Temperature as a potent driver of regional forest drought stress and tree mortality

Abstract

As the climate changes, drought may reduce tree productivity and survival across many forest ecosystems; however, the relative influence of specific climate parameters on forest decline is poorly understood. We derive a forest drought-stress index (FDSI) for the southwestern United States using a comprehensive tree-ring data set representing AD 1000-2007. The FDSI is approximately equally influenced by the warm-season vapour-pressure deficit (largely controlled by temperature) and cold-season precipitation, together explaining 82% of the FDSI variability. Correspondence between the FDSI and measures of forest productivity, mortality, bark-beetle outbreak and wildfire validate the FDSI as a holistic forest-vigour indicator. If the vapour-pressure deficit continues increasing as projected by climate models, the mean forest drought-stress by the 2050s will exceed that of the most severe droughts in the past 1,000 years. Collectively, the results foreshadow twenty-first-century changes in forest structures and compositions, with transition of forests in the southwestern United States, and perhaps water-limited forests globally, towards distributions unfamiliar to modern civilization.

Full text

ARTICLES
PUBLISHED ONLINE: 30 SEPTEMBER 2012 | DOI: 10.1038/NCLIMATE1693
Temperature as a potent driver of regional forest
drought stress and tree mortality
A. Park Williams1*, Craig D. Allen2, Alison K. Macalady3,4, Daniel Griffin3,4, Connie A. Woodhouse3,4,
David M. Meko4, Thomas W. Swetnam4, Sara A. Rauscher5, Richard Seager6,
Henri D. Grissino-Mayer7, Jeffrey S. Dean4, Edward R. Cook6, Chandana Gangodagamage1,
Michael Cai8and Nate G. McDowell1
As the climate changes, drought may reduce tree productivity and survival across many forest ecosystems; however, the
relative influence of specific climate parameters on forest decline is poorly understood. We derive a forest drought-stress
index (FDSI) for the southwestern United States using a comprehensive tree-ring data set representing AD 1000–2007. The
FDSI is approximately equally influenced by the warm-season vapour-pressure deficit (largely controlled by temperature) and
cold-season precipitation, together explaining 82% of the FDSI variability. Correspondence between the FDSI and measures
of forest productivity, mortality, bark-beetle outbreak and wildfire validate the FDSI as a holistic forest-vigour indicator. If the
vapour-pressure deficit continues increasing as projected by climate models, the mean forest drought-stress by the 2050s will
exceed that of the most severe droughts in the past 1,000 years. Collectively, the results foreshadow twenty-first-century
changes in forest structures and compositions, with transition of forests in the southwestern United States, and perhaps
water-limited forests globally, towards distributions unfamiliar to modern civilization.
Recent declines in forest productivity and tree survival have
been documented at many sites globally and attributed to
water limitation1,2. Forest declines may be accelerating in
many regions because warming has amplified water limitation3–10.
We describe forest declines due to water limitation as drought-
induced declines, whether by lack of precipitation or by increased
evaporative demand. In the southwestern United States (SWUS;
Supplementary Fig. S1), drought impacts on forests have been
relatively severe since the late 1990s (refs 5,11). This period has
been punctuated by the highest temperatures in the observed
record, positively influencing atmospheric moisture demand12. The
ongoing heat-driven drought, in combination with a multitude of
ecophysiological data documenting regional forest processes, makes
the SWUS an ideal natural experimental target13 for examining
the influence of heat-induced drought on productivity and tree
mortality in drought-sensitive forests.
Climate modelling experiments collectively project the SWUS to
become warmer and more arid as greenhouse-gas concentrations
rise and the sub-tropical high-pressure zones expand and shift
polewards14,15. Water balance has long been known to substantially
limit tree growth and influence forest disturbances in the SWUS;
however, the relative importance of temperature and precipitation
requires clarification16. Temperature should be expected to influ-
ence water balance because it exponentially influences atmospheric
evaporative demand (Supplementary Fig. S2). Uncertainty regard-
ing the relative roles of evaporative demand versus precipitation in
dictating forest drought-stress and tree mortality has limited our
1Earth & Environmental Sciences Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA, 2US Geological Survey, Fort Collins
Science Center, Jemez Mountains Field Station, Los Alamos, New Mexico 87544, USA, 3School of Geography and Development, University of Arizona,
Tucson, Arizona 85721, USA, 4Laboratory of Tree-ring Research, University of Arizona, Tucson, Arizona 85721, USA, 5Theoretical Division, Los Alamos
National Laboratory, Los Alamos, New Mexico 87545, USA, 6Lamont-Doherty Earth Observatory of Columbia University, Palisades, New York 10964, USA,
7Laboratory of Tree-Ring Science, Department of Geography, The University of Tennessee, Knoxville, Tennessee 37996, USA, 8Space Data Systems, Los
Alamos National Laboratory, Los Alamos, New Mexico 87545, USA. *e-mail: parkwilliams@lanl.gov.
ability to anticipate future impacts to forests, which might otherwise
be estimated through global climate model output. Here we de-
velop a tree-ring-based index of forest drought-stress that explicitly
resolves the contributions of the vapour-pressure deficit (VPD;
difference between the actual- and saturation-vapour pressure,
largely controlled by temperature) and precipitation. We link this
index to the most comprehensive set of regional forest productivity
and disturbance records assembled so far. We use an ensemble
of climate-model projections of VPD and precipitation to project
future levels of forest drought-stress and place them in the context
of historic events known to have caused widespread tree mortality.
Forest drought-stress and climate
Annual tree-ring widths reflect variability in the environmental
stressors that limit growth17. We used tree-ring records to develop
an index of annual forest stress for dominant southwestern tree
species, ad 1000–2007. The index is based on 335 collections of site-
specific tree-ring width measurements (13,147 specimens) from
throughout the SWUS and surrounding regions (Supplementary
Fig. S1; see Supplementary Information for all methods and data
sources). Nearly all chronologies represent the three most abundant
conifer species in the SWUS: piñon (Pinus edulis), ponderosa pine
(P. ponderosa) and Douglas-fir (Pseudotsuga menziesii). Sites reflect
a range of elevations along landscape moisture gradients.
We call the first-principal-component time series of ring-width
chronologies the SWUS FDSI (Fig. 1a) because it is highly
representative of interannual variability in ring width across
NATURE CLIMATE CHANGE |ADVANCE ONLINE PUBLICATION |www.nature.com/natureclimatechange 1
© 2012 Macmillan Publishers Limited. All rights reserved.

ARTICLES NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE1693
Predictive contribution of VPD (%)
Predictive contribution of precipitation (%)
Correlation (r)
1896–2007
1896–1951
1952–2007
Less drought stress
More drought stress
FDSI
r = 0.91
Year
1900 1920 1940 1960 1980 2000
1
¬2
¬1
0
2
1009080706050403020100
100 90 80 70 60 50 40 30 20 10 0
0.6
0.7
0.8
0.9
a
b
Figure 1 |Correlation between the FDSI and climate. a, The annual FDSI
derived from tree ring-width index records (red, 1896–2007) and
estimated with climate data using equation (1) (black, 1896–2012). See
Supplementary Fig. S3 for estimated confidence ranges in the FDSI values.
b, Curves show correlation between the estimated and the actual FDSI,
allowing predictive contributions of the warm-season VPD and cold-season
precipitation to vary from 0 to 100% and 100 to 0%, respectively. The
straight lines connect optimal correlations with axes. Grey areas represent
95% confidence intervals.
space and species (Supplementary Fig. S1) and because the index
correlates well with observed drought-related climate variables.
See Supplementary Fig. S3 for sample size and error. The FDSI
is unique because it explicitly represents regionally coherent
tree-growth variability calculated from all available ring-width
records from the three main conifer species in the SWUS. In
contrast, climate-reconstruction approaches (for example, ref. 18)
are tuned to represent a specific drought variable during a specific
season, using only ring-width records that correlate well with the
drought variable of interest.
Considering observed climate data during 1896–2007, the FDSI
most strongly correlates with the log of cold-season precipitation
(positive relationship) and the warm-season VPD (a combination
of the previous August–October and growing-season May–July
VPD, negative relationship). The FDSI can be estimated using the
following formula:
Estimated FDSI =0.44[zscore(ln(Pndjfm))]
−0.56[zscore((VPDaso +VPDmjj)/2)](1)
where the subscripts are the initials of months, underlined initials
indicate months of the previous year, and zscore indicates time-
series standardization so values during 1896–2007 have a mean of
zero and a standard deviation of one (Supplementary Information).
Coefficients in equation (1) indicate that the warm-season VPD
and cold-season precipitation (P) account for 56% and 44% of the
predictive power, respectively (derived visually in Fig. 1b). Com-
bined, these variables account for 82% of the tree-ring-derived FDSI
variability (p<0.0001, 95% confidence: 0.74 ≤R2≤0.88). Relative
contributions of the warm-season VPD and cold-season precipita-
tion were stable throughout the instrumental period (Fig. 1b).
The seasons when precipitation and the VPD are best
correlated with the FDSI are consistent with our knowledge
of plant physiology. Stemwood growth in the SWUS is most
strongly dependent on soil–water recharge from cold-season
precipitation13,19,20. Warming in spring and summer causes the VPD
to increase exponentially (Supplementary Fig. S2) and soil moisture
decreases through evapotranspiration. Increased VPD coupled with
limited soil moisture increases the potential for hydraulic failure
(collapse of water columns within xylem cells) and can force
prolonged stomatal closure, decreasing photosynthesis, growth rate
and carbohydrate reserves16,21,22. Correlation between growth and
the VPD does not extend into August, consistent with observations
of a mid- to late-summer shutdown of radial growth among SWUS
trees (for example, refs 17,19). Instead, the FDSI is negatively
correlated with the August–October VPD of the previous year.
Although these months are not entirely within the warm season, the
mechanisms by which the previous autumn VPD limits tree growth
are similar to those of the warm season. When conditions allow,
photosynthesis continues during August–October after cambial
shutdown, allowing allocation of carbohydrates to reserves that
influence radial growth during the following growing season17. The
VPD during this period can also influence soil-moisture recharge
by subsequent cold-season precipitation17.
The FDSI reflects variability in regional-scale vegetation growth,
as supported by the strong correlation between the FDSI and
the satellite-derived regional normalized difference vegetation
index (NDVI) during 1981–2012 (Fig. 2a). The FDSI and NDVI
both capture ongoing decline in productivity following the mid-
1990s as well as a minimum in 2002. Turn-of-the-century
drought conditions reduced productivity and ecosystem uptake of
atmospheric carbon throughout western North America10.
Tree mortality
Ongoing drought-driven decline in regional productivity is associ-
ated with widespread tree mortality in the SWUS. Considering the
three focal species in this study, FDSI and NDVI minima in 2002
were followed by an approximate doubling of the proportion of
dead individuals in the SWUS (Fig. 2b). Other studies5,23,24 docu-
ment widespread post-2002 mortality for piñon, which occupies the
most arid habitat of the three focal species, but the same is true for
ponderosa pine and Douglas-fir.
Much of the conifer mortality in the 2000s was caused by
factors influenced by bark-beetle attack and wildfire9,11,13,25–28.
Bark-beetle populations seem to grow during relatively warm
periods and specifically target drought-stressed trees with weakened
defences9,25. Considering aerial survey data for 1997–2011, we find
a negative temporal correspondence between the 2-year running
FDSI and the SWUS area that experienced bark-beetle-induced
mortality of at least 10 trees per acre (Fig. 2c). Although a longer
record is needed to confirm the relationship between the FDSI and
beetle-induced tree mortality, the multi-year relationship between
the area impacted by bark beetles and the FDSI makes sense
because bark-beetle populations grow and decline over multiple
seasons25. Importantly, the logarithmic scale of the left-hand yaxis
in Fig. 2c suggests that incremental increases in drought conditions
promote exponential increases in the potential for beetle-induced
tree mortality. SWUS bark-beetle outbreaks may also be influenced
by forest-stand density, which has generally increased in the past
century owing to forest management practices13,29–32.
Our results also indicate a strong exponential correspondence
between forest drought-stress and satellite measurements of
forest and woodland area burned by wildfire (Fig. 2d). This
is consistent with regional analyses of fire scars left on tree
rings before observed records (for example, refs 13,26,33).
Using SWUS fire-scar data from a network of sites more than
four times as dense as those used by previous studies (for
example, ref. 13), we find that the probability of an extensive
wildfire year in the SWUS was exponentially related to the FDSI
during ad 1650–1899 (Fig. 2d inset). These analyses indicate that
2NATURE CLIMATE CHANGE |ADVANCE ONLINE PUBLICATION |www.nature.com/natureclimatechange
© 2012 Macmillan Publishers Limited. All rights reserved.

NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE1693 ARTICLES
r = 0.83 ¬2
¬1
0
1
2
1980 1990 2000
Year
2010
Wildfire area (km2)
r = ¬0.84
r = ¬0.82
Bark-beetle area (km2)
1
0
¬1
Piñon
Ponderosa pine
Douglas-fir
1¬1 0
5
10
15
20
25
1
0
¬1
1
10
100
1,000
10,000
2
0
¬2
10
100
1,000
10,000
25
50
%
75
a
b
c
d
0.35
0.40
0.45
NDVIPercentage dead
FDSI FDSI2-yr FDSI6-yr FDSI
Figure 2 |Measurements of forest productivity and mortality overlaid on
the FDSI (red, right yaxis). a, The annual average late-June to
early-August NDVI calculated from satellite (1981–1999: AVHRR,
2000–2012: MODIS) imagery. b, Annual forest inventory and analysis
measurements of the percentage of standing dead trees in the SWUS for
the three most common conifer species. Error bars represent standard
deviation of the percentage dead when each year’s forest inventory and
analysis measurements are randomly resampled 1,000 times
(Supplementary Information). c, Aerial-survey-derived estimates of the
area where ≥10 trees per acre were killed by bark-beetle attack.
d, Satellite-derived moderately and severely burned forest and woodland
area in the SWUS. See Supplementary Information and Supplementary
Fig. S4 for methods to calculate burned area. The inset shows the
percentage of years within a given FDSI class that were top-10% fire-scar
years during AD 1650–1899 (the horizontal line is at the expected
frequency of 10%, bins are 0.25 FDSI units wide). In all panels, the FDSI
values for 2008–2012 (open red squares) were estimated by applying
climate data to equation (1). Note the inverted yaxes for the FDSI in b–d.
drought has been, and remains, a primary driver of widespread
wildfires in the SWUS.
Given the exponential relationships established between the
FDSI and tree mortality, severe drought events before the observed
record probably coincided with widespread tree mortality. As
observed climate and mortality data are unavailable for much of
the past millennium, we use the FDSI record to identify other
severe drought events likely to have caused widespread mortality
since ad 1000 (Fig. 3). A drought event is defined as any period
greater than three years when the mean FDSI is negative, the FDSI
is not positive for two consecutive years34, and the FDSI is less
than two standard deviation units below the 1896–2007 mean for
at least one year. Drought-event strength is the sum of the FDSI
values during the event. Updating the FDSI for 2008–2012 with
the FDSI values estimated from equation (1), three drought events
have occurred within the observed climate record: the present
drought (2000–2012, the fifth strongest since ad 1000), 1945–1964
(the sixth strongest) and 1899–1904 (the seventeenth strongest;
Fig. 3). The prolonged 1945–1964 event was indeed associated with
extensive tree mortality in the SWUS as indicated by documentation
FDSI
¬1.0
¬0.5
0.0
0.5
1.0
1000 1200 1400 1600
Year
1800 2000
Figure 3 |Eleven-year smoothed FDSI for AD1000–2012. Black area: 95%
confidence range of the FDSI, representing the range of FDSI values
expected if all 335 chronologies were available. Vertical grey areas highlight
drought events.
of bark-beetle outbreaks30,35, anomalously large wildfires31,32 and
widespread die-off of conifers30,31,35. The 1899–1904 drought was
also associated with forest declines36, although little documented.
Before the 1900s, the 1572–1587 event was the most recent
event exceeding the severity of the present event (Fig. 3). This
megadrought event37,38 ranks as the fourth most severe since
ad 1000 and the most severe since 1300. Although direct mortality
observations are not available for the 1500s event, studies of forest
age structure document a scarcity of trees on today’s landscape
that began growing before the late 1500s (refs 13,31). As lifespans
of SWUS conifers often greatly exceed 400 years, the scarcity of
trees preceding the 1500s event indicates that intense drought
conditions probably led to deaths of a large proportion of trees
living at the time. Before the late 1500s, the correspondence between
records of conifer pollen buried at archaeological sites and tree-ring
widths39 suggests that widespread tree mortality (indicated by
pollen) co-occurred with massive droughts in the 1200s (indicated
by tree rings). Notably, ancient Puebloan populations and land-use
practices were in great flux during this time, confounding the
attribution of a dominant cause of the 1200s forest decline40.
Future forest drought-stress
The ongoing VPD-dominated drought event (Fig. 4a) is consistent
with climate-model projections (phase 3 of the Coupled Model
Intercomparison Project (CMIP3)) of increasing warm-season
VPD (∼3.6% decade−1) throughout the rest of the twenty-first
century in response to business-as-usual greenhouse-gas emissions
scenarios41 (SRES A2; Fig. 4a and Supplementary Fig. S6 for
alternative emissions scenarios: SRES A1B and B1). Dynamically
downscaled (0.5◦geographic resolution) model projections indicate
similar increases in the VPD (Fig. 4a and Supplementary Infor-
mation). Furthermore, most models project a slight decrease in
cold-season precipitation during the second half of this century
(∼−1.25% decade−1, Fig. 4c). Applying model projections to
equation (1), all models indicate negative FDSI trends throughout
the twenty-first century (Fig. 4d). By 2050, the ensemble mean FDSI
is consistently more severe than that of any megadrought since at
least ad 1000 (megadrought conditions are surpassed by 2070 in the
most optimistic B1 emissions scenario, Supplementary Fig. S6d).
Notably, projections of the FDSI are more severe than projections
of gross water balance (precipitation–evaporation) because water-
balance projections are influenced more by decreased cold-season
precipitation than by increased warm-season VPD (ref. 15).
FDSI projections suggest that SWUS forest drought-stress is
entering a new era where natural oscillations such as those apparent
in Fig. 3 are superimposed on, and overwhelmed by, a trend
towards more intense drought stress. As the VPD diverges from the
range of observed variability, nonlinear effects may alter drought
impacts on forests (for example, Fig. 5 in ref. 42). During the
observed record, equation (1) was a better predictor of the FDSI
NATURE CLIMATE CHANGE |ADVANCE ONLINE PUBLICATION |www.nature.com/natureclimatechange 3
© 2012 Macmillan Publishers Limited. All rights reserved.

ARTICLES NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE1693
s.d. anomaly
Warm-season
VPD (kPa)
0
2
4
6
Warm-season
Tmax (°C)
Cold-season P (mm)
FDSI
1.2
1.4
1.6
1.8
23
25
27
29
50
150
250
–4
¬2
0
2
0
2
4
6
1900 1950 2000 2050
Year
2100
a
b
c
d
¬2
¬1
0
1
2
3
¬4
¬2
0
2
Figure 4 |Observed and modelled climate and forest drought-stress.
a–d, The warm-season VPD (a), warm-season Tmax (b), cold-season
precipitation (c) and the FDSI (d). Black: observed records. Coloured bold
lines: CMIP3 ensemble mean values. Shading around time series: inner
50% of CMIP3 values. Green time series: 2042–2069 dynamically
downscaled NARCCAP ensemble mean values. Horizontal brown line and
shading in dshow the mean and 95% confidence FDSI values of the most
severe 50% of years during the 1572–1587 megadrought. The horizontal
grey lines show the anomaly in standard deviations from the observed
1896–2007 mean (right yaxis). See Supplementary Figs S5 and S6 for
individual model projections and alternative emissions scenarios.
in years of relatively high drought-stress (Supplementary Fig. S7).
This may indicate that forest sensitivity intensifies as drought
intensifies, consistent with exponential relationships between the
FDSI and tree mortality (Fig. 2c,d). Interestingly, the observed
intensification of drought sensitivity during high drought-stress
years was mainly due to heightened sensitivity to variability in
cold-season precipitation (Supplementary Fig. S7a,b,g). This may
mean that cold-season precipitation will gain relative importance as
drought intensifies in the coming decades. To account for this and
other possible nonlinear effects, we estimated the future FDSI where
the relative predictive contributions of cold-season precipitation
and the warm-season VPD are forced to vary. Reducing the future
predictive contribution of the warm-season VPD to 25%—less than
half the observed contribution—and increasing the contribution
of cold-season precipitation to 75%, the ensemble mean FDSI is
still estimated to equal or exceed 1500s megadrought levels by the
2060s in a business-as-usual emissions scenario (Supplementary
Fig. S8a). In the hypothetical case that climate models have over-
predicted VPD trends by a factor of two, possibly influenced
by model misrepresentation of SWUS monsoon characteristics,
megadrought levels are still surpassed during the late twenty-first
century (Supplementary Fig. S8d).
Importantly, forest-decline events are particularly sensitive to
extreme conditions (Fig. 2c,d). As widespread forest decline seems
to have occurred during the 1572–1587 megadrought, we treat the
mean FDSI during the most extreme half of the years during this
period as a forest-stress benchmark signifying a level of extreme
Percentage
Percentage
¬5 ¬4 ¬3 ¬2
FDSI
¬1 0 1
2000–2012
a
b
90
70
50
30
10
90
70
50
30
10
1000 1200 1400 1600 1800
Year
2000
Figure 5 |Extreme drought stress. a, Cumulative distribution functions of
tree-ring derived FDSI during AD 1000–2007 (black) and model-projected
FDSI during AD 2000–2100 for the A2 (red), A1B (green) and B1 (blue)
emissions scenarios. Brown line: mean FDSI during the most extreme half
of the 1572–1587 megadrought. b, Fifty-year running frequency of annual
FDSI values more negative than the mean FDSI during the most negative
half of the years during the 1572–1587 megadrought. The colours in b
represent the same as in a. Shaded areas: 95% confidence ranges for
tree-ring-derived values and inner-quartile values for model ensemble
projections.
drought-stress likely to correspond with widespread forest decline
(benchmark FDSI =−1.41). Although the 1200s megadrought
was longer, the 1500s megadrought was more extreme. During
ad 1000–2007, 4.8% of the FDSI values were more negative than the
1500s benchmark, and the highest 50-year frequency of benchmark
years was 18% during 1247–1296. During the present drought
event, 4 of the 13 years (31%) qualify as benchmark years. On the
basis of ensemble mean projections, 59% of years will be benchmark
years during 2000–2100 assuming the A2 emissions scenario
(Fig. 5a), and the frequency of benchmark years is projected to reach
approximately 80% during the second half of this century (Fig. 5b).
Assuming the most optimistic emissions scenario (B1), this value
is 53%. Very extreme FDSI values of less then −3 (unprecedented
during 1000–2012) are projected to occur with a frequency of ap-
proximately 20% in the twenty-first century (A2 scenario, Fig. 5b).
Projections of increased forest drought-stress and tree mortality
are relevant throughout the SWUS, not only to the most drought-
prone sites. Recent bark-beetle- and wildfire-induced tree mortality
has occurred approximately uniformly among a wide variety of
SWUS sites ranging from very drought prone to less drought prone
than average (Fig. 6a,b; see Supplementary Methods). Mortality
has been less common within the ∼20% of forest area least
prone to drought (high cold-season precipitation, low warm-
season VPD). Observed exponential relationships between the
FDSI and forest-decline processes suggest that less drought-prone
SWUS forests may become progressively more vulnerable to forest
decline processes and mortality as the warm-season VPD increases.
Furthermore, regeneration of common conifer species in the SWUS
generally occurs in pulses linked to wet/cool conditions13, and
often requires the presence of parent tree sources32. Loss of mature
trees across increasingly large areas due to high-severity fires
and bark-beetle-induced mortality (Fig. 2c,d and Supplementary
Fig. S9b), coupled with ongoing and projected increases in drought
stress due to climate change (Fig. 4), means that species-specific
tree regeneration needs are increasingly less likely to be met after
disturbance43. This increases the risks of long-term forest structural
4NATURE CLIMATE CHANGE |ADVANCE ONLINE PUBLICATION |www.nature.com/natureclimatechange
© 2012 Macmillan Publishers Limited. All rights reserved.

NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE1693 ARTICLES
Site-specific long-term average FDSI
2.5
5.0
7.5
10.0
PercentagePercentagePercentage
12.5
a
b
c
5
10
15
20
0.0
2.5
5.0
7.5
10
¬6 ¬5 ¬4 ¬3 ¬2 ¬1 0 1 2 3 4 5 6
Figure 6 |Where have trees died? The xaxis represents a long-term
drought-stress gradient among SWUS forest grid cells. The grid cells with
the most severe long-term drought-stress are on the left side of plots.
a, The percentage of grid cells in each drought-stress class with ≥10 trees
per acre killed by bark beetles during 1997–2011. b,The percentage of grid
cells in each drought-stress class where moderate or severe wildfire
occurred during 1984–2011. Horizontal dotted black lines in aand b
indicate expected percentages if these mortality processes were spatially
uniform. c, Probability distribution function of the average FDSI during
1896–2007 among SWUS forest grid cells. The site-specific FDSI (xaxis)
estimated using equation (1). The methods are described in the
Supplementary Information. Grey and white shading is intended to assist
with interpretation.
and compositional changes and type conversions from forests to
shrublands or grasslands (for example, refs 32,44).
Conclusions
The warm-season VPD was at least as important as cold-season
precipitation in dictating SWUS forest drought-stress during
1896–2007. The warm-season VPD has been particularly high since
2000 and is the primary driver of an ongoing drought-stress event
that is more severe than any event since the late 1500s megadrought.
The present event has been associated with regional-scale declines in
canopy greenness and tree survival, due in part to large bark-beetle
outbreaks and increasingly large wildfires. On the basis of an
ensemble of climate-model projections, continued increases in the
warm-season VPD will by the mid-twenty-first century force mean
annual SWUS forest drought-stress levels to exceed the severity of
the strongest megadroughts since at least ad 1000. Importantly,
the warm-season VPD is largely driven by the maximum daily
temperature (Tmax) in the SWUS. As such, warm-season Tmax
is nearly as effective as the warm-season VPD at predicting the
FDSI (Supplementary Fig. S10). The importance of the VPD and
temperature in dictating the FDSI, combined with the relatively
high confidence in the projections of continued warming in the
SWUS (ref. 45), translates into a high confidence in projections
of intensified forest drought-stress. The strong correspondence
between forest drought-stress and tree mortality suggests that
intensified drought-stress will be accompanied by increased forest
decline. Importantly, human forest-management practices have
profoundly influenced the regional wildfire regime over the past
century46 and future practices will continue to influence the impacts
of drought on wildfire behaviour. Furthermore, there are many
complex interactions not accounted for in this study, including in-
teractions between disturbance processes29. We therefore constrain
our quantitative projections to the FDSI and do not forecast abso-
lute magnitudes of forest area impacted by bark beetles or wildfire.
The implications of this study extend beyond the well-studied
SWUS region. Given that the ongoing SWUS drought event is prob-
ably a product of both natural and anthropogenic forcing47, it serves
as a natural experiment where the recent forest response to drought
may serve as a harbinger of how drought-sensitive forests globally
may respond to warming1, with implications regarding terrestrial
carbon budgets10. Model projections48,49 and observations50 of a
poleward expansion of the subtropics indicate that forests near the
poleward edges of the subtropics may be particularly vulnerable
to enhanced drought-induced tree mortality. This study indicates
that if warming continues, increasing VPD and drought stress will
continue to cause twenty-first-century readjustments to the SWUS
forest structure, composition and distribution through amplified
disturbance processes that have become increasingly evident
regionally in recent decades (for example, Supplementary Fig. S9).
Given the reproductive and dispersal limitations of dominant native
tree species, climate-driven amplification of forest drought-stress
and associated disturbance processes can be expected to force
many landscapes in the SWUS and probably elsewhere towards
vegetation-type conversions, with species distributions quite
different from those familiar to modern civilization.
Received 21 March 2012; accepted 28 August 2012;
published online 30 September 2012
References
1. Allen, C. D. et al. A global overview of drought and heat-induced tree mortality
reveals emerging climate change risks for forests. Forest Ecol. Manage. 259,
660–684 (2010).
2. Zhao, M. & Running, S. W. Drought-induced reduction in global terrestrial
net primary production from 2000 through 2009. Science 329, 940–943 (2010).
3. Van Mantgem, P. J. et al. Widespread increase of tree mortality rates in the
western United States. Science 323, 521–524 (2009).
4. Beck, P. S. A. et al. Changes in forest productivity across Alaska consistent with
biome shift. Ecol. Lett. 14, 373–379 (2011).
5. Breshears, D. D. et al. Regional vegetation die-off in response
to global-change-type drought. Proc. Natl Acad. Sci. USA 102,
15144–15148 (2005).
6. Phillips, O. L. et al. Drought sensitivity of the Amazon rainforest. Science 323,
1344–1347 (2009).
7. Lewis, S. L., Brando, P. M., Phillips, O. L., van der Heijden, G. M. F. &
Nepstad, D. The 2010 amazon drought. Science 331, 554 (2011).
8. Peng, C. et al. A drought-induced pervasive increase in tree mortality across
Canada’s boreal forests. Nature Clim. Change 1, 467–471 (2011).
9. Bentz, B. J. et al. Climate change and bark beetles of the western United States
and Canada: Direct and indirect effects. Bioscience 60, 602–613 (2010).
10. Schwalm, C. R. et al. Reduction in carbon uptake during turn of the century
drought in western North America. Nature Geosci. 5, 551–556 (2012).
11. Williams, A. P. et al. Forest responses to increasing aridity and warmth in the
southwestern United States. Proc. Natl Acad. Sci. USA 107, 21298–21294 (2010).
12. Weiss, J. L., Castro, C. L. & Overpeck, J. T. Distinguishing pronounced
droughts in the southwestern United States: Seasonality and effects of warmer
temperatures. J. Clim. 22, 5918–5931 (2009).
13. Swetnam, T. W. & Betancourt, J. L. Mesoscale disturbance and ecological
response to decadal climatic variability in the American Southwest. J. Clim. 11,
3128–3147 (1998).
14. Diffenbaugh, N. S., Ashfaq, M. & Scherer, M. Transient regional climate
change: Analysis of the summer climate response in a high-resolution,
century-scale ensemble experiment over the continental United States.
J. Geophys. Res. 116, D24111 (2011).
15. Seager, R. et al. Model projections of an imminent transition to a more arid
climate in southwestern North America. Science 316, 1181–1184 (2007).
16. McDowell, N. G. et al. Mechanisms of plant survival and mortality during
drought: Why do some plants survive while others succumb to drought?
New Phytol. 178, 719–739 (2008).
17. Fritts, H. C. Tree Rings and Climate (Academic, 1976).
18. Cook, E. R., Woodhouse, C. A., Eakin, C. M., Meko, D. M. & Stahle, D. W.
Long-term aridity changes in the western United States. Science 306,
1015–1018 (2004).
NATURE CLIMATE CHANGE |ADVANCE ONLINE PUBLICATION |www.nature.com/natureclimatechange 5
© 2012 Macmillan Publishers Limited. All rights reserved.

ARTICLES NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE1693
19. McDowell, N. G., Allen, C. D. & Marshall, L. Growth, carbon-isotope
discrimination, and drought-associated mortality across a Pinus ponderosa
elevational transect. Glob. Change Biol. 16, 399–415 (2010).
20. St. George, S., Meko, D. M. & Cook, E. R. The seasonality of precipitation
signals embedded within the North American Drought Atlas. Holocene 20,
983–988 (2010).
21. Adams, H. D. et al. Temperature sensitivity of drought-induced tree mortality
portends increased regional die-off under global change-type drought.
Proc. Natl Acad. Sci. USA 106, 7063–7066 (2009).
22. McDowell, N. G. et al. The interdependence of mechanisms underlying
climate-driven vegetation mortality. Trends Ecol. Evol. 26, 523–532 (2011).
23. Shaw, J. D., Steed, B. E. & DeBlander, L. T. Forest inventory and analysis (FIA)
annual inventory answers the question: What is happening to pinyon-juniper
woodlands? J. Forestry 103, 280–285 (2005).
24. Huang, C., Asner, G. P., Barger, N. N., Neff, J. C. & Floyd, M. L. Regional
aboveground live carbon losses due to drought-induced tree dieback in
piñon–juniper ecosystems. Remote Sens. Environ. 114, 1471–1479 (2010).
25. Raffa, K. F. et al. Cross-scale drivers of natural disturbances prone to
anthropogenic amplification: The dynamics of bark beetle eruptions. Bioscience
58, 501–517 (2008).
26. Grissino Mayer, H. D. & Swetnam, T. W. Century scale climate forcing of fire
regimes in the American Southwest. Holocene 10, 213–220 (2000).
27. Westerling, A. L., Hidalgo, H. G., Cayan, D. R. & Swetnam, T. W. Warming
and earlier spring increase western US forest wildfire activity. Science 313,
940–943 (2006).
28. Furniss, R. L. & Carolin, V. M. Western Forest Insects (United States Department
of Agriculture (USDA) Forest Service (FS), 1977).
29. Allen, C. D. Interactions across spatial scales among forest dieback, fire,
and erosion in northern New Mexico landscapes. Ecosystems 10,
797–808 (2007).
30. Allen, C. D. & Breshears, D. D. Drought-induced shift of a forest-woodland
ecotone: Rapid landscape response to climate variation. Proc. Natl Acad. Sci.
USA 95, 14839–14842 (1998).
31. Betancourt, J. L., Pierson, E. A., Rylander, K. A., Fairchild-Parks, J. A. &
Dean, J. S. in Managing Piñon-juniper Ecosystems for Sustainability and Social
Needs (eds Aldon, A. F. & Shaw, D. W.) 42–62 (USDA Forest Service, 1993).
32. Savage, M. & Mast, J. N. How resilient are southwestern ponderosa pine forests
after crown fires? Can. J. Forest Res. 35, 967–977 (2005).
33. Swetnam, T. W. & Betancourt, J. L. Fire-southern oscillation relations in the
southwestern United States. Science 249, 1017–1020 (1990).
34. Herweijer, C., Seager, R., Cook, E. R. & Emilie-Geay, J. North American
droughts of the last millennium from a gridded network of tree-ring data.
J. Clim. 20, 1353–1376 (2007).
35. Potter, L. D. Phytosociological study of San Augustin Plains, New Mexico.
Ecol. Monogr. 27, 114–136 (1957).
36. Plummer, F. G., Rixon, T. F. & Dodwell, A. Forest Conditions in the Black Mesa
Forest Reserve, Arizona. Series H (Government Printing Office, 1904).
37. Stahle, D. W. et al. Tree-ring data document 16th century megadrought over
North America. EOS Trans. 81, 121–121 (2000).
38. Grissino-Mayer, H. D. in Tree Rings, Environment, and Humanity (eds Dean, J.
S., Meko, D. M. & Swetnam, T. W.) 191–204 (Radiocarbon, 1996).
39. Dean, J. S. The medieval warm period on the southern Colorado Plateau.
Climatic Change 26, 225–241 (1994).
40. Fall, P. L., Kelso, G. & Markgraf, V. Paleoenvironmental reconstruction
at Canyon del Muerto, Arizona, based on principal-component analysis.
J. Archaeol. Sci. 8, 297–307 (1981).
41. IPCC Special Report on Emissions Scenarios (eds Nakicenovic, N. & Swart, R.)
(Cambridge Univ. Press, 2000).
42. Williams, A. P., Michaelsen, J., Leavitt, S. W. & Still, C. J. Using tree rings to
predict the response of tree growth to climate change in the continental United
States during the twenty-first century. Earth Interact. 14, 1–20 (2010).
43. Jackson, S. T., Betancourt, J. L., Booth, R. K. & Gray, S. T. Ecology and
the ratchet of events: climate variability, niche dimensions, and species
distributions. Proc. Natl Acad. Sci. USA 106, 19685–19692 (2009).
44. Goforth, B. R. & Minnich, R. A. Densification, stand-replacement wildfire, and
extirpation of mixed conifer forest in Cuyamaca Rancho State Park, southern
California. Forest Ecol. Manage. 256, 36–45 (2008).
45. Räisänen, J. CO2-induced climate change in CMIP2 experiments:
Quantification of agreement and role of internal variability. J. Clim. 14,
2088–2104 (2001).
46. Marlon, J. R. et al. Long-term perspective on wildfires in the western USA.
Proc. Natl Acad. Sci. USA 109, E535–E543 (2012).
47. Cayan, D. R. et al. Evolution toward greater droughts in the SW United States.
Proc. Natl Acad. Sci. USA 107, 21271–21276 (2010).
48. Held, I. M. & Soden, B. J. Robust responses of the hydrological cycle to global
warming. J. Clim. 19, 5686–5699 (2006).
49. Seager, R., Naik, N. & Vecchi, G. A. Thermodynamic and dynamic mechanisms
for large-scale changes in the hydrological cycle in response to global warming.
J. Clim. 23, 4651–4668 (2010).
50. Seidel, D. J., Fu, Q., RanDel, W. J. & Reichler, T. S. J. Widening of the tropical
belt in a changing climate. Nature Geosci. 1, 21–24 (2008).
Acknowledgements
The work was supported by LANL-LDRD and DOE-BER. We acknowledge contributors
to the International Tree-Ring Databank and funding by the NSF (grant 0823090) for
tree-ring data. We thank contributors of fire-scar data to the FACS database, accessed
with assistance from E. Bigio. Unpublished fire-scar data donated by C. Aoki, P. Brown,
E. Heyerdahl, P. Iniguez, M. Kaib and R. Wu. J. Paschke provided access to USFS
FHTET data. M. Brown provided access to GIMMS AVHRR NDVI data. Dynamically
downscaled model climate data came from NARCCAP, funded by NSF, DOE, NOAA
and EPA. We appreciate constructive comments from P. Brown, K. Cavanaugh,
M. Crimmins, P. Fulé, S. Garrity, J. Grahame, D. Gutzler, J. Hicke, X. Jiang, S. Leavitt,
M. Massenkoff, A. Meddens, J. Michaelsen, C. Millar, B. Osborn, H. Powers, T. Rahn,
N. Stephenson, C. Still, C. Tague and C. Xu.
Author contributions
A.P.W., C.D.A., A.K.M., D.G., C.A.W., D.M.M., T.W.S., S.A.R., R.S., M.C. and N.G.M.
conceived and designed the experiments. A.P.W. performed the experiments. A.P.W.
and E.R.C. analysed the data. A.K.M., D.G., C.A.W., C.G., D.M.M., T.W.S., S.A.R.,
H.D.G-M., J.S.D. and E.R.C. contributed data. A.P.W., C.D.A., A.K.M., D.G., C.A.W.,
D.M.M., T.W.S., S.A.R., R.S., H.D.G-M. and N.G.M. wrote the paper.
Additional information
Supplementary information is available in the online version of the paper. Reprints and
permissions information is available online at www.nature.com/reprints.
Correspondence and requests for materials should be addressed to A.P.W.
Competing financial interests
The authors declare no competing financial interests.
6NATURE CLIMATE CHANGE |ADVANCE ONLINE PUBLICATION |www.nature.com/natureclimatechange
© 2012 Macmillan Publishers Limited. All rights reserved.