12. M. Gustafsson, L. Frommhold, Astron. Astrophys. 400,
1161 (2003).
13. H. C. Harris et al., Astrophys. J. 524, 1000 (1999).
14. S. T. Hodgkin et al., Nature 403, 57 (2000).
15. B. R. Oppenheimer et al., Astrophys. J. 550, 448 (2001).
16. L. R. Bedin et al., Astrophys. J. 624, L45 (2005).
17. B. M. S. Hansen, Phys. Rep. 399, 1 (2004).
18. I. N. Reid, J. E. Gizis, Astron. J. 116, 2929 (1998).
19. R. G. Gratton et al., Astron. Astrophys. 408, 529 (2003).
20. S. K. Leggett, F. Allard, P. H. Hauschildt, Astrophys. J.
509, 836 (1998).
21. S. Lepine, M. M. Shara, R. M. Rich, Astrophys. J. 602,
L125 (2004).
22. I. R. King, J. Anderson, A. M. Cool, G. Piotto, Astrophys. J.
492, L37 (1998).
23. I. Baraffe, G. Chabrier, F. Allard, P. H. Hauschildt, Astron.
Astrophys. 327, 1054 (1997).
24. The models in (23) have been updated (available at
http://perso.ens-lyon.fr/isabelle.baraffe).
25. G. Chabrier, I. Baraffe, Annu. Rev. Astron. Astrophys. 38,
337 (2000).
26. Supported by NASA/HST grant GO-10424 (J.A., B.M.S.H.,
I.R.K., J.S.K., R.M.R., M.M.S.), a NASA Hubble Fellowship
(J.S.K.), the U.S.-Canada Fulbright Fellowship Committee
(H.B.R.), the Natural Sciences and Engineering Research
Council of Canada (H.B.R.), and the University of British
Columbia. H.B.R. thanks I. Ozier for fruitful discussions
on CIA, as well as UCLA for support during his extended
visit during which most of this paper was written. This
research is based on NASA/ESA Hubble Space Telescope
observations obtained at the Space Telescope Science
Institute, which is operated by the Association of
Universities for Research in Astronomy Inc. under NASA
contract NAS5-26555. These observations are associated
with proposal GO-10424.
31 May 2006; accepted 18 July 2006
10.1126/science.1130691
Warming and Earlier Spring Increase
Western U.S. Forest Wildfire Activity
A. L. Westerling,
1,2
*
H. G. Hidalgo,
1
D. R. Cayan,
1,3
T. W. Swetnam
4
Western United States forest wildfire activity is widely thought to have increased in recent decades,
yet neither the extent of recent changes nor the degree to which climate may be driving regional
changes in wildfire has been systematically documented. Much of the public and scientific
discussion of changes in western United States wildfire has focused instead on the effects of 19th-
and 20th-century land-use history. We compiled a comprehensive database of large wildfires in
western United States forests since 1970 and compared it with hydroclimatic and land-surface data.
Here, we show that large wildfire activity increased suddenly and markedly in the mid-1980s, with
higher large-wildfire frequency, longer wildfire durations, and longer wildfire seasons. The greatest
increases occurred in mid-elevation, Northern Rockies forests, where land-use histories have
relatively little effect on fire risks and are strongly associated with increased spring and summer
temperatures and an earlier spring snowmelt.
W
ildfires have consumed increasing
areas of western U.S. forests in recent
years, and fire-fighting expenditures
by federal land-management agencies now
regularly exceed US$1 billion/year (1). Hun-
dreds of homes are burned annually by wild-
fires, and damages to natural resources are
sometimes extreme and irreversible. Media re-
ports of recent, very large wildfires (9100,000
ha) burning in western forests have garnered
widespread public attention, and a recurrent
perception of crisis has galvanized legislative
and administrative action (1–3).
Extensive discussions within the fire-
management and scientific communities and
the media seek to explain these phenomena, fo-
cusing on either land-use history or climate as
primary causes. If increased wildfire risks are
driven primarily by land-use history, then eco-
logical restoration and fuels management are
potential solutions. However, if increased risks
are largely due to changes in climate during
recent decades, then restoration and fuels treat-
ments may be relatively ineffective in reversing
current wildfire trends (4, 5). We investigated
34 years of western U.S. (hereafter, Bwestern[)
wildfire history together with hydroclimatic
data to determine where the largest increases
in wildfire have occurred and to evaluate how
recent climatic trends may have been important
causal factors.
Competing explanations: Climate versus
management. Land-use explanations for in-
creased western wildfire note that extensive
livestock grazing and increasingly effective fire
suppression began in the late 19th and early
20th centuries, reducing the frequency of large
surface fires (6–8). Forest regrowth after ex-
tensive logging beginning in the late 19th cen-
tury, combined with an absence of extensive
fires, promoted forest structure changes and bio-
mass accumulation, which now reduce the
effectiveness of fire suppression and increase the
size of wildfires and total area burned (3, 5, 9).
The effects of land-use history on forest struc-
ture and biomass accumulation are, however,
highly dependent upon the ‘‘natural fire re-
gime’’ for any particular forest type. For exam-
ple, the effects of fire exclusion are thought to
be profound in forests that previously sustained
frequent, low-intensity surface fires [such as
Southwestern ponderosa pine and Sierra Neva-
da mixed conifer (2, 3, 10, 11)], but of little or
no consequence in forests that previously sus-
tained only very infrequent, high-severity
crown fires (such as Northern Rockies lodge-
pole pine or spruce-fir (1, 5, 12)].
In contrast, climatic explanations posit that
increasing variability in moisture conditions
(wet/dry oscillations promoting biomass growth,
then burning), and/or a trend of increasing
drought frequency, and/or warming temperatures
have led to increased wildfire activity (13, 14).
Documentary records and proxy reconstructions
(primarily from tree rings) of fire history and
climate provide evidence that western forest
wildfire risks are strongly positively associated
with drought concurrent with the summer fire
season and (particularly in ponderosa pine–
dominant forests) positively associated to a
lesser extent with moist conditions in anteced-
ent years (13–18). Variability in western cli-
mate related to the Pacific Decadal Oscillation
and intense El Nin˜o/La Nin˜a events in recent
decades along with severe droughts in 2000 and
2002 may have promoted greater forest wildfire
risks in areas such as the Southwest, where
precipitation anomalies are significantly influ-
encedbypatternsinPacific sea surface tem-
perature (19–22). Although corresponding
decadal-scale variations and trends in climate
and wildfire have been identified in paleo
studies, there is a paucity of evidence for such
associations in the 20th century.
We describe land-use history versus climate
as competing explanations, but they may be
complementary in some ways. In some forest
types, past land uses have probably increased the
sensitivity of current forest wildfire regimes to
climatic variability through effects on the quan-
tity, arrangement, and continuity of fuels. Hence,
an increased incidence of large, high-severity
fires may be due to a combination of extreme
droughts and overabundant fuels in some forests.
Climate, however, may still be the primary
driver of forest wildfire risks on interannual to
decadal scales. On decadal scales, climatic
means and variability shape the character of the
vegetation [e.g., species populations and their
drought tolerance (23) and biomass (fuel)
continuity (24), thus also affecting fire regime
responses to shorter term climate variability].
On interannual and shorter time scales, climate
variability affects the flammability of live and
dead forest vegetation (13–19, 25).
High-quality time series are essential for
evaluating wildfire risks, but for various reasons
(26), previous works have not rigorously docu-
mented changes in large-wildfire frequency for
1
Scripps Institution of Oceanography, La Jolla, CA 92093,
USA.
2
University of California, Merced, CA 95344, USA.
3
U.S. Geological Survey, La Jolla, CA 92093, USA.
4
Labo-
ratory of Tree-Ring Research, University of Arizona, Tucson,
AZ 85721, USA.
*To whom correspondence should be addressed. E-mail:
awesterling@ucmerced.edu
RESEARCH ARTICLES
18 AUGUST 2006 VOL 313 SCIENCE www.sciencemag.org940

western forests. Likewise, detailed fire-climate
analyses for the region have not been conducted
to evaluate what hydroclimatic variations may be
associated with recent increased wildfire activity,
and the spatial variations in these patterns.
We compiled a comprehensive time series
of 1166 large (9400 ha) forest wildfires for
1970 to 2003 from federal land-management
units containing 61% of western forested areas
(and 80% above 1370 m) (26) (fig. S1). We
compared these data with corresponding hydro-
climatic and land surface variables (26–34)to
address where and why the frequency of large
forest wildfire has changed.
Increased forest wildfire activity. We
found that the incidence of large wildfires in
western forests increased in the mid-1980s
(Fig. 1) [hereafter, ‘‘wildfires’’ refers to large-
fire events (9400 ha) within forested areas only
(26)]. Subsequently, wildfire frequency was
nearly four times the average of 1970 to 1986,
and the total area burned by these fires was
more than six and a half times its previous
level. Interannual variability in wildfire fre-
quency is strongly associated with regional
spring and summer temperature (Spearman’s
correlation of 0.76, P G 0.001, n 0 34). A
second-order polynomial fit to the regional
temperature signal alone explains 66% of the
variance in the annual incidence of these fires,
with many more wildfires burning in hotter
than in cooler years.
The length of the wildfire season also
increased in the 1980s (Fig. 1). The average
season length (the time between the reported
first wildfire discovery date and the last wild-
fire control date) increased by 78 days (64%),
comparing 1970 to 1986 with 1987 to 2003.
Roughly half of that increase was due to earlier
ignitions, and half to later control (48% versus
52%, respectively). Later control dates were no
doubt partly due to later ignition dates, given
that the date of the last reported wildfire ig-
nition increased by 15 days, but a substantial
increase in the length of time the average
wildfire burned also played a role. The average
time between discovery and control for a wild-
fire increased from 7.5 days from 1970 to 1986
to 37.1 days from 1987 to 2003. The annual
length of the fire season and the average time
each fire burned were also moderately corre-
lated with the regional spring and summer tem-
perature (Spearman’s correlations of 0.61 (P G
0.001) and 0.55 ( P G 0.001), respectively.
The greatest increase in wildfire frequency
has been in the Northern Rockies, which account
for 60% of the increase in large fires. Much of
the remaining increase (18%) occurred in the
Sierra Nevada, southern Cascades, and Coast
Ranges of northern California and southern
Oregon (‘‘Northern California,’’ in fig. S2). The
Pacific Southwest; the Southern Rockies; the
Northwest; coastal, central, and southern Califor-
nia; and the Black Hills each account for 11%,
5%, 5%, G1%, and G1%, respectively. Interest-
ingly, the Northern Rockies and the Southwest
show the same trend in wildfire frequency
relative to their respective forested areas. How-
ever, the Southwest’s absolute contribution to the
western regional total is limited by its smaller
forested area relative to higher latitudes.
Increased wildfire frequency since the mid-
1980s has been concentrated between 1680 and
2590 m in elevation, with the greatest increase
centered around 2130 m. Wildfire activity at
these elevations has been episodic, coming in
pulses during warm years, with relatively little
activity in cool years, and is strongly associated
with changes in spring snowmelt timing, which
in turn is sensitive to changes in temperature.
Fire activity and the timing of the spring
snowmelt. As a proxy for the timing of the
spring snowmelt, we used Stewart and col-
leagues’ dates of the center of mass of annual
flow (CT) for snowmelt-dominated streamflow
gauge records in western North America (32–34).
The annual wildfire frequency for the region is
highly correlated (inversely) with CT at gauges
across the U.S. Pacific Northwest and interior
West, indicating a coherent regional signal of
wildfire sensitivity to snowmelt timing (Fig. 2).
Fig. 2. (A) Pearson’s rank correlation between annual western U.S. large (9400 ha) forest wildfire
frequency and streamflow center timing. x axis, longitude; y axis, latitude. (B) Average frequency of
western U.S. forest wildfire by elevation and early, mid-, and late snowmelt years from 1970 to
2002. See Fig. 1B for a definition of early, mid-, and late snowmelt years.
Fig. 1. (A) Annual fre-
quency of large (9400 ha)
western U.S. forest wild-
fires (bars) and mean
March through August
temperature for the west-
ern United States (line)
(26, 30). Spearman’s rank
correlation between the
two series is 0.76 (P G
0.001). Wilcoxon test for
change in mean large–
forest fire frequency after
1987 was significant (W 0
42; P G 0.001). (B)First
principle component of
center timing of stream-
flow in snowmelt domi-
nated streams (line).
Low (pink shading), mid-
dle (no shading), and
high (light blue shading)
tercile values indicate
early, mid-, and late tim-
ing of spring snowmelt,
respectively. (C) Annual
time between first and last large-fire ignition, and last large-fire control.
RESEARCH ARTICLES
www.sciencemag.org SCIENCE VOL 313 18 AUGUST 2006 941