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

Abstract

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 inter-annual 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.

Full text

fpls-09-01964 January 8, 2019 Time: 16:34 # 1
ORIGINAL RESEARCH
published: 08 January 2019
doi: 10.3389/fpls.2018.01964
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@irstea.fr
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, ˇ
Cufar K,
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
Early-Warning Signals of Individual
Tree Mortality Based on Annual
Radial Growth
Maxime Cailleret1,2*, Vasilis Dakos3, Steven Jansen4, Elisabeth M. R. Robert5,6,7 ,
Tuomas Aakala8, Mariano M. Amoroso9,10 , Joe A. Antos11 , Christof Bigler1,
Harald Bugmann1, Marco Caccianaga12 , Jesus-Julio Camarero13 , Paolo Cherubini2,
Marie R. Coyea14 , Katarina ˇ
Cufar15 , Adrian J. Das16, Hendrik Davi17,
Guillermo Gea-Izquierdo18 , Sten Gillner19, Laurel J. Haavik20,21, Henrik Hartmann22 ,
Ana-Maria Here ¸s23,24 , Kevin R. Hultine25, Pavel Janda26, Jeffrey M. Kane27 ,
Viachelsav I. Kharuk28,29, Thomas Kitzberger30,31, Tamir Klein32, Tom Levanic33,
Juan-Carlos Linares34 , Fabio Lombardi35, Harri Mäkinen36, Ilona Mészáros37 ,
Juha M. Metsaranta38 , Walter Oberhuber39, Andreas Papadopoulos40 ,
Any Mary Petritan2,41 , Brigitte Rohner2, Gabriel Sangüesa-Barreda42 , Jeremy M. Smith43 ,
Amanda B. Stan44 , Dejan B. Stojanovic45, Maria-Laura Suarez46, Miroslav Svoboda26 ,
Volodymyr Trotsiuk2,26,47, Ricardo Villalba48 , Alana R. Westwood49, Peter H. Wyckoff50
and Jordi Martínez-Vilalta5,51
1Department of Environmental Systems Science, Forest Ecology, Institute of Terrestrial Ecosystems, ETH Zürich, Zurich,
Switzerland, 2Swiss Federal Institute for Forest, Snow and Landscape Research – WSL, Birmensdorf, Switzerland, 3CNRS,
IRD, EPHE, ISEM, Université de Montpellier, Montpellier, France, 4Institute of Systematic Botany and Ecology, Ulm
University, Ulm, Germany, 5CREAF, Cerdanyola del Vallès, Catalonia, Spain, 6Ecology and Biodiversity, Vrije Universiteit
Brussel, Brussels, Belgium, 7Laboratory of Wood Biology and Xylarium, Royal Museum for Central Africa, Tervuren, Belgium,
8Department of Forest Sciences, University of Helsinki, Helsinki, Finland, 9Consejo Nacional de Investigaciones Científicas y
Técnicas, CCT Patagonia Norte, Río Negro, Argentina, 10 Instituto de Investigaciones en Recursos Naturales, Agroecología y
Desarrollo Rural, Sede Andina, Universidad Nacional de Río Negro, Río Negro, Argentina, 11 Department of Biology,
University of Victoria, Victoria, BC, Canada, 12 Dipartimento di Bioscienze, Università degli Studi di Milano, Milan, Italy,
13 Instituto Pirenaico de Ecología (IPE-CSIC), Zaragoza, Spain, 14 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,
15 Biotechnical Faculty, University of Ljubljana, Ljubljana, Slovenia, 16 United States Geological Survey, Western Ecological
Research Center, Sequoia and Kings Canyon Field Station, Three Rivers, CA, United States, 17 Ecologie des Forêts
Méditerranéennes (URFM), Institut National de la Recherche Agronomique, Avignon, France, 18 Centro de Investigación
Forestal (CIFOR), Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria, Madrid, Spain, 19 Institute of Forest
Botany and Forest Zoology, TU Dresden, Dresden, Germany, 20 USDA Forest Service, Forest Health Protection, Saint Paul,
MN, United States, 21 Department of Entomology, University of Arkansas, Fayetteville, AR, United States, 22 Department
of Biogeochemical Processes, Max Planck Institute for Biogeochemistry, Jena, Germany, 23 Department of Forest Sciences,
Transilvania University of Brasov, Bras
"ov, Romania, 24 BC3 – Basque Centre for Climate Change, Leioa, Spain, 25 Department
of Research, Conservation and Collections, Desert Botanical Garden, Phoenix, AZ, United States, 26 Faculty of Forestry
and Wood Sciences, Czech University of Life Sciences, Prague, Czechia, 27 Department of Forestry and Wildland Resources,
Humboldt State University, Arcata, CA, United States, 28 Sukachev Institute of Forest, Siberian Division of the Russian
Academy of Sciences, Krasnoyarsk, Russia, 29 Siberian Federal University, Krasnoyarsk, Russia, 30 Department of Ecology,
Universidad Nacional del Comahue, Río Negro, Argentina, 31 Instituto de Investigaciones en Biodiversidad y Medioambiente,
Consejo Nacional de Investigaciones Científicas y Técnicas, Río Negro, Argentina, 32 Department of Plant and Environmental
Sciences, Weizmann Institute of Science, Rehovot, Israel, 33 Department of Yield and Silviculture, Slovenian Forestry Institute,
Ljubljana, Slovenia, 34 Department of Physical, Chemical and Natural Systems, Pablo de Olavide University, Seville, Spain,
35 Department of Agricultural Science, Mediterranean University of Reggio Calabria, Reggio Calabria, Italy, 36 Natural
Resources Institute Finland (Luke), Espoo, Finland, 37 Department of Botany, Faculty of Science and Technology, University
of Debrecen, Debrecen, Hungary, 38 Northern Forestry Centre, Canadian Forest Service, Natural Resources Canada,
Edmonton, AB, Canada, 39 Department of Botany, University of Innsbruck, Innsbruck, Austria, 40 Department of Forestry
and Natural Environment Management, Technological Educational Institute of Stereas Elladas, Karpenisi, Greece, 41 National
Institute for Research and Development in Forestry “Marin Dracea”, Voluntari, Romania, 42 Departamento de Ciencias
Agroforestales, EiFAB, iuFOR – University of Valladolid, Soria, Spain, 43 Department of Geography, University of Colorado,
Boulder, CO, United States, 44 Department of Geography, Planning and Recreation, Northern Arizona University, Flagstaff,
AZ, United States, 45 Institute of Lowland Forestry and Environment, University of Novi Sad, Novi Sad, Serbia, 46 Grupo
Ecología Forestal, CONICET – INTA, EEA Bariloche, Bariloche, Argentina, 47 Department of Environmental Systems Science,
Institute of Agricultural Sciences, ETH Zürich, Zurich, Switzerland, 48 Laboratorio de Dendrocronología e Historia Ambiental,
Frontiers in Plant Science | www.frontiersin.org 1January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 2
Cailleret et al. Early-Warning Signals of Tree Mortality
Instituto Argentino de Nivología, Glaciología y Ciencias Ambientales, CCT CONICET Mendoza, Mendoza, Argentina,
49 Boreal Avian Modelling Project, Department of Renewable Resources, University of Alberta, Edmonton, AB, Canada,
50 Department of Biology, University of Minnesota, Morris, Morris, MN, United States, 51 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 inter-
annual 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 drought-
induced 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
Frontiers in Plant Science | www.frontiersin.org 2January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 3
Cailleret et al. Early-Warning Signals of Tree Mortality
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. Early-
warning 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-to-
year 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 bark-
beetles, 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
Frontiers in Plant Science | www.frontiersin.org 3January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 4
Cailleret et al. Early-Warning Signals of Tree Mortality
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 time-
series (∼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 time-
window 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
COR20 until the last year with complete ring formation, i.e., the
year before tree death. We determined whether absolute values
in SD20, AR120 , and COR20 calculated during the last 20 years
preceding mortality (SD20f, AR120f, and COR20ffor final values)
were significantly different than those during any other previous
20-year period.
As SD20 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 meanRW20 as a fixed explanatory variable.
The same approach was used for AR120 and COR20 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
Frontiers in Plant Science | www.frontiersin.org 4January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 5
Cailleret et al. Early-Warning Signals of Tree Mortality
FIGURE 1 | Example of early-warning signals of tree mortality based on ring-width (RW) data from two Abies alba trees from Mont Ventoux, France (Cailleret et al.,
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.
Frontiers in Plant Science | www.frontiersin.org 5January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 6
Cailleret et al. Early-Warning Signals of Tree Mortality
averaging results to analyze the overall temporal dynamics in
growth metrics. Bootstrap resampling procedures were then used
to test if the LMM residuals for SD20f, AR120f, and COR20f
significantly differed from zero (500 re-samplings).
SD20 and meanRW20 were log-transformed unlike AR120 and
COR20 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 SD20f, AR120f, and COR20fbetween
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−SDBHf≤2.5 cm), or (ii) with a similar
mean RW during the 20-year period before the mortality year
(diffD−SmeanRW20f≤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 SD20f,
AR120f, and COR20fbetween conspecific dying and surviving
trees (diffD−SSD20f, diffD−SAR120f, and diffD−SCOR20f) 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−SRW20f) and
in DBH (diffD−SDBHf) 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−SSD20f, diffD−SAR120f, and diffD−SCOR20f
FIGURE 2 | Temporal change in SD20 (A), AR120 (B), and COR20 (C) before
death averaged for all dying trees and calculated on the original and
detrended RW data. We also show the temporal change in the residuals of the
linear mixed-effects models fitted to these metrics (right y-axes). Shaded
areas represent the 95% confidence intervals of the means. Note that COR20
values were not calculated on not-detrended RW data.
Frontiers in Plant Science | www.frontiersin.org 6January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 7
Cailleret et al. Early-Warning Signals of Tree Mortality
values in the theoretical situation in which dying trees have
similar meanRW20fand DBHfas surviving ones.
Sampling Scheme
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 COR20
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
SD20 calculated on detrended RW data started decreasing
around 30 years before tree death (Figure 2A). This trend in
SD20 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 (resSD20f) 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
resSD20 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 (resAR120f;
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 resAR120 values was not exclusive to the end of a
FIGURE 3 | Variation in the residuals of SD (A), AR1 (B), and COR (C)
calculated over the last 20-year period of the detrended ring-width time series
preceding tree death (resSD20f, resAR120f, and resCOR20f) among mortality
sources and species groups. Error bars depict 95% confidence intervals of
the mean residuals, which were determined from 500 bootstrap resamplings
of the original dataset.
Frontiers in Plant Science | www.frontiersin.org 7January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 8
Cailleret et al. Early-Warning Signals of Tree Mortality
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 resAR120
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−S
SD) based on LMMs adjusted for growth rate (meanRW20f)
and size effects (DBHf) 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
meanRW20f), but this was never consistent across mortality
drivers or sampling schemes (Figures 4C,D and Supplementary
Appendix G).
Finally, we found predominantly lower COR20ffor 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), species-
dependent 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
Frontiers in Plant Science | www.frontiersin.org 8January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 9
Cailleret et al. Early-Warning Signals of Tree Mortality
FIGURE 4 | Mean difference in SD20f(A,B), AR120f(C,D), and COR20f(E,F) values between dying and surviving trees predicted by the linear mixed-effects models
(LMMs) fitted to the original dataset, fixing diffD−SRW20fand diffD−SDBHfat zero. Positive values mean that dying trees showed higher SD20f, AR120f, or COR20f
compared to conspecific surviving trees. Standardization was based on similar meanRW20f(Left) and similar DBHf(Right). Error bars depict 95% confidence intervals
of the predicted mean differences, which were determined from 500 bootstrap resamplings. Estimates of the LMMs are available in Supplementary Table H1.
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.
Frontiers in Plant Science | www.frontiersin.org 9January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 10
Cailleret et al. Early-Warning Signals of Tree Mortality
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 inter-
correlated (Supplementary Appendix C), which can lead to a
misinterpretation of the results. For instance, the decrease in
SD20 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 (AR120f) was generally higher than
average AR120 (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 AR120f
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.
Frontiers in Plant Science | www.frontiersin.org 10 January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 11
Cailleret et al. Early-Warning Signals of 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, bark-
beetle 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 multi-
annual growth rates and trends (Cailleret et al., 2017), the inter-
annual 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, Kˇ
C,
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 tree-
ring 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). Kˇ
C was funded by the Slovenian Research Agency
(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
Frontiers in Plant Science | www.frontiersin.org 11 January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 12
Cailleret et al. Early-Warning Signals of Tree Mortality
REFERENCES
Adams, H. D., Zeppel, M. J., Anderegg, W. R., Hartmann, H., Landhäusser,
S. M., Tissue, D. T., et al. (2017). A multi-species synthesis of physiological
mechanisms in drought-induced tree mortality. Nat. Ecol. Evol. 1, 1285–1291.
doi: 10.1038/s41559-017- 0248-x
Allen, C. D., Breshears, D. D., and McDowell, N. G. (2015). On underestimation
of global vulnerability to tree mortality and forest die-off from hotter
drought in the Anthropocene. Ecosphere 6, 1–55. doi: 10.1890/ES15-
00203.1
Allen, C. D., Macalady, A. K., Chenchouni, H., Bachelet, D., McDowell, N.,
Vennetier, M., et al. (2010). A global overview of drought and heat-induced tree
mortality reveals emerging climate change risks for forests. For. Ecol. Manage.
259, 660–684. doi: 10.1016/j.foreco.2009.09.001
Amoroso, M. M., Daniels, L. D., and Larson, B. C. (2012). Temporal patterns of
radial growth in declining Austrocedrus chilensis forests in Northern Patagonia:
the use of tree-rings as an indicator of forest decline. For. Ecol. Manage. 265,
62–70. doi: 10.1016/j.foreco.2011.10.021
Anderegg, W. R., Berry, J. A., and Field, C. B. (2012). Linking definitions,
mechanisms, and modeling of drought-induced tree death. Trends Plant Sci.
17, 693–700. doi: 10.1016/j.tplants.2012.09.006
Anderegg, W. R., Hicke, J. A., Fisher, R. A., Allen, C. D., Aukema, J., Bentz, B.,
et al. (2015a). Tree mortality from drought, insects, and their interactions in a
changing climate. New Phytol. 208, 674–683. doi: 10.1111/nph.13477
Anderegg, W. R., Kane, J. M., and Anderegg, L. D. (2013a). Consequences of
widespread tree mortality triggered by drought and temperature stress. Nat.
Clim. Change 3, 30–36. doi: 10.1038/nclimate1635
Anderegg, W. R., Klein, T., Bartlett, M., Sack, L., Pellegrini, A. F., Choat, B.,
et al. (2016a). Meta-analysis reveals that hydraulic traits explain cross-species
patterns of drought-induced tree mortality across the globe. Proc. Natl. Acad.
Sci. U.S.A. 113, 5024–5029. doi: 10.1073/pnas.1525678113
Anderegg, W. R., Martinez-Vilalta, J., Cailleret, M., Camarero, J. J., Ewers, B. E.,
Galbraith, D., et al. (2016b). When a tree dies in the forest: scaling climate-
driven tree mortality to ecosystem water and carbon fluxes. Ecosystems 19,
1133–1147. doi: 10.1007/s10021-016- 9982-1
Anderegg, W. R., Plavcová, L., Anderegg, L. D., Hacke, U. G., Berry, J. A.,
and Field, C. B. (2013b). Drought’s legacy: multiyear hydraulic deterioration
underlies widespread aspen forest die-off and portends increased future risk.
Glob. Change Biol. 19, 1188–1196. doi: 10.1111/gcb.12100
Anderegg, W. R., Schwalm, C., Biondi, F., Camarero, J. J., Koch, G., Litvak, M., et al.
(2015b). Pervasive drought legacies in forest ecosystems and their implications
for carbon cycle models. Science 349, 528–532. doi: 10.1126/science.aab1833
Augusto, L., Davies, T. J., Delzon, S., and Schrijver, A. (2014). The enigma of
the rise of angiosperms: Can we untie the knot? Ecol. Lett. 17, 1326–1338.
doi: 10.1111/ele.12323
Bates, D., Mächler, M., Bolker, B., and Walker, S. (2014). Fitting linear mixed-
effects models using lme4. arXiv preprint arXiv: 1406.5823. Available at: https:
//arxiv.org/abs/1406.5823.
Bigler, C., and Bugmann, H. (2004). Predicting the time of tree death using
dendrochronological data. Ecol. Appl. 14, 902–914. doi: 10.1890/03-5011
Bigler, C., and Rigling, A. (2013). Precision and accuracy of tree-ring-based death
dates of mountain pines in the Swiss National Park. Trees 27, 1703–1712.
doi: 10.1007/s00468-013- 0917-6
Boden, S., Kahle, H. P., von Wilpert, K., and Spiecker, H. (2014). Resilience of
Norway spruce (Picea abies (L.) Karst) growth to changing climatic conditions
in Southwest Germany. For. Ecol. Manage. 315, 12–21. doi: 10.1016/j.foreco.
2013.12.015
Bowman, D. M., Brienen, R. J., Gloor, E., Phillips, O. L., and Prior, L. D. (2013).
Detecting trends in tree growth: not so simple. Trends Plant Sci. 18, 11–17.
doi: 10.1016/j.tplants.2012.08.005
Brandt, L. A., Butler, P. R., Handler, S. D., Janowiak, M. K., Shannon, P. D., and
Swanston, C. W. (2017). Integrating science and management to assess forest
ecosystem vulnerability to climate change. J. For. 115, 212–221. doi: 10.5849/
jof.15-147
Bréda, N., Huc, R., Granier, A., and Dreyer, E. (2006). Temperate forest trees and
stands under severe drought: a review of ecophysiological responses, adaptation
processes and long-term consequences. Ann. For. Sci. 63, 625–644. doi: 10.1051/
forest:2006042
Brienen, R. J., Zuidema, P. A., and During, H. J. (2006). Autocorrelated growth
of tropical forest trees: unraveling patterns and quantifying consequences. For.
Ecol. Manage. 237, 179–190. doi: 10.1016/j.foreco.2006.09.042
Brodribb, T. J., Bowman, D. J., Nichols, S., Delzon, S., and Burlett, R. (2010). Xylem
function and growth rate interact to determine recovery rates after exposure to
extreme water deficit. New Phytol. 188, 533–542. doi: 10.1111/j.1469-8137.2010.
03393.x
Bunn, A. G., Jansma, E., Korpela, M., Westfall, R. D., and Baldwin, J. (2013).
Using simulations and data to evaluate mean sensitivity as a useful statistic
in dendrochronology. Dendrochronologia 31, 250–254. doi: 10.1016/j.dendro.
2013.01.004
Büntgen, U., Esper, J., Frank, D. C., Treydte, K., Schmidhalter, M., Nicolussi, K.,
et al. (2005). “The effect of power transformation on RCS–evidence from
three millennial length alpine chronologies,” in Proceedings of the International
conference Tree rings in archaeology, climatology and ecology, TRACE,
Birmensdorf, 141–149.
Büntgen, U., Krusic, P. J., Verstege, A., Sangüesa-Barreda, G., Wagner, S.,
Camarero, J. J., et al. (2017). New tree-ring evidence from the Pyrenees reveals
Western Mediterranean climate variability since medieval times. J. Climate 30,
5295–5318. doi: 10.1175/JCLI-D- 16-0526.1
Cailleret, M., Bigler, C., Bugmann, H., Camarero, J. J., Cufar, K., Davi, H., et al.
(2016). Towards a common methodology for developing logistic tree mortality
models based on ring-width data. Ecol. Appl. 26, 1827–1841. doi: 10.1890/15-
1402.1
Cailleret, M., Jansen, S., Robert, E. M., DeSoto, L., Aakala, T., Antos, J. A., et al.
(2017). A synthesis of radial growth patterns preceding tree mortality. Glob.
Change Biol. 23, 1675–1690. doi: 10.1111/gcb.13535
Cailleret, M., Nourtier, M., Amm, A., Durand-Gillmann, M., and Davi, H. (2014).
Drought-induced decline and mortality of silver fir differ among three sites in
Southern France. Ann. For. Sci. 71, 643–657. doi: 10.1007/s13595-013- 0265-0
Camarero, J. J., Gazol, A., Sangüesa-Barreda, G., Oliva, J., and Vicente-Serrano,
S. M. (2015). To die or not to die: early warnings of tree dieback in response to
a severe drought. J. Ecol. 103, 44–57. doi: 10.1111/1365-2745.12295
Carrer, M., and Urbinati, C. (2006). Long-term change in the sensitivity of tree-
ring growth to climate forcing in Larix decidua.New Phytol. 170, 861–872.
doi: 10.1111/j.1469-8137.2006.01703.x
Carvalhais, N., Forkel, M., Khomik, M., Bellarby, J., Jung, M., Migliavacca, M., et al.
(2014). Global covariation of carbon turnover times with climate in terrestrial
ecosystems. Nature 514, 213–217. doi: 10.1038/nature13731
Cherubini, P., Fontana, G., Rigling, D., Dobbertin, M., Brang, P., and Innes, J. L.
(2002). Tree-life history prior to death: two fungal root pathogens affect tree-
ring growth differently. J. Ecol. 90, 839–850. doi: 10.1046/j.1365-2745.2002.
00715.x
Choat, B., Brodribb, T. J., Brodersen, C. R., Duursma, R. A., López, R., and Medlyn,
B. E. (2018). Triggers of tree mortality under drought. Nature 558, 531–539.
doi: 10.1038/s41586-018- 0240-x
Cook, E. R., and Peters, K. (1997). Calculating unbiased tree-ring indices for the
study of climatic and environmental change. Holocene 7, 361–370. doi: 10.1177/
095968369700700314
Csank, A. Z., Miller, A. E., Sherriff, R. L., Berg, E. E., and Welker, J. M. (2016). Tree-
ring isotopes reveal drought sensitivity in trees killed by spruce beetle outbreaks
in south-central Alaska. Ecol. Appl. 26, 2001–2020. doi: 10.1002/eap.1365
Dai, L., Vorselen, D., Korolev, K. S., and Gore, J. (2012). Generic indicators for loss
of resilience before a tipping point leading to population collapse. Science 336,
1175–1177. doi: 10.1126/science.1219805
Dakos, V., Carpenter, S. R., Brock, W. A., Ellison, A. M., Guttal, V., Ives, A. R.,
et al. (2012a). Methods for detecting early warnings of critical transitions in
time series illustrated using simulated ecological data. PLoS One 7:e41010.
doi: 10.1371/journal.pone.0041010
Dakos, V., Carpenter, S. R., van Nes, E. H., and Scheffer, M. (2015). Resilience
indicators: prospects and limitations for early warnings of regime shifts. Philos.
Trans. R. Soc. B 370:20130263. doi: 10.1098/rstb.2013.0263
Dakos, V., Van Nes, E. H., D’Odorico, P., and Scheffer, M. (2012b). Robustness of
variance and autocorrelation as indicators of critical slowing down. Ecology 93,
264–271. doi: 10.1890/11-0889.1
D’Arrigo, R., Wilson, R., Liepert, B., and Cherubini, P. (2008). On the ‘divergence
problem’ in northern forests: a review of the tree-ring evidence and possible
causes. Glob. Planet. Change 60, 289–305. doi: 10.1016/j.gloplacha.2007.03.004
Frontiers in Plant Science | www.frontiersin.org 12 January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 13
Cailleret et al. Early-Warning Signals of Tree Mortality
Das, A. J., Stephenson, N. L., and Davis, K. P. (2016). Why do trees die?
Characterizing the drivers of background tree mortality. Ecology 97, 2616–2627.
doi: 10.1002/ecy.1497
Dobbertin, M. (2005). Tree growth as an indicator of tree vitality and of tree
reaction to environmental stress: a review. Eur. J. For. Res. 124, 319–333.
doi: 10.1007/s10342-005- 0085-3
Dobbertin, M., and Brang, P. (2001). Crown defoliation improves tree mortality
models. For. Ecol. Manage. 141, 271–284. doi: 10.1016/S0378-1127(00)00335-2
Drake, J. M., and Griffen, B. D. (2010). Early warning signals of extinction in
deteriorating environments. Nature 467, 456–459. doi: 10.1038/nature09389
Esper, J., Schneider, L., Smerdon, J. E., Schöne, B. R., and Büntgen, U. (2015).
Signals and memory in tree-ring width and density data. Dendrochronologia
35, 62–70. doi: 10.1016/j.dendro.2015.07.001
Fischer, E. M., Beyerle, U., and Knutti, R. (2013). Robust spatially aggregated
projections of climate extremes. Nat. Clim. Change 3, 1033–1038. doi: 10.1038/
nclimate2051
Frank, D., Esper, J., and Cook, E. R. (2006). “On variance adjustments in tree-
ring chronology development,” in Tree Rings in Archaeology, Climatology and
Ecology, TRACE, Vol. 4, eds I. Heinrich, H. Gärtner, M. Monbaron, and
G. Schleser, 56–66.
Franklin, J. F., Shugart, H. H., and Harmon, M. E. (1987). Tree death as an
ecological process. BioScience 37, 550–556. doi: 10.2307/1310665
Fritts, H. C. (1976). Tree Rings and Climate. London: Academic Press .
Galiano, L., Martínez-Vilalta, J., and Lloret, F. (2011). Carbon reserves and canopy
defoliation determine the recovery of Scots pine 4 yr after a drought episode.
New Phytol. 190, 750–759. doi: 10.1111/j.1469-8137.2010.03628.x
Galiano, L., Timofeeva, G., Saurer, M., Siegwolf, R., Martínez-Vilalta, J.,
Hommel, R., et al. (2017). The fate of recently fixed carbon after drought
release: towards unravelling C storage regulation in Tilia platyphyllos and Pinus
sylvestris.Plant Cell Environ. 40, 1711–1724. doi: 10.1111/pce.12972
Garcia-Forner, N., Sala, A., Biel, C., Savé, R., and Martínez-Vilalta, J. (2016).
Individual traits as determinants of time to death under extreme drought in
Pinus sylvestris L. Tree Physiol. 36, 1196–1209. doi: 10.1093/treephys/tpw040
Gea-Izquierdo, G., Guibal, F., Joffre, R., Ourcival, J. M., Simioni, G., and Guiot, J.
(2015). Modelling the climatic drivers determining photosynthesis and carbon
allocation in evergreen Mediterranean forests using multiproxy long time series.
Biogeosciences 12, 3695–3712. doi: 10.5194/bg-12-3695-2015
Gea-Izquierdo, G., Viguera, B., Cabrera, M., and Cañellas, I. (2014). Drought
induced decline could portend widespread pine mortality at the xeric ecotone
in managed mediterranean pine-oak woodlands. For. Ecol. Manage. 320, 70–82.
doi: 10.1016/j.foreco.2014.02.025
Gessler, A., Cailleret, M., Joseph, J., Schönbeck, L., Schaub, M., Lehmann, M., et al.
(2018). Drought induced tree mortality–a tree-ring isotope based conceptual
model to assess mechanisms and predispositions. New Phytol. 219, 485–490.
doi: 10.1111/nph.15154
Gillner, S., Rüger, N., Roloff, A., and Berger, U. (2013). Low relative growth
rates predict future mortality of common beech (Fagus sylvatica L.). For. Ecol.
Manage. 302, 372–378. doi: 10.1016/j.foreco.2013.03.032
Girard, F., Vennetier, M., Guibal, F., Corona, C., Ouarmim, S., and Herrero, A.
(2012). Pinus halepensis Mill. crown development and fruiting declined with
repeated drought in Mediterranean France. Eur. J. For. Res. 131, 919–931.
doi: 10.1007/s10342-011- 0565-6
Guada, G., Camarero, J. J., Sánchez-Salguero, R., and Cerrillo, R. M. N. (2016).
Limited growth recovery after drought-induced forest dieback in very defoliated
trees of two pine species. Front. Plant Sci. 7:418. doi: 10.3389/fpls.2016.00418
Guillemot, J., Francois, C., Hmimina, G., Dufrêne, E., Martin-StPaul, N. K.,
Soudani, K., et al. (2017). Environmental control of carbon allocation matters
for modelling forest growth. New Phytol. 214, 180–193. doi: 10.3389/fpls.2016.
00418
Hagedorn, F., Joseph, J., Peter, M., Luster, J., Pritsch, K., Geppert, U., et al. (2016).
Recovery of trees from drought depends on belowground sink control. Nat.
Plants 2:16111. doi: 10.1038/nplants.2016.111
Hartl-Meier, C., Zang, C., Büntgen, U., Esper, J., Rothe, A., Göttlein, A., et al.
(2015). Uniform climate sensitivity in tree-ring stable isotopes across species
and sites in a mid-latitude temperate forest. Tree Physiol. 35, 4–15. doi: 10.1093/
treephys/tpu096
Hartmann, H., Moura, C. F., Anderegg, W. R., Ruehr, N. K., Salmon, Y., Allen,
C. D., et al. (2018). Research frontiers for improving our understanding of
drought-induced tree and forest mortality. New Phytol. 218, 15–28. doi: 10.
1111/nph.15048
Here¸s, A. M., Camarero, J. J., López, B. C., and Martínez-Vilalta, J. (2014).
Declining hydraulic performances and low carbon investments in tree rings
predate Scots pine drought-induced mortality. Trees 28, 1737–1750. doi: 10.
1007/s00468-014-1081-3
Heres, A. M., Martínez-Vilalta, J., and López, B. C. (2012). Growth patterns in
relation to drought-induced mortality at two Scots pine (Pinus sylvestris L.)
sites in NE Iberian Peninsula. Trees 26, 621–630. doi: 10.1007/s00468-011-
0628-9
Herguido, E., Granda, E., Benavides, R., García-Cervigón, A. I., Camarero, J. J., and
Valladares, F. (2016). Contrasting growth and mortality responses to climate
warming of two pine species in a continental Mediterranean ecosystem. For.
Ecol. Manage. 363, 149–158. doi: 10.1016/j.foreco.2015.12.038
Hülsmann, L., Bugmann, H., Cailleret, M., and Brang, P. (2018). How to kill a tree:
empirical mortality models for 18 species and their performance in a dynamic
forest model. Ecol. Appl. 28, 522–540. doi: 10.1002/eap.1668
Johnson, D. M., McCulloh, K. A., Woodruff, D. R., and Meinzer, F. C. (2012).
Hydraulic safety margins and embolism reversal in stems and leaves: Why are
conifers and angiosperms so different? Plant Sci. 195, 48–53. doi: 10.1016/j.
plantsci.2012.06.010
Jump, A. S., Ruiz-Benito, P., Greenwood, S., Allen, C. D., Kitzberger, T.,
Fensham, R., et al. (2017). Structural overshoot of tree growth with climate
variability and the global spectrum of drought-induced forest dieback. Glob.
Change Biol. 23, 3742–3757. doi: 10.1111/gcb.13636
Kane, J. M., and Kolb, T. E. (2014). Short-and long-term growth characteristics
associated with tree mortality in southwestern mixed-conifer forests. Can. J.
For. Res. 44, 1227–1235. doi: 10.1139/cjfr-2014-0186
Körner, C. (2015). Paradigm shift in plant growth control. Curr. Opin. Plant Biol.
25, 107–114. doi: 10.1016/j.pbi.2015.05.003
Kuznetsova, A., Brockhoff, P. B., and Christensen, R. H. B. (2017). LmerTest
package: tests in linear mixed effects models. J. Stat. Softw. 82, 1–26. doi: 10.
18637/jss.v082.i13
Lenton, T. M., Dakos, V., Bathiany, S., and Scheffer, M. (2017). Observed trends
in the magnitude and persistence of monthly temperature variability. Sci. Rep.
7:5940. doi: 10.1038/s41598-017- 06382-x
Lin, Y. S., Medlyn, B. E., Duursma, R. A., Prentice, I. C., Wang, H., Baig, S., et al.
(2015). Optimal stomatal behavior around the world. Nat. Clim. Change 5,
459–464. doi: 10.1038/nclimate2550
Lloret, F., Escudero, A., Iriondo, J. M., Martínez-Vilalta, J., and Valladares, F.
(2012). Extreme climatic events and vegetation: the role of stabilizing processes.
Glob. Change Biol. 18, 797–805. doi: 10.1111/j.1365-2486.2011.02624.x
Lloret, F., Keeling, E. G., and Sala, A. (2011). Components of tree resilience: effects
of successive low-growth episodes in old ponderosa pine forests. Oikos 120,
1909–1920. doi: 10.1111/j.1600-0706.2011.19372.x
Macalady, A. K., and Bugmann, H. (2014). Growth-mortality relationships in
piñon pine (Pinus edulis) during severe droughts of the past century: shifting
processes in space and time. PLoS One 9:e92770. doi: 10.1371/journal.pone.
0092770
Manion, P. D. (1991). Tree Disease Concepts. Englewood Cliffs, NJ: Prentice-Hall.
Martín-Benito, D., Cherubini, P., Del Río, M., and Cañellas, I. (2008). Growth
response to climate and drought in Pinus nigra Arn. trees of different crown
classes. Trees 22, 363–373. doi: 10.1007/s00468-007-0191-6
Martínez-Vilalta, J., and Lloret, F. (2016). Drought-induced vegetation
shifts in terrestrial ecosystems: the key role of regeneration dynamics.
Glob. Planet. Change 144, 94–108. doi: 10.1016/j.gloplacha.2016.
07.009
Martínez-Vilalta, J., Sala, A., Asensio, D., Galiano, L., Hoch, G., Palacio, S., et al.
(2016). Dynamics of non-structural carbohydrates in terrestrial plants: a global
synthesis. Ecol. Monogr. 86, 495–516. doi: 10.1002/ecm.1231
McDowell, N., Allen, C. D., and Marshall, L. (2010). Growth, carbon-isotope
discrimination, and drought-associated mortality across a Pinus ponderosa
elevational transect. Glob. Change Biol. 16, 399–415. doi: 10.1111/j.1365-2486.
2009.01994.x
Mencuccini, M., and Munné-Bosch, S. (2017). “Physiological and biochemical
processes related to ageing and senescence in plants,” in The Evolution of
Senescence in the Tree of Life, eds R. P. Shefferson, O. R. Jones, and R.
Salguero-Gómez (Cambridge: Cambridge University Press), 257–283.
Frontiers in Plant Science | www.frontiersin.org 13 January 2019 | Volume 9 | Article 1964

fpls-09-01964 January 8, 2019 Time: 16:34 # 14
Cailleret et al. Early-Warning Signals of Tree Mortality
Mérian, P., and Lebourgeois, F. (2011). Size-mediated climate–growth
relationships in temperate forests: a multi-species analysis. For. Ecol. Manage.
261, 1382–1391. doi: 10.1016/j.foreco.2011.01.019
Millar, C. I., Westfall, R. D., and Delany, D. L. (2007). Response of high-elevation
limber pine (Pinus flexilis) to multiyear droughts and 20th-century warming,
Sierra Nevada, California, USA. Can. J. For. Res. 37, 2508–2520. doi: 10.1139/
X07-097
Millar, C. I., Westfall, R. D., Delany, D. L., Bokach, M. J., Flint, A. L., and Flint,
L. E. (2012). Forest mortality in high-elevation whitebark pine (Pinus albicaulis)
forests of eastern California, USA; influence of environmental context, bark
beetles, climatic water deficit, and warming. Can. J. For. Res. 42, 749–765.
doi: 10.1139/x2012-031
Mueller, R. C., Scudder, C. M., Porter, M. E., Talbot Trotter, R., Gehring, C. A.,
and Whitham, T. G. (2005). Differential tree mortality in response to severe
drought: evidence for long-term vegetation shifts. J. Ecol. 93, 1085–1093.
doi: 10.1111/j.1365-2745.2005.01042.x
Nesmith, J. C., Das, A. J., O’Hara, K. L., and van Mantgem, P. J. (2015). The
influence of prefire tree growth and crown condition on postfire mortality of
sugar pine following prescribed fire in Sequoia National Park. Can. J. For. Res.
45, 910–919. doi: 10.1139/cjfr-2014-0449
Ogle, K., Whitham, T. G., and Cobb, N. S. (2000). Tree-ring variation in pinyon
predicts likelihood of death following severe drought. Ecology 81, 3237–3243.
doi: 10.1890/0012-9658(2000)081[3237:TRVIPP]2.0.CO;2
Olde Rikkert, M. G., Dakos, V., Buchman, T. G., de Boer, R., Glass, L., Cramer,
A. O., et al. (2016). Slowing down of recovery as generic risk marker for
acute severity transitions in chronic diseases. Crit. Care Med. 44, 601–606.
doi: 10.1097/CCM.0000000000001564
Pace, M. L., Carpenter, S. R., and Cole, J. J. (2015). With and without warning:
managing ecosystems in a changing world. Front. Ecol. Environ. 13, 460–467.
doi: 10.1890/150003
Palacio, S., Hernández, R., Maestro-Martínez, M., and Camarero, J. J. (2012). Fast
replenishment of initial carbon stores after defoliation by the pine processionary
moth and its relationship to the re-growth ability of trees. Trees 26, 1627–1640.
doi: 10.1007/s00468-012- 0739-y
Pangle, R. E., Limousin, J. M., Plaut, J. A., Yepez, E. A., Hudson, P. J., Boutz,
A. L., et al. (2015). Prolonged experimental drought reduces plant hydraulic
conductance and transpiration and increases mortality in a piñon–juniper
woodland. Ecol. Evol. 5, 1618–1638. doi: 10.1002/ece3.1422
Pedersen, B. S. (1998). The role of stress in the mortality of midwestern oaks
as indicated by growth prior to death. Ecology 79, 79–93. doi: 10.1890/0012-
9658(1998)079[0079:TROSIT]2.0.CO;2
Pellizzari, E., Camarero, J. J., Gazol, A., Sangüesa-Barreda, G., and
Carrer, M. (2016). Wood anatomy and carbon-isotope discrimination
support long-term hydraulic deterioration as a major cause of drought-
induced dieback. Glob. Change Biol. 22, 2125–2137. doi: 10.1111/gcb.
13227
Peters, R. L., Groenendijk, P., Vlam, M., and Zuidema, P. A. (2015). Detecting
long-term growth trends using tree rings: a critical evaluation of methods. Glob.
Change Biol. 21, 2040–2054. doi: 10.1111/gcb.12826
Plavcová, L., Hoch, G., Morris, H., Ghiasi, S., and Jansen, S. (2016). The amount
of parenchyma and living fibers affects storage of nonstructural carbohydrates
in young stems and roots of temperate trees. Am. J. Bot. 103, 603–612.
doi: 10.3732/ajb.1500489
R Core Team (2017). R: A Language and Environment for Statistical Computing.
Vienna: R Foundation for Statistical Computing.
Raffa, K. F., Aukema, B. H., Bentz, B. J., Carroll, A. L., Hicke, J. A., Turner, M. G.,
et al. (2008). Cross-scale drivers of natural disturbances prone to anthropogenic
amplification: the dynamics of bark beetle eruptions. BioScience 58, 501–517.
doi: 10.1641/B580607
Reyer, C. P., Leuzinger, S., Rammig , A., Wolf,A., B artholomeus, R. P., Bonfante, A.,
et al. (2013). A plant’s perspective of extremes: terrestrial plant responses to
changing climatic variability. Glob. Change Biol. 19, 75–89. doi: 10.1111/gcb.
12023
Rogers, B. M., Solvik, K., Hogg, E. H., Ju, J., Masek, J. G., Michaelian, M., et al.
(2018). Detecting early warning signals of tree mortality in boreal North
America using multiscale satellite data. Glob. Change Biol. 24, 2284–2304.
doi: 10.1111/gcb.14107
Royer, P. D., Cobb, N. S., Clifford, M. J., Huang, C. Y., Breshears, D. D., Adams,
H. D., et al. (2011). Extreme climatic event-triggered overstorey vegetation loss
increases understorey solar input regionally: primary and secondary ecological
implications. J. Ecol. 99, 714–723. doi: 10.1111/j.1365-2745.2011.01804.x
Sánchez-Salguero, R., Navarro, R. M., Camarero, J. J., and Fernández-Cancio, Á
(2010). Drought-induced growth decline of Aleppo and maritime pine forests
in south-eastern Spain. For. Syst. 19, 458–470. doi: 10.5424/fs/2010193-9131
Sangüesa-Barreda, G., Linares, J. C., and Camarero, J. J. (2015). Reduced
growth sensitivity to climate in bark-beetle infested Aleppo pines: connecting
climatic and biotic drivers of forest dieback. For. Ecol. Manage. 357, 126–137.
doi: 10.1016/j.foreco.2015.08.017
Scheffer, M., Bascompte, J., Brock, W. A., Brovkin, V., Carpenter, S. R., Dakos, V.,
et al. (2009). Early warning signals for critical transitions. Nature 461, 53–59.
doi: 10.1038/nature08227
Schiestl-Aalto, P., Kulmala, L., Mäkinen, H., Nikinmaa, E., and Mäkelä, A.
(2015). CASSIA–a dynamic model for predicting intra-annual sink demand
and interannual growth variation in Scots pine. New Phytol. 206, 647–659.
doi: 10.1111/nph.13275
Schulman, E. (1956). Dendroclimatic Changes in Semiarid America. Tucson:
University of Arizona Press.
Steppe, K., Niinemets, Ü, and Teskey, R. O. (2011). “Tree size- and age-related
changes in leaf physiology and their influence on carbon gain,” in Size- and
Age-Related Changes in Tree Structure and Function, eds F. C. Meinzer, B.
Lachenbruch, and T. E. Dawson (Dodrecht: Springer), 235–254. doi: 10.1007/
978-94-007-1242-3
Suarez, M. L., Ghermandi, L., and Kitzberger, T. (2004). Factors predisposing
episodic drought-induced tree mortality in Nothofagus–site, climatic sensitivity
and growth trends. J. Ecol. 92, 954–966. doi: 10.1111/j.1365-2745.2004.00941.x
Timofeeva, G., Treydte, K., Bugmann, H., Rigling, A., Schaub, M., Siegwolf, R., et al.
(2017). Long-term effects of drought on tree-ring growth and carbon isotope
variability in Scots pine in a dry environment. Tree Physiol. 37, 1028–1041.
doi: 10.1093/treephys/tpx041
Trumbore, S., Brando, P., and Hartmann, H. (2015). Forest health and global
change. Science 349, 814–818. doi: 10.1126/science.aac6759
Veraart, A. J., Faassen, E. J., Dakos, V., van Nes, E. H., Lürling, M., and Scheffer, M.
(2012). Recovery rates reflect distance to a tipping point in a living system.
Nature 481, 357–359. doi: 10.1038/nature10723
Verbesselt, J., Umlauf, N., Hirota, M., Holmgren, M., Van Nes, E. H., Herold, M.,
et al. (2016). Remotely sensed resilience of tropical forests. Nat. Clim. Change 6,
1028–1031. doi: 10.1038/nclimate3108
von Arx, G., Arzac, A., Fonti, P., Frank, D., Zweifel, R., Rigling, A., et al. (2017).
Responses of sapwood ray parenchyma and non-structural carbohydrates of
Pinus sylvestris to drought and long-term irrigation. Funct. Ecol. 31, 1371–1382.
doi: 10.1111/1365-2435.12860
Wissel, C. (1984). A universal law of the characteristic return time near thresholds.
Oecologia 65, 101–107. doi: 10.1007/BF00384470
Yin, J., and Bauerle, T. L. (2017). A global analysis of plant recovery performance
from water stress. Oikos 126, 1377–1388. doi: 10.1111/oik.04534
Zeppel, M. J., Harrison, S. P., Adams, H. D., Kelley, D. I., Li, G., Tissue, D. T.,
et al. (2015). Drought and resprouting plants. New Phytol. 206, 583–589.
doi: 10.1111/nph.13205
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
Kˇ
C.
Copyright © 2019 Cailleret, Dakos, Jansen, Robert, Aakala, Amoroso, Antos,
Bigler, Bugmann, Caccianaga, Camarero, Cherubini, Coyea, ˇ
Cufar, Das, Davi,
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.
Frontiers in Plant Science | www.frontiersin.org 14 January 2019 | Volume 9 | Article 1964