Long-term forest resilience to climate change indicated by mortality, regeneration and growth in semi-arid southern Siberia

Abstract

Several studies have documented that regional climate warming and the resulting increase in drought stress have triggered increased tree mortality in semi-arid forests with unavoidable impacts on regional and global carbon sequestration. Although climate warming is projected to continue into the future, studies examining long-term resilience of semi-arid forests against climate change are limited. In this study, long-term forest resilience was defined as the capacity of forest recruitment to compensate for losses from mortality. We observed an obvious change in long-term forest resilience along a local aridity gradient by reconstructing tree growth trend, disturbance history and investigating post-disturbance regeneration in semi-arid forests in southern Siberia. In our study, with increased severity of local aridity, forests became vulnerable to drought stress, and regeneration first accelerated and then ceased. Radial growth of trees during 1900-2012 was also relatively stable on the moderately arid site. Furthermore, we found that smaller forest patches always have relatively weaker resilience under the same climatic conditions. Our results imply a relatively higher resilience in arid timberline forest patches than in continuous forests; however, further climate warming and increased drought could possibly cause the disappearance of small forest patches around the arid treeline. This study sheds light on climate change adaptation and provides insight into managing vulnerable semi-arid forests. This article is protected by copyright. All rights reserved.

Full text

Long-term forest resilience to climate change indicated by
mortality, regeneration, and growth in semiarid southern
Siberia
CHONGYANG XU
1
, HONGYAN LIU
1
, OLEG A. ANENKHONOV
2
,
ANDREY YU KOROLYUK
3
, DENIS V. SANDANOV
2
,LARISAD.BALSANOVA
2
,
BULAT B. NAIDANOV
2
and XIUCHEN WU
4
1
College of Urban and Environmental Sciences and MOE Laboratory for Earth Surface Processes, Peking University, Beijing
100871, China,
2
Institute of General and Experimental Biology, Siberian Branch, Russian Academy of Sciences, Ulan-Ude 670047,
Russia,
3
Central Siberian Botanical Garden, Siberian Branch, Russian Academy of Sciences, Novosibirsk 630090, Russia,
4
College
of Resources Science and Technology, Beijing Normal University, Beijing 100875, China
Abstract
Several studies have documented that regional climate warming and the resulting increase in drought stress have
triggered increased tree mortality in semiarid forests with unavoidable impacts on regional and global carbon seques-
tration. Although climate warming is projected to continue into the future, studies examining long-term resilience of
semiarid forests against climate change are limited. In this study, long-term forest resilience was defined as the capac-
ity of forest recruitment to compensate for losses from mortality. We observed an obvious change in long-term forest
resilience along a local aridity gradient by reconstructing tree growth trend and disturbance history and investigating
postdisturbance regeneration in semiarid forests in southern Siberia. In our study, with increased severity of local
aridity, forests became vulnerable to drought stress, and regeneration first accelerated and then ceased. Radial growth
of trees during 1900–2012 was also relatively stable on the moderately arid site. Furthermore, we found that smaller
forest patches always have relatively weaker resilience under the same climatic conditions. Our results imply a rela-
tively higher resilience in arid timberline forest patches than in continuous forests; however, further climate warming
and increased drought could possibly cause the disappearance of small forest patches around the arid tree line. This
study sheds light on climate change adaptation and provides insight into managing vulnerable semiarid forests.
Keywords: arid timberline, arid tree line, climate change, forest resilience, forest–steppe, patch size, southern Siberia
Received 22 July 2016; revised version received 5 November 2016 and accepted 16 November 2016
Introduction
Semiarid forests distributed along the arid timberline
(Rotenberg & Yakir, 2010) are particularly vulnerable to
even slight increases in water stress in seasonally or
episodically water-limited environments (Liu et al.,
2013). Warming-induced increases in both vapor
pressure deficits and heat waves have been widely
documented to result in tree growth decline and forest
die-off in semiarid forests (Allen et al., 2010). Such
drought-kill disturbance may reduce forest canopy and
biomass, and even cause rapid and pronounced vegeta-
tion declines in semiarid regions, which may further
feedback to local/regional climate (Huxman et al., 2005;
Bonan, 2008; Chasmer et al., 2011; Levine et al., 2016).
Therefore, answering the question of whether semiarid
forests are resilient to drought-kill disturbance is cru-
cial to understand permanent vegetation change under
local and even global climate dynamics.
Forest resilience is the ability of a forest ecosystem to
maintain fundamental characteristics, such as carbon
pools, composition, and structure, under climate
change (Medvigy et al., 2009; Levine et al., 2016).
Thereby, it maintains both short-term resilience, such
as increased water-use efficiency with drought (Ponce-
Campos et al., 2013), and long-term resilience, such as
equilibrium between rates of key processes (e.g., forest
regeneration and mortality) for forests (Folke et al.,
2004; Walker et al., 2004; Zhou et al., 2013). Rapid cli-
mate change is predicted to have a negative impact on
ecosystem resilience in humid forests, with decreased
tree radial growth, increased tree mortality but
unchanged recruitment rates and vegetation shifts
(Allen & Breshears, 1998; Van Mantgem & Stephenson,
2007; Malhi et al., 2009; Van Mantgem et al., 2009;
Enquist & Enquist, 2011; Zhou et al., 2013; Levine et al.,
2016). However, studies of resilience of semiarid forests
against changing climatic conditions are very limited.
Due to the relatively lower diversity of dominant tree
species in the canopy, the resilience ability of semiarid
Correspondence: Hongyan Liu, tel./fax: +86 10 62759319,
e-mail: lhy@urban.pku.edu.cn
2370 ©2016 John Wiley & Sons Ltd
Global Change Biology (2017) 23, 2370–2382, doi: 10.1111/gcb.13582

forests in mid- and high latitudes in the northern hemi-
sphere is mainly determined by the tree radial growth
trends and the capacity of recruitment to compensate
for forest loss from mortality.
The Inner Asian forest–steppe dominates the south-
ernmost fringe of the Siberian taiga. Similar to alpine
regions, continuous forests here are fragmented into
forest patches within the arid timberline ecotone, and
water-deficit stress along the aridity gradient can cause
a further shift to a sparse savanna-like pattern beyond
the aridity limits for forest vegetation, which is consid-
ered as arid tree line (Stevens & Fox, 1991). Water limi-
tations and the highly fragmented landscapes with
forest patches of various sizes coaffect tree growth sen-
sitivity to a changing climate (Vygodskaya et al., 2002;
Dulamsuren et al., 2010, 2013; Chenlemuge et al., 2015).
Parts of Inner Asia, including southern Siberia, have
recently experienced rapid warming with increasingly
frequent and severe drought events, and this warming
pattern is projected to continue into the future (Li et al.,
2009; Sheffield et al., 2009; Liu et al., 2013). Forest mor-
tality and ceased regeneration have been documented
recently on the southern edge of taiga (Gilbert et al.,
2001; Dulamsuren et al., 2010, 2013; Lloyd et al., 2011;
Kharuk et al., 2013a,b). These studies have mostly
focused on forest mortality and regeneration sepa-
rately. However, little is known about whether semi-
arid forests here can regenerate effectively to
compensate for the reported increasing tree mortality
under climate warming. Likewise, although climatic
influence on forest mortality or regeneration has been
studied in several previous works, the effects of for-
est patch size, an important determinant of the forest
interior environment and tree growth in semiarid
regions (Chenlemuge et al., 2015), have been rela-
tively less thoroughly studied. Thus, the resilience
pattern along continuous forests, arid timberline for-
ests, and arid tree line forests remains unclear. Here
we hypothesized that semiarid forests were not resili-
ent to climate change from long-term perspective,
with increasing mortality and ineffective recruitment
to compensate for forest loss, and resilience pattern
in southern Siberia should decrease along both cli-
mate and forest patch size gradient.
In this study, we evaluated the resilience of semiarid
forests in the southern margin of the Siberian taiga by
reconstructing disturbance history and investigating
the postdisturbance regeneration status. To test the
above hypotheses, we compared differences in forest
resilience indicated by forest mortality, regeneration,
and growth on three sites located along an aridity gra-
dient with three different patch sizes on each site to fig-
ure out how climate and patch size affect forest
resilience in southern Siberia.
Materials and methods
Study area and field work
Transbaikalia is a mountainous region of southern Siberia sit-
uated to the east of or ‘beyond’ (trans-) Lake Baikal in Russia.
Our study area was chosen in the forest–steppe ecotone of
Transbaikalia, between the southern margin of the Siberian
taiga and the Inner Asian steppe. Larix sibirica and Pinus syl-
vestris are the most dominant tree species forming widespread
light coniferous forests in this region. The study sites are
located within elevations of about 650–900 m a.s.l. Mean
annual temperature is about 5to2°C, and mean annual
precipitation is about 240–380 mm (Table 1).
Three sites along an aridity gradient were chosen to sample
in this region in 2013 (Fig. 1): Armak (ARM) near the continu-
ous forest, Khoshoon-Uzurh (KHO) at the arid timberline, and
Dyrestui (DYR) near the arid tree line. Three plots with similar
topographical features but of different forest patch sizes
(large, medium, and small, Table 1) were established on each
study site. Three to five subplots of 20 m 910 m in size were
set up from the upper to lower borders of the forest patch
within each plot. At each subplot, we measured the height
and d.b.h. (diameter at breast height) of every tree. For trees
that were too small to measure d.b.h., diameter at base (about
5 cm from ground) was measured instead (Qi et al., 2014). All
dead trees were also counted. Two cores, one parallel to slope
and another parallel to contour, were taken for all trees with
d.b.h. >5 cm and height >200 cm. Cores were carefully col-
lected to ensure that each core reached the pith of the tree.
Cones on the ground and seeds both in cones and in the soil
were collected in five grids in the four corners and in the cen-
ter of each subplot to assess the potential regeneration ability
of the forest. No obvious evidence of fire, insect attack, or
Table 1 Site characteristics, including location, dominant
tree species, approximate area of each patch, mean annual
precipitation (MAP) and temperature (MAT), and mean
annual PDSI (mPDSI)
Sites Armak
Khoshoon-
Uzurh Dyrestui
Site-ID ARM KHO DYR
Dominant
tree
species
Larix
sibirica
L. sibirica
(accompanied
by P. sylvestris)
Pinus
sylvestris
Latitude (°N) 50.578 51.232 50.566
Longitude
(°E)
104.633 107.507 105.964
Altitude
(m a.s.l.)
939 877 668
MAP (mm) 380 243 272
MAT (°C) 4.8 5.2 2.7
mPDSI 0.2 1.2 1.4
Patch size (ha)
large/
medium/
small
20.8/8.5/0.8 24.0/6.3/2.1 14.3/4.1/1.9
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
SEMIARID FOREST RESILIENCE TO CLIMATE CHANGE 2371

human activity (e.g., logging and grazing) was found on the
three sites.
A total of 580 relev
es (100 m
2
each, mostly 10 m 910 m)
were sampled for the vegetation analysis: 175 on ARM, 180 on
KHO, and 225 on DYR. We avoided areas with significant
gypsy moth (Lymantria dispar asiatica) damage, so vegetation
could be considered as being relatively undisturbed.
Relev
es on each site were plotted along different hillsides
(Anenkhonov et al., 2015).
Estimation of habitat wetness with indicator values of
plants
General evaluation of ecotopological differentiation of wetness
conditions favorable for forest vegetation was analyzed with
the ‘Hill model’ (Anenkhonov et al., 2015). For that reason, an
indicator value for moisture for every relev
e on each site was
calculated as the average of the total importance values of all
species registered within the plot and analyzed through the
model. The indicator value represented the optimal position
of a species along a wetness gradient, ranging from 1 (abso-
lutely dry) to 120 (aquatic habitats) (Anenkhonov et al., 2015).
The Hill model is a method of evaluating the totality of relev
es
and provides a means for presenting vegetation distribution
for rugged landscapes, such as small hills and mountains. This
approach was developed using ARCVIEW 3.2 software (www.e
sri.com) specifically to assist in recognizing the spatial struc-
ture of landscapes where plant communities are variably
distributed along surfaces according to different aspects,
slope, and shape.
Chronology development and history reconstruction
Cores were dried, sanded with successively finer grades of
sandpaper, and measured using the TSAPwin system (Qi et al.,
2014). Quality of cross-dating was then checked using COFECHA
software (Holmes, 1983). Series that were not correlated with
the entire dataset were discarded. A total of 896 cores from 463
trees were available after cross-dating. Mean ages of trees,
number of samples from each plot, common statistics such as
expressed population signal (EPS), and mean sensitivity are
listed in Table 2. The EPS, a measure of chronology reliability,
was generally above 0.84, a generally accepted level (Liu et al.,
2013). Mean sensitivities ranged from 0.39 to 0.59. Trees on the
DYR site were relatively older than those on the other two sites.
To test whether semiarid forests are resilience to climate
change from radial growth perspective (Imbert & Portecop,
2008; Salzer et al., 2009), basal area increment (BAI) during
1900–2012 was calculated based on measured ring widths as
the following equation:
BAI ¼pðr2
tr2
t1Þ
where r
t
is the radial radius at year tand r
t1
is the radial
radius at year t1 (Monserud & Sterba, 1996). BAI of all trees
within each plot was averaged into a master BAI series
(Qi et al., 2014).
Fig. 1 Study area and location of sampling sites (red dots). Subpanel at right shows the relative areas of forest patches with different
sizes (Image from Google Earth, captured May 15, 2013). Photographs below show forest mortality in the study region (taken August,
2013). ARM, Armak; KHO, Khoshoon-Uzurh; DYR, Dyrestui. [Colour figure can be viewed at wileyonlinelibrary.com].
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
2372 C. XU et al.

Age structures are essentially records of both establishment
and subsequent survival to the time of sampling, which can
be used to roughly represent regeneration history (Szeicz &
Macdonald, 1995; Wu et al., 2014). We defined the number of
rings at 1.3 m in height (breast height) plus 10 years (an esti-
mated time for trees to reach breast height) as adult tree age
(Dulamsuren et al., 2013; Wu et al., 2014). For dead adult trees
with rotten cores, ages were estimated through linear relation-
ships of d.b.h. and tree ages. For other small trees, seedlings,
and saplings, tree age was estimated by counting the number
of branches (Liang et al., 2011). Forest recruitment history for
different plots was investigated by summarizing the age of
trees into age distributions at 5-year intervals (Wang et al.,
2006; Wu et al., 2014). We assessed the forest age structures for
ensemble plots.
A percentage growth-change analysis (Nowacki & Abrams,
1997) was used to assess the growth release patterns for the
nine plots using the following equation:
%GC ¼M2M1
M1
100
where %GC represented percentage growth change between
preceding and subsequent 10-year means, and M
1
and M
2
were the preceding and subsequent 10-year ring width means,
respectively. A master growth-change chronology for each
plot was built by averaging the percentage growth change for
all trees in each year. The percent of growth change reaching
50% was defined as a minor growth release, and major growth
releases were identified when %GC ≥100% (Wu et al., 2014).
As we cannot obtain the exact mortality history from tree-ring
chronologies, this growth release history method is suitable
for roughly evaluating possible canopy disturbance history
(Wu et al., 2014).
Possible climatic effects on growth release were assessed by
analyzing simple linear correlations between the percentage
growth changes (%GC) and the mean annual Palmer Drought
Severity Index (PDSI, http://www.cru.uea.ac.uk/). We
assessed the correlation between %GC and PDSI
diff
, where
PDSI
diff
is the difference between preceding and subsequent
10-year PDSI means of an entire year (Wu et al., 2014).
Measurement of postdisturbance natural regeneration
Besides disturbance and regeneration history reconstruction,
we also investigated forest regeneration status after a big
drought-kill disturbance. Increased forest mortality in the
neighboring regions has been observed since 2006 (Kharuk
et al., 2013a). Here we evaluated the disturbance degree of
each plot by estimating both total mortality ratio and
mortality ratio in each age group from 0 to 120 years with
20-year intervals.
The density of seedlings (d.b.h <2 cm, height <50 cm) and
saplings (2 cm ≤d.b.h. <5 cm, 50 cm ≤height <500 cm)
was used as an indicator to assess forest regeneration status.
Amounts of cones and seeds in the soil are considered as an
indicator of the potential regeneration ability for the forest
(Brock & Rogers, 1998; Moles & Drake, 1999). We kept seeds
in moist sand at 4 °C in the dark before the germination exper-
iment. To evaluate the vitality of seeds, germination experi-
ments were carried out in laboratory. Seeds of L. sibirica and
P. sylvestris were put on the moist filter paper in Petri dishes.
Three Petri dishes served as replicate samples; twenty seeds
were put on each Petri dish as subsamples. The Petri dishes
were exposed at a constant relative humidity of 60%. Temper-
ature was set of 15 °C at ‘night’. During the ‘day’, temperature
was set to 20 °C and photosynthetic photon flux density was
set to 300 lmol m
2
s
1
(13 h). The samples were kept in a
growth chamber for 16 days, and germinated seeds were
counted daily (Dulamsuren et al., 2013).
Statistical analysis
Pearson correlation analysis between age structure and the
percentage of growth change and between the percentage of
growth change and PDSI
diff
was conducted to analyze rela-
tionships between regeneration history, disturbance history,
and drought. The generalized linear model (GLM) was
applied to evaluate the influence of aridity and patch sizes on
the difference in forest disturbance and regeneration ability
among the three sites. GLM has been widely used in ecological
studies which allow data with distributions other than normal
to be used (Nelder & Baker, 1972). Kruskal–Wallis test was
applied on each site, respectively, to assess the influence of
patch size under the same climatic condition, which is a non-
parametric test for data with non-normal distribution to ana-
lyze the variance. Kruskal–Wallis test was applied based on
the 13 subplots (five subplots at large patch, five subplots at
medium patch, and three subplots at small patch) on each site.
Results
Assessments of habitat aridity on the three sites
Analysis revealed more humid condition for the ARM
site and driest condition on the DYR site (Fig. 2). In
Table 2 Statistical chronologies of the nine plots
Sites
No. of
cores
(trees)
Average tree
age
(age range)
Mean
sensitivity EPS
ARM-L 88 (49) 42 (10–122) 0.45 0.986
ARM-M 140 (71) 56 (25–97) 0.49 0.977
ARM-S 94 (51) 32 (18–60) 0.56 0.989
KHO-L 112 (57) 83 (20–137) 0.46 0.952
KHO-M 121 (61) 60 (24–173) 0.58 0.975
KHO-S 59 (32) 47 (19–68) 0.57 0.833
DYR-L 93 (47) 82 (51–144) 0.39 0.986
DYR-M 163 (82) 58 (34–90) 0.44 0.978
DYR-S 26 (13) 73 (31–151) 0.44 0.868
Seedlings and saplings were included in average tree age and
age range. L, M, and S indicate large, medium, and small
patches on the three sites. ARM, Armak; KHO, Khoshoon-
Uzurh; DYR, Dyrestui.
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
SEMIARID FOREST RESILIENCE TO CLIMATE CHANGE 2373

contrast, the KHO site is moderately arid and is charac-
terized by high ecological ‘contrast’ due to a large pro-
portion of dry steppes with an abundance of
xeromesophytic forest communities. Ecotopological site
conditions favorable for forest vegetation (according to
Hill models) are continuous on the ARM and KHO sites
and relatively fragmented on the DYR site (Fig. 2).
Tree recruitment history and disturbance history
The tree age–d.b.h. relationships at the nine patches
were statistically significant (P<0.01), except the large
and small patches on the KHO site (P=0.18 and 0.77,
Table 3). These relationships were used to roughly esti-
mate the ages of trees with rotten cores at each patch,
respectively.
Age structures are generally different within the
three sites. On the ARM site, there was one episodic
recruitment peak for the large and small patches,
respectively, in 1975–2005 and 1970–1980. A bimodal
recruitment pattern was observed at the medium patch,
with the first peak centered during 1935–1960, and the
second during about 1975–1985 (Fig. 3a). Compara-
tively, a different recruitment pattern was observed on
the KHO site, with slight increases but no obvious peak
during 1900–2005 at the three patches (Fig. 3b). How-
ever, after 2005, the percentage of young trees at the
large patch of the KHO site sharply increased from
7.5% to 26.4%. The three patches on the DYR site all
had one recruitment peak, which occurred at approxi-
mately the same time during 1925–1950 (Fig. 3c). Nota-
bly, few saplings were found after 1975. Only the small
patch had recruitment after 2005, with young trees
accounting for 14.6%.
There was a common growth release pattern on the
ARM and DYR sites, with major peaks (exceeding
100%) during 1915–1945 and one or two smaller peaks
(lower than 50%) since the 1970s (Fig. 3d–f). For the
ARM site, there was one obvious peak (exceeding
500%) during 1918–1920 at the large patch. Then in
approximately 1940, the growth releases for both large
and medium patches on the ARM site had peaks with
average sizes of more than 50%. Since the 1970s, only
one peak with an average size of 50% occurred (in
2000) for the large patch. For the small patch on the
Fig. 2 Hill models for vegetation on (a) ARM, (b) KHO, and (c) DYR sites. Outer circle represents flat slopes (up to 5°), and other circles
show more steep slopes (from 5°to 15°, more than 15°). Colors indicate the indicator value for moisture (IVM): 44–48 (dry bunchgrass
steppe), 48–52 (typical bunchgrass steppe with abundant herbs), 52–56 (meadow steppe and xerophytic forests), and 56–60 (dry
meadow and grassy xeromesophytic forests). ARM, Armak; KHO, Khoshoon-Uzurh; DYR, Dyrestui. [Colour figure can be viewed at
wileyonlinelibrary.com].
Table 3 Coefficients of linear relationship between tree age
and d.b.h. for ARM, KHO, and DYR sites
ab r
2
Pdf
ARM-L 2.28 6.39 0.63 <0.01 49
ARM-M 0.85 44.84 0.14 <0.01 71
ARM-S 0.97 20.37 0.09 <0.05 51
KHO-L 0.78 70.39 0.04 0.18 57
KHO-M 1.47 38.91 0.10 <0.01 61
KHO-S 0.28 43.22 0.01 0.77 32
DYR-L 0.92 60.46 0.14 <0.01 47
DYR-M 0.74 43.03 0.25 <0.01 82
DYR-S 1.87 29.52 0.26 <0.01 13
‘a’and‘b’ indicate the coefficients of y=b+ax. r
2
and Pindicate
the correlation coefficients between tree age and d.b.h. df
indicates the degree of freedom for each linear regression. L, M,
and S indicate large, medium, and small patches on three sites.
ARM, Armak; KHO, Khoshoon-Uzurh;DYR,Dyrestui.
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
2374 C. XU et al.

ARM site, there were two peaks of growth release with
relatively low average sizes (50% and 42%, respec-
tively) during 1960–2000 (Fig. 3d). For the three patches
on the KHO site, there is a common pattern with four
major peaks occurring, respectively, during 1911–1917,
1930–1940, 1965–1974, and 1988–2004, but the average
sizes of growth release were generally less than 100%
(Fig. 3e). For each patch of the DYR site, major growth
release peaks with average sizes ranging from 191% to
50% were found during 1915–1945. Another recent
common major growth release peak occurred around
1980 at the three patches, but the average sizes were
quite different for large (78%), medium (17%), and
small (10%) patches, respectively (Fig. 3f).
Forest recruitment history for the ARM and DYR
sites correlates significantly with the percentage of
growth change (P<0.01), except for large patches on
the ARM site (P=0.68, Fig. 4a). However, there was
not a significant relationship between forest recruit-
ment history and percent of growth change for the
three patches of the KHO site, with correlation coeffi-
cients 0.03, 0.10, and 0.12 (P>0.05, Fig. 4a). The GLM
results showed that aridity class could explain 20.48%
of the differences’ variation in correlation between
growth release and recruitment history, while patch
size gradient could explain 5.66%, and their interaction
explained 23.88%.
PDSI
diff
was shown to have a close relationship with
percentage growth changes on the KHO site (Fig. 4b).
The correlation coefficients were 0.72, 0.50, and 0.59 for
the large, medium, and small patches. However, no sig-
nificant relationship was found on the ARM site
(P>0.01). Weak relationships were found on the DYR
site, with coefficients below 0.4. Aridity gradient
explained 14.02% of the differences’ variation in
correlation between percent of growth change and
PDSI
diff
. Patch size gradient could explain 17.61%, and
the explanation ratio of their interaction was 2.15%,
according to the GLM results.
Demographic statistics of postdisturbance regeneration
Forest mortality ratios on the ARM and DYR sites ran-
ged from 10% to 40% (ARM: 19.2%, 37.1%, and 9.8%;
DYR: 9.8%, 5.0%, and 23.7%, Fig. 5a). The mortality
ratios on the KHO site were much higher than those
on the other two sites (47.9% vs. 22.0% and 12.8%,
Fig. 5a), with a significant increase (P<0.1) in mortal-
ity ratio along with decreasing patch size within the
KHO site, indicated by Kruskal–Wallis test. For the
ARM site, the mortality ratio of trees in the 0–40 age
class (40–80%) was much higher than those of older
age classes (around 20%, Fig. 5b). While similar results
were observed for large and medium forest patches on
the DYR site, dead trees at the small patch of the DYR
site were mainly in the 60- to 100-year age class
(Fig. 5b). For the large patch on the KHO site, the mor-
tality ratios of trees 0–80 years old were generally
Fig. 3 History of tree recruitment on (a) ARM, (b) KHO, and (c) DYR sites, and the growth release pattern (%GC) on the (d) ARM, (e)
KHO, and (f) DYR sites. Bars give the distribution percentages of trees in different age classes (5-year interval). Lines indicate the aver-
age size of %GC for all trees at each patch. %GC indicates the percentage of growth change. Two horizontal dash lines indicate the
major (100%) and minor (50%) growth releases. ARM, Armak; KHO, Khoshoon-Uzurh; DYR, Dyrestui. [Colour figure can be viewed at
wileyonlinelibrary.com].
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
SEMIARID FOREST RESILIENCE TO CLIMATE CHANGE 2375

around 15–37%, but for medium and small patches,
the mortality ratio of trees older than 80 years was
about 50–94% (Fig. 5b).
Densities of living adult trees (d.b.h. >5 cm,
height >500 cm) on the three sites were similar
(Fig. 6a–c). Densities of both seedlings and saplings on
the KHO site were clearly higher than the other two
sites, especially for the large patch, which had the high-
est density of seedlings (0.07 trees m
2
, Fig. 6b). The
densities of living adult trees, seedlings, and saplings
on the KHO site were obviously decreasing along the
patch size gradient, while the same pattern was not
observed on the other two sites.
Amounts of cones decreased along the patch size gra-
dient on each site (Fig. 7a, ARM: 56–35 cone m
2
,
KHO: 19–5 cone m
2
, and DYR: 37–22 cone m
2
). The
average amount of cones on the KHO site was clearly
lower than that on the other two sites. However, only
cones with seeds inside were found on the ARM site
(Fig. 7b), with a significantly higher amount at the large
patch than at the other two patches (P<0.1). The
amount of total seeds in the soil was generally lower
than 150 grain m
2
on the ARM and DYR sites, and
significant differences were found within the DYR site,
as suggested by the Kruskal–Wallis test (Fig. 7c). On
the KHO site, the amounts of soil seeds at the large and
medium patches reached 300–400 grain m
2
, but for
the small patch, the amounts of seeds in the soil bank
were significantly lower (82 grain m
2
,P<0.1). The
amounts of viable seeds showed a similar pattern
among the three sites, with clearly higher amounts of
valid seeds for large and medium patches on the KHO
site than those on the other two sites (Fig. 7d). How-
ever, the amount of valid seeds for the small patch of
the KHO site was still much lower than that for the
large and medium patches. Few valid seeds were found
at the three patches of the DYR site, especially for the
small patch.
Tree growth trends during 1900–2012
For each patch, its annual BAI is shown in Fig. 8. On
the ARM site, tree growth for the large patch is obvi-
ously larger than that for the medium and small
patches (Fig. 8a). Since 1990, annual BAI for the three
patches of the ARM site were observed to decline. Tree
growths for the large and medium patches of the KHO
site were relatively stable during 1900–1990, despite a
slight decline after 1990 (Fig. 8b). However, the annual
BAI at the small patch had a relatively short lifetime
with larger variability and obviously increased since
1990. Little difference was found among the annual
BAI for the three patches on the DYR site, with a slight
decline during 2000–2012 (Fig. 8c).
Discussion
Climate-determined forest resilience patterns
Our findings partially rejected our hypothesis, showing
a relatively high resilience of semiarid forests on the
moderately arid site. Besides, the result showed a clear
spatial pattern in semiarid forest resilience, indicated
by the difference between forest mortality, regenera-
tion, and radial growth, which is obviously affected by
local arid conditions and forest structure (e.g., patch
size).
There are several related regional studies on either
forest mortality or regeneration. For example, increas-
ing tree mortality in the Siberian dark-coniferous and
birch forests was reported based on dendrochronology
and remote-sensing results (Kharuk et al., 2013a,b). A
decline in regeneration has also been observed for
Siberian larch in northern Mongolia, which is close to
the southern Siberian taiga of our study region (Dulam-
suren et al., 2010). These studies tend to show declines
in forest regeneration under drought, which is different
from the increased regeneration observed under
Fig. 4 Correlation relationship (indicated by Pearson correla-
tion coefficients) between (a) the percent of growth change (%
GC) and age structure, and (b) between the percentage of
growth change (%GC) and PDSI
diff
for each patch on the three
sites. ARM, Armak; KHO, Khoshoon-Uzurh; DYR, Dyrestui.
[Colour figure can be viewed at wileyonlinelibrary.com].
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
2376 C. XU et al.

moderately arid condition in our studies. However,
whether forest regeneration can compensate for the
reported tree mortality remains poorly quantified in
these semiarid forests, adding to one of the major
uncertainties on site-level resilience of the southern
Siberian taiga. In our study, differences in forest mor-
tality and regeneration patterns among the three sites
along an aridity gradient and within the three patches
on each site were used to quantify differences in forest
resilience along with diverse local environments.
The tree recruitment process on the least arid ARM
site was episodic and mainly driven by disturbance his-
tory, which is consistent with a previous study in east-
ern Kazakhstan for the same forest type (Dulamsuren
et al., 2013). Weak correlation between disturbance his-
tory and PDSI
diff
indicated drought was not the main
factor driving canopy disturbance. Some possible natu-
ral or anthropogenic factors (e.g., fire, windstorms, or
logging) possibly affected canopy disturbance; how-
ever, we cannot know the complete disturbance histo-
ries of these forests (Wu et al., 2014).
On the moderately arid KHO site, regeneration pro-
cesses were shown to be successive with no obvious
peak. Meanwhile, results also showed an increase in
regeneration ability with relatively higher amounts of
soil seeds, seedlings, and saplings than those on the
other two sites. A high mortality ratio and strong corre-
lation between PDSI
diff
and disturbance indicated that
Fig. 5 (a) Total mortality ratios and (b) mortality ratio for each age group from 0 to 120 years with 20-year interval on ARM, KHO, and
DYR sites. a, b, and c in the panels indicate the significant difference between forests with different patch sizes on each site tested by
Kruskal–Wallis test (P<0.1). L, M, and S indicate large, medium, and small patches on the three sites. ARM, Armak; KHO, Khoshoon-
Uzurh; DYR, Dyrestui. [Colour figure can be viewed at wileyonlinelibrary.com].
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
SEMIARID FOREST RESILIENCE TO CLIMATE CHANGE 2377

forest canopies on the KHO site were easily disturbed
by drought, which could have altered the site condi-
tions with increased light availability and reduced com-
petition stress on young trees, which further favors the
regeneration process (Wu et al., 2014; Chenlemuge
et al., 2015; Elkin et al., 2015). These kinds of
disturbance could not induce regeneration peaks, but
could force the trees to keep recruiting. Thus, moderate
aridity stress seems to have different influences on
semiarid forest mortality and regeneration, severely
killing adult trees but promoting seed germination and
sapling survival.
Fig. 6 Tree density of living adult trees, saplings, and seedlings on (a) ARM, (b) KHO, and (c) DYR sites. L, M, and S indicate large,
medium, and small patches on the three sites. ARM, Armak; KHO, Khoshoon-Uzurh; DYR, Dyrestui. [Colour figure can be viewed at
wileyonlinelibrary.com].
Fig. 7 Amount of cones (a), amount of seeds in cones (b), amounts of total seeds in soil (c), and amounts of valid seeds (d) on three
sites with large, medium, and small patch size. a, b, and c in the panels indicate the significant difference between forests with different
patch size on each site tested by Kruskal–Wallis test (P<0.1). L, M, and S indicate large, medium, and small patches on the three sites.
ARM, Armak; KHO, Khoshoon-Uzurh; DYR, Dyrestui. [Colour figure can be viewed at wileyonlinelibrary.com].
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
2378 C. XU et al.

However, on the most arid DYR site, forest regenera-
tion was episodic and highly correlated with distur-
bance before the 1950s, but ceased since the 1970s,
despite one minor growth release in 1980. Similar
reduced or ceased regeneration was also observed in
previous studies in Central and Inner Asia, which is
partly explained by increasing topsoil desiccation in a
warmer climate and a high drought susceptibility of
larch seed germination (Dulamsuren et al., 2010, 2013;
Wu et al., 2014). Our results also showed a lack of soil
seeds on the DYR site, especially a lack of valid seeds
in the soil, which implied low potential regeneration
ability. Notably, both the driest environment and the
entire plot with large trees on the DYR site indicated a
high risk of drought-induced tree mortality (Mcdowell
& Allen, 2015). However, our study showed a relatively
lower mortality ratio on the DYR site than that on the
KHO site, which could be explained by the relatively
higher drought tolerance of P. sylvestris (Anenkhonov
et al., 2015).
Semiarid forests were shown to have a certain degree
of resilience to climate change, with timely recruitment
after a severe disturbance. However, our results
implied such resilience may track a threshold-based
nonlinear trajectory associated with local climatic con-
ditions. When the local climate is exceedingly arid for
seedling and sapling survival, regeneration will cease.
Thus, local climatic conditions are likely to be an impor-
tant determinant affecting semiarid forest resilience to
climate change.
Patch size partially affected semiarid forest resilience
In addition to local arid conditions, patch size was
demonstrated to play an important role on forest resili-
ence under dry climate. The results of Kruskal–Wallis
test showed a significant increase in mortality ratio and
a decrease in regeneration ability with decreasing patch
size on the two drier sites, which indicated that tree
growth and recruitment at smaller patch sizes may be
more vulnerable to increasing drought stress (when the
local climate is sufficiently dry). Higher edge–interior
ratio and reduced canopy cover make the interior envi-
ronment warmer and drier at small forest patches,
which can result in more severe water deficit induced
by increasing temperatures than at large patches. Such
soil desiccation may have a negative effect on seed ger-
mination and seedling survival (Dulamsuren et al.,
2013; Mcdowell et al., 2013). Besides, adult trees at the
small patches may grow fast and have high hydraulic
conductivity at the sapling stage as a result of reduced
competition for water and light. However, large-dia-
meter xylem conduits, one of the fundamental anatomi-
cal characteristics of high hydraulic conductivity, are
more prone to drought-induced embolism, which can
cause hydraulic failure and tree mortality (Maherali
et al., 2006; Brodribb & Cochard, 2009; Domec et al.,
2010; Chenlemuge et al., 2015). Such environmental
characteristics and physiological features of trees at
small forest patches may lead to weak resilience to
drought-kill disturbance.
The high explanations of residuals in GLM results
suggested that there are several other potential factors
affecting forest mortality and regeneration, for instance,
tree species. In our study region, dominant tree species
are different among sites, with L. sibirica on the ARM
site, L. sibirica accompanied by P. sylvestris on the KHO
site, and P. sylvestris on the DYR site. Besides, complex
processes at multiple vertical levels from within soil
layers up to the canopy could also affect forest resili-
ence (Thompson et al., 2009). But we tend to believe
that climatic conditions were likely a more general indi-
cator of long-term resilience in semiarid forests, as we
found a clear pattern with reduced resilience along
decreasing patch size for the same species on the same
site, and patch size was also an indicator of moisture
condition in the forest interior. The clear pattern on for-
est resilience along both the climatic condition gradient
and patch size gradient indicated that climatic condi-
tion was more important for forest long-term resilience
Fig. 8 Basal area increment (BAI) chronology of each patch on (a) ARM, (b) KHO, and (c) DYR sites. Gray areas are the variations in
BAI of all trees on each site (average BAI standard variation). ARM, Armak; KHO, Khoshoon-Uzurh; DYR, Dyrestui. [Colour figure can
be viewed at wileyonlinelibrary.com].
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
SEMIARID FOREST RESILIENCE TO CLIMATE CHANGE 2379

than other factors, although some factors such as differ-
ent tree species might cause large differences in short-
term resilience (Choat et al., 2012).
Tree growth as an evidence of relatively stable resilience
Tree growth for all sites as a whole had slightly
declined since 1990, which is consistent with reports
from several related studies (Dulamsuren et al., 2010,
2013; Liu et al., 2013). The recent drought of the 1990s
and 2000s documented in previous study has resulted
in large declines in radial growth rates in semiarid
Inner Asia (Liu et al., 2013). Different growth patterns,
however, were observed among the three sites, with
relatively more stable trends of lifetime growth on the
moderately arid site than that on the other two sites.
Radial growth of all trees on the ARM site had a
declined trend since 1990, suggesting a possible reduc-
tion in the forest resilience. Meanwhile, tree growth on
the KHO site is much stable than those on the other
two sites, except the unstable tree growth for the small
patch with a clear increase since 1990. Tree growth
trend on the DYR site declined since 2000, which was
similar with that on the ARM site.
Although we are unable to reconstruct changes in
forest ecosystem functions or essential taxonomic com-
position and structures adopted as resilience indicator
in some previous studies (Folke et al., 2004; Walker
et al., 2004), the result of stable radial growth trends
and high mortality followed by effective recruitment
suggested that semiarid forests under moderately arid
conditions have been stable, which indicated a rela-
tively high resilience to climate change.
Forests at the arid timberline are more resilient to climate
change
Based on the results of semiarid forest resilience in this
study, our result suggested a conceptual model regard-
ing differences in forest resilience between continuous
forests, arid timberline forests, and arid tree line forests
(Fig. 9). Along the aridity gradient, forest size
decreased, from continuous forests to forest patches at
the arid timberline, then to sparse savanna-like patterns
at the arid tree line. While the change in forest resili-
ence is nonlinear with increased mortality and nonlin-
ear recruitment, it was manifested as being higher in
arid timberline forests and lower in continuous forests
and arid tree line forests, which partially rejected our
hypothesis.
Several previous works have documented that con-
tinuous forests, for instance, tropical rainforests, boreal
forests, and temperate forests in Europe, show
decreased resilience against climate change and have
transitioned from moist forests to dry forests or treeless
states (Scheffer et al., 2012; Drobyshev et al., 2013; Zhou
et al., 2013; Grossiord et al., 2014; Levine et al., 2016). In
this study, the arid timberline forests were shown to be
more resilient than the continuous forests, with rela-
tively higher regeneration ability than tree mortality.
Although frequent and severe droughts increase the
frequency of canopy disturbance and enhance mortality
among adult trees, successive regeneration induced by
altered site conditions could recruit tree individuals
and stabilize forest ecosystems, which is likely to be
regarded as an adaptive strategy to the arid conditions.
However, increased regeneration ability could not pro-
vide an enlargement of forest patch area in semiarid
regions, as seedlings and saplings are likely to be killed
by severe drought due to shallow rooting (Mcdowell
et al., 2013; O’Brien et al., 2014; Zhang et al., 2015).
Lower resilience at smaller forest patches suggests that
forests near the arid tree line, which are sparse and
small in size due to the extremely arid conditions,
might display weak resilience to climate change.
Ceased regeneration and increased mortality could pos-
sibly cause forest replacement by grasslands or shrub-
lands along the arid tree line (Fig. 9).
In summary, forests at the arid timberline may have
relatively higher resilience from the perspective of for-
est demography (i.e., balance of tree mortality and
regeneration) than continuous forests and forests at the
arid tree line, which is very important for permanent
vegetation succession in semiarid regions. Under the
projections of increasing climatic aridity in the future,
small forest patches near the driest tree line might be
Fig. 9 Conceptual model of changes in resilience between con-
tinuous forests, arid timberline forests, and arid tree line forests.
Black spots roughly show the relative location of the three sites.
ARM, Armak; KHO, Khoshoon-Uzurh; DYR, Dyrestui. [Colour
figure can be viewed at wileyonlinelibrary.com].
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
2380 C. XU et al.

potentially replaced by shrubs or grasses, and continu-
ous forests may shift into fragmented forests or treeless
vegetation (Scheffer et al., 2012; Levine et al., 2016),
while semiarid forests at the arid timberline may main-
tain a relatively stable state, which may have a great
influence on local and regional carbon sequestration
(Rotenberg & Yakir, 2010; Schimel, 2010). Comprehen-
sive understanding of the resilience of semiarid forests
to changing climate from multiperspectives, including
demography and physiology, is urgently needed and
vital for more accurate prediction of future semiarid
forest dynamics.
Acknowledgements
This research was granted by National Natural Science
Foundation of China (NSFC 41325002 and 41530747).
References
Allen CD, Breshears DD (1998) Drought-induced shift of a forest-woodland ecotone:
rapid landscape response to climate variation. Proceedings of the National Academy
of Sciences,95, 14839–14842.
Allen CD, Macalady AK, Chenchouni H et al. (2010) A global overview of drought
and heat-induced tree mortality reveals emerging climate change risks for forests.
Forest Ecology and Management,259, 660–684.
Anenkhonov OA, Korolyuk AY, Sandanov DV, Liu HY, Zverev AA, Guo DL (2015)
Soil-moisture conditions indicated by field-layer plants help identify vulnerable
forests in the forest-steppe of semi-arid Southern Siberia. Ecological Indicators,57,
196–207.
Bonan GB (2008) Forests and climate change: forcings, feedbacks, and the climate
benefits of forests. Science,320, 1444–1449.
Brock MA, Rogers KH (1998) The regeneration potential of the seed bank of an
ephemeral floodplain in South Africa. Aquatic Botany,61, 123–135.
Brodribb TJ, Cochard H (2009) Hydraulic failure defines the recovery and point of
death in water-stressed conifers. Plant Physiology,149, 575–584.
Chasmer L, Quinton W, Hopkinson C, Petrone R, Whittington P (2011) Vegetation
canopy and radiation controls on permafrost plateau evolution within the discon-
tinuous permafrost zone, Northwest Territories, Canada. Permafrost and Periglacial
Processes,22, 199–213.
Chenlemuge T, Dulamsuren C, Hertel D, Schuldt B, Leuschner C, Hauck M (2015)
Hydraulic properties and fine root mass of Larix sibirica along forest edge-interior
gradients. Acta Oecologica,63,28–35.
Choat B, Jansen S, Brodribb TJ et al. (2012) Global convergence in the vulnerability of
forests to drought. Nature,491, 752–755.
Domec JC, Schafer K, Oren R, Kim HS, McCarthy HR (2010) Variable conductivity
and embolism in roots and branches of four contrasting tree species and their
impacts on whole-plant hydraulic performance under future atmospheric CO
2
concentration. Tree Physiology,30, 1001–1015.
Drobyshev I, Gewehr S, Berninger F, Bergeron Y (2013) Species specific growth
responses of black spruce and trembling aspen may enhance resilience of boreal
forest to climate change. Journal of Ecology,101, 231–242.
Dulamsuren C, Hauck M, Khishigjargal M, Leuschner HH, Leuschner C (2010)
Diverging climate trends in Mongolian taiga forests influence growth and regener-
ation of Larix sibirica.Oecologia,163, 1091–1102.
Dulamsuren C, Wommelsdorf T, Zhao FJ, Xue YQ, Zhumadilov BZ, Leuschner C,
Hauck M (2013) Increased summer temperatures reduce the growth and regenera-
tion of Larix sibirica in southern boreal forests of eastern Kazakhstan. Ecosystems,
16, 1536–1549.
Elkin C, Giuggiola A, Rigling A, Bugmann H (2015) Short- and long-term efficacy of
forest thinning to mitigate drought impacts in mountain forests in the European
Alps. Ecological Applications,25, 1083–1098.
Enquist BJ, Enquist CA (2011) Long-term change within a Neotropical forest: assess-
ing differential functional and floristic responses to disturbance and drought. Glo-
bal Change Biology,17, 1408–1424.
Folke C, Carpenter S, Walker B, Scheffer M, Elmqvist T, Gunderson L, Holling CS
(2004) Regime shifts, resilience, and biodiversity in ecosystem management.
Annual Review of Ecology, Evolution, and Systematics,35, 557–581.
Gilbert GS, Harms KE, Hamill DN, Hubbell SP (2001) Effects of seedling size, El Ni ~
no
drought, seedling density, and distance to nearest conspecific adult on 6-year sur-
vival of Ocotea whitei seedlings in Panam
a. Oecologia,127, 509–516.
Grossiord C, Granier A, Ratcliffe S et al. (2014) Tree diversity does not always
improve resistance of forest ecosystems to drought. Proceedings of the National
Academy of Sciences,111, 14812–14815.
Holmes RL (1983) Computer-assisted quality control in tree-ring dating and measure-
ment. Tree-Ring Bulletin,43,69–78.
Huxman TE, Wilcox BP, Breshears DD et al. (2005) Ecohydrological implications of
woody plant encroachment. Ecology,86, 308–319.
Imbert D, Portecop J (2008) Hurricane disturbance and forest resilience: assessing
structural vs. functional changes in a Caribbean dry forest. Forest Ecology and Man-
agement,255, 3494–3501.
Kharuk VI, Im ST, Oskorbin PA, Petrov IA, Ranson KJ (2013a) Siberian pine decline
and mortality in southern Siberian Mountains. Forest Ecology and Management,310,
312–320.
Kharuk VI, Ranson KJ, Oskorbin PA, Im ST, Dvinskaya ML (2013b) Climate induced
birch mortality in Trans-Baikal lake region, Siberia. Forest Ecology and Management,
289, 385–392.
Levine NM, Zhang K, Longo M et al. (2016) Ecosystem heterogeneity determines the
ecological resilience of the Amazon to climate change. Proceedings of the National
Academy of Sciences,113, 793–797.
Li JB, Cook ER, D’arrigo R, Chen FH, Gou XH (2009) Moisture variability across
China and Mongolia: 1951–2005. Climate Dynamics,32, 1173–1186.
Liang EY, Wang YF, Eckstein D, Luo TX (2011) Little change in the fir tree-line posi-
tion on the southeastern Tibetan Plateau after 200 years of warming. New Phytolo-
gist,190, 760–769.
Liu HY, Williams AP, Allen CD et al. (2013) Rapid warming accelerates tree
growth decline in semi-arid forests of Inner Asia. Global Change Biology,19,
2500–2510.
Lloyd AH, Bunn AG, Berner L (2011) A latitudinal gradient in tree growth response
to climate warming in the Siberian taiga. Global Change Biology,17, 1935–1945.
Maherali H, Moura CF, Caldeira MC, Willson CJ, Jackson RB (2006) Functional coor-
dination between leaf gas exchange and vulnerability to xylem cavitation in tem-
perate forest trees. Plant, Cell & Environment,29, 571–583.
Malhi Y, Arag~
ao LE, Galbraith D et al. (2009) Exploring the likelihood and mecha-
nism of a climate-change-induced dieback of the Amazon rainforest. Proceedings of
the National Academy of Sciences,106, 20610–20615.
Mcdowell NG, Allen CD (2015) Darcy’s law predicts widespread forest mortality
under climate warming. Nature Climate Change,5, 669–672.
Mcdowell NG, Ryan MG, Zeppel MJ, Tissue DT (2013) Feature: improving our
knowledge of drought-induced forest mortality through experiments, observa-
tions, and modeling. New Phytologist,200, 289–293.
Medvigy D, Wofsy SC, Munger JW, Hollinger DY, Moorcroft PR (2009) Mecha-
nistic scaling of ecosystem function and dynamics in space and time: ecosys-
tem demography model version. Journal of Geophysical Research: Biogeosciences,
114, G01002.
Moles AT, Drake DR (1999) Potential contributions of the seed rain and seed bank to
regeneration of native forest under plantation pine in New Zealand. New Zealand
Journal of Botany,37,83–93.
Monserud RA, Sterba H (1996) A basal area increment model for individual trees
growing in even-and uneven-aged forest stands in Austria. Forest Ecology and Man-
agement,80,57–80.
Nelder JA, Baker R (1972) Generalized linear models. Encyclopedia of Statistical
Sciences,135, 370–384.
Nowacki GJ, Abrams MD (1997) Radial-growth averaging criteria for reconstructing
disturbance histories from presettlement-origin oaks. Ecological Monographs,67,
225–249.
O’Brien MJ, Leuzinger S, Philipson CD, Tay J, Hector A (2014) Drought survival of
tropical tree seedlings enhanced by non-structural carbohydrate levels. Nature Cli-
mate Change,4, 710–714.
Ponce-Campos GE, Moran MS, Huete A et al. (2013) Ecosystem resilience despite
large-scale altered hydroclimatic conditions. Nature,494, 349–352.
Qi ZH, Liu HY, Wu XC, Hao Q (2014) Climate-driven speedup of alpine treeline for-
est growth in the Tianshan Mountains, Northwestern China. Global Change Biology,
21, 816–826.
Rotenberg E, Yakir D (2010) Contribution of semi-arid forests to the climate system.
Science,327, 451–454.
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
SEMIARID FOREST RESILIENCE TO CLIMATE CHANGE 2381

Salzer MW, Hughes MK, Bunnb AG, Kipfmueller KF (2009) Recent unprecedented
tree-ring growth in bristlecone pine at the highest elevations and possible causes.
Proceedings of the National Academy of Sciences,106, 20348–20353.
Scheffer M, Hirota M, Holmgren M, Van Nes EH, Chapin FS (2012) Thresholds for
boreal biome transitions. Proceedings of the National Academy of Sciences,109,
21384–21389.
Schimel DS (2010) Drylands in the earth system. Science,327, 418–419.
Sheffield J, Andreadis KM, Wood EF, Lettenmaier DP (2009) Global and continental
drought in the second half of the twentieth century: severity-area-duration analy-
sis and temporal variability of large-scale events. Journal of Climate,22, 1962–1981.
Stevens GC, Fox JF (1991) The causes of treeline. Annual Review of Ecology and System-
atics,22, 177–191.
Szeicz JM, Macdonald GM (1995) Recent white spruce dynamics at the sub-
arctic alpine treeline of north-western Canada. Journal of Ecology,83, 873–885.
Thompson I, Mackey B, McNulty S, Mosseler A (2009) Forest Resilience, Biodiversity,
and Climate Change, Vol 43. Technical Series. Secretariat of the Convention on Bio-
logical Diversity, Montreal.
Van Mantgem PJ, Stephenson NL (2007) Apparent climatically induced increase of
tree mortality rates in a temperate forest. Ecology Letters,10, 909–916.
Van Mantgem PJ, Stephenson NL, Byrne JC et al. (2009) Widespread increase of tree
mortality rates in the western United States. Science,323, 521–524.
Vygodskaya NN, Schulze ED, Tchebakova NM et al. (2002) Climatic control of stand
thinning in unmanaged spruce forests of the southern taiga in European Russia.
Tellus,54B, 443–461.
Walker B, Holling CS, Carpenter SR, Kinzig A (2004) Resilience, adaptability and
transformability in social-ecological systems. Ecology and Society,9,5.
Wang T, Zhang QB, Ma KP (2006) Treeline dynamics in relation to climatic variability
in the central Tianshan Mountains, northwestern China. Global Ecology and Biogeog-
raphy,15, 406–415.
Wu XC, Liu HY, He LB et al. (2014) Stand-total tree-ring measurements and forest
inventory documented climate-induced forest dynamics in the semi-arid Altai
Mountains. Ecological Indicators,36, 231–241.
Zhang J, Huang SM, He FL (2015) Half-century evidence from western Canada shows
forest dynamics are primarily driven by competition followed by climate. Proceed-
ings of the National Academy of Sciences,112, 4009–4014.
Zhou GY, Peng CH, Li YL et al. (2013) A climate change-induced threat to the ecologi-
cal resilience of a subtropical monsoon evergreen broad-leaved forest in Southern
China. Global Change Biology,19, 1197–1210.
©2016 John Wiley & Sons Ltd, Global Change Biology,23, 2370–2382
2382 C. XU et al.