal pattern, severity, and duration, thereby demonstrating the complexity of this ecological phenomenon.

## Spatiotemporal Patterns at the Landscape Scale

These different patterns manifest themselves by an expansion of the spatial extent of the affected area over the 20th century. This dynamic is confirmed at a wider temporal scale as outbreaks in the 19th century were less synchronous and presented a lower diffusion rate (Jardon et al., 2003). This could be the result of a long-term forest transformation process. Baskerville (1975) described SBW as a super silviculturist, killing overstory trees and promoting the development of shade-tolerant species, such as balsam fir (Morin, 1994; Morin and Laprise, 1997). Even if the anthropogenic influences on SBW dynamics remain a matter of debate (MacLean, 2016), we know that fire suppression, clear cutting, and insecticide spraying tend to favor the development of fir, SBW's most vulnerable host (MacLean, 1980). This forest transformation process, enhanced by the diminution of fire frequency since the end of the Little Ice Age (Drobyshev et al., 2017), could have a major role in explaining the onset of the first SBW outbreak in the early 20th century (Bergeron and Leduc, 1998; Jardon et al., 2003). Furthermore, this outbreak occurred during one of the driest decades of the last century "1910–1920" (Girardin et al., 2009), reducing host trees' vigor, and favoring their susceptibility to subsequent stresses, such as fire or SBW outbreaks (Flower et al., 2014; Berdanier and Clark, 2016).

The full spatial extent of this outbreak is described here for the first time at the landscape level using dendrochronological data and including sites from both fir and spruce domain. From 1909 to 1921, this outbreak affected mainly the fir domain south of the 50th parallel. Several local infestations were recorded in the years 1905–1909, prior to the synchronized outbreak in the southwestern portion in 1911 (**Supplementary Figure S1a**). This

synchronization at a regional and landscape scale could stem from favorable weather conditions (Moran effect) (Myers, 1998; Royama et al., 2005) combined with the exchange of eggs by moth dispersion (Williams and Liebhold, 2000). The northern part of the study area was later affected moderately and synchronously during the 1920s. A second infestation (1935–1965) was smaller in extent, had a milder impact in more northern latitudes. However, this event does present the same pattern than the first outbreak with local infestations occurring during a few years prior to a synchronized outbreak at a wider scale. Our data tend to confirm the theory of Blais (1983) and suggest that an outbreak following a stand-replacing epidemic event will have a lower impact due to the establishment of less vulnerable younger stands. Finally, the 1968–1988 epidemic was the largest, most synchronous, and best-documented SBW outbreak in eastern Canada (Morin and Laprise, 1990; Morin et al., 2007). Its dynamics in the spruce domain presented a very different pattern from the earlier outbreaks, and it appears to have reached almost all study sites. Furthermore, although spruce are less vulnerable to SBW than fir (MacLean, 1980), many spruce stands were significantly impacted by this outbreak.

### Factors Involved in Spatiotemporal Patterns at Multiples Scales

These differences in spatiotemporal patterns could be explained by many factors, one of the most important being climate. In fact, as climate influences SBW population dynamics, it also causes stress to host stands causing them to be more vulnerable to subsequent biotically induced disturbances (De Grandpré et al., 2018a). Greenbank (1956) demonstrated that hot and dry summers influenced the onset of the 1912 and 1949 SBW outbreaks in New Brunswick. Hot summer temperatures are required for the insect to complete its life cycle (Pureswaran et al., 2015). Drought can also increase host vulnerability by enhancing the carbohydrate content of leaves (Mattson and Haack, 1987). The northern portion of our study area is characterized by cold and short summers that, at present, prevent the establishment of endemic populations. Early frosts in the northern stands prevent eggs from hatching (Pureswaran et al., 2015). Thus, the outbreak impacts observed on trees from the more northern sites during the 1920s could be evidence for the arrival of immigrant populations from southwestern Quebec. Recently, the use of weather radar has allowed the identification of such mass exodus events (Boulanger et al., 2017a). However, the 1968–1988 outbreak provides a different picture as severe and synchronous local epidemics were recorded in the early phases (1970–1971) of the outbreak within the North Shore region before any growth reduction was recorded in the fir domain (1972) (**Figure 4** and **Supplementary Figure S1e**). This phenomenon matches with the distribution of our regression analysis residuals, discarding the hypothesis of a climatic event influence as a unique factor of growth reduction (**Figure 6**). This could therefore be the first evidence of the presence of endemic populations of SBW north of the 50th parallel. In addition, as black spruce phenology and overall shoot length increase in response to experimental warming (Bronson et al., 2009), this could have reduced the phenological asynchrony between the insect and host, thereby making northern sites more suitable for infestation. These arguments give weight to the hypothesis of a northward shift in the extent of SBW outbreaks during the 20th century (Régnière et al., 2012; Pureswaran et al., 2015). Gray (2013) identified the North Shore region and Gaspé Peninsula as the areas having the highest increases in outbreak severity and duration; this agrees with our dataset from the North Shore and with the onset area of the current infestation. Unfortunately, our dataset did not contain sites from the Gaspé Peninsula.

SBW dynamics recorded at the regional scale differ between the northern and southern portions of the study area. In the spruce domain, we observed more cumulative years of growth reduction than what was observed in the southern regions (**Figure 5**). However, we have found only a few occurrences of severe SBW impacts on black spruce in comparison to the fir domain where white spruces were periodically (and severely) affected. First, black spruce is less vulnerable to SBW than white spruce or balsam fir. This is due to an asynchrony of approximately 14 days in budburst phenology causing high rates of mortality for the second instar larvae trying to feed on this host (Nealis and Régnière, 2004). Despite the fact that white spruce buds burst in a time frame more similar to balsam fir than black spruce, white spruce also produces more buds that develop and lignify faster (Nealis and Régnière, 2004). According to MacLean (2016), balsam fir and white spruce, being less resistant species, are more prone to secondary mortality agents, such as shoestring root rot, Armillaria mellea (Vahl ex Fries), which occurs in most defoliated trees. In addition, southern mixed forest benefit from a greater diversity of SBW natural enemies, which could also explain the lower frequency and shorter duration of outbreaks in this zone (Cappuccino et al., 1998; Campbell et al., 2008).

#### Methodological and Forest Management Implications

Similarities were observed between our spatiotemporal patterns and the aerial surveys of defoliation in the area (Gray et al., 2000; Gray and Mackinnon, 2006). Although these surveys were not conducted specifically to measure SBW impacts on black spruce at these latitudes, they still show an important spread of the epidemic in 1974, especially in the spruce–moss domain. It is possible that dendrochronology detects epidemic thresholds earlier as it is more sensitive, and the technique is better suited to black spruce. Indeed, growth reductions can be observed for defoliation levels that are not detectable with aerial surveys, which are categorial (none, light, moderate, severe).

The dendroecological approach has shown its effectiveness in the study of past insect outbreak dynamics (Jardon et al., 2003; Morin et al., 2007; Boulanger et al., 2012). Given the large amount of dendrochronological data that has been published during the last decades, large-scale meta-analyses are increasingly important and can provide a complete portrait of historical SBW outbreaks, placing recent events into a larger spatiotemporal context, and completing the existing monitoring proxies. Therefore, to provide a more relevant understanding of the SBW dynamics,

future sampling efforts should be more homogeneous with sampling sites evenly distributed and with uniform sampling methods. Furthermore, the inclusion of northern latitude sites demonstrated its potential for improving our understanding of past outbreak patterns, but also the methodological difficulties associated with the inclusion of a secondary host in the analysis. The inclusion of such chronologies will be challenging, nonetheless, given the difficulties in accessing sampling sites, finding old trees able to provide long chronologies, and, moreover, the difficulty of finding non-host trees to refine the epidemic signal. We recommend continuing the collection of more samples from northern latitude sites, considering new areas (e.g., Ontario and Gaspe peninsula), as well as adding other species affected by SBW (e.g., balsam fir). Getting more details across the historical distribution area of SBW will improve the resolution of the existing dendroecological database. Based on our results concerning the potential links between climate and SBW activity (**Figure 6**), we suggest that future research should be developed to better discriminate the interactions connecting climate anomalies, as a triggering or interrupting factor of growth reduction, to SBW and its hosts dynamics.

Natural disturbance regimes are an integral part of boreal forest ecosystems, and silvicultural methods are now attempting to emulate their impacts by adapting more appropriate harvesting treatments (Kuuluvainen and Grenfell, 2012; Montoro Girona et al., 2018a). Many studies have focused on the role, impact, and frequency of fire cycles on the management of boreal ecosystems (Bergeron et al., 2002; Kuuluvainen and Grenfell, 2012); however, the understanding of the role and impacts of insect outbreaks remains incomplete, in particular at a larger scale (De Grandpré et al., 2018b; Robert et al., 2018). Understanding SBW disturbance regimes at the landscape-scale and implementing effective management strategies requires to define outbreak dynamics in both time and space (Bouchard and Pothier, 2010). Currently, sylvicultural practices aim to imitate fire disturbances promoting large clearcuts (Hunter, 1993; National Forest Database [N.F.D], 2016. In order to reduce boreal forest vulnerability to SBW outbreaks, some authors proposed to adapt sylvicultural treatments and forest management promoting the harvesting of the most susceptible stands such as mature fir stands (MacLean, 1980, 1996; Sainte-Marie et al., 2014) and favor uneven-aged spruce and mixed stands through silvicultural practices such as partial cuttings (Bergeron et al., 2017). Integrative multipledisturbance research is needed to better understand the climatic and ecological context of insect outbreaks and to identify the type of interactions that occur during these events; as such, adequate management strategies can be developed in accordance with the forest structure at regional and local scales.

#### CONCLUSION

Natural disturbance regimes define forest ecosystems by influencing their structure, species composition, and functional processes. The evaluation of outbreak periods during the last century demonstrated that SBW is a major disturbance event in eastern Canada, affecting large surfaces and having an impact on forest ecosystem dynamics. Landscape-scale reconstruction of the spatiotemporal patterns of SBW outbreaks in eastern Canadian forests highlighted three outbreaks during the 20th century, each having different spatiotemporal patterns, duration, and severity. This study revealed the diversity and complexity of outbreak dynamics over time as well as the importance of meta-analyses for better understanding the SBW patterns at the landscape scale and evaluating the impacts on forest ecosystems. Furthermore, this study represents a major contribution to forest ecology providing valuable data from remote sites located at the limit of commercial forests.

Under climate change, natural disturbances regimes and species' distributions are expected to be altered. Based on dendroecological approaches, we demonstrated evidence of SBW activity north of the 50th parallel, adding weight to the hypothesis of a northward shift in the extent to the outbreaks during the 20th century (Régnière et al., 2012; Pureswaran et al., 2015). Finally, improving our understanding of natural disturbance cycles at multiple scales should be a priority for assessing boreal forest adaptation and modification to future climate change.

## AUTHOR CONTRIBUTIONS

LN, MG, HM, and YB conceptualized and investigated the study. LN performed the data curation and administered the project. LN and MG designed the methodology, wrote the original draft of the manuscript, and edited the manuscript. HM provided the resources. HM and MG supervised the study. LN, MG, HM, and YB reviewed the manuscript. HM and YB contributed to the funding.

#### FUNDING

Funding was provided by the Natural Sciences and Engineering Research Council (NSERC) of Canada, the Canada Research Industrial Chairs Program, Canada Foundation for Innovation, and the Fonds de Recherche de Québec-Nature et Technologies obtained by HM and YB.

### ACKNOWLEDGMENTS

We thank V. Bergeron and G. Grosbois for logistical help and essential support as well as J. Hjältén and A. Hof for their suggestions on an earlier version of this manuscript. We also thank M.-J. Tremblay for technical advice with the dendroecological methods and M. Hay for verifying the English in the text.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018.01905/ full#supplementary-material

### REFERENCES

fpls-09-01905 December 20, 2018 Time: 15:36 # 13


Holmes, R. (1983). Program COFECHA User's Manual. Laboratory of Tree-Ring Research. Tucson, AZ: The University of Arizona.


Disturbance Ecology: The Process and the Response eds E. A. Johnson, and K. Miyanishi (New York, NY: Elsevier), 555–577.



1375–1388. doi: 10.1002/(SICI)1097-0088(199910)19:12<1375::AID-JOC427> 3.0.CO;2-0


**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 Navarro, Morin, Bergeron and Girona. 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.

# Early-Warning Signals of Individual Tree Mortality Based on Annual Radial Growth

Maxime Cailleret1,2 \*, Vasilis Dakos<sup>3</sup> , Steven Jansen<sup>4</sup> , Elisabeth M. R. Robert5,6,7 , Tuomas Aakala<sup>8</sup> , Mariano M. Amoroso9,10, Joe A. Antos11, Christof Bigler<sup>1</sup> , Harald Bugmann<sup>1</sup> , Marco Caccianaga12, Jesus-Julio Camarero13, Paolo Cherubini<sup>2</sup> , Marie R. Coyea14, Katarina Cufar ˇ <sup>15</sup>, Adrian J. Das16, Hendrik Davi<sup>17</sup> , Guillermo Gea-Izquierdo18, Sten Gillner19, Laurel J. Haavik20,21, Henrik Hartmann<sup>22</sup> , Ana-Maria Here ¸s23,24, Kevin R. Hultine25, Pavel Janda26, Jeffrey M. Kane<sup>27</sup> , Viachelsav I. Kharuk28,29, Thomas Kitzberger30,31, Tamir Klein32, Tom Levanic<sup>33</sup> , Juan-Carlos Linares34, Fabio Lombardi35, Harri Mäkinen36, Ilona Mészáros<sup>37</sup> , Juha M. Metsaranta38, Walter Oberhuber39, Andreas Papadopoulos<sup>40</sup> , Any Mary Petritan2,41, Brigitte Rohner<sup>2</sup> , Gabriel Sangüesa-Barreda42, Jeremy M. Smith<sup>43</sup> , Amanda B. Stan44, Dejan B. Stojanovic45, Maria-Laura Suarez46, Miroslav Svoboda<sup>26</sup> , Volodymyr Trotsiuk2,26,47, Ricardo Villalba48, Alana R. Westwood49, Peter H. Wyckoff<sup>50</sup> and Jordi Martínez-Vilalta5,51

<sup>1</sup> Department of Environmental Systems Science, Forest Ecology, Institute of Terrestrial Ecosystems, ETH Zürich, Zurich, Switzerland, <sup>2</sup> Swiss Federal Institute for Forest, Snow and Landscape Research – WSL, Birmensdorf, Switzerland, <sup>3</sup> CNRS, IRD, EPHE, ISEM, Université de Montpellier, Montpellier, France, <sup>4</sup> Institute of Systematic Botany and Ecology, Ulm University, Ulm, Germany, <sup>5</sup> CREAF, Cerdanyola del Vallès, Catalonia, Spain, <sup>6</sup> Ecology and Biodiversity, Vrije Universiteit Brussel, Brussels, Belgium, <sup>7</sup> Laboratory of Wood Biology and Xylarium, Royal Museum for Central Africa, Tervuren, Belgium, <sup>8</sup> Department of Forest Sciences, University of Helsinki, Helsinki, Finland, <sup>9</sup> Consejo Nacional de Investigaciones Científicas y Técnicas, CCT Patagonia Norte, Río Negro, Argentina, <sup>10</sup> Instituto de Investigaciones en Recursos Naturales, Agroecología y Desarrollo Rural, Sede Andina, Universidad Nacional de Río Negro, Río Negro, Argentina, <sup>11</sup> Department of Biology, University of Victoria, Victoria, BC, Canada, <sup>12</sup> Dipartimento di Bioscienze, Università degli Studi di Milano, Milan, Italy, <sup>13</sup> Instituto Pirenaico de Ecología (IPE-CSIC), Zaragoza, Spain, <sup>14</sup> Centre for Forest Research, Département des Sciences du Bois et de la Forêt, Faculté de Foresterie, de Géographie et de Géomatique, Université Laval, Québec, QC, Canada, <sup>15</sup> Biotechnical Faculty, University of Ljubljana, Ljubljana, Slovenia, <sup>16</sup> United States Geological Survey, Western Ecological Research Center, Sequoia and Kings Canyon Field Station, Three Rivers, CA, United States, <sup>17</sup> Ecologie des Forêts Méditerranéennes (URFM), Institut National de la Recherche Agronomique, Avignon, France, <sup>18</sup> Centro de Investigación Forestal (CIFOR), Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria, Madrid, Spain, <sup>19</sup> Institute of Forest Botany and Forest Zoology, TU Dresden, Dresden, Germany, <sup>20</sup> USDA Forest Service, Forest Health Protection, Saint Paul, MN, United States, <sup>21</sup> Department of Entomology, University of Arkansas, Fayetteville, AR, United States, <sup>22</sup> Department of Biogeochemical Processes, Max Planck Institute for Biogeochemistry, Jena, Germany, <sup>23</sup> Department of Forest Sciences, Transilvania University of Brasov, Bras " ov, Romania, <sup>24</sup> BC3 – Basque Centre for Climate Change, Leioa, Spain, <sup>25</sup> Department of Research, Conservation and Collections, Desert Botanical Garden, Phoenix, AZ, United States, <sup>26</sup> Faculty of Forestry and Wood Sciences, Czech University of Life Sciences, Prague, Czechia, <sup>27</sup> Department of Forestry and Wildland Resources, Humboldt State University, Arcata, CA, United States, <sup>28</sup> Sukachev Institute of Forest, Siberian Division of the Russian Academy of Sciences, Krasnoyarsk, Russia, <sup>29</sup> Siberian Federal University, Krasnoyarsk, Russia, <sup>30</sup> Department of Ecology, Universidad Nacional del Comahue, Río Negro, Argentina, <sup>31</sup> Instituto de Investigaciones en Biodiversidad y Medioambiente, Consejo Nacional de Investigaciones Científicas y Técnicas, Río Negro, Argentina, <sup>32</sup> Department of Plant and Environmental Sciences, Weizmann Institute of Science, Rehovot, Israel, <sup>33</sup> Department of Yield and Silviculture, Slovenian Forestry Institute, Ljubljana, Slovenia, <sup>34</sup> Department of Physical, Chemical and Natural Systems, Pablo de Olavide University, Seville, Spain, <sup>35</sup> Department of Agricultural Science, Mediterranean University of Reggio Calabria, Reggio Calabria, Italy, <sup>36</sup> Natural Resources Institute Finland (Luke), Espoo, Finland, <sup>37</sup> Department of Botany, Faculty of Science and Technology, University of Debrecen, Debrecen, Hungary, <sup>38</sup> Northern Forestry Centre, Canadian Forest Service, Natural Resources Canada, Edmonton, AB, Canada, <sup>39</sup> Department of Botany, University of Innsbruck, Innsbruck, Austria, <sup>40</sup> Department of Forestry and Natural Environment Management, Technological Educational Institute of Stereas Elladas, Karpenisi, Greece, <sup>41</sup> National Institute for Research and Development in Forestry "Marin Dracea", Voluntari, Romania, <sup>42</sup> Departamento de Ciencias Agroforestales, EiFAB, iuFOR – University of Valladolid, Soria, Spain, <sup>43</sup> Department of Geography, University of Colorado, Boulder, CO, United States, <sup>44</sup> Department of Geography, Planning and Recreation, Northern Arizona University, Flagstaff, AZ, United States, <sup>45</sup> Institute of Lowland Forestry and Environment, University of Novi Sad, Novi Sad, Serbia, <sup>46</sup> Grupo Ecología Forestal, CONICET – INTA, EEA Bariloche, Bariloche, Argentina, <sup>47</sup> Department of Environmental Systems Science, Institute of Agricultural Sciences, ETH Zürich, Zurich, Switzerland, <sup>48</sup> Laboratorio de Dendrocronología e Historia Ambiental,

#### Edited by:

Veronica De Micco, University of Naples Federico II, Italy

#### Reviewed by:

Louis S. Santiago, University of California, Riverside, United States Minhui He, Northwest Institute of Eco-Environment and Resources (CAS), China

\*Correspondence:

Maxime Cailleret maxime.cailleret@usys.ethz.ch

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 12 September 2018 Accepted: 18 December 2018 Published: 08 January 2019

#### Citation:

Cailleret M, Dakos V, Jansen S, Robert EMR, Aakala T, Amoroso MM, Antos JA, Bigler C, Bugmann H, Caccianaga M, Camarero J-J, Cherubini P, Coyea MR, Cufar K, ˇ Das AJ, Davi H, Gea-Izquierdo G, Gillner S, Haavik LJ, Hartmann H, Here ¸s A-M, Hultine KR, Janda P, Kane JM, Kharuk VI, Kitzberger T, Klein T, Levanic T, Linares J-C, Lombardi F, Mäkinen H, Mészáros I, Metsaranta JM, Oberhuber W, Papadopoulos A, Petritan AM, Rohner B, Sangüesa-Barreda G, Smith JM, Stan AB, Stojanovic DB, Suarez M-L, Svoboda M, Trotsiuk V, Villalba R, Westwood AR, Wyckoff PH and Martínez-Vilalta J (2019) Early-Warning Signals of Individual Tree Mortality Based on Annual Radial Growth. Front. Plant Sci. 9:1964. doi: 10.3389/fpls.2018.01964

Instituto Argentino de Nivología, Glaciología y Ciencias Ambientales, CCT CONICET Mendoza, Mendoza, Argentina, <sup>49</sup> Boreal Avian Modelling Project, Department of Renewable Resources, University of Alberta, Edmonton, AB, Canada, <sup>50</sup> Department of Biology, University of Minnesota, Morris, Morris, MN, United States, <sup>51</sup> Departament de Biologia Animal, de Biologia Vegetal i d'Ecologia, Universitat Autònoma de Barcelona, Cerdanyola del Vallès, Spain

Tree mortality is a key driver of forest dynamics and its occurrence is projected to increase in the future due to climate change. Despite recent advances in our understanding of the physiological mechanisms leading to death, we still lack robust indicators of mortality risk that could be applied at the individual tree scale. Here, we build on a previous contribution exploring the differences in growth level between trees that died and survived a given mortality event to assess whether changes in temporal autocorrelation, variance, and synchrony in time-series of annual radial growth data can be used as early warning signals of mortality risk. Taking advantage of a unique global ring-width database of 3065 dead trees and 4389 living trees growing together at 198 sites (belonging to 36 gymnosperm and angiosperm species), we analyzed temporal changes in autocorrelation, variance, and synchrony before tree death (diachronic analysis), and also compared these metrics between trees that died and trees that survived a given mortality event (synchronic analysis). Changes in autocorrelation were a poor indicator of mortality risk. However, we found a gradual increase in interannual growth variability and a decrease in growth synchrony in the last ∼20 years before mortality of gymnosperms, irrespective of the cause of mortality. These changes could be associated with drought-induced alterations in carbon economy and allocation patterns. In angiosperms, we did not find any consistent changes in any metric. Such lack of any signal might be explained by the relatively high capacity of angiosperms to recover after a stress-induced growth decline. Our analysis provides a robust method for estimating early-warning signals of tree mortality based on annual growth data. In addition to the frequently reported decrease in growth rates, an increase in inter-annual growth variability and a decrease in growth synchrony may be powerful predictors of gymnosperm mortality risk, but not necessarily so for angiosperms.

Keywords: tree mortality, ring-width, forest, growth, resilience indicators, drought, biotic agents, variance

#### INTRODUCTION

Episodes of tree mortality associated with drought and heat stress have been reported in many forested biomes over the last decades (Allen et al., 2010; Hartmann et al., 2018), and are expected to increase under ongoing climate change in many regions (Allen et al., 2015). Forest dieback can induce multiple changes in forest functions and dynamics (Franklin et al., 1987; Anderegg et al., 2013a, 2016b), including rapid shifts in vegetation composition (Martínez-Vilalta and Lloret, 2016) or significant changes in terrestrial carbon sequestration with resulting feedbacks to the climate system (e.g., Carvalhais et al., 2014). In addition to the direct loss of individuals, tree mortality may also reduce forest regeneration capacity by decreasing the number of potential reproductive individuals, and by modifying the micro-environmental conditions and biotic interactions (e.g., Mueller et al., 2005; Royer et al., 2011). Being able to forecast when and where tree mortality episodes are likely to occur is thus a prerequisite for effective and adaptive forest management, especially under progressively warmer and drier conditions (Pace et al., 2015; Trumbore et al., 2015).

Evaluating individual tree mortality risk requires reliable indicators that reveal temporal changes in tree vitality (Allen et al., 2015; Hartmann et al., 2018). Such information can be provided by physiological and anatomical data. Both abrupt and long-term declines in hydraulic conductivity caused by droughtinduced xylem embolism (Anderegg et al., 2013b; Adams et al., 2017; Choat et al., 2018) or changes in wood anatomical features (e.g., lower lumen area; Here¸s et al., 2014; Pellizzari et al., 2016) may indicate impending tree death. In association with low whole-plant conductivity, reduced carbon assimilation and depletion of stored carbohydrates may also occur due to the decline in stomatal conductance and leaf area, particularly for gymnosperms (Galiano et al., 2011; Pangle et al., 2015; Adams et al., 2017). The determination of such mechanistic indicators is, however, costly, and temporally and spatially limited. Therefore, other approaches have been used to identify changes in tree health and mortality risk, such as temporal changes in crown

defoliation (Dobbertin and Brang, 2001), or in radial growth rates (e.g., Pedersen, 1998; Bigler and Bugmann, 2004; Dobbertin, 2005; Camarero et al., 2015; Hülsmann et al., 2018). Ring-width (RW) data are especially suitable, as they provide retrospective and long-term information about tree radial growth at an annual resolution, and can be applied effectively at an affordable cost to a large number of trees, sites, and species.

A recent synthesis reported either abrupt or long-term reduction in growth rates before death in most tree mortality events recorded in dendrochronological studies worldwide (Cailleret et al., 2017). However, this decrease in growth before mortality was not ubiquitous, and its detection was subject to important methodological constraints, especially related to the sampling design (Cailleret et al., 2016). Therefore, additional metrics that go beyond changes in absolute growth rates are needed to identify individuals at high risk of mortality. Earlywarning signals (EWS) have been proposed to characterize (ecological) systems that are approaching a critical transition, i.e., a sudden and persistent shift in a system's state (Scheffer et al., 2009). EWS are caused by the gradual decrease in the recovery rate of a system after a perturbation – called "critical slowing down" (Wissel, 1984) – and have been identified prior to population extinction in experiments under increasing levels of stress (e.g., Drake and Griffen, 2010; Dai et al., 2012; Veraart et al., 2012). Tree death can be considered as system failure (Anderegg et al., 2012), and can be viewed as a critical transition caused by the combined changes in the intensity, frequency and duration of stress factors (Dakos et al., 2015), and high sensitivity of the tree to these specific stresses (Brandt et al., 2017). This would be somewhat analogous to recent applications of critical transitions theory to human physiology, where health failures at the individual level can be anticipated with EWS (Olde Rikkert et al., 2016). In fact, the growth rate decline observed in most trees before mortality may be typical of such "critical slowing down" phenomenon, which can be captured by an increase in temporal autocorrelation and variance in time series of variables reflecting the functioning of the system (Scheffer et al., 2009; Dakos et al., 2012b), and by a decrease in their synchrony with the environment. These EWS would, respectively, reveal that the state of the system at any given moment becomes more and more like its recent past state, increasingly affected by shocks, and less able to track the environmental fluctuations (Scheffer et al., 2009).

Several studies have reported that RW time series of dying or declining individual trees tend to show increasing temporal autocorrelation and variance over time or higher values than surviving individuals (e.g., Ogle et al., 2000; Suarez et al., 2004; Millar et al., 2007; Kane and Kolb, 2014; Camarero et al., 2015; see **Supplementary Appendix A**), especially in the case of drought-induced mortality (McDowell et al., 2010; Heres et al., 2012; Gea-Izquierdo et al., 2014; Macalady and Bugmann, 2014). However, it remains unclear whether rising growth variance and autocorrelation can be used as EWS for tree mortality. First, other studies have reported opposite trends (e.g., Pedersen, 1998; Millar et al., 2012), or contrasting results depending on the study species (Camarero et al., 2015), sites (Ogle et al., 2000), and tree size (Herguido et al., 2016). Second, finding a common trend comparing results across different case studies can be difficult, as methodologies vary among studies, especially for the quantification of the inter-annual variability in growth. This aspect is fundamental, as opposite relationships could be obtained when using the standard deviation (SD) or the mean sensitivity (i.e., the mean relative change in RW between two consecutive rings; see Bunn et al., 2013) to characterize year-toyear variability in RW series (Gillner et al., 2013; Macalady and Bugmann, 2014). Similarly, Camarero et al. (2015) did not find any consistent change in growth synchrony between declining and healthy trees among species.

Here, we tested whether EWS based on annual radial growth data can be used as universal indicators of tree mortality. We used a unique, pan-continental database that contains paired growth time series for dead and surviving trees from nearly 200 sites, including data for 13 angiosperm and 23 gymnosperm species. In particular, we measured temporal changes in tree growth variance, temporal autocorrelation, and synchrony (correlation among trees) after removing any effect driven by changes in absolute growth rates, which had been studied in a previous publication (Cailleret et al., 2017). We analyzed temporal changes in the properties of RW chronologies of individual trees that died during a given stress event (diachronic approach on dying trees), and compared the resulting patterns to those from trees that survived this specific event (synchronic approach). Contrary to standard tree growth analysis that explores trends in RW chronologies, our approach here is to estimate changes in the dynamic properties of these time series (e.g., autocorrelation structure) that can be used as proxies of tree mortality risk. The methodology we develop may assist in using such proxies for assessing individual tree resilience.

# MATERIALS AND METHODS

## Tree-Ring Width Chronologies

We used the pan-continental tree-ring width (mm) database compiled by Cailleret et al. (2017), which includes 58 published and unpublished datasets for which (i) both dying and surviving trees growing together at the same site were cored, (ii) all individual chronologies had been successfully cross-dated, (iii) mortality was proximally induced by stress (e.g., drought, competition, and frost) and biotic agents in an endemic phase (e.g., bark beetles, defoliator insects, fungi, acting as predisposing or contributing factor), and not by abrupt abiotic disturbances such as windthrow, fire, or flooding, which may kill trees irrespective of their vitality and previous growth patterns (but see Nesmith et al., 2015). We grouped the datasets into four groups according to the main mortality sources determined by the authors of each study: (i) 'drought' corresponds to mortality caused by a single or several drought events without obvious impact of biotic agents; (ii) 'biotic' includes sites in which mortality was induced primarily by biotic factors, including barkbeetles, defoliator insects, and/or fungal infection; (iii) 'drought and biotic' when the impact of biotic agents (including mistletoes and wood-borers) was associated with drought; (iv) and the group 'others' includes snow break, frost events, high competition

intensity, and cases in which mortality were not evident or not specified.

The database analyzed here slightly differs from Cailleret et al. (2017) as some sites for which we previously did not find any pair of dying/surviving tree with similar diameter at breast height (DBH) are considered in the present analysis, and as we excluded trees with less than 20 measured rings (see below). A total of 36 gymnosperm and angiosperm species were studied, with an overrepresentation of gymnosperms (64% of the species and 86% of the sites). Pinaceae was the most represented family, followed by Fagaceae. Overall, the dataset analyzed in the main text included 3065 dead trees and 4389 living trees growing at 198 sites mostly in boreal, temperate, and Mediterranean biomes of North America and Europe. More details on the sampling methods and on the assessments of the mortality sources, tree cambial age, DBH, and the year of death are available in **Supplementary Appendix B** and in Cailleret et al. (2017).

#### Growth Metrics

Following Dakos et al. (2012a) and Camarero et al. (2015), we estimated levels and trends of Standard Deviation (SD) and first-order autocorrelation (AR1) in detrended RW time series of individual trees (**Figure 1**). Contrary to most dendrochronological studies, where AR1 is calculated using raw RW time series (e.g., Martín-Benito et al., 2008; Esper et al., 2015; Hartl-Meier et al., 2015), chronologies were detrended to correct for decadal to centennial trends, including decadal decreases in growth rates that are commonly observed prior to mortality (Cailleret et al., 2017). Such negative growth trends would automatically lead to increasing trends in AR1 before tree death (**Figure 2B** and **Supplementary Appendix C**), irrespective of the potential intrinsic change in the AR1 properties related to changes in tree vitality. In addition, we calculated the Pearson correlation (COR) coefficient between individual time series and the site chronology (**Figure 1**). In contrast to the study by Camarero et al. (2015), where COR coefficients corresponded to the correlations between separated mean chronologies of 'declining' and 'non-declining' trees, we analyzed COR values between each individual detrended time series of dying trees and the corresponding site- and species-specific chronology (including both dying and surviving trees), to reduce potential biases at sites where few living trees had been sampled. Site chronologies were derived using the bi-weight robust mean of the individual residual chronologies (**Figure 1**) to reduce the importance of outliers. This is particularly important when sample size is low, which is the case for some of our sites (**Supplementary Appendix B**).

As we aimed at analyzing temporal changes in growth SD and AR1, and at comparing them among trees with different ages, sizes, or growth rates, two precautionary measures were taken to detrend the RW data. (1) Most tree-ring- based studies remove size-effects on the RW data while keeping climate-induced decadal to centennial changes in growth rates using negative exponential curves or using the Regional Curve Standardization method (e.g., Peters et al., 2015; Büntgen et al., 2017). In contrast, we used smoothing splines which are more flexible and more adapted to remove decadal trends (Cook and Peters, 1997). As SD and AR1 values are highly sensitive to the bandwidth of the Gaussian kernel regression (see **Supplementary Appendix D**), this one was fixed at 15 years rather than proportional to the length of the time-series. Indeed, the latter approach would bias the comparison among trees with different length of the timeseries (∼different ages). As we specifically focused on the end of the RW time series, our analysis is prone to edge-effects that can emerge from Gaussian detrending (e.g., D'Arrigo et al., 2008; see **Supplementary Appendix E**). Thus, the sensitivity of our results to the bandwidth length was also assessed (**Supplementary Appendix D**). (2) We used residuals (differences between the original (raw) RW data and the smoothing spline from the Gaussian kernel regression) rather than ratios as done in traditional dendrochronological studies. In this way, the output chronology is centered on zero, is still heteroscedastic, and does not include annual outliers when RW is close to zero, which often occurs in dying trees. In contrast, most dendrochronological studies using RW data calculate ratios to get series that are centered on one and are assumed to be homoscedastic (see Cook and Peters, 1997; Büntgen et al., 2005; Frank et al., 2006; **Supplementary Figure C2**). To detect short-term (∼decadal) but still robust changes in growth metrics, SD, AR1 and COR were calculated within a 20-year moving time-window (hereafter SD20, AR120, and COR20). Trees with fewer than 20 rings were thus discarded from this analysis. Other lengths of the moving timewindow were tested and showed similar results (**Supplementary Appendix F**).

### Detecting Trends in Growth Metrics Before Tree Mortality

Our dataset allowed us to follow two approaches for estimating EWS that helped us to increase the robustness of our conclusions and to assess potential methodological biases. The first approach was based on the analysis of the temporal changes in growth patterns of dying trees (diachronic approach), and the second on the comparison between dying and surviving individuals coexisting at the same site (synchronic approach).

#### Temporal Change in Growth Metrics of Dying Trees

For each of the 3065 dying trees, we calculated SD20, AR120, and COR<sup>20</sup> until the last year with complete ring formation, i.e., the year before tree death. We determined whether absolute values in SD20, AR120, and COR<sup>20</sup> calculated during the last 20 years preceding mortality (SD20<sup>f</sup> , AR120<sup>f</sup> , and COR20<sup>f</sup> for final values) were significantly different than those during any other previous 20-year period.

As SD<sup>20</sup> calculated on the detrended chronology was still positively related to mean growth rate calculated over the same period (meanRW20; see **Supplementary Appendix C**), we did not directly analyze this metric, but instead we analyzed the residuals of a linear mixed-effect model (LMM) fitted to the overall dataset with meanRW<sup>20</sup> as a fixed explanatory variable. The same approach was used for AR1<sup>20</sup> and COR<sup>20</sup> to center them on zero, which allows for an easier comparison among trees, species, and periods with different mean growth rates. This is especially important as our sampling is not equal in terms of mean tree age per species, which could lead to problems when

2014). The Standard Deviation (SD), first-order autocorrelation (AR1) and Pearson correlation coefficients (COR) were calculated on the original (Left) and detrended (Right) RW data using 20-year moving time windows.

averaging results to analyze the overall temporal dynamics in growth metrics. Bootstrap resampling procedures were then used to test if the LMM residuals for SD20<sup>f</sup> , AR120<sup>f</sup> , and COR20<sup>f</sup> significantly differed from zero (500 re-samplings).

SD<sup>20</sup> and meanRW<sup>20</sup> were log-transformed unlike AR1<sup>20</sup> and COR<sup>20</sup> values because their distributions were normal. As each tree species may have different SD and AR1 values for a similar growth rate (e.g., higher AR1 values are expected for evergreen species; Anderegg et al., 2015b), and COR values may depend on the number of trees used to derive the reference chronology, random effects were estimated for the intercept and the slope with species crossed with site as a grouping factor.

#### Differences in Growth Metrics Between Conspecific Dying and Surviving Trees

Although RW data were detrended using Gaussian filtering before calculating SD20, AR120, and COR20, temporal changes in these metrics could be affected by site-specific decadal-scale changes in environmental conditions (e.g., change in climatic conditions or in canopy dynamics; Brienen et al., 2006; Carrer and Urbinati, 2006; Esper et al., 2015), regardless of individual intrinsic changes in tree vitality. Thus, to account for this possibility, we compared SD20<sup>f</sup> , AR120<sup>f</sup> , and COR20<sup>f</sup> between conspecific dying and surviving trees for each mortality event, i.e., for each combination of species, site, and mortality year (see Cailleret et al., 2017).

For each dying tree, two approaches were followed for selecting comparable conspecific surviving trees from the same site: we only considered trees (i) with a similar DBH at the given mortality year (difference in final DBH between dying and surviving trees diffD−SDBH<sup>f</sup> ≤ 2.5 cm), or (ii) with a similar mean RW during the 20-year period before the mortality year (diffD−SmeanRW20<sup>f</sup> ≤ 5%). In cases where none of the surviving trees fulfilled this condition, the corresponding dying tree was discarded. Following these two approaches, we considered 2887 (94.2% of the dying trees) and 2093 (68.3%) pairs of trees, respectively. On the one hand, comparing trees with similar DBH removes both geometric and structural (∼size) effects (see Bowman et al., 2013). For instance, large and dominant trees tend to show more plastic growth than small and suppressed ones (Martín-Benito et al., 2008; Mérian and Lebourgeois, 2011). On the other hand, comparing trees with similar mean RW removes mathematical effects related to changes in growth rate (see **Supplementary Appendix C**), and allows us to detect the presence of growth-based EWS in case of unchanging growth level before tree death (relative to the surviving trees). Thus, these two sampling approaches may individually bias the results, but they are complementary and should be considered together.

On both datasets, we analyzed if the differences in SD20<sup>f</sup> , AR120<sup>f</sup> , and COR20<sup>f</sup> between conspecific dying and surviving trees (diffD−SSD20<sup>f</sup> , diffD−SAR120<sup>f</sup> , and diffD−SCOR20<sup>f</sup> ) were significantly different from zero for all species groups and mortality sources using LMMs and bootstrapping methods. For each of these response variables, we fitted a LMM considering the species group and mortality source as interactive fixed effects. As size or geometric effects could remain, we also included the difference in final mean RW (diffD−SRW20<sup>f</sup> ) and in DBH (diffD−SDBH<sup>f</sup> ) as fixed effects. Random effects were estimated for the intercept with species crossed with site as grouping factor. Direct age effects were not considered here assuming that senescence only marginally affects tree function (Mencuccini and Munné-Bosch, 2017). LMMs were finally used to predict diffD−SSD20<sup>f</sup> , diffD−SAR120<sup>f</sup> , and diffD−SCOR20<sup>f</sup>

values in the theoretical situation in which dying trees have similar meanRW20<sup>f</sup> and DBH<sup>f</sup> as surviving ones.

#### Sampling Scheme

fpls-09-01964 January 16, 2019 Time: 14:15 # 7

To account for the heterogeneity in the number of dying trees per site and per species in the dataset, we used two resampling procedures (Cailleret et al., 2017). First, we randomly sampled with replacement the same number of dying trees (diachronic approach) or the same number of dying-surviving pairs (synchronic approach) for each of the 36 species. Second, a similar approach was followed to provide the same weight in the calibration dataset for each of the 198 sites. With both approaches, each species or each site contributes equally to the results, which minimizes the bias related to under-sampling or over-sampling of specific sites or species (**Supplementary Appendix G**).

#### Theoretical Expectations

Finally, to detect which combinations of temporal trends in SD and AR1 can be expected when growth rates gradually decrease (commonly reported for dying trees), we generated theoretical RW time series based on simple growth models that included (i) an autocorrelation component, (ii) a long-term change in the mean, and (iii) some noise reflecting the environmental stochasticity (**Supplementary Appendix E**).

The calculation of moving SD20, AR120, and COR<sup>20</sup> values, and LMM analyses were performed using the packages earlywarnings (Dakos et al., 2012a), lme4 (Bates et al., 2014), and lmerTest (Kuznetsova et al., 2017) of the open-source software R (R Core Team, 2017).

#### RESULTS

#### Temporal Changes in Growth Metrics of Dying Trees

SD<sup>20</sup> calculated on detrended RW data started decreasing around 30 years before tree death (**Figure 2A**). This trend in SD<sup>20</sup> was related to the general reduction in mean RW, as both variables are highly correlated (**Supplementary Appendix C**). After removing the effect of the mean RW using a LMM, SD residuals revealed an increase in inter-annual variability of RW before trees died (**Figure 2A**). The variability calculated for the 20-year period before mortality (resSD20<sup>f</sup> ) was generally higher than during the rest of the lives of dying trees (**Figure 3**). For gymnosperms, this pattern was significant irrespective of the mortality cause and of the method used to account for the heterogeneity in sample properties (**Figure 3A** and **Supplementary Appendix G**). In addition, the increase in variability was even stronger in the last 10-year period before mortality (**Supplementary Appendix F**). Results were less clear for angiosperms. Although variability was generally significantly higher at the end of an angiosperm's life, this pattern was not present for all sources of mortality (e.g., when mortality was caused by both drought and biotic agents, **Figure 3A**), and resSD<sup>20</sup> did not monotonically increase toward the end of a tree's life (**Supplementary Figure G1B**).

The first-order autocorrelation increased on average before tree death both in detrended RW chronologies (AR120) and in the residuals of the LMMs (resAR120) (**Figure 2B**). In fact, the residual AR1 (after removing both growth level and trend effects, **Supplementary Appendix C**) was higher than zero in the final 20-year period preceding tree death (resAR120<sup>f</sup> ; **Figure 3B**). However, this was mostly true for gymnosperms (except when mortality was caused by both drought and biotic agents in samples including equal number of dying trees per species; **Supplementary Appendix G**), and such level of positive resAR1<sup>20</sup> values was not exclusive to the end of a

gymnosperm's life (**Supplementary Figure G1C**). Thus, the high AR1 values calculated during the 20-year period before gymnosperm mortality should not be interpreted as an exclusive response indicative of impending tree death. In the case of angiosperms, no significant or monotonic change in resAR1<sup>20</sup> was observed consistently before mortality (**Figure 3B** and **Supplementary Figure G1D**).

On average, Pearson correlations calculated between individual RW time series of dying trees and site chronologies decreased gradually ca. 30 years before death (**Figure 2C**). However, residual correlation values (resCOR20; after correcting for mean RW, **Supplementary Appendix C**) were not consistently below zero or lower than any previous period across mortality sources, species groups, or sampling strategies (**Figure 3C** and **Supplementary Appendix G**). The only exceptions were mortality caused by both drought and biotic agents for angiosperms and mortality caused by other factors in gymnosperms (**Figure 3C** and **Supplementary Figure G2**).

# Differences in Temporal Changes of Growth Metrics Between Conspecific Dying and Surviving Trees

Dying trees generally showed higher variability in growth in the last 20 years of their lives compared to surviving trees. Estimated differences in variance between dying and surviving trees (diffD−<sup>S</sup> SD) based on LMMs adjusted for growth rate (meanRW20<sup>f</sup> ) and size effects (DBH<sup>f</sup> ) were significantly higher than zero in most cases for both angiosperms and gymnosperms and across mortality drivers, except when trees were killed by biotic agents (**Figures 4A,B**). This result was generally robust to different sampling schemes (unbalanced original dataset in **Figure 4** vs. equal weight among species or sites in **Supplementary Appendix G**). Dying gymnosperms showed more consistent effects, although the magnitude of the SD difference between dying and surviving trees was generally higher for angiosperms (**Figures 4A,B**).

Contrary to variance, autocorrelation did not significantly differ between dying and surviving trees. In specific cases, differences were significantly higher than zero (e.g., gymnosperms for drought-induced mortality and pairing by meanRW20<sup>f</sup> ), but this was never consistent across mortality drivers or sampling schemes (**Figures 4C,D** and **Supplementary Appendix G**).

Finally, we found predominantly lower COR20<sup>f</sup> for dying trees than surviving ones (**Figures 4E,F**). This pattern was largely consistent and of similar magnitude for every mortality source for gymnosperms, but it was less clear for angiosperms, as some differences in correlation (e.g., when biotic agents were the main mortality source) strongly depended on the sampling strategy, i.e., on the species and sites considered (**Supplementary Appendix G**).

# DISCUSSION

We found a gradual increase in inter-annual growth variability and a decrease in growth synchrony during the ∼20-year period before mortality. These trends were more robust for gymnosperms than for angiosperms, irrespective of the main cause of mortality. However, this result only partly conforms to the patterns that are expected to characterize systems prior to transitions due to critical slowing down (Scheffer et al., 2009; Dakos et al., 2012b), as no consistent changes in growth autocorrelation was detected for either taxonomic group.

# Mechanisms Underlying the Differences Between Angiosperms and Gymnosperms

The increase in growth variance (for a given growth level) of dying gymnosperms may indicate an increase in susceptibility to external influences such as climatic factors or pathogen diseases (e.g., Csank et al., 2016; Timofeeva et al., 2017). In addition, their growth seems to be less coupled to high-frequency climate fluctuations than surviving gymnosperms, as revealed by the decrease in growth synchrony with the overall site chronology (Fritts, 1976; Boden et al., 2014). Both changes may be associated with small-scale differences in atmospheric conditions and in water availability that may become more important under stress, and with alterations in carbon allocation patterns, which may reflect the higher sensitivity of gymnosperms' carbon economy to stress events (Adams et al., 2017). Some studies have shown stronger stomatal control and reduced non-structural carbohydrate (NSC) concentrations in tissues of dying conifers, relative to coexisting surviving individuals (Galiano et al., 2011; Timofeeva et al., 2017). For instance, Pinus sylvestris saplings survived experimental drought longer when keeping assimilation rates relatively high, even at the expense of higher water loss (Garcia-Forner et al., 2016). Associated changes in xylogenesis phenology are also likely to be important. Compared to healthy trees, defoliated pines showed a delay in the onset and reduction in the duration of cambial activity (Guada et al., 2016). Such physiological responses could explain the observed higher growth variability in dying trees that goes along with a different synchrony relative to surviving individuals.

In contrast, no consistent increase in growth variance was observed for angiosperms. This is in line with reported small and short-term reductions in tree growth before angiosperm death (Cailleret et al., 2017). Several reasons may explain the lack of growth-based signals in angiosperms, including greater functional diversity (Augusto et al., 2014), speciesdependent responses to tree size compared to gymnosperms (Steppe et al., 2011), the relatively loose coupling between hydraulic failure and carbon depletion during drought (Adams et al., 2017), and their high recovery rates once favorable environmental conditions prevail after drought (Augusto et al., 2014; Anderegg et al., 2015b; Yin and Bauerle, 2017). Compared with gymnosperms, angiosperms generally have a higher capacity to (i) store NSC in their wood parenchyma (Plavcová et al., 2016), (ii) rebuild NSC pools owing to their higher stomatal conductance (Lin et al., 2015) and growth efficiency, and (iii) replace conducting area via new xylem growth (Brodribb et al., 2010), resprouting (Zeppel et al., 2015), and potentially by refilling embolized xylem conduits (Johnson et al., 2012). In

addition, all gymnosperms studied are evergreen species, whereas most analyzed angiosperms are deciduous (except Nothofagus betuloides, Nothofagus dombeyi, and Tamarix chinensis) which may make them less dependent on previous-year leaf area and growth efficiency. The relatively low number of angiosperm species included in our study, together with the higher variation in leaf and growth strategies (e.g., diffuse- vs. ring-porous species) and in recovery performance across species relative to gymnosperms (Cailleret et al., 2017; Yin and Bauerle, 2017) may have also contributed to the lack of consistent increases in variance before tree mortality.

The lack of change in AR1 for both taxonomic groups may be explained by antagonistic effects of the stress-induced changes in key components of growth autocorrelation. On the one hand, the growth dependency on NSC reserves may induce lagged responses ('growth memory'; Schulman, 1956; Esper et al., 2015; Timofeeva et al., 2017; von Arx et al., 2017). On the other hand, reductions in hydraulic conductivity through xylem embolism and lower production of new functional xylem (Brodribb et al., 2010), as well as reductions in overall crown area, or in leaf size, number and longevity (Bréda et al., 2006; Girard et al., 2012; Jump et al., 2017), may reduce the importance of lag effects.

Finally, species-specific changes in water and carbon economy, during and after high stress levels (Galiano et al., 2017), can explain the lack of a consistent change in AR1 preceding tree death. For instance, after intense drought, carbon assimilates may be invested into storage and restoration of root functions rather than into stem growth (Palacio et al., 2012; Hagedorn et al., 2016; Martínez-Vilalta et al., 2016), and the allocation priority level varies among species (Galiano et al., 2017).

### Methodological Considerations

Our results did not agree with some previous studies that showed that declining/dying trees had higher radial growth variance, autocorrelation, and synchrony than healthy/surviving ones, or showed an increase of these growth metrics before death (e.g., Sánchez-Salguero et al., 2010; Amoroso et al., 2012; Camarero et al., 2015; Cailleret et al., 2016). They also indicate that the contrasting results obtained among previous studies (**Supplementary Appendix A**) may be due to methodological choices. In addition to the prescriptions that are inherent to the characteristics of our database, e.g., regarding the inequality in sample sizes among sites and species (**Supplementary Appendix G**), or the potential biases related to the assessment of the year of tree death (see Bigler and Rigling, 2013) or to the measurement of narrow rings, there are three particularly important elements to consider, which we discuss in the following paragraphs.

First, if one aims at understanding the ecological mechanisms behind changes in the variance (quantified here with SD) and autocorrelation of ring-width chronologies, the effects of tree size, growth level, and growth trend should be removed or accounted for. All these growth-related metrics are highly intercorrelated (**Supplementary Appendix C**), which can lead to a misinterpretation of the results. For instance, the decrease in SD<sup>20</sup> calculated on raw RW data before tree death was caused by the gradual decrease in RW increment, and thus did not indicate an intrinsic decrease in growth sensitivity to inter-annual changes in environmental conditions (**Figure 2A**). Four procedures can be used to account for these effects: (i) detrending the RW time series to remove part of the low- and medium-frequency fluctuations, (ii) removing the remaining effects of growth rate on the composite SD, AR1 and COR individual time series using mixed-effects models, (iii) comparing dying and surviving trees with similar size or growth rate, and (iv) including the remaining differences in size and growth rate between dying and surviving trees of a given pair as an additional explanatory variable in the statistical models. As in all dendrochronological analyses, the detrending method should be carefully selected (e.g., Esper et al., 2015). For instance, the bandwidth of the kernel regression smoother should be constant among trees and should have an adequate length to capture enough medium-frequency (∼decadal-scale) variability (**Supplementary Appendix D**) while minimizing end-effect biases (**Supplementary Appendix E**). Also, and in contrast to classical dendroclimatic studies that aim at getting homoscedastic growth time series by calculating ratios (Cook and Peters, 1997; Frank et al., 2006), the heteroscedasticity of growth residuals needs to be retained. As using one or the other approach may lead to opposite trends (**Supplementary** **Appendix C**), differences are to be preferred over ratios (see also Scheffer et al., 2009; Dakos et al., 2012a).

Second, it is always advisable to combine both diachronic and synchronic approaches to control for potential biases that are typical of field data; i.e., to focus on the temporal change in growth metrics of dying trees before they actually die, and on the comparison between coexisting trees that died and survived a specific mortality event (see also Gessler et al., 2018). Still, the synchronic approach is prone to artifacts, due to the fact that the group of 'surviving' trees at a given mortality event, which are used as a control, may include trees that died shortly after the stress event. On the other hand, using the diachronic approach only is not sufficient to disentangle changes in growth patterns that are caused by variations in tree functions or in environmental conditions (e.g., mortality of neighbors). For instance, first-order temporal autocorrelation calculated for the 20-year period before the death of gymnosperms (AR120<sup>f</sup> ) was generally higher than average AR1<sup>20</sup> (**Figure 3B**), which could indicate that high AR1 is associated with impending tree death. However, it cannot be used as a predictive tool, as high AR1 values were also observed during other periods of the trees' lives, and because conspecific trees that survived the mortality event showed similar AR120<sup>f</sup> values (**Figures 4C,D**).

Third, the unexpected lack of significant and meaningful differences in growth-based EWS among the mortality groups considered here (see Cailleret et al., 2017) highlights the need for a more precise determination of the mortality source(s) in the field. It is now well accepted that tree mortality is a phenomenon induced by multiple biotic and abiotic drivers with strong interdependencies (Manion, 1991; Anderegg et al., 2015a), and rarely occurs because of one single factor. Trees in the 'drought' category might actually belong in 'drought-biotic,' and trees in the 'others' category might belong in the 'biotic agents' category (Das et al., 2016). In addition to information on climate, soil, and stand characteristics, detailed pathological data would be highly needed as biotic factors are involved in many individual mortality reports (Das et al., 2016).

# Application of Early-Warning Signals of Tree Mortality Based on Radial Growth

Our results expand previous assessments of the association between tree radial growth and mortality risk based on the direct effects of (absolute) growth rates (cf. Cailleret et al., 2017) by focusing on subtler properties of the growth time series. Overall, we found that an increase in inter-annual growth variability and a low growth synchrony could be used as EWS of gymnosperm mortality. Because these results were clear even after accounting for any indirect effect driven by changing growth levels, high growth variability and low synchrony could be used as independent diagnostics to identify gymnosperm trees or populations at high risk of mortality. However, these trends were much less consistent for angiosperms, and we did not find significant changes in autocorrelation prior to mortality. Hence, our results do not support the idea that critical slowing down indicators in radial growth data can be used as universal early warnings for tree mortality.

There are many reasons why early-warning indicators based on radial growth metrics may not be accurate indicators of stress-induced tree mortality. First, although we did not detect any consistent difference in growth metrics between mortality sources, some types of mortality stress may be too abrupt to be reflected in gradual changes in tree-ring width, and can occur without previous warning. For example, fungal diseases, barkbeetle outbreaks, or intense droughts can kill trees irrespective of their vitality, or at least, irrespective of their previous radial growth (Cherubini et al., 2002; Raffa et al., 2008; Sangüesa-Barreda et al., 2015; Cailleret et al., 2017). Second, for a similar stress event, there is a large variety in the type and timing of responses among and within species (Jump et al., 2017) that may confound detection of common changes in growth sensitivity. Third, annual radial growth may not be the most appropriate variable to derive such early warnings, as it is not only dependent on tree carbon and water status, but also on the environmental influences on sink activity (Körner, 2015). Other xylem-based physiological, anatomical, hydraulic, and isotopic properties that can be measured in tree rings may provide complementary information on tree mortality probability (e.g., Here¸s et al., 2014; Anderegg et al., 2016a; Csank et al., 2016; Pellizzari et al., 2016; Timofeeva et al., 2017; Gessler et al., 2018). Fourth, despite recent developments (Gea-Izquierdo et al., 2015; Schiestl-Aalto et al., 2015; Guillemot et al., 2017), we lack mechanistic models of cambial activity based on sink demand, carbon uptake and reserves and water relations, which can go beyond simplistic formulations to produce clear expectations of ring-width dynamics before mortality (cf. **Supplementary Appendix E**). Finally, depending on which state variable(s) are affected by the environmental 'noise' and by the change in tree vitality, the temporal trends in AR1 and in SD prior to the transition can vary (Dakos et al., 2012b). For instance, the simple autoregressive models we developed to simulate decreasing growth rate over time, highlighted that all combinations of SD and AR1 trends can theoretically occur (**Supplementary Appendix E**). Considering that climate modifies tree growth based on multiple direct and indirect pathways (e.g., via changes in cambial activity and in the water and carbon economy), the relationship between climate variability and growth autocorrelation and variance is not straightforward. Similarly, the SD metric integrates both tree resistance and recovery to specific events that could be independently analyzed (Lloret et al., 2011; Dakos et al., 2015).

Climate change is predicted to modify mean temperature and precipitation, but also to increase the inter-annual variability and persistence of climatic fluctuations (Fischer et al., 2013; Lenton et al., 2017), and to modify the population dynamics of biotic agents (Allen et al., 2015). Several physiological thresholds can be exceeded during extreme biotic or abiotic conditions (e.g., during drought; Adams et al., 2017), which may ultimately lead to individual tree mortality, and potentially to widespread forest decline in many regions (Lloret et al., 2012; Reyer et al., 2013; Allen et al., 2015). However, we still lack a general set of mechanistic and empirical EWS of tree mortality at the individual scale (Gessler et al., 2018) that could be used to complement the signals used for detecting dieback at the forest stand or landscape scales (e.g., Verbesselt et al., 2016; Rogers et al., 2018). Based on a rich pan-continental ring-width database of dying and surviving trees, and by combining diachronic and synchronic approaches, our results highlight that in addition to the analysis of the multiannual growth rates and trends (Cailleret et al., 2017), the interannual variability of the growth time series can be used to assess mortality risk, particularly for gymnosperm species.

### AUTHOR CONTRIBUTIONS

MC, VD, and JM-V conceived the ideas and designed the methodology. MC, TA, MA, JA, CB, HB, J-JC, PC, MRC, KC, ˇ AD, HD, GG-I, SG, LH, HH, A-MH, KH, PJ, JK, VK, TKi, TKl, TL, J-CL, FL, HM, IM, JM, WO, AP, AMP, BR, GS-B, JS, AS, DS, M-LS, MS, VT, RV, AW, PW, and JM-V collected the treering data. MC, SJ, ER, and JM-V compiled and cleaned the ring-width database. MC analyzed the data and led the writing of the manuscript with inputs from VD and JM-V. All authors contributed critically to the drafts and gave final approval for publication.

## ACKNOWLEDGMENTS

This study generated from the COST Action STReESS (FP1106) financially supported by the EU Framework Programme for Research and Innovation Horizon 2020. We would like to thank Don Falk (University of Arizona) and two reviewers for their valuable comments, all the colleagues for their help while compiling the database, and Louise Filion, Michael Dorman, and Demetrios Sarris for sharing their datasets. MC was funded by the Swiss National Science Foundation (project number 140968). ER was funded by the Research Foundation – Flanders (FWO, Belgium) and got support from the EU Horizon 2020 Programme through a Marie Skłodowska-Curie IF Fellowship (No. 659191). KC was funded by the Slovenian Research Agency ˇ (ARRS) Program P4-0015. IM was funded by National Research, Development and Innovation Office, project number NKFI-SNN-125652. AMP was funded by the Ministry of Research and Innovation, CNCS – UEFISCDI, project number PN-III-P1-1.1-TE-2016-1508, within PNCDI III (BIOCARB). GS-B was supported by a Juan de la Cierva-Formación grant from MINECO (FJCI 2016-30121). DS was funded by the project III 43007 financed by the Ministry of Education and Science of the Republic of Serbia. AW was funded by Canada's Natural Sciences and Engineering Research Council and Manitoba Sustainable Development. JM-V benefited from an ICREA Academia Award. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the United States Government.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018.01964/ full#supplementary-material

#### REFERENCES

fpls-09-01964 January 16, 2019 Time: 14:15 # 12



drought-induced tree and forest mortality. New Phytol. 218, 15–28. doi: 10. 1111/nph.15048


Manion, P. D. (1991). Tree Disease Concepts. Englewood Cliffs, NJ: Prentice-Hall.



increases understorey solar input regionally: primary and secondary ecological implications. J. Ecol. 99, 714–723. doi: 10.1111/j.1365-2745.2011.01804.x


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

The handling Editor declared a past co-authorship with the authors JC, PC, and KC. ˇ

Copyright © 2019 Cailleret, Dakos, Jansen, Robert, Aakala, Amoroso, Antos, Bigler, Bugmann, Caccianaga, Camarero, Cherubini, Coyea, Cufar, Das, Davi, ˇ Gea-Izquierdo, Gillner, Haavik, Hartmann, Here¸s, Hultine, Janda, Kane, Kharuk, Kitzberger, Klein, Levanic, Linares, Lombardi, Mäkinen, Mészáros, Metsaranta, Oberhuber, Papadopoulos, Petritan, Rohner, Sangüesa-Barreda, Smith, Stan, Stojanovic, Suarez, Svoboda, Trotsiuk, Villalba, Westwood, Wyckoff and Martínez-Vilalta. 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.

# Projecting Tree Species Composition Changes of European Forests for 2061–2090 Under RCP 4.5 and RCP 8.5 Scenarios

Allan Buras1,2 \* and Annette Menzel1,3

<sup>1</sup> Professorship of Ecoclimatology, Technische Universität München, Freising, Germany, <sup>2</sup> Land-Surface-Atmosphere-Interactions, Technische Universität München, Freising, Germany, <sup>3</sup> Institute of Advanced Study, Technische Universität München, Garching, Germany

#### Edited by:

Veronica De Micco, University of Naples Federico II, Italy

#### Reviewed by:

Martina Pollastrini, Università degli Studi di Firenze, Italy Minhui He, Northwest Institute of Eco-Environment and Resources (CAS), China

> \*Correspondence: Allan Buras allan@buras.eu

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 05 November 2018 Accepted: 20 December 2018 Published: 11 January 2019

#### Citation:

Buras A and Menzel A (2019) Projecting Tree Species Composition Changes of European Forests for 2061–2090 Under RCP 4.5 and RCP 8.5 Scenarios. Front. Plant Sci. 9:1986. doi: 10.3389/fpls.2018.01986 Climate change poses certain threats to the World's forests. That is, tree performance declines if species-specific, climatic thresholds are surpassed. Prominent climatic changes negatively affecting tree performance are mainly associated with so-called hotter droughts. In combination with biotic pathogens, hotter droughts cause a higher tree vulnerability and thus mortality. As a consequence, global forests are expected to undergo vast changes in the course of climate change. Changed climatic conditions may on the one hand locally result in more frequent dieback of a particular tree species but on the other hand allow other—locally yet absent species—to establish themselves, thereby potentially changing local tree-species diversity. Although several studies provide valuable insights into potential risks of prominent European tree species, we yet lack a comprehensive assessment on how and to which extent the composition of European forests may change. To overcome this research gap, we here project future tree-species compositions of European forests. We combine the concept of climate analogs with national forest inventory data to project the tree-species composition for the 26 most important European tree species at any given location in Europe for the period 2061–2090 and the two most relevant CMIP5 scenarios RCP 4.5 and RCP 8.5. Our results indicate significant changes in European forests species compositions. Species richness generally declined in the Mediterranean and Central European lowlands, while Scandinavian and Central European high-elevation forests were projected an increasing diversity. Moreover, 76% (RCP 4.5) and 80% (RCP 8.5) of the investigated locations indicated a decreasing abundance of the locally yet most abundant tree species while 74 and 68% were projected an increasing tree-species diversity. Altogether, our study confirms the expectation of European forests undergoing remarkable changes until the end of the 21st century (i.e., 2061–2090) and provides a scientific basement for climate change adaptation with important implications for forestry and nature conservation.

Keywords: tree-species vulnerability, climate-smart forests, forest-management adaptation, climate change, CMIP5 climate projections, climate analogs

**Abbreviations:** CMIP 5, Coupled Model Intercomparison Project Phase 5; RCP, relative concentration pathway.

# INTRODUCTION

fpls-09-01986 January 7, 2019 Time: 19:46 # 2

In the course of climate change, the World's forests will likely undergo significant changes. This is because the frequency of heat waves likely will increase (IPCC, 2014) which in combination with long-lasting drought spells results in so called "global change type droughts," "hotter droughts," or compound events and eventually an increased tree mortality (Breshears et al., 2005; Allen et al., 2010, 2015; Martínez-Vilalta et al., 2012; Cailleret et al., 2017; Zscheischler and Seneviratne, 2017; Buras et al., 2018; Choat et al., 2018; Zscheischler et al., 2018). Consequently, tree species distributional ranges and the composition of forests will likely change (Pastor and Post, 1988; Hogg and Hurdle, 1995; Sykes and Prentice, 1996; Rigling et al., 2013; Fekete et al., 2017; Scherrer et al., 2017). Because of changing distributional tree species ranges and typically long rotation periods (often more than 100 years), forests have to be adapted to anticipated climate conditions by altering management strategies and selecting better adapted and therefore more resilient tree species (Halofsky et al., 2018; Sousa-Silva et al., 2018). In this context, foresters, stakeholders, and policy makers are targeting at so-called climatesmart forests (Halofsky et al., 2018; Nabuurs et al., 2018).

Already nowadays, the impact of hotter droughts on forests is visible (Breshears et al., 2005; Allen et al., 2010, 2015; Cailleret et al., 2017). Regarding European forests, the economically most important and spatially most abundant tree species of Scots pine (Pinus sylvestris, L.) and Norway spruce [Picea abies, (L.) Karst.] are regionally experiencing increased mortality rates (Bigler et al., 2006; Kohler et al., 2010; Lévesque et al., 2013; Rigling et al., 2013; Zang et al., 2014; Huang et al., 2017; Martínez-Sancho et al., 2017; Rehschuh et al., 2017; Buras et al., 2018). But also the widespread European beech (Fagus sylvatica, L.) is likely to suffer (Scharnweber et al., 2011, 2013; Dulamsuren et al., 2017; Huang et al., 2017; Walentowski et al., 2017) while other abundant species such as oak [Quercus robur, L. and Quercus petraea, (Matt.) Liebl.] may benefit from climate change (Scharnweber et al., 2011, 2013; Perkins et al., 2018).

Given their dominance within European forest ecosystems as well as their economic importance, it is crucial to evaluate the vulnerability of these tree species to anticipated climate change. Moreover, since some of these species are likely to decline under climate change, the key question arises, which other tree species feature a higher resilience to anticipated climate conditions and thus may be considered suitable alternatives (Walentowski et al., 2017). Several case studies have already provided valuable insights on potential reactions of selected species to projected climate (Sykes and Prentice, 1996; Dulamsuren et al., 2010, 2017; Scharnweber et al., 2011, 2013; Lévesque et al., 2013; Rigling et al., 2013; Zang et al., 2014; Rehschuh et al., 2017; Walentowski et al., 2017; Buras et al., 2018). However, these studies lean on different methodological approaches ranging from tree-ring analyses (Scharnweber et al., 2011, 2013; Lévesque et al., 2013; Zang et al., 2014; Dulamsuren et al., 2017; Rehschuh et al., 2017; Buras et al., 2018), over common-garden experiments also known as provenance trials (Martinez-Meier et al., 2008; Huang et al., 2017), to various types of species-distribution models (Sykes and Prentice, 1996; Hijmans and Graham, 2006; Walentowski et al., 2017) and are often restricted to a certain region and/or a relatively small number of tree species.

Another means to estimate the suitability of a given tree species under different climate conditions is related to the concept of climate analogs. Climate analogs have previously been used to assess climate-change induced ecosystem and land-use vulnerability and land-use change as well as to identify tree species or provenances which are more resilient to anticipated climate change (Hogg and Hurdle, 1995; Ford et al., 2010; Leibing et al., 2013; Luedeling et al., 2014). The idea of climate analogs is to identify regions in space and time which feature more or less similar climatological properties in comparison to a selected location at a given time. Since these regions feature similar climatological properties, they likely also provide suitable growing conditions for the same tree-species assemblages, as long as the considered climate parameters are relevant for tree ecophysiology. Identifying regions which currently feature climate conditions that are analogous to the projected conditions of a given site and determining the tree species which successfully survive under these "future analogs" thus may allow for depicting those tree species which are able to cope with anticipated climate conditions.

Under this framework, we here combine the concept of climate analogs with an ensemble of downscaled climate projections (CMIP5) and a recently established European forest inventory (Mauri et al., 2017) to project tree species composition changes of European forests under RCP 4.5 and RCP 8.5 scenarios until the end of the 21st century. By doing so, we seek to (1) provide detailed insights into the possible development of Europe's most abundant tree species, (2) identify tree species which may potentially become more relevant for European forestry under anticipated conditions, and (3) outline the potential change in local tree-species abundance and composition which is linked to tree-species diversity. Thereby, we (4) aim at identifying potential hotspots of forest vulnerability which should be given special attention in the context of adapting European forests to climate change.

#### MATERIALS AND METHODS

#### Data

#### Statistically Downscaled Climate Projections

For the climate analog computations, we made use of statistically downscaled CMIP5 climate projections. Considering 16 different models, we downloaded historical (1961–1990) and future (2061–2090) scenario projections of monthly minimum, mean, and maximum temperature as well as monthly precipitation sums. Regarding the future projections, we considered the RCP scenarios RCP 4.5 and RCP 8.5. Thus, altogether 192 datasets (4 variables × 16 models × 3 scenarios) were obtained, which were downloaded via various Earth System Grid Federation data nodes<sup>1</sup> (for details see **Supplementary Table S1**). The selection of the particular 16 models as well as the particular period 2061– 2090 was based on the availability of relevant climate projections

<sup>1</sup>https://esgf.llnl.gov/

covering a period of 30 years at the end of the 21st century at the ESGF data nodes. We are aware of that choosing only 16 CMIP5 models, may lead to a certain underrepresentation of projection spread (McSweeney and Jones, 2016). In this context, McSweeney and Jones (2016) recommended using at least 13 models to account for model uncertainty since they on average represent more than 80% of the overall spread of CMIP5 climate projections. Therefore, it seems likely that the selected 16 models represent model variability to a high degree.

Since the downloaded models had a rather coarse spatial resolution (in the order of 0.75–3◦ ) and moreover different resolutions leading to differing overlaps of grid cells, we statistically downscaled all projections using Climate Research Unit gridded monthly minimum, mean, and maximum temperature [CRU TS version 4.01<sup>2</sup> (Harris et al., 2014)] as well as Global Precipitation Climatology Center gridded monthly precipitation sums [GPCC<sup>3</sup> (Rudolf and Ziese, 2011)] to a common 0.5◦ spatial resolution. Statistical downscaling was undertaken using the delta-method. That is, for the baseline period with temporal overlap between gridded and projected climate data (i.e., 1961–1990) we for each projected variable (CMIP5) and grid cell computed its mean monthly climatological difference to the corresponding gridded variable (CRU and GPCC, respectively). Thereby, we for each variable obtained a grid at 0.5◦ spatial resolution, representing the difference between the 1961–1990 monthly climatology of projected and gridded climate variables. In case of uneven spatial overlap of the two different grids, we computed the corresponding zonal mean. These climatological differences were subsequently added to the CMIP5 climate projections (both historic and future scenarios) to finally obtain climate projections at 0.5◦ spatial resolution and referenced to the baseline period 1961–1990. Due to this downscaling procedure, the temperature and precipitation climatology of all historic projections matched exactly those of the gridded data within the baseline period. However, the temporal patterns varied among models in dependence of particular model specifications which cause the typically observed spread of model projections (McSweeney and Jones, 2016).

Statistically downscaled projections were further used to compute ecologically meaningful variables related to growing season temperatures, length of the growing season, climatic water balance, and seasonality (Metzger et al., 2013). Computation of such generally more meaningful variables with regard to plant ecology is needed to better represent the actual "climate envelopes" of tree species (Hijmans and Graham, 2006). In order to simplify the multidimensionality of the obtained data, we computed the spatiotemporal mean for each of the obtained variables (i.e., for each grid cell and time step) across the 16 selected models. To select the most appropriate variables for the definition of climate analogs, we finally tested a large number of various variable combinations regarding their ability to represent the well-known Köppen-Geiger climate classification (Köppen, 1900; Geiger, 1961) as well as two more recent global climate classifications (Olson et al., 2001; Metzger et al., 2013). This was done by comparing corresponding classifications representative of the respective variable selection (computed to comprise 32 climate zones using the partitioning around medoids method) with the considered, existing classifications (i.e., Köppen-Geiger, Metzger, and Olson). As a measure of match, we computed the socalled Kappa-statistic and used the threshold values introduced by Monserud and Leemans (1992) which revealed good to very good agreement (Kappa ranging from 0.58 to 0.79) of the finally selected variable combination with the chosen classifications. Following this approach, we eventually selected 11 variables which are listed and described in detail in **Supplementary Table S2**. The corresponding climate classification is shown in the supplementary (**Supplementary Figure S1**). The spatial extent of the generated climate parameters was finally matched with the forest inventory data (see Section "Forest Inventory Data"), i.e., longitude ranging from 10.75◦ W to 32.75◦ E and latitude ranging from 35.75◦ N to 70.75◦ N resulting in altogether 3,949 considered grid cells.

#### Forest Inventory Data

As a baseline for current tree species distributions across Europe, we downloaded the recently published EU-Forest dataset (Mauri et al., 2017). This dataset contains altogether 588,983 tree-species records (for details see Mauri et al., 2017). Due to low spatial and climatological representativity we did not consider data from the Canary Islands, finally resulting in 586,862 records representing altogether 242 different tree species distributed across Europe. Since many of these species were only represented at very few locations and thus would not allow for a meaningful representation of their current potential distribution, we decided to only include the 26 species which each at least represented 1% of the total dataset. By doing so, we eventually considered 82% of the EU-Forest dataset for our analyses (for an overview on the selected species see **Supplementary Table S3**).

#### Data Processing and Statistical Analyses Computation of Climate Analogs

To map recent and potential future species distributions we computed climate analogs and merged those with the EU-forest dataset (see Section "Mapping Current and Future Tree-Species Distributions"). Here, we used climate analogs to identify regions which currently feature climate conditions that are more or less similar to the projected conditions of a given site. Depending on the selected target projection we in the following call these climates current analogs if considering current climates that are analogous to the current climate of a given site vs. future analogs if considering current climates that are analogous to the projected future climate of a given site.

Prior to defining current and future analogs, we standardized (z-transformed) each of the selected 11 variables separately. For the standardization of historic and future projections we used the mean and SD of the historic baseline period (1961–1990). Thereby, we for each variable obtained records representing unitless SDs from the variable-specific historic mean. Doing so allowed for equally weighing the variables while retaining the

<sup>2</sup>https://crudata.uea.ac.uk/cru/data/hrg/

<sup>3</sup>https://www.esrl.noaa.gov/psd/data/gridded/data.gpcc.html

offset of projected future climates in relation to the historic baseline period.

To define current/future climate analogs, we then computed the Manhattan distance (Black, 2006) between the standardized current/future climate variables of a given site and the historic climate variables of all considered grid cells. Consequently, each of the considered grid cells obtained a value which resembled the climatic distance (considering all 11 variables) of that grid cell to the selected site and projection. That is, the lower this climatic distance, the more similar the climate of the respective grid cell to the projected current/future climate of the selected site.

Finally, we had to define a threshold of climatic distance below which a climate is considered analogous to the climate at a location of interest. For reasons of objectivity and reproducibility we based this threshold on the average first percentile of current climate distances computed over all grid cells. That is, we for each of the 3,949 grid cells computed its climatic distance to the current climate of the remaining 3,948 grid cells and extracted the first percentile of the grid-cell specific climate distances. The mean over all 3,949 first percentiles of gridcell specific climate distances we took as an objective and uniform measure for defining climate analogs. A verification of the first percentile was performed manually by computing 100 different climate analogs and visually comparing the climate diagrams (Walter and Lieth, 1960) of analogous grid cells with those of the selected site. Since climate diagrams in most cases matched very well, we considered the selected threshold of the first percentile as suitable. We also assessed other percentiles, e.g., 5 permille and the second percentile but thereby we either obtained too small climate analog regions for several locations (5 permille) or included areas, whose extreme climate diagrams did not match well within the analogous region (second percentile). The match between actual and projected historic species distributions may be considered a further validation

green colors.

procedure (see Section "Mapping Current and Future Tree-Species Distributions").

#### Mapping Current and Future Tree-Species Distributions

The EU-forest dataset was used as the basis for mapping current and future tree-species distributions. First of all, we had to transform the point-based records into a gridded format that matched the resolution and extent of computed climate analogs (see Section "Computation of Climate Analogs"). Therefore, we computed the relative abundance of a given species within each grid cell. For this, all EU-forest records within a specific grid cell were pooled by species and the relative contribution of each species computed, thereby obtaining values between 0 and 1. That is, if a species had no records within that grid cell it would obtain the value 0, whereas if a species was the only species occurring within that cell it would obtain the value 1, while if a species would co-occur with other species it would obtain a value between 0 and 1 reflecting its proportional contribution to grid-cell specific tree species abundances.

In the next step, these relative abundance maps were merged with the climate analogs. Thereby, we obtained the occurrence probability of a given species at a specific site with a certain climate projection (historic or future). For this, we first computed the according site- and projection-specific (historic or future) climate analog and secondly averaged the relative abundances of the corresponding analog region per species. That is, if a species expressed fairly high (low) abundance in most of the analog grid cells, it would also gather a high (low) average abundance for the respective site and projection. This procedure was carried out for each of the 26 selected species and applied to each of the 3,949 grid cells, to finally obtain European-wide occurrence probabilities of the 26 most abundant tree species under current and future climate projections. **Figure 1** depicts a schematic flow-chart of our methodological approach.

To assess how well the projected relative abundance probabilities represented observed relative abundances as obtained from EU-forest, we for each of the 26 species computed Spearman's rank correlations between observed and projected current relative abundance. Correlation scores from 0.43 to 0.76 (mean 0.58) indicated moderate to good agreement between observed and projected relative abundance. Observed and projected current relative abundance maps are shown along with the species-specific correlations between observations and projections in the supplementary information (**Supplementary Figures S2–S8**).

#### Evaluation

To visualize the change of European forests tree species composition (see Section "Mapping Current and Future Tree-Species Distributions") we first of all plotted current and future relative abundance probabilities for the four currently most abundant European tree species, i.e., P. sylvestris (Scots pine), P. abies (Norway spruce), F. sylvatica (European beech), and Q. robur (English oak). In combination, these four species make up 44% of the considered forest inventory data. As a proxy for tree-species relevance for forestry within the different scenarios, we computed the mean relative abundance probability for each species across all grid cells for current and future conditions. That is, species with high mean relative abundance probabilities in a given scenario feature a high relevance for forestry and vice versa. As a measure of change for the general appearance of forests, we moreover computed the relative change of the currently most dominant tree species for each of the 3,949 grid cells. That is, for each grid cell the currently dominant tree species was determined and its relative abundance probability of the two future scenarios for the same grid cell was put in relation to its

current relative abundance probability. Finally, as a measure of tree-species diversity changes, we for each grid cell computed Shannon's H (Shannon, 1948) and expressed the absolute change (i.e., the difference) of the two future scenarios in relation to the current value. All analyses were performed in, R' (version 3.3.3., R Core Team, 2017) extended for the packages SPEI (Beguería and Vicente-Serrano, 2013) and lattice (Sarkar, 2008).

#### RESULTS

The four most abundant tree species indicated prominent changes in relative abundance probabilities, particularly for RCP 8.5 projections (**Figures 2**, **3**). All of the four species indicated a declining abundance probability at southern latitudes and lower elevations. Based on declining abundance probabilities, Scots pine indicated a retreat from Central and Southeast Europe to the higher elevations of the Alps and the Carpathians, as well as Northern Europe. The same was observed for Norway spruce. However, for this species the retreat appeared even more pronounced, i.e., under RCP 8.5 projections it completely disappeared from Central European lowlands with last refugia in the Alps and the Carpathians as well as in Scandinavia north of 60◦ latitude. European beech also indicated a northward migration, i.e., abundance probabilities significantly decreased over large parts of Central Europe but increased in Southern Scandinavia. Finally, English oak retreated from its southernmost locations such as large parts of France and the Pannonian Basin and increased its abundance in Southern Scandinavia.

While Scots pine and Norway spruce expressed a decrease in the order of 5 (RCP 4.5) to 15% (RCP 8.5) of mean relative abundance probability across Europe, European beech and English oak retained their mean relative abundance probability for both scenarios (**Figure 3**). Despite possible losses in abundance probability, these four tree-species remained the most abundant across Europe. Tree species with remarkable increases (2–6%) of abundance probabilities were Quercus ilex, Pinus pinaster, Pinus halepensis, Pinus nigra, Castanea sativa, and

The inserted values refer to the cumulative frequencies of losses (left of black vertical line) and gains (right of black vertical line) for RCP 4.5 (orange) and RCP 8.5 (red).

Quercus pubescens, all of which seemed to expand their current distributional range from the Mediterranean to Central Europe. In contrast, declining species mostly featured a northward migration of their distributional range rather than an expansion (**Figure 3** and **Supplementary Figures S9–S14**).

Projected relative abundance changes of currently dominant tree species depicted large losses for most of Europe. However, large areas in the Mediterranean as well as some higher elevations in England and Scandinavia expressed increasing abundances. That is, for many European regions, projections indicated the currently most abundant tree species to experience a significant decline in abundance (**Figure 4** upper panel). More specifically, under RCP 4.5 77% of all grid cells expressed losses which increased to 80% under RCP 8.5. However, under the more extreme RCP 8.5 scenario, many more grid cells expressed remarkable losses, i.e., for more than 30% of grid cells less than 50% of current relative abundance of the dominant species were retained (**Figure 4** lower panel).

Shannon's H expressed features complementary to relative abundance changes (**Figure 5** vs. **Figure 4**). That is, while treespecies diversity decreased in the Mediterranean, it generally increased in Central and Northern Europe, thereby mirroring the immigration of new species into those regions (**Figure 5** upper panels). Under RCP 4.5 74% of grid cells expressed increasing tree-species diversity while it was only 68% under RCP 8.5 (**Figure 5** lower panel). However, under RCP 8.5 the distribution was shifted toward higher increases in tree-species diversity, particularly in Scandinavia.

# DISCUSSION

#### The Projected Face of European Forests

Based on our projections, it seems likely that the appearance of European forests will change remarkably until the end of the 21st century. Most prominent changes are related to the northward migration of the four most abundant species, i.e., Scots pine, Norway spruce, European beech, and English oak and at the same a remarkable decrease in abundance in Central Europe, particularly regarding Norway spruce and Scots pine (**Figure 2**). Other currently abundant species such as Betula pubescens, Betula pendula, Q. petraea, and Sorbus aucuparia featured similar projections (**Figure 3** and **Supplementary Figures S9– S11**). Considering whole Europe, the relative importance of Scots pine and Norway spruce—the currently economically most important tree species—decreased, while for European beech and English oak, the relative importance remained unchanged (**Figure 3**). Tree species that were projected to partly fill the resulting gaps were Q. ilex, P. pinaster, P. halepensis, P. nigra, C. sativa, and Q. pubescens (**Figure 3** and **Supplementary Figures S9–S14**). For large parts of Central and Northern Europe, the currently dominant tree species experienced remarkable reductions in abundance, particularly for RCP 8.5 while for most Mediterranean areas an increase in abundance of the currently dominant species was projected (**Figure 4**). The projected migrations and changes in abundance resulted in an increasing tree-species diversity in most regions north of the Alps and a decreasing tree-species diversity in the Mediterranean (**Figure 5**).

In combination, these results outline potential hotspots of vulnerability, particularly in Southern France, Spain, and on the Italian peninsula due to both decreasing frequencies of currently most dominant tree species as well as decreasing tree-species diversity even under the less extreme RCP 4.5 scenario (**Figures 4**, **5**). Other potential hotspots of forest vulnerability—though to a lower degree—are located in Bulgaria and Romania as well as the Pannonian Basin, particularly under the RCP 8.5 scenario. For these hotspots, it according to our projections seems likely that the currently most abundant tree species will experience a remarkable decline in abundance, while at the same, species diversity will decrease. Consequently, these regions should be given special attention in the context of adapting local forests to anticipated climate change and tree-species should be selected carefully since the species-portfolio of those regions is likely to decline.

# Species Projections in the Scientific Context

The projected changes of the four most abundant species are in line with recent research. First of all, for Scots pine an increasing mortality at its southern distributional margin was observed over the last decades, e.g., on the Iberian peninsula after a severe drought in 2005 (Galiano et al., 2010, 2011), in inner-alpine dry valleys, for instance in 1998 and 2003 (Rebetez and Dobbertin, 2004; Bigler et al., 2006; Pichler and Oberhuber, 2007), and recently also Central European lowlands after the hot and dry summer of 2015 (Buras et al., 2018). Secondly, Norway spruce is well-known for its high droughtsusceptibility (Kohler et al., 2010; Lévesque et al., 2013; Zang et al., 2014; Rehschuh et al., 2017). Since the frequency and intensity of drought spells are projected to increase in Central Europe (Horton et al., 2015; Pfleiderer and Coumou, 2018), the disappearance of Norway spruce from Central European lowlands in our projections seems reasonable. Finally, our projections suggested beech to express a more pronounced decline in Central Europe compared to English oak, although both species were relatively less affected in comparison to Scots pine and Norway spruce. On the one hand, this is in line with Zang et al. (2014), who reported a lower drought susceptibility of beech in comparison to Norway spruce. On the other hand, the higher drought susceptibility of beech in comparison to English oak has been reported on the basis of dendro-ecological investigations (Scharnweber et al., 2011, 2013; Walentowski et al., 2017) and species distribution models (Walentowski et al., 2017).

Considering less abundant species, the projected northward expansion of Q. ilex also finds support in the literature. Based on tree-ring analyses, Q. ilex was shown to decline under increasing water stress, but to benefit from increasing winter temperatures (Gea-Izquierdo et al., 2011), which is in line with the projected northward expansion and decline in the driest regions in Spain (**Supplementary Figure S9**).

Also, the increasing importance of Q. pubescens in Central European lowlands (**Supplementary Figure S13**), finds support in the literature. That is, pubescent oak has been reported to potentially replace Scots pine in dry inneralpine valleys due to a comparably better adaptation to dry conditions (Eilmann et al., 2009; Rigling et al., 2013). The projected northward migration of P. nigra (**Supplementary Figure S11**) is supported by studies which reported a relatively high drought-susceptibility of black pine and increasing trends in needle defoliation in the Mediterranean which reflects its sensitivity to increasing temperatures (Linares and Tíscar, 2010; Carnicer et al., 2011). Moreover, its cold-hardiness was reported to be in the range of Central European species which renders it suitable for the projected northward migration (Kreyling et al., 2012). Also for P. pinaster and P. halepensis an increased leaf defoliation has been observed, indicating a declining performance of those species in the Mediterranean (Carnicer et al., 2011) which is also reflected in their projected northward migration in our study (**Supplementary Figure S10**). In conclusion, several of the presented projected changes of European tree-species distributions are supported by independent studies.

#### Constraints and Limitations

Despite the support by independent studies, the methodological approach behind the presented projected tree-species distributions features certain constraints and limitations. Most of all, the rather coarse resolution of 0.5◦ yet needs further improvement. This is because at such coarse resolution, elevational differences are only poorly resolved—particularly in the inner-alpine valleys but also in other mountain areas such as the Norwegian fjords. Therefore, projected relative abundances of these regions have to be interpreted carefully. For instance, while it seems likely that Norway spruce and Scots pine refugia will remain at higher elevations in the Alps, these species will likely disappear from the dry inner alpine-valleys. To overcome these limitations, future studies should consider

integration of higher resolved climate projections such as the EURO-CORDEX ensembles with 11 km spatial resolution (Jacob et al., 2014) or the WORLDCLIM projections with 1 km spatial resolution (Fick and Hijmans, 2017). However, since we wanted to account for model uncertainty by incorporating a comparably large number of different simulations and moreover encountered computational limitations, we here opted for downscaled projections derived from 16 different models. But even if increasing the spatial resolution to 1 km, unresolved sub-scale processed would remain. A focal point in the context of sub-scale processes is the forest edge, which recently was shown to feature a higher drought-induced mortality of Scots pine within the first 50 m of forest-edge distance (Buras et al., 2018) that may also hold true for other tree species. However, given the specific micro-climatic conditions at the forest edge (Chen et al., 1993, 1999), incorporation of these effects into projections of tree-species distributions that rely on gridded climate projections only remains a demanding challenge. Consequently, the forest edge needs to be given further attention in the context of climate-change induced tree vulnerability.

Another shortcoming of the presented approach is related to the current relative abundance of species. For instance, Picea sitchensis is yet mainly distributed in the United Kingdom and Denmark, which may introduce certain biases regarding its projected distribution. That is, although it also is capable of growing under more continental conditions, the high abundance in the observational data mainly restricts it to coastal climates in the projected relative abundances. This is also the main reason, why we focused our evaluation on the four most abundant tree species. However, to also include species with projected increasing relative abundances, we decided to consider all species which cover at least 1% of the EU-forest data. Another constraint in this context is related to other theoretically important tree species that were not considered in our approach due to a marginal representation in the EU-forest data. For instance, Walentowski et al. (2017) reported Acer campestre, Sorbus torminalis, Sorbus aria, Ulmus minor, and Tilia platyphyllos to be well adapted to anticipated climate conditions, all of which were not considered here due to low representation. Within this context it moreover is important to mention, that based on the utilized forest-inventory data, we were not able to incorporate tree species from the Mediterranean coast of North Africa, which may qualify as potential alternative tree species for the European Mediterranean under projected climate conditions. Therefore, the decline in tree-species diversity for Mediterranean regions has to be interpreted in the context of the underlying data. That is, the depicted declining treespecies diversity in those regions highlights a substantial loss in diversity regarding Europe's 26 currently most abundant tree species, and thus that local foresters should consider non-native tree species from outside of Europe as potential alternatives.

The rigorous definition of the climate analog threshold to the average first percentile over all European grid cells also has certain caveats. Although the validation assessments confirmed this threshold superior in comparison to others, some likely errors resulted as for instance the projected increase in relative abundance of beech and oak on the Kola Peninsula (**Figure 2**). One way to cope with this, is to incorporate additional climate parameters into the computation of climate distances. However, in the extensive trials undertaken to identify the best-suited variable combination, we eventually decided for the current selection, which may be considered the best compromise among available variables, particularly since the match between actual and projected current tree-species distributions were fairly good (see Section "Mapping Current and Future Tree-Species Distributions" and **Supplementary Figures S2–S8**).

Our projections of future tree-species distributions were restricted to the RCP 4.5 and RCP 8.5 scenarios. However, this does not represent the full range of available climate scenarios for the end of the 21st century. For instance, the Paris agreement obtained at the conference of the parties 21 aimed at limiting global warming to well below 2◦C and ideally 1.5◦C (Schleussner et al., 2016). However, the scenarios RCP 4.5 and RCP 8.5 on average result in a global warming of 2.4 and 4.1◦C, respectively. Although this is higher than the aim of the Paris agreement, these projections appear to be realistic since it based on the current nationally stated mitigation ambitions seems likely that global temperatures will increase by 3◦C until the end of the 21st century (IPCC, 2018).

Finally, our approach mainly relies on climatic properties of analog regions, while pathogens such as fungi and bark beetles were not considered. Although the risks associated with potential pathogens to some degree are reflected in current species abundances, other specific diseases are not included. On the one hand increasing drought frequency is considered an inciting factor for specific fungi and bark beetles that indirectly determine tree-species distributions via mortality (Bigler et al., 2006; Rehschuh et al., 2017; Buras et al., 2018) and likely are incorporated in our climatebased approach. But on the other hand, specific diseases or beetles which are only indirectly related to climate such as the Asian longhorn beetle or the fungi responsible for the remarkable ash die-back in Europe are likely not covered by our approach (MacLeod et al., 2002; Kirisits and Freinschlag, 2012).

Despite these limitations, the presented projections of relative species abundances may be considered a valuable estimate of anticipated forest change. Outlining the most prominent changes of tree-species distributions and identifying associated hotspots provides stake-holders, foresters, and ecologists with valuable insights into upcoming changes of forest compositions. However, given their currently relatively low abundance, the performance of some of the identified alternative species under climate change (such as C. sativa) should be investigated in more detail.

# CONCLUSION AND OUTLOOK

Our projections of tree-species distributions highlight prominent changes for Europe's currently most abundant tree species. That is, based on our projections and due to increasing risks of drought-induced mortality Scots pine and Norway spruce

will likely become unprofitable and consequently significantly decrease their abundance in Central European lowland forests in course of the anticipated climate change. Depending on the magnitude of climate change (RCP 4.5 vs. RCP 8.5), European beech and English oak may sustain in large parts of Central Europe. Potential alternative tree species identified here, such as Q. ilex, P. nigra, P. halepensis, P. pinaster, and C. sativa may be considered a meaningful replacement of locally declining tree species. However, other species which were not integrated into our approach because of low representation in the EU-forest dataset and were shown to be potential replacement species in other studies should also be taken into consideration. Moreover, our study identified hotspots of forest vulnerability in South France, Spain, Italy, the Pannonian Basin, Bulgaria, and Romania. Particularly in those areas, foresters, stakeholders, and nature conservationists should pay special attention to adapt local forests to anticipated changes well in time by selecting tree species that are adapted to anticipated climate conditions. To further refine our picture about future forests, subsequent studies should aim at incorporating higher resolved gridded climate projections into climate analog computations and extent the forest-inventory beyond Europe.

#### DATA AVAILABILITY STATEMENT

Upon publication of the manuscript, the datasets generated for this study can be found in the Github repository of AB. The R-code used for computation of climate analogs will soon be released as an "R"-package entitled "ClimCong."

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

AB and AM designed the study. AB performed all statistical analyses and wrote the manuscript with valuable contributions by AM.

## FUNDING

This study has received funding from the Bavarian State Ministry for Food, Agriculture, and Forestry (StMELF, Grant Number M029).

### ACKNOWLEDGMENTS

We acknowledge the World Climate Research Programme's Working Group on Coupled Modeling, which is responsible for the utilized CMIP5 projections, and we thank the climate modeling groups (listed in **Supplementary Table S1** of this paper) for producing and making available their model output. Moreover, we are grateful for the valuable discussions with Dr. Christian Kölling while designing the study.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018.01986/ full#supplementary-material


vulnerability research. Wiley Interdiscip. Rev. Clim. Change 1, 374–392. doi: 10.1002/wcc.48


availability suggests high vulnerability of Norway spruce and European larch. Glob. Change Biol. 19, 3184–3199. doi: 10.1111/gcb.12268


oak in dry Alpine forests. Glob. Change Biol. 19, 229–240. doi: 10.1111/gcb. 12038


**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 Buras and Menzel. 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.

# River Regulation Causes Rapid Changes in Relationships Between Floodplain Oak Growth and Environmental Variables

Maksym Netsvetov<sup>1</sup> \*, Yulia Prokopuk<sup>1</sup> , Radosław Puchałka<sup>2</sup> , Marcin Koprowski<sup>2</sup> , Marcin Klisz<sup>3</sup> and Maksym Romenskyy4,5

<sup>1</sup> Department of Phytoecology, Institute for Evolutionary Ecology, National Academy of Sciences of Ukraine, Kiev, Ukraine, <sup>2</sup> Department of Ecology and Biogeography, Faculty of Biology and Environmental Protection, Nicolaus Copernicus University, Torun, Poland, ´ <sup>3</sup> Department of Silviculture and Genetics, Forest Research Institute, Raszyn, Poland, <sup>4</sup> Department of Life Sciences, Faculty of Natural Sciences, Imperial College London, London, United Kingdom, <sup>5</sup> Department of Zoology, Stockholm University, Stockholm, Sweden

#### Edited by:

Veronica De Micco, University of Naples Federico II, Italy

#### Reviewed by:

Minhui He, Northwest Institute of Eco-Environment and Resources (CAS), China Jesús Julio Camarero, Spanish National Research Council (CSIC), Spain

> \*Correspondence: Maksym Netsvetov disfleur76@live.fr

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 08 October 2018 Accepted: 21 January 2019 Published: 05 February 2019

#### Citation:

Netsvetov M, Prokopuk Y, Puchałka R, Koprowski M, Klisz M and Romenskyy M (2019) River Regulation Causes Rapid Changes in Relationships Between Floodplain Oak Growth and Environmental Variables. Front. Plant Sci. 10:96. doi: 10.3389/fpls.2019.00096 The radial growth of pedunculate oak (Quercus robur), a species often ecologically dominating European deciduous forests, is closely tied up with local environmental variables. The oak tree-ring series usually contain a climatic and hydrologic signal that allows assessing the main drivers of tree growth in various ecosystems. Understanding the climate-growth relationship patterns in floodplains is important for providing insights into the species persistence and longevity in vulnerable riverine ecosystems experiencing human-induced hydrology alteration. Here, we use 139 years long instrumental records of local temperature, precipitation, and water levels in the Dnipro River in Kyiv to demonstrate that the implementation of river regulation has decoupled the established relationship between the radial growth of floodplain oak and local hydro-climatic conditions. Before the river flow has been altered by engineering modifications of 1965–1977, the water level in the Dnipro River was the key driver of oak radial growth, as reflected in the tree-ring width and earlywood width. The construction of two dams has altered the seasonal distribution of water level diminishing the positive effect of high water on oak growth and subsequently reversing this trend to negative, resulting from a seasonal ground water surplus. The decrease in the correlation between oak growth indices and the river's water level in April–June was unprecedentedly rapid and clearly distinguishable among other changes in the growth-to-climate relationship. Our findings further demonstrate that trees growing in areas exposed to urban development are the most susceptible to downside effects of river regulation.

Keywords: Quercus robur, tree-ring, intra-annual ring-width, temperature, precipitation, water level, floodplain

# INTRODUCTION

Riparian forests play a pivotal role in ecological processes at various scales (Naiman et al., 2005) and represent the most productive terrestrial ecosystems in the world. Yet, they are considered to be highly vulnerable, and without planned adaptation exposed to the grave consequences of climate change and human impact (Capon et al., 2013). The susceptibility of riparian ecosystems to

natural disturbances often results in their degradation and alteration caused, among other factors, by hydrological modifications such as installation of dams and levees, water extraction, etc. (Stella and Bendix, 2019). River regulation sways a number of hydro-geomorphic processes (Stella et al., 2013) translating its overarching intra-system impact on growth of floodplain trees, density of forest stands, their structure and composition (Madera and Úradn ˇ ccek, 2001 ˆ ; Rodríguez-González et al., 2010; Smith et al., 2013; Gee et al., 2014; Stojanovic et al., ´ 2015; Weissbrod and Binder, 2017).

Disentangling the intricacies of long-term responses of individual trees, stands or entire ecosystems to river flow modification requires thorough retrospective investigations employing reliable proxies with at least annual precision accuracy. In temperate zones, the yearly variation of tree radial growth serves an effective tool for extracting the high-frequency environmental signal as well as event-driven disturbance information (St. George, 2010a,b; Esper et al., 2015). Although a number of woody plants compositing a particular ecosystem are potentially useful for developing the ring-width records, the species that contributes the most to the system's structure and functions is of the key interest in retrospective ecological and climatological analyses. In European forests and woodlands, pedunculate oak (Quercus robur L.), a long-lived species, performs important ecosystem-wide functions (Nilsson et al., 2002) dominating forests growing in varying hydro-climatic and soil conditions (Ducousso and Bordacs, 2004; Ellenberg, 2009). Although European flooded forests are typically dominated by black alder or several co-dominant species, pedunculate oak particularly often composes riverine ecosystems in the areas exposed to short-term floods (Klimo and Hager, 2000; Didukh and Aloshkina, 2012). Unlike many other woody plants, the high ecological plasticity of this species enables a direct comparison of the climatic signal from nearby habitats varying by local conditions, e.g., floodplain sites adjacent to out of the valley areas.

The oak ring-width, intra-annual ring-width, and xylem anatomy measurements proved to be promising in assessing the growth-to-climate relationships in bog sites (Scharnweber et al., 2015; García-González and Souto-Herrero, 2017) and in areas experiencing flooding (Gricar et al., 2013 ˇ ; Hafner et al., 2015; Kames et al., 2016; Tumajer and Treml, 2016; Okonski, 2017 ´ ), seasonal waterlogging (Rozas and García-González, 2012; Scharnweber et al., 2013) or human-induced hydrology modification (Singer et al., 2013; Gee et al., 2014; Stojanovic et al., 2015 ´ ; Tumajer and Treml, 2017; Zheng et al., 2017). However, only a limited number of studies have actually employed at least some of these tools to study the growth-hydrology relationships in pedunculate oak or in other ring-porous tree species (St. George and Nielsen, 2003; Kames et al., 2016; Koprowski et al., 2018). Even fewer works have looked into the impact of hydrology modifications on growth of floodplain trees and its relationships with environmental drivers (Predick et al., 2009; Gejková and Polákova, 2012 ˘ ; Stella et al., 2013; Gee et al., 2014; Weissbrod and Binder, 2017).

In this study, we provide a comprehensive assessment of the effect of the implementation of river regulation in Kyiv, Ukraine, on relationships between the floodplain oak radial growth and local hydro-climatic conditions. Based on detailed, 139 years long, instrumental meteorological and water level records, we quantify the pedunculate oak intra-annual ring width chronologies for almost all sites embedded in the Dnipro River's floodplain in Kyiv. Given a strong regional climate signal exhibited at all study sites, we dissect the original series using chronology from an unflooded reference forest, and applying an adjustment technique similar to that from Meco and Baisan (2001) to all study sites' series, thereby isolating the floodplainspecific signal. Based on these data, we assess the following hypotheses: (1) Floodplain sites' intra-annual ring-width chronologies share a specific climate/hydrology signal distinct from an adjacent area out of the valley; (2) Implementation of river regulation alters the relationships between floodplain oak growth and climate/hydrology; (3) Upon flood-preventing river modification, the oak growth is substantially relying on the river's water level during spring (April–May) as it coincides with the early xylem development stage.

# MATERIALS AND METHODS

#### Sampling Sites

Tree core sampling was undertaken at five sites (**Table 1** and **Figure 1**) in Kyiv, Ukraine. Three study sites are located in the Dnipro River floodplain adjacent to the river stream channel: Zhukiv island (Zhu, a nature reserve), Dubysche (Dub, an oak forest), and Bychok (Byc, a reserve forest). Soil in this area is sandy alluvial and the forest floor is formed by a thick litter layer. Another study site, the Lisnyky forest botanical reserve (Lis), is part of the flooded lowland between two tributaries of the Dnipro River – the Siverka River and the Petil River. In the 1950s, the tributary rivers have been modified through drainage and ditch network maintenance that significantly depleted their flow regime. This area historically experiences seasonal inundation and backwater due to flooding events at the main stem river. The soil type here is podsolgley. Our reference site, the forest of Feofania [Feo (ref)] is located about 5 km west and 65–90 m upslope of the Dnipro River. The area does not experience seasonal floods and the groundwater level is consistently 5–20 m below the surface throughout the year. The soil in the forest is clay podsol. Byc is the only site among studied that is situated in a built-up area gradually developed since the 1950s.

#### Tree-Ring Data

At the study sites, only visibly healthy and dominant or co-dominant oaks were selected for tree-ring sampling. Using a 5-mm-diameter increment borer, we extracted two to three cores from each tree at 1.3 m stem height. All obtained samples were glued onto the wooden supports, surfaced using blade, and scanned with a flatbed scanner (Epson Perfection V37) at a resolution of 3200 dpi. Tree ring-widths were measured to the nearest 0.01 mm resolution with AxioVision 4.9.1 software (Carl Zeiss). The cores were crossdated and merged by simple arithmetic averaging to obtain individual

#### TABLE 1 | Sampling sites information.

fpls-10-00096 February 2, 2019 Time: 18:20 # 3


MSSL, mean sample segment length.

series. To reduce a non-climate-related signal, the individual series were detrended using a smoothing spline with a wavelength of 0.67 series duration and then standardized to produce dimensionless ring-width indices (RWI). The site-level chronologies were then built by calculating robust means for the individual series prewhitened with a best-fit first-order autoregressive model. The quality of crossdating was checked with COFECHA software (Holmes, 1983) and dplR package (Bunn, 2010) in R 3.5.1 (R Core Team, 2018). The dimensionless earlywood (EW) and latewood (LW) width chronologies, were developed from the subsets of trees that highly correlated (r > 0.7, p < 0.01) with the corresponding site-level COFECHA master-series. In LW analysis, we used its indices adjusted with respect to the known dependence of LW on EW (Meco and Baisan, 2001). Adjusted LW series were produced by fitting the linear regression of LW to EW (for correlations between intra-annual width series see **Supplementary Figure 1**). To pinpoint the tree growth variability inherent to floodplains (Byc, Zhu, Dub, and Lis), we applied a simple linear regression, i.e., an adjustment of RWI, EWI, and LWI (with I standing for indices) study sites' chronologies to the corresponding reference site [Feo (ref)] chronologies. To find a year of a possible change in the mean growth rate due to river regulation, we applied the change point analysis (Killick and Eckley, 2014) to raw RW, EW, and LW chronologies. In all further analyses, we used the adjusted series from the flooded sites.

#### Climate and Water Level Data

The Dnipro River has been modified in the 1920–1970s at its Ukraine stretch to generate hydroelectricity, provide water supply for agricultural needs, and mitigate extreme flooding events (Vyshnevsky, 2005). In Kyiv, the river flow modification system has included installation of dams and levees upstream (1964) and downstream (1977) of the city, which substantially moderated both the overbank and low water. We obtained data on climate and hydrology in Kyiv from the Central Geophysical Observatory (CGO, 50.40 N, 30.34<sup>0</sup> E, 166 m a.s.l.), Kyiv, Ukraine. The available data spanned the period from 1877 through 2015. Based on the times of the main engineering modifications of the Dnipro River (1964 and 1977), we have identified two distinct study periods within our continuous (1877–2015) climate and hydrology records: before river regulation (1877–1964, first period) and during river regulation (1978–2015, second period).

According to hydrology data, the largest floods in Kyiv (**Figure 1**) occurred in springs of 1877, 1908, 1917, 1931, 1970 (first period), and 1979 (second period) and have in general been attributed to rapid temperature changes and high amounts of precipitation in the river's catchment area above the city. The lowest water level, under natural fluctuations, was recorded

in November of 1921 (88.74 m a.s.l.). Following establishment of the first dam below Kyiv in 1964, the Dnipro River's water level has decreased significantly, falling to its historical minimum (88.58 m a.s.l.) in the summer-autumn of 1972. In December 1977, another dam above Kyiv was implemented causing Dnipro's mean annual water level to increase by 0.85 m while lowering its annual maximum by 1.32 m. It is worth to note that the engineering modification of the Dnipro River has altered the magnitude of the water level fluctuations rather than the seasonal water level distribution (**Figure 2**). Since 1978, the most prominent changes in the river water level were its seasonal decrease in April–May and augmentation in July–February. The latter is likely attributed to the increasing trends in June and September precipitation as well as in autumn-winter temperatures (Netsvetov et al., 2018).

The climate data (recorded at CGO) have showed higher monthly temperatures in Kyiv during the second period as compared to the first period (p < 0.05, t-test for means) for all months except June, July, September, and November. The average monthly temperature was −5.9 and −3.7◦C in January and 20.0 and 20.5◦C in July for the periods of 1878–1964 and 1978–2015, respectively. The annual temperature averaged 7.2◦C and 8.5◦C (p < 0.05) during the first and second period, respectively.

FIGURE 2 | The distribution of monthly total precipitation (A), mean air temperature (B), and maximum water level (C) of the Dnipro River in Kyiv for the periods before (1877–1964) and after (1978–2015) the onset of the river regulation. In the box-and-whisker plots, the lower and upper hinges indicate the 25th and 75th percentiles, the horizontal lines denote the median values, the whiskers extend from the hinges to the largest and smallest values within the 1.5 inter-quartile range, and the points indicate outliers.

The average total annual precipitation increased significantly (p < 0.05) from 590 mm in the first period to 628 mm in the second period, although its average monthly values have not showed a significant change. The month of occurrence of maximal precipitation value has shifted from July in the first period to June in the second period. The lowest precipitation value for all years since 1978 has been consistently recorded in January (**Figure 2**). The change in temperature in Kyiv was recognized being in line with the overall temperature trend in the northern part of the river's catchment area (Vyshnevsky, 2005).

# Growth-to-Hydrology and Growth-to-Climate Relationships

We assessed relationships between oak growth and environmental variables, water level and climate, using bootstrapped correlation and response functions (Fritts et al., 1971; Biondi and Waikul, 2004). To establish the growth-to-climate relationship, we used a climatic window spanning from June of the previous growing season through September of the current growing season. The water level data were processed with a climatic window extending from April to September of the current year, thereby capturing a complete oak radial growth season in Kyiv. Given the multicollinearity of climatic and hydrological variables, to identify the drivers of oak growth in the first (1878–1964) and second (1978–2015) time periods, we used the stationary response function. For those variables where the stationary response function was found to have a statistically significant (p < 0.05) relationship with tree growth either over one or two periods, we computed the running correlation function with a 23-year sliding interval and tested its output for spurious low-frequency modulations (Gershunov et al., 2001). The calculations were performed using the 'treeclim' package (Zang and Biondi, 2013) for R. We then employed the locally weighted polynomial regression, implemented as a function in the 'SiZer' package (Sonderegger, 2018) for R, which yields the estimated smooth function. To classify the changes in oak growth-to-driver correlations, we used the estimated derivative with confidence intervals (Fan and Gijbels, 1996; Sonderegger et al., 2009). This allowed defining the intervals with the most pronounced changes in growth-to-driver correlations as minima and maxima of the first derivative.

#### RESULTS

#### Chronologies Strength and Covariance

Over the continuous period of 1877–2015, the mean intraannual tree ring-width in the three measurement categories (radial width – RW, earlywood portion – EW, and latewood portion – LW) was higher for the floodplain sites, excluding Byc (**Table 2** and **Figure 3**), as compared to the reference site Feo (ref). For all study sites, the change point analysis has not detected any changes in RW, EW, or LW that could be consequent of the Dnipro River regulation. The only change was identified for Lys' RW and LW (p < 0.01) in 1956, well before the engineering modification of the Dnipro River stretch. The first-order autoregression coefficient (A1), a measure quantifying


The reference interval is 1877–2015. RW, EW, and LW stand for the radial width of a ring, the earlywood part and the latewood part, respectively; EPS is the expressed population signal; Reff is the mean correlation between trees; SD is the standard deviation; MS is the mean sensitivity; A1 denotes the first order autocorrelation; PC1 and PC2 are the first two principal components; raw and adj denote raw and adjusted series, respectively.

the persistence of chronologies, ranged from 0.46 to 0.68 for all categories. Across all sites, including Feo (ref), A1 was consistently highest in RW. Overall, A1 was lower for the reference site's chronology than for the flooded sites' records. The year-to-year variation, expressed by the standard deviation (SD) and mean sensitivity (MS), was considerably lower in EW than in LW and RW for all sites. The mean between-tree correlation (Reff), was in the ranges of 0.35 [Feo (ref)] – 0.43 (Dub) for RWI, 0.32 (Zhu) – 0.43 (Lis) for EWI, and 0.42 (Zhu) – 0.6 (Byc and Lis) for LWI. The expressed population signal (EPS) over the entire period for all measurement categories was greater than 0.85, suggesting a sufficient confidence level in all chronologies (**Table 2**).

**Figure 4** presents results of principal component analysis (PCA). The correlation between series from the Dnipro's floodplain and from the reference site was positive for all same-category raw chronologies (see also **Supplementary Figure 1**), implying that a common regional climatic signal was intrinsic to the series from all sites. The first principal component PC1 explained 51% of the total intra-annual ring-width variance between the sites, and the second component PC2 explained 25% of the variance. Yet, as many as first four principal components were required to extract ca. 90% of the total inertia. The ring-width and the intra-annual ring-width chronologies from all sites were effectively correlated with the first component that represented a mix of the common signal (observed at all sites) and the specific signal inherent to the floodplain sites (**Figure 4A**). Detrending and adjusting the series for flooded sites resulted in a slightly weaker MS and in a substantial decrease in correlation with the reference site, denoting the absence of the regional climatic signal in the adjusted data from the floodplain sites (**Table 2**). In the PCA performed on adjusted data (**Figure 4B**), PC1 represented the intra-annual ring-width variation mainly at the flooded sites, while PC2 corresponded to the reference site. PC1 and PC2 explained 36 and 15% of the total inertia in the adjusted data, respectively, and the first six principal components combined ca. 90% of the variance.

#### Relationships Between Oak Growth and Environmental Variables

The response function analysis revealed significant relationships between oak growth at the reference site [Feo (ref)] and environmental variables both for the current and preceding growing seasons (**Figure 5**). During the entire period of 1877–2015, Feo (ref) RWI were strongly driven by prior-December and current April precipitation and were not significantly influenced by either air temperature or Dnipro's water level (WL). Feo (ref) EWI ring width indices were significantly tied with prior season's August precipitation and temperature (negatively correlated), although the effect of precipitation was significant only in the first period (1878–1964). The decrease in Feo (ref) EWI over the second period (**Figure 5**), was also associated with the prior-July temperature. The significant relationships between Feo (ref) LWI and environmental variables spanned only for the first period. Over these years, prior-December-January and current May precipitation affected latewood growth positively, while prior-Jun precipitation, as well as October and December temperature had an overall negative effect on LWI.

The response function analysis applied to flooded sites' data admitted numerous significant coefficients in oak

grow-to-environment relationships during the first period (**Figure 5**). The Dnipro River's WL, under natural fluctuations, was the main driver for floodplain oaks growth, except in Lis, which is adjacent to Dnipro's tributaries. April WL had a significant effect on Zhu EWI, Byc RWI and EWI, Dub RWI and LWI. May WL affected both RWI and EWI growth at all floodplain sites, and also LWI in Dub. June WL had a significant influence only on Zhu LW. In the second period, upon implementation of both the upstream and downstream river regulation systems, the Dnipro's WL effect on floodplain oak growth has diminished and subsequently reversed, i.e., in April for Byc RWI, and in May for Byc RWI and LWI. High WL in May has also become superfluous for latewood formation in Lis.

In the first period, precipitation had a positive overall effect on oak growth in the flooded sites, limited to Zhu RWI, LWI, and Byc RWI in August. More relationships were found for temperature, e.g., positive effect on Lis RWI, EWI, LWI for current May, on Byc RWI and LWI for current August, on Zhu EWI for prior-November and the negative effect on Byc EWI for prior-June. After 1977, only prior-November and current January precipitation drove Byc earlywood and latewood formation,

while Zhu latewood growth was controlled by current June temperature.

The Gershunov test on temporal stability of the running correlation function coefficients confirms significant changes in relationships between oak growth and environmental variables only for WL (**Figure 6**). The most rapid changes occurred twice: the running correlation coefficient increased in the period of 1903–1922 and decreased during years 1971–1973. Weak correlations in the early 20th century varied in time across different sites and the reason for this remains opaque. The more coherent fall in correlation coefficients was associated with the implementation of the Dnipro regulation systems near Kyiv between 1964 and 1977. The most prominent changes were observed at Byc for RWI and EWI, where significant positive correlations with April and May WL have rapidly reversed to significantly negative correlations within the interval offset of only 3 years. The correlations were at their lowest value throughout the whole period (1978–2000) that followed the completion of river modification constructions. Similar but less abrupt changes in correlations were also registered for Zhu and Dub.

#### DISCUSSION

In this study, we used tree ring width and intra-annual ring-width chronologies (RWI, EWI, and LWI) that were additionally adjusted on reference data allowing us to accurately extract the floodplain-specific signal. All resulted floodplain chronologies contained a strong hydrologic signal, however, the most significant correlations were obtained for RWI an EWI series (**Figures 5**, **6**). This complements some of the findings of earlier studies, which suggested that high variability of LW series reflects a strong climatic signal (Garciá-González and Eckstein, 2003). Great variability of LW might be attributed to a wide range of factors synchronizing tree growth and, consequently, LW might trace climatic information weaker than the less variable EW and its anatomical features (Fonti and Garciá-González, 2008). Despite differences among individual pedunculate oaks in the onset of cambial activity and duration of early vessel formation (Puchałka et al., 2016, 2017), an advantage of using the EW features is a relatively short period of their growth leading to limited imprinted climatic signal. In our work, strong correlations between EWI and WL, particularly in the first period, could be related to the overlap between the high water season and the period of earlywood formation.

A remarkably rapid change in relationships between oak growth and Dnipro's WL was observed at all flooded sites, highlighting the scale of impact of river regulation on floodplain forest growth (**Figure 6**). Two dams on the Dnipro River near Kyiv have altered river's annual water level oscillation by shrinking its variance throughout the season, lowering the mean water level during April–May and elevating it in June–February. This resulted in the unexpected reversal of the positive correlation between floodplain oaks growth and April–May WL as well as in more predictable dampening of growth-to-June-WL correlation. Despite the lack of data on flooded oak growth-to-hydrology relationships in the literature, the positive correlation of oak growth with WL before river regulation is in line with the results reported for ring-porous species at floodplains sites, e.g., for Quercus lyrata RW (Gee et al., 2014), Q. robur RW (Okonski, 2017 ´ ), and LW (Hafner et al., 2015), for Fraxinus excelsior basal area increment (Singer et al., 2013) and RW (Koprowski et al., 2018). The relatively short

1878–2015 (see section "Materials and Methods" for details).

WL rise during the early growing period favors oaks' growth as inundations contribute to soil saturation by water and its enrichment by nutrients. This seems to play a key role at the sites where both floods and low water occurred during the radial growth season, certainly being the case for the studied forests before river regulation. The positive correlation between oak growth and river WL is also intrinsic to floodplains where human impact on hydrology has caused a decrease in a water-table level (Stojanovic et al., 2015 ´ ). In contrast, a negative influence exerted on oaks by floods during early growing season under river regulation resembles a similar effect attributed to the areas liable to floods (Scharnweber et al., 2015) or prone to artificially increased soil water-table (Koval et al., 2015). The physiological consequences of root anoxia (Kozlowski, 1997; Kreuzwieser et al., 2004) and mycorrhiza death (Vasilas et al., 2004) that follow the period of prolonged water excess are among main causes of bog oaks growth depression (Sass-Klassen and Hanraets, 2006; Scharnweber et al., 2015) and seem to be also relevant to our results for the second study period.

The registered growth-climate relationships confirmed water surplus at the flooded sites in the post-regulated period, when growth (Byc EWI and LWI) became negatively correlated with precipitation. Despite the warming trend in regional climate, only June temperatures have significantly and positively affected the oak growth [Zhu LWI (**Figure 5**)]. The weakening of the relationship between non-adjusted RWI from Lis and temperature in May has been earlier considered attributed to the known warming trend and lowering of the evaporation demand, although no linear trend has been found for precipitation in Kyiv (Netsvetov et al., 2018). The positive correlation with temperature has been also reported in other studies of flooded areas (Hafner et al., 2015; Tumajer and Treml, 2016) and areas without water deficiency (Gejková and Polákova, 2012 ˘ ; Tumajer and Treml, 2017), although neither of these works has considered the effect of hydrology modification.

Generally, the modification of river flow is known to affect floodplain ecosystems through alteration of natural disturbances (Nilsson and Berggren, 2000), e.g., flooding. Our findings coupled with earlier data suggest multiple pathways to changes in the growth-to-hydrology/climate relationships after river regulation. The growth responses observed in a single valley (Stella et al., 2013) or even at a single site (Scharnweber et al., 2015; Tumajer and Treml, 2017) may differ significantly, emphasizing the crucial importance of microsite conditions for the trees' ability to withstand artificial hydrology changes. In the Dnipro's valley in Kyiv, all chronologies shared a similar hydrological signal but differed in the extent of changes in correlations following the damming of the river. The oak stands at Lis and Byc were found to be the most sensitive, demonstrating a strong negative LW-WL or RWI-WL relationships (**Figure 5**). These sites experienced hydrology alteration caused by developments in their immediate neighborhood (Byc): a levee construction along the Dnipro's coast (Lis) and modifications of tributaries and intermittent streams (at both sites). Such factors likely impact tree growth, influencing the levels of transient reservoirs and, thus, soil-water availability in the root zone. This bias, however, cannot be reduced in our study as there are no data available on the local groundwater level. Our previous study has demonstrated that the last decades' hydrology alteration has caused Lis oaks to suffer from both yearly growing season inundations and late-summer low-water (Netsvetov et al., 2018). The oscillation in the water table can exacerbate a severe drought impact as the EW-vessel

size and density are governed by soil saturation in water during xylogenesis (Copini et al., 2016) and may became unfit to low water availability later in the season (Tumajer and Treml, 2016).

Our results show that floodplain oaks' RWI, EWI, and LWI adjusted on the reference site's data contain a sufficient hydrologic signal allowing us to discriminate between pre- and post-regulation periods in the river management. The rate and strength of the oak intra-annual ring-width response to river regulation varies among different study sites and is highest in the areas that have suffered from human impact on local hydrology. Though our findings show that the signal from pedunculate oak growth is an effective tool in assessing the impact of hydrological modification on riparian trees, to gain additional insight into the floodplain forests' vulnerability, future researches should also consider the inherent site- and microsite-based variation in environmental conditions and ideally employ data on growth of multiple dominant tree species.

#### DATA AVAILABILITY

The datasets supporting the conclusion of this study are available from the corresponding author upon request.

#### REFERENCES


### AUTHOR CONTRIBUTIONS

YP, MN, and MR contributed to conception and design of the study. YP organized the datasets. YP and MN performed the statistical analysis. MN produced the figures. MN, YP, RP, MKl, and MKo drafted the first version of the manuscript. MR, MN, and YP revised and finalized the manuscript with input from RP, MKo, and MKl. All authors read and approved the final version of the article.

#### ACKNOWLEDGMENTS

RP and MKo acknowledge the support through the Polish National Science Center grant no. 2017/27/B/NZ8/ 00316.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.00096/ full#supplementary-material



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

Copyright © 2019 Netsvetov, Prokopuk, Puchałka, Koprowski, Klisz and Romenskyy. 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.

# Limitations at the Limit? Diminishing of Genetic Effects in Norway Spruce Provenance Trials

Marcin Klisz<sup>1</sup> \*, Allan Buras<sup>2</sup> , Ute Sass-Klaassen<sup>2</sup> , Radosław Puchałka<sup>3</sup> , Marcin Koprowski<sup>3</sup> and Joanna Ukalska<sup>4</sup>

<sup>1</sup> Department of Silviculture and Genetics, Forest Research Institute, Sêkocin Stary, Poland, <sup>2</sup> Forest Ecology and Forest Management, Wageningen University & Research, Wageningen, Netherlands, <sup>3</sup> Faculty of Biology and Environment Protection, Nicolaus Copernicus University, Toruñ, Poland, <sup>4</sup> Biometry Division, Department of Econometrics and Statistics, Faculty of Applied Informatics and Mathematics, Warsaw University of Life Sciences, Warsaw, Poland

#### Edited by:

Giovanna Battipaglia, Università degli Studi della Campania Luigi Vanvitelli, Italy

#### Reviewed by:

Peter Prislan, Slovenian Forestry Institute, Slovenia Susanne Suvanto, Natural Resources Institute Finland (Luke), Finland

> \*Correspondence: Marcin Klisz m.klisz@ibles.waw.pl

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 05 October 2018 Accepted: 26 February 2019 Published: 13 March 2019

#### Citation:

Klisz M, Buras A, Sass-Klaassen U, Puchałka R, Koprowski M and Ukalska J (2019) Limitations at the Limit? Diminishing of Genetic Effects in Norway Spruce Provenance Trials. Front. Plant Sci. 10:306. doi: 10.3389/fpls.2019.00306 Provenance trials are used to study the effects of tree origin on climate-growth relationships. Thereby, they potentially identify provenances which appear more resilient to anticipated climate change. However, when studying between provenance variability in growth behavior it becomes important to address potential effects related to site marginality in the context of provenance trials. In our study we focus on provenance-specific climate sensitivity manifested under marginal growth conditions. We hypothesized that the provenance effects are masked if trials are located at marginal environmental conditions of the natural species distribution. Under this framework, we investigate 10 Norway spruce provenances growing at two contrasting locations, i.e., a relatively drought-prone site in western Poland (at the climatic margin of Norway spruce's natural distribution) and a mild and moist site in northeastern Poland (within its natural range). Combining principal component analysis with climate-growth relationships, we found distinguishable growth patterns and climate correlations among provenances. That is, at the mild and moist north-eastern site, we observed provenance-specific growth patterns and thus a varying drought susceptibility. In contrast, at the dryer western site, provenance-specific growth patterns were less pronounced and all provenances expressed a common and strong sensitivity to drought. Our results indicate that the genetic specificity of growth reactions diminishes toward the distributional margins of a given species. We conclude that the climate conditions at the margins of a species' distribution are constraining tree growth independently of tree origin. Because of this, the marginality of a site has to be considered when evaluating climate sensitivity of provenances within trials. As a consequence, the yet different responses of provenances to adverse growing conditions may synchronize under more extreme conditions in course of the anticipated climate change.

Keywords: radial growth, G × E, Picea abies, drought, phenotypic plasticity

# INTRODUCTION

fpls-10-00306 March 13, 2019 Time: 13:44 # 2

Predicted climate change and increased frequency of extreme weather events may result in sudden, large scale tree-dieback (Allen et al., 2010; Anderegg et al., 2012, 2015; Choat et al., 2012) rather than a gradual shift of species distribution. Sudden, large scale tree dieback is especially relevant at the margins of a species distributional range (Mátyás, 2006). This becomes particularly important if predicted rates of evolutionary response of sessile, long-living organisms (such as trees) are much slower than the predicted rate of climate change (Etterson et al., 2001). In this context, phenotypic plasticity is crucial to survive under changing environmental conditions since it results in a higher fitness of phenotypes which are better adapted to prevailing conditions (Thompson, 1991; Dudley, 2004). In recent years the frequency of drought and heat events has been increasing, however expressing a large spatial heterogeneity (Steinkamp and Hickler, 2015). Moreover, trees response to drought stress also shows great diversity among forest types, tree species, and provenances (Alberto et al., 2013; Vicente-Serrano et al., 2014). Given this variability and the associated uncertainty, forestry has to face the challenge to select tree species and provenances which are able to cope with diverse climate conditions (Bolte et al., 2009).

Although Norway spruce (Picea abies (L.) H. Karst) has expressed drought-induced dieback since the 1970s it yet is considered one of the key tree species for European forestry (Roberts et al., 1989). Several hypotheses indicate different biotic and abiotic environmental factors determining spruce decline (Grodzki, 2010; Hentschel et al., 2014). According to recent studies on climate adaptation of tree species in Europe, Norway spruce is supposed to be more vulnerable to climate change than other more drought-tolerant species e.g., Abies alba Mill. and Pseudotsuga menziesii (Mirb.) Franco (Lévesque et al., 2013; Zang et al., 2014; Vitali et al., 2018). Furthermore, water shortage during the growing season considerably increased Norway spruce vulnerability to bark beetle attack (Ips typographus L.) (Netherer et al., 2015). In addition, a recent projection of European treespecies distributions, indicates a large decline of Norway spruce abundance in Central Europe (Buras and Menzel, 2019). Thus, the risk of maladaptation to current and future climate is likely, unless forest management strategies incorporate climatebased seed transfer, thereby utilizing the phenotypically more suitable provenances (Isaac-Renton et al., 2014; Frank et al., 2017). In this context, soil properties may also significantly influence Norway spruce growth performance (Lévesque et al., 2013; Rehschuh et al., 2017).

The observed and projected negative effect of climate change on the condition and productivity of forests justifies incorporating an adaptation strategy into forest management. The most powerful tool for studying the genotype and environment interaction effect (G × E) in the context of tree adaptation are provenance trials and common garden experiments (Mátyás, 1994; Pâques, 2013). Although studying between-provenance variation on replicant experiments may provide useful simulation of environmental change over time (Mátyás, 1997) the diminishing effect of marginal environmental conditions can make it impossible to detect a provenancespecific, climate-related adaptive response. Within this context, marginal environmental conditions refer to climatic properties that represent the climatic margins of a species distributional range (Mellert et al., 2016). Differences in adaptive traits both within and among populations are observed even at the climatic margin of the species distribution (Skrøppa and Johnsen, 1999; Savolainen et al., 2004). However, under optimal growing conditions differences among provenances are pronounced, whereas under more adverse growing conditions, differences among phenotypes may diminish (Mátyás, 1994). These findings suggest a tendency toward uniform growth reactions under unfavorable (i.e., marginal) environmental conditions regardless of the genotype. That is, depending on the marginality of the trial or common garden site, phenotypic differences might be reduced (at the margins) or pronounced (in the center of the species distributional range). Consequently, site-specific environmental conditions related to climate but also soil properties and hydrology should be taken into consideration (Mátyás and Yeatman, 1992). According to Liebig's law of the minimum (Liebig, 1840) plant productivity reflects the variation of the limiting factor (e.g., growing season temperature at high elevations and latitudes), which may cause betweenprovenance adaptive variation to become negligible. Moreover, if the adaptation process is modified by multiple environmental factors, the rate and the course of the process of adaptation is limited by the pace of the "slowest" factor, i.e., the factor which features the slowest rate of change in relation to other environmental factors (Blackman, 1905). This may be explained by the effect of the relatively slower factor (with a slower growth rate) to condition the influence of other factors. In this context, a complex limiting factor is drought stress as a secondary effect of different types of climate induced stress, e.g., heat waves, soil water depletion, intensive solar radiation, and their combination (Taiz and Zeiger, 2010). Consequently, when tracing phenotypic adaptation of trees to drought not only intensity but also duration of the prevailing stress factor should be considered.

Within this context, we here address a central question related to the appropriate selection of tree provenances for climate-smart forestry (Nabuurs et al., 2018):

Does species-specific marginality of a site affect provenancespecific climate sensitivity within provenance trials and if so, how?

Based on existing studies outlined above, we hypothesize that provenance-specific growth reactions will diminish toward the climatic marginality of the considered tree species. If so, this would have important implications for provenance trials in the context of adapting forestry to the anticipated climate change.

#### MATERIALS AND METHODS

#### Study Sites and Sample Acquisition

The study material is part of the International Union of Forest Research Organizations (IUFRO) Spruce provenance trial established in 1972 within the framework of a cooperation among research institutes from 10 European countries and

Canada (Matras, 2009). 10 Polish provenances included in the provenance trial series represent the southern, northern and central range of Norway spruce in Poland (**Figure 1**) and the highest possible genetic variation among Polish provenances (Klisz et al., 2019). Out of altogether four provenance trials located in Poland, two were selected: Kórnik (KR; N 52.237291◦ E 17.076364◦ ) and Knyszyn (KN; N 53.327119◦ E 23.060735◦ ).

The site selection aimed at representing the diversity of climatic conditions and site conditions in relation to the natural distribution area of Norway spruce. Although the average annual climate parameters and climatic water balance (CWB) of the growing season of both sites were similar, the Walter–Lieth annual aridity index (WAI) indicates that KR is a relatively drought-prone site while KN is mild and moist (29.6 and 25.7, respectively). Data from weather stations in Gorzów and Białystok, for the reference period 1973– 2015 [National Oceanic and Atmospheric Administration agency (NOAA/NCEI/CWC<sup>1</sup> )] was used to determine the climatic conditions of provenance trials (**Figure 2**). Climatic conditions of KR can be considered more marginal/limiting according to the ecological requirements of Norway spruce (Tjoelker et al., 2007). To explore climate-growth relationships, total sum of monthly precipitation (P), and mean monthly temperature (T) from Gorzów and Białystok meteorological stations (years 1981–2014) were obtained from the European Climate Assessment and Dataset (ECA&D) project (Klein Tank et al., 2002) for KR and KN, respectively. Walter– Lieth annual aridity index (WAI) was calculated for each site to determine their relative water surplus or deficit. To detect periods with negative water balance, we further computed the standardized precipitation evapotranspiration index (SPEI), while the Palmer drought severity index (PDSI) was extracted from the Climate Explorer<sup>2</sup> . The PDSI varies slowly underlining the accumulation effect of long-term hydrological drought (Palmer, 1965). Integrated over three and six month periods, SPEI (SPEI3 and SPEI6, respectively) is defined as a standardized difference between monthly precipitation and potential evapotranspiration (PET) (Beguería and Vicente-Serrano, 2013; Beguería et al., 2014). Moreover, for each site we computed the monthly CWB for the period 1981–2014 (**Figure 3**). According to Thornthwaite (1948), CWB is the difference between total precipitation and PET. Due to the limited availability of meteorological data, the Hargreaves equation was used to estimate PET,

<sup>1</sup>http://www.noaa.gov/

<sup>2</sup>https://climexp.knmi.nl

FIGURE 1 | Location of study sites. Blue triangles – experimental sites, red circles – Norway spruce provenances. KR, KN – Kórnik and Knyszyn testing site; BA, BO, PR, NR, IS, WS, RY, OR, WT, and ZL Norway spruce provenances; green area – natural distribution area of Norway spruce (European Forest Genetic Resources Programme [EUFORGEN], 2009).

**114**

FIGURE 3 | Temporal variations of main climatic parameters: Palmer Drought Severity Index (PDSI; A and G), standardized precipitation evapotranspiration index integrated over three (SPEI3; B and H) and six months period (SPEI6; C and I), climatic water balance (CWB; D and J), monthly mean temperature (Tmean; E and K) and monthly precipitation sum (Prec; F and L). Panels A–F: Knyszyn site, panels F–K: Kórnik site. Common pointer years (CPY, more than five provenances) are indicated by light gray (positive) and dark gray (negative) poligons.

which is an alternative for the FAO 56 Penman–Monteith method (Hargreaves and Samani, 1985; Allen et al., 1998):

$$PET = 0.408 \times 0.0023 \left( T\_{\text{mean}} + 17.8 \right) \left( T\_{\text{max}} - T\_{\text{min}} \right)^{0.5} R\_{\text{a}} \tag{1}$$

where PET is the potential evapotranspiration, Tmean is the mean of Tmax and Tmin, R<sup>a</sup> is extraterrestrial radiation, and 0.408 being a factor to convert units of MJ m−<sup>2</sup> day−<sup>1</sup> into mm day−<sup>1</sup> .

Regarding soil fertility, sites were relatively similar, i.e., a Mollic Gleysol in KR and a Cambisol in KN (IUSS Working Group, 2015). The provenance trials were designed according to the scheme of a random block design with four replicates, with 3-year-old seedlings with 1.65 × 1.45 m spacing. For each of the 10 provenances, 15 trees were sampled at breast height (1.3 m) for two increment cores at the end of the growing season in 2015, resulting in 150 trees from KN but only 134 from KR. The lower sample size in KR is to be explained by the fact that for three provenances (BO, NR, and RY) the number of trees suitable for sampling was lower than ten since we avoided sampling of suppressed trees, characterized by symptoms of fungal infection as well as trees located near gaps.

# Ring-Width Analyses

#### Sample Preparation To prepare the sample material for tree-ring analysis, cores

were manually surfaced, polished with abrasive paper (grain size up to 1000) and scanned (Epson Expression XL12000) at 1200 dpi resolution. Ring widths were measured with an accuracy of 0.01 mm and then cross-dated with "CooRecorder" and "CDendro" software (version 9.0, Cybis, 2017). The two series per tree were averaged into one series resulting in altogether 274 ring-width series. Individual tree-ring series were detrended using a cubic-smoothing spline with a 50% frequency cut-off at 30 years (Cook and Peters, 1981; Speer, 2010). To remove temporal autocorrelation and to emphasize the high-frequency signal (year-to-year variability) of the tree-ring series, the firstorder autoregressive model (aka prewhitening) was applied to each series (Cook and Kairiukstis, 1992) finally resulting in indexed ring-width series (RWI). Using RWI, mean chronologies were computed per provenance and site using a biweight robust mean, resulting in altogether 20 provenance-chronologies.

#### Tree-Ring Data

To characterize and qualify site chronologies, Gleichläufigkeit (glk aka coefficient of coherence, Eckstein and Bauch, 1969; Buras and Wilmking, 2015); mean sensitivity (MS, indicator of general climate sensitivity of growth) and mean inter-series correlation (mean rbar, an indicator of the strength of the common signal in growth series from individual trees within a stand) were calculated on the basis of ring-width series (Douglass, 1920; Wigley et al., 1984; Cook and Kairiukstis, 1992).

#### RW Variation Between Sites and Provenances

To determine the effects of site, provenance, and year on radial growth variation, a generalized estimating equation (GEE) was used (Liang and Zeger, 1986). The following model was applied:

$$g(\mu\_{\rm ijk}) = S\_{\rm i} + P\_{\rm j} + Y\_k + S \times P\_{\rm ij} + S\_{\rm i} \times Y\_{\rm ik} + P\_{\rm j} \times Y\_{\rm jk} \tag{2}$$

where g(µijk) is the identity link function, µijk is the singletree ring-width index mean for the ijth site × provenance combination in the kth year (k = 1,. . ., 31); S<sup>i</sup> is the effect of the ith site (i = 1,2); P<sup>j</sup> is the effect of the jth provenance (j = 1,...,10); S × Pij is the site × provenance interaction effect; S × Yik is the site × year interaction effect; P<sup>j</sup> × Yjk is the provenance × year interaction effect and Yik is the repeated measurements year effect. We used GEE since it efficiently handles data with repeated measurements as well as possible inter-correlations among data at the tree level (285 trees – clusters).

To estimate model parameters, "sandwich" (empirical) estimators were used, which are asymptotically unbiased even if the correlation structure is unknown. However, choosing a working correlation matrix that is closer to the truth improves the efficiency of estimates. We assumed compound symmetry for the data from the same tree. The analyses were done with SAS/STAT 14.3 software, and the GLIMMIX procedure was followed (Software suite developed [SAS] Institute, 2017).

#### Provenance Grouping

To group mean RWI provenance chronologies according to their similarity, a hierarchical clustering using Euclidean distance (root sum-of-squares of differences) as similarity measure and Ward (1963) clustering method with the criterion proposed by Murtagh and Legendre (2014) was applied. Four different clustering methods, single and complete linkage, the unweighted pair group method with arithmetic mean (UPGMA) and Ward's method were tested according to their clustering structure of the dataset (Kaufman and Rousseeuw, 2008). Finally, Ward's method was chosen, since it expressed the highest value of the agglomerative coefficient.

To investigate provenance-specific individualistic growth reaction of trees, we for each site computed a refined version of the principal component gradient analysis (PCGA, Buras et al., 2016). That is, we performed a pairwise PCGA only adding the RWI of two provenances for each of the possible provenance combinations per site. In each pairwise PCGA, Wilcoxon rank-sum test (Wilcoxon, 1945) was used to test whether the polar coordinates of RWI-loadings express a provenance-specific location shift which would indicate that the considered provenances express provenance-specific growth patterns in comparison to each other.

#### Provenance-Specific Climate Sensitivity

A pointer-year analysis was carried out using single-tree RWI to test whether the growth reactions to extreme events varied between provenances and sites (Schweingruber et al., 1990). The "Neuwirth" method, a window size of 5 years, and a series threshold of 65% were used as criteria for weak, strong, and extreme events, where the intensity classes refer to Cropper values of >1, >1.28 and >1.645, respectively (Cropper, 1979; Neuwirth et al., 2007). Years in which at least six of the ten provenances indicated a pointer year at the specific site were defined as common pointer years (CPY). Furthermore, to identify climatic drivers of growth variability, indexed provenance chronologies were correlated with previous year March through current year October

temperature, precipitation, SPEI3, SPEI6, PDSI and CWB for 1981 through 2014.

#### Bioclimatic Variation Between Sites and Provenances

To characterize the climatic distance between the provenance origins and the trial sites, a principal component analysis (PCA) using 19 bioclimatic parameters, related to monthly and seasonal precipitation and temperature, was performed (**Tables 1**, **2**). For this, climate data for the two trial sites KR and KN as well as for the 10 Norway spruce origins were extracted from BIOCLIM 1.4 (Hijmans et al., 2005) at a spatial resolution of 2.5 arcmin taken from the www.worldclim.org website. The selected variables contained data concerning bioclimatic indices calculated on the basis of monthly, seasonal (i.e., three-month periods) and annual values of precipitation and temperatures

TABLE 1 | Pearson correlation coefficients between climatic variables and the first three major components, eigenvalues and variation explained.


for the period 1970–2000. **Tables 1** and **2** provide a simplified overview on the bioclimatic variables. For a detailed description of the bioclimatic variables, we refer to Hijmans et al. (2005). The variables with the strongest impact on the distribution of the provenances and sites along the principal components was identified on the basis of Pearson correlation coefficients. Clustering of provenances according to their locations on the PCA biplot were determined visually.

All analyses were computed in "R" (R Core Team, 2015). Extraction of climatic data values for spruce sites was done with biovars function in the R package "dismo" 1.1-4 (Hijmans et al., 2017). PCA analyses as well as the biplot were created with fviz\_pca\_biplot functions from the "FactoMineR" 1.41 package (Husson et al., 2015). Detrending, chronology building, and calculation of chronology statistics were performed using the "dplR" package 1.6.4 (Bunn, 2008). Clustering analyses were calculated using "cluster" 2.0.7-1 package (Rousseeuw et al., 2018). Pointer-year analysis was done by the "pointRes" 1.1.3 package (van der Maaten-Theunissen et al., 2015). Integrated over three and six month periods, SPEI were calculated using "SPEI" 1.7 package (Beguería and Vicente-Serrano, 2013).

### RESULTS

#### Characteristics of RW Chronologies

The mean sensitivity, and mean inter-series correlation varied between provenance chronologies, however, they generally featured higher values for the chronologies from KR than from the KN site (**Table 3**). In KR, a strong growth depression between 2003 and 2006 highlighted a bark beetle (I. typographus L.) outbreak (**Figure 4**). The results of GEE analysis (Eq. 2) for RWI confirmed the significant effect of site, provenance, and year (P = 0.001, P < 0.001, P < 0.001, respectively). Moreover, the interaction effects between site and provenance, site and year, as well as provenance and year were significant (**Table 4**; in all cases P ≤ 0.001).

#### Provenance Grouping

The results of the cluster analysis confirmed grouping of provenance chronologies mainly by site. However, the provenance grouping patterns within the sites were different (**Figure 5**). That is, at KN provenance WT took a special position relative to the remaining provenances. Moreover, the eastern provenances BA and ZL took a distinct position within the largest sub-cluster. In contrast, at KR four of the five southern provenances (WS, RY, WT, and IS) indicated a separate cluster (**Figure 5**). The pairwise PCGA distinguished pairs of provenances differing in terms of high-frequency growth patterns. That is, at KR the provenances WS and WT were highlighted as significantly different from the other provenances (**Figure 6A** and **Supplementary Figure S1**). The outstanding position of the provenance WT was also present at KN, although here the other, clearly distinguishable provenance was BA (**Figure 6B** and **Supplementary Figure S2**). In general, site KN expressed a higher diversity among provenances according to pairwise PCGA in comparison to KR. In other words, the

#### TABLE 2 | Mathematical definition of bioclimatic indices.

fpls-10-00306 March 13, 2019 Time: 13:44 # 7


TABLE 3 | Chronology statistics of Norway spruce provenances.


N – number of trees, TRW – mean tree-ring width, glk – Gleichläufigkei, MS – mean sensitivity, rbt – mean inter-series correlation.

differentiation between provenances was much stronger in KN compared to KR, where provenances featured more similar growth patterns.

#### Climate-Growth Relationships

The CPY clearly separated the two sites from each other since there was no similar CPY for both sites. Interestingly, the two CPY at the KN site were negative (2000 and 2003) while those at the KR site were positive (1987, 1991, and 1997) (**Figures 3**, **7**). However, it is noteworthy, that despite no CPY in 2003 at KR, all chronologies were characterized by a growth depression beginning in 2003 (**Figure 4**). Climate correlations mainly differentiated provenance trials, indicating a high diversity of the sites in terms of climatic conditions. However, some climatic drivers seemed to be common for most provenances at both sites (e.g., SPEI3, SPEI6 and PDSI for July for RWI or T of previous April, P and SPEI3; **Figure 8** and **Supplementary Figure S3A**). Namely, RWI chronologies were mainly influenced by drought conditions of summer and autumn months, in particular over the period May – August (**Figures 8C,D**). However, for the generally drier KR site an additional sensitivity to September – October drought-indices was apparent (**Figure 8D** and **Supplementary Figure S3A**). Moreover, T, P, and CWB appeared as climate parameters mainly responsible for the growth of provenances in KR site. In particular, T, P and CWB of previous August and September as well as P and the CWB of May and June, were significantly correlated with RWI (**Figures 8A,B** and **Supplementary Figure S3B**). Among the correlations with drought indices (mainly SPEI3 and SPEI6 but also PDSI and CWB), the provenance WT was clearly separated from the other provenances in KN. The SPEI3 and SPEI6 indices, and to a lesser extent PDSI and CWB for the months of the preceding growing season, correlated negatively with RWI. In comparison, climate correlations revealed a more homogeneous growth response of provenances in KR compared to KN, underlining the impression derived from PCGA that between-provenance growth patterns are more similar in KR and more heterogeneous in KN (**Figure 8**).



Num Df – nominator degrees of freedom; Den Df – denominator degrees of freedom; F – test statistic; Site – experimental site effect; Prov – provenace effect; Year – year effect.

#### Climate-Shift Effect

The PCA of the 19 bioclimatic variables representing the two trial sites as well as the 10 Norway spruce provenance origins successfully allowed for reducing the multidimensionality to two dimensions. That is, the first two principal components (PC) were able to explain 91.98% of the overall variance (**Figure 9** and **Table 1**). The first PC explained 75.35% of variance and was positively correlated (r > 0.25) with precipitation (bio12- 14, bio16-19) and negatively correlated (-0.26 < r < -0.20) with annual mean temperature and mean temperature of the growing season (bio1, bio4-11). The second component (PC2) explained 16.63% of variance (**Table 1**) and was positively correlated (0.32 < r < 0.51) with the mean monthly temperature range, isothermality, mean temperature of the coldest quarter and precipitation seasonality (bio2, bio3, bio11, and bio15, respectively). Furthermore, PC2 was negatively correlated with temperature seasonality (bio4, r < -0.23).

The northern provenances (BO, PR, and NR) formed a cluster, indicating their similarity in terms of their climate origin, while another cluster was formed by the southern provenances (IS, RY, WS and OR) (**Figure 9**). However, the observed spread of southern provenances was much wider compared to northern provenances. Only the eastern provenances (BA and ZL) and the provenance from the Tatry Mountains (WT) were not included in any cluster. WT was located on the periphery of the PC biplot, which resembles that it originates from the highest location above sea level. It is important to highlight the mutual location of the provenances and the two sites. The distance between the sites and the provenances in relation to the PC axes provides information about the effect of the climatic shift associated with the change of the climate at the trial sites in comparison to the respective provenance origin. For instance, the distance between the scores of the two sites to the provenance WT depicts that

it experienced the greatest climatic shift. Moreover, bioclimatic conditions differentiating the origin of the WT provenance to the KN site were mostly related to PC1 while at the KR site they were related to both PC1 and PC2.

# DISCUSSION

Our study confirmed the importance of site marginality for provenance-specific climate sensitivity within provenance trials, showing that optimal climate conditions enhance provenance differentiation (Knyszyn site) while adverse growing conditions resulting from a negative water balance may reduce the effect of provenance-specific growth patterns (Kórnik site). In particular the higher homogeneity of growth patterns between provenances as indicated by the PCGA as well as the more homogeneous climate correlations point toward a weaker differentiation between provenances at KR, i.e., the site with marginal climate conditions. These findings indicate, that provenance-specific growth reactions likely diminish toward the climatic marginality of the considered tree species. This has two important implications in the context of adapting forestry to climate change. First of all, based on our study, site marginality seems to be an important factor to consider when evaluating

data from provenance trials. Secondly, even though provenances may express contrasting responses to adverse climate conditions nowadays, their response to anticipated climate conditions may be less diverse. Nonetheless, foresters should aim at selecting provenances which currently cope best with adverse conditions, but it seems noteworthy that even for those provenances, speciesspecific climatic margins cannot be surpassed. To study the extent to which these provenances can cope with anticipated climate change, marginal sites might be considered candidates of particular interest.

## Individualistic Growth Response and Provenance Clustering

According to the pairwise principal component gradient analysis of provenances, two provenances were characterized by a distinct growth behavior: WT and WS in Kórnik and WT and BA in Knyszyn (**Supplementary Figures S1, S2**, respectively). The uniform individualistic growth reaction of trees representing the provenance WT indicates relatively low within-population variability and therefore confirms the long-term effect of the selection pressure (Dudley, 2004). Moreover, in case of WT Liebig's law of the minimum can be applied both at the level of provenance chronologies as well as at the individual tree level (Stine and Huybers, 2017). This suggests that almost all trees of WT responded to a common growth factor and not as in case of other provenances to different local growth factors (Chapin et al., 2011). In turn, WS and BA provenances were only characterized by a homogeneous single-tree reaction under specific environmental conditions. As for WT, this likely relates to their bioclimatic distance to the trial Knyszyn in the case of BA and the bioclimatic distance to the trial Kórnik in the case of WS (**Figure 9**). This observation may be interpreted as an effect of artificially transferred populations (Eriksson et al., 1980).

The cluster analyses clearly separated the two experimental sites from each other (**Figure 5**). This confirms results of previous studies that covered a much larger series of provenance trials and which observed a site-specific clustering, too (Karlsson et al., 2001; Suvanto et al., 2016). If we assume that one of the main drivers of Norway spruce growth behavior is water balance – which remarkably differs between Kórnik and Knyszyn – then site-related provenance grouping seems reasonable (Lévesque et al., 2013; Rehschuh et al., 2017). The special position of WT in the cluster analysis likely relates to the special bioclimatic position of this provenance which originates from the highest elevation with the highest annual precipitation sums, hence seems to be determined by the ecological distance between climatic conditions at the location of origin and location of planting, i.e., the Knyszyn site (**Figure 9** and **Table 1**). Surprisingly, in the sub-cluster representing Kórnik the provenance WT formed a small cluster with the southern provenances WS, RY and IS (**Figure 5**). This however is in line with Koprowski and Zielski (2006) who found that Norway spruce provenances from the so-called "spruceless area" have more similar climate-growth relationships compared to those from southern Poland.

However, another reason for the observed grouping pattern may be the distribution of genetic diversity in Norway spruce related to migration events (Chen et al., 2019). A recent study on Norway spruce pollen distribution in sediments confirmed that the recolonization of Poland by spruce occurred from both the north-west and the south-east (Kupryjanowicz et al., 2018). Moreover, during the last two centuries a massive import of tree seeds in Europe and a shift from natural regeneration to manual seeding significantly affected the genetic structure of Norway spruce populations (Myking et al., 2016).

#### Climate-Driven Between-Provenance Variation

Based on the performed climate correlations, drought appeared as the main growth-limiting factor for all provenances (**Figure 8** and **Supplementary Figure S3**). Nevertheless, a site-related variation in the sensitivity to drought stress was indicated by different CPYs in Kórnik compared to Knyszyn (**Figure 3**).

(panel A), precipitation (panel B), SPEI3 (panel C), SPEI6 (panel D) over the period 1981–2014 at the two sites: KR, KN. Colors represent correlation coefficient, non-significant correlations are not represented (white), <sup>∗</sup> , ∗∗ and ∗∗∗ demarcate a significance level of correlation (p < 0.05, p < 0.01 and p < 0.001, respectively).

Two negative CPYs in Knyszyn likely relate to severe drought events in 2000 and 2003 widely reported throughout Europe (e.g., Rebetez et al., 2006; He et al., 2018). In contrast, in Kórnik three positive CPYs were recorded in 1987, 1991, and 1997, of which at least one (1997) may have been the result of high precipitation sums in western and southern Poland (Knight and Le Comte, 1998; Kundzewicz et al., 1999). However, based on the inspection of indexed ring-width series (**Figure 4**) it is obvious that Norway spruce in Kórnik also reacted negatively to the heat wave of 2003. That is, all provenance chronologies in KR showed a growth decline in 2003 which was followed by a growth depression that lasted until 2006. Evidently, the metric on which pointer year analysis is based (Cropper values of a moving window spanning five years) was unable to identify 2003 as a pointer year because the values after 2003 also were comparably low. From this we conclude that the effect of the 2003 heat wave was even more pronounced in Kórnik and eventually resulted in a bark-beetle outbreak which underlines the drought susceptibility and affectedness of trees in Kórnik. Thus, the widely known effect of the cumulative impact of stress load may effectively diminish the genetic variability of provenance-specific growth patterns (Rolland and Lempérière, 2004; Netherer et al., 2015). According to Marini et al. (2012), Norway spruce is more prone to bark beetle outbreaks when growing under climate conditions warmer than those of its historical climatic range.

Among provenances in Knyszyn which did not reduce growth in response to 2003 drought year were WT and WS, both characterized by a long ecological distance to Knyszyn site conditions (**Figures 7A**, **9**). In contrast, provenances BO, PR, and RY did not react to the less severe drought in 2000. Hence the susceptibility of provenances to extreme weather events under moderate climate conditions (as in Knyszyn) appears to be a more complex process, related to adaptation to local conditions in relation to their climate origin – a phenomenon which is known as the transfer effect (Mátyás, 1994).

While exploring between-provenance variation in climate sensitivity the effect of mortality has to be considered since according to Mátyás and Bozic (2009) the genetic tolerance limit of adaptation to adverse climatic conditions may end up in mass mortality. That is, initial effects of climatic extremes such as drought may increase the population's sensitivity to other abiotic and biotic threats eventually leading to increased rates of mortality (Mátyás et al., 2010). In course of the growth depression observed in KR following the hot and dry summer of 2003 (**Figure 4**), an increased mortality of Norway spruce was observed, which underlines the importance of mortality at marginal sites. Unfortunately, we lack detailed data to assess whether specific provenances were particularly affected by drought-triggered bark-beetle infestation and eventual die-back. However, it provenance-specific tree vulnerability to bark-beetle infestation seems possible, wherefore future investigations should consider obtaining according information when evaluating data from provenance trials.

Regarding the climate correlations, a detailed analysis of the provenance susceptibility to drought conditions (mainly visible in the SPEI integrated over three and six month periods but also temperature) demonstrates a relatively uniform response of all provenances on the Kórnik site and, in turn, a differentiated reaction on Knyszyn site (**Figures 8A,C,D**). In the case of Knyszyn site, the distinct climate reaction of the WT provenance is clearly visible, but interestingly, unlike in case of the principal component gradient analysis, is not observable in Kórnik. Thus, a significant effect of provenance and site interaction as obtained with the generalized linear model (**Table 4**) is supported by the climate correlations, however most likely due to the separate reaction of the WT provenance. In conclusion, it seems that provenances expressed more diverse growth patterns and climate correlations in Knyszyn, while growth patterns and climate correlations were more homogeneous in Kórnik. The higher homogeneity of growth patterns and correlations in Kórnik we interpret as a common reaction to more adverse – in this case dryer – growing conditions.

## Site and Provenance Interaction and Phenotypic Plasticity

Significant effects of the site, provenance and year, and their interactions in the generalized linear model suggest different climate and site-related adaptive responses of the provenances (**Table 4**). Significant effects of site and provenance interaction may indicate between-provenances variation in plasticity (Dudley, 2004). The site and provenance interaction was mainly found to be significant in studies covering a wide range of environmental conditions (Karlsson et al., 2001; Stojnic et al., 2015; Suvanto et al., 2016), however it generally appears when in one environment genetic variation is almost unnoticeable while it is high in another one (Windig et al., 2004). This is confirmed by our study, where at the site of adverse growing conditions (Kórnik) genetic variation was barely noticeable while under optimal conditions (Knyszyn) between-provenance variation was pronounced. This is consistent with results of Androsiuk and Urbaniak (2014) who stated that the location of a provenance trial influences the level of genetic polymorphism and the patterns of inter-population differentiation.

#### CONCLUSION

Although common garden experiments are considered one of the best tools to simulate the response of trees to anticipated climatic conditions, interpreting the results requires acknowledging the complexity of the phenotypic adaptation process. Besides the intra- and inter-population genetic variation, it is essential to consider the transfer effect expressed in the phenotypic reaction in dependence of the bioclimatic distance between the climate of provenance origin and the climate of the trial site. Moreover, the climatically determined site potential defines the match between actual environmental conditions and ecological needs of the species. That is, the limiting climate conditions of the experimental site may diminish differences of climate-growth responses among provenances which in turn may hamper the appropriate selection of phenotypically stable populations. Consequently, some provenances may better cope with extreme weather under moderate climate conditions, while all provenances homogeneously will respond with growth depression to adverse conditions. In view of the above, it seems reasonable to select sites with moderate climate conditions for provenance trials when aiming at identifying phenotypes which cope best with single extreme events. On the other hand, when seeking to explore the general performance of a species under

#### REFERENCES


extreme climate conditions (thus prevailing over several years), sites with more adverse climatic conditions should be considered.

#### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

#### AUTHOR CONTRIBUTIONS

MKl and AB performed ring width analyses and PCGA analyses. JU was in charge of the generalized linear model. MKl, MKo, and RP performed the climatic analyses. MKl and RP performed the bioclimatic analyses. MKl wrote the first draft of the manuscript. All authors gave a substantial contribution to the conception and design of the study, wrote specific sections of the manuscript, and contributed to manuscript revision, read and approved the submitted version.

#### FUNDING

AB was supported by the German Academic Exchange Service (DAAD). MKo and RP was supported by a grant from the Polish National Science Centre (2017/27/B/NZ8/00316). MK was supported by the Forest Research Institute statutory aid No. 24.02.38 of the Ministry of Science and Higher Education in Poland and Science.

#### ACKNOWLEDGMENTS

We are grateful for the valuable comments provided by the reviewers. This research was performed under the Forest Research Institute statutory aid No. 24.02.38 of the Ministry of Science and Higher Education in Poland and Science. This research was linked to activities conducted within the COST FP1106 "STReESS" network and SUSTREE – Interreg project "Conservation and sustainable utilization of forest tree diversity in climate change."

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.00306/ full#supplementary-material

mortality reveals emerging climate change risks for forests. For. Ecol. Manag. 259, 660–684. doi: 10.1016/j.foreco.2009.09.001



Cramer, A. V. Morgan, H. C. Prentice, and J. R. M. Allen (Berlin: Springer-Verlag), 357–370. doi: 10.1007/978-3-642-60599-4


abies," in Forest Genetics and Sustainability, ed. C. Mátyás (Dordrecht: Springer), 49–58.


**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 Klisz, Buras, Sass-Klaassen, Puchałka, Koprowski and Ukalska. 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.

fpls-10-00306 March 13, 2019 Time: 13:44 # 15

# Tree Circumference Changes and Species-Specific Growth Recovery After Extreme Dry Events in a Montane Rainforest in Southern Ecuador

Volker Raffelsbauer<sup>1</sup> , Susanne Spannl<sup>2</sup> , Kelly Peña<sup>3</sup> , Darwin Pucha-Cofrep<sup>3</sup> , Kathy Steppe<sup>4</sup> and Achim Bräuning<sup>1</sup> \*

1 Institute of Geography, Friedrich Alexander University Erlangen-Nürnberg, Nuremberg, Germany, <sup>2</sup> Department of Plant Physiology, University of Bayreuth, Bayreuth, Germany, <sup>3</sup> Laboratorio de Dendrocronología y Anatomía de la Madera, Carrera de Ingeniería Forestal, Universidad Nacional de Loja, Loja, Ecuador, <sup>4</sup> Laboratory of Plant Ecology, Department of Plants and Crops, Faculty of Bioscience Engineering, Ghent University, Ghent, Belgium

#### Edited by:

Giovanna Battipaglia, Università degli Studi della Campania Luigi Vanvitelli Caserta, Italy

#### Reviewed by:

Alexandros Galanidis, University of the Aegean, Greece Ze-Xin Fan, Xishuangbanna Tropical Botanical Garden (CAS), China

> \*Correspondence: Achim Bräuning achim.braeuning@fau.de

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 08 October 2018 Accepted: 05 March 2019 Published: 22 March 2019

#### Citation:

Raffelsbauer V, Spannl S, Peña K, Pucha-Cofrep D, Steppe K and Bräuning A (2019) Tree Circumference Changes and Species-Specific Growth Recovery After Extreme Dry Events in a Montane Rainforest in Southern Ecuador. Front. Plant Sci. 10:342. doi: 10.3389/fpls.2019.00342 Under drought conditions, even tropical rainforests might turn from carbon sinks to sources, and tree species composition might be altered by increased mortality. We monitored stem diameter variations of 40 tree individuals with stem diameters above 10 cm belonging to eleven different tree genera and three tree life forms with highresolution dendrometers from July 2007 to November 2010 and additionally March 2015 to December 2017 in a tropical mountain rainforest in South Ecuador, a biodiversity hotspot with more than 300 different tree species belonging to different functional types. Although our study area receives around 2200 mm of annual rainfall, dry spells occur regularly during so-called "Veranillo del Niño" (VdN) periods in October-November. In climate change scenarios, droughts are expected with higher frequency and intensity as today. We selected dry intervals with a minimum of four consecutive days to examine how different tree species respond to drought stress, raising the question if some species are better adapted to a possible higher frequency and increasing duration of dry periods. We analyzed the averaged species-specific stem shrinkage rates and recovery times during and after dry periods. The two deciduous broadleaved species Cedrela montana and Handroanthus chrysanthus showed the biggest stem shrinkage of up to 2 mm after 10 consecutive dry days. A comparison of daily circumference changes over 600 consecutive days revealed different drought responses between the families concerning the percentage of days with stem shrinkage/increment, ranging from 27.5 to 72.5% for Graffenrieda emarginata to 45–55% for Podocarpus oleifolius under same climate conditions. Moreover, we found great difference of recovery times after longer-lasting (i.e., eight to ten days) VdN drought events between the two evergreen broadleaved species Vismia cavanillesiana and Tapirira guianensis. While Vismia replenished to pre-VdN stem circumference after only 5 days, Tapirira needed 52 days on average to restore its circumference. Hence, a higher frequency of droughts might increase inter-species competition and species-specific mortality and might finally alter the species composition of the ecosystem.

Keywords: stem diameter variations, dendrometer, tropical mountain rainforest, drought recovery, tree life form

# INTRODUCTION

fpls-10-00342 March 21, 2019 Time: 19:38 # 2

Global change, especially climate change, affects forests worldwide, with adverse effects on biodiversity and ecological services like carbon sequestration. Hence, understanding forest responses to climate variability is key to conservation and protection of forest ecosystems (Anderson-Teixeira et al., 2015). Especially, selective mortality of species can trigger long-term shifts in forest communities (Anderegg et al., 2013). Climate induced changes in forest composition are therefore an important scientific topic since several decades (Overpeck et al., 1990). Especially, drought as one of the most frequent climatic extremes on a global scale (Allen et al., 2009, 2015) and the accompanied increase in atmospheric vapor pressure deficit is, due to its partly devastating impact on forests, still under intense debate (Sass-Klaassen et al., 2016). According to climate modeling results (Fu et al., 2013), drought frequency and intensity in the tropics will increase in the near future. Recent studies on gross primary production and ecosystem respiration (Cavaleri et al., 2017) revealed that tropical forests may shift from carbon sinks to sources under drought conditions (Qie et al., 2017). Monitoring stem radial increment with high-resolution dendrometers can provide useful information if trees are in a state of active growth and carbon uptake, or if they are in a state of cambial inactivity due to drought stress (Steppe et al., 2015). However, there is still limited knowledge regarding the seasonal occurrence of positive or negative radial diameter or circumference variations of tree stems, especially at hourly or daily scale (De Swaef et al., 2015). Furthermore, the possibilities of the interpretation of dendrometer readings considering plant physiology are not fully exploited yet (Zweifel, 2016). These deficiencies are in particular valid for the tropics. Although a number of studies deals with tree growth in the tropics on a seasonal or annual level (e.g., Shimamoto et al., 2016; Wagner et al., 2016; Xu et al., 2016), only few studies analyze stem diameter variations of tropical trees on an intra-daily scale (e.g., Volland-Voigt et al., 2009; Butz et al., 2016; Spannl et al., 2016). To directly monitor the impact of drought on a forest and tree species specific response, it is necessary to have a close look at the individual tree level.

Disentangling the various factors leading to stem circumference changes, such as swelling and shrinkage of phloem, xylem and bark due to water potential and elastic properties of tissues (e.g., Rosner et al., 2009; Robert et al., 2014) and radial growth including cambial division and cell expansion are still a topic of ongoing research (e.g., Chan et al., 2015; Mencuccini et al., 2017) varying for each species. Modeling approaches offer one solution to solve this problem (Hölttä et al., 2010; Zweifel et al., 2014; De Swaef et al., 2015; Schippers et al., 2015; Steppe et al., 2015, 2016; Cocozza et al., 2018), however, complex models require numerous parameters with some of them, e.g., initial turgor phloem turgor pressure or elasticity of non-lignified xylem cells being challenging to acquire under field conditions. A different approach is defining growth empirically, if the diameter of a tree exceeds its previous day's maximum (e.g., Deslauriers et al., 2003; Butz et al., 2016; van der Maaten et al., 2016; Zweifel et al., 2016). Although very promising, this approach has a limited capability considering extreme climatic events and considering drought coping strategies (e.g., succulence). Due to the constraints of the mentioned approaches we focus in the present study on analyzing daily stem diameter responses to drought between different tropical tree species and families of different plant functional types regarding their resilience and ability to cope with dry spells of varying length.

Analyzing the impact of climatic extremes on tree circumference provides important insights into tree functioning and carbon budget. Hence, analyzing responses of stem girth to extreme meteorological events is an important variable in climate-growth-modeling and helps to improve existing models (Siegmund et al., 2016). Stem circumference changes are thus a valuable indicator for analyzing drought legacy effects and carbon cycle balance (Anderegg et al., 2015) and tree mortality risks which rapidly increase under extreme drought conditions (Meir et al., 2015).

In this paper we analyze one of the largest dendrometer datasets existing in the tropics. Our study is separated into three parts. After examining the general response of stem size variations to climate parameters, we first analyzed how the different tree species in a moist tropical mountain forest in South Ecuador respond to different durations of rare dry spells. Second, we analyzed species-specific differences in growth dynamics and drought response over a 600-day period. Third, we compared recovery from drought in two contrasting tree species Tapirira and Vismia and draw inferences about possible consequences for forest composition in view of expected climate change. All add up to the main research question, how different humid mountain rainforest tree species respond to drought stress and if some species are better adapted to a possible higher frequency and increasing duration of dry periods.

# MATERIALS AND METHODS

# Study Site and Local Climate

The study was conducted in a lower montane rainforest (Homeier, 2004) on the eastern declivity of the Cordillera Real in southern Ecuador. The study site is located at the northern slope of the Podocarpus National Park (3◦ 580 S, 79◦ 040W) on an elevation of ca. 2000 m a.s.l. The soils at the study site are very heterogeneous, with humic cambisols dominating on the slopes (Wilcke et al., 2008). The studied forest is located in a biodiversity hotspot and hosts more than 280 tree species, with special importance of the families Melastomataceae, Lauraceae, Rubiaceae, and Euphorbiaceae (Homeier et al., 2004).

We used climate datasets used from several local meteorological stations with a maximal distance to our study site of approximately 500 m. Mean annual air temperature is 15.5◦C (Fries et al., 2009), and average annual rainfall is 2200 mm. Over the year, the region experiences three different seasons characterized by changing wind directions and precipitation amounts: between January and April, humid air masses from SE and NE directions dominate, with an average rainfall of around 800 mm. During May to August, northeasterly winds dominate, bringing about 1000 mm precipitation. Finally, during

September to December, only 400 mm of rainfall are recorded. In this phase the climate phenomenon "Veranillo del Nino" (VdN) may occur, which is caused by a pronounced low pressure system east of the Andes weakening the dominating trade winds over our study area. This leads to reduced cloud cover and higher amounts of solar irradiance at our study site (Emck, 2007; Bendix et al., 2008, Rollenbeck et al., 2011). Due to increased outgoing longwave radiation (OLR) under clear sky during the night, the daily temperature range increases to almost 25K, spanning from 2.4◦C during nighttime to 27.1◦C during daytime.

Longer dry intervals (7–9 days) almost exclusively occur during VdN events, except one drought event during 30.06.2010– 06.07.2010 (**Supplementary Table S1**). In oral lore of local farmers, VdN reoccurs every year in the first week of November, but its timespan has become more irregular in recent years. The preconditions for its occurrence, i.e., a change in the dominant easterly wind direction are also fulfilled in October, December and January (Emck, 2007), and four of the other registered dry spells arose in those months. Other dry spells with four or more days without precipitation occurred in July and August (**Supplementary Table S1**).

#### Dendrometer Measurements and Tree Species

The data for this study were collected during two different study periods. From July 2007 to November 2010, tree growth dynamics and radial stem changes were measured every 30 min with electronic point dendrometers (Type DR, Ecomatic, Germany). From March 2015 to December 2017, logging band dendrometers (LBDs) with a built-in thermometer (DRL26, EMS Brno) were used. For comparability of the two dendrometer types, we converted the data of the point dendrometers into circumference values using the formula: circumference = 2 × π × radius, assuming symmetrical stem geometries of the studied individuals. Due to latest calibration tests, the used dendrometer types show different thermal sensor coefficients of 3.29092 µm◦C −1 for the type DR and 0.0054 µm ◦C −1 for the type DRL26, respectively (von der Crone et al., pers. comm., work under review). Since the maximum temperature differences between consecutive nights in our study area do not exceed 3◦C, no temperature correction was needed for the values derived from band dendrometers. Due to the higher thermal sensor coefficient of 3.29 µm◦C <sup>−</sup><sup>1</sup> of the point dendrometers, their maximum error is the range of up to 62 µm in circumference. An error in this range does not question our main results. Data recorded by dendrometers of DR-type have only been used in the first part of the study. In all cases, dendrometers were installed at breast height (approx. 1.30 m). In case of thick-barked species, parts of the outer bark were removed without injuring the cambial zone to minimize the influence of bark swelling and shrinking due to water uptake or loss. Since active growth by wood formation and stem swelling due to water uptake cannot be differentiated by dendrometer data we will use the term "increment" in the following for any kind of stem diameter increase. To investigate the impact of drought on different species we selected dry intervals with a minimum of four consecutive days without rainfall during the periods July 2007 to November 2010 and March 2015 to December 2017. We analyzed the averaged stem shrinkage rates during periods from 4 to 9 days. To calculate stem shrinkage, only the daily maximum stem diameter values (Deslauriers et al., 2007) were compared. In order to create a more robust dataset, the values regarding the dry spells include the data of all droughts with the minimum amount of days without precipitation.

In total, 40 tree individuals with stem diameters above 10 cm belonging to eleven different tree genera and three tree life forms were studied (**Table 1**). Stem diameter variations of trees of the same genus that belong to the same plant functional type were averaged to increase replication, this applies for the genera of the Lauraceae family, namely Persea, Nectandra, and Ocotea. Hereafter, we use the genus names for the groups of trees comprising more than one species and the species names, if the genus is represented by only one species. For Prumnopitys, our data regarding dry spells is limited to 6 days due to data logger problems.

#### Statistical Analysis

The dendrometer data were scanned for outliers and measurement artifacts (exceeding 0.5 mm stem circumference change within 30 min). For stem shrinkage and increment calculation we took the maximum daily circumference/diameter values and subtracted the values from the subsequent day.


TABLE 1 | Characteristics of studied tree species equipped with electronic dendrometers.

The distinguished groups for the second part of the study (**Figure 2**) differed significantly from each other (tested by Welch ANOVA test). We standardized the data with the formula z = (x-µ) /σ, where x is the measurement value, µ is the arithmetic mean and σ is the standard deviation. For better visual comparison, we divided the outcome by 20. The relationships between changes in stem circumference and climatic variables were tested with linear correlations. All statistical analyses were conducted with the R programming language (R Development Core Team, 2015) thought the software RStudio version 3.2.2.

# RESULTS

## Tree Species Response to Dry Spells of Different Length

Evergreen broadleaved and coniferous species showed a lower rate of stem diameter shrinkage in relation to dry interval length than the deciduous broadleaved species H. chrysanthus and C. montana, with average circumference losses of 1.876 and 2.190 mm after nine dry days, respectively (**Figure 1**). G. emarginata lost least of their circumference, with a maximum loss of only 0.290 mm after nine consecutive dry days. T. guianensis first showed a circumference loss similar to most of the evergreen broadleaved species, but after the fifth rainless day shrinkage rates strongly increased and summed up to 1.528 mm after the ninth rainless day. Also noticeable is the strong circumference loss of V. cavanillesiana from the eighth to the ninth day without rainfall. The two coniferous species showed no differences compared to most of the evergreen broadleaved species.

A closer look to the day-to-day circumference changes (**Figure 2**) reveals that for the two deciduous broadleaved species the circumference loss was constantly increasing with increasing drought duration. T. guianensis showed lower losses from the fifth to sixth and seventh to eighth day compared to the day before and shrunk 0.662 mm on the last day, which was the strongest observed shrinkage of all species. V. cavanillesiana showed a strong loss of circumference of 0.279 mm on the ninth day, which was more than double the amount of any day before. The shrinkage of Inga acreana declined after the seventh day, and also Nectandra and Persea showed maximum shrinkage from the sixth to the seventh rainless day.

To make the values more comparable, we used z-transformation (**Figure 3**). Cedrela montana also showed the highest standardized change but Handroanthus chrysanthus moved closer to the other species. Graffenrieda emarginata which showed the lowest shrinkage in absolute values now blended in with the other data.

#### Species-Specific Response of Stem Size Variations to Climate Variables, Differences of Growth Dynamics and Drought Response

For the comparison of the overall growth behavior of the studied species, we selected a timespan from 12.04.2016 to 01.12.2017 (599 days) in which no missing dendrometer measurement values, which might compromise the results, occurred. The species Handroanthus chrysanthus, Cedrela montana, Inga acreana, Prumnopitys montana and the genera Persea and Nectandra had to be excluded from this comparison because they

deciduous broadleaved (db, lightblue).

were not instrumented during the second measurement period from March 2015 to December 2017.

In **Table 2**, correlations between daily stem circumference changes and different climate parameters are shown. Mean relative humidity and vapor pressure deficit showed the highest correlation values. Temperature parameters and global radiation were also significantly correlated. Precipitation of the current day showed higher correlations with stem size variations than precipitation of the previous day. Soil moisture was measured at a site close to our study trees (ca. one kilometer distance,


TABLE 2 | Correlations between daily circumference changes of different tree species with climate parameters.

rHmean, daily mean relative humidity; P, daily precipitation; P-1, previous day precipitation; T Mean, daily mean temperature; T min, daily minimum temperature; T max, daily maximum temperature; VPD, vapor pressure deficit; gRad, global radiation. <sup>∗</sup>95% significance; ∗∗99% significance; ∗∗∗99.9% significance.

same elevation). Over the year, soil water content measured 10 cm below ground only varied between 35.7 and 48.7% over the year (Moser et al., 2008). Factors dampening soil water content fluctuations are the high amount of annual precipitation, the regular rainfall distribution (apart from the short dry spells), the high clay content of the deeply developed soils, and the dense rainforest canopy preventing solar insulation of the soil surface. Despite these factors, soil water content and soil moisture tension are crucial variables for trees and influence water dependent systems e.g., by buffering drought phases or prolonging the experienced drought past the first rainfalls until the soil are replenished. Because soil moisture data are not available for our study area over the studied time periods, we are not able to quantify this possible error.

In spite of the identical climatic conditions, differences in the percentages of days with increment or shrinkage are apparent. G. emarginata showed the highest percentage of days with stem increment (68.1%) and also the lowest percentage of days with shrinkage (28.5%). These values are followed by V. cavanillesiana (65/29.8%), T. guianensis (62.3/35.8%), Ocotea (58.3/38.7%) and P. oleifolius (55.5/40.3%). The percentages missing to 100% is stagnation of circumference (**Supplementary Figure S1**).

For the comparison of the maximum daily increment and maximum daily shrinkage we averaged the three highest values to lower the influence of outliers. **Table 3** shows a distinction into three groups: Ocotea with equal maximum values of shrinkage and increment, G. emarginata and T. guianensis with the maximum daily shrinkage exceeding the maximum daily increment, and V. cavanillesiana and P. oleifolius with dominant maximum daily increment values. In contrast to the maximum values, Ocotea and G. emarginata are the only genera which showed higher values of mean daily shrinkage than mean daily increment (**Table 3**).

When considering cumulated changes, the species showing the highest increment also showed the highest shrinking rates (**Table 3**). The mean net circumference gain was 6.44 mm for Ocotea, 3.82 mm for G. emarginata, 3.03 mm for P. oleifolius, 8.53 mm for T. guianensis, and 7.76 mm for V. cavanillesiana, respectively. For better visualization of the species differences, the figures respective to **Table 3** can be found in the **Supplementary Figures S2–S4**.

# Species-Specific Recovery From Drought

When comparing species and families during the 60 days without precipitation over the study period of 539 days, all groups showed very similar percentages of days with increment, ranging from 15.8 to 18.4%. Due to this homogenous behavior during dry days, we assumed that an important difference in the response to climate may exist following the dry spells. We therefore analyzed in detail the recovery times of the different species. Two species are standing out because of their contrasting response during and after drought, namely T. guianensis and V. cavanillesiana.

Both species showed a very consistent response pattern among different individuals (**Figure 4**). Their responses considering the VdN drought lasting from 13 to 23 November 2016 were at first very similar, starting with a steady circumference loss which was more pronounced in T. guianensis. However, very different responses occured concerning the replenishing phase after the first rainfall on the 23 November 2016. While the circumference of all V. cavanillesiana individuals rose immediately, the individuals of T. guianensis showed a much slower pattern of stem replenishment. The most distinct differenence beween the two species was the time needed to reach the stem diameter preceding the VdN phase. On average, V. cavanillesiana needed only 4 days to regain its pre-drought circumference. In contrast, T. guianensis regained only 50% of its



I\_max, Average daily maximum increment; S\_max, Average daily maximum shrinkage; I\_me, Average daily mean increment; S\_me, Average daily mean shrinkage; I\_sum, Average cumulated increment; S\_sum, Average cumulated shrinkage; SD, standard deviation.

losses within 4 days, and it took until early- January (45 days) until the pre-drought stem diameter was regained, but already at the beginning of December the species showed a flattening of the stem increment curve (F).

# DISCUSSION

The rates of stem circumference shrinkage in relation to the dry time interval, the net circumference change between subsequent days and the standardized shrinkages (**Figures 1–3**) highlighted that evergreen broadleaved and coniferous species are able to regulate stem water loss after a couple of sunny and dry days, probably by the closure of leaf stomata (Ying et al., 2015). If the duration of a dry spell exceeds four consecutive days, differences between families became apparent. G. emarginata appeared very well adapted, showing the smallest absolute stem diameter shrinkage but standardized, it turned more to average. P. oleifolius showed the second smallest stem diameter change in response to ongoing drought, very similar to Ocotea and Inga. V. cavanillesiana responded very alike, but the daily changes of the last three dry days indicated differences in response, probably due to change in droughtcoping adaptations like different degrees of leaf-water potential regulation (Ryan, 2011). The decreasing circumference loss between the seventh to the eighth dry day and the much higher loss on the ninth day could be an indication that a threshold of water loss was reached. The same assumption applies to T. guianensis, showing the maximum change on a single day of all investigated species.

The deciduous broadleaved species H. chrysanthus and C. montana share the ability of leaf shedding to regulate the risk of water loss by enhanced transpiration. The latter species showed during the timespans in which most of the dry spells occur around 50% foliage (Bendix et al., 2006), but both species showed the highest values of circumference loss from the very beginning of a dry spell. Even after 9 days without precipitation, stem circumferences constantly decreased with still increasing rates. This led us to speculate that if dry intervals occur during the growing period, when both deciduous species are fully foliated, these species may suffer even more from drought. As visualized by the standardized values C. montana is more sensitive and vulnerable to droughts compared to H. chrysanthus.

The significant correlations of precipitation, relative humidity and vapor pressure deficit with our stem circumference measurements are consistent with previous studies (e.g., Deslauriers et al., 2007; Bräuning et al., 2009; Volland-Voigt et al., 2009; Butt et al., 2014). The majority of the climatic parameters are highly inter-correlated, like e.g., global radiation and temperature.

Consistent with the small absolute response during the dry spells, G. emarginata showed the highest percentage of days with stem increment. However, when shrinkage occurred, the maximum and the mean shrinkage values exceeded the respective increment value. This seems not

problematic due to the high percentage of days with net increment, and hence we do not classify the species as vulnerable. G. emarginata is also the second slowest growing species with the lowest cumulated increment but also with the lowest cumulated shrinkage. Slow growth may be related to drought tolerance as resource-conservative strategy (Ouédraogo et al., 2013).

P. oleifolius showed the lowest percentage of days with increment, but no peculiarities were found in its responses during drought. The increment maxima and mean values were both higher than their shrinkage values. The mean net growth during the analyzed timespan was the lowest with 3.03 mm. Hence, Podocarpus is demanding concerning climate conditions, but this is compensated by higher mean and maximum increment values during suitable climate conditions.

Ocotea is also unobtrusive regarding its dry spell responses and show an average percentage of days with increment compared to the other species. Maximum shrinkage values are equal to the increment maxima. The daily mean increment is lower than the mean shrinkage. This could be, in contrast to G. emarginata, potentially problematic for the family because of the lower percentage of days with increment.

V. cavanillesiana showed some distinctive features during drought, i.e., the percentage of days with shrinkage is only 29.8%. The values of mean and maximum increment are higher than the respective shrinkage. With 7.76 mm, the species' net growth is the second highest of the analyzed species. T. guianensis shows the same features as V. cavanillesiana, but more intense. Overall, T. guianensis showed the highest maximum shrinkage values of all investigated species, but also net growth is highest, with 8.53 mm.

The flattening of the growth curve for T. guianensis indicates a switch from stem water replenishment into a usual growth pattern (**Figure 4**). The lower absolute stem circumference may be explained by embolism of xylem cells which reduces on the one hand the ability of expansion of the stem by replenishment of xylem cells with water and hence build-up of turgor, and on the other hand causes hydraulic dysfunction lowering the conductance of the stem. Recovery from a drought can be interfered by xylem cavitation (Luo et al., 2016). Other drought stress effects that may negatively affect post-drought growth despite favourable growth conditions include damaged organelle structures, decreased photosynthetic activity or induced chlorophyll degradation (Ying et al., 2015). T. guianensis seems therefore to be vulnerable to longer or higher frequency droughts. In our study, we determined similar responses between T. guianensis individuals but it is important to take into consideration that embolism resistance may vary intraspecifically (Anderegg, 2015). To sum up our findings and to answer the main research question, we can state that there is a difference in drought response between deciduous and evergreen trees regarding the absolute amount of circumference loss. However, when considering standardized values of stem shrinkage, the deciduous C. montana shows high drought sensitivity and vulnerability. Also Tapirira becomes more vulnerable with increasing duration of dry spells. Regarding the species' responses during the 600 days drought and non-drought comparison, we consider Ocotea and again Tapirira as most vulnerable to drought. Our last analysis verified Tapirira being highly vulnerable to longer lasting droughts, resulting in a non-replenishable loss of stem circumference.

To make inferences about possible effects of the detected differences in drought response about possible species composition changes of the entire ecosystem or about speciesspecific mortality risk, a key research need is determination of species-specific thresholds of xylem embolism which are recoverable or not (Hartmann et al., 2018). Understanding the mechanisms of drought response, survival and mortality will be critical for predicting tree response to a changing climate (Ryan, 2011). This needs to be taken into consideration also regarding the carbon sink or source issue because special patterns of moist tropical forest carbon storage are primarily driven by mortality (McDowell et al., 2018).

Of interest for future research is if nonlethal droughts can stimulate drought resilience, since plants typically respond by acclimation of key hydraulic parameters like leaf area (reduction), root and sapwood area (increase), and cavitation resistance (increase), which should protect trees against future droughts (McDowell et al., 2008). Considering that the intact forest sink is declining in size, and that tropical forests may switch to become a net carbon source (Mitchard, 2018), more research in different tropical forest ecosystems is of utmost importance.

#### DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

# AUTHOR CONTRIBUTIONS

AB and VR developed the concept of the manuscript. VR and KP collected the data in the field. VR and SS contributed statistics and graphics. All authors contributed to the text writing.

# FUNDING

The financial support granted by the German Science Foundation (DFG) through PAK823-825 "Platform for Biodiversity and Ecosystem Monitoring and Research in South Ecuador" (BR1895/23-1) is gratefully acknowledged.

# ACKNOWLEDGMENTS

We thank the Ecuadorian Ministry of the Environment (MAE) for the research permission and Nature and Culture International (NCI) for logistic support.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.00342/ full#supplementary-material

FIGURE S1 | Percentage of different circumference changes (%) for different genera.

#### REFERENCES


FIGURE S2 | Daily increment and shrinkage maxima (mm) for different genera with standard deviations.

FIGURE S3 | Daily increment and shrinkage means (mm) for different genera with standard deviations.

FIGURE S4 | Daily increment and shrinkage sums (mm) for different genera with standard deviations.

TABLE S1 | Number of dry intervals and the dates they occurred.



**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 Raffelsbauer, Spannl, Peña, Pucha-Cofrep, Steppe and Bräuning. 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.

# One Century of Forest Monitoring Data in Switzerland Reveals Speciesand Site-Specific Trends of Climate-Induced Tree Mortality

Sophia Etzold<sup>1</sup> \*, Kasia Zieminska ´ 1 , Brigitte Rohner<sup>1</sup> , Alessandra Bottero1,2 , Arun K. Bose<sup>1</sup> , Nadine K. Ruehr<sup>3</sup> , Andreas Zingg<sup>1</sup> and Andreas Rigling1,4

<sup>1</sup> Swiss Federal Institute for Forest, Snow and Landscape Research WSL, Birmensdorf, Switzerland, <sup>2</sup> SwissForestLab, Birmensdorf, Switzerland, <sup>3</sup> Institute of Meteorology and Climate Research – Atmospheric Environmental Research, Karlsruhe Institute of Technology, Garmisch-Partenkirchen, Germany, <sup>4</sup> Institute of Terrestrial Ecosystems, ETH Zurich, Zurich, Switzerland

#### Edited by:

Sebastian Leuzinger, Auckland University of Technology, New Zealand

#### Reviewed by:

Martina Pollastrini, Università degli Studi di Firenze, Italy Cyrille Barthélémy Karl Rathgeber, INRA Centre Nancy-Lorraine, France Henrik Hartmann, Max-Planck-Institut für Biogeochemie, Germany

> \*Correspondence: Sophia Etzold sophia.etzold@wsl.ch

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 12 November 2018 Accepted: 26 February 2019 Published: 22 March 2019

#### Citation:

Etzold S, Zieminska K, Rohner B, ´ Bottero A, Bose AK, Ruehr NK, Zingg A and Rigling A (2019) One Century of Forest Monitoring Data in Switzerland Reveals Speciesand Site-Specific Trends of Climate-Induced Tree Mortality. Front. Plant Sci. 10:307. doi: 10.3389/fpls.2019.00307 Climate-induced tree mortality became a global phenomenon during the last century and it is expected to increase in many regions in the future along with a further increase in the frequency of drought and heat events. However, tree mortality at the ecosystem level remains challenging to quantify since long-term, tree-individual, reliable observations are scarce. Here, we present a unique data set of monitoring records from 276 permanent plots located in 95 forest stands across Switzerland, which include five major European tree species (Norway spruce, Scots pine, silver fir, European beech, and sessile and common oak) and cover a time span of over one century (1898–2013), with inventory periods of 5–10 years. The long-term average annual mortality rate of the investigated forest stands was 1.5%. In general, species-specific annual mortality rates did not consistently increase over the last decades, except for Scots pine forests at lower altitudes, which exhibited a clear increase of mortality since the 1960s. Temporal trends of tree mortality varied also depending on diameter at breast height (DBH), with large trees generally experiencing an increase in mortality, while mortality of small trees tended to decrease. Normalized mortality rates were remarkably similar between species and a modest, but a consistent and steady increasing trend was apparent throughout the study period. Mixed effects models revealed that gradually changing stand parameters (stand basal area and stand age) had the strongest impact on mortality rates, modulated by climate, which had increasing importance during the last decades. Hereby, recent climatic changes had highly variable effects on tree mortality rates, depending on the species in combination with abiotic and biotic stand and site conditions. This suggests that forest species composition and species ranges may change under future climate conditions. Our data set highlights the complexity of forest dynamical processes such as long-term, gradual changes of forest structure, demography and species composition, which together with climate determine mortality rates.

Keywords: drought, competition, stand basal area, climate change, tree size, mortality

# INTRODUCTION

fpls-10-00307 March 21, 2019 Time: 11:16 # 2

Forest ecosystems play a crucial role in maintaining the balance of land-atmosphere cycles, sequestering carbon, fostering biodiversity and providing esthetic value to humanity (Bonan, 2008; Pan et al., 2011; Millar and Stephenson, 2015). Hence, understanding forest mortality is a pressing task of our times in order to predict and regulate the future development of forests. Although forest mortality is a natural process within forest stand dynamics (Franklin et al., 1987), mortality rates have been reported to increase in recent years due to climatic changes (Allen et al., 2010; McDowell et al., 2018), and are predicted to increase further in the future (Steinkamp and Hickler, 2015).

Drivers of tree mortality are diverse and intertwined with one another. The decline-disease theory is a conceptual framework relating tree mortality to a sequence of abiotic and biotic factors, which can be of predisposing, inciting or contributing nature (Manion, 1981; Houston, 1984). Predisposing factors (e.g., age or air pollutants) devitalize a tree over a long time period. Inciting factors, such as drought or insect defoliation, are short-term events that considerably reduce tree vigor. Finally, the occurrence of contributing factors (e.g., additional drought events, fungi) determines the fate of the weakened tree. Contributing factors can act in a long term, but can also cause a fast die-off when a certain threshold is reached (e.g., bark beetle infestations). Forest stand characteristics play an important role in forest mortality. For example, higher stand basal area and subsequent competition have been shown to increase mortality rates (Bradford and Bell, 2017; Young et al., 2017). Mortality risk also depends on tree or stand age (van Mantgem et al., 2009; Cailleret et al., 2017; Neumann et al., 2017) with usually a U-shaped curve of mortality probability in relation to age and/or tree size (Coomes and Allen, 2007; Lines et al., 2010). In recent years, changing climatic conditions, such as increasing drought intensity and frequency as well as rising temperatures, have been found to strongly influence forest mortality rates (van Mantgem et al., 2009; Allen et al., 2010; Anderegg et al., 2013). Disturbances that are often a consequence of changing climate, such as more numerous and severe fires (Brando et al., 2014; Stephens et al., 2018) or damaging insect outbreaks associated with storms and droughts (Weed et al., 2013; Kolb et al., 2016; Wood et al., 2018) have also been linked to increased forest mortality. Given this diversity of factors, mortality rates vary depending on the geographical location, its climate and consequent vegetation structure. Nevertheless, an increase in forest mortality due to climatic changes has been observed across continents, under different climatic conditions and across diverse forest types, from gymnosperm to angiosperm dominated ones, and from tropical to boreal regions (Allen et al., 2010; Anderegg et al., 2015; McDowell et al., 2018).

Drought-induced tree mortality can develop after substantial long periods of suboptimal water supply. The underlying physiological mechanisms include substantial damages to the tree hydraulic system through air entering the water conducting tissues of trees (Tyree and Sperry, 1989) under high xylem tensions induced by low soil moisture and high evaporative demand during drought. If these so called embolisms are substantial, then trees are not able to recover (Hartmann et al., 2018) following drought release and may subsequently die (Anderegg et al., 2016; Adam et al., 2017). Trees have developed multiple physiological strategies, at root, stem and leaf level, which allow them to cope with drought stress and to avoid hydraulic failure to a certain degree. For example, they can grow deep roots for better access to water, build xylem that is robust to embolism and/or control water loss via stomatal closure or leaf shedding (Martin-StPaul et al., 2017; Choat et al., 2018).

Growing recognition of the impacts of extreme droughts on forest ecosystems in Europe has spurred scientific attention on how major tree species including Scots pine (Pinus sylvestris L.), Norway spruce [Picea abies (L.) H. Karst], silver fir (Abies alba Mill.), European beech (Fagus sylvatica L.) and oak [Quercus petraea (Matt.) Liebl., Quercus robur L.] will cope with the expected changes in climate. These species follow different physiological strategies to cope with drought stress and exhibit different levels of drought tolerance (Breda et al., 2006). Among these species, Scots pine usually occurs on the driest sites used for commercial forestry in Central Europe. Compared to Norway spruce and silver fir, Scots pine is considered to be more drought tolerant. This is presumably due to its early stomatal closure to avoid xylem embolism under mild-moderate drought (Martinez-Vilalta et al., 2004) in combination with a deep rooting system (Richardson, 2000). Despite these strategies, Scots pine may not be well adapted to the combined effects of drought and heat (Dobbertin et al., 2005; Giuggiola et al., 2010) or to several consecutive drought events (Bigler et al., 2006) and is further vulnerable to diverse drought-related pests and diseases (Wermelinger et al., 2008). Norway spruce has repeatedly been described as particularly susceptible to drought (Zang et al., 2012, 2014; Pretzsch et al., 2013), which is likely due to its shallow root system and its vulnerability to drought-related insect outbreaks. There is an agreement that Norway spruce is already negatively affected by the impacts of climate change, which will be more pronounced in the future (Hanewinkel et al., 2013; Zang et al., 2014). Silver fir, however, has been shown to be more resilient to drought stress and associated phenomena like bark beetle outbreak or storm damages (Zang et al., 2014; George et al., 2015; Vitali et al., 2017). European beech is another drought sensitive species (Leuschner et al., 2001; Gessler et al., 2007; Charru et al., 2010). Several studies show a strong decline in radial growth of beech trees in response to decreasing water availability particularly for the Mediterranean region (Jump et al., 2006; Piovesan et al., 2008). However, beech has also shown a high physiological plasticity and adaptability to changing growing conditions (Vitasse et al., 2014; Cocozza et al., 2016; Stojnic et al., 2018), and the acclimation of beech to future climatic changes is highly uncertain. Oaks are generally considered to be well adapted to drought, due to their high resistance to xylem embolism (Robert et al., 2017; Lobo et al., 2018), their high capacity to withdraw water from the soil also under dry conditions by deep root systems and their ability to maintain water potential gradients along the soil-stem-crown continuum (Zweifel et al., 2007), as well as their xeromorphic leaf structure. Moreover, they are able to rapidly resume assimilation after periods of water deficiency (Kubiske and Abrams, 1993; Kuster et al., 2013).

Across Europe, a considerable number of studies found increasing mortality rates due to drought and heatwave events in Spain (Peñuelas et al., 2001; Martínez-Vilalta and Piñol, 2002), France (Bréda et al., 2006; Vennetier et al., 2007), Switzerland (Bigler et al., 2006), Poland (Siwecki and Ufnalski, 1998), Norway (Solberg et al., 2004), and Russia (Kauhanen et al., 2008; Ogibin and Demidova, 2009). However, opposite trends in mortality rates have also been reported. When comparing the period of 1900–1960 with 1960–2000, mortality rates of Norway spruce in Germany did not change, while for beech, mortality rates decreased by 17% (Pretzsch et al., 2014b). Similarly, annual mortality rates of four angiosperm species in Sweden were also found to be smaller in recent decades (1988–2013) than earlier (1912–1988, Hytteborn et al., 2017). Our current understanding on forest mortality trends is strongly influenced by the time span of available data. A long-term perspective is necessary to obtain a comprehensive picture of forest dynamics, especially when dealing with long-living organisms, such as trees.

It is also important to make a distinction between mortality rates of individual species and mortality rates of entire forest ecosystems. Studies investigating physiological mechanisms driving mortality usually focus on species-specific mortality, and often cover a relatively short time span, limited sample size and geographical range (e.g., Chao et al., 2008; Mitchell et al., 2013). In contrast, ecosystem-scale mortality rates are typically assessed across multi-year or multi-decadal time span, covering a large sample size and geographical range (e.g., Williams et al., 2013; Neumann et al., 2017; Huelsmann et al., 2018). The two perspectives complement each other and are essential for improving our current understanding and future predictions of tree and forest mortality across the globe.

The majority of ecosystem-scale studies assess mortality rates across one or two decades (e.g., Monserud and Sterba, 1999; Bradford and Bell, 2017; Pillet et al., 2018; Rogers et al., 2018) or after a specific disturbance event (e.g., Breshears et al., 2005; Michaelian et al., 2011; Brodrick and Asner, 2017; Reed et al., 2018). However, long-term mortality trends can add a unique perspective and significantly improve our understanding of forest dynamics. Long-term monitoring requires extensive effort, resources and commitment; hence, the data are extremely scarce. We are aware of only three studies reporting tree mortality rates of seven European species and spanning a time frame of at least 100 years (Pretzsch et al., 2014a,b; Hytteborn et al., 2017). An additional challenge is to assess mortality across broad climatic gradients and across a more diverse suite of species. For example, a recent geographically comprehensive study encompassed entire Europe, but included only two species and covered one decade (Neumann et al., 2017).

In this study, we address these knowledge gaps by compiling a dataset spanning ca. 120 years of inventory data for five dominant European tree species in Switzerland: Scots pine, Norway spruce, silver fir, European beech, sessile and common oak. Despite the fact that our study is geographically restricted to Switzerland, the large heterogeneity of the Swiss landscape enables an assessment of forest mortality under a wide range of environmental conditions. For example, the 276 sampled plots spanned a wide altitudinal (∼1800 m) and precipitation gradient (∼800 mm during growth season). We combined longterm inventory data of forest mortality, stand characteristics such as basal area and mean stand diameter at breast height (used as proxy for stand age), and high-resolution climate data to: (i) assess annual tree mortality rates and temporal trends of five dominant tree species across Switzerland over the last century, and (ii) identify the main drivers of mortality, out of a set of 12 predictors, including climate, stand characteristics and topography. We hereby focused on the impact of summer drought conditions on mortality, being aware that also other climatic factors, such as winter conditions, might have an impact on forest mortality at certain sites. We hypothesized that drought-induced mortality is species-specific with high droughtinduced mortality rates in spruce and beech and lowest mortality rates in more drought-resistant oak and pine. Within species we expected different responses to drought along environmental gradients, due to different limiting factors that depend on site conditions (e.g., temperature vs. water availability). For instance, pine or spruce growing at dry (lowland) sites and showing already climate related declines should exhibit higher drought induced mortality rates compared to wet or high-altitude sites. On the other hand, it has been shown that, e.g., beech trees that are adapted to a generally lower water availability are more resistant to occurring droughts compared to beech trees growing under generally good water supply (Grossiord et al., 2014; Kunz et al., 2018). Therefore, we would expect a higher drought-induced mortality on wet sites compared to drier sites for species with a high adaptive capacity, e.g., beech and oak.

#### MATERIALS AND METHODS

#### Forest Inventory Data

The study includes data from three long-term data sources of Swiss forest growth monitoring networks (**Supplementary Table S1**): Swiss Experimental Forest Management plots network (EFM), Swiss Long-Term Forest Ecosystem monitoring network (LWF), and the Swiss Nature Reserves Network (SNR). Within all three networks, the diameter at breast height (DBH, 1.3 m height) and the status (dead/alive) of all marked trees larger than a network-specific DBH threshold were recorded during repeated inventory campaigns (**Supplementary Table S1**). For consistency, only trees with DBH > 5 cm were included in this study.

The EFM plots were established between 1887 and the early 1900s, and currently comprise ca. 390 yield plots across Switzerland in order to study the development of forests in a changing environment under the influence of forest management (cf., Schütz and Zingg, 2010). Inventory intervals ranged from 3 to 22 years (mean 7 years). The LWF plots were established in 1995 and are part of the intensively monitored Level II plot network of the International Co-operative Programme on Assessment and Monitoring of Air Pollution Effects on Forests (ICP Forests) in Europe (Ferretti et al., 2010). The LWF plots are assessed every 3–6 years (mean 5 years) since 1995 according to a harmonized sampling protocol (Dobbertin and Neumann, 2010; Hug et al., 2011). The SNR network currently consists of

49 unmanaged forest reserves throughout Switzerland (Brang et al., 2006, 2011; Wunder et al., 2007). Repeated inventory campaigns of core plots within the nature reserves have been conducted every 5–20 years (mean 12 years) since the late 1940s. It needs to be noted that the permanent plots of the three monitoring networks were assessed at different intervals (3–22 years, on average 9 years), for different inventory years and time periods, and that the number of plots increased with time.

#### Study Sites

A total of 276 permanent plots within 95 forest sites (**Supplementary Table S1** and **Figure 1**) were selected according to the following criteria: (1) at least two inventories per plot were available; (2) the plots were not affected – as far as documented – by major disturbances or management interventions; for the EFM, only plots with natural thinning (A grade according to IUFRO rules, Versuchsanstalten, 1902) or plots with low management intensity, where suppressed, fallen or dying stems are removed from below but without impact on crown layer (B grade), were included, so that the measured mortality rates should reflect as far as possible the natural mortality rates; (3) plots consisted of at least 20% of the basal area of one of the following species: Scots pine (Pinus sylvestris L.), Norway spruce [Picea abies (L.) H. Karst], silver fir (Abies alba Mill.), European beech (Fagus sylvatica L.) or oak [Quercus petraea (Matt.) Liebl., Quercus robur L.]. The two oak species were treated as one entity and are referred throughout the text as 'oak', due to their morphological similarities and tendency to hybridize, leading to their disputed taxonomical identity (Kissling, 1980; Aas, 1998; Muir et al., 2000).

The plots within the SNR and the EFM forest sites were considered as separate plots, despite their geographical proximity (on average 100 m apart, ranging from 3 to 423 m), as their stand structure and abiotic characteristics were very heterogeneous

FIGURE 1 | Location of plots from EFM (red symbols), LWF (green symbols) and SNR (blue symbols) networks included in this study. In a number of locations, several plots are located in close proximity to each other. See Supplementary Material for maps showing locations of each species separately (Supplementary Figure S1).

(cf. Brang et al., 2011). The individual plots covered an area of 0.02–3.47 ha (mean 0.5 ha). Out of the total of 276 plots, 175 were pure stands (basal area of a given species > 70%) and 101 mixed stands. Among the pure stands, 47 were dominated by pine, 23 by spruce, 10 by fir, 42 by beech, 50 by oak, and 3 by other species (for these plots, species with remaining 20–30% basal area were included in analyses). According to annual changes in tree numbers and basal area as described in Brang et al. (2011), 40 stands were classified to the juvenile forest phase, 17 stands to the young forest phase, 178 to the optimal phase and 41 to the old growth phase. The mean stand age of the 100 forest stands with age information was 130 years (ranging from 45 to 360 years). In total, the data series used in this study consisted of 2–17 inventories per plot (on average 9) from 1898 to 2013. Average data series length was 45 years and the longest series comprised 113 years from 1898 to 2011. In total, data of 247,529 trees were included.

#### Climate Data

All site-specific climate data (temperature, precipitation, drought stress indicators) for the time period from 1901 to 2012 were obtained with a daily resolution from Remund et al. (2016). Climate data from 1901 to 1930 were interpolated from CRU grid data "CRU TS Version 1.2" (Mitchell et al., 2004) based on the change factor method, and since 1931 from observational plots (MeteoSchweiz) based on the Shepard's Gravity interpolation method (Zelenka et al., 1992).

For all study plots within one forest site identical climate data were used. All climate variables were calculated for the growth period from May to September. The minimal site water balance (SWBmin) during the growth period was used as drought indicator as it was found to be one of the best predictors of mortality in a comparison of various multivariate models for beech and spruce in Switzerland (Braun, 2016). In addition, SWBmin resulted in the best agreement with soil water content measured in 10–20 cm soil depth for a subset of LWF sites used in this study, relative to other drought indices (R = 0.48, data not shown). SWBmin was calculated according to Grier and Running (1977) as sum of the daily differences between precipitation sum and potential evapotranspiration with field capacity as starting value. The detailed methodology is described here: http://www.wsl.ch/staff/ niklaus.zimmermann/programs/amls/swb.aml. Lower SWBmin, denotes higher drought stress. In this study, the minimal value for the growth season was applied.

## Ecoregions

Within each species, plots were categorized into two distinct groups, referred to as "ecoregions" (**Table 1** and **Supplementary Figure S1**). The grouping was intended to emphasize the diverse mortality patterns in forests growing under contrasting climate conditions. Recent studies have shown that temperature and drought have divergent effects on high elevation vs. low elevation forests, as well as on temperature-limited vs. water-limited forests (Jolly et al., 2005; Lindner et al., 2010; Sarris et al., 2011). For Scots pine in the inner-Alps in Switzerland, mortality was found to be highest below 1000 m a.s.l., which was also related to drought

TABLE 1 | Stand characteristics for 1960–2012 period summarizing mean values and the 0.05–0.95% quantile range (in parentheses) for plots within a given species-ecoregion.


SWBmin, minimum Site Water Balance during the growth season; Temp, mean temperature during the growth season; Precip, mean total precipitation during the growth season; BA, basal area; mDBH, stand mean DBH as indicator of stand age; Mortality, annual mortality rate. Letters ('a', 'b') indicate differences in climate and stand characteristics between ecoregions per species, tested by an unpaired t-test. Significant coefficients are indicated as follows ∗∗∗P < 0.001, ∗∗P < 0.01, <sup>∗</sup>P < 0.05.

(Rigling et al., 2013). We divided plots so that both ecoregions per species contained approximately equal number of plots. The resulting ecoregions encompassed low and high-altitude plots for pine (threshold at 1000 m) and spruce (1300 m). The grouping of pine and spruce by elevation resulted also in significantly different climate conditions with warmer temperatures in the lowlands, and for spruce also drier conditions, compared to the highlands (**Table 1**). For beech, oak and fir, which grow predominantly at lower elevations, we distinguished dry and wet plots based on the SWBmin (beech, fir SWBmin = 100 mm, oak SWBmin = 50 mm).

#### Mortality Rates

Based on the individual tree data, annual mortality rates m were calculated at a population level per inventory period (interval) as follows:

$$m = \left(1 - \left(\left(\frac{N\_t}{N\_0}\right)^{\frac{1}{\bar{\tau}}}\right)\right) \times 100\tag{1}$$

where N<sup>0</sup> and N<sup>t</sup> are the numbers of living trees at the beginning and end of the interval, respectively, and t is the inventory interval in years (Sheil and May, 1996). Mortality rates m were only calculated for populations with N<sup>0</sup> > 10. In populations with different mortality probabilities, mortality rates decline with the length of the inventory interval, because the fraction of trees with a higher mortality probability declines faster than the fraction of trees with a low mortality probability. Therefore, mortality rates calculated from diverse interval lengths were compared for subpopulations for which homogeneous mortality probabilities may be assumed (Sheil and May, 1996). To account for the different interval length in our data set, we followed the approach adopted by Rohner et al. (2012) and calculated mortality rates for subpopulations, i.e., for each tree species and for three DBH-classes (small, medium, large) separately, assuming homogeneous mortality probability for these subpopulations. DBH-classes for each plot and each time period were defined so that all three DBH-classes contained the same number of trees.

To allow for comparison among plots, we standardized the mortality rate m per plot and inventory period by dividing it by the mean mortality rate of the plot during the total observation period:

$$rm\_i = \frac{m\_i}{\text{mean}(m)}\tag{2}$$

where rm is the normalized mortality (%), m is the mortality rate as in Equation (1) and <sup>i</sup> is the inventory period.

#### Statistical Analyses

We distinguished two time periods: the full time period covered by the inventory data set of 1898 to 2013, and a shorter time period including only the recent decades since 1960. This was done because (i) many plots (especially within the LWF and SNR networks) were established later than 1960, and (ii) climate in Switzerland started to change around 1960 (**Figure 2**). Hereby, inventory data were available for the period 1898–2013 and climate data for the period 1901–2012. Therefore, for analyses based only on inventory data, we considered the whole length of the inventory time series, whereas for analyses including also climate data we used inventory data only for 1901–2012 to match with the time-period of climate data availability.

To assess the temporal trend of mortality, generalized linear mixed effects models (GLMM) for proportions with annual mortality rate as response variable were used, i.e., logistic regression models assuming binomial distribution (Zuur et al., 2009). Random effects were defined for the intercept with study plots nested in forest sites (see "Study sites" section) as grouping factor. We used two different model approaches: (1) We fitted one model per species and ecoregion with year at the beginning of the inventory period and DBH-class as fixed effects (Equation 3). (2) In order to detect whether mortality within the three DBH-classes followed different (e.g., counteracting) trends, we fitted additional GLMMs for the three DBH-classes separately (Equation 4).

$$\mathbf{Y}\_{\rm ijk} = \alpha + \beta\_{\rm l} \times \mathbf{Y}\_{\rm ijk} + \beta\_{\rm 2} \times \mathbf{DBH} \text{-class}\_{\rm ijk}$$

$$+ \mathbf{a}\_{\rm j} + \mathbf{a}\_{\rm j|k} + \varepsilon\_{\rm ijk} \tag{3}$$

Where Yijk is the annual mortality rate per observation i, study site j and study plot k, α is the intercept, Year is the start

smoother was added.

year of inventory, and DBH-class is a factor with three levels (small, medium, large). Fixed effects parameters are given as β, with β<sup>1</sup> indicating the temporal trend of the mortality time series and β<sup>2</sup> indicating the additive effect of DBH-class. a<sup>j</sup> and aj|<sup>k</sup> are site and plot random effects, assuming normal distribution with mean = 0 and σ<sup>j</sup> 2 and σj|<sup>k</sup> 2 , respectively.

$$\mathbf{Y}\_{\rm ijk,\ DBH\text{-class}} = \alpha + \beta\_1 \times \mathbf{Year}\_{\rm ijk} + \mathbf{a}\_{\rm j} + \mathbf{a}\_{\rm j|k} + \varepsilon\_{\rm ijk} \tag{4}$$

where Yijk,DBH-class is the annual mortality rate per observation i, study site j and study plot k of a given DBH-class (small, medium or large). All other parameters are the same as described for Equation 3.

Multivariate relationships between explanatory variables and temporal changes in mortality rates were analyzed using GLMMs as described for Equations 3 and 4, but considering a pool of 8 initial predictors (stand characteristics, plot properties, and climate variables) and 4 interactions as fixed effects (Equation 5). All possible combinations of models were calculated per species and ecoregion. To avoid oversaturation of a model, a maximum of 5 (fir) and 7 (other species) explanatory variables were included simultaneously (resulting in 310 and 560 potential models, respectively). Preliminary analysis indicated that an additional error structure to account for plot spatial autocorrelation did not improve model performance and was not incorporated into the final model.

$$Y\_{\rm ijk} = \alpha + \beta\_1 \times \varkappa\_{1\rm ijk} + \beta\_2 \times \varkappa\_{2\rm ijk} + \dots + \beta\_n \times \varkappa\_{n\rm ijk} + a\_{\rm j}$$

$$+ a\_{\rm j|k} + \varepsilon\_{\rm ijk} \tag{5}$$

where x1...x<sup>n</sup> are the included explanatory variables as described below. All other parameters are the same as described for Equation 3.

We considered the following initial predictors: Basal area (BA, m<sup>2</sup> ha−<sup>1</sup> ) was recorded at the beginning of each inventory period and indicated the competitive state in the stand. As stand age was not available for each plot, we used the mean stand DBH as a proxy for stand age (mDBH, cm). Plot-wise regressions of mDBH against stand age for 100 plots with available stand age information resulted in a median adj. R <sup>2</sup> of 0.96. Further, we included the DBH-classes and topographic properties, such as altitude (m), slope (degree) and aspect (as factor with four levels) as fixed effects. Mean temperature (Temp, ◦C) and SWBmin (mm) as drought indicator during the growth period were calculated for all inventory periods and then related to the long-term average (1Temp, 1SWBmin). All numerical variables (BA, mDBH, slope,

altitude, 1Temp, 1SWBmin) were standardized. We tested for interactions between BA and/or mDBH and the climate variables (1Temp and 1SWBmin). Only plots with at least two inventory periods available were included in the GLMMs (n = 224). To account for collinearity of explanatory variables only models with a maximum correlation coefficient between explanatory variables smaller than 0.7 were considered. Models were ranked according to their corrected AIC, AICc (Hurvich and Tsay, 1989). In case of several comparable models (Akaike weight of the best model < 0.9) an average model of all models with a 1AICc (calculated from the model with lowest AICc) < 4 was calculated (Burnham and Anderson, 2002). Model residuals were visually checked for heterogeneity vs. fitted values and included variables.

All statistical analyses were performed in R statistical software (R Development Core Team, 2011) with the function glmer (package 'lme4') for GLMMs and dredge and model.avg (package 'MuMin') for model selection and model averaging, respectively.

#### RESULTS

#### Overview of Annual Mortality Rates

Average annual mortality rate for all species across the ∼120 years was 1.5%, and a considerable variation in mortality rates across species and ecoregions (dry and wet, high and low altitudes) was observed (**Table 1** and **Figure 3**). Oak and fir had highest mean annual mortality rates of 2 and 1.8%, respectively, pine and beech had 1.3%, and spruce had the lowest mortality rate of 1% (**Table 2**). However, mortality rates of individual plots ranged from 0% to over 13%, with majority of plots (95%) averaging below 6%.

#### Driving Factors of Annual Mortality Rates

Out of an initial set of 8 predictor variables and 4 interactions, GLMMs were calculated for each species and ecoregion to identify the drivers of temporal variation in tree mortality (**Table 3**), which are presented in detail in the following.

#### Stand Properties

Basal area (BA) was the most important and consistent predictor of annual mortality rates for all species and ecoregions. Increasing annual mortality rates were correlated with higher BA (**Table 3** and **Figure 4**). The strongest effect was observed for fir and oak growing on dry sites and beech from wet sites. BA was not significant for only two species-ecoregions, pine at low altitudes and spruce at high altitudes. In the latter case, however, interaction of BA with climate variables was significant (**Table 3**).

Mean DBH (mDBH, proxy for stand age) had a significant effect in six out of the ten species-ecoregion combinations and both negative and positive relationships were observed (**Table 3**). Increased mortality rates were associated with higher mDBH in pine from low altitudes, spruce from high altitudes, and fir and oak both from dry regions, while for spruce at low altitudes, and oak and beech in wet regions, increased mortality rates were related to lower mDBH. Annual mortality rates varied considerably between the three DBH-classes. Overall, smallest trees had the largest annual mortality rates, while largest trees had, in most cases, lower mortality rates (**Table 3**).

### Climate Conditions

Changes in the minimum site water balance (1SWBmin) and in temperature (1Temp) had variable effects on annual mortality rates, and were significant in four species-ecoregions for 1SWBmin and for 1Temp (**Table 3** and **Figure 4**). Increasing temperatures were associated with increasing mortality only in oak from wet regions, while mortality decreased in pine at high altitudes, fir and oak both in dry regions. The strongest negative impact of 1SWBmin on annual mortality rates was recorded for pine in the lowlands and a weaker effect for spruce in lowlands, thus mortality increased under drier conditions. In contrast, for pine and spruce at high altitudes no significant impact of 1SWBmin on annual mortality rates was found. Fir mortality was not related to 1SWBmin in neither of the two ecoregions. For beech and oak, mortality rates were positively associated with 1SWBmin, indicating an increase in mortality under wet conditions. For beech at dry sites the relationship was only marginally significant, however, 1SWBmin was included in 24 of the best 25 models. Although the relationship of 1SWBmin and mortality was not significant in the averaged model for beech at wet sites, 1SWBmin was included in the best ranked models either as significant single parameter or in a significant interaction (**Supplementary Table S2**). For both, beech and oak, mortality rates at a given 1SWBmin appeared to be slightly higher on wet sites compared to dry sites (**Figure 4**), for beech especially in denser stands (cf. **Supplementary Figure S3**), but differences were generally small. Larger differences were observed for 1Temp. For beech, only wet sites showed a negative temperature effect on mortality in interaction with stand age, whereas mortality was not related to 1Temp on dry sites. For oak, mortality was increased at higher 1Temp on wet sites, whereas the sign was reversed at dry sites (**Figure 4**).

The impact of 1SWBmin and/or 1Temp on mortality was dependent on mDBH for all species, particularly at low and dry sites (**Supplementary Figures S2**, **S3**). In general, the effect of 1SWBmin was stronger in old stands, except for pine at low altitudes with a reversed interaction. The effect of 1Temp negatively interacted with mDBH, indicating a greater vulnerability of young stands relative to old stands to increased temperature. At high-altitude and wet sites, the magnitude of the climatic effect (1SWBmin and/or 1Temp) on mortality was stronger for dense forests stands (i.e., higher stand BA) (**Table 3** and **Supplementary Figure S3**). This was consistent for all species.

#### Altitude and Site Topography

Altitude, slope and aspect were only occasionally related to mortality (**Table 3**) with no clear pattern. For instance, for beech growing on wet sites, mortality rates were negatively correlated with altitude, while for pine from low altitude regions, the pattern was opposite. The steepness of the slope was significantly related to mortality for beech from wet and fir from dry regions. Aspect was negatively related to mortality rates for beech from both dry and wet regions and positively in pine from high altitudes. For

from results shown in Table 2.

gray band ('n obs/plots') indicates the number of observations included/number of plots monitored during the respective inventory interval. Inventory intervals ranged from 3 to 22 years and were carried out in different years for different plots. Because of the different inventory intervals and dates, the temporal trajectories can differ



N, number of plots. <sup>∗</sup>Sum of N per species is higher than overall plot number since more than one species occurred in some plots.

all other species-ecoregions the relationships of mortality with topography were not significant.

# Temporal Trends of Annual Mortality Rates

Temporal trends of annual mortality rates varied considerably across species and ecoregions throughout the entire study period (starting in ∼1900s, 1910s, 1920s, and 1930s depending on the species and ecoregion). Generalized mixed effect models indicated that over the last ∼120 years, pine and spruce from high altitudes and fir from dry regions exhibited the highest increase of annual mortality rates by 1.6, 1.0, and 1.3% per year, respectively (**Table 4**). For other species and ecoregions either no significant or very small shifts were observed (<±0.5%). For pine in lowlands and beech in dry regions, time series started later than 1960 and were therefore not included in this analysis.

Throughout 1960–2013, pine in lowlands showed the highest increase of mortality rates by 2.3% per year, followed by spruce in both ecoregions, with an increase by 1.8 and 1.6% per year at low and high altitudes, respectively (**Table 4**). Fir from dry regions, beech from wet regions and oak from both dry and wet regions showed decline in mortality rates with most pronounced shift in oak from dry regions (−4.2% per year). In the remaining three species-ecoregions the change was not significant.

Over the last ∼120 years, mortality of trees within the large and small DBH-classes exhibited opposite trends: mortality of large trees increased in seven species-ecoregions, while mortality of small trees decreased, increased or showed no significant trend (**Table 4**). Since 1960, this tendency became even more pronounced with an increase of mortality in large trees for six species-ecoregions, and a decrease in small trees for five speciesecoregions or no significant changes. Exceptions were fir at dry sites, where all DBH-classes showed increases in mortality, and in oak where a general decrease of mortality across the DBHclasses was observed. These opposing tendencies also emerged when comparing mortality rates of DBH-classes between two periods before and after 1960, although changes were not always significant due to large data heterogeneity (**Figure 5**).

Despite notable differences between species in temporal trends of annual mortality (**Figure 3** and **Table 4**) and heterogeneous data (different number of plots, inventory dates and intervals between the inventories), the five species displayed


temperature during the growth season.

similar fluctuations in normalized mortality rates (**Figure 6** and **Supplementary Table S3**). Normalized mortality rates express annual mortality rate relative to the long-term average annual mortality rate. Although annual mortality rates changed only mildly during the last ∼120 years (**Table 4**), it is noteworthy that in the most recent three decades (1980–2010) the crossspecies average mortality rate did not drop below the 100-year average. In addition, the frequency of mortality peaks in the last three decades was also higher than earlier. The magnitude of the peaks, however, decreased probably due to the higher plot number in recent years. There were also distinct differences in the temporal mortality trends (cf. **Table 4**). For instance, the temporal trajectory of beech mortality correlated well with all four other species (**Figure 6** and **Supplementary Table S3**), while trajectory of oak mortality was well associated with only two other species: beech and pine.

#### DISCUSSION

In this study, we assessed mortality rates of five dominant species in Swiss forests throughout the last century (1898–2013) and examined factors driving mortality change. The complex study design with different points in time of forest inventories as well as differing length of the inventory periods within and across the three monitoring networks challenges statistical analysis and generates limitations on data interpretation. Nevertheless, some very clear patterns emerged from the one century-long data set. Since 1960, mortality rates of pine in lowlands and of spruce increased significantly, whereas those of oak trees decreased (**Table 4**). Fir and beech showed only minor or not significant changes in mortality rates. Stand properties, especially BA, were the most important factors influencing variability of mortality rates, while climate impact (temporal changes in minimum site water balance and temperature) was inconsistent across species and ecoregions, and depended on interactions with mDBH (i.e., an indicator of stand age) and BA (indicating competition) (**Table 3**). Thus, we could not conclusively confirm our initial hypothesis, that oak and pine would be most tolerant, and beech and spruce would be most vulnerable to drought, at least not in absence of other stand-related factors. The impact of drought and heat on mortality appeared rather to result from a combination of species and site effects, and being further modulated by competition and stand age. The long-term data presented here highlight the complex nature of the processes shaping forest dynamics. Throughout one century, unmanaged

TABLE 4 | Results of the GLMMs with annual mortality rate as response variable, with first year of the inventory period as fixed effect, and plot nested in forest site (see Materials and Methods, Equations 3 and 4) as random effect.


Models were calculated for each DBH-class (small, medium, large) separately, as well as for all trees together. For the latter the DBH-class was additionally included in the model as fixed effect. 'Low' and 'High' refers to low and high altitudes. NA, not enough data available for period before 1960. Significant coefficients are indicated as follows ∗∗∗P < 0.001, ∗∗P < 0.01, <sup>∗</sup>P < 0.05, '.'P < 0.1, n.s., not significant.

forests gradually alter their structure, demography and in some cases also their species composition. Consequently, structural and demographical changes in combination with climate affected mortality rates in our study, implying that a variety of factors need to be accounted for in mortality studies at the ecosystem scale. Below, we discuss the annual mortality rates, the effect of competition and climate variability on mortality, and its impact on the future development of forests in Switzerland.

# Annual Mortality Rates

Average annual mortality rate through the entire study period was 1.5% but varied considerably across species and plots (**Figure 3** and **Tables 1**, **2**). On average, mortality rates of various tree species in Europe range between 0.1 and 2.9 %, with a median of 0.5% (**Supplementary Table S5**), which is lower compared to mortality rates in this study. Beyond the overall inherent, high variability of mortality rates across forests, this may be due to several methodological and conceptual reasons: (1) the 5 cm DBH threshold applied here was lower than in other studies, and this might have increased the average mortality rate of the investigated populations, since mortality tends to be higher in small trees (Rohner et al., 2012; Rigling et al., 2013); (2) The data set included a high proportion of young stands, for which mortality rates are usually higher compared to forests in the optimum phase (Oliver and Larson, 1996); (3) The long study period of over 100 years may have increased the probability of mortality events, although fluctuations of mortality rates across different decadal time periods were small (**Supplementary Table S4**); (4) Our study encompassed a large number of plots across broad altitudinal and climate gradients, including also plots at species distribution margins with exceptionally high mortality, rising the overall average mortality rates; (5) Finally, the absence of management in our study sites might explain the relatively high mortality rates, especially in small and medium size tree classes. Thus, a direct comparison of mortality rates across different studies should be interpreted with caution. These and other factors challenge a direct comparison of absolute mortality rates across different studies, while relative changes of mortality rates as well as their driving factors should be comparable across studies.

# Drivers of Mortality

Mortality rates were best explained by a combination of stand properties and climate variables, whereas altitude and topography were less significant. In concordance with other studies (Condés and del Río, 2015; Pillet et al., 2018) our results showed that when climate and competition considered together a larger portion of variation in mortality rates can be explained, as when considered separately.

# Effects of Stand Characteristics on Mortality Rates

Basal area, and thus competition, was the strongest predictor variable of mortality rates. For all species and ecoregions a consistent effect of increasing mortality with higher BA was found (**Table 3** and **Figure 4**), as high stand density leads to increased competition for resources, which consequently may result in higher mortality (Giuggiola et al., 2013; Ruiz-Benito et al., 2013; Condés and del Río, 2015; Bradford and Bell, 2017; Young et al., 2017).

In all species, mortality rates were largest among the smallest diameter trees, while the largest trees had lowest mortality rates, as reported also in many other studies (Monserud and Sterba, 1999; Mantgem and Stephenson, 2007; van Mantgem et al., 2009; Peng et al., 2011; Neumann et al., 2017) and could be explained by smaller trees being outcompeted by larger trees (Coomes and Allen, 2007) or by trees of the same size class. The impact of stand age (implied by mean DBH) on mortality rates was significant for most species and ecoregions, but with variable sign and strength, which might reflect differences in stand development of the included plots.

# Effects of Climate on Mortality Rates

Climate variables, i.e., changes in temperature and SWBmin, had contrasting influence on mortality rates, depending on species and ecoregion. For spruce and pine, mortality of lowland forests increased with increasing dryness (negative 1SWBmin). Contrary to our initial hypothesis, this effect was most conspicuous for pine, for which SWBmin was the overall strongest predictor of annual mortality rates (**Table 3** and **Figure 4**). Increased mortality of pine due to drought became, however, an increasingly observed pattern in the dry valleys of the Valais, in Grisons and the Churer Rheintal in Switzerland (Dobbertin et al., 2005;

FIGURE 5 | Annual mortality rate by species and DBH-class for time periods before and after 1960. Asterisks indicate statistically significant differences in mortality rates per DBH-class and species between the two time periods (∗∗∗P < 0.001, ∗∗P < 0.01, and <sup>∗</sup>P < <0.05, unpaired t-test on normally distributed data). Number of plots included is the same as in Table 2.

Bigler et al., 2006; Rigling et al., 2006, 2013; Wohlgemuth and Rigling, 2014). Although pine can recover after a few incidents of drought (Eilmann et al., 2010; Eilmann and Rigling, 2012), several consecutive dry years can lead to a reduction in vitality and associated higher susceptibility of trees to pests, pathogens and parasites (Bigler et al., 2006; Rigling et al., 2010). Although Scots pine is generally drought-tolerant, but - growing already on dry sites – an increase in frequency and severity of drought events may force it beyond its physiological limits. Given that pine has been introduced by humans outside its native range and was promoted until 1950 (Gimmi et al., 2010), it is now being outcompeted by indigenous or invasive species, or species which are better adapted to dry conditions, like pubescent oak (Rigling et al., 2013). Spruce in the lowlands has also been found to be susceptible to drought (Vanoni et al., 2016a,b) and is expected to be substituted by more drought-tolerant species in the future (Hanewinkel et al., 2013). However, we did not find a very strong impact of drought on spruce mortality. Instead,

BA and mDBH were more influential. This might be due to the fact that we excluded plots that were substantially affected by storm damages or bark beetle attacks. These plots might have been also experienced severe drought beforehand since the intensity of the attack is often amplified by preceding drought events (Meier et al., 2014; Kolb et al., 2016). Thus, secondary factors, such as bark beetles, affect spruce mortality rates often to a larger extent than drought itself. Additionally, in contrast to the lowland pine plots, which were restricted to very dry regions, lowland spruce plots exhibited on average a relatively high soil water availability (**Table 1**). Contrary to the lowlands, mortality rates of pine and spruce at higher altitudes were not related to drought, but increased under low temperatures, probably due to direct low-temperature limitation (Körner, 1998, 2003; Vanoni et al., 2016a) combined with indirect temperature stress such as frost damage, winter desiccation, or low-temperature photoinhibition (Barbeito et al., 2012). Similar patterns were also observed for Scots pine growing at higher altitudes in Spain (Ruiz-Benito et al., 2013).

For silver fir trees, drought did not show any significant influence on annual mortality (**Table 3**). This pattern confirms other studies from the Churer Rheintal in Switzerland, where the detrimental impact of drought on growth of silver fir trees is mitigated by deep soils with higher water holding capacity and northern exposition (Wohlgemuth and Rigling, 2014). On drier sites, the strong impact of low temperatures on mortality rates, could presumably be explained by the sensitivity of silver fir to frost (Lebourgeois et al., 2010). In summary, silver fir in Switzerland seems to be able to cope with the current level of experienced droughts and is still within its physiological boundaries, as also observed in other parts of Europe (George et al., 2015).

Increasing mortality with higher water availability was found for beech and, even more noticeably, for oak (**Table 3** and **Figure 4**). This pattern is counter-intuitive, especially for beech on dry sites, although it was also observed in other studies (Rohner et al., 2012; Nothdurft, 2013; Ruiz-Benito et al., 2013), and might be related to other concurrent abiotic conditions. Presumably, higher water availability in moist years can lead to an increase of total leaf area and subsequent overshadowing of short, young trees limiting their access to light and water, consequently resulting in their increased mortality. Beech is often considered to be drought sensitive (Gessler et al., 2007) and, in general, performs better on wetter sites (Aranda et al., 2000; Gessler et al., 2007). A recent study also found that beech growth appears to be very location specific, with growth decline at lower altitudes and growth increase at higher altitudes (Dulamsuren et al., 2017). In concordance with this, we observed higher beech mortality at lower altitudes compared to higher altitudes.

It has been observed that beech (Vitasse et al., 2014) and also oak (Jensen and Hansen, 2010; Grossiord et al., 2014) trees can adapt very well to dry conditions, possibly due to their deep rooting penetration, efficient stomatal control (Nothdurft, 2013), and high degree of evolutionary adaptability (Roloff and Grundmann, 2008). Thus, higher mortality rates per given 1SWBmin at wet sites compared to dry sites for beech and oak partly confirm the hypothesis that drought resistance of trees increases with lower site water availability (Kunz et al., 2018). This might be due to the high degree of genetic adaptability. However, differences related to SWB were small and might also be associated with generally higher turn-over at sites with higher water availability. The temperature effect on mortality of the two species clearly differed between dry and wet sites of the two species, indicating that trees that are adapted to dry and warm conditions might be able to cope better with occurring heat waves as projected for the future. For oak species, the impact of pathogens related to wet conditions and frost (Führer, 1998; Gaertig et al., 2005) but also to drought (Wood et al., 2018) is an additional relevant factor driving mortality (Haavik et al., 2015).

# Effects of Stand Characteristics and Climate on Mortality Rates

The inclusion of an interaction term of BA or mDBH (i.e., stand age) with climate variables improved the model parsimony for pine, spruce, fir and oak, indicating that the effect of climatic variables varies across stand structures and developmental stages (**Table 3**). These effects were more pronounced on dry and lowland sites compared to wet and highland sites (**Supplementary Figures S2**, **S3**). Our results indicate that the impact of drought and/or temperature on tree mortality increased with competition (i.e., higher BA), which has also been reported for many tree species in Spain (Ruiz-Benito et al., 2013). In addition, pine has been found to perform better after stand density reduction on xeric sites in Switzerland (Giuggiola et al., 2013, 2018), further indicating that competition is an important factor in driving mortality and demography (Kohler et al., 2010; Sohn et al., 2016). For age, the pattern reversed along the temperature gradient and young stands were generally more strongly affected by increased temperatures compared to old stands. This might be due to the fact that young stands are usually denser and prone to self-thinning. Our results indicate that heat and/or drought may trigger the self-thinning process in young stands.

#### Temporal Trends in Annual Mortality

Temporal trends in annual mortality rates varied depending on the data grouping factor (e.g., species, ecoregions, DBHclasses, **Figures 3**, **5**, **6** and **Table 4**) partly due to methodological reasons (i.e., data averaging and integration procedures). For example, discrepancies between temporal trajectories in GLMMs (**Table 4**) and figures (**Figures 3**, **5**) originated from the fact that models used actual dates of inventories and interval lengths and accounted for plots as grouping factor, while figures show mortality estimates averaged per decade or even larger time periods and/or per species and ecoregions. Nevertheless, some clear patterns emerged. Species-specific trends of annual mortality rates across the ∼120 years were modest, but became larger during the second half of the century (**Figure 3** and **Table 4**). Spruce and pine exhibited increasing mortality rates, while fir on dry sites, oak on dry and wet sites, and, to a lesser degree, beech on wet sites showed decreasing mortality rates. Thus, the normalized mortality rates (**Figure 6**) averaged over all species showed a slight,

but consistent, increasing trend in mortality across all species, but with a large uncertainty at the beginning of the study period due to low plot number. The steady increase in the normalized mortality rates since 1900, rather than an obvious break point at the start of pronounced climatic changes around 1960 (**Figure 2**), suggests that the increase in mortality is more strongly driven by gradually changing stand parameters (such as density and age, as shown in the previous paragraphs) and to a lesser extent by climate. However, it is also noteworthy that fluctuations of normalized mortality rates of all species are remarkably similar, especially since the 1960s (**Figure 6** and **Supplementary Table S3**). Thus, mortality peaks were for all species relatively (given the heterogeneity of the data structure) synchronous every 10–20 years, especially during the recent decades. Additionally, since the 1980s, annual mortality rates were consistently, although only slightly, above the long-term average. Temporal disturbance patterns strongly synchronized across landscapes have been also observed in temperate forests in Europe during 1986–2016 and were related to preceding drought and storm events (Senf and Seidl, 2018). Thus, forest stand properties, such as BA, might drive the overall long-term trend of forest mortality, while climate modulates these trends with an assumed increasing importance during the recent decades.

Our results also demonstrate that trends in annual mortality rates differed depending on tree size and even run in opposing directions. Mortality rates of small trees tended to decrease or did not change, while mortality rates of large trees generally increased over time (**Figure 5** and **Table 4**). This might be a stand aging effect, since observed forests aged for up to 120 years throughout the study period and run through different stand developmental stages. Moreover, recent studies in temperate forests indicate that susceptibility to drought increases with tree age or size (Carrer and Urbinati, 2004; Lloret et al., 2011; Ding et al., 2017), probably due to higher risk of hydraulic failure (Bennett et al., 2015; McDowell and Allen, 2015) or higher costs involved in growing new functional xylem after the drought has passed (Trugman et al., 2018). As a consequence, an increased mortality of large trees may have allowed for higher survival of young trees. However, also contradictory observations have been made, showing higher susceptibility to drought of young trees (Mantgem and Stephenson, 2007; Colangelo et al., 2017). More research is needed to clarify this context dependency (Ding et al., 2017).

#### Future Implications

Contrarily to our initial hypothesis, we found a significant drought-induced mortality only in pine at low altitudes and, to a lesser extent, in spruce from low altitudes (**Table 3**). It appears, that drought-tolerant pine, growing already on dry sites, was severely impacted by drought events, but that fir, beech and oak, growing mainly on soils with a good water holding capacity, seemed not yet to be affected by the occurrence of drought in Switzerland. They appear to be quite well adapted to the current conditions within their ecological niche. This pattern is contrary to many studies reporting rising mortality rates due to drought and heat (Breshears et al., 2005; Gitlin et al., 2006; Allen et al., 2010, 2015; Anderegg et al., 2013; Gustafson and Sturtevant, 2013; Holmgren et al., 2013; Reichstein et al., 2013; Bradford and Bell, 2017; Chen et al., 2018; McDowell et al., 2018). One reason for this discrepancy may stem from the fact that many studies reported mortality after a specific drought or heatwave event and/or were restricted to relatively short time intervals of several years, including a large climatic disturbance event (for example, Martínez-Vilalta and Piñol, 2002; Dobbertin, 2005; Landmann and Dreyer, 2006; Ogibin and Demidova, 2009). The inventory intervals of five or more years also may dilute the climate signal on mortality (Dobbertin et al., 2005; Huelsmann et al., 2018). Nevertheless, our results show that mortality rates of five dominant tree species in Switzerland increased only slightly over the last ∼120 years, which could mainly be related to changes in stand structure. This suggests that Swiss forests have been resilient to recent climates change so far (with the exception of some hotspots) and that instead of an abrupt transition of forests, changes in species composition might occur more gradually and subtly. Nevertheless, the amplified effect of drought and heat under competition might indicate potentially strong changes in the demographic structure and species composition of forests in the future (Ruiz-Benito et al., 2013).

It is predicted that the future will bring longer and more intense droughts in many regions across Europe (Fischer and Schär, 2010; Dai, 2013; CH2014-Impacts, 2014). The resilience of trees to climatic changes requires further investigation, for instance, how quickly and effectively trees recover from episodes of drought and heatwaves. Some studies suggest that drought tolerant species may shift to sites that are becoming more arid, as these species are already acclimated to drier conditions. This shift may eliminate species that are already at their ecophysiological limits (Lévesque et al., 2013, 2014; Rigling et al., 2013). Moreover, species specific responses to insect attacks, pest and pathogens, may shape the structure of future forests. For example, in Germany and Austria, an increase in mortality risk is predicted especially for spruce at low altitudes, while beech and oak may be more robust to climate change (Lexer et al., 2002; Nothdurft, 2013; Ding et al., 2017), which could be supported by the observations from Switzerland reported in this study.

#### DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

#### AUTHOR CONTRIBUTIONS

AR, SE, and AZ conceived the study. SE analyzed the data with support from BR, AB, AKB, NKR, and AR. SE and KZ wrote the manuscript with contribution from all other co-authors.

#### FUNDING

This study was funded by the Swiss Federal Office for the Environment FOEN and the Swiss Federal Institute for Forest, Snow and Landscape Research WSL as part of the research program 'Forests and Climate Change.' NKR received funding by the German Research Foundation through its Emmy Noether Programme (RU 1657/2-1).

### ACKNOWLEDGMENTS

fpls-10-00307 March 21, 2019 Time: 11:16 # 15

We would like to thank Jan Remund from meteotest for providing the climate data. We also would like to thank the networks EFM, LWF, and NKF for providing the data, and all the field

#### REFERENCES


teams, who did the extensive monitoring work since the last 100 years. This study is dedicated to Matthias Dobbertin (†2012) who initiated this project.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.00307/ full#supplementary-material


and show lower growth prior to tree death. Front. Plant Sci. 8:135. doi: 10.3389/ fpls.2017.00135


Forests and Forest Management to Changing Climate with Emphasis on Forest Health: A Review of Science, Policies, and Practices", Umeå.


Apennines, Italy. Glob. Chang. Biol. 14, 1265–1281. doi: 10.1111/j.1365-2486. 2008.01570.x


im Klimawandel. Birmensdorf: Eidg. Forschungsanstalt für Wald, Schnee und Landschaft WSL.


**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 Etzold, Zieminska, Rohner, Bottero, Bose, Ruehr, Zingg and ´ Rigling. 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.

# Ecosystem Services Related to Carbon Cycling – Modeling Present and Future Impacts in Boreal Forests

Maria Holmberg<sup>1</sup> \*, Tuula Aalto<sup>2</sup> , Anu Akujärvi<sup>1</sup> , Ali Nadir Arslan<sup>2</sup> , Irina Bergström<sup>1</sup> , Kristin Böttcher<sup>1</sup> , Ismo Lahtinen<sup>1</sup> , Annikki Mäkelä<sup>3</sup> , Tiina Markkanen<sup>2</sup> , Francesco Minunno<sup>3</sup> , Mikko Peltoniemi<sup>4</sup> , Katri Rankinen<sup>1</sup> , Petteri Vihervaara<sup>1</sup> and Martin Forsius<sup>1</sup>

<sup>1</sup> Finnish Environment Institute (SYKE), Helsinki, Finland, <sup>2</sup> Finnish Meteorological Institute (FMI), Helsinki, Finland, <sup>3</sup> Department of Forest Sciences, University of Helsinki, Helsinki, Finland, <sup>4</sup> National Resources Institute (LUKE), Helsinki, Finland

#### Edited by:

Giovanna Battipaglia, Università degli Studi della Campania Luigi Vanvitelli, Italy

#### Reviewed by:

Roberto Cazzolla Gatti, Tomsk State University, Russia Miguel Montoro Girona, Swedish University of Agricultural Sciences, Sweden

> \*Correspondence: Maria Holmberg maria.holmberg@ymparisto.fi

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 07 October 2018 Accepted: 05 March 2019 Published: 26 March 2019

#### Citation:

Holmberg M, Aalto T, Akujärvi A, Arslan AN, Bergström I, Böttcher K, Lahtinen I, Mäkelä A, Markkanen T, Minunno F, Peltoniemi M, Rankinen K, Vihervaara P and Forsius M (2019) Ecosystem Services Related to Carbon Cycling – Modeling Present and Future Impacts in Boreal Forests. Front. Plant Sci. 10:343. doi: 10.3389/fpls.2019.00343 Forests regulate climate, as carbon, water and nutrient fluxes are modified by physiological processes of vegetation and soil. Forests also provide renewable raw material, food, and recreational possibilities. Rapid climate warming projected for the boreal zone may change the provision of these ecosystem services. We demonstrate model based estimates of present and future ecosystem services related to carbon cycling of boreal forests. The services were derived from biophysical variables calculated by two dynamic models. Future changes in the biophysical variables were driven by climate change scenarios obtained as results of a sample of global climate models downscaled for Finland, assuming three future pathways of radiative forcing. We introduce continuous monitoring on phenology to be used in model parametrization through a webcam network with automated image processing features. In our analysis, climate change impacts on key boreal forest ecosystem services are both beneficial and detrimental. Our results indicate an increase in annual forest growth of about 60% and an increase in annual carbon sink of roughly 40% from the reference period (1981–2010) to the end of the century. The vegetation active period was projected to start about 3 weeks earlier and end ten days later by the end of the century compared to currently. We found a risk for increasing drought, and a decrease in the number of soil frost days. Our results show a considerable uncertainty in future provision of boreal forest ecosystem services.

Keywords: forest growth, carbon sink, vegetation active period, JSBACH, PREBAS, model, continuous monitoring, webcam

# INTRODUCTION

Ecosystem services (ES) are defined as the contributions that ecosystems make to human wellbeing (e.g., Costanza et al., 1997; Daily and Matson, 2008). The ES concept has become widely used and serves to emphasize the dependency of human society's welfare on natural ecosystems (MEA, 2005; Costanza et al., 2017). A wide variety of ES are found in forests, which cover about 30% of global terrestrial area (FAO, 2018). Forests provide timber, and food, they conserve biodiversity, regulate

water resources, and provide recreational opportunities (Shvidenko et al., 2005; Saastamoinen et al., 2014). Forests are essential factors of the global carbon (C) cycle, with an important role in regulating atmospheric concentrations of carbon dioxide (CO2) (Kurz et al., 2013). Boreal forests represent 29% of global forests (Shvidenko et al., 2005), comprising the circumpolar vegetation zone of high northern latitudes that covers one of the world's largest biogeoclimatic areas (Brandt et al., 2013). The ES provided by boreal forests thus benefit human society both locally and on the global scale (Gauthier et al., 2015). Boreal forests account for about 20% of the global C sinks (Pan et al., 2011). Thereby boreal forests provide a climate regulating ES with bearing on climate change mitigation on the global level, although trade-offs are also recognized (Luyssaert et al., 2018).

Estimates of recent trends in above-ground biomass C indicate an increase for boreal forests (Liu et al., 2015). The stand and landscape level characteristics of the boreal forests of North America, Fennoscandia and Russia vary between the regions because of historical and current regional differences in natural disturbances and management practices (Gauthier et al., 2015; Kuuluvainen and Gauthier, 2018). Statistics from two neighboring boreal countries show increasing trends in roundwood increment in Finland (Vaahtera et al., 2018) and Sweden (Skogsdata, 2018). The future rate of C uptake by forests depends on how ambient temperature, land use and resource management practices evolve (Ahlström et al., 2012; Shao et al., 2013; Lappalainen et al., 2016). Rapid climate warming projected for the boreal zone (IPCC, 2013) has been observed in North America (McKenney et al., 2006; Price et al., 2013), Fennoscandia (Mikkonen et al., 2015; Hedwall and Brunet, 2016), and Russia (Schaphoff et al., 2016), which may have significant consequences for future boreal forest ES. Management practices may interact with climate change with consequences for forest biodiversity (Tremblay et al., 2018; Cadieux et al., 2019). Schaphoff et al. (2016) found that the impacts of climate change on Russia's boreal forests are often superimposed by other environmental and societal changes, while Kalliokoski et al. (2018) estimated likely increases in future forest productivity in Finland, although with a high uncertainty. Climate change may affect the occurrence and extent of natural disturbances, such as insect outbreaks and fire (Schaphoff et al., 2016), e.g., Navarro et al. (2018) report a northward shift of spruce budworm (Choristoneura fumiferana) during the 20th century. Earlier thermal growing season has been found to be associated with earlier onset of biospheric C uptake, whereas earlier termination of biospheric activity was associated to later termination of thermal growing season (Barichivich et al., 2012). Warmer springs have consequences for bird reproduction (Ludwig et al., 2006; Wegge and Rolstad, 2017) and for moth multivoltinism, in combination with warmer summers (Pöyry et al., 2011). Less severe winter colds may promote the expansion of forest pests (Fält-Nardmann et al., 2018). Warmer winters may also have impacts on the possibilities for winter harvest on drained peatlands (Hökkä et al., 2016) and for winter recreation (Neuvonen et al., 2015).

The pathway from ecosystems, their biophysical structure, processes and functions to their benefit and value to human society, is illustrated by the ES cascade model, which portrays ES as emerging from the functional and structural properties of the ecosystem (Potschin and Haines-Young, 2011). In the Common International Classification of ES (CICES), services are classified into provisioning, regulating and maintenance, and cultural services (Haines-Young and Potschin, 2018). Several studies have discussed mapping of ES provision potential at different scales (Burkhard et al., 2009; Vihervaara et al., 2010; Maes et al., 2013; Albert et al., 2016a). The European Commission seeks to improve the basis for implementing the EU Biodiversity Strategy for 2020 (European Commission, 2011) by encouraging national ES assessments of the member states of EU through its flagship project MAES (European Commission, 2018). National accounting of natural capital needs information from national ES accounts and ongoing development of natural-capital accounts also supports national ES assessments (Schröter et al., 2016). Saastamoinen et al. (2014) applied the CICES hierarchy on the boreal forest ES in Finland, reporting their results in a conceptual and historical context. Mononen et al. (2016) developed a framework of ES indicators for Finland that complies with both national circumstances and international typologies such as the cascade model (Potschin and Haines-Young, 2011) and the CICES framework (Haines-Young and Potschin, 2018). ES are increasingly used to inform policies, and frameworks including ES analysis for decision support are becoming available (Albert et al., 2016b). Potentially highly promising avenues for underpinning resource allocation decisions are based on detailed place-based analyses of supply and demand of ES (e.g., Kopperoinen et al., 2014; Vauhkonen and Ruotsalainen, 2017). On the other hand, broad, unspecified ES may be more easily adopted by policy actors, according to Van Oudenhoven et al. (2018).

Our aim is to illustrate potential impacts of climate change on boreal forest ES. In this paper we (i) apply two dynamic ecosystem models (JSBACH and PREBAS) driven by a set of climate change scenarios to simulate present day and future values of a set of biophysical variables; (ii) compare simulated present day values with available observations and statistical data; (iii) suggest interpretations of the biophysical variables as ES; (iv) discuss the consequences of changes in these biophysical variables on future ES of boreal forests (**Supplementary Figure 1**).

### MATERIALS AND METHODS

#### Ecosystem Models

We applied the land ecosystem model JSBACH (Raddatz et al., 2007; Reick et al., 2013; Goll et al., 2015; Gao et al., 2016) and the stand growth model PREBAS (Valentine and Mäkelä, 2005; Peltoniemi et al., 2015b; Minunno et al., 2016, 2019) for the land area of mainland Finland, using input data in spatial resolution ranging from 16 m (forest resources, PREBAS), 0.1◦ (land surface characteristics, JSBACH) to a 0.2◦ × 0.1◦ longitudelatitude grid (climate data). We report simulated results for a set of biophysical variables for four time periods (1981– 2010, 2011–2040, 2041–2070, 2071–2100). Two variables reflect directly ecosystem C fluxes: gross primary productivity (GPP; gCm−<sup>2</sup> yr−<sup>1</sup> ) and net ecosystem exchange (NEE; gCm−<sup>2</sup> yr−<sup>1</sup> ).

Ecosystem phenology is described by three variables: the length of vegetation active period (VAPlength; days), when terrestrial vegetation is assimilating C through photosynthesis, the start (end) of vegetation active period (VAPstart; VAPend; days from January 1st, when the daily level of photosynthesis first exceeds, or returns to below 15% of its summertime value). Stemwood growth (m<sup>3</sup> yr−<sup>1</sup> ) was simulated with PREBAS, and the number of summer dry days (soil moisture < 5th percentile of reference period) and number of winter days with soil frost were simulated with JSBACH. The models have been compared earlier and found to produce similar GPP estimates at local and national level (Peltoniemi et al., 2015a). Here we use the two models in parallel to provide further information about the changes in biophysical variables under climate change, and also in supplementing each other for variables predicted by only one of the models. We present the results as boxplots and tabulated percentile values, and cumulative distribution functions aggregated over the whole area, as well as time series and maps for some variables.

JSBACH is a land surface model of an earth system model of Max Planck institute for meteorology (MPI-MET) and describes the biogeophysical processes that regulate the balances of water and CO2. The storage of water and C into the ecosystem as well as their release to the atmosphere is regulated by the climatic variables. Land vegetation is divided into plant functional types (PFT), and for our domain (the Finnish mainland) we based the PFT distribution on Finnish CORINE land cover data (CLC, 2012), which contains information about soil type, thus providing soil characteristics consistent with the vegetation cover data. Seasonal development of leaf area index is regulated by air temperature and soil moisture with PFT specific maximum leaf area index as a limiting value. For the generation of the seasonal cycle, PFTs are divided in summergreen, evergreen, grass and crop phenology types (Böttcher et al., 2016). Photosynthesis is described according to Farquhar et al. (1980), using PFT-specific parameters for the maximum carboxylation (Vmax) and electron transport (Jmax) rates. Global parameter values were used in this study, and as JSBACH was run without explicit nitrogen cycle, the mean values at 25◦C (Vmax25) were applied, with Jmax25 = 1.9 · Vmax25 for all PFTs (Wullschleger, 1993; Kattge et al., 2009). The photosynthetic rate is resolved first under non-water-stressed conditions to attain photosynthetic activity (Schulze et al., 1994), and limitations in water availability are accounted for Knorr and Heimann (1995) and Knorr (2000). The radiation absorption within the vegetation canopy is calculated for three layers (Sellers, 1985), accounting for clumping of the leaves in sparse canopies (Knorr, 2000). In addition to the canopy processes the model consists of a 5-layer soil moisture description (Hagemann and Stacke, 2015) and the Yasso soil C module (Goll et al., 2015). We adopted a formulation that delays the beginning of photosynthetic activity of evergreen species in spring (Kolari et al., 2007), as according to Böttcher et al. (2016) the start date of the photosynthetically active season of coniferous evergreens in the model is ahead of the observed. We decreased the threshold of the temperature sum regulating the bud-break from 4 to 2◦C in accordance to findings by Böttcher et al. (2016). Furthermore, a condition that reduces stomatal conductance under supersaturation was removed because it falsely prohibits photosynthesis under conditions of very high humidity.

In addition, simulations were performed with PREBAS, a C-balance based stand growth and gas exchange model (Valentine and Mäkelä, 2005; Peltoniemi et al., 2015b; Minunno et al., 2019). In PREBAS, photosynthesis (GPP) and evapotranspiration are calculated using a light-useefficiency approach linked to soil moisture and driven by daily environmental inputs of radiation, temperature, vapor pressure deficit, precipitation and ambient CO<sup>2</sup> concentration (Peltoniemi et al., 2015b; Minunno et al., 2016; Kalliokoski et al., 2018). GPP is allocated to mean-tree growth and respiration at an annual time step, and allocation of growth to different tree components is based on conservation of structural constraints, e.g., the pipe model (Valentine and Mäkelä, 2005). Tree mortality due to crowding is included in the model. The growth module updates canopy leaf area index which is used as input to the gas exchange module in the following year. To calculate NEE, PREBAS has been linked with the soil C model YASSO07 through annual litter inputs (Liski and Westman, 1995; Tuomi et al., 2009). In addition to weather data, PREBAS requires information on the initial state of the simulated forest, and forest management actions, including thinning, clearcut and regeneration, to obtain a realistic dynamic pattern of forest development. The relatively simple structure of the model allows us to apply it at a large regional scale and make simulations of C and water fluxes that are both climate and management sensitive. Here, we defined forest management on the basis of current management recommendations in Finland (Sved and Koistinen, 2015; Minunno et al., 2019). The simulations were carried out by linking PREBAS to information on soil fertility, stand basal area, mean height and mean diameter, derived from 16 m resolution multisource forest inventory data maps (Tomppo et al., 2014; Mäkisara et al., 2016; LUKE, 2018a,b). The forest resource maps were divided into 16 km grid cells. Within each grid cell, simulations were carried out for 50 forest classes with the highest proportional areas, accounting also for the remaining classes by distributing their area to these dominant classes. The forest classes represented different combinations of dominant tree species, age, basal area and mean height. Categories of soil fertility were herb-rich heath, mesic heath, sub-xeric heath, xeric heath, and poorer types (Cajander, 1949). To account for mineral and peat soil types, we used two sets of soil parameters, one describing typical mineral soil, and one for a soil layer with high soil water retention capacity (depth = 430 and 1000 mm, respectively), as PREBAS does not simulate organic soils as such. The simulated annual values for each class were aggregated to grid cell level based on 16 m resolution forest maps. Climate data from the nearest climate grid point were used for each 16 km forest simulation grid cell.

We compared the simulated annual NEE for the period 1981– 2010 with reported values of the C sink of Finnish forests (Official Statistics of Finland, 2019). The simulated stemwood growth values for the period 2009–2012 were compared to observed forest growth for the same period. The simulated VAPstart days were compared to estimates based on webcam images (Arslan et al., 2017; Peltoniemi et al., 2018a,b), as well as to

those estimated from satellite observations (SYKE Vegetation Phenology 2001–2017).

#### Climate Change Scenarios

The climate change scenario data through years from 1981 to 2100 were obtained as the output from five global climate models (GCMs; CanESM2, CNRM-CM5, GFDL-CM3, HadGEM2-ES and MIROC5) from the fifth phase of the Coupled Model Intercomparison Project (CMIP5; Meehl et al., 2009; Taylor et al., 2012). We used output for climate models driven by three representative concentration pathways (RCPs), that lead to radiative forcing levels of 2.6, 4.5 and 8.5 W m−<sup>2</sup> by the end of the century (Moss et al., 2010; Van Vuuren et al., 2011). These pathways include one mitigation scenario (RCP2.6), one stabilization scenario (RCP4.5) and one high emission scenario (RCP8.5) (Van Vuuren et al., 2011). The resulting climate variables were down-scaled to a 0.2◦ × 0.1◦ longitude-latitude grid by a quantile-quantile type bias correction algorithm for daily mean temperature, relative humidity, shortwave radiation and wind speed (Räisänen and Räty, 2013) and parametric quantile mapping for daily precipitation (Räty et al., 2014). Gridded harmonized FMI meteorological data by Aalto et al. (2013) were used. Long-wave radiation data were interpolated by a bilinear interpolation method. Global mean CO<sup>2</sup> concentrations from the RCPs 2.6, 4.5 and 8.5 were linearly interpolated to monotonously increase through the calendar years. The changes of temperature and precipitation from a baseline period 1981–2010 to periods 2011–2040, 2041–2070, and 2071–2099 in Finland varied considerably between data from five down-scaled CMIP5 climate models (**Figure 1**). PREBAS was run with all three RCPs, while JSBACH input was taken from RCP4.5 and RCP8.5. The RCP8.5 output of HadGEM2-ES was not included in JSBACH runs.

#### Ecosystem Services

On the basis of the analysis of Saastamoinen et al. (2014) and Mononen et al. (2016), we identified a set of key boreal forest ES. In the provisioning services section of the CICES classification, we selected productivity and supply of harvestable wood as key benefits to human society (**Table 1**). Forestry land covers 86% of the total land area of Finland, and forest industry products accounted for 20% of the total value of Finnish goods exports in 2017 (Vaahtera et al., 2018). Gross primary production, or the rate at which terrestrial vegetation assimilates CO<sup>2</sup> in photosynthesis, (GPP gC m−<sup>2</sup> yr−<sup>1</sup> ) was simulated by JSBACH and PREBAS, and forest stemwood growth (m<sup>3</sup> yr−<sup>1</sup> ) by PREBAS.

For the regulation and maintenance section of the CICES classification, the key service studied was related to regulating



climate through C sequestered by growing vegetation. The corresponding benefit to society was avoided CO<sup>2</sup> forcing (**Table 1**). The proxy for CO<sup>2</sup> forcing, NEE (gC m−<sup>2</sup> yr−<sup>1</sup> ), is the difference between the total respiration rate of the ecosystem, and GPP. When the flux is from the atmosphere to the ecosystem, NEE is negative, and when the flux is from the ecosystem to the atmosphere, NEE is positive. In 2017, the Finnish total greenhouse gas emissions were 55.5 million metric tons CO<sup>2</sup> eq, including energy, industrial, agriculture, and waste sectors. In contrast, the net removal from the atmosphere by forest land was 27.0 million metric tons CO2 eq. Altogether, the net sink of the land use, land use change, and forestry sector (LULUCF) was 20.4 million metric tons CO<sup>2</sup> eq (Official Statistics of Finland, 2019). Climate change impacts on future values of NEE were simulated by JSBACH and PREBAS.

Risk of productivity losses related to summer drought, and decreasing opportunities for winter harvest on drained peatlands were considered as examples of regulating services. JSBACH simulations of the number of dry summer days, and soil frost days, were used for the climate impact projections. We used the length and the timing of the vegetation active period as proxies for both a regulating service (the maintenance of habitats), and a cultural service (recreational opportunities). The impact of climate change on the length (VAPlength, number of days) and timing of the vegetation active period (VAPstart, VAPend, day of year) was simulated with JSBACH and PREBAS.

Current potential ES provision was estimated as the median values of the simulated biophysical variables for the reference period (1981–2010). The impact of climate change on ES provision was interpreted as future changes in simulated biophysical variables compared to reference period median values. Positive changes in GPP, stemwood growth and number of soil frost days are in our analysis associated with benefits for human society. Positive values of NEE represent a flux from the ecosystem to the atmosphere (CO<sup>2</sup> emissions), while negative values of NEE (CO<sup>2</sup> removals) represent avoided CO<sup>2</sup> forcing. Thus, in our analysis, positive changes in the simulated values of NEE are considered losses, and negative changes benefits to human society. Increasing frequencies of dry summer days are similarly considered losses. An increase in the length of the vegetation active period, however, may lead to both benefits and losses.

#### RESULTS

#### Provisioning Services: GPP and Supply of Harvestable Wood

GPP was projected to increase with climate change, more so with higher radiative forcing (RCP8.5), and more toward the end of the century (**Figure 2**). For the low emission scenario (RCP2.6) hardly any difference between the time periods can be seen in the simulated GPP values. Differences between the medium and the high emission scenario simulations (RCP4.5 and RCP8.5) are evident only from the mid-century onward (**Figure 2** and **Supplementary Figure 2**). The estimated median increase in GPP was 34% for the mid-century time period (2041–2070), and 46% for the end of the century (2071–2100), over all radiative forcing levels (**Supplementary Table 1**). Current observed mean annual growth in 15 forestry regions of Finland compared well (R <sup>2</sup> = 0.95) with PREBAS simulated results for the same period (2.0–7.6 m<sup>3</sup> ha−<sup>1</sup> yr−<sup>1</sup> ; **Supplementary Figure 3**). In the climate change projections for mean annual forest growth, all scenarios give similar results for the first time period (2011–2040, **Figure 2** and **Supplementary Figure 4**), whereas the differences become clear later, especially toward the end of the century. The highest increase in mean annual growth was projected for the north of Finland (**Figure 3**). The simulated country-wide median forest growth increased from 5.6 m<sup>3</sup> ha−<sup>1</sup> yr−<sup>1</sup> in 1981–2010 by 2.7 to 8.3 m<sup>3</sup> ha−<sup>1</sup> yr−<sup>1</sup> in 2041–2070, and by 3.4 to 8.7 m<sup>3</sup> ha−<sup>1</sup> yr−<sup>1</sup> in 2071–2100 (**Supplementary Table 1**), corresponding to an increase of 48% by mid-century, and 60% by the end of the century.

# Regulating Services

#### Avoided CO<sup>2</sup> Forcing

Interpreted as an ES, a more negative value of NEE means that the ecosystems of the region have a potential to mitigate the radiative forcing by atmospheric CO2, which is the most important contributor of human induced climate change. Across Finland, the simulated median annual NEE was –50 gC m−<sup>2</sup> yr−<sup>1</sup> for the period 1981–2010 (**Supplementary Table 1**). Reported current national total annual C sink of forest biomass and soil varied in the range –35 to –18 Mt CO<sup>2</sup> eq yr−<sup>1</sup> in the period 1990 to 2017 (Official Statistics of Finland, 2019). Averaged over total forestry land (262,000 km<sup>2</sup> ), the reported range corresponds

to annual removals of –36 to –19 gC m−<sup>2</sup> yr−<sup>1</sup> . Our simulated reference period annual C sink thus clearly exceeds the range of reported current values (Official Statistics of Finland, 2019).

Simulations with the low emission scenario (RCP2.6), which include only PREBAS results, represent clearly larger C sink values (more negative NEE) than simulations with the other scenarios, for all time periods. The high emission scenario (RCP8.5) shows the largest range in simulated NEE results (**Figure 2**). In projections across Finland from both ecosystem models and all climate scenarios, the median removal of CO<sup>2</sup> increased with 48 and 39% by mid-century and the end of the century, respectively. The corresponding simulated median values were NEE –74.1 and –66 gC m−<sup>2</sup> yr−<sup>1</sup> for the period 2041–2070 and 2071–2100 (**Supplementary Table 1**).

#### Increasing Risk for Drought and Decreasing Opportunities for Winter Harvest

Warming temperatures, changing precipitation patterns (**Figure 1**) and increasing forest growth (**Figures 2**, **3**) altered the water availability in the JSBACH simulations. The risk for drought was slightly increasing for all scenarios. Much of the change is due to earlier snow melt and thus possibility for earlier drought events. Decreasing precipitation was projected for some climate models in winter for all time periods, and also in summer for the end of the century (**Figure 1**). The mean number of dry summer days was approximated to around 4 days for the reference period. By the end of the century (2071–2100), the mean simulated number of dry summer days increased to about 15 days in the south and to 10 days in the north (**Supplementary Table 2**). In southern Finland, simulations for the mid-century gave even higher number of dry days (23). In the JSBACH simulations, the number of soil frost days was decreasing in all simulations (**Supplementary Table 2**). The changes were largest in southern Finland under the high forcing scenario (RCP8.5), from 134 days in 1981–2010 to 25 days in 2071–2100. This means the opportunities for winter harvest on drained peatlands are decreasing with all scenarios, more in the south than in the north.

#### Regulating and Cultural Services: Maintaining Habitats, Nature Tourism

Estimates of the start dates of the vegetation active period on the basis of webcam images were similar to those simulated by the JSBACH model (see section "Webcam Network" in **Supplementary Material** and **Supplementary Figure 6**). The vegetation active period ended slightly later according to the JSBACH results than based on the webcam images. VAPstart in Finland occurred in median on 28 April in all simulations for the present period (1980–2010) (**Supplementary Table 1**). In comparison, for the period 2001–2017, VAPstart in Finland was determined from satellite observations (SYKE, 2018). The satellite-derived median VAPstart occurred on 13 April in evergreen coniferous forest. JSBACH simulations included all PFTs whereas PREBAS simulations included forested areas only. Thus, the simulated present day VAPstart seems to be slightly ahead of satellite observations with a median VAPstart on 5 May for all forest types (**Figure 4**). In comparison to reference conditions, the vegetation active period started (VAPstart) about 2 weeks earlier in the mid-century, and 3 weeks earlier toward the end of the century. This means spring would start earlier in April than at present. The vegetation active period ended (VAPend) by the end of the century in early October, compared to late September in the reference conditions (**Figure 2** and **Supplementary Table 1**). In the climate change simulations, the length of the vegetation active period (VAPlength) increased from 162 days with about 20 days in the mid-century and with 30 days by the end of the century (**Figure 2** and **Supplementary Table 1**).

#### DISCUSSION

#### Provisioning Services: GPP and Supply of Harvestable Wood

We aggregated the results from the two ecosystem models in order to account for the uncertainty inherent in the use of different assumptions, e.g., JSBACH simulations do not account for forest management. In this paper we focus on presenting the results on the whole-country level. Data on grid-based results of the biophysical variables for each ecosystem model and each

climate scenario separately are available online (Metadata, 2018a,b). These online metadata descriptions include links to the Climateguide.fi service, where maps of the grid-based results can be found by selecting the category "Terrestrial ecosystems," and the variable of interest "GPP," "NEE," etc. The maps are displaying results from each ecosystem model (JSBACH, PREBAS), climate model, and climate forcing level RCP separately. Seasonal patterns of GPP simulated by PREBAS and JSBACH aggregated to the national level were almost identical, and temporal trends of annual totals were also parallel. This is remarkable in light of earlier results, as ecosystem models tend to vary quite much in their GPP estimates globally and in the northern latitudes (Anav et al., 2013; Shao et al., 2013). However, it has to be noted that the current version of the JSBACH photosynthesis model has been modified for boreal conditions using partly the same empirical evidence as PREBAS (e.g., Mäkelä et al., 2004; Kolari et al., 2007).

The climate change simulations of forest growth did not account for growth restrictions due to limited access to nutrients, mainly nitrogen or phosphorus. Forest growth in higher CO<sup>2</sup> concentrations may become nutrient limited in nutrient deficient sites (e.g., Wieder et al., 2015). As PREBAS projections of future growth were made assuming current forest practices continue in the future (Sved and Koistinen, 2015; Minunno et al., 2019), here no provision was made for potential adaptive responses by the forest owners, e.g., aiming for more variation in stand structure, stand age or silvicultural management (Blennow, 2012; Montoro Girona et al., 2017), introducing new tree species or varieties (Laakkonen et al., 2018; Sousa-Silva et al., 2018), or striving to maximize positive synergies between ecosystem services (Kuuluvainen and Gauthier, 2018). Current total volume of growing tree stock in Finland is 2.5 billion m<sup>3</sup> with a total annual increment of 107 million m<sup>3</sup> , which corresponds to annual average 6.8 and 3.2 m<sup>3</sup> ha−<sup>1</sup> in southern and northern Finland, respectively. Total volume of roundwood harvested from forests was a record high 72.4 million m<sup>3</sup> in 2017, about 17% higher than the average of the preceding ten-year period (Vaahtera et al., 2018). Current forest policies in Finland aim to increase the annual cut to over 80 million m<sup>3</sup> in the next decades. Our simulations indicate that these plans could be feasible in a changing climate, however, increased cutting levels would also imply a lower growing stock than that in our simulations which assumed continuing the current harvest intensity. This would have further implications on C sequestration that need to be explored separately. We did not study the impact of climate change and cutting regimes on biodiversity indicators, such as the amount of dead wood, but others have found significant effects of bioenergy extraction on dead wood levels (e.g., Hof et al., 2018) and conflicting objectives of wood production and biodiversity conservation (Mönkkönen et al., 2014; Angelstam et al., 2019). In their comprehensive review on the retention approach for forestry, Lindenmayer et al. (2012) argued that the practice of permanently retaining significant elements of the original forest is crucial for maintaining multiple forest values, such as biodiversity and carbon stocks. Recently, partial harvest methods were found adequate for regeneration for sustainable forest management in Canadian boreal forests (Montoro Girona et al., 2017, 2018).

#### Regulating Services Avoided CO<sup>2</sup> Forcing

Regarding NEE, JSBACH and PREBAS models produced fairly different patterns especially under climate change – PREBAS predicted more negative NEE values than JSBACH (MONIMET,

2017; Peltoniemi et al., unpublished results). We assume that the differences are caused by different approaches to describing tree growth and management in these models, JSBACH assuming steady-state conditions and PREBAS accounting for forest growth and management explicitly. The harvests lead to a drain of stemwood from the forest which reduces the amount of C entering the soil as coarse woody debris, and therefore reduces the rate of heterotroph respiration from the soil. In our presentation, the NEE projections aggregate results from these models with different assumptions concerning vegetation and forest management, and the variance of the results reflect uncertainties in process descriptions and parameterizations. Forests remain a C sink in our simulation results.

It has been shown that future development of NEE varies much according to ecosystem model and applied climate forcing. According to Ahlström et al. (2012), simulations with a dynamic global vegetation model (LPJ-GUESS), forced by 18 CMIP5 climate projections, showed net release by 2100 for the boreal region. However, the projections were not in full agreement with each other. The magnitude of the release varied, and even turned to uptake in some simulations. Further, according to an ensemble of CMIP5 models both uptake and release of C by land ecosystems is accelerated in the twenty-first century (Shao et al., 2013). The boreal latitudes could become a major C sink by 2100 in many models. It should also be recognized that there is a trade-off situation between the ES supply of "harvestable wood" and "avoided radiative forcing." According to calculations by the Finnish Climate Change Panel, meeting the Finnish targets for implementing the Paris climate change agreement would, in addition to stringent GHG emission reductions, also require a large increase in the C sinks (Finnish Climate Change Panel, 2018). Similar measures would be required also at the global level (Rockström et al., 2017; Rogelj et al., 2018). Intensification of biomass removals from forests may invoke harmful impacts on forest productivity, biodiversity, soil quality, and climate change mitigation potential (Aherne et al., 2012; Forsius et al., 2016; Mäkelä et al., 2016; Soimakallio et al., 2016; Vanhala et al., 2016; Eyvindson et al., 2018).

#### Increasing Risk for Drought and Decreasing Opportunities for Winter Harvest

In boreal and northern conditions, dry and warm summers have been found to increase drought-related symptoms in trees, such as defoliation or discoloration (Muukkonen et al., 2015), limit tree radial growth (Aakala and Kuuluvainen, 2010) and contribute to severe disturbances (Aakala et al., 2011). Satellitederived estimates of vegetation growth trends indicate a clear negative response to recent dry summer conditions (Piao et al., 2011). Shorter periods of frozen soil increase wind damages to forest, as the trees are less firmly anchored in non-frozen soil. Wind-thrown trees may also serve as starting points for bark beetle outbreaks (Pukkala, 2018). In boreal forests, timber is mainly harvested with heavy machinery, which can operate only if the soil has a sufficient bearing capacity (Suvinen, 2006). Higher and more variable air temperatures in winter, coupled with longer and more frequent rain events, may cause pronounced edge effects by skidding trails (Montoro Girona et al., 2016), or lead to periods during which the sites on drained peatlands are not accessible (Kokkila, 2013).

#### Regulating and Cultural Services: Maintaining Habitats, Nature Tourism

For regulating ES, earlier start of vegetation active period means earlier opportunities for birds, insects and other species, but is also linked to the risk for coincident occurrences of cold spells that may be detrimental. For cultural ES, climate warming is expected to lead to earlier opportunities for nature tourism linked to the coming of spring, such as bird watching and observing shoot growth, bud bursting and flowering. However, potential detrimental long-term impacts of climate change on species diversity and conservation are also well recognized (e.g., Thomas et al., 2013; Virkkala et al., 2013; Heikkinen et al., 2015; Virkkala, 2016; Cadieux et al., 2019). For regulating ES, later end of vegetation activity may mean decreased leaching of N in autumn. For provisioning and cultural ES, later end of vegetation may be associated with improved opportunities for the growth and picking of mushrooms and berries, although the success of these species is mostly regulated by local weather conditions.

In terms of regulating ES, this means that there is a positive impact on the reproduction and survival of birds, insects and other species that are dependent on forest vegetation activity. There might be negative impacts due to mismatch of animal production and food supply, or better match between insects and host plants and resulting damages that could alter ecosystem composition (Foster et al., 2013; Pureswaran et al., 2015). The spruce bark beetle (Ips typographus), a serious pest in boreal forests may benefit from warmer temperatures (Annila, 1969) and earlier occurrence of spring (Lange et al., 2006; Zang et al., 2015), increasing the risks for declining forest health (Blomqvist et al., 2018). With the lengthening of the vegetation active period, there is also the option for spatial migration of plant and animals to more northern areas (e.g., Jeganathan et al., 2014).

Summertime nature tourism in the Nordic countries, e.g., for fishing, biking, hiking, kayaking, bird and animal watching, was perceived as interesting activities by respondents in an international survey (MEK, 2010). According to the Finnish national outdoor recreation inventory, almost 70% of annual nature visits were made during the 5 months from May to September, the most popular outdoor recreation activities being walking, swimming, spending time in nature and at a recreational home, picking wild berries, cycling, boating, fishing, cross-country skiing and mushroom picking (Neuvonen and Sievänen, 2011). According to Lankia et al. (2015), the overall value of recreational nature visits was considerable in comparison to other land uses. It is likely that an increase in VAPlength would be beneficial to human society. It should, however, be noted that as the increase in VAPlength may be accompanied by increased probability of spring and autumn rains (Ruosteenoja et al., 2016), the recreational value may not be realized. Here, we do not report simulated future changes in number of snow cover days in response to climate change. It is clear, however, that the occurrence of earlier

start of spring and later beginning of autumn season are coupled to shorter winter periods, leading to losses as regards the opportunities for winter tourism. Cross-country skiing is an important recreational activity in Finland (Neuvonen and Sievänen, 2011) and Neuvonen et al. (2015) presented an interactive tool to study the implications of climate change for cross-country skiing. A considerable proportion of recreational nature visits are made to nature conservation areas, and in the case these visits constitute an increased pressure on the nature conservation areas, the goals of conservation and recreation may be conflicting. It is more likely, however, that the goals of forestry and conservation are colliding. The recreational experience is not independent of the maintenance of habitats, and we are aware of the risk of double attribution in estimating quantitative values for these services, e.g., in monetary terms. It was, however, beyond the scope of our work to assess quantitative values for the services we have studied. In the case of the qualitative analysis carried out here, we think we are justified to use the vegetation active period characteristics to illustrate both the potential for habitat maintenance and recreation opportunities.

For the near decades, even in mid-century, the three emission levels (RCP) gave broadly similar projected temperature increase and change in precipitation on the annual and country-wide level (**Figure 1**). Different climate models, however, gave varying responses especially for seasonal changes. Extreme warming (>5 ◦C) was projected by some climate models by the end of the century for the high emission scenario. Ruosteenoja et al. (2016) call the late century response to RCP8.5 "an alarm signal, demonstrating the furious climatic changes that would be expected if the mitigation of greenhouse gas emissions were totally neglected." Although spatially explicit data have been used to drive the ecosystem models, the results we report here are mainly estimates aggregated over the whole country. We realize that important detailed information is lost in such an aggregation, especially concerning tradeoffs between ES, and the spatial distribution and balance of ES supply and demand (Potschin and Haines-Young, 2013; Vauhkonen and Ruotsalainen, 2017). The value of boreal forest future ES has also not been addressed in this paper. We have qualitatively described future ES on the basis of simulated biophysical quantities such as forest growth and CO<sup>2</sup> exchange with the atmosphere. Analyzing current and future valuation of ES by forest owners and other stakeholders would be needed in order to introduce boreal forest ES into a natural capital framework (Schröter et al., 2016; Lai et al., 2018), or to assess forest carbon policies (Pohjola et al., 2018). Such considerations are, however, beyond the scope of our work.

#### CONCLUSION

In our analysis, climate change impacts on key boreal forest ES are both beneficial and detrimental. Increased GPP is beneficial, leading to increased C sequestration, as well as increased forest growth and improved supply of harvestable wood. Warmer temperatures lead, however, also to higher ecosystem respiration, which in some instances cause increasing fluxes of CO<sup>2</sup> from vegetation to atmosphere (NEE > 0). Although GPP is steadily increasing in our simulations, there are times in early and late 21st century when GPP is smaller than ecosystem respiration and annual NEE becomes positive. Here, the assumptions concerning future conditions included only different climate change scenarios, not considering any adaptation measures, e.g., regarding changing forestry practices. Thus, future forest growth simulations with PREBAS reflect a continuing business as usual response of the forestry sector. This means the current forestry practices are assumed to continue, and possible shifts toward higher harvest rates to accommodate increased demand for bioenergy have not been considered here. Possible increased risk of forest damage was also not accounted for, although increasing probabilities of more frequent insect outbreaks and wind throw damages have been recognized. There is still a large uncertainty in predicting future forest growth, since the models disregard natural disturbances, and are not constrained by potential nutrient limitations (N, P and base cations). These limitations call for further model developments as well as biomass and flux data to inform process based modeling. Our current simulation results likely represent the high end of the future growth estimates. We also noted the need to consider a balanced approach between forest harvesting, C sequestration potential and nature based ecosystem services in forested ecosystems.

While some of the studied ES are expected to improve with climate change (biomass growth, nature tourism), the projected decrease in number of soil frost days is expected to decrease the opportunities for winter harvest in forestry especially in the south. Future levels of net ecosystem exchange of CO<sup>2</sup> are uncertain; some simulations indicate decreasing radiative forcing levels, while other simulations lead to increasing levels. Forests remain C sinks, however, in all simulations. The risk for summer drought is slightly increasing in the whole country. Climate warming is expected to lead to earlier opportunities for springtime nature tourism, and may improve opportunities for autumn mushroom and berry picking. The opportunities for winter sports are decreasing while hiking opportunities are increasing with earlier occurrences of spring.

Our results indicate the range of uncertainty in future provision of boreal forest ES. The estimates described in this paper may provide input to comprehensive systems for integrated natural capital accounting, as well as sharing such information via hubs such as National Clearing House Mechanism for the Convention on Biological Diversity (Finnish Ecosystem Service Indicators, 2015). Furthermore, the examples in this paper may highlight the potential for advancing ES monitoring, such as improving information on ecosystem variables for instance by high-resolution Earth Observation or novel in situ observation networks such as webcams.

#### DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

### AUTHOR CONTRIBUTIONS

fpls-10-00343 March 22, 2019 Time: 18:51 # 11

MH planned and drafted the manuscript. MH, TA, AnA, AlA, TM, AM, FM, MP, KB, and KR planned the framework. TA and TM provided and interpreted JSBACH model results. AM, MP, and FM provided and interpreted PREBAS model results. IL provided the web interface of the model results. KB contributed revised material on phenology and performed GIS analysis. AnA, IB, KR, PV, and MF contributed and revised material on ecosystem services.

#### FUNDING

We acknowledge the funding contributions and support by the projects LIFE12 ENV/FIN/000409 Climate change indicators and vulnerability of boreal zone applying innovative observation and modeling techniques MONIMET; the Academy of Finland

## REFERENCES


Strategic Research Council project SRC 2017/312559 IBC-CARBON; and the Academy of Finland Center of Excellence (307331), CARB-ARC (285630), and OPTICA (295874).

# ACKNOWLEDGMENTS

We appreciate the support by two reviewers through their detailed comments to the manuscript. We thank Pekka Vanhala, Finnish Environment Institute, for helpful discussions on the topics of the paper.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.00343/ full#supplementary-material



global-scale terrestrial biosphere models. Glob. Change Biol. 15, 976–991. doi: 10.1093/jxb/44.5.907



from forest condition monitoring and GIS analysis. Boreal Environ. Res. 20, 172–180.



**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 Holmberg, Aalto, Akujärvi, Arslan, Bergström, Böttcher, Lahtinen, Mäkelä, Markkanen, Minunno, Peltoniemi, Rankinen, Vihervaara and Forsius. 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.

# Wood Growth in Pure and Mixed Quercus ilex L. Forests: Drought Influence Depends on Site Conditions

Enrica Zalloni<sup>1</sup> , Giovanna Battipaglia<sup>2</sup> , Paolo Cherubini3,4, Matthias Saurer<sup>3</sup> and Veronica De Micco<sup>1</sup> \*

<sup>1</sup> Department of Agricultural Sciences, University of Naples Federico II, Naples, Italy, <sup>2</sup> Department of Environmental, Biological and Pharmaceutical Sciences and Technologies, University of Campania "Luigi Vanvitelli", Caserta, Italy, <sup>3</sup> Swiss Federal Research Institute WSL, Zurich, Switzerland, <sup>4</sup> Department of Forest and Conservation, Sciences Forest Sciences Center, The University of British Columbia, Vancouver, BC, Canada

#### Edited by:

Jian-Guo Huang, University of Chinese Academy of Sciences (UCAS), China

#### Reviewed by:

Pei-Li FU, Xishuangbanna Tropical Botanical Garden (CAS), China Martina Pollastrini, University of Florence, Italy

> \*Correspondence: Veronica De Micco demicco@unina.it; veronica.demicco@unina.it

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 14 December 2018 Accepted: 15 March 2019 Published: 02 April 2019

#### Citation:

Zalloni E, Battipaglia G, Cherubini P, Saurer M and De Micco V (2019) Wood Growth in Pure and Mixed Quercus ilex L. Forests: Drought Influence Depends on Site Conditions. Front. Plant Sci. 10:397. doi: 10.3389/fpls.2019.00397 Climate response of tree-species growth may be influenced by intra- and inter-specific interactions. The different physiological strategies of stress response and resource use among species may lead to different levels of competition and/or complementarity, likely changing in space and time according to climatic conditions. Investigating the drivers of inter- and intra-specific interactions under a changing climate is important when managing mixed and pure stands, especially in a climate change hot spot such as the Mediterranean basin. Mediterranean tree rings show intra-annual density fluctuations (IADFs): the links among their occurrence, anatomical traits, wood growth and stable isotope ratios can help understanding tree physiological responses to drought. In this study, we compared wood production and tree-ring traits in Quercus ilex L. dominant trees growing in two pure and two mixed stands with Pinus pinea at two sites in Southern Italy, on the basis of the temporal variation of cumulative basal area, intrinsic water use efficiency (WUE<sup>i</sup> ), δ <sup>18</sup>O and IADF frequency in long tree-ring chronologies. The general aim was to assess whether Q. ilex trees growing in pure or mixed stands have a different wood production through time, depending on climatic conditions and stand structure. The occurrence of dry climatic conditions triggered opposite complementarity interactions for Q. ilex growing with P. pinea trees at the two sites. Competitive reduction was experienced at the T site characterized by higher soil water holding capacity (WHC), lower stand density and less steep slope than the S site; on the opposite, high competition occurred at S site. The observed difference in wood growth was accompanied by a higher WUE<sup>i</sup> due to a higher photosynthetic rate at the T site, while by a tighter stomatal control in mixed stand of S site. IADF frequency in Q. ilex tree rings was linked to higher WUEi, thus to stressful conditions and could be interpreted as strategy to cope with dry periods, independently from the different wood growth. Considering the forecasted water shortage, inter-specific competition should be reduced in denser stands of Q. ilex mixed with P. pinea. Such findings have important implications for forest management of mixed and pure Q. ilex forests.

Keywords: mediterranean, tree rings, water use efficiency, δ <sup>13</sup>C, δ <sup>18</sup>O, basal area increment

# INTRODUCTION

fpls-10-00397 March 29, 2019 Time: 18:52 # 2

Wood growth in mixed vs. pure stands depends on several factors, such as species composition, stand density, age and climatic conditions (Forrester, 2014). During the development of mixed stands, there might be changes in the dominance of species with different growth and physiological strategies (Forrester, 2015). Interactions between species could be ascribed to competition, with a negative effect of one species on the other, to competitive reduction, when inter-specific competition is less than intra-specific one because of a differentiation in resource use strategies, or to facilitation, with a positive effect of one species on the other (Forrester, 2014). Many studies have shown that mixed stands under stressful conditions (e.g., very high stand density, poor water supply, drought or nutrient shortage), are more productive than pure ones (Amoroso and Turnblom, 2006; Erickson et al., 2009; Pretzsch et al., 2013a,b; del Río et al., 2014), following the assumption of the stress gradient hypothesis (SGH). SGH hypothesis suggests that facilitation is favored when the environmental conditions become harsher (Bertness and Callaway, 1994). However, this is not always the case and mixed stands are not always better adapted to climate constraints if compared to monospecific stands. Complementarity, which is the set of competition and facilitation interactions possibly occurring in mixed and pure populations, may show variations due to climatic factors depending on: the different species reactivity to stand density (Forrester et al., 2013), tree size (Forrester, 2015), site conditions (Binkley, 2003; Pretzsch et al., 2010; Coates et al., 2013; Dieler and Pretzsch, 2013), microclimatic differences (Lebourgeois et al., 2013), and the severity of climatic extremes.

The Mediterranean region is foreseen to be strongly affected by global warming, leading to enhanced drought stress for trees in many ecosystems (Giorgi, 2006; Somot et al., 2008; IPCC, 2017). The increased intra-annual frequency and duration of drought periods in the Mediterranean Basin may lead to changes in water use efficiency (WUE) depending on the species, stand density, tree size and age, and growth rate (Brienen et al., 2017). The latter is reported to scale positively with WUE (Huxman et al., 2008). However, fast-growing trees in mixed stands could suffer from drought more than slower growing trees in monocoltures since they generally use more water (Law et al., 2002; Schume et al., 2004; Forrester, 2015). Tree responses to the changing environmental conditions can be reconstructed with the study of tree-ring features in chronologies of tree-ring width, anatomical traits or stable isotope composition (McCarroll and Loader, 2004; Cufar, 2007 ˇ ; Fonti et al., 2010). The combination of carbon and oxygen stable isotope analysis with tree-ring growth provides information about tree ecophysiological processes in response to stress, suggesting which physiological process, namely carbon uptake or water loss, prevailed in determining the variation in WUEi (Scheidegger et al., 2000), expecially in severely water-limited ecosystems (Gessler et al., 2014; Altieri et al., 2015; Moreno-Gutiérrez et al., 2015; Battipaglia et al., 2016b). Grossiord et al. (2014a,b) found that the stand-level δ <sup>13</sup>C declined with increasing diversity in temperate beech and thermophilous deciduous forests but not in hemiboreal, mountainous beech and Mediterranean forests. Within the Mediterranean region, studies analyzing the complementarity effects between mixed and pure stands are scarce. Grossiord et al. (2014c) found that Quercus cerris L. trees did not reduce transpiration in response to drought when growing in pure stands, but significantly reduced transpiration and increased WUE<sup>i</sup> in mixed stands with Quercus petraea (Mattuschka) Liebl. Battipaglia et al. (2017) showed a higher wood productivity and WUE in mixed stands of Quercus robur L. and Alnus cordata Loisel. in comparison with Q. robur pure stands, due to the positive N-fixation effect of A. cordata. Understanding which factors drive inter- and intra-specific interactions under a changing climate is necessary when managing mixed and pure stands, since one of the priority in forestry is to acquire knowledge on the capability of different forest ecosystems to adapt to short- and long-term climatic variability (Brooker, 2006), especially in so called climate-change hot spots such as the Mediterranean. Quercus ilex L. forests widely occur throughout the Mediterranean basin, both in pure stands or in mixed forests with Mediterranean pines such as Pinus pinea L. (Terradas, 1999), differing in light demand, root system and physiological strategies in response to drought. It is still unknown whether mixed stands would be more capable to acclimate to forecasted increase in intra-annual climate variability in the Mediterranean, if compared to pure stands.

Mediterranean trees often form peculiar anatomical traits in tree rings called intra-annual density fluctuations (IADFs), which have been linked to intra-annual frequency of dry periods (De Micco et al., 2016). They have been considered either an hydraulic adjustment of trees to drought or a strategy to take advantage of favorable conditions of growth after a drought event (Battipaglia et al., 2016a). Finding the link between IADF occurrence and facilitation or competition effects at different sites, under different micro-climatic conditions, may be useful to add insights on the ecological role of these tree-ring traits. In this study, we aimed to (1) analyze the dynamics of complementarity effects of Q. ilex dominant trees growing in a pure and in a mixed stand with P. pinea at two study sites differing for tree age, stand density, slope and soil characteristics, on the basis of tree-ring growth and stable isotope ratio variations, (2) analyze the different tree growth response to climatic factors, (3) find the link between wood anatomical recurrent traits, such as IADFs and tree growth in the different study sites. In order to reach these aims, we investigated the temporal variation of cumulative basal area, intrinsic WUE (WUEi) assessed through δ <sup>13</sup>C and δ <sup>18</sup>O in tree rings (Moreno-Gutiérrez et al., 2012; Altieri et al., 2015; Battipaglia et al., 2016b), in each pure stand in comparison with mixed ones, calculating annual indexes of complementarity. We hypothesize that: (1) Q. ilex tree growth is higher in pure than in P. pinea-mixed stands, accompanied by a higher WUE<sup>i</sup> ; (2) precipitation is the main factor influencing Q. ilex tree growth at all the Mediterranean study sites; (3) IADFs occur where wood growth is lower, because linked to stressful conditions of growth rather than to favorable ones.

#### MATERIALS AND METHODS

fpls-10-00397 March 29, 2019 Time: 18:52 # 3

#### Study Sites

The study sites are located within the Mediterranean region, in the Vesuvio National Park, southeast from Naples, Southern Italy. The two sites are located on two opposite slopes, one in the southwest-faced "Tirone Alto-Vesuvio" Forest State Reserve and the other on the northeast-faced Mount Somma slopes, differing by stand density, slope, aspect (**Table 1**) and soil characteristics. In each site, a pure Q. ilex stand (TP – Tirone Pure stand; SP – Somma Pure stand) and a mixed Q. ilex-P. pinea stand (TM – Tirone Mixed stand; SM – Somma Mixed stand) with comparable age of trees, soil and stand characteristics were sampled (**Figure 1**). The stands are forests and P. pinea trees were planted.

Both pure stands (TP and SP) are dominated by Q. ilex with an understory consisting in Q. ilex trees and the sporadic presence of Robinia pseudoacacia L., a non-native invasive species. Both mixed stands (TM and SM) are covered by P. pinea dominant trees with a Q. ilex understory and the sporadic presence of R. pseudoacacia. SP and SM stands are characterized by smaller trees and have higher stand density and steeper slope than TP and TM (**Table 1**). Moreover, P. pinea trees of the SM stand are taller than Q. ilex trees, while P. pinea and Q. ilex trees have a similar height at TM (**Table 1**). At the S site, total stand density and canopy cover are higher at the pure than at the mixed stand, while the slope is less steep at the pure than at the mixed one. At the T site, total stand density and canopy cover are lower while slope is steeper at the pure than at the mixed stand.

As regards soil, there were no significant differences in water content (WC), available water capacity (AWC) and water holding capacity (WHC) between the mixed and pure stand at each site. However, there were significant differences between the two sites in terms of AWC and WHC, with both the parameters higher at the T site (AWC mean value = 33.07 ± 12.87%; WHC mean value = 24.27 ± 7.33%) in comparison to S site (AWC mean value = 20.27 ± 5.33%; WHC mean value = 16.71 ± 3.79%) (P < 0.05). WC, AWC, and WHC were determined in autumn by taking six samples per site (three samples per each stand) and following standard procedures (USDA, 1996).

The climate is Mediterranean with dry summer and mild winter. Climate data of minimum, maximum and mean monthly temperature and total monthly precipitation from the nearest meteorological stations were interpolated and compared to the CRU TS3.23 gridded dataset at 0.5◦ resolution data (Harris et al., 2014). Since the correlation between the two data series was significant (as shown in Zalloni et al., 2018a), we used the CRU climate data for the analyses. Mean annual temperature and precipitation of the period 1985–2005 selected for statistical analysis are 16.4◦C and 710 mm, respectively (**Figure 1**). The wettest month is November, with an average of 114 mm of cumulative precipitation, while the driest month is August, with an average of 24 mm of cumulative precipitation and the highest temperature of 30◦C. The lowest mean temperatures are recorded in January, with an average of 9 ◦C (**Figure 1**). A dry season lasts from the middle of May to the end of August.

#### Tree-Ring Growth Analysis

Two cores per tree were extracted at breast height from 20 dominant Q. ilex trees per stand in the "Tirone Alto-Vesuvio" site (T) and from 14 dominant Q. ilex trees per stand in the Mount Somma site (S). Being the sites in a Natural Park, the minimum number of trees to get a good EPS value was sampled. The cores were collected during September 2015 at the mixed sites, while during September 2016 at the pure ones. The number of cored trees per site is different because of the different availability of dominant trees. The cores were air dried, mounted on wooden supports and sanded. A Leica MS5 light microscope (Leica Microsystems, Germany) fitted with a LINTAB measuring system (Frank Rinn, Heidelberg, Germany) was used to measure ring-width chronologies with a resolution of 0.01 mm. After being visually cross-dated, tree-ring width chronologies were statistically checked with the TSAP-Win (Time Series Analysis and Presentation; Rinntech) and COFECHA (Holmes, 1983) softwares. Mean tree-ring width chronologies were developed per each stand. The Dendrochronology Program Library within the


TABLE 1 | Coordinates, altitude and structure features of the four selected stands.

TP and TM are the pure and the mixed stand, respectively, at the "Tirone Alto-Vesuvio" site, while SP and SM are the pure and the mixed stand, respectively, of the Mount Somma site.

software R (dplR; Bunn, 2008, 2010) was used to calculate the expressed population signal (EPS) (Wigley et al., 1984), the mean RBAR (that is the is the mean correlation coefficient among treering series) and the signal-to-noise ratio (SNR) in order to assess chronology quality (**Table 2**).

In order to compare radial growth of the dominant trees between stands, correctly dated tree-ring width chronologies were converted into tree basal area increment (BAI) chronologies with the following formula:

$$BAI\_t = \pi R\_{(t^2)} - \pi R\_{(t-1^2)},\tag{1}$$

where R<sup>t</sup> and Rt−<sup>1</sup> are the stem radius at the end and at the beginning of the annual increment, respectively, and BAI<sup>t</sup> is the annual ring area at year t. Cumulative mean basal area was then calculated for each stand summing the average basal area. BAI instead of ring-width time series were chosen because they reduce tree-size and age effect on growth trends, keeping the high and low frequency signals of tree-ring width series at the same time (Tognetti et al., 2000; Biondi and Qeadan, 2008).

#### Stable C and O Isotope Analysis

Five correctly dated cores of Q. ilex without defects per stand were chosen for isotopic analyses. Carbon and oxygen stable isotope analysis were conducted over the common period 1985– 2005 for all the stands, where a change in wood growth was found between pure and mixed stands at both sites. Tree rings were manually split with annual resolution using a scalpel under a dissection microscope, and the derived samples of the five cores per species and per stand were then pooled together in order to maximize sample size. Preliminary analyses showed that comparable results are obtained by using either whole wood or cellulose (Borella et al., 1998; Korol et al., 1999; Barbour et al., 2001; Warren et al., 2001; Loader et al., 2003; Verheyden et al., 2005; Weigt et al., 2015), thus we decided to proceed on whole wood, without any chemical pretreatment. The collected samples were milled with a centrifugal mill, weighted in silver capsules (aliquots of 0.8/1.0 mg) and pyrolyzed at 1450◦C, (PYRO-cube, Elementar, Hanau, Germany). The annual δ <sup>13</sup>C and δ <sup>18</sup>O values of the obtained CO were



TP and TM are the pure and the mixed stand, respectively, at the "Tirone Alto-Vesuvio" site, while SP and SM are the pure and the mixed stand, respectively, at the Mount Somma site.∗Expressed population signal (EPS) is a measure of the common variability in a chronology and it is commonly acceptable for value >0.85; ∗∗RBAR is the mean correlation coefficient among tree-ring series, ranging from −1 to +1 (the higher the value, the stronger is the underlying common signal); ∗∗∗The signal-to-noise ratio (SNR) informs about the ratio between the signal (short-term variation) and the noises (long-term variation) contained in chronologies.

determined simultaneously by a Delta Plus XP isotope ratio mass spectrometer (ThermoFinnigan MAT, Bremen, Germany) via a pyrolysis unit by a ConFlo III interface (ThermoFinnigan MAT). A subset of samples that covered the whole range of the expected δ <sup>13</sup>C values was measured again via oxygen combustion with an EA1110 elemental analyzer (CE Instruments, Milan, Italy) coupled to a Delta-S isotope ratio mass spectrometer (ThermoFinnigan MAT), in order to make a correction of the δ <sup>13</sup>C values. The δ <sup>13</sup>C signal obtained by pyrolysis is dampened because of "memory effects" compared to the more usually measured one obtained by oxygen combustion (Woodley et al., 2012). The formula used to correct the pyrolysis δ <sup>13</sup>C data was the following: δ <sup>13</sup>Ccorr = 1.2526 × δ <sup>13</sup>Cpyro + 5.0032, where δ <sup>13</sup>Ccorr is the corrected final δ <sup>13</sup>C value and δ <sup>13</sup>Cpyro is the value measured by pyrolysis and corrected with internal standards. Furthermore, δ <sup>13</sup>C values were corrected for the Suess effect, which is a shift in the atmospheric concentrations of carbon isotopes due to increasing fossil-fuel derived CO<sup>2</sup> (Keeling, 1979). The corrected series were used for the subsequent statistical analyses.

#### WUE<sup>i</sup> Calculation From δ <sup>13</sup>C

Isotopic <sup>13</sup>C-fractionation during CO2-fixation can be calculated as:

$$
\delta^{13}\mathcal{C}\_{\text{plant}} = \delta^{13}\mathcal{C}\_{\text{air}} - a - (b - a) \* \left(\mathfrak{c}\_{\text{i}}/\mathfrak{c}\_{\text{a}}\right),
\tag{2}
$$

where δ <sup>13</sup>Cair is the carbon isotope ratio of atmospheric CO2, a is the fractionation factor due to CO<sup>2</sup> diffusion through stomata (4.4%), b is the fractionation factor due to the Rubisco enzyme during photosynthesis (27.0h), <sup>c</sup><sup>i</sup> is the intercellular leaf CO<sup>2</sup> concentration, c<sup>a</sup> is the atmospheric CO<sup>2</sup> concentration and δ <sup>13</sup>Cplant is the carbon isotope ratio of plant organic matter, e.g., in tree-rings. WUE<sup>i</sup> chronologies for each stand were then calculated following the formula reported by Ehleringer and Cerling (1995):

$$WUE\_{\mathrm{i}} = A/g\_{\mathrm{s}} = (c\_{\mathrm{a}} - c\_{\mathrm{i}})/1.6,\tag{3}$$

where A is the photosynthetic rate, g<sup>s</sup> is the stomatal conductance and 1.6 is the ratio of diffusivity of water and CO<sup>2</sup> in the atmosphere. This can be solved as c<sup>i</sup> is known from Eq. (2). In particular, the following formula was used:

$$WUE\_{\mathbf{i}} = \left(c\_{\mathbf{a}} - c\_{\mathbf{i}}\right) / 1.6 = \left[c\_{\mathbf{a}} - c\_{\mathbf{a}}\left(\Delta - a/b - a\right)\right] 1/ 1.6$$

$$= c\_{\mathbf{a}} \left[\left(1 - \left(\Delta - a/b - a\right)\right) 1/ 1.6\right],\tag{2}$$

where 1 is the carbon isotope discrimination which represents the difference between δ <sup>13</sup>Cair and δ <sup>13</sup>Cplant, and using Eq. (1) c<sup>i</sup> is equivalent to

$$c\_{\mathfrak{a}}\left[\left(\Delta - a\right)/\left(b - a\right)\right],\tag{4}$$

while c<sup>a</sup> annual values were taken from the NOAA database ( <sup>1</sup>Mauna Loa station). The parameter 1 was calculated as:

$$
\Delta = \left( \\$^{13}\text{C}\_{\text{air}} - \\$^{13}\text{C}\_{\text{plant}} \right) / \left( 1 + \\$^{13}\text{C}\_{\text{plant}} / 1000 \right). \tag{5}
$$

δ <sup>13</sup>Cair values were taken from the ones estimated by McCarroll and Loader (2004) and the measured ones available online<sup>2</sup> , while δ <sup>13</sup>Cplant are the values measured in tree rings of our samples.

#### Complementarity Calculations

In order to assess inter-specific facilitation and competition interactions for comparison of wood growth, WUE<sup>i</sup> and δ <sup>18</sup>O of Q. ilex in pure and mixed stands, an annual index of complementarity was calculated for the period 1985-2005 for each site with the following formula (Forrester, 2015; Battipaglia et al., 2017):

$$\text{Complementarity} \left( \% \right) = \left[ \left( X\_{\text{M}} - X\_{\text{P}} \right) / X\_{\text{P}} \right] \* 100,\tag{6}$$

where X is annual basal area, WUE<sup>i</sup> or δ <sup>18</sup>O, M is related to mixed stands and P is related to pure stands. The index is positive when wood growth, WUE<sup>i</sup> or δ <sup>18</sup>O are higher in mixed than in pure stands, while negative when they are higher in pure than in mixed stands.

To compare the two sites, in terms of WUE<sup>i</sup> , δ <sup>18</sup>O, and BAI, characterized by different number of samples, U-test was used through SPSS 13.0 statistical package (SPSS Inc., Chicago, IL, United States) (Spiegel, 1975).

#### IADF Frequency Analysis

Intra-annual density fluctuation occurrence was detected within the rings of all the Q. ilex dated cores under a reflected light microscope. IADFs were identified by detecting variations in cell lumen area, frequency and wall density different from the "standard" transition from earlywood to latewood of Q. ilex described in Wheeler (2011), as found in Campelo et al. (2007) and defined in Zalloni et al. (2018b) (**Figure 2**). Relative annual IADF frequency chronologies of each stand were calculated as the ratio between the number of cores with an IADF and the total number of cores for each year. Stabilized annual IADF frequency chronologies were then calculated according to Osborn et al. (1997) as f = F ∗ n <sup>0</sup>.<sup>5</sup> where F is the relative IADF frequency value and n is the total number of cores for each year, in order to stabilize the variance overcoming the problem of the changing sample depth over years. A percentage of IADF occurrence was calculated for each stand as the number of rings with IADF on the number of total rings for the period 1985–2005.

#### Climate Analysis

The period 1985–2005 was selected for statistical analysis in order to match BAI data with isotope ones for comparisons, considering that in those years (more specifically 1996– 1997) a change in wood growth was found between pure and mixed stands at both sites. In order to investigate the relations between growth traits and climate parameters, a Pearson's linear correlation function analysis (P < 0.05) was implemented between cumulative mean annual BAI, WUE<sup>i</sup> , and δ <sup>18</sup>O annual values of the whole study period (1985– 2005) and temperature and precipitation data. A Pearson's

<sup>1</sup>http://www.esrl.noaa.gov/

<sup>2</sup>http://www.esrl.noaa.gov/gmd/

FIGURE 2 | Light microscopy views of cross sections of tree rings of Quercus ilex without (a) and with IADFs (b). Arrowheads point the boundaries of tree rings; arrows point to the IADF. Bar: 100 µm.

linear correlation function analysis (P < 0.05) was also implemented between mean annual BAI of the period 1985– 1996, mean annual BAI of the period 1997–2005 and temperature and precipitation data, in order to investigate whether and what climate factor significantly influenced tree growth at the pure and mixed stands of the two sites, and their ecophysiological responses. Temperature and precipitation data were seasonally grouped from December of the previous year to February of the next year, in order to certainly cover all the season (winter, spring, summer, autumn, and winter again) of the current year which could influence tree-ring growth in Mediterranean species (Cherubini et al., 2003; Vieira et al., 2015; Balzano et al., 2018). The analyses were performed using Excel©.

# RESULTS

#### Tree-Ring Growth, WUE<sup>i</sup> and δ <sup>18</sup>O Trends

The dendrochronological characteristics of Q. ilex tree-ring width chronologies for the four stands are summarized in **Table 2**. Treering chronologies of Q. ilex trees covered the timespan from 1949

to 2014 at the two stands of the T site, while the shorter timespan (from 1966 to 2014) was found at the S site (**Table 2** and **Figure 3**).

The mean annual BAI for the period 1985–2005 of SM stand was the lowest (1063.31 ± 311.26 mm<sup>2</sup> , mean value ± standard error), followed by TP (1257.38 ± 360.67 mm<sup>2</sup> ), SP (1274.51 ± 298.34 mm<sup>2</sup> ), and TM stands (1665.08 ± 407.62 mm<sup>2</sup> ).

The cumulative BAI of wood growth of the dominant trees of the stands revealed an opposite shift in wood growth of pure and mixed stands between the two sites from the year 1997 to 2014 (**Figure 3**). More specifically, a wood growth increase of dominant trees in the mixed compared to the pure stand was recorded at the T site (**Figure 3A**), while the opposite was found at the S site (**Figure 3B**). At the T and S sites, the basal area of Q. ilex accounted, respectively, for the 38.51 and 14.54% of the total basal area of the mixed stand.

At the T site, WUE<sup>i</sup> was significantly higher along the whole study period in the dominant trees of the mixed (mean value = 84.15 ± 2.22 µmol mol−<sup>1</sup> ) compared to the pure stand (mean value = 78.04 ± 2.86 µmol mol−<sup>1</sup> ) (P < 0.05) (**Figure 4A**). At the S site, this applied in the 71.43% of the cases (SM mean value = 79.91 ± 2.37 µmol mol−<sup>1</sup> ; SP mean value = 77.2 ± 2.9 µmol mol−<sup>1</sup> ) (P < 0.05) (**Figure 4B**).

At each site, δ <sup>18</sup>O values were similar between dominant trees growing in pure and mixed stands, while significantly absolute higher values of δ <sup>18</sup>O were found in both pure and mixed stand at the S site (SP mean value = 25.75 <sup>±</sup> 0.49h; SM mean value = 25.87 <sup>±</sup> 0.5h) compared to the T site (TP mean value = 25.04 <sup>±</sup> 0.29h; TM mean value = 25.05 <sup>±</sup> 0.36h) (P < 0.05) (**Figures 4C,D**).

Finally, a Pearson's linear correlation function analysis (P < 0.05) implemented between WUE<sup>i</sup> and δ <sup>18</sup>O values, showed a significant positive correlation only at the SM stand, during the period 1997–2005 (Pearson's correlation value = 0.95).

#### Complementarity Effects

After the year 1992, the complementarity effect analysis pointed out the difference between facilitation/competition interactions of the dominant trees of mixed and pure stands at the two sites. More specifically, from 1992 to 2005, Q. ilex wood growth was higher in the mixed than in the pure stand at the T site, while it was higher in the pure than in the mixed stand at the S site (**Figure 5**).

WUE<sup>i</sup> was found to be substantially higher in the dominant trees in mixed than in pure stands during the whole study period at both sites (**Figure 6A**).

Finally, the wood of the mixed stand is more enriched in δ <sup>18</sup>O than the pure one at the S site in most years, while an unclear pattern was shown for the complementarity index based on δ <sup>18</sup>O values of the T site (**Figure 6B**).

#### IADF Frequency

The highest percentage of occurrence of IADFs in tree-rings of dominant Q. ilex trees, for the period 1985–2005, was found in both the stands of the S site. In particular, the highest one was recorded in the mixed stand of the S site (26.41%), followed by the pure one (16.61%), then the mixed stand of the T site (14.76%) followed by the pure one, which showed a very low IADF frequency (1.79%).

#### Climate Influence

Climate analysis with cumulative mean annual BAI of the whole study period did not show significant correlations for any of the stands. The same lack of significant correlation was found for the analysis cumputed for the period 1985–1996. On the contrary, a significant influence of summer (from June to August) and autumn (from September to November) precipitation on cumulative mean annual BAI of the period 1997–2005 was found for all the stands, with higher Pearson's coefficient values (r) for autumn (TP = 0.82; TM = 0.82; SP = 0.85; SM = 0.85) than for summer precipitation (TP = 0.69; TM = 0.71; SP = 0.71; SM = 0.71) (P < 0.05). Climate analysis with WUE<sup>i</sup> and δ <sup>18</sup>O of the whole study period showed that precipitation was the main influencing factor. WUE<sup>i</sup> was positively correlated with precipitation of winter of the previous year (from December of the previous year to February of the current year) and negatively with precipitation of current spring (from March to May) for the TM stand (r = 0.43 and −0.47, respectively), while negatively correlated with autumn precipitation for the SP stand (r = −0.45) (P < 0.05). δ <sup>18</sup>O was negatively driven by autumn precipitation

for all the stands (r = −0.48 for TM; r = −0.57 for SP; r = −0.44 for SM) except for the TP one, where winter precipitation (from December of the current year to February of the next year) was the driving factor (r = −0.51) (P < 0.05). WUE<sup>i</sup> was also positively correlated with summer temperature for the TM stand (r = 0.61) (P < 0.05).

# DISCUSSION

## Different Q. ilex Wood Growth of Pure and Mixed Stands Within Each Site

Trends in Q. ilex cumulative BAI for the period 1985–2005, suggested different wood growth dynamics of trees growing at the two sites. More specifically, starting from the year 1997, the slope of wood growth trends were inverted between mixed and pure stands within each site; the mean BAI of Q. ilex of the mixed stand at the T site presented higher wood growth compared to the pure stand, while at the S site, the situation was exactly the opposite, with the higher growth of the pure stand than the mixed one. Such an inversion also fits well with the temporal variability observed in the BAI-based complementarity indexes. Summer and autumn precipitation seem to have driven the observed shift in wood growth between dominant trees in pure and mixed stands, since climate correlations with cumulative BAI showed no significant influences until 1996, while summer and autumn precipitation affected tree growth of all the stands starting from 1997. After the year 1996 a decrease in both summer and autumn precipitation accompanied by an increase in temperature is recorded (**Supplementary Figure S1**), which leads to drier conditions that have possibly triggered the complementarity interactions.

In water-limited Mediterranean ecosystems, water availability is the main factor affecting wood growth of Q. ilex, leading

to changes in complementarity interactions, as shown by the high value of the BAI-based complementarity index at the T site associated with high summer precipitation in 1995. Therefore, with the occurrence of drier climatic conditions, at the T site an interaction effect of competitive reduction has been likely experienced, thus resulting in increased Q. ilex wood growth in the mixed than pure stand. This interaction, i.e., facilitation, between different species growing in the same stand supports several studies which found mixed stands with increased wood growth compared to monocoltures, being facilitated by segregation niche (including many processes like the interspecific differences in phenology and physiology that reduce the competition for resources) (Roupsard et al., 1999; Moore et al., 2011; Schwendenmann et al., 2015). Different species growing in mixed stands also likely use different water sources due to differences in root architecture (Cherubini et al., 2003; Schume et al., 2004; Schwendenmann et al., 2015). In our study case, observed phenomena might be linked to different root systems with Q. ilex extracting water from deeper soil layers than P. pinea, or by different water use strategies. Indeed, the anisohydric species Q. ilex resists drought, thus behaving differently from the isohydric P. pinea which avoids drought to save water (Mayoral et al., 2015; Zalloni et al., 2018b). Differences in Q. ilex wood growth in pure and mixed stands, together with the occurrence of drier climatic conditions after 1997, were also found at the S site, even if with an opposite trend: competition outweighted any complementary effects in the mixed stand, with a reduced wood growth in Q. ilex compared to pure stand. Tougher conditions of growth with higher density and slope, and a soil with a lower WHC could have concurred to make Q. ilex more affected by P. pinea competitiveness at the S site. Moreover, stand density is in favor of P. pinea in SM stand. This assumption would be in contrast with the SGH, as well as with the CSR strategy theory, which suggest that facilitation in spite of competition increases between species when site conditions are harsher (Bertness and Callaway, 1994; Grime, 2007). However, it would instead agree with the resource-ratio theory described by Tilman (1985, 2007), which implies that inter-specific competition may be stronger where soil fertility and moisture is lower, as also showed by Trinder et al. (2012) for grassland species and by Coates et al. (2013) for Picea glauca (Moench) Voss associated with Pinus contorta Douglas ex Loudon, 1838. To further support this theory: Hunt et al. (1999) found that facilitation effects decreased with increasing stand density in Eucalyptus nitens H.Deane & Maiden stand mixed with Acacia dealbata Link, 1822 in Australia; del Río and Sterba (2009) showed a lower growth in mixed than in pure stands of Pinus sylvestris L., 1753 and Quercus pyrenaica Willd. in Spain driven by forest density. As a late successional species, Q. ilex at the pure stand at the S site could have increased growth compared to the pine-oak ecosystem (Crow, 1988; Urbieta et al., 2011), moving toward a state of climax community which is better adapted to stressed Mediterranean conditions of growth (Sheffer, 2012).

# Ecophysiological Responses of Pure and Mixed Stands of Q. ilex

Precipitation seems to be the most important limiting factor in controlling Q.ilex WUE<sup>i</sup> . Temperature showed only one significant correlation with WUE<sup>i</sup> , indicating little influence on inter-annual variations in water use efficiency. This is in agreement with several previous studies on Quercus species in the Mediterranean region (Ferrio et al., 2003; Ferrio and Voltas, 2005; Andreu et al., 2008; Maseyk et al., 2011). Autumn and winter precipitation seem to play a key role and represent the typical period for soil recharge in the Mediterranean area (Pumo et al., 2008).

The analysis of the WUE<sup>i</sup> and the δ <sup>18</sup>O together with their relative complementarity indexes, revealed that Q. ilex dominant trees in mixed stands had a higher WUE<sup>i</sup> at similar δ <sup>18</sup>O at both sites; moreover tree rings of dominant trees of both the stands at the S site were more enriched in <sup>18</sup>O than those at the T one. This could indicate a tighter stomatal control in trees growing at the S than the T site, probably linked to its drier conditions with a soil characterized by less WHC thus a higher vapor pressure deficit at the leaf level (Roden and Ehleringer, 2000; Barbour et al., 2002). However, the higher δ <sup>18</sup>O at the S than at the T site could be also due to the fact that trees growing at the S site are younger and may rely on water (mainly precipitation) from upper soil layers, compared to the trees at the T site, which tend to capture less enriched water from deep soil horizons (Dawson et al., 2002). Further, a difference in WUE<sup>i</sup> not associated with a difference in δ <sup>18</sup>O indicates that the high WUE<sup>i</sup> observed in Q. ilex trees of the

mixed stands was due to higher photosynthetic rates rather than lower stomatal conductance (Scheidegger et al., 2000). The processes that improve light and nutrient availability or uptake, which are driven by inter-specific differences in mixed stands, can enhance WUE<sup>i</sup> enabling the plants to increase photosynthesis and make more efficient use of water resources (Forrester, 2015). Kunert et al. (2012) and Schwendenmann et al. (2015) found, respectively, a higher WUE in wood growth, calculated as the ratio between annual wood increment and water use, and a higher diversity in the water uptake depth in mixed stands than in monocoltures of tropical plants due to complementary water use. Forrester et al. (2010)showed an enhanced WUE due to increased N and P availability and light absorption in mixed stands which increased photosynthesis in Eucalyptus globulus growing with Acacia mearnsii. A high WUE<sup>i</sup> could be associated with the high wood growth (Binkley et al., 2004; Binkley, 2012), as we found at the mixed stand at the T site. On the other hand, the higher WUEi found in the dominant trees in the mixed compared to the pure stand at the S site, did not determine an increase in tree growth, in agreement with other studies showing the lack of correlation between WUE<sup>i</sup> and growth (Maseyk et al., 2011; Peñuelas et al., 2011; Battipaglia et al., 2013; Moreno-Gutiérrez et al., 2015), or even warming-induced growth reductions in spite of increasing WUE<sup>i</sup> (Peñuelas et al., 2008; Linares and Camarero, 2012; Granda et al., 2014) for several Mediterranean species during drought periods. Indeed, carbon resources may be allocated to reproduction, to primary growth or just to other tissues such as roots (Dewar et al., 1994). During drought periods carbon investments in below-ground growth are in fact of higher priority than the above ground structures (Kotzlowski and Palladry, 2002) because below-ground growth is favored to guarantee water uptake (Saxe et al., 1998). The decrease in Q. ilex wood growth at the SM stand, although the enrichment of WUE<sup>i</sup> , could be due to reduced stomatal conductance after increasing warmingrelated drought, as Brito et al. (2016) showed for P. canariensis in Spain. Indeed, the positive correlation found between δ <sup>13</sup>Cderived WUE<sup>i</sup> and δ <sup>18</sup>O for dominant trees growing at SM, suggests that g<sup>s</sup> played a significant role (Scheidegger et al., 2000; Moreno-Gutiérrez et al., 2012). According to the observed cumulative BAI reduction, the less favorable growth conditions at the S site, with a higher tree density, a steeper slope and a lower soil WHC of the topsoil, could have concurred to intensify the drought-induced stomatal closure reducing transpiration in the mixed stand, at the price of reducing net assimilation rate, as Brito et al. (2014) showed for P. canariensis at a treeline site with low soil WHC. Q. ilex trees growing in the mixed stand at the S site were probably more affected by competition, given that P. pinea trees presence prevailed. Furthermore, young Mediterranean trees could be more sensitive to limiting climatic conditions than older ones (Rozas et al., 2009, 2013; Vieira et al., 2009; Brito et al., 2016; Zalloni et al., 2016), confirming the hypothesis that the younger Q. ilex trees at the S site suffered from competition with P. pinea rather than being benefited from facilitation. Coherently, WUE<sup>i</sup> and δ <sup>18</sup>O-based complementarity indexes showed that competition prevailed over facilitation for dominant trees in the mixed stand at the S site, where the higher WUE<sup>i</sup> was, however, accompained by higher <sup>18</sup>O ratios compared to trees in the pure stand, suggesting a tighter stomatal control of Q. ilex mixed with P. pinea, which was not shown for Q. ilex growing alone. Conte et al. (2018) found that Fagus sylvatica growing in a mixed stand with P. sylvestris had high WUEi but low productivity not only due to competition but also due to other factors, such as nutrient limitation and forest management.

The highest percentage of IADFs was found in tree rings of Q. ilex dominant plants growing at the S site, where harsher growth conditions, due to the higher stand density, steeper slope, a soil with a lower WHC, and a tighter stomatal control, were observed. A higher frequency of IADFs in tree rings enriched in <sup>18</sup>O at a site with drier growth conditions, compared to a wetter site, was also found in Erica arborea L. tree-rings by Battipaglia et al. (2014), showing that the formation of these peculiar wood anatomical traits is an indicator of the ability of trees to face stressful conditions. Furthermore, within the two sites, more IADFs occurred in tree rings in dominant trees in the mixed stands compared to the ones in the pure ones: the high IADF occurrence could thus be also related to the higher WUE<sup>i</sup> recorded in tree rings of Q. ilex growing at the mixed stands. A high WUE<sup>i</sup> is often influencing the ability of a species to withstand water stress (Battipaglia et al., 2014), and interpreted as an adaptation to drought-prone environments (Raven, 2002). In this view, the higher IADF frequency in tree rings of mixed stands than pure ones, accompanying the higher WUE<sup>i</sup> , is a further support that IADFs should be considered a sign of the ability of a species to avoid stress conditions; in such a way, a positive carbon balance under dry conditions would be maintained through a high WUE, regardless of differences in wood growth.

#### CONCLUSION

The observed differences between wood growth of Q. ilex dominant trees in pure and mixed stands growing at two sites, highlighted the importance of local site conditions in determining the inter- and intra-specific interactions underlying the growth response to environmental variability. The occurrence of drier climatic conditions from 1997 was shown to trigger opposite complementarity interactions for Q. ilex growing with P. pinea trees at the two sites characterized by different soil WHC, stand density and slope. Competitive reduction was experienced at the site with higher soil WHC, lower stand density and less steep slope, while competition became a limiting factor at the other site. WUE<sup>i</sup> increased in trees of both mixed stands at the two sites, but the isotopes showed completely different ecophysiological processes behind tree growth. At the T site, the increase in WUE<sup>i</sup> was mainly related to higher photosynthetic rates that lead to an increase in wood growth. Differently, at the S site, WUE<sup>i</sup> increase was related to a more conservative strategy saving water through stomata closure, thus not leading to wood growth increase. IADF frequency in Q. ilex treerings seemed to be linked to stressful conditions rather than to favorable ones, and could be interpreted as an adaptation aimed at avoiding dry periods, independently from wood growth

differences. The analysis of a combination of different tree-ring parameters helped to find plausible physiological causes of the observed interactions. The findings of this study case highlight the importance of considering site conditions in planning forest management strategies in the view of forecasted increase in water shortage for mixed and pure forests of Q. ilex and P. pinea. Based on our results, at those specific sites, thinnings of P. pinea mixed stands with Q. ilex, where trees are young and stand density is high, could be a good choice to limit inter-specific competition for resources and to promote Q. ilex wood growth. On the contrary, when good conditions of stand density are present, promoting the co-existence of Q. ilex and P. pinea could facilitate complementarity in resource use, while thinning pure Q. ilex stands could limit intra-specific competition. To draw general strategies in planning forest management, further case studies, which also take dominated trees into account, are needed. These would help assessing the influence of stand structure, soil and environmental conditions on complementarity interactions in Mediterranean Q. ilex mixed stands, also analyzing IADF occurrence as an indicator of species capability to avoid stressful conditions.

# DATA AVAILABILITY

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

# REFERENCES


# AUTHOR CONTRIBUTIONS

EZ, GB, and VDM conceived and designed the study. EZ performed sampling and analyses and wrote the main part of the manuscript. GB, VDM, PC, and MS contributed to the analyses. All authors contributed to interpretation of the overall data and wrote specific parts, made a critical revision of the whole text, and approved the submitted version of the manuscript.

#### ACKNOWLEDGMENTS

The authors thank Dr. Angela Balzano, Dr. Francesco D'Ambrosio, and Dr. Enrico Anzano for their help during fieldwork. The authors also wish to thank the Vesuvius National Park for access to the protected areas and logistic support.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.00397/ full#supplementary-material

FIGURE S1 | Summer (from June to August) (a) and autumn (from September to November) (b) total precipitation (in blue) and mean temperature (in red) of the CRU TS3.23 gridded dataset at 0.5◦ resolution data (Harris et al., 2014) for the period 1985–2005. The gray bar indicates a decrease in both summer and autumn precipitation accompanied by an increase in temperature after the year 1996.

stem hydraulic properties in Pinus pinea L. Tree Physiol. 36, 1019–1031. doi: 10.1093/treephys/tpw034





**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 Zalloni, Battipaglia, Cherubini, Saurer and De Micco. 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.

# Compensatory Growth of Scots Pine Seedlings Mitigates Impacts of Multiple Droughts Within and Across Years

#### Hannes Seidel<sup>1</sup> \*, Michael Matiu1,2 and Annette Menzel1,3

<sup>1</sup> Professorship of Ecoclimatology, Department of Ecology and Ecosystem Management, TUM School of Life Sciences Weihenstephan, Technical University of Munich, Freising, Germany, <sup>2</sup> Institute for Earth Observation, EURAC Research, Bolzano, Italy, <sup>3</sup> Institute for Advanced Study, Technical University of Munich, Garching bei München, Germany

#### Edited by:

Giovanna Battipaglia, Università degli Studi della Campania "Luigi Vanvitelli", Italy

#### Reviewed by:

Arun Bose, Swiss Federal Institute for Forest, Snow and Landscape Research (WSL), Switzerland Peter Prislan, Slovenian Forestry Institute, Slovenia

> \*Correspondence: Hannes Seidel hseidel@wzw.tum.de

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 08 October 2018 Accepted: 04 April 2019 Published: 24 April 2019

#### Citation:

Seidel H, Matiu M and Menzel A (2019) Compensatory Growth of Scots Pine Seedlings Mitigates Impacts of Multiple Droughts Within and Across Years. Front. Plant Sci. 10:519. doi: 10.3389/fpls.2019.00519 Tree seedling resistance to and recovery from abiotic stressors such as drought and warming are crucial for forest regeneration and persistence. Selection of more resilient provenances and their use in forest management programs might alleviate pressures of climate change on forest ecosystems. Scots pine forests in particular have suffered frequent drought-induced mortality, suggesting high vulnerability to extreme events. Here, we conducted an experiment using potted Scots pine seedlings from ten provenances of its south-western distribution range to investigate provenancespecific impacts of multiple drought events. Seedlings were grown under ambient and elevated temperatures for 1.5 years and were subjected to consecutive droughts during spring and summer. Growth (height, diameter, and needle) and spring phenology were monitored during the whole study period and complemented by biomass assessments (bud, needle, wood, and needle/wood ratio) as well as measurements of chlorophyll fluorescence and of needle stable carbon isotope ratio. Phenology, growth and biomass parameters as well as carbon isotope ratio and their (direct) responses to reoccurring droughts differed between provenances, indicating genotypic adaptation. Seedling growth was plastic during drought with intra- and inter-annual compensatory growth after drought stress release (carryover effects), however, not fully compensating the initial impact. For (smaller) seedlings from southern/drier origins, sometimes greater drought resistance was observed which diminished under warmer conditions in the greenhouse. Warming increased diameter growth and advanced phenological development, which was (partly) delayed by drought in 2013, but advanced in 2014. Earlier phenology was linked to higher growth in 2013, but interestingly later phenology had positive effects on wood and needle biomass when subjected to drought. Lastly, stable carbon isotope ratios indicated a clear drought response of carbon assimilation. Droughtinduced reduction of the photosystem II efficiency was only observed under warmer conditions but showed compensation under ambient temperatures. Besides these direct drought impacts, also interactive effects of previous drought events were shown, either reinforcing or sometimes attenuating the actual impact. Thus, depending on amount and timing of events, Scots pine seedlings, particularly from southern origins, might be well adapted and resilient to drought stress and should be considered when discussing assisted migration under changing climatic conditions.

Keywords: provenances, growth timing, plasticity, resilience, assisted migration, Pinus sylvestris

#### INTRODUCTION

fpls-10-00519 April 22, 2019 Time: 17:40 # 2

Responses of temperate tree species to water shortage and rising temperatures are manifold, comprising molecular, physiological, and structural responses (Chaves et al., 2003; Niinemets, 2010). On the molecular level, gene expression pathways of molecules that maintain cell turgor and integrity such as abscisic acid, proline, soluble sugars, heat shock proteins or anti-stress proteins are stimulated (Peñuelas et al., 2013). On the physiological level, drought can reduce photosynthetic activity, stomatal conductance, transpiration, sap flow and carbon assimilation, while increasing water use efficiency (Tognetti et al., 1997; Rennenberg et al., 2006; de Miguel et al., 2012; Arend et al., 2013; Klein et al., 2013). Notably, slight to moderate warming has been observed to trigger the opposite response, e.g., an increase in net photosynthesis, stomatal conductance and specific hydraulic conductivity, and a decrease in water use efficiency (Maherali and DeLucia, 2000; Arend et al., 2013; Zhao et al., 2013). Droughtinduced structural responses include leaf shedding, reductions of leaf size, changes in xylem conduit size, declines in leaf/sapwood area ratio, and reductions in total growth along with changes in resource allocation from shoots to roots (Irvine et al., 1998; Peña-Rojas et al., 2005; Eilmann et al., 2009; Martínez-Vilalta et al., 2009; Bryukhanova and Fonti, 2013; Kuster et al., 2013; Taeger et al., 2013a, 2015).

Temperature-growth relationships, on the other hand, are not always uniform: dendroecological studies on Scots pine and Norway spruce have found negative correlations between ring width increment and temperature during the growing season (Pichler and Oberhuber, 2007; Zang et al., 2012), while experimental studies have revealed negative, positive or no effects of temperature on above-ground growth (Olszyk et al., 1998; Arend et al., 2011; Wu et al., 2011; Taeger et al., 2015). Therefore, Saxe et al. (2001) have suggested that temperature responses could be species- or even provenancespecific. Nevertheless, common growth responses to rising temperatures include increasing leaf area and xylem conduits, and decreasing root/shoot ratios and leaf/sapwood area ratios (Maherali and DeLucia, 2000; Way and Oren, 2010).

With future warming along with higher severity and frequency of droughts (Kirtman et al., 2013), climatic pressures are predicted to intensify. Anatomical and physiological changes induced by warmer growing seasons may render trees more susceptible to drought, as reviewed by Way (2013). Since vapor pressure deficit is positively correlated with temperature, warming would furthermore increase the evaporative demand of the atmosphere and thus increase transpiration (Allen et al., 1998; Cinnirella et al., 2002).

The timing of water availability plays a major role for tree growth since foliage, height and diameter show seasonal growth patterns in temperate regions (Anderegg et al., 2013). For example, radial growth of deciduous trees may be most sensitive to early season water deficit whereas that of conifers may also be influenced by late season water deficit (Hanson and Weltzin, 2000). Species that flush early may finish shoot and leaf expansion before drought occurs, while species that exhibit continuous growth and multiple flushes are more strongly affected by seasonal drought events (Bréda et al., 2006). Additionally, past climatic events can have long-lasting effects on tree growth. For instance, current year diameter growth has been shown to be positively correlated with water availability of the previous year (Eilmann et al., 2009; Michelot et al., 2012). Similarly, shoot elongation, leaf number and size depend on the climatic conditions of bud formation in the previous year (Bréda et al., 2006). Furthermore, drought may delay or advance leaf phenology in the following year (Bernal et al., 2011; Misson et al., 2011; Peñuelas et al., 2012; Spieß et al., 2012; Swidrak et al., 2013; Vander Mijnsbrugge et al., 2016). Thus, drought events can have long-lasting impacts on radial, twig and leaf growth that may take several years until pre-drought growth conditions are fully regenerated (Becker, 1989; Orwig and Abrams, 1997; Stribley and Ashmore, 2002; Bréda et al., 2006).

However, compensatory responses following a drought event within and across years are also reported, such as increased radial and shoot growth, additional flushing, earlier leaf development, delayed leaf senescence, and increased cell enlargement rates (Spieß et al., 2012; Taeger et al., 2013b; Balducci et al., 2016; Turcsán et al., 2016; Vander Mijnsbrugge et al., 2016). Scots pine for instance, has the ability to adjust leaf/sapwood area ratio, leafspecific hydraulic conductivity, total leaf area and conduit size in response to drought (Sterck et al., 2008; Martínez-Vilalta et al., 2009). Additionally, acclimation facilitates enhanced resistance to re-occurring stress events (Seidel and Menzel, 2016).

Scots Pine (Pinus sylvestris L.) has a wide-ranging distribution from Spain to Scandinavia to the far east of Russia, covering a number of climatically contrasting environments (boreal to Mediterranean) (Boratynski, 1991). Its general response to water shortage is a drought avoidance strategy characterized by strong stomatal control reducing water loss through the needles (Irvine et al., 1998). This strategy implies reduced photosynthetic carbon gain (Mitchell et al., 2013) and renders Scots pine particularly susceptible to long drought events, even if these events are not severe (Anderegg et al., 2013). Overall, drought has negative effects on Scots pine shoot elongation and radial increment (Irvine et al., 1998; Taeger et al., 2013b). Colonization of fundamentally different climatic regions has

favored local adaptations of Scots pine provenances (Boratynski, 1991; Reich and Oleksyn, 2008), resulting in different responses to drought regarding seedling establishment, mortality, shoot elongation and radial increment, stomatal conductance, and drought resistance (Richter et al., 2012; Taeger et al., 2013a,b, 2015; Seidel and Menzel, 2016; Seidel et al., 2016). Although Scots pine is considered to be a drought resistant species (Ellenberg, 1988), numerous drought-induced mortality events of Scots pine forests have been documented in the last decades (Allen et al., 2010). Recovery from disturbances as well as forest regeneration and persistence might depend on seedling vitality and, furthermore, be crucial to alleviate climate change impacts. Thus, forest management might consider assisted migration of suitable provenances adapted to warmer and/or drier climates of the species' distribution range (Millar et al., 2007; Bussotti et al., 2015).

Climate-growth relationships of Scots pine have been frequently investigated (Oberhuber et al., 1998; Linderholm, 2001; Rigling et al., 2002; Pichler and Oberhuber, 2007; Martínez-Vilalta et al., 2008; Michelot et al., 2012; Sánchez-Salguero et al., 2012, 2015; Zang et al., 2012), while provenance-specific drought impacts on growth are less studied (Cregg and Zhang, 2001; Richter et al., 2012; Taeger et al., 2013a,b, 2015). Knowledge about the influence of seasonal drought events on provenance-specific intra- and inter-annual growth and ecophysiological responses of Scots pine seedlings is lacking in particular.

We therefore ask the following research questions: (1a) How do Scots pine seedlings respond to drought (direct drought effects), (1b) do provenances differ in their drought response, and (1c) are those differences linked to the climate at their origin assuming local adaptation? (2) Are drought impacts additive or can Scots pine seedlings acclimate to drought, e.g., through growth reductions, in order to increase resistance to subsequent drought events within and across years (interactive drought effects)? (3) After drought stress release, do negative effects of drought persist or do Scots pine seedlings recover and even compensate negative effects (carryover drought effects)? In this study, we conducted a seasonal drought experiment using potted seedlings of 10 European Scots pine provenances to evaluate the intra- and inter-annual impact on growth and ecophysiological response across one and a half years. Drought conditions were simulated in the spring and summer of the 1st year and in the spring of the 2nd year. Drought periods were intermitted by wellwatered conditions. Additionally, seedlings were grown under ambient temperature and under passively elevated temperatures in a greenhouse.

#### MATERIALS AND METHODS

#### Plant Material and Growing Conditions

We used Scots pine seedlings of 10 provenances originating from their south-western distribution, namely from Poland (PL9: Suprasl), Germany (D8: Mittel-/Ostdt. Tiefland, D6: Hauptsmoorwald, D7: Alpenkiefer), Hungary (HU14: Plantage Pornoapati), Italy (I4: Emilia Romagna), France (F12: Mont Ventoux, F3: Prealpes du Sud), Spain (ES1: Alto Ebro) and


Bulgaria (BG10: Garmen). Hereafter, provenances are referred using their abbreviations, which combine the international vehicle registration or country code (ISO 3166-1 alpha-2) and an internal number (**Table 1**). An isozyme study of Taeger et al. (2013a) revealed that German provenances, PL9 and BG10 belong to the same gene pool, whereas I4, F12, F3, and ES1 each belong to a separate gene pool. Annual mean temperature and sum of precipitation at the origin of seed sources ranged between ∼3 and 11◦C and ∼600 and 1,500 mm, respectively (**Table 1**). Aside from D7 and HU14, which were from seed orchards, seeds were collected from autochthonous populations. In April 2012, 232–241 one-year-old seedlings per provenance were potted in 3 l pots containing a peat substrate ('Basismischung Bayer. Staatsforsten AöR', Klasmann-Deilmann GmbH, Geeste, Germany) at the Gewächshauslaborzentrum (GHL) near Freising, Germany. Half of the plants per provenance were placed in a vegetation hall (a glassed roofed building with open side walls), and the second half of seedlings was arranged in a greenhouse, creating two different temperature regimes: close to ambient conditions in the vegetation hall and a 3◦C passive warming in the greenhouse [see Seidel and Menzel (2016) for detailed information]. Hereafter, if not specified otherwise, statements apply to the greenhouse as well as to the vegetation hall. The seedlings were randomly arranged on three tables within each building, with similar numbers of individuals per provenance on each table. Plants were acclimated to the local growing conditions and, based on biweekly soil moisture measurements, they were kept well-watered using a time-controlled dripping system until the experimental drought manipulations were conducted from March 2013 to June 2014. Due to partial harvest of Scots pine seedlings in November 2013 and the use of individuals from specific provenances for another experiment (Lüpke et al., 2016), only around 80 individuals from seven provenances each remained available in 2014 to be distributed across the two buildings and watering regimes of 2013. In March 2014, these remaining seedlings were replanted into 20 l pots containing substrate of identical composition as before and were again randomly arranged on tables within each building, with similar numbers of individuals per provenance and watering regime on each table. For further analyses of the experiment conducted in 2014, these seedlings were grouped according to their latitude into northern (>45◦N) and southern (<45◦N) provenances (**Table 1**), termed region in the subsequent manuscript. The climate at the origin of northern provenances shows a distinct precipitation maximum in summer whereas there is a distinct precipitation minimum at the origin of southern provenances (**Supplementary Figure S1**).

#### Drought Treatments

Drought conditions were simulated during a spring and a summer period in 2013 and a spring period in 2014 (**Figure 1**). In between all seedlings were well-watered (recovery periods). The design of the time-controlled dripping system enabled a maximum of four different seasonal watering groups per year.

In March 2013, 28–31 seedlings per provenance and building were assigned to a well-watered control treatment (wet–wet), a spring drought treatment (March 22 to June 14, dry–wet), a summer drought treatment (July 10 to August 21, wet–dry) or a spring and summer drought treatment (dry–dry). Hereafter, seasonal drought treatments are referred to in shortened form (i.e., spring drought or summer drought). During the drought treatments soil moisture was adjusted to oscillate around the permanent wilting point (pF 4.2) through an initial dry-off period and a subsequent addition of small amounts of water when necessary (**Supplementary Figure S2**). The permanent wilting point was derived from water retention curves following the pressure plate method described by Richards (1941) and corresponded to 12 Vol% soil moisture, which was achieved during spring for around 5 weeks in the vegetation hall and 6 weeks in the greenhouse, and during summer for around 4 weeks in both buildings (**Figure 1** and **Supplementary Figure S2**). All seedlings received equal amounts of water after the drought treatments, which were considered to be a recovery period until the next drought treatment began.

In 2014, 28–40 individuals per building, regional provenance group (north and south) and drought treatment group of 2013 (wet–wet, wet–dry, dry–wet, and dry–dry) were exposed to a drought treatment from March 23 to June 23 by totally withholding irrigation. The soil moisture fell below the permanent wilting point for around 5 weeks in the vegetation hall and 6 eeks in the greenhouse (**Figure 1** and **Supplementary Figure S2**).

Soil moisture was monitored twice a week using a hand-held soil moisture sensor (UMP1, Umwelt-Geräte-Technik GmbH, Müncheberg, Germany) on 240 pots equally spread across provenances and treatments (**Supplementary Figure S2**). Due to technical problems with the dripping system during summer 2014, we did not analyze the recovery period after the 2014 drought treatment.

#### Phenological, Morphological, and Ecophysiological Measurements Phenology

Phenology was monitored weekly on the terminal buds of the main shoot of all individuals from March 3 to June 28, 2013, and from March 17 to June 2, 2014 (**Figure 1**). We recorded the onset of three different phenophases which were classified as bud break (first green tissue visible between bud scales), needle unfolding (first needles emerging from needle sheaths) and needles unfolded (all needles have penetrated through needle sheaths) (Hack et al., 1992).

#### Height, Diameter, Needle Length and Above-Ground Dry Weight

Morphology of seedlings was monitored during five periods in 2013 and one period in 2014; namely from January 3 to 7, 2013 (before the start of the growing season), from June 10 to 14, 2013 (at the end of the spring drought period), from July 8 to 10 and from August 19 to 23, 2013 (before and at the end of the summer drought period), in autumn from November 4 to 9 in 2013, and on June 23 and 24, 2014 following the spring drought period (**Figure 1**). Height was measured from the substrate surface to the terminal tip of each plant using a digital caliper in 2013 and a folding ruler in 2014. Diameter was determined using a

digital caliper at the height of the pot rim, while needle length was measured using a customized ruler at 2 mm intervals. Mean needle length per individual was determined using ten needles equally distributed along the terminal shoot of the respective year. Height and diameter were assessed on all 2,384 pine individuals whereas mean needle length was evaluated for twelve individuals per building, provenance and drought treatment group. From November 11 to 13, 2013, we harvested 20 seedlings per building, provenance and drought treatment group, to assess the dry weight of above-ground compartments. Seedlings were thus cut at the root collar, oven dried at 60◦C for 48 h and then separated into wood, needles and buds.

#### Quantum Efficiency of Photosystem II

Quantum efficiency of photosystem II (PSII) was used as a stress indicator of the photochemical efficiency since it is sensitive to drought conditions (Maxwell and Johnson, 2000; Ashraf and Harris, 2013). Quantum efficiency of PSII was measured on one dark-adapted needle pair of the current year shoot per individual with a continuous excitation fluorimeter (Pocket PEA, Hansatech Instruments Ltd., Pentney, United Kingdom) from midnight until 4 am. We took measurements on nine individuals per building, provenance and drought treatment group in 2013 (720 individuals in total) and on all individuals in 2014 (560 individuals in total). Sampling was conducted on identical individuals in eight periods, six times during the summer drought in 2013, once 3 weeks after the summer drought in 2013 (recovery) and once at the end of the spring drought in 2014 (**Figure 1**). Due to the large amount of individuals, each sampling period consisted of two consecutive nights during which equal numbers of randomly chosen individuals per building, provenance and drought treatment group were measured.

#### Stable Carbon Isotope Ratio of Needles

We selected the four provenances D8, I4, ES1, and BG10 that had a low genetic relationship and differed in their drought response (Taeger et al., 2013a). Needle samples were taken at the end of the summer drought period in 2013 (**Figure 1**). We collected ten needle pairs from the current year main shoot of 10 random individuals per building, drought treatment group and provenance, resulting in 320 samples. Needles were oven dried at 60◦C for 48 h, grounded with liquid nitrogen and weighed before using an elemental analyzer (Euro EA 3000, Eurovector S.p.A, Milan, Italy) and mass spectrometer (Isoprime, GV Instruments Ltd., Manchester, United Kingdom) for isotopic analyses.

#### Statistical Analyses

Linear models using generalized least squares (nlme; Pinheiro et al., 2015) implemented in the R version 3.3.2 (R Core Team, 2016) were applied separately to the vegetation hall and the greenhouse to analyze the effects of provenance, region, and drought on (1) phenology, (2) height, diameter and needle growth, (3) wood, needle, bud and total above ground biomass as well as the needle to wood ratio, (4) quantum efficiency of PSII after the summer drought in 2013 and during the drought in 2014, (5) stable carbon isotope ratio, and (6) to analyze the association between structural response (growth, biomass) and physiological response (stable carbon isotope ratio). For the analysis of the quantum efficiency of PSII during the summer drought in 2013 we used linear mixed effects models (nlme, Pinheiro et al., 2015), with the individual seedlings represented as random variables to account for repeated measurements.

Intra-annual growth of 2013 was analyzed for multiple periods until the growth of respective compartments was completed (**Supplementary Figure S3**); these included January to June and June to July for height growth, January–June, June–July, July–August, and August–November for diameter growth, and January–June, June–July, and July–August for needle growth. Additionally, we analyzed the annual growth of 2013 and growth from November 2013 to June 2014.

The most complex statistical models (full models) included all possible explanatory terms of provenance or region and drought treatments as factorial dummy variables. Only those drought treatments were included in the full models which occurred before and during the respective time or period of measurement of the dependent variable. We included the additional covariates phenophase and/or morphological measures at the beginning of the experiment to evaluate the provenance-specific drought response of height growth, diameter growth and biomass independent of growth potential and growth timing. Therefore, models of height growth included the covariates height that was measured in January along with bud break, i.e., the start of shoot elongation. Diameter growth models included bud break and needle unfolding apart from the diameter measured in January, as the beginning of radial growth can be related to the start of shoot or needle growth depending on environmental conditions

(Swidrak et al., 2013). Needle growth models contained the phenophase of needle unfolding, and biomass models used bud break and needle unfolding as additional covariates. All possible high-order interactions of categorical dummy variables and all possible interactions of current year drought treatments with continuous covariates were included in the statistical full models except for needle growth to prevent overfitting because of a smaller sample size.

Models evaluating the association between structural and physiological response included drought treatments and their interaction as well as provenance, stable carbon isotope ratio and their interaction.

Diagnostic plots were checked for heteroscedasticity and nonnormal distribution of the residuals; variance function structure classes were applied whenever needed. The full models were then simplified based on minimizing the AIC using the drop1 function (stats; R Core Team, 2016). Furthermore, the significance of single terms was proven by performing a type III ANOVA using the Anova function (car; Fox and Weisberg, 2011). These test statistics can be found in **Supplementary Tables S1–S14**. Pairwise comparisons of treatments, provenances, regions and their interactions were computed with the lsmeans function adjusting p-values with the "fdr" method (Lenth, 2016).

If the drought impact was still detectable after drought stress release, we considered it a carry-over drought effect. If a previous drought altered the resistance to a subsequent drought event either through acclimation or additive impacts, we called it interactive drought effect.

When analyses of growth parameters revealed a differing drought response for provenances or regional groups, we calculated their absolute and relative response magnitude. Particular response magnitudes were calculated by pairwise subtracting each individual's growth of the control group from each individual's growth of the drought group. Analyses of response magnitudes were done using generalized least squares (nlme; Pinheiro et al., 2015) with provenance or region as explanatory variable and the lsmeans function (Lenth, 2016) for pairwise comparisons adjusting p-values with the "fdr" method.

Adaptation of provenances' mean annual growth, mean biomass and the aforementioned mean drought response to local site aridity [aridity index (AI) = ratio of precipitation to potential evapotranspiration) was analyzed by conducting simple linear regressions with annual AI, growing season AI, summer AI and the minimum AI of the year using the lm function (stats; R Core Team, 2016). D7 was excluded from this study because AI values exceed the AI values of the second moist site by more than 100% (**Table 1**).

#### RESULTS

#### Influence of Provenance and Drought on Phenology 2013

Mean bud break, needle unfolding and needles unfolded in 2013 occurred 25, 17, and 11 days earlier in the warmer greenhouse than in the vegetation hall, respectively (**Supplementary Figure S4**) and mostly varied with provenance (except needle unfolding in the greenhouse; **Supplementary Table S1**). Except for bud break in the greenhouse, the onset of all phenophases was earliest for PL9 (**Supplementary Figure S4**), e.g., in the vegetation hall mean PL9 bud break was almost 6 days earlier than F12 (p = 0.001) and mean needle unfolding 5 days earlier than ES1. Bud break in general and needle unfolding in the greenhouse occurred before the start of the spring treatment 2013 (**Figure 1**) and were consequently not affected by the drought. The spring drought then delayed needle unfolding by 1.3 days (p < 0.001; **Supplementary Table S1** and **Supplementary Figure S4B**) exclusively in the vegetation hall and after that more strongly the later phenological phase of needles unfolded in both buildings by more than 5 days (see also **Figure 4A**).

#### Annual Variations of Seedling Growth and Wood, Needle and Bud Biomass in 2013

Mean annual height and needle growth was comparable in the vegetation hall and in the greenhouse, whereas overall mean annual diameter growth was 0.5 mm higher in the greenhouse than in the vegetation hall (**Figure 2**). Height, diameter and needle growth varied with provenance (**Supplementary Tables S2–S4**). The southern provenances generally had lower height and needle growth (**Figures 2A,B,P,Q**; at least p < 0.02). In the vegetation hall, ES1 had the smallest diameter growth and differed significantly (p < 0.05) from D8, I4 and BG10 which had the highest diameter growth (**Figure 2H**). In the greenhouse, I4 had the highest diameter growth (**Figure 2J**) and significantly exceeded (at least p < 0.04) the growth of PL9, D6, HU14, F12, and ES1. All measured biomass parameters significantly varied with provenance (**Supplementary Table S5**), but did not show a clear regional pattern and thus are not presented in more detail. In some cases, the provenances from seed orchards (D7 and HU14) are among the better performing provenances (**Supplementary Figure S5**).

The direct negative effect of spring drought on mean height growth was larger in the warmer greenhouse (−64 mm, −26%) than in the vegetation hall (−46 mm, −20%), although in the vegetation hall this effect depended on the provenance, i.e., northern provenances had a higher absolute reduction than southern ones (at least p < 0.002; **Figures 2A,C** and **Supplementary Figure S6A**), but still grew taller than southern provenances. Spring and summer drought both separately significantly reduced total above-ground biomass (p < 0.001, **Figure 3A** and **Supplementary Table S5**) and dry weight of above-ground wood and buds (**Supplementary Figures S7A,C** and **Supplementary Table S5**).

Interacting effects of spring and summer drought were observed for annual diameter growth, needle growth, needle biomass and the needle to wood ratio (**Figures 2I,K,R,T** and **Supplementary Tables S3–S5**). In the vegetation hall, we observed a stepwise decline in diameter growth from spring (−0.5 mm, −11%) to summer drought (−0.8 mm, −18%) to spring and summer drought (−1.2 mm, −25%), whereas in the greenhouse, spring and summer drought caused a similar decrease (p < 0.001) by around −0.5 mm (−10%) in contrast

FIGURE 2 | Estimated model effects of explanatory variables included in the final model on (A–G) annual height growth, (H–O) annual diameter growth, and (P–U) annual needle growth in 2013 in (A,D,E,H,I,L,M,P,R,S) the vegetation hall and in (B,C,F,G,J,K,N,O,Q,T,U) the greenhouse. Shown are fitted mean values and 95% confidence intervals for each variable holding all other variables constant around their mean. Abbreviations in (I,K,R,T) denote wet (w) and dry (d) conditions during the spring and summer treatment. Provenances and treatments sharing the same lowercase letter in the same color within buildings are not different at a significance level of 0.05. Significance levels of asterisk are ∗∗∗p < 0.001, ∗∗p < 0.01.

to −1.1 mm (−21%) for both drought treatments (p < 0.001; **Figures 2I,K**). Needle growth decreased under summer drought conditions in the vegetation hall (**Figure 2R**) while under greenhouse conditions only the combination of both droughts had a negative effect (p < 0.001; **Figure 2T**). The pattern of annual needle growth was also reflected in needle dry weight (**Supplementary Figure S7B** and **Supplementary Table S5**). The needle to wood ratio of seedlings increased with drought regardless of season, but less strongly when only subjected to a summer drought (at least p < 0.002; **Figure 3B** and **Supplementary Table S5**).

Initial height and diameter in January 2013 promoted respective annual growth when seedlings got well-watered in spring, but had no effect on annual diameter growth or even slightly decreased annual height growth under spring drought conditions (**Figures 2D,F,L,N** and **Supplementary Tables S2**, **S3**). Later bud break resulted in lower annual height growth (**Figures 2E,G** and **Supplementary Table S2**). This effect was stronger in the greenhouse (−1.9 mm/day) than in the vegetation hall (−1.0 mm/day). Later bud break decreased annual diameter growth, except for the summer drought group, and increased the needle to wood ratio just under warmer conditions in the greenhouse (**Figure 2O** and **Supplementary Tables S3**, **S5**). Needle unfolding interacting with spring and summer drought improved the model fit of diameter growth in the vegetation hall, although the slopes of the treatment groups were not significantly different from the well-watered control group (**Figure 2M**). Later needle unfolding generally resulted in shorter needles, with a weaker response under spring drought conditions (**Figures 2S,U** and **Supplementary Table S4**) and increased bud weight in the greenhouse (**Supplementary Table S5**). A later start of needle unfolding had positive effects on wood and needle biomass when subjected to summer drought and combined spring and summer drought conditions, respectively (**Supplementary Table S5**). Additionally, later needle unfolding decreased the needle to wood ratio in the spring drought treatment (greenhouse) and summer drought treatment (vegetation hall) (**Supplementary Table S5**).

## Intra-Annual Variations of Seedling Growth in 2013 With Provenance, Morphology and Phenology

Height growth from January to June and June to July as well as needle growth from January to June, June to July, and July to August clustered provenances according to their origin (north and south, **Supplementary Figures S8A,B,E–G**). In general, southern provenances grew less and were less affected during drought periods in terms of absolute height growth in the vegetation hall as well as absolute needle growth from July to August in the greenhouse (**Supplementary Figures S6A,B**). In contrast, intra-annual diameter growth was not associated with a north–south pattern (**Supplementary Figures S8C,D**).

During the growth period from January to June, taller seedlings grew more under well-watered conditions but were more strongly affected by drought conditions. A later bud break generally resulted in smaller seedlings (**Supplementary Table S2**).

The initial diameter measured in January increased intraannual diameter growth under well-watered conditions, but decreased it under drought. In the vegetation hall, a later start of needle unfolding resulted in less intra-annual diameter growth from January to June (only in the control) and from August to November, but promoted diameter growth during the period from June to July. In the greenhouse, seedlings that started to unfold needles later had smaller diameter growth from July to August (**Supplementary Table S3**).

Except during July and August, later start of needle unfolding resulted in shorter needles in the greenhouse. Needle growth of seedlings responded less during drought in the vegetation hall, and when released from drought from June to July in the greenhouse (**Supplementary Table S4**).

### Direct, Interactive and Carryover Drought Effects on Intra-Annual Growth in 2013

During spring, drought significantly reduced height, diameter and needle growth, and delayed needles unfolded in the vegetation hall and the greenhouse by 5.2 and 2.0 days, respectively (p < 0.001; **Figures 4A–D**). Although the absolute growth was higher in the greenhouse, the relative reduction was similar in both buildings and ranged from 22 to 28%. During drought in summer, diameter growth even decreased by 75% in the greenhouse (p < 0.001; **Figure 4E**).

In the vegetation hall, the reduction of diameter growth during the summer treatment period was influenced by previous spring treatment conditions (**Figure 4F**). These interactive drought effects (dependency of the effect of a drought on a previous

drought) were detected not only for July to August diameter growth, but also needle growth (**Supplementary Tables S3**, **S4** and **Figure 4G**). In the vegetation hall, diameter growth from July to August was significantly lower (p < 0.001) in individuals experiencing the summer drought, but seedlings that additionally experienced the spring drought before being subjected to the summer drought had 70% higher (!) diameter growth (p = 0.024) than seedlings that only experienced the summer drought (**Figure 4F**). From July to August in both buildings, the summer drought alone decreased needle growth by around 30% compared to the control (at least p < 0.008), while seedlings that experienced the spring and summer drought showed resistance and grew similarly to control seedlings (**Figure 4G**). Seedlings that experienced drought during spring exhibited a 72% (vegetation hall) and 102% (greenhouse) increase of July to August needle growth (p < 0.001; **Figure 4G**) compared to the control.

Diameter growth of pine seedlings was influenced by drought periods following the release from respective drought stress (carryover drought effects; **Figures 4H–J**). In the greenhouse, the negative effect of the spring drought persisted for diameter growth from June to July and July to August (p < 0.001; **Figures 4H,I**). In the spring drought group, diameter growth from August to November was reduced by 12% in the vegetation hall (p = 0.04), but increased by 26% in the greenhouse compared to control seedlings (p < 0.001; **Figure 4J**). The summer drought had positive effects after stress release on diameter growth from August to November in both buildings (p < 0.001, **Figure 4J**).

#### Relation of Growth in 2013 and Drought Response to Aridity at the Origin of Provenances

In the vegetation hall, annual height and needle growth could be best predicted by minimum AI (p < 0.001; **Figure 5**). In the greenhouse, minimum AI could just predict annual height growth (p < 0.001; **Figure 5**). For biomass parameters, only total above-ground biomass responded to summer AI (p = 0.05). For all analyzed parameters, the coefficients of determination decreased when provenances were grown under higher temperatures in the greenhouse.

In the vegetation hall, absolute and relative spring drought response of January to June and annual height growth could be significantly predicted by summer AI and minimum AI

excluded from the analyses, because its AI values exceed the AI values of the second moist site by more than 100%.

(**Figure 6**; p < 0.001 for absolute growth response and p < 0.05 for relative growth response). In the greenhouse, only absolute summer drought response of July to August needle growth was significantly explained by minimum AI (**Figure 6**; p < 0.05). Annual AI and growing season AI were not related to any growth, biomass or drought response parameter.

#### Phenology in 2014

more than 100%.

Mean phenological development in 2014 started 10– 14 days earlier in the greenhouse than in the vegetation hall (**Supplementary Figure S9**). Buds of the southern provenances broke around 2 days earlier than buds of the northern provenances (at least p < 0.0013; **Supplementary**

needle growth from January to June and from July to August. Black color indicates the relationships in the vegetation hall whereas red color shows the relationships in the greenhouse. D7 was excluded from the analyses, because

its AI values exceed the AI values of the second moist site by

**Figures S9A,C** and **Supplementary Table S6**). Previous year spring drought advanced bud break by almost 2 days in the vegetation hall (p = 0.018; **Supplementary Figure S9B**) and by around three and a half days in the greenhouse (p < 0.001; **Supplementary Figure S9D**). The phenophases needle unfolding and needles unfolded responded to current year drought conditions in the vegetation hall, whereas they differed between region and previous year drought conditions in the greenhouse. Therefore, in the vegetation hall, drought advanced needle unfolding (p = 0.05) and the day when all needles were unfolded (p = 0.018) by more than 1 day (**Supplementary Figures S9E,H** and **Supplementary Table S6**). In the greenhouse, southern provenances started to unfold needles (p = 0.018) and had needles unfolded (p = 0.001) around one and half have days earlier than northern provenances (**Supplementary Figures S9F,I** and **Supplementary Table S6**). Finally, the previous year spring and summer drought, respectively, advanced needle unfolding (p < 0.001) and needles unfolded (p = 0.001) by approximately 2 days.

# Seedling Growth Variation With Provenance, Drought and Phenology in 2014

Seedlings in both buildings showed similar overall mean height growth, increased overall mean diameter growth (+1.2 mm) but decreased overall mean needle growth (−8,7 mm) in the vegetation hall compared to the greenhouse (**Supplementary Tables S7–S9**). Northern provenances grew more in height and had longer needles (at least p < 0.002; **Figures 7A,C,Q,S** and **Supplementary Tables S7, S9**), but grew less in diameter than southern ones (**Figures 7I,K** and **Supplementary Table S8**).

Current year drought decreased height and diameter growth (p < 0.001) with a stronger decrease of height growth in the vegetation hall than in the greenhouse (**Figures 7B,D,J,L** and **Supplementary Tables S7, S8**). The reduction of absolute needle growth by current year drought was stronger in the northern provenances than in the southern provenances (**Figures 7Q–T**, **Supplementary Table S9** and **Supplementary Figure S10**).

Carryover drought effects of the previous year summer drought still reduced height growth of well-watered seedlings in the vegetation hall (−10% or −35 mm; p < 0.001) (**Figure 7B** and **Supplementary Table S7**).

Positive carryover effects of the previous year spring drought and additionally of the previous year summer drought were apparent for height and needle growth, respectively (**Figures 7E,G,U–X** and **Supplementary Tables S7, S9**). Furthermore, in the vegetation hall, seedlings that experienced at least one drought event in the previous year grew 23–30% more (p < 0.001) in diameter in 2014 (**Figure 7M** and **Supplementary Table S8**).

Later bud break resulted in lower height growth in both buildings (**Figures 7F,H** and **Supplementary Table S7**) and in slightly lower diameter growth in the greenhouse (p = 0.02; **Figure 7O** and **Supplementary Table S8**). Later onset of needle unfolding resulted in lower diameter growth and shorter needles under well-watered conditions (**Figures 7N,P,R,T** and **Supplementary Tables S8, S9**). In general the phenological effects on height and needle growth were more pronounced in the vegetation hall than in the greenhouse.

#### Influence of Provenance and Drought on Quantum Efficiency of Photosystem II and Stable Carbon Isotope Ratio

Quantum efficiency of PSII during the summer 2013 was not influenced by the actual drought in the vegetation hall, but was lower under drought in the greenhouse (p < 0.001; **Supplementary Figure S11A** and **Supplementary Table S10**). In the vegetation hall, previous spring drought had a positive effect on quantum efficiency (p < 0.001), whereas in the greenhouse quantum efficiency was still negatively affected (p = 0.04; **Supplementary Figure S11A**). The provenances I4, ES1, and F3 had the lowest quantum efficiency during the summer treatment period (**Supplementary Figure S12**).

In the vegetation hall after summer drought stress release, previous spring or summer drought experience increased quantum efficiency (at least p < 0.02; **Supplementary Figure S11B** and **Supplementary Table S10**). In the greenhouse, summer drought reduction of quantum efficiency still persisted after drought stress release (p = 0.001, **Supplementary Figure S11B** and **Supplementary Table S10**). In 2014, southern provenances had lower quantum efficiencies than northern provenances (p < 0.008). Overall, quantum efficiency was significantly reduced (p < 0.001) by drought (**Supplementary Figure S13** and **Supplementary Table S11**).

The differences of stable carbon isotope ratios between provenances and treatments were more distinct in the greenhouse than in the vegetation hall, although stable carbon isotope ratio followed the same patterns (**Supplementary Figure S14** and **Supplementary Table S12**). Therefore, the provenances D8 and BG10 showed less carbon discrimination than the provenances I4 and ES1 (**Supplementary Figure S14**). A reduction of carbon discrimination was caused by summer drought and was intensified by spring drought (**Supplementary Figure S12**).

# Association Between Physiological and Structural Response

Annual diameter growth and biomass showed a positive relationship with δ13C while accounting for drought effects (**Figure 8** and **Supplementary Tables S13, S14**). In the vegetation hall, annual height growth was not associated with stable carbon isotope ratio, whereas in the greenhouse, this association varied between provenances (**Figure 8** and **Supplementary Tables S13, S14**). Height growth of D8 and I4 decreased, while height growth of ES1 and BG10 increased with rising stable carbon isotope ratio. Drought treatments generally had a negative influence on growth and biomass (**Supplementary Tables S13, S14**). Since results of drought and provenance influences on growth and biomass were similar to the results described in "Annual variations of seedling growth and wood, needle and

FIGURE 7 | Estimated model effects of explanatory variables included in the final model on (A–H) height growth, (I–P) diameter growth, and (Q–X) needle growth from November 2013 to June 2014 in (A,B,E,F,I,J,M,N,Q,R,U,V) the vegetation hall and in (C,D,G,H,K,L,O,P,S,T,W,X) the greenhouse. (E,F) Height, (M) diameter and (U–X) needle growth show compensatory growth to previous year drought. Shown are fitted mean values and 95% confidence intervals for each variable holding all other variables constant around their mean. Abbreviations in (B,M) denote wet (w) and dry (d) conditions during the drought periods. Regional provenances groups and treatments sharing the same lowercase letter in the same color within buildings are not different at a significance level of 0.05. Levels of significance are <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, nsp > 0.05.

bud biomass in 2013", we kindly refer to this chapter for more detailed information.

#### DISCUSSION

#### Provenance Effects

Phenology, growth traits, biomass and carbon isotope ratios differed between provenances, indicating genotypic variation. An analysis of isozymes of most of the provenances used in our experiment found differences in genetic variability and genetic diversity (Taeger et al., 2013a). Provenance-specific growth patterns could thus be a reflection of adaptations to the climatic conditions at their origins, especially since site aridity index and diverse growth parameters were positively correlated. Respectively, various drought growth responses were lower in provenances from drier origins especially when assessed in absolute terms.

Onset of phenophases in 2013, particularly of needle unfolding and needles unfolded in the vegetation hall, occurred earlier in northern than in southern provenances, agreeing with results from a previous study by Taeger et al. (2015). Generally, forcing requirements can differ between populations (Hannerz et al., 2003) and provenances from colder origins often require lower temperature sums to trigger bud break (Berninger, 1997). However, this pattern could not be confirmed in our greenhouse experiment of spring 2013, most likely because temperatures never fell below 0◦C, causing higher forcing requirements when the natural chilling requirements of P. sylvestris were not met (Laube et al., 2014). In contrast, onsets of phenophases in 2014 were earlier for southern than for northern provenances. All provenances in the southern group of 2014 originated from high elevations with cooler mean spring temperatures than those at the origins of the northern group (Seidel and Menzel, 2016). This difference between low and high elevation populations may be comparable to the differences between southern and northern latitude populations. The results of phenology in 2014 can be confirmed by additional analyses of phenophases in 2013 with the same groups of provenances as used in 2014 (results not shown).

Southern provenances generally showed smaller height and needle growth, but also a smaller drought-induced reduction of height growth and needle growth in 2013 and 2014, respectively. Oleksyn et al. (1992, 1998, 1999) and Semerci et al. (2017) found similar differences in overall height and needle growth between northern and southern provenances. The provenancespecific height and needle growth suggests a lower phenotypic response but also a better adaptation to drought by individuals from southern provenances. This is supported by our finding that height and needle growth is positively correlated with aridity at the site of their origin. Since tree height and needle area are positively correlated (Xiao and Ceulemans, 2004; Jagodzinski ´ and Kałucka, 2008), the reduction of stem and needle growth itself might decrease evaporative water loss and thus increase resistance to drought. Additionally, smaller trees have a lower risk of suffering hydraulic dysfunction because of physical and anatomical aspects. Forces to lift water in xylem conduits against gravity and conduit resistance are lower in smaller sized trees,

reducing the risk of a water column collapse and thus the impairment of water transport by xylem cavitation (McDowell and Allen, 2015). Furthermore, conduit size increases with the distance to the stem tip (Anfodillo et al., 2006, 2013; Olson et al., 2014), resulting in a higher risk of xylem cavitation (Martínez-Vilalta et al., 2009; Sterck et al., 2012). It has been shown on a global scale and for Scots pine seedlings in particular, that tree height is linked to a higher risk of drought-induced mortality (Bennett et al., 2015; Seidel and Menzel, 2016). The drought adaption of southern provenances is also reflected in the stable carbon isotope ratios of needles, similar to Pinus pinaster (Correia et al., 2008). Southern provenances (I4, ES1, and BG10) show more negative values than D8, suggesting lower water use efficiency (Jones, 2013), higher maximum net photosynthesis (DeLucia and Schlesinger, 1991) and higher stomatal conductance (DeLucia et al., 1988) during carbon fixation. This loose stomatal control might be of advantage in arid climates since photosynthesis can continue for longer periods. Moreover, ES1 and BG10 which originate from drier sites than D8 and I4, can translate higher water use efficiency to higher height growth, whereas higher water use efficiency in D8 and I4 is linked to reduced height growth. This might further suggest provenance-specific adaptation to drought, since a positive relationship between water use efficiency and height growth could indicate higher assimilation, while a negative relationship would indicate stronger stomatal control (Marguerit et al., 2014). A looser stomatal control of the southern provenances, and thus a partial exploitation of water resources in pots, might also be seen through the slightly lower values of photosynthetic efficiency we recorded in 2014.

A provenance-specific drought response of height growth was detectable in the vegetation hall for the period from January to June and for the annual period in 2013, but not from November 2013 until June 2014. In a provenance trial in Poland (Oleksyn et al., 2001) duration of Scots pine shoot elongation was about 60–70 days and reached its maximum growth rate after 42–48 days. In our experiment, the time span between bud break and maximum drought (soil moisture below the permanent wilting point) ranged between 36 and 40 days (greenhouse in 2013, during 2014) and just 13 days for the vegetation hall in 2013. The height growth of seedlings in the vegetation hall in 2013 was thus affected by intensive drought for a much longer time, while seedlings in all other cases could have partially escaped drought impacts, diluting our observations of provenance-specific drought responses. Experiencing higher drought stress exposure could also be an explanation for the provenance-specific differences in drought response, e.g., of needle growth. Higher temperatures in the greenhouse increased the mean and maximum vapor pressure deficit by 0.18–0.35 kPa and 1.2–1.45 kPa, intensifying stress during the spring and summer drought; in contrast to 2013, the drought treatment groups of 2014 did not receive any water at all during the drought period, leading to soil moisture values well below the species' permanent wilting point for several weeks.

Diameter growth and biomass parameters also differed between provenances, but without as clear of a pattern as height and needle growth. Most prominent differences include the superiority of Alpenkiefer (D7) and Plantage Pornoapati (HU14), whose seeds came from seed orchards aiming at profitable growth and biomass production. The high variability of provenances in biomass production was also confirmed by other authors (Oleksyn et al., 1999, 2000), however, their reported relationship of total above-ground biomass with latitude was driven by a broader latitudinal range than in our study. This may imply that diameter growth is not under a pronounced climate-related selective pressure compared to height and needle growth, as indicated by our study results. Lastly, we found no pattern between needle dry weight and needle growth. This disagreement might be explained by the observation that specific leaf area is variable among individuals from the same provenance and does not follow a particular pattern (Taeger et al., 2015). Additionally, there are differences in biomass allocation to above-ground

BG10 (Garmen, Bulgaria).

compartments between trees grown under non-drought and drought conditions (DeLucia et al., 2000).

Thus, in accordance with recent literature findings, our results strongly suggest that provenances differ in their drought response especially with regards to phenology, height and needle growth as well as stable carbon isotope ratio (research question 1b, c). Some of these differences were linked to their climatic origin assuming local adaptation. However, these relationships of growth and biomass parameters diminished under warmer conditions in the greenhouse, which might ultimately hinder the selection of provenances suitable for assisted migration with ongoing climate change.

#### Drought and Building Effects

Direct drought effects, i.e., the response of Scots pine seedlings to drought, are manifold (research question 1a): seasonal drought treatments directly affected almost all phenological, growth and ecophysiological traits except bud break in 2013, phenophases in the greenhouse in 2014 and quantum efficiency of the PSII in the vegetation hall in 2013. We mainly attribute building effects to the higher temperatures measured in the greenhouse consequently increasing the mean and maximum vapor pressure deficit by 0.33 and 1.4 kPa during frost free periods. This higher evaporative demand has probably caused more distinct differences of carbon discrimination between provenances in the greenhouse. However, radiation could be controlled by automated shading to prevent over-heating of the greenhouse in summer what might have caused higher needle carbon isotope ratios induced by reduced radiation (Brendel et al., 2003).

Current year drought delayed the phenological development of buds and leaves in 2013, but advanced it in 2014. The delay of phenology might be due to low water potentials impeding tissue formation (Hsiao and Acevedo, 1974; Peñuelas et al., 2013). Bernal et al. (2011) also observed advanced spring phenology during drought and attributed it to a lack of transpirational cooling and thus earlier phenological development. This mechanism might be constrained by the reduced growth caused by low water potentials. In the vegetation hall, soil moisture was lower and mean and maximum vapor pressure deficits were 0.1 and 0.8 kPa higher during the mean onset of phenophases in 2013 than in 2014, suggesting higher drought stress during phenological development in 2013. This might have caused the switch from drought-induced advance to drought-induced delay of phenology. Spring phenology in the greenhouse, apart from needles unfolded in 2013, did not respond to the current year drought treatments since the mean onset of phenophases occurred before the severe drought conditions around the permanent wilting point. Depending on the phenophase, onset dates were 10–20 days earlier in the greenhouse, matching the well-known advance of phenology with higher temperature (Reyer et al., 2013).

Growth reduction under drought might not be due to a limitation of photosynthesis since tissue growth commonly decreases ahead of carbon assimilation (Muller et al., 2011), indicating a sink rather than a source limitation (Körner, 2015); a drought-related growth decline of Scots pine has been observed even though the pool of carbon assimilates increased (Gruber et al., 2011; Bachofen et al., 2017). Drought generally decreases turgor pressure or induces the production of growth regulators, which in turn reduces cell division and expansion (Hsiao and Acevedo, 1974; Peñuelas et al., 2013). Nevertheless, our findings show that drought increased the stable carbon isotope ratio and decreased the quantum efficiency of the PSII, suggesting stomatal closure along with a limitation of the PSII, thereby restricting photosynthesis (DeLucia et al., 1988; Ashraf and Harris, 2013).

A lacking influence of summer drought on quantum efficiency of the PSII in the vegetation hall in 2013 can be explained by drought severity. Since the mean and maximum vapor pressure deficit was 0.35 and 1.2 kPa lower than in the greenhouse, the stress for seedlings was smaller (Williams et al., 2012). Thus, drought stress in summer 2013 in the vegetation hall might have been too low to induce an inhibition of PSII, as suggested in a similar study on Norway spruce (Pukacki and Kaminska-Ro ´ zek, 2005 ˙ ).

Drought conditions increased the needle to wood dry weight ratio indicating that needle growth was less sensitive to drought than wood growth. This finding is contradictory to other studies since the ratio between transpiring and water transporting tissue should decrease with drier conditions (Poyatos et al., 2007; Martínez-Vilalta et al., 2009) and it may not solely be related to different life stages (young vs. adult) or study conditions (experiment with potted individuals vs. field studies).The reduction of a considerable amount of foliage can be a long lasting process (Dobbertin, 2005), whereas the adjustment of the xylem can be very fast (Bryukhanova and Fonti, 2013); thus, we might have missed the ultimate drought impact on needle to wood biomass ratio since we only measured biomass during the 1st year of drought.

Elevated temperatures have been shown to decrease height and diameter growth of mature Scots pine in the field (Martínez-Vilalta et al., 2008; Reich and Oleksyn, 2008; Michelot et al., 2012), but there was no warming effect in Scots pine and Ponderosa pine seedling experiments (Maherali and DeLucia, 2000; Taeger et al., 2015), confirming a change of sensitivity to environmental influences with ontogeny (Niinemets, 2010). In our study we did not find obvious differences in height growth between the cooler vegetation hall and the warmer greenhouse, although in a previous study height growth of PL9, D7, and F12 was lower in the greenhouse than in the vegetation hall (Seidel and Menzel, 2016). If the reduced set of provenances had been considered in our current study, the results presented here would be similar (data not shown). The sensitivity of trees to environmental influences can change with ontogeny and thus alter their responses to climatic conditions (Niinemets, 2010). In contrast to temperature-insensitive height growth, diameter growth under warmer conditions in the greenhouse in 2013 was higher than under cooler temperatures in the vegetation hall, although this pattern was reversed in 2014. The overall differences in annual diameter growth in 2013 matched the increased growth from August to November, suggesting a longer growth period induced by higher temperatures (Peltola et al., 2002; Rossi et al., 2014). Spring radial growth in the greenhouse in 2014 might have been constrained by the depletion

of carbohydrate reserves during the longer growing period in 2013. Early wood formation relies on stored carbohydrates, which are affected by late wood formation in the previous year (Oberhuber et al., 2011).

Compared to the existing literature, our study was unique in showing interactive effects of previous drought events, either in the same year (spring summer) or in the subsequent year (2013 and 2014) on seedling response (see research question 2). For several parameters we could show that previous drought experience reduced the impact of a following extreme event.

#### Carryover Effects and Compensation

Related to our third research question, the results clearly revealed drought-related carryover effects in the greenhouse for (a) intra-annual diameter growth, (b) for the efficiency of PSII and (c) in the vegetation hall for inter-annual shoot growth. Naturally, recovery of water potentials takes longer under warmer conditions (Balducci et al., 2016), prolonging the time to reach the necessary cell turgor pressure for cell division and expansion (Hsiao and Acevedo, 1974) and thereby inhibiting diameter growth, as might have been the case after our spring drought in 2013. Most likely, different timings of bud set in the greenhouse and in the vegetation hall modulated the influence of the summer drought in 2013 on shoot growth in 2014. Notably, current year shoot growth potential in Scots pine is related to previous year conditions during bud formation (Wareing, 1956; Junttila and Heide, 1981) as well as to water availability during the previous summer (Jansons et al., 2015). Elevated temperatures for example, can delay bud dormancy and bud formation (Wareing, 1956; Strømme et al., 2016). Taeger et al. (2013a) showed that 50% of bud set under greenhouse conditions was achieved in September. Consequently, in our study, bud set might have occurred earlier in the vegetation hall and overlapped with the summer drought, thereby reducing growth in the following year. The efficiency of PSII was obviously not recovered after the more severe spring and summer drought in the greenhouse, indicating damage of the PSII reaction centers (Ashraf and Harris, 2013). Following drought stress release in the vegetation hall, efficiency of PSII was higher in drought stressed than in nonstressed seedlings. We assume that one of the drivers of increased photoprotection was triggered during the drought treatment and led to an increased efficiency of the PSII once seedlings were well-watered (Derks et al., 2015).

Phenology in 2014 was advanced by previous year drought when not impaired by current year water availability. This could be beneficial for Scots pine by escaping unfavorable conditions during spring. Similar results have been documented for various oak species (Spieß et al., 2012; Kuster et al., 2014). Since environmental conditions during bud formation influence succeeding year growth (Wareing, 1956; Junttila and Heide, 1981) it is likely that phenological development could also be affected, although we are not aware of any study describing a mechanism behind earlier phenology induced by previous year drought.

Pine seedlings responded with intra- and inter-annual compensatory growth of height, diameter and needles after drought stress release, thus affirming our research question 3. Enhanced shoot growth upon re-watering after drought was observed in Quercus petraea within and across years (Spieß et al., 2012; Turcsán et al., 2016). Arend et al. (2016) as well as Pflug et al. (2018) observed stimulated net-photosynthesis in formerly drought stressed Fagus sylvatica saplings until the end of the vegetation period, partly counterbalancing previous drought effects. Numerous studies show an accumulation of nonstructural carbohydrates when water availability is low (Gruber et al., 2011; Muller et al., 2011; Bachofen et al., 2017; Piper et al., 2017). Since phloem transport of non-structural carbohydrates can be impaired under insufficient water supply (Sala et al., 2010), the release from drought stress might induce changes in carbon allocation from source to sink tissue and thus provoke an increased growth rate. This is in line with recent findings, which show that the non-structural carbon pool size is positively related to ring-width growth (von Arx et al., 2017). Allocation dynamics of non-structural carbon to different organs varies during the year (Rosas et al., 2013), indicating that drought timing may influence the growth response. This can be clearly seen in our study for the inter-annual response of height growth in relation to the previous year spring and summer drought. Nevertheless, compensatory growth might differ between ontogenetic stages since the ratio between currently assimilated carbon and carbon pools is higher in seedlings compared to mature trees (Niinemets, 2010).

# CONCLUSION

In this study we demonstrate that Scots pine seedlings show a highly plastic (direct) response of phenology, growth and ecophysiological parameters to reoccurring drought events. Our results suggest that intra-annual compensatory growth, however, is not sufficient to fully offset the drought-induced reduction of annual growth, and can only help to mitigate these impacts. Nonetheless, we were able to identify carryover effects and show that this compensatory growth can also occur on an inter-annual time scale. Interactive effects of multiple droughts may have the ability to render seedlings resistant against negative direct impacts. Additionally, we were able to show that the timing of drought, in relation to phenology, modulates the influence on seedling growth and phenology itself. Lastly, our findings suggest that southern provenances of Scots pine are better adapted to drought conditions than northern ones; the former display a less severe drought response and exhibit morphological characteristics associated with drought resistance. Although southern provenances appear to be less productive as a result of lower height and needle growth, it may actually render them more resilient to extreme climatic events and highlights the apparent trade-off between productivity and drought resistance. The predictability of provenances' drought performance through climatic parameters might nonetheless be constrained by higher temperatures in the future. Warmer temperatures could counteract plastic responses due to intensified drought conditions and shifting phenophases, leading to indirect drought effects. More studies are needed to better understand the relationship between carbon assimilation, allocation, storage and use during and after drought conditions,

a hot topic in current climate change and forestry research (Granda et al., 2017). The influence of drought timing and phenology on inter- and intra-annual tree growth should also be further investigated.

# AUTHOR CONTRIBUTIONS

HS collected data, contributed to the experimental design, analyzed and interpreted the data, and wrote the manuscript. MM contributed to data analyses and interpreted the data. AM contributed to the conception of the work, interpreted the data, and wrote the manuscript.

#### FUNDING

This work was financed by the European Research Council under the European Union Seventh Framework Programme (FP7/2007–2013/ERC Grant Agreement No. 282250). This work was supported by the German Research Foundation (DFG) and

#### REFERENCES


the Technical University of Munich (TUM) in the framework of the Open Access Publishing Program.

#### ACKNOWLEDGMENTS

We thank Phillip Falk and numerous student workers for their untiring commitment during the fieldwork and Steffen Taeger, Andreas Ludwig (BaySF), and the ASP/Teisendorf for providing plant and seed material. We further thank the team of the GHL Dürnast for carefully handling all plant material during the overarching drought experiment and Sofie Hemprich for her help with language editing.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.00519/ full#supplementary-material




ozone pollution over 10 years. For. Ecol. Manage. 157, 191–204. doi: 10.1016/ S0378-1127(00)00665-4


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

Copyright © 2019 Seidel, Matiu and Menzel. 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.

# Fire Severity Influences Ecophysiological Responses of Pinus pinaster Ait

#### Francesco Niccoli, Assunta Esposito, Simona Altieri and Giovanna Battipaglia\*

Department of Environmental, Biological and Pharmaceutical Sciences and Technologies, Università degli Studi della Campania Luigi Vanvitelli, Caserta, Italy

The effect of fire severity on Pinus pinaster growth and ecophysiological responses was evaluated in four burned sites of Vesuvio National Park, Southern Italy. After the wildfire of 2017, when over 1300 hectares of vegetation, mainly P. pinaster woods, were destroyed, four sites were selected according to the different degree of fire severity and a multidisciplinary approach based on tree rings, stable isotopes and percentage of crown scorched or consumed was applied. All the sampled trees in the burned sites showed a decrease in tree growth in 2017, in particular in the latewood at high-severity site. The dendrochronology analyses showed that several individuals experienced and endured higher fire severity in the past compared to 2017 fire. Further δ <sup>13</sup>C and δ <sup>18</sup>O underlined the ecophysiological responses and recovery mechanisms of P. pinaster, suggesting a drastic reduction of photosynthetic and stomata activity in the year of the fire. Our findings demonstrated that P. pinaster growth reduction is strictly linked to the percentage of crown scorch and that even trees with high level of crown scorched could survive. In all the burned sites the high temperatures and the time of exposure to the flames were not sufficient to determine the death of the cambium and all the trees were able to complete the 2017 seasonal wood formation. This data can contribute to define guidelines to managers making post-fire silvicultural operations in pine forest stands in the Mediterranean Basin.

#### Edited by:

Andreas Rigling, Swiss Federal Institute for Forest, Snow and Landscape Research (WSL), Switzerland

#### Reviewed by:

Daniel Moya, Universidad de Castilla La Mancha Albacete, Spain Peter Prislan, Slovenian Forestry Institute, Slovenia

#### \*Correspondence:

Giovanna Battipaglia giovanna.battipaglia@unicampania.it

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 17 January 2019 Accepted: 09 April 2019 Published: 26 April 2019

#### Citation:

Niccoli F, Esposito A, Altieri S and Battipaglia G (2019) Fire Severity Influences Ecophysiological Responses of Pinus pinaster Ait. Front. Plant Sci. 10:539. doi: 10.3389/fpls.2019.00539 Keywords: fire severity, Pinus pinaster (maritime pine), stable isotope of O and C, tree ring, post-fire effects

#### INTRODUCTION

Forest fires are a critical issue in the Mediterranean basin, experiencing increasing frequency and intensity in the last decades (IPCC, 2014).

The impact of fire on trees can cause damage to the canopy, trunk, and root system (Ducrey et al., 1996). The single or combined effect of these damages can reduce the vigor of the plant and trigger a temporary reduction in growth (Battipaglia et al., 2014a) or can lead to tree mortality if the fire is particularly destructive (Brown, 2000). Understanding post-fire responses of trees is a crucial issue in planning forest management actions of burned area in the short term (Bovio et al., 2017) and fire risk reduction at the medium and long term (Battipaglia et al., 2017). It is wellknown that the variation in plant responses to fire is linked to the species-specific heat sensitivity (Catry et al., 2010) and to fire regime. Indeed, several plant species are able to tolerate forest fires of medium and low severity, thanks to their adaptive traits that guarantee their survival: very thick bark and needles, deep root system, self-pruning capacity and particular structure of the crown

**203**

(Brown, 2000; Madrigal et al., 2019). Fire sensitivity is most often studied in pine species due to their wide distribution range (Fernandes et al., 2008; Espinoza et al., 2019). An extensive literature deals with pine post-fire responses and mortality prediction after wildfire or prescribed burnings, mainly in North America (e.g., Peterson and Arbaugh, 1986a; Ryan and Reinhardt, 1988; Stephens and Finney, 2002; Beverly and Martell, 2003; McHugh and Kolb, 2003; Fowler and Sieg, 2004; Kobziar et al., 2006; Sieg et al., 2006; Breece et al., 2008; Hood et al., 2010), but only limited information is related to fire resistance of European pines (see Fernandes et al., 2008). In particular, several papers analyzed post-fire tree mortality or recruitment, while very few concerns the ecophysiological responses of pines (mainly Pinus halepensis and Pinus sylvestris) after fire (Beghin et al., 2011; De Micco et al., 2013; Battipaglia et al., 2014a,b, 2016; Valor et al., 2018). However, little is known about the post-fire ecophysiological response of Pinus pinaster Ait., a widespread and fire-prone species which is also economically and ecologically important in Mediterranean region (Botelho et al., 1998; Vega et al., 2008, 2010; Catry et al., 2010). Previous studies addressed post-fire recruitment (Calvo et al., 2008; Vega et al., 2008, 2010), or post-fire regeneration (Maia et al., 2012) of P. pinaster, highlighting contrasting results for this species and its resistance to fire. Indeed, Fernandes et al. (2008) reported a considerable fire resistance of this species due to the bark depth and to the high temperature requested for needle necrosis in comparison to Pinus pinea. While Catry et al. (2010), using a mortality model, suggested that P. pinaster is more vulnerable to fire than other pines, such as P. pinea because of its crown architecture. Finally, Vega et al. (2010), found consistent differences in P. pinaster mortality both in relation to site and fire conditions. Thus, it becomes extremely important to better understand ecophysiological responses to fire of this species and its link with fire severity. In this study we used a multidisciplinary approach of dendrochronology, stable isotopes and percentage of crown scorch to assess the effect of a large wildfire, occurred in Italy during July 2017, when over 1300 hectares of vegetation, mainly P. pinaster woods, were destroyed and where different degrees of fire damage were recorded (Battipaglia et al., 2017). Dendrochronology allowed, not only, to reconstruct historical fire dynamics (Camarero et al., 2019) in the study area and involving sampled trees, but also to verify the growth related to the fire year 2017. Carbon and oxygen stable isotopes were used to understand the complex ecophysiological processes involved in post-fire responses, linking the possible reduction of carbon assimilation to crown damage or to reduction in stomatal activity (Battipaglia et al., 2014a,b, 2016). Fire severity, considered as a measure of the immediate impact of fire on trees, was evaluated in terms of the degree of crown scorch, cambium or root damage and height of burn on the trunk (Van Wagner, 1973; Moreno and Oechel, 1989; Lentile et al., 2006; Keeley, 2009).

This research aims to improve our understanding of the wildfire effects on P. pinaster, with particular focus on the strategies and mechanisms that this species is able to adopt to face the passage of fire.

We aim, not only, to verify the link between fire severity and P. pinaster ecophysiological responses, but also to understand the possibility of its survival in the short and long term. We hypothesize a tight relation between fire severity, in particular crown scorch, and tree damage in the short term, with needle defoliation triggering growth reduction and decrease in photosynthetic activity. Further, we hypothesize that stable isotopes can help assessing the ecophysiological processes activated by fire and can contribute to determine possible species survival. In particular, carbon stable isotope can be considered a proxy of changes in the gas exchange processes and can be used to estimate the ratio between photosynthetic activity (A) and stomatal conductance (gs) (Farquhar et al., 1982). Whereas δ <sup>18</sup>O can help elucidating the independent effect of A and gs on δ <sup>13</sup>C (Scheidegger et al., 2000). Indeed, high fire severity can damage the crown, reducing leaf surface and altering the plant's photosynthetic rate and efficiency (Beghin et al., 2011).

This research appeared extremely important from a forest management point of view, since the right assessment of actual tree damage and a better prediction of post-fire tree survival could avoid cutting down a scorched tree that could not be dead. Indeed harvesting trees, when is not necessary, could alter the post-fire germination and, generally, the carbon cycle and the ecosystem biodiversity.

#### MATERIALS AND METHODS

#### Study Area

The study area is the "Tirone Alto Vesuvio Nature Reserve" within the Vesuvio National Park (**Figure 1**), which covers a total area of 8482 hectares and suffers a strong anthropic pressure: almost 400000 inhabitants live in the 13 municipalities around the protected area. The Park, is characterized by a large plantation forest: while the Northern slope is covered by tall deciduous trees, the Southern side presents a large plantation dominated by P. pinaster Aiton and P. pinea L, with the presence of Pinus halepensis Mill, Pinus nigra Arnold (Picariello et al., 2000) and scattered patches of typical Mediterranean macchia vegetation. The study area shows the typical Mediterranean climate, with hot and dry summers and rainy winters with mild temperatures. In the last 40 years the average temperatures have gradually increased from 14 to 16◦C and the rainfall has been very erratic: some years (1980, 1984, 1996, 2005, 2009, 2010) have been characterized by heavy rains, others have been extremely dry (1977, 1999, 2003, 2012, 2016, 2017) (The Royal Netherlands Meteorological Institute, KNMI database Climate Explorer<sup>1</sup> , Trouet and Van Oldenborgh, 2013 and **Supplementary Figure S1**, data acquired on January 2018). The frequent drought seasons, the presence of strong winds and particularly flammable vegetation, associated with a strong anthropic pressure, make the area particularly vulnerable to forest fires (Sibilio et al., 2002).

To evaluate the effects induced by the fire of July 2017 on P. pinaster plantation, four sampling sites (around 1 ha) were chosen within the Tirone Alto Vesuvio Nature Reserve, from areas burned with different fire severity (**Table 1**). Fire severity

<sup>1</sup>https://climexp.knmi.nl

TABLE 1 | Sites coordinates, historical fire events, tree average diameter at breast height (DBH), tree height and severity parameters (percentage of crown reduction, height of scorch on the trunk) of 2017 wildfire.


HSS, high severity site; MSS, medium severity site; LSS, low severity site; CS, control site.

was evaluated based on parameters such as: presence and height of scorch on the trunk, presence of fire scars on the trunk, percentage of consumed crown and presence of damage to the roots related to 2017 fire.

Fire severity was high in the high severity site (HSS), in which the individuals of P. pinaster showed important damages to the canopy (70% crown reduction) and scorch on the trunk (over four meters). The severity was medium in the medium severity site (MSS), where the plants showed partial damage of crown (10%) and a scorch on the trunk that not exceed three meters. In addition, low severity was estimated in the low severity site (LSS), where individuals presented mild damage to the needles (average reduction of 2%) and scorch of the trunk of two and a half meters. Finally, individuals were sampled in control site (CS): an area situated into the Nature Reserve but where individuals did not show any scorch or damage to the trunk roots or to the crown trees, since they were not affected by the fire event of 2017.

Evidences of past fire events related to the four study sites were collected using the official fire reports, from 1988 to 2016, of the "Carabinieri Forestali per la Biodiversità (UTCB) of Caserta," a governmental agency, acting since 1974, as a park ranger force.

In addition to the large fire of the 2017, two surface fires, in 2015 and 2007, occurred in the MSS. Further in August 1993, an important wildfire (more than 240 he was burned) occurred on the Southern slope of Vesuvio, including the MSS and LSS sites. This information was, subsequently, related to dendrochronological (especially to fire scars) and isotope data. Finally, the LSS site was subjected to prescribed fires in 2014 and 2016, promoted by the Campania Region (Battipaglia et al., 2017).

#### Sampling and Processing of Tree Cores

In December 2017, for each study site, 15 dominant trees (trees with crowns extending above the general level of the main canopy and receiving full light from above and partly from the sides-

Adams et al., 1994) were sampled with a 5 mm increment borer (Haglöfs, Långsele, Sweden), and from each individual two cores were collected, at 1.3 m above ground, at east and west direction.

The total of 120 collected cores were fixed on specific wooden supports and, subsequently, subjected to a sanding process, to facilitate tree-ring identification. The measurement of the treering growth was carried out through the LINTAB system: a stereo-microscope connected to a computer, in which a specific software called TSAP-Win is installed and allows to elaborate a series of representative curves of each individual plant growth trend. After measuring each sample, first a visual comparison, then a statistical synchronization of the curves, known as crossdating, was performed using the Gleichaeufigkeit index (GLK), which evaluates the correlation between the different series (Eckstein and Bauch, 1969).

Successively, the data were statistically analyzed using the COFECHA software, which allowed checking the quality of crossdating, providing indications on the number of years to add or subtract to the chronologies.

Finally, the elementary chronologies have been standardized using ARSTAN (Cook and Holmes, 1986). Standard chronologies were detrended using the smoothing spline function, with a 40-year step for the chronologies of MSS site and a 10-year step for those relating to the HSS, LSS, and CS sites. The mean chronology was calculated through an arithmetic mean, while the stabilize variance was determined through the Keith Briffa rbar-weighted method. Finally, a different running bar was used for each site, allowing to maximize the expressed population signal (EPS).

## Sample Pretreatment and Isotopic Analysis

For each site five cores were selected for isotope analysis using the Gleichaeufigkeit (GLK) which evaluates the correlation between the different series (Eckstein and Bauch, 1969). The five selected cores presented the best cross-dating (GLK > 0.70) with the corresponding average chronology. The annual rings of the last 30 years (1988–2017) were manually removed and divided into late-wood and early-wood. Only for the CS site, with the youngest trees, the isotopic analyzes were performed across a shorter time period (1994–2017). Further analyzes were performed on an individual of the MSS who had an obvious fire scar (see **Supplementary Figure S2**) belonging to a past fire event.

Subsequently, the samples were milled using a pulverizing mill and were weighed precisely and encapsulated in a tin or silver capsules for the carbon or oxygen isotopic measurements, respectively.

The carbon and oxygen stable isotope composition was measured by continuous-flow isotope ratio mass spectrometry (Delta V Advantage, Thermo Scientific) at IRMS lab of University of Campania. The δ <sup>13</sup>C series were corrected for the fossil fuel combustion effect (Francey et al., 1999).

#### Climate Analysis

Climate data of minimum, maximum, and mean monthly temperature and total monthly precipitation for the period 1975–2017 were downloaded from the CRU TS3.23 gridded dataset at 0.5◦ resolution data (Harris et al., 2014) since Zalloni et al. (2018) demonstrated for the same area the high significant correlation with local meteo stations. A Pearson's linear correlation function analysis (P < 0.05), performed using Excel, was implemented between climate data and treering width and isotope data. Temperature and precipitation data were seasonally grouped from December of the previous year to February of the next year, in order to cover all the season which could influence tree-ring growth in Mediterranean species (Cherubini et al., 2003; Balzano et al., 2018). Those relationships help clarify the role of climate, namely temperature and precipitation, on past tree growth and to understand how much it influenced tree growth in CS site, not affected by fire.

# RESULTS

#### Dendrochronological Data

The standardized average chronologies (**Figure 2**) of the individual study sites showed a discrete synchronization: of the 43 years analyzed, the growth trend was common for 25 years. The good synchronization of the chronologies of the study areas is confirmed by the EPS value > 0.85 (**Table 2**). Also, the average sensitivity values of the four sites are very similar (always < 0.25): the sampled trees can be considered compliant, even if it is necessary to underline that the plants of the control area have a particularly high MS.

The HSS chronologies (**Figure 2A**), showed a good consistency in growth trend (EPS = 0.9; GLK > 60; MS = 0.12). In all individual chronologies, 1999 presented a simultaneous growth increase while a drastic growth reduction was observed in 2008, 2012, and 2017.

The MSS and LSS sites (**Figures 2B,C**) are very similar both in terms of age (mean age MSS = 38 years, mean age LSS = 37 years) and in terms of growth. Some years present a common increase of growth (e.s., in 1989, 1992, 1998, 2002) or a common decrease of growth (es., in 1993, 1997, 2000, 2003). In particular, a significant difference in growth between sites was observed in the period 2014–2016 (**Figure 2C**), in which there was an increase in tree growth in the individuals of the LSS site.

Finally, the CS (**Figure 2D**) presented the youngest trees (average age = 16 years) with positive growth peaks during 1999, 2002, 2006, 2010, 2014, 2016 and negative growth in 2000, 2003, 2005, 2007, 2011, 2013, 2015.

The tree-ring width showed that in 2017 all the sampled trees recorded a decrease of growth both in early-wood and latewood, in comparison to the previous year 2016 (**Figure 3**). In particular, in comparison to 2016, at HSS the total decrease in growth was 55,6% (early-wood decrease of 49,8%, late-wood 72%) while at the MSS of 55,7% (early-wood decrease of 65%, late-wood 38,5%) and LSS of 49,5% (early-wood decrease of 63,2%, late-wood 27,5). Finally, at CS the total decrease of the growth was 62,4% (earlywood of 66,4%, late-wood 50,8%).

The tree-ring widths of 2017 of each individual of the LSS, MSS (see **Supplementary Figure S3**) and HSS sites (**Figure 4**), were related to the percentage of destroyed crown. In the

FIGURE 2 | Individual standardized chronologies in gray and average chronologies in black for (A) high severity site (HSS), (B) medium severity site (MSS), (C) low severity site (LSS), (D) control site (CS).

TABLE 2 | Statistical parameters of the indexed average chronologies of the four areas.


TRW, tree ring width; EPS, expressed population signal; MS, mean sensitivity.

HSS a strong negative correlation was found, suggesting that the decrease in growth is directly proportional to the amount of needles destroyed by the fire: the plants that have a 90% crown reduction, showed a strong decrease in growth. In particular, relating the percentage of scorched crown with the

early-wood and late-wood ring-width, we observed that the highest negative correlation was found with late-wood. Indeed, the 2017 fire occurred in July, the period when the late-wood is normally formed.

#### Carbon Isotopes

The HSS presented a δ <sup>13</sup>C mean value, recorded between 1983 and 2017, of <sup>−</sup>26.3 <sup>±</sup> 0,4<sup>h</sup> (**Figure 5A**), at MSS and LSS the average δ <sup>13</sup>C, measured between 1975 and 2017, was <sup>−</sup><sup>26</sup> <sup>±</sup> 0,14<sup>h</sup> for MSS and <sup>−</sup>26.01 <sup>±</sup> 0,3 for LSS (**Figures 5B,C**), finally in the CS the C isotope values, for the period 1994 and 2017, was <sup>−</sup>25.7 <sup>±</sup> 0,05<sup>h</sup> (**Figure 5D**). The

δ <sup>13</sup>C site-chronologies showed different trends with only MSS and LSS presenting a high correlation (r = 0.26, P < 0.05). However, some years influenced the δ <sup>13</sup>C values of the four sites in the same way: negatively during 1993, 2012, 2014, positively during 1994, 2006.

While the 1993 fire drastically lowered the δ <sup>13</sup>C in all sites, the fires of 2007 and 2015 caused the decrease only in MSS site.

The different trends are also evident for δ <sup>13</sup>C measured in the early-wood and in the late-wood of the four sites (**Figure 5**) with the exception of particular years (2000, 2001, 2012, 2014, 2016, and 2017 for early-wood and, 1990, 1993, 1994, 1996, 2005, 2006, 2013 for late-wood), in which the isotopic values showed a similar response (relationships rise or fall at the same time). During the 2017 the δ <sup>13</sup>C values recorded in whole wood and particularly in late-wood resulted to increase at the HSS and MSS sites, compared to the previous year. At the LSS the δ <sup>13</sup>C of whole wood was almost unchanged in 2017 compared to 2016, with slightly decrease in late-wood while at CS we observed a decrease in δ <sup>13</sup>C.

#### Oxygen Isotopes

The δ <sup>18</sup>O values of the four sites varied in a range between 23.8 and 28.39h for the study period (**Figure 6**). The four chronologies showed a low synchronization among each other with only few years presenting the same trend, i.e., 2001, 2002, 2011, 2012, and 2017 both in whole wood and in each wood compartment (early-wood and late-wood). In particular δ <sup>18</sup>O of 2017 strongly increased in all the sites.

#### Relationships Among Climate, Growth, and Isotopes

The mean temperature, especially during the summer (JJA = June, July, August), was negatively correlated with the annual growth of the plants of all the four sites (**Table 3**). While the mean temperature of spring (MAM = March, April, and May) and autumn (SO = September, October) led to an increase of δ <sup>13</sup>C in the sampled individuals. The mean temperature of spring (MAM) and of summer period (JJA)

FIGURE 5 | δ <sup>13</sup>C chronologies (± standard deviation) for early-wood (gray line), late-wood (dashed gray line), and for the whole wood (black line) for (A) high severity site (HSS), (B) medium severity site (MSS), (C) low severity site (LSS), and (D) control site (CS).

site (HSS), (B) medium severity site (MSS), (C) low severity site (LSS), and (D) control site (CS).

TABLE 3 | Correlations between whole width (TRW), early-wood (TRW EW), late-wood (TRW LW), δ <sup>13</sup>C of whole wood (δ <sup>13</sup>C TOT), δ <sup>13</sup>C C of early-wood (δ <sup>13</sup>C EW), δ <sup>13</sup>C of late-wood (δ <sup>13</sup>C LW), δ <sup>18</sup>O of whole wood (δ <sup>18</sup>O TOT), d δ <sup>18</sup>O, of early-wood (δ <sup>18</sup>O EW) and δ <sup>18</sup>O of late-wood (δ <sup>18</sup>O LW) of the four sites, with monthly mean temperature (Tmean) (MAM, March, April, and May; JJA, June, July, and August; SO, September and October; ND, November and December) and total annual precipitation (Ptot).


R indicates the correlation coefficient of Pearson, with the corresponding significance (∗P < 0.05, ∗∗P < 0.01; ∗∗∗P < 0.001) calculated according to Student's t-test.

triggered the increase of δ <sup>18</sup>O. A high positive correlation was found between total annual rainfall and tree growth of HSS, LSS and CS sites. While negative correlation resulted between total precipitation and the late-wood δ <sup>13</sup>C of LSS site.

#### Isotopic Response of an Individual With 1993 Fire Scar

The analysis of stable isotopes, focused in the period 1992– 2001, showed how, due to the fire event of 1993, the δ <sup>13</sup>C and the δ <sup>18</sup>O increased sharply (**Figures 7A,B**). The carbon isotopic composition spanned from <sup>−</sup>28.24<sup>h</sup> of the spring '92 to <sup>−</sup>23.51<sup>h</sup> of summer '93, while oxygen ranged from 20.33h of 1992 to 23.39h of summer '93. After 1993, the δ <sup>13</sup>C values were stabilized while the δ <sup>18</sup>O continued to increase, reaching, in 1995, a value of 28h. The normal trend was recovered only in 1998.

### DISCUSSION

### Fire Effect on Tree Growth

Pinus pinaster trees sampled in the three burned sites (HSS, MSS, and LSS) showed a decrease in tree growth in 2017, in particular in the late-wood at HSS. Indeed, the fire event occurred in July, when the late-wood is formed in several woody species growing on Vesuvio area (Balzano et al., 2018). Reduction in growth has been described as a short term effect of fire in many species and across different ecosystems (e.g., Hoffmann and Solbrig, 2003; Werner, 2005; Werner et al., 2006; Goldammer, 2007; Murphy et al., 2010; De Micco et al., 2013; Battipaglia et al., 2014b) and it has been interpreted as an abrupt narrowing of growth rings after fire (Schweingruber, 1993; Stahlea et al., 1999) mainly due to cambium or crown damage (Pausas et al., 2003). When the tree is seriously damaged, the formation of new cell can be interrupted in the injured sector of the trunk or branch and fire scars can be formed (De Micco et al., 2014). No fire scars have been observed after the 2017 fire in all the burned sites, further all the sampled trees, including the one growing in the HSS, were able to complete the 2017 seasonal wood formation.

For P. pinaster, the bark is considered the main adaptive trait in response to forest fires (Ryan et al., 1994). As for P. pinea, the bark of the analyzed species is laminated, a peculiarity that allows to the outer layers of the trunk a gradual exfoliation during combustion, which contributes not only to the dispersion of heat, but also to an increase in the time necessary for killing the vascular cambium (Peterson and Ryan, 1986b; Rego and Rigolot, 1990; Battipaglia et al., 2016). The bark thickness is strongly dependent on tree age (Fernandes and Rigolot, 2007), indeed its diameter doubles when the plant reaches 10 years of age. The maritime pine (P. pinaster) growing at the study sites presented a DBH (diameter at breast height) > 20 cm, with a fairly thick bark (between 1.5 and 4 cm), which guarantees a good defense against fire (Burrows et al., 2000). This could explain why, even if the sampled individuals in the HSS, MSS, LSS presented a considerable bole char height, reaching the 4 m in the HSS, no fire scar were formed.

For the majority of conifers crown volume damaged is considered the most important factor determining mortality (Sieg et al., 2006). Previous studies (Freitas, 1995) demonstrates that the needles of the P. pinaster are less susceptible to thermal trauma than those of Pinus halepensis and P. pinea. In fact, maritime pines are able to survive at temperatures between 55 and 65◦C, for a time of exposure to the flame of 60 s. Also, the apical buds show a remarkable resistance to combustion: the protection offered by the surrounding long needles shield the gem from the connective heat of the flame (Michaletz and Johnson, 2006). Therefore, it seems evident that the burning of the needle in P. pinaster is not always a symptom of the destruction of the crown: even in the case of complete defoliation, the plant has been found able to survive (De Ronde, 1983). Experimental prescribed

burning, carried out on the maritime pine by in McCormick (1976), have allowed to report the damage to the foliage with the respective growth rate, showing that, in adult individuals, a slight damage of the crown does not lead to a drastic decline in growth. On the contrary, when the crown damage exceeds 25%, it is possible to find a substantial decrease in growth. Our findings are in agreement with McCormick's studies: the plants of the MSS and LSS that showed a reduction of the crown of 10% did not show particular growth decreases, while the individuals sampled at HSS, which experienced a conspicuous damage to the crown (in some trees the defoliation was equal to 90%) showed a sharp decrease in growth. Further, the ability of P. pinaster to survive, although the presence of a large fire scar and a possible severe defoliation, is demonstrated by the individual sampled in the MSS presenting the 1993 fire scar. Dendrochronological analyses of that individual highlighted that the plant survived to the 1993 wildfire growing until nowadays. According to the treering dating, that individual, similarly to the rest of the pines in the stand, in 1993 was more than 10 years old with a bark certainly greater than 1 cm. Ryan et al. (1994) applying a prescribed fire on P. pinaster stand showed that, with a bark greater than 1 cm of thickness, a scar cannot represent a significant contribution to the tree mortality.

The reduction in growth recorded in 2017, in comparison to the 2016, was evident also in the individuals sampled in the CS and this could be due to the extreme high temperatures recorded in 2017. Indeed, this year has been considered as an extreme dry year (World Meteorological Organization, 2018).

The climate- growth correlations reported in **Table 3** showed the importance of temperature on the tree ring width for the whole period. The high temperatures can cause a change in the constant kinetics of the RuBisCo, leading to a consequent decrease in carboxylation rate (Medlyn et al., 2002) and thus to a reduction in tree growth. On the other hand, also precipitation affected the growth of the sampled trees: indeed, in all the four chronologies a reduction in growth has been found after particular dry years: for example, the low rainfall in 1999 (455 mm/year) determined a drastic decline in growth in 2000 in all the trees.

In addition to the climatic factors, competition can also play an important role in the growth rate: dendrochronological analyzes have highlighted how the prescribed burning applications, carried out in 2014 and 2016 in the LSS, have led to a sharp increase in the growth of maritime pine. Prescribed burning, in addition to reducing in competition, through the biomass reduction of the herbaceous and shrubby species, has determined a large nutrients mineralization, ensuring to the dominant individuals to take advantage of the favorable conditions. Similar results were also observed in other prescribed burning studies performed on individuals of Pinus halepensis (Battipaglia et al., 2014b; Valor et al., 2018) and P. pinea (Battipaglia et al., 2016).

# Ecophysiological Responses of P. pinaster to Fire

The isotopic analyses related to 2017 of the four sampled sites allowed to better understand the different processes triggering trees responses to fire. Indeed, the growth reduction found in all the trees was due to changes in ecophysiological mechanisms, mainly related to fire severity. Trees of HSS and MSS sites showed in 2017 a significant increase of δ <sup>18</sup>O and a slight increase of δ <sup>13</sup>C especially of late-wood, compared to previous years. The increase in the oxygen isotopic ratio may be associated with a decrease in stomatal activity, whilst the increase of δ <sup>13</sup>C can indicate a lower partial pressure of CO<sup>2</sup> in the intracellular spaces of the leaf, due to both photosynthetic activity and stomatal conductance. Therefore, considering both isotopes, according to the dual-isotope approach (Scheidegger et al., 2000) we could hypothesize that the plants to protect themselves from strong stress condition, due to fire, and as a consequences of the serious observed crown damage, closed their stomata and lowered their photosynthetic activity. These conditions, added to the high temperatures and drought conditions through the growing season, explain the observed reduction in growth of 2017 tree ring. Those results are in agreement with the studies carried out by Battipaglia et al. (2014b) on of Pinus halepensis, in several forest fires in Southern France, that showed a simultaneous increase in δ <sup>13</sup>C and δ <sup>18</sup>O in occurrence of the fire events. In the same studies the authors demonstrated that the Pinus halepensis, considered a species highly vulnerable to fire, was able to survive and to recovery from fire damages, presenting a low mortality rate. To better understand the probability of death of maritime pine, exposed to high severity fire, the study carried out on the individual with the deep fire scar, was crucial. The study showed that, in the periods before the 1993 fire, the

tree with the fire scar, located in the MSS, was not subjected to stressful climatic conditions since in the period between 1985 and 1992, precipitation and temperatures were rather constant with low anomalies. In the year of fire, the values of δ <sup>13</sup>C and δ <sup>18</sup>O appeared to drastically increase (**Figure 7**) and, therefore, the ecophysiological responses were similar to that recorded in the HSS individuals after the 2017 fire.

In the following years from 1993 onwards, while the δ <sup>13</sup>C decreased, returning to the average values, the δ <sup>18</sup>O showed a progressive increase until 1995. This combination suggests that, although the stomatal conductance gradually stabilized over time, a severe defoliation of the plants, at least in the first years after fire, resulted in a lower photosynthetic capacity (Battipaglia et al., 2016), leading a lower growth rate (as recorded by the dendrochronological analyzes). After that period, the growth and the ecophysiological processes of that individual returned to the average values.

The individuals sampled at MSS and LSS showed a strong synchronization in the δ <sup>13</sup>C and δ <sup>18</sup>O chronologies along all the time period (R = 0.26 and R = 0.24, respectively). However, between 2014 and 2016, the application of the prescribed burning in the LSS determined, in the plants of that site, an increase of δ <sup>13</sup>C while the δ <sup>18</sup>O remained unchanged. The variation of the δ <sup>13</sup>C of the plants of the LSS can be connected to an increase of the photosynthetic activity, triggered by prescribed burning. A positive effect of fire was recorded also in the 2007 when a surface fire has affected individuals of the MSS and when a decrease of δ <sup>13</sup>C and δ <sup>18</sup>O was recorded until 2009, indicating a possible favorable effect of reduction in competition for water among survived plants, as found in a study by Beghin et al. (2011). On the other hand the August 2015 fire, which involved the MSS plants, did not result in significant changes of the isotopic response or drastic changes in growth rate. This allows us to hypothesize that the analyzed pinewood was only marginally affected by this event.

Finally, the isotopic analyses of CS trees showed, in 2017, a moderate decrease of δ <sup>13</sup>C and a drastic increase in δ <sup>18</sup>O. This variation, compared to the previous year, can indicate a lower photosynthetic activity and an unchanged stomatal conductance (Scheidegger et al., 2000) probably related to the extreme hot conditions experienced in that summer by all the plants. The climate-growth correlations demonstrated how the minimum temperature of the summer period (r = −0.56∗∗), and, in particular, of August (r = −0.66∗∗∗), has a negative influence on the growth of studied trees. The high temperatures, indeed, besides determining a reduction of the chlorophyll pigments and the denaturation of some thioacid proteins, can cause the inhibition of the photosynthetic process (Bussotti et al., 2012). When air temperature exceeds the compensation point (temperature at which the amount of CO<sup>2</sup> fixed by photosynthesis is equivalent to that released by respiration), the photosynthetic process is not able to replace the carbon used for respiration. This process implies a consumption of carbohydrates reserve that led the plant to a slowdown or a stop of its growth (Di Toppi et al., 2018). Climate conditions could also represent a cumulative factor of stress contributing to a delayed mortality. Catry et al. (2010) assessing the vegetative condition of 1040 burned trees from 11 different species during the first years after the 2003 wildfire, demonstrated how the tree mortality largely increased after the 2005 drought. Their results emphasize the importance of monitoring the tree health in the following years. The burned trees in the study sites showed, at the end of 2018, a very limited mortality rate, with only the 2–10% of plants reducing their vigor. We will continue monitoring the area and the analyzed trees since our work want to be a contribution on the fire ecology of the P. pinaster species, underling the importance of evaluating the ecophysiological responses of the species to fire severity in order to valuate the most suitable and effective silvicultural operations mainly in post-fire forest management. Indeed, P. pinaster trees affected by fire, with needles scorched or consumed, are not necessarily dead and before to remove every blackened trees, it is important to assess the real damage, waiting for the next growing season, in particular in Natural parks or in areas not close to roads or anthropic settlements. Savage logging could produce severe damage to soil stability and biodiversity, and in particular mechanical actions can largely increase seedling mortality (e.g., Martinez-Sanchez et al., 1995; Fernandes et al., 2008), negatively affecting the ratios final seedling density/initial seedling density and final seedling density/total viable seed dispersal (Vega et al., 2008).

# AUTHOR CONTRIBUTIONS

GB conceived and designed the study. FN performed sampling and analyses. SA and AE contributed to the analyses. FN and GB wrote the main part of the manuscript. All authors contributed to interpretation of the overall data and wrote specific parts, made a critical revision of the whole text, and approved the submitted version of the manuscript.

#### ACKNOWLEDGMENTS

The authors thank Giuseppina Vicario for help during fieldwork. The authors wish to thank the Carabinieri Forestali per la Biodiversità (UTCB) of Caserta and Vesuvio National Park for access to the protected areas and logistic support. The authors also thank the Valere Program of University of Campania "L. Vanvitelli".

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.00539/ full#supplementary-material

FIGURE S1 | Trend of total annual precipitation in green (expressed in millimeters/year) and average temperatures in blue (expressed in ◦C) recorded from 1975 to 2017 and coming from the KNMI Climate Explorer database.

FIGURE S2 | Individual of P. pinaster belonging to the MSS which has an important lesion on the trunk due to a previous fire, dated 1993.

FIGURE S3 | Relationship between ring growth of 2017 and the percentage of crown scorch of the MSS trees (indicated with in yellow symbols) and LSS trees (indicated in red).

#### REFERENCES

fpls-10-00539 April 25, 2019 Time: 16:16 # 10


characteristics and post-fire effects. Int. J. Wildland Fire 15, 319–345. doi: 10. 1071/WF05097


coperture vegetali," in Proceedings of the XII Congresso Nazionale della Società Italiana di Ecologia – S.It.E atti, (Urbino).


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

The handling Editor is currently organizing a Research Topic with one of the authors GB and confirms the absence of any other collaboration.

Copyright © 2019 Niccoli, Esposito, Altieri and Battipaglia. 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.

# The Vulnerability of Qilian Juniper to Extreme Drought Events

#### *Xiaofeng Wang1,2, Bao Yang1\* and Fredrik Charpentier Ljungqvist3,4*

*1 Key Laboratory of Desert and Desertification, Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences, Lanzhou, China, 2 University of Chinese Academy of Sciences, Beijing, China, 3 Department of History, Stockholm University, Stockholm, Sweden, 4 Bolin Centre for Climate Research, Stockholm University, Stockholm, Sweden*

Identifying which trees are more vulnerable to extreme climatic events is a challenging problem in our understanding of forest and even ecosystem dynamics under climate change scenarios. As one of the most widely distributed tree species across the arid and semi-arid northeastern Tibetan Plateau, Qilian juniper (*Juniperus przewalskii* Kom.), is the main component of the local forest ecosystem, providing critical insurance for the ecological security of the surrounding areas. However, this species's ability to cope with climate extremes (especially drought) has not been adequately assessed. Here, we apply a dendroecological approach that considers indices of resistance and resilience to quantify the vulnerability of Qilian junipers to the extreme drought events of 1957, 1966, 1979, and 1995. A total of 532 Qilian juniper trees from different age stages (100–1,100 years) and altitudes [3,500–4,000 m above sea level (a.s.l.)] were studied to assess their response characteristics during these four drought extremes. We conclude that drought extremes have a significant negative impact on the growth of Qilian juniper. The oldest Qilian junipers at the lower altitudes constituted the most vulnerable populations across the northeastern Tibetan Plateau and were characterized by the lowest resistance values, the narrowest annual rings, and the highest proportion of missing rings during the four drought years. Tree resilience after droughts was strongly related to the intensity of the drought event and did not change with tree age or elevation. A threshold of tree tolerance to drought may exist, with the more vulnerable tree individuals (e.g., the oldest Qilian junipers from lower altitudes) being exposed to the highest mortality risk when drought intensity exceeds the threshold value. Such a threshold needs further consideration, through the study of trees that have died (or are about to die) due to extreme droughts.

#### Keywords: Qilian juniper, vulnerability, drought, age effect, elevation, global warming

# INTRODUCTION

Droughts can reduce tree growth and forest productivity through changes in photosynthesis rate (Grassi and Magnani, 2005; Hinckley et al., 1979), carbon assimilation (Chaves, 1991; Lawlor and Cornic, 2002), phenology (Misson et al., 2011), tree morphology (Abrams et al., 1992; Aspelmeier and Leuschner, 2006), and others, with adverse impacts on ecosystem stability. Severe droughts may push tree growth decline beyond its biological thresholds (Folke et al., 2004), triggering widespread tree dieback (Camarero et al., 2015; Hoffmann et al., 2011) and even tree mortality (Allen et al., 2010; Anderegg et al., 2013). However, tree growth performance during extreme drought episodes is complicated and varies with species, age, size, population features, and the geographical distribution

#### *Edited by:*

*Giovanna Battipaglia, University of Campania Luigi Vanvitelli, Italy*

#### *Reviewed by:*

*Peter Prislan, Slovenian Forestry Institute, Slovenia Philipp Hochreuther, University of Erlangen Nuremberg, Germany*

> *\*Correspondence: Bao Yang yangbao@lzb.ac.cn*

#### *Specialty section:*

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

*Received: 06 March 2019 Accepted: 29 August 2019 Published: 27 September 2019*

#### *Citation:*

*Wang X, Yang B and Ljungqvist FC (2019) The Vulnerability of Qilian Juniper to Extreme Drought Events. Front. Plant Sci. 10:1191. doi: 10.3389/fpls.2019.01191*

**214**

of the trees (Gazol and Camarero, 2016; Lebourgeois et al., 2013; Merlin et al., 2015; Vitali et al., 2017), which limits our understanding of the mechanism of tree response to drought. Research is increasingly focusing on areas threatened by severe droughts and heat waves to investigate drought damage based on the diverse physiological and distribution characteristics of trees (Brouwers et al., 2013; Rennenberg et al., 2006; Tyree et al., 1994) with the aim of assessing the reaction of these trees to drought events. Nonetheless, our knowledge of tree response to drought extremes is still limited, especially in some long-standing vulnerable and sensitive environments such as the northeastern Tibetan Plateau (NETP).

Qilian juniper (*Juniperus przewalskii* Kom.) is the main component of natural forests on the NETP and is widely distributed up to elevations of 3,500–4,100 m above sea level (a.s.l.) on the sunny and partly sunny slopes (Shao et al., 2005) of the Gobi Desert margins. Although their distribution is relatively scattered, it is one of the most critical components of the local fragile terrestrial ecosystem due to its wide distribution and the high number of trees, which play a crucial role in maintaining ecological stability and preventing desert expansion or erosion. However, these trees are threatened by insufficient water supply. Many dendrochronological studies conducted in this region have consistently found that radial growth of Qilian juniper is primarily limited by water shortage (Gou et al., 2013; Gou et al., 2015; Qin et al., 2015), and growth decline and mortality related to droughts have also occurred recently (Fang et al., 2015; Yu et al., 2015; Liang et al., 2016a), despite slight increases in annual precipitation during the past few decades (Li et al., 2008; Shi et al., 2007). When the wetting trend terminates, or if a transition to drier conditions occurs due to rising global temperatures (similar to other parts of inner Asia), the forests in this region will immediately be exposed to more severe water deficits than at present, exposing those trees with low drought resistance and resilience to a greater risk. Although many studies have considered the relationships between radial growth of Qilian juniper and the climate conditions on the NETP (Liu et al., 2009; Shao et al., 2005; Shao et al., 2009; Shao et al., 2010; Yang et al., 2013; Yang et al., 2014; Yang et al., 2017a; Yang et al., 2017b; Yang et al., 2019; Zhang et al., 2003), most focus on the reconstruction of paleoclimatic conditions. Fang and Zhang (2018) explored the resilience of Qilian juniper after four extreme drought events on the NETP and concluded that tree resilience to drought has increased over the past few decades. Unfortunately, their study did not involve the differentiated response of tree individuals to drought extremes, and the ability of trees to cope with drought remains insufficiently known. We now need to compare how the responses to hydraulic deficits of trees under different site conditions (e.g., altitude) and within different age bands vary, and identify which trees are more resistant and resilient to water shortage, making them more likely than other trees to survive after drought extremes.

To answer these questions, we studied 532 Qilian juniper trees from diverse age stages (100–1,100 years) and altitudes (3,500–4,000 m a.s.l.) to explore the variance in tree vulnerability to extreme droughts and to assess the main factors influencing the ability of trees to resist drought events. More specifically, the resilience indices presented by Lloret et al. (2011), which measure tree resilience, resistance, and recovery using ring width data from individual trees, were adopted to quantify the vulnerability characteristics of trees suffering from drought extremes in 1957, 1966, 1979, and 1995. The generalized linear model and superposed epoch analysis (SEA; Chree, 1913; Chree, 1914; Lough and Fritts, 1987) are used to verify the reliability of the initial results obtained in our study.

# MATERIALS AND METHODS

#### Tree-Ring Data and Dendroecological Analysis

In order to investigate the possible disparate responses among different Qilian juniper individuals under drought stress, we re-examined published data from the NETP and classified treering width series according to the following tree age criteria: age class 1 (AC1), 100–300 years; age class 2 (AC2), 301–500 years; age class 3 (AC3), 501–700 years; age class 4 (AC4), 701–900 years; and age class 5 (AC5), 901–1,100 years. We divided each age class into high altitude (3,800–4,000 m a.s.l.) and low altitude (3,500–3,700 m a.s.l.). Data from a total of 532 trees of Qilian juniper were assimilated from reported studies (Shao et al., 2009; Yang et al., 2014), distributed across 18 sites in Wulan, Delingha, Tianjun, and Dulan (**Figure 1; Table 1**). To maintain consistency among all tree samples, we used only one increment core per tree; thus, the mean ring width sequence was used for trees having two cores. Due to the high altitude, the cold and dry climate, and the barren soil, other tree species generally do not thrive, such that almost all of the trees growing on the NETP are Qilian junipers. These scattered trees, together with alpine meadows and shrubs distributed between the bare rocks or in the lower valleys, constitute the local plant cover (**Figure 2**). From the perspective of ecosystem stability, the species composition in the NETP is relatively simple and has little ability to buffer external disturbance. Further detailed information on the topography, climate, and hydrology can be found in previous dendrochronological works (Shao et al., 2009; Yang et al., 2014).

Each tree-ring width series was detrended with a 10-year cubic smoothing spline in the R-environment (R Core Team, 2017) using the function "detrend" in the "dplR" package (Bunn, 2008) to eliminate non-climatic trends and preserve higher interannual variability (Cook and Peters, 1981). This step transforms the raw tree-ring width sequence into a standard and dimensionless ring width index. A mean standard chronology (STD) was established using the biweight robust mean of the detrended ring widths of each age class (Cook, 1985) with the function "chron" for the overlapping period 1950–2000. The chronologies were characterized by mean ring widths, Gleichläufigkeit (GLK), mean sensitivity (MS), and first-order autocorrelation (AR1) (Fritts, 1976). GLK represents the similarity in signals between chronologies (Schweingruber, 1988). MS measures the mean percentage change in year-to-year growth variations. AR1 represents the influence of the previous year's growth on the current year. MS and AR1 were calculated using ring width series, while GLK was based on the STD chronology.

FIGURE 1 | The location of the study area and sampling sites. The Qilian juniper site names and other geographic information refer to Shao et al. (2009) and Yang et al. (2014).

TABLE 1 | Descriptive statistics for the Qilian juniper chronologies developed in this study. The sample sites of DLH1–6, WL1–4, and TJ1 were obtained from Shao et al. (2009) and MNT, XDC, XHC, SBP, ZHG, SGX, and NDC from Yang et al. (2014). Rchron-PDSI refers to the correlation coefficient between tree-ring chronologies and June scPDSI(the self-calibrated Palmer Drought Severity Index (1850-2012 scPDSI dataset, http://www.cgd.ucar.edu/cas/catalog/climind/pdsi.html); Dai, 2011) during the study period. PMR refers to the percent of missing rings during drought years compared to all tree-ring width data for the four selected drought years. MS and AR1 represent mean sensitivity and first-order autocorrelation, respectively. AC, age class.


# Selection of Extreme Drought Years Corresponding to Tree Growth Decline

Drought triggered by anomalously low precipitation and/or high temperature might yield exceptionally narrow tree rings over a large area in moisture-limited environments (He et al., 2018; Wang et al., 2015). The following two steps were applied to identify extreme drought years and ensure that they were primarily responsible for the sharp decline of tree growth (rather than other growth disturbances such as fire or insect pests).

1) Many previous dendrochronological studies in the adjacent area indicated that tree growth is primary restricted by the water shortage in June (Qin et al., 2013; Shao et al., 2005; Shao et al., 2010; Yang et al., 2014). Higher temperatures will enhance potential evapotranspiration, resulting in a further increase of moisture deficit during the same period (**Figure 3B**). Hence, we adopted the scPDSI sequence, which includes precipitation and surface air temperature (Dai, 2011), to identify drought episodes associated with tree growth decline.

FIGURE 2 | An overview of the distribution characteristics of Qilian junipers in the study area. The typical Qilian juniper individuals on the NETP (A) and their barren growth environment and dotted distribution (B, C, D).

Firstly, we performed correlation analysis between the treering chronology of each age class and scPDSI sequence, using the DendroClim2002 software (Biondi and Waikul, 2004), to assess drought responses of tree growth. Next, considering that the fastest radial growth period and main contribution to the annual ring width occurred in June (Gou et al., 2013; Zhang et al., 2018), we examined the scPDSI of each June during the study period and considered those years with values below 1.5 standard deviations from the mean as alternative drought years for subsequent analysis.

2) The "pointer year" (Schweingruber et al., 1990), referring to a year with notable growth variance occurring at the stand level, was calculated from the ring width index chronology of each age class. In this study, we only focused on growth declines associated with drought events. The pointer years were calculated using the R package "pointRes" (van der Maaten-Theunissen et al., 2015), with a threshold defined as 60% of tree-ring series showing a growth decrease of at least 30% compared to the average growth during the 2 preceding years.

We employed the drought years obtained from the first step to check the growth characteristics of tree-ring width series during these drought episodes and used the years with significant tree growth decline identified by the second step to control the number of extreme drought years ultimately adopted. Moreover, we ranked the intensity of the four drought years selected above based on precipitation and temperature anomalies during June, and ranking the driest year as fourth, the second driest year as third, and so forth, for further analysis.

## Quantification of Tree Vulnerability to Drought

The vulnerability of individual trees to drought events was quantified through the resistance (Rt), recovery (Rc), and resilience (Rs) indices adopted from Lloret et al. (2011). Resistance describes the magnitude of the growth reduction during a drought year compared to the pre-drought years and measures the ability of trees to "buffer" drought impacts. Recovery measures the growth difference between the postdrought period and the drought year and quantifies the capability of the tree to recover from drought disturbance. Resilience measures the difference in tree growth before and after a drought year and quantifies the capacity to reach predisturbance performance levels.

These three indices were calculated using the following equations:

Resistance (Rt) = Dr / PreDr

Recovery (Rc) = PostDr / Dr

#### Resilience (Rs) = PostDr / PreDr

The "PreDr" and "PostDr" values were calculated with the average tree-ring width during the 2 years before and after a drought event. The "Dr" variable refers to the tree-ring width for the drought year. All calculations follow the algorithm described above and were conducted using the raw ring width sequence on each individual tree. We used the 2-year period as the reference time interval for two reasons: first, we found that the growth of trees generally returns to its previous level within 2 years after the drought event, and second, this time scale allows full consideration of the lag effects of drought while avoiding interference with other disturbance factors (such as the next drought event). Considering that the relationship among these resilience indices is relatively intuitive (resilience = resistance × recovery), we only used the resistance and resilience indices for further analyses.

SEA was also utilized to evaluate tree growth response to drought extremes and, more importantly, to verify the results that we obtained above. SEA was applied to each age class by generating composites of tree-ring width indices from lag –2 to lag +2 years relative to the average of the previous 2 years before each drought event, to obtain the range of growth decline. A bootstrap resampling procedure (using 1,000 random sets) was applied to estimate whether tree growth in these drought years was statistically significantly different from the random sets of other years (lag +1 year). All the analyses were performed using R version 3.4.3 (R Core Team, 2017).

#### Data Analysis

To facilitate the analysis, we performed cyclic elimination on the calculated resistance and resilience values by using the mean plus/ minus 3 standard deviations rule (Limpert et al., 2001) to remove the presence of a very few outliers (less than five data points for each index), which we found to have no significant effect on the results. The differences in each index between three age classes and between two elevation gradients were assessed by the non-parametric Kruskal–Wallis analysis of variance (ANOVA) test (Kruskal and Wallis, 1952; unlike other regression methods, this one does not require the original data to be normally distributed) and the Dunn– Bonferroni test (Dunn, 1961) for *post hoc* comparisons.

The importance of age and elevation to resistance and resilience was assessed through a generalized linear model. In order to more accurately assess possible controlling factors of resistance and resilience, the sorted drought levels, latitude, and longitude, as well as tree age and elevation, were all added to our model. Ultimately, we operated the model with the data groups AC1, AC3, and AC5, and verified the accuracy of the model with the actual values from groups AC2 and AC4.

#### RESULTS

#### Radial Growth and Tree-Ring Chronologies

Parallel radial growth patterns were detected among all age classes of Qilian juniper over the period 1950–2000 (**Figure 3A**). Whether at higher or lower sites, nearly all younger trees showed wider ring widths compared to older trees throughout the study period (mean ring width ranges from 0.46 to 0.27 mm for high altitude and 0.36 to 0.23 mm for low altitude). Moreover, trees growing at higher elevations showed wider ring widths compared to those of the same age at lower elevations. In particular, trees growing at lower altitudes or/and belonging to older age classes exhibited a greater proportion of missing rings during drought years than that of other tree individuals (**Table 1**). The mean AR1 of tree-ring width series was 0.27, showing that, during the study period, the growth conditions of the previous year had no significant effect on the tree-ring width of the current year. The MS of all trees exceeded 0.4, indicating a higher interannual variability of radial growth. The high GLK (0.87) among all chronologies suggests that the growth of trees within our study was highly consistent and that the chronologies are suitable for climate-growth analysis.

#### Extreme Drought Years

Based on the selection methods described above, the drought years associated with notably slow growth of Qilian junipers are 1957, 1966, 1979, and 1995. It should be noted that rapid growth decline has also occurred in 1998, but it was probably due to other factors than drought and thus not included in the subsequent analysis as a drought event. These four drought years were ranked according to the precipitation and temperature anomalies in June (with respect to 1961–1990). Of these, the most severe drought year was 1995, followed in descending order by 1966, 1979, and 1957 (**Figure 3B**). Moreover, the correlation analysis results of tree-ring chronologies and scPDSI showed that the correlation coefficient in June was the highest during the whole hydrological year (**Table 1**); therefore, using the scPDSI value of June to identify extreme drought years is a reasonable approach.

#### Changes of Resistance and Resilience Over Time

The resistance of Qilian juniper to extreme drought for all age classes showed a gradually declining tendency, with the resistance in 1995 being the lowest among the four extreme drought years (–13%, –44%, and –6% for low-latitude trees and –21%, –43%, and –9% for high-altitude trees compared to 1957, 1966, 1979, respectively, based on the algorithm (Rt1957 − Rt1995)/Rt1957; **Figures 4A, C**). Regardless of the age and elevation, the resilience of almost all trees remained nearly constant during the three drought years of 1957, 1966, and 1979. However, a sudden drop in 1995 is evident (Rs decreased by 38% compared to the average of the previous three drought years; **Figures 4B**, **D** and **5B**), with 90% of all sample trees showing resilience values less than 1 and 65% less than 0.75. Thus, both resistance and resilience values in 1995 were the lowest among the four drought years.

#### Comparison of the Radial Growth Responses in Different Age Classes

The resistance of Qilian juniper to extreme drought events showed a descending trend with increasing tree age (**Figures 4A**, **C**; **Table 2**). The oldest trees showed the lowest resistance during nearly all drought years (the average resistance of trees in AC5 was 11%/54% lower than in AC1 for high/low altitude; these differences were statistically significant in 1957 and 1995 for high altitude and in 1966 and 1979 for low altitude, with *p* < 0.05). Consistent results were obtained with the generalized linear model (**Figure 5A**). Likewise, the results from the SEA indicated that superposed tree-ring width indices during drought years were strongly reduced compared to the average of the previous 2 years and that the magnitude of the decrease was proportional to the age of the trees (**Figures 6A**, **C**). Furthermore, the proportion of missing rings during drought years in older Qilian junipers was higher than that of younger trees (**Figures 6B**, **D**), indicating that the older trees react more strongly to drought events. The resilience of Qilian juniper to drought events was somewhat less clear and did not change with increasing tree age (**Table 2**).

#### Effect of Altitude on Resistance and Resilience Variability

Qilian junipers from lower-altitude sites showed smaller resistance values than those from higher-altitude sites within the same age class (**Figures 7A**, **C**, **E**; **Table 2**), which is especially evident in AC5 (**Figure 7E**). We note that the age structures of the sampled trees were similar at high and low altitudes (**Figure 6E**) and that comparison of resistance at different altitudes was performed within the same age group: hence, the difference in resistance between trees at different altitudes was not directly caused by the age effect. Furthermore, results of the SEA and the number of detected missing rings indicated that the magnitude of growth decline during drought events was greater at lower elevation (**Figures 6B**, **D**). Changes in resilience with altitude did not follow a distinct pattern (**Figures 7B**, **D**, **F**) for Qilian juniper, but trees with resilience > 1 accounted for 69% at higher altitude and 64% at lower altitude. The location (latitude and longitude) cannot explain the variations of resilience; while latitude did have an effect on resistance, its effect was much weaker than tree age and altitude.

#### DISCUSSION

#### Age Effect on Drought Responses

Our results revealed varying resistance levels among the different age groups of Qilian junipers subjected to extreme

*post hoc* tests, *p* < 0.05). Red circles represent outliers, and vertical lines show the whiskers for the 5th and 95th percentiles of the data distribution.

TABLE 2 | The impacts of age, elevation, drought intensity, and latitude for the resistance and resilience of Qilian juniper to extreme drought events, calculated using the generalized linear models (GLMs), where DI = drought intensity, b = regression coefficient, SE = standard error, T = corresponding T statistic (for the partial test of H0: b(i) = 0), and p(T) = corresponding significance value.


drought events. These changes may be primarily associated with physiological characteristics as well as diverse drought influences on water/nutrient supply and demand relationships at different age stages. From the perspective of demand, older trees possess a more complex structure and much greater volume of aboveground woody tissues, with a thicker trunk and more abundant foliage, which increases the autotrophic respiration and consumes more carbon to meet the demand of normal life activities (Granda et al., 2017; Lucas-Borja and Vacchiano, 2018; Ryan and Yoder, 1997).

Regarding the supply, the lower hydraulic conductivity within the trunks and branches, coupled with the low level of photosynthetic capacity and stomatal conductance in older trees, may mainly explain their lower resistance to extreme drought events. In particular, the hydraulic resistance increases with tree age/height (Dawson et al., 1990; Ryan et al., 2006). For older Qilian junipers, their greater size means xylem fluid needs to be transported upward for a longer distance to reach the top/lateral branches and leaves, which requires higher tension in the xylem water column (Ryan and Yoder, 1997) compared to younger trees. When severe drought occurs during the growing season, this hydraulic architecture increases the risk of cavitation or embolism in the xylem of older trees (Hacke et al., 2001; Johnson et al., 2012; Savi et al., 2015; Tyree and Sperry, 1989), further

impeding xylem hydraulic function and exacerbating the water stress. In response, blade stomata of these trees close earlier to prevent further loss of xylem water. This, in turn, prevents carbon dioxide uptake and causes a decrease in the photosynthesis capacity, thus limiting the carbon assimilation in older trees. These factors reduce the competitiveness of older trees under drought conditions and ultimately trigger carbon starvation and growth decline or even tree mortality (Dulamsuren et al., 2019; McDowell et al., 2008; Sevanto et al., 2014). Many studies have suggested that photosynthesis and stomatal conductance of old trees are lower than those of younger individuals (Grulke and Miller, 1994; Kull and Koppel, 1987; Schoettle, 1994), consistent with the above arguments. Therefore, older trees in moisturelimited environments with less water and nutrient supply are already closer to a critical point during normal years, and thus less resistant to extreme drought events.

Contrary to our results, a few studies have shown that older trees, with their well-developed root systems, can more easily absorb inorganic salts and water from a deeper and broader extent of the soil; this helps to buffer water stress during drought (Fritts, 1976; Krämer, 1996; Rozas et al., 2009; Wu et al., 2013). However, this effect was not observed in Qilian juniper at our study sites. Although root depth and lateral root extension increased with aboveground canopy size, root architectural traits are more closely related to water infiltration depth to allow more effective absorption of soil water in arid environments (Schenk and Jackson, 2002). Evidence supporting this view is the characteristic of Qilian juniper as a shallow-rooted species,

with rooting depths ranging from 40 to 100 cm below the ground surface (Gou et al., 2013; Xu et al., 2011). Thus, while older Qilian junipers possess well-developed rooting systems compared to younger trees, their greater extent is mainly lateral and is not significantly deeper. Those well-developed roots of older Qilian junipers are produced to absorb water more effectively under normal hydrological conditions. However, based on the theory that smaller roots are more vulnerable to cavitation than stems (Sperry and Ikeda, 1997), it is these trees that are more exposed to the danger of cavitation during the dry periods, intensifying the poor water availability of older trees.

#### The Influence of Altitude on Radial Growth Reactions

The change in altitude theoretically affects the distribution of heat and moisture resources. In general, thermal stress increases, whereas water stress decreases with altitudes in alpine environments (Cavieres et al., 2006; Gao et al., 2017; Wilson et al., 2005). Greater water availability at higher altitudes than at lower altitudes allows trees at higher altitudes not only to grow more rapidly under normal hydrological years but also to more effectively buffer the adverse effects of water deficits during drought years. Our results showed that the average treering width at lower altitudes was invariably narrower, and that resistance to drought was significantly lower, than those at higher altitudes, consistent with the results discussed above.

The percentage of missing rings, which can reveal the severity of drought stress on tree growth and which is inevitably linked to

tree mortality (Liang et al., 2016a), was higher at lower elevations in our study. The radial growth of all tree samples used in this study was mainly limited by water deficit. Given that trees at different elevations will suffer different degrees of drought, due to changes in the distribution of precipitation and temperature along the altitudinal gradient during the same drought event, this may indicate that environmental conditions at lower altitudes are less favorable (lower water availability) than those at higher altitudes, prompting the trees in these areas to react more strongly to the extreme drought events. It is worth noting that the relatively higher temperatures at lower elevations could theoretically deepen the seasonal thaw depth of the permafrost in summer, thus alleviating the water deficit in these areas to some extent. However, the maximum thaw depth of the permafrost in this region is beyond 1.5 m (Huang et al., 2011), while the soil layer and the roots of Qilian juniper are mainly concentrated within the uppermost 1 m (Gou et al., 2013; Xu et al., 2011). Therefore, the changing contribution of summer permafrost melting to soil moisture along the elevation gradient does not ultimately change the pattern of lower-elevation sites facing greater moisture deficits.

The slightly increasing precipitation during the study period might have promoted tree growth at higher elevations, leading to increasing canopy cover and stem density; this could have intensified the competition among trees for soil moisture, causing a higher frequency of missing rings as observed by Liang et al. (2016a) in the central Qilian Mountains. However, this effect was not seen in our study, presumably because the canopy density of Qilian juniper within our study sites (**Figure 2**) is far less than that in the Qilian Mountains (0.3 *vs*. 0.6–0.8) (Qin et al., 2013; Liang et al., 2016a), and there is no distinct intraspecific competition among trees at higher altitudes.

## Climate Threshold–Related Growth Decline

Our results demonstrated that the resilience of Qilian juniper to extreme drought is not directly related to the elevation, age, latitude, or longitude of trees, but instead showed a significant negative correlation with the intensity of drought events. The mean resilience value fluctuated around 1, as shown in 1957, 1966, and 1979, indicating that these droughts did not seriously injure the Qilian juniper; indeed, the trees showed rapid recovery to pre-drought growth levels after the droughts. Resilience values significantly lower than 1 (e.g., in 1995) suggested that drought intensity was particularly high, with a strongly adverse impact on these trees, such that the majority of trees had not yet returned to their pre-drought growth status within 1 to 2 years following the drought episodes. Some of the above variability may also be attributed to patterns of water deficit at regional scales.

The average anomalies of precipitation and temperature (**Figure 3B**) showed that a strong negative precipitation anomaly was superposed on a positive temperature anomaly during June 1995 at both Delingha and Chaka meteorological stations (–29.3 and –23.3 mm for June precipitation anomaly and +2.3°C and +1.7°C for June temperature anomaly in Delingha and Chaka, respectively), far exceeding the magnitude of temperature and precipitation anomalies in 1957, 1966, and 1979. Therefore, the 1995 growing season was the driest for Qilian juniper during the study period. Indeed, 1995 was documented as one of the driest years on the NETP in other studies (Fang et al., 2012; Tian et al., 2007; Liang et al., 2016a). Thus, the Qilian juniper samples suffered from more severe water shortage in 1995 than during any of the other three drought years. The 1995 drought was associated with significant growth decline and longer recovery period. Our results indicate that almost all trees used in our study survived all four extreme drought events. However, in making this interpretation, we may have overlooked the important aspect that all tree samples we used were from living trees; consequently, those trees that had died after a specific drought event were not included in subsequent samples. There are precedents for tree mortality due to high-intensity droughts in many parts of the world (Allen et al., 2010; Anderegg et al., 2013; McDowell et al., 2008). Therefore, we can infer from our results that if a very severe drought event occurs during the growing season (like that in June 1995), and if it exceeds a level for some trees, then those trees will die first, while the other more drought-tolerant individuals may survive after the drought.

Unfortunately, such thresholds are rarely studied through dendrochronology, and it is difficult to obtain robust estimates. Theoretically, it is possible for resilience to equal zero; this can be interpreted as the failure of some trees to form a radial increment during the 2 or more years after the extreme drought (i.e., missing rings) and includes the possibility that the tree died. A resilience of zero was not observed in our study, which may be due to the inherent weakness of the evaluation method that we adopted (Lloret et al., 2011). Therefore, a priority for future work will be to seek these dead/dying tree samples that are most vulnerable to drought events and compare their data with surviving trees to analyze their growth status before and after drought, with the aim of better understanding tree resilience to drought and the threshold of tree mortality.

#### Considerations of Future Forest Variability Under Global Warming

The increasing temperature and precipitation associated with global warming in our study region during recent decades, coupled with a possible carbon dioxide fertilization effect (Piao et al., 2012), have been beneficial to the radial growth of Qilian juniper on the NETP. This has been demonstrated by the slightly upward trend in tree-ring width sequences during the study period. However, the slight rise in precipitation has not changed the overall status of water shortage on the NETP. Furthermore, a growing number of studies have shown that greater variability of precipitation under global warming will enhance the intensity and frequency of extreme drought events (Kim and Byun, 2009; Trenberth et al., 2014). This will undoubtedly be a greater test for Qilian junipers growing in the arid Tibetan Plateau of inner Asia, since our results showed that the recovery ability of Qilian juniper after drought events is closely related to the drought intensity.

In addition, there is general consensus that the growth of Qilian juniper over broad areas of the lower/middle altitudes is mainly limited by water availability. However, Zhang et al. (2015) directly addressed this issue by collecting Qilian juniper samples in the uppermost 20% of the forest belt and concluded that low temperature rather than drought limits the radial growth of Qilian juniper at the upper tree-line (~4,250 m a.s.l., higher than in all of the other related studies). With continued global warming, trees growing near the upper tree-line may benefit from increasing temperatures as thermal conditions become less restrictive on tree growth, thereby leading to an upward migration of the upper tree-line (Liang et al., 2016b). According to our results, the younger Qilian junipers at lower altitudes are the most vulnerable groups to drought events, while the younger individuals at higher altitude are more resistant. Ongoing global warming and associated expansion of the forest belt to higher elevations will have two main consequences: one is that the warming process is equivalent to expanding the range of lowelevation Qilian junipers, causing the older trees in these areas to be exposed to more severe water stress conditions; the other is to increase the rate of forest recruitment at high altitudes, where younger trees are more resistant to drought and are under less severe drought stress. However, if this trend continues, the upper tree-line will eventually be constrained by other factors (e.g., terrain, or the height at which precipitation no longer increases with elevation). From that time onward, the forest belt will no longer be able to expand upward, yet the intensity of water shortage would continue increasing under ongoing global warming, which is equivalent to placing a larger number of trees into the "lower-altitude environment" (warmer and drier), which would have an adverse effect on local forest ecological security.

#### CONCLUSIONS

We studied Qilian junipers that are widely distributed across the NETP to explore the vulnerability of different tree individuals to severe drought events. In conclusion, we found that drought extremes have a significant adverse impact on the radial growth of Qilian juniper. The effects of age and elevation both strongly affect the ability of Qilian junipers to cope with drought extremes: specifically, older trees from lower elevations are more vulnerable to drought and may even be exposed to a higher risk of mortality should water shortages become aggravated in the future. Meanwhile, the resilience of Qilian junipers to extreme drought showed no connection with tree age, elevation, latitude, or longitude, but was closely correlated with drought intensity. Thus, we can reasonably

#### REFERENCES


speculate that, although all the Qilian juniper samples used in this study had successfully recovered from the four selected drought events, there are some very vulnerable tree individuals (e.g., some of the oldest trees at lower altitudes) that may already be dead if the intensity of one of those drought episodes exceeded their endurance limit; these individuals will not have been included in our samples. This is important in the context of global warming, since increased precipitation variability may enhance the intensity and frequency of extreme drought events. Our study revealed which tree individuals are more vulnerable to drought extremes, enabling identification of those trees most seriously affected by drought so that their growth and even death processes can be monitored. This would not only provide the possibility of estimating threshold conditions for tree mortality but also enable us to better understand the dynamics of forests and even ecosystems across the NETP under the background of climate warming.

# DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

# AUTHOR CONTRIBUTIONS

BY and XW designed the study. XW analyzed the data and wrote the first version of manuscript. XW, BY, and FL revised the manuscript and approved the submitted version.

# FUNDING

This study is supported by the National Nature Science Foundation of China (NSFC grant nos. 41520104005 and 41325008), and by the Belmont Forum and JPI-Climate Collaborative Research Action "INTEGRATE" (NSFC grant no. 41661144008). FL is supported by the Swedish Research Council (Vetenskapsrådet, grant no. 2018-01272).

# ACKNOWLEDGMENTS

We are grateful to Prof. Sergio Rossi for his guidance in data analysis and to Linzhou Xia for his excellent fieldwork.


Zhang, Q. B., Cheng, G., Yao, T., Kang, X., and Huang, J. (2003). A 2,326-year tree-ring record of climate variability on the Northeastern Qinghai–Tibetan Plateau. *Geophys. Res. Lett.* 30, 1739-1742. doi: 10.1029/2003GL017425

**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 Wang, Yang and Ljungqvist. 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.*

# Unraveling Shade Tolerance and Plasticity of Semi-Evergreen Oaks: Insights From Maritime Forest Live Oak Restoration

#### *Emily C. Thyroff1, Owen T. Burney2, Michael V. Mickelbart3 and Douglass F. Jacobs1\**

1 Department of Forestry and Natural Resources, Hardwood Tree Improvement and Regeneration Center, Purdue University, West Lafayette, IN, United States, 2 John T. Harrington Forestry Research Center, New Mexico State University, Las Cruces, NM, United States, 3 Department of Botany and Plant Pathology, Purdue University, West Lafayette, IN, United States

#### Edited by:

Giovanna Battipaglia, University of Campania Luigi Vanvitelli, Italy

#### Reviewed by:

Enrique Andivia, Complutense University of Madrid, Spain Lina Fusaro, Sapienza University of Rome, Italy

> \*Correspondence: Douglass F. Jacobs djacobs@purdue.edu

Specialty Section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Genetics

Received: 07 July 2019 Accepted: 01 November 2019 Published: 20 November 2019

#### Citation:

Thyroff EC, Burney OT, Mickelbart MV and Jacobs DF (2019) Unraveling Shade Tolerance and Plasticity of Semi-Evergreen Oaks: Insights From Maritime Forest Live Oak Restoration. Front. Plant Sci. 10:1526. doi: 10.3389/fpls.2019.01526

Quercus spp. (oaks) are generally intermediate in shade tolerance, yet there is large variation within the genus in shade tolerance and plasticity in response to varying resource availability. Ecophysiological knowledge specific to semi-evergreen Quercus spp. from subtropical maritime forests is lacking relative to temperate deciduous oaks. We studied the influence of light availability and plant competition on leaf physiology and performance of semi-evergreen Quercus virginiana on a barrier island along the US southern Atlantic coast. Seedlings were underplanted in pine (Pinus taeda) plantation stands with varying overstory density (clear-cut, heavy thin, light thin, and non-thinned; creating a gradient of understory light availability) and vegetation (no competition removal or herbaceous competition removal) treatments. After 2 years, seedling survival was higher with increasing light availability (clear-cut = heavy thin > light thin > non-thinned). Seedling growth (i.e., diameter, height, and crown width) increased similarly with increasing thinning intensity, while vegetation control was mainly beneficial to seedling growth in clear-cuts. These responses were partially explained by foliar nitrogen and leaf trait measurements, which followed the same pattern. Q. virginiana seedlings demonstrated high plasticity in their ability to acclimate to varying resource availability, as indicated by light response curves, specific leaf area, stomatal density, stomatal pore index, and maximum theoretical stomatal conductance. Light compensation and saturation points illustrated seedling capacity to increase net CO2 assimilation with increased light availability. Leaves on trees in the high light environment had the highest net CO2 assimilation, stomatal density, stomatal pore index, maximum theoretical stomatal conductance, and lowest specific leaf area. Although we demonstrated the relative shade tolerance of Q. virginiana in lower light environments (i.e., heavy and light thin plots), this semi-evergreen species shows high plasticity in capacity to respond to varying resource availability, similar to other Quercus spp. from mesic and Mediterranean environments.

Keywords: Quercus virginiana, canopy openness, ecophysiology, gas exchange, leaf traits, light acclimation, plant competition, forest regeneration

# INTRODUCTION

Plasticity is an adaptive strategy to promote survival and fitness of long-lived forest tree species that may experience several environmental changes and associated stresses throughout their lifespan (Cavender-Bares and Ramírez-Valiente, 2017; Gil-Pelegrín et al., 2017a). Phenotypic plasticity, both physiological and morphological, affects life history traits and contributes to the large distribution ranges (Gratani et al., 2003; Niinemets, 2015) and wide ecophysiological variation (Niinemets and Valladres, 2006; Gil-Pelegrín et al., 2017b) of *Quercus* spp. (oaks). From deciduous to semi-evergreen or evergreen *Quercu*s spp., plasticity has commonly been identified as a trait contributing to greater drought, cold, and shade tolerance (Valladares et al., 2002; Gimeno et al., 2009; Limousin et al., 2012; Ramírez-Valiente et al., 2017). Plasticity relates to how populations may respond to climate change as well as the development of improved management activities (Paquette et al., 2007; Gimeno et al., 2009). Plasticity is also important in understanding mechanisms that lead to regeneration and restoration success (Benito-Garzón et al., 2013; Lawson and Michler, 2014; Löf et al., 2019).

*Quercus* spp. are generally intermediate in shade tolerance and underplanting may provide an effective means to restore these species (Dey et al., 2012). Increased light and soil moisture resulting from overstory density reduction may benefit planted *Quercus* seedlings, but also pioneer species (Dey et al., 2008; Dey et al., 2012; Kern et al., 2012; Villar-Salvador, 2016). Pioneer species that acclimate rapidly and take advantage of higher light levels are particularly competitive, often suppressing oak seedling survival and growth (Paquette et al., 2006; Dey et al., 2008; Gardiner et al., 2010). Limited light, nutrients, and water resources from competition can negatively affect physiological processes, inhibiting seedling performance (Salifu et al., 2009; Grossnickle, 2012). Removal of competing vegetation, therefore, has potential to channel limited resources to planted seedlings, yet can be logistically prohibitive (Wagner and Zasada 1991; Fleming et al., 2006). Alternatively, maintaining partial overstory may introduce sufficient light to optimize growth in the target species, while restricting faster growing competition (Elliott and Swank, 1994; Paquette et al., 2006; Brown et al., 2014).

To facilitate forest restoration, there has been increased investigation in converting pine plantations to diverse hardwood forests using clear-cutting or thinning followed by planting of desired species (Parker et al., 2001; Gómez-Aparicio et al., 2009; Löf et al., 2010; Villar-Salvador, 2016; Lesko and Jacobs, 2018). Overstory removal treatments affect light, temperature, soil moisture, and soil compaction in complex feedback loops, which are dependent on species and ecosystems (Canham et al., 1990; Madrigal-González et al., 2017; Soto et al., 2017). Thus, understanding how silvicultural treatments affect plasticity and adaptive potential of target species regeneration may accelerate restoration processes by increasing availability of light, water, and nutrients.

Maritime forests of the subtropical US southern Atlantic coast, characterized by the dominant, semi-evergreen *Quercus virginiana* L., represent a case study where conversion of pine plantations back to diverse native hardwood forests may facilitate the restoration of associated ecosystem services (Albers and Alber, 2003; Jones et al., 2013). Within the range of *Q. virginiana*, there has been centuries of human land transformation, particularly on the more stable land where maritime forests develop (Bratton and Miller, 1994; Bellis, 1995; Fox et al., 2007; Jones et al., 2013). A fraction of the original estimated land area of maritime forests remains at approximately 39,000 ha, which has created interest to protect and restore maritime forests (Mathews et al., 1980; Lopazanski et al., 1988).

Many agricultural lands that were originally maritime forests on the US southern Atlantic coast were abandoned and more recently planted into pine plantations (i.e., *Pinus taeda* L.) for commercial investment and to minimize erosion (Fox et al., 2007; Brockerhoff et al., 2008). Pine plantations tend to perform poorly when exposed to inherent coastal stressors and abandoned monoculture pine plantations with low genetic diversity are particularly prone to disease and outbreaks of southern pine beetles (*Dendroctonus frontalis* Zimm.), which are an economically destructive forest pest due to exponential outbreaks (Conner et al., 2005; Fox et al., 2007; Brockerhoff et al., 2008; Watson et al., 2013; Nowak et al., 2015; Asaro et al., 2017). Clear-cuts are used to salvage residual timber value and reduce continual spread of active outbreak sites, while overstory thinning helps to minimize future outbreaks in at-risk stands (Belanger et al., 1993; Watson et al., 2013; Asaro et al., 2017). The complete or partial removal of the pine overstory provides an opportunity to restore maritime forest by regenerating *Q. virginiana.*

Our objective was to better understand the underlying physiological mechanisms that drive plant structure, physiology, and function of *Q. virginiana*, as a semi-evergreen oak. *Quercus virginiana* L. has a broad distribution across maritime edaphic site factors compared to other maritime oaks (Cavender-Bares and Pahlich 2009), suggesting potential for high plasticity. While the effects of varying resources such as temperature and precipitation has been studied in *Q virginiana* (Kurtz et al., 2013; Ramírez-Valiente et al., 2015; Cavender-Bares and Ramírez-Valiente, 2017), the species response to light and competition has yet to be explored. We experimentally evaluated the relative influence of pine overstory density and vegetation control treatments on *Q. virginiana* seedling performance. We hypothesized that *Q. virginiana* survival, growth, and leaf development would peak in the thinned treatments, when competition was controlled, reflective of the relative shade tolerance of most *Quercus* spp. Under this scenario, seedlings should show higher net CO2 assimilation and greater growth and development due to increased light compared to the control, while avoiding excessive light in the clear-cut.

#### MATERIALS AND METHODS

#### Experimental Site

The experiment was conducted on the north end of St. Simon's Island, Georgia at Cannon's Point Preserve (N 31°15'29" W 81°20'45"), which is a 246-ha wilderness tract with approximately 50 ha dominated by abandoned pine plantations (mostly *P. taeda* L. with some *P. elliotti* Englem.). Tree rings and cores indicated that the pine stands were approximately 50 years old. In 2015 and 2016, areas of natural and planted pines affected by southern pine beetles were clear-cut to salvage timber and reduce the southern pine beetle outbreak.

Soils at Cannon's Point Preserve are a mixture of fine sandy soils dominated by Mandarin fine sand and Cainhoy fine sand, 0–5% slopes. Pottsburg sand and Rutledge fine sand are also present (NRCS, 2017). At each plot, four soil samples were composited to evaluate physical and chemical characteristics using Mehlich III extraction (Brookside Laboratories, New Brennan, Ohio). Soil characteristics were similar with slight differences creating variability across replicate blocks (**Table 1**).

This region receives an average annual precipitation of 114 cm and average annual temperature was 20.0°C. During the study period 2017–2018, average annual precipitation was 90 cm and temperature was 21.0°C (Sapelo Island National Estuarine Research Reserve Meterological Monitoring, 2018; U.S. Climate Data, 2018). Hurricane Irma (September 2017; 7 months into experiment) resulted in temporarily increased precipitation, saltwater inundation, salt spray, and strong winds across the region, yet no damage to our experimental site.

#### Experimental Design and Treatments

A randomized complete block design with a split-plot structure was used for this study. The whole plot factor consisted of four overstory densities (clear-cut, heavy thin, light thin, and nonthinned). The subplot factor was two levels of vegetation control (no vegetation removal and 2 years of vegetation removal). A total of 25 seedlings were planted within each subplot. All treatment combinations were replicated by four blocks resulting in 800 total seedlings.

Overstory density treatments were randomly applied to a 66 × 44 m area. Within the treated area, 26 × 14 m research plots were established. All plots were fenced (2.5 m height) to exclude white-tailed deer (*Odocoileus virginianus* Zimm.) as herbivory is cited as a limiting site factor in maritime forest restoration (Thyroff et al., 2019). Overstory density treatments were installed by modifying the basal area of the original pine overstory. Target basal areas were clear-cut at 0 m2 ha−1, heavy thin at 4–9 m2 ha−1, light thin at 18–23 m2 ha−1, and non-thinned at 27+ m2 ha−1. Logging operations to implement overstory treatments were completed in December 2016. Target basal areas were monitored by a contracted forester and logger. Additionally,


all mid-story trees, understory vegetation, and large slash were removed immediately after the harvest activity to reduce possible confounding effects. Subplots requiring vegetation control over 2 years were done so throughout both growing seasons using mechanical methods (i.e., brush saws and hand clippers).

# Plant Material

One-year-old *Q. virginiana* bare-root seedlings were planted in February 2017. Seedlings were obtained from Superior Trees in Lee, Florida with a Louisiana seed source. From baseline morphology analysis (*n* = 20), mean seeding diameter was 5 mm (± 0.20), mean seedling height was 54 cm (± 2.00), and root to shoot dry mass (g) ratio was 0.89 (± 0.76). Seedlings were sorted prior to planting and randomly assigned treatments. Seedlings were hand planted with planting bars at 2-m spacing. To maintain planting density and interspecific seedling competition, a perimeter of buffer trees was planted 2 m from the research seedlings.

# Plot Characteristics

For each plot, basal area and canopy closure data were collected in summer 2017. All mature tree species within each fenced plot and the buffer areas around them were identified and measured for diameter at breast height (DBH; 1.37 m) to the nearest centimeter. DBH of mature trees were used to calculate the basal area. Three hemispherical photographs were taken under homogeneous diffuse sky conditions and across the centerline, working from west to east cardinal directions, at approximately 4.7-m intervals (1/3 of the plot width). Photographs were analyzed with CIMES software (Gonsamo et al., 2011) to determine percent canopy closure.

To record soil moisture and temperature, each plot had an Em50 digital data logger with two 5TM sensors (Decagon, Pullman, Washington) located in the subplot center. Each sensor was installed at a depth of 25 cm and recorded measurements every 2 h. Four of the plots (all of block 1) had additional sensors installed; photosynthetically active radiation (PAR) sensors to monitor light and VP4 sensors to capture air temperature and relative humidity recorded every 2 h. At the peak of vegetation cover on site (September 2018), five seedlings from each treatment (160 total) were randomly selected for a 1-m2 plot vegetation survey to assess percent competing vegetation cover, mean height of competition, and top competing species within each plot.

### Seedling Performance

At the time of planting (February 2017), measurements of ground line diameter and height to last live bud were recorded. At the end of the growing seasons (November 2017 and 2018), survival, diameter, height, and crown width were recorded. Survival was recorded as a binary response; "alive" included seedlings with any number of green leaves. At the end of the second growing season (November 2018), foliar nitrogen (N) was determined by randomly sampling seven seedlings per subplot, resulting in 224 total sampling units. Three leaves per seedling were collected and composited, dried at 60°C for 72 h, weighed, pulverized in vials with stainless steel balls, and analyzed with an ECS 4010 CHNSO Analyzer (Costech, Valencia, California).

## Ecophysiology and Leaf Trait Measurements

Gas exchange, specific leaf area (SLA), stomatal density, stomatal pore index (SPI), and maximum theoretical stomatal conductance (gmax) were measured during the growing season (June 2018). Leaf gas-exchange was measured with a portable LI-6400XT (LI-COR Biosciences, Lincoln, Nebraska) to create light response curves. Two *Q. virginiana* seedlings were randomly selected per subplot, resulting in 64 total sampling units. One upper-canopy, fully expanded, recently mature leaf per tree was measured between the hours 10:00 and 14:00. Light levels used to create light response curves were: 1,600, 1,400, 1,200, 1,000, 800, 600, 400, 300, 200, 100, 50, 0 (µmol m−2 s−1). Infrared gas analyzers of the LI-6400XT (IRGAs; reference and sample) were matched at the beginning and end of each light curve measurement. Relative humidity (~ 60%), vapor pressure deficit (<3.0 kPa), and temperature (leaf and block) were monitored for consistency. The gas exchange data point was taken after sample gas values (H2O and CO2) and net CO2 assimilation were stable, based on coefficient of variation. *Q. virginiana* leaves did not fully fill the 2 × 3 cm LI-6400XT leaf chamber, therefore, gas-exchange measurements were adjusted for actual leaf areas. Leaf areas were determined from a photo of the leaf in the chamber using ImageJ (National Institutes of Health, Bethesda, Maryland). Light response curves were created by plotting net CO2 assimilation (AN, μmol CO2 m−2 s−1) against PAR. The curves were fitted to a non-rectangular hyperbola (SigmaPlot V11.0, Systat Software, San Jose, California). Methodology to calculate final parameters from the model followed Chartier and Prioul (1976). Final parameters were used to calculate light compensation (µmol m−2 s−1) and light saturation (µmol m−2 s−1) points.

SLA, stomatal density, SPI, and gmax were sampled from the same four selected *Q. virginiana* seedlings per plot used for gasexchange measurements. Three selected, upper-canopy, fully expanded, recently mature leaves were used for each seedling. In the non-thinned overstory some seedlings did not have many leaves, therefore in those cases only two leaves were collected. SLA was calculated by dividing leaf area by leaf mass (cm2 g−1). Collected whole leaves were scanned to measure leaf area (cm2 ) using ImageJ. Leaves were dried at 60°C for 48 h then weighed for leaf mass (g).

Impressions of the abaxial leaf surface were made in the middle of each leaf, midway between the midrib and the leaf margin. Leaf impressions were made on microscope slides using cyanoacrylate. Five leaf impression images (DCM 900 microscope CMOS Camera, Oplenic Optronics, Hangzhou, China) were taken of a 0.19 × 0.14 mm (0.0266 mm2 ) area under 40× magnification using a microscope (BH-2 microscope, Olympus, Tokyo, Japan). Stomatal counts were conducted using ImageJ and the cell counter plug-in (Kurt De Vos, University of Sheffield). For unbiased counting, all whole stomata were counted within the impression image area and stomata partially within the image were only counted on the top and right sides of the image area. Stomatal density (mm−2) was calculated by dividing the number of stomata in the image by image area. To calculate SPI and gmax, stomatal lengths and widths were calculated from the leaf impressions using ImageJ and using the Feret's diameter measurement (**Supplementary Appendix A**). SPI was calculated by multiplying stomatal density by stomatal length squared, while gmax (mmol m−2 s−1) was calculated using stomatal length and width following equations from McElwain et al. (2016) methodology.

#### Statistical Analyses

All data was analyzed with R software version 3.5.3 (R Core Team, 2019) using: lme4 package (Bates et al., 2015) for general linear models, linear regressions, and logistic regression; nlme (Pinheiro et al., 2018) package for repeated measures models; multcomp package (Hothorn et al., 2008) for pairwise comparisons. Plot characteristics (basal area, canopy closure, and data loggers) were analyzed with general linear mixed models, with overstory as the fixed factor and block as a random factor. A logistic regression model was used to analyze survival with overstory, vegetation control, and interaction as fixed factors and block as a random factor. Diameter, height, and crown width were analyzed with repeated measures general linear mixed models with overstory, vegetation control, time, and resulting interactions as fixed factors; block and individual tree as random factors. Foliar N, light response curves including light compensation/saturation points, leaf traits (i.e., SLA, stomatal density, SPI, gmax), and vegetation survey dependent variables were analyzed separately with general linear mixed models, with overstory, vegetation control, and interaction as fixed factors and block as a random factor. Linear regression models were used to analyze light saturation points and growth parameters. Residuals from all response variables were tested to ensure normality and homogeneity of variance. Crown width did not meet assumptions and data was square root transformed. For all analyses, when significant treatment effects were detected (p ≤ 0.05), Tukey's HSD test was used to test for pairwise comparisons (α = 0.05). All statistical output results are provided in **Appendices B** and **C**. Although the number of sampling units from each experimental unit varied across measurements (per details above), the number of experimental units was always n = 4.

# RESULTS

## Plot Characteristics

Basal area (m2 ha−1), canopy closure (%), light (PAR), mean air temperature (°C), and mean soil temperature (°C) followed a progression of overstory density (**Table 2**). Clear-cuts had the lowest basal area and canopy closure, resulting in greatest PAR. This pattern was consistent along the light progression with thinned plots having intermediate basal area and canopy closure. Non-thinned plots had the greatest basal area and canopy closure, resulting in the lowest PAR. Mean air and soil temperature increased with increased thinning density resulting in highest temperatures in clear-cut plots. From April 2017 to May 2018, average soil moisture was consistently greater in vegetation control subplots than non-vegetation control subplots. Soil moisture peaked in late summer/early autumn and was often greatest in heavy thin plots, followed by clear-cut and light thin, and lastly non-thinned plots (**Figure 1**).


TABLE 2 | Mean (± SE) target basal area from logging operation, basal area from stand inventory, canopy closure from hemispherical photos.

Mean (± SE) PAR levels, air temperature, and soil temperature from data loggers. Different letters indicate significant differences among treatments (α = 0.05).

The interaction of overstory and vegetation control was significant for percent vegetation cover (F3,54 = 6.84, p = 0.001) as the effect of vegetation control was different between the overstory treatments. Vegetation control decreased percent vegetation cover in clear-cut, heavy thin, and light thin plots, but had no effect in non-thinned plots (**Figure 2**). Additionally, height of competing vegetation was 58.3 cm (± 14.4) in non-vegetation control subplots compared to 18.8 cm (± 3.0) in vegetation control subplots (F1,57 = 35.56, p < 0.001). Top competing species included: *Ilex vomitoria*, *Paspalum notatum, Rubus trivialis, Vitis rotundifolia,* and *Morella cerifera.*

### Seedling Performance

Initial height and diameter of planted seedlings were similar across all treatments with an average height of 48 cm (± 0.9) and an average diameter of 3.8 mm (± 0.1). Overall survival after two growing seasons was 75.5% (± 1.9). The treatment interaction was not significant, however the main effect of overstory was significant with increased survival with increased thinning intensity (X2 3,794 = 9.86, p = 0.020); clear-cut and heavy thin plots had the highest survival at 81.5 and 81.0%, light thin was intermediate at 73.0%, and non-thinned had the lowest survival at 65.5%.

The interaction of overstory, vegetation control, and time was significant for diameter (F6,595 = 13.30, p < 0.001), height (F6,1217 = 3.22, p = 0.004), and crown width (F3,587 = 13.00, p < 0.001). The effect of vegetation control differed among the overstory density

treatments and the effects varied over time. Overall diameter, height, and crown width increased with increased thinning intensity. Vegetation control was most beneficial for seedlings in clear-cut plots followed by heavy thin plots and had no effect in non-thinned plots (**Figure 3**).

Seedling diameter increased by 400% in clear-cut plots with vegetation control and 200% in clear-cut plots without vegetation control, while marginal growth occurred in nonthinned plots regardless of vegetation control treatment. Seedling height and crown width increased by 200% in clearcut plots with vegetation control and 100% in clear-cut plots without vegetation control, while non-thinned plots had little growth regardless of treatment. Dieback occurred frequently in the non-thinned plots, which resulted in negative relative heights after two growing seasons.

#### Ecophysiology and Leaf Trait Measurements

significant differences among treatments (α = 0.05).

For foliar N, only main effects were significant showing increased foliar N with increased thinning intensity (F3,217 = 24.06, p < 0.001) and increased foliar N with vegetation control (F1,217 = 34.94, p < 0.001). While the interaction was non-significant the overstory

main effect was significant for the calculated light compensation (F3,57 = 8.10, p < 0.001) and light saturation points (F3,57 = 23.56, p < 0.001). With increased thinning intensity, light compensation and saturation points increased. Net CO2 assimilation was greatest in clear-cut plots, intermediate in heavy and light thin, lowest in nonthinned plots (**Figure 4**), and positively related to growth parameters (**Supplementary Appendix D**). Additionally, the vegetation control main effect was significant only for light saturation point with greater light saturation points in subplots with vegetation control compared to subplots without vegetation control (F1,57 = 5.56, p = 0.022).

Different letters indicate significant differences among treatments (α = 0.05).

For SLA, stomatal density, SPI, and gmax, only main effects were significant. With increased thinning intensity, SLA decreased (F3,57 = 12.60, p < 0.001), while stomatal density (F3,57 = 21.20, p < 0.001), SPI (F3,57 = 24.19, p < 0.001), and gmax (F3,57 = 24.48, p < 0.001) all increased (**Figure 5**). Vegetation control resulted in decreased SLA (F1,57 = 6.28, p = 0.015), and increased stomatal

FIGURE 4 | Mean light response curves (net photosynthesis plotted by photosynthetically active radiation) of Quercus virginiana seedlings planted either in clear-cut, heavy thin, light thin, or no thin plots taken in June 2018. Different letters indicate significant differences among treatments (α = 0.05).

density (F1,57 = 5.37, p = 0.024), SPI (F1,57 = 6.55, p = 0.013), and gmax (F1,57 = 5.69, p = 0.020). More abaxial trichomes (i.e., leaf hairs) occurred on stomatal density impression images from clear-cut plots followed by heavy thin, light thin, and lastly no trichomes in non-thinned plots (F3,57 = 6.38, p < 0.001; **Figure 6**).

## DISCUSSION

#### Light and Competing Vegetation

Removal of stand basal area affects environmental characteristics such as light, temperature, and soil moisture, which influence seedling performance (Gil-Pelegrín et al., 2017b; Villar-Salvador, 2016). In our study, the pattern reflected in the PAR results (clear-cut had greatest PAR, followed by the thinned overstories, and lastly non-thinned) corresponded with increased seedling survival, diameter, height, crown width, and foliar N. Growth parameters were consistently greatest in clear-cut plots, intermediate in thinned plots, and lowest in non-thinned plots, rather than peaking in the thinned treatments as we hypothesized. Foliar N, which is an essential macronutrient for seedling establishment and performance (Abrams and Mostoller, 1995; Soto et al., 2017), was significantly greater in leaves of seedlings in clear-cut plots followed by the thinned overstories, indicating greater accessibility of this limiting resource to seedlings (Kobe, 2006; Uscola et al., 2015). Similar to other studies, seedlings in less dense overstory treatments were able to use available resources, and thereby increase photosynthesis and growth (Gómez-Aparicio et al., 2006; Cooper et al., 2014; Soto et al., 2017). Supporting our results, *Q. virginiana* has been reported as a fast growing species that preferentially allocates resources to aboveground biomass development (Cavender-Bares and Pahlich, 2009).

In a dense overstory, such as the non-thinned plots, soil moisture and nutrients (e.g., N as seen with low foliar N concentrations) are typically limited because of canopy tree

FIGURE 5 | Mean (± SE) (A) specific leaf area (cm2 g−1), (B) stomatal density (stomata mm−2), (C) stomatal pore index, and (D) maximum theoretical stomatal conductance (gmax, mmol m−2 s−1) of Quercus virginiana seedlings planted in clear-cut, heavy thin, light thin, or no thin plots. Different letters indicate significant differences among treatments (α = 0.05).

dominance and competition (Cooper et al., 2014). Seedlings in non-thinned plots with dense overstories may have arrested development due to low light levels preventing their progression from seedlings to saplings (Zavala et al., 2011; Soto et al., 2019). Artificially regenerated *Q. virginiana* did not appear to benefit from the shade of the pine plantations as identified with *Q. ilex* L. (Gómez-Aparicio et al., 2009); however, this advantage may be expressed during a prolonged drought or on drier coastal sites or if studying natural regeneration. Thus, effects may also be due to the density of the pine plantations (33.9 m2 ha−1 basal area) in our study. Sites with pine overstory densities closer to the heavy thin treatment (16.9 m2 ha−1 basal area) may see the benefit of directly underplanting in a plantation.

As hypothesized, we found a consistent interaction between overstory and vegetation control treatments, with vegetation control benefiting growth parameters in clear-cut plots and not in non-thinned plots (**Figure 3**). Without thinning, light and soil moisture were more limiting to *Q. virginiana* performance than understory vegetation competition. Semi-evergreen *Quercus* spp. have varying photosynthetic responses to different environments, with a range of growth rates and shade tolerance (Gratani et al., 2003; Cavender-Bares and Ramírez-Valiente, 2017; Villar et al., 2017). Within the clear-cut and heavy thinning treatments, higher PAR contributed to understory vegetation becoming more abundant and thus more competitive (Ter-Mikaelian et al., 1999). In our study, vegetation control promoted seedling growth, particularly on sites with more light (i.e., clearcut and heavy thin). The significant interaction illustrated a shift in pressure from light limited environments to resource limited environments (i.e., greater foliar N concentrations and increased soil moisture in vegetation control subplots). Overall, our results align with past studies indicating that vegetation control enhances seedling establishment and performance, particularly when released from dense overstories (Wagner and Zasada, 1991; Fleming et al., 2006). In cases where competing vegetation is expected to be high and vegetation control is not feasible, however, maintaining an overstory may reduce competition and the need for costly and sometimes controversial (i.e., herbicide) competing vegetation removal (Thiffault and Roy, 2011; Löf et al., 2014; Villar-Salvador, 2016).

#### Ecophysiology and Leaf Trait Responses

Trees are long lived organisms that may experience several environmental changes and associated stresses and plasticity is an adaptive strategies to promote survival under environmental fluctuations (Cavender-Bares and Ramírez-Valiente, 2017; Gil-Pelegrín et al., 2017a). Our results indicate that *Q. virginiana*

can acclimate to varying environments as maximum net CO2 assimilation occurred in clear-cuts (**Figure 4**), which aligns with the increased growth and foliar N in clear-cut plots. With higher light compensation and saturation points, seedlings in clearcuts took longer to achieve positive CO2 assimilation, but were able to utilize increased PAR with higher net CO2 assimilation rates. Similar to studies with other oaks, *Q. virginiana* seedlings responded to the other overstory treatments in a linear progression with respect to light availability (Cavender-Bares and Pahlich, 2009; Cooper et al., 2014). Seedlings in the nonthinned plots reached positive net CO2 assimilation quickest at the lowest PAR; however, non-thinned seedlings also reached light saturation point quicker at a lower PAR. Net CO2 assimilation, therefore, was limited and lowest for seedlings in non-thinned plots. Greater net CO2 assimilation rates were supported by more stomata, not larger stomata (**Supplementary Appendix A**), resulting in greater gas exchange potential as seen with stomatal density, SPI, and gmax. Ability to regulate stomatal development (stomatal density and SPI) and capacity for high gmax rates leads to seedling acclimation to a wide range of environments. Additionally, seedlings in more shaded environments with decreased SPI can allow for higher water use efficiency (Yoo et al., 2009; McElwain et al., 2016; Liu et al., 2017). In other studies, Mediterranean evergreen oaks such as *Quercus oleoides* Schltdl. & Cham. and *Quercus ilex* have also shown phenotypic plasticity in growth rates, CO2 assimilation, and leaf development (Cavender-Bares and Ramírez-Valiente, 2017; Gil-Pelegrín et al., 2017b; Ramírez-Valiente et al., 2017). Niinemets and Valladares (2006) classified *Q. virginiana* as drought-tolerant and intermediate in shade tolerance. Not only is maximizing performance under high-light conditions beneficial, but also the capability of shade adaptation is beneficial to seedling survival (Valladares and Niinemets, 2008). Similarly, *Quercus velutina* Lam. is a drought-tolerant and light demanding oak for which Ashton and Berlyn (1994) showed great leaf anatomical plasticity with high net CO2 assimilation compared to other temperate deciduous oaks.

Leaf variation of *Quercus* spp. tends to be on the lower end of the leaf economic spectrum, aligning with early to midsuccessional classification (Wright et al., 2004; Niinemets and Valladres, 2006; Gil-Pelegrín et al., 2017b). Following a conservative resource strategy, lower SLA leaves on semievergreen oaks can help to maintain function and extend photosynthesis (Cavender-Bares and Ramírez-Valiente, 2017). Thicker leaves are also more resistant to environmental stressors such as aridity, freezing temperatures, and solar radiation (Cavender-Bares and Ramírez-Valiente, 2017; Gil-Pelegrín et al., 2017b; Pemán et al., 2017). Whereas SLA was negatively associated with CO2 assimilation, stomatal density was positively associated with CO2 assimilation. Similarly, Ramírez-Valiente et al. (2017) found that SLA was negatively associated with net CO2 assimilation and *Q.* seedling growth. In clearcut plots with low SLA (i.e., smaller, thicker leaves), seedling growth was greatest. Higher stomatal density and SPI increases gas exchange potential and gmax, which increases net CO2 assimilation (Wright et al., 2004; McElwain et al., 2016). Along with increased gas exchange potential, comes increased risk of desiccation; therefore, a trade-off is necessary to maximize performance, which was seen in the opposite associations of SLA and stomatal density with net CO2 assimilation. Trichomes are commonly found in many *Quercus* spp. and trichomes in the abaxial surface may be a mechanism to reduce water loss (Bickford, 2016; Gil-Pelegrín et al., 2017b) as seen in the higher light plots (**Figure 6**). Additionally, a smaller leaf typically has a thinner leaf boundary layer, which facilitates cooling in drier and warmer climates (e.g., clear-cuts) (Gil-Pelegrín et al., 2017b). In our experiment, *Q. virginiana* showed phenotypic plasticity for growth rates, gas exchange, and leaf development in response to silvicultural treatments that modified seedling environments as was observed in similar evergreen oak regeneration studies (Ramírez-Valiente et al., 2015).

## CONCLUSIONS

We demonstrated that *Q. virginiana* is capable of acclimating to varying resource availability, with a high tolerance to full sunlight. Developmental responses of underplanted *Q. virginiana* did not follow the predicted trend of peaking in the heavy/light thin pine plantation overstory, but rather in clearcuts. Nonetheless, thinning is among the most effective practices for preventing or mitigating pine beetle outbreaks by creating barriers to population growth and spread (Nowak et al., 2015; Asaro et al., 2017). Additionally, a canopy buffering-effect in thinned stands may be more beneficial for oaks on drier sites or during prolonged drought (García-Plazaola et al., 2017). Thus, rather than prescribing a single treatment (e.g., clear-cut) that optimizes *Q. virginiana* development, an alternative may be to prescribe several treatments that result in cost-effective restoration and reduce the need for costly competing vegetation control. Forest species with high plasticity and rapid growth, such as *Q. virginiana* in our study, will obtain rapid canopy closure, allowing favorable conditions for establishment of associated mid-and late-seral maritime forest species (Gómez-Aparicio et al., 2009; Bertacchi et al., 2016), helping to create resilient, diverse, and complex forests (Löf et al., 2019).

### REFERENCES


# DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

# AUTHOR CONTRIBUTIONS

ET designed and installed the experiment, collected and analyzed the data, and co-wrote the paper. OB helped to design and install the experiment, contributed to data analysis and interpretation, and co-wrote the paper. MM assisted with stomatal data collection and interpretation, and co-wrote the paper. DJ supervised the research, helped to design the experiment, and co-wrote the paper.

# FUNDING

Funding support was provided by the USDA National Institute of Food and Agriculture, McIntire Stennis projects IND011535 and NMSU1002447, Hardwood Tree Improvement and Regeneration Center, Fred M. van Eck Forest Foundation, John T. Harrington Forestry Center, and St. Simon's Land Trust.

# ACKNOWLEDGMENTS

St. Simon's Land Trust provided access to research sites. In kind support from the St. Simon's Land Trust, The Nature Conservancy, and Georgia Department of Natural Resources. Thank you to Stephanie Knox, Susan Shipman, Andrei Toca, Michael Gosney, Mercedes Uscola, Michael Szuter, Caleb Reddick, Madeline Montague, Edward Oehlman, Bailey Elkins, Noel Mano, Sofia Sifnaios, Jacob Thompson, and Tate Holbrook for valuable assistance with this project.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.01526/ full#supplementary-material


R Core Team (2019). R: a language and environment for statistical computing.


insights to overcome arrested succession. *For. Ecol. Manage.* 445, 26–36. doi: 10.1016/j.foreco.2019.05.004


**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 Thyroff, Burney, Mickelbart and Jacobs. 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.*