Regional vegetation die-off in response
to global-change-type drought
David D. Breshearsa,b, Neil S. Cobbc, Paul M. Richd, Kevin P. Pricee,f, Craig D. Alleng, Randy G. Baliceh, William H. Rommei,
Jude H. Kastensf,j, M. Lisa Floydk, Jayne Belnapl,m, Jesse J. Andersonc, Orrin B. Myersn, and Clifton W. Meyerd
aSchool of Natural Resources, Institute for the Study of Planet Earth, and Department of Ecology and Evolutionary Biology, University of Arizona,
Tucson, AZ 85721-0043;cMerriam Powell Center for Environmental Research and Department of Biological Sciences, Northern Arizona University,
Flagstaff, AZ 86011;dEarth and Environmental Sciences Division, andhEnvironmental Stewardship Division, University of California–Los Alamos
National Laboratory, Los Alamos, NM 87545; Departments ofeGeography andjMathematics, University of Kansas, Lawrence, KS 66045;fKansas
Applied Remote Sensing Program, 2101 Constant Avenue, Lawrence, KS 66047-3759;gFort Collins Science Center, U.S. Geological Survey,
Jemez Mountains Field Station, Los Alamos, NM 87544;iForest, Rangeland, and Watershed Stewardship, Colorado State University, Fort Collins, CO 80523;
kEnvironmental Studies Program,Prescott College, 220 Grove Avenue, Prescott, AZ 86301;lSouthwest Biological Science Center, U.S. Geological Survey,
Moab, UT 84532;mNatural Resource Ecology Laboratory, Colorado State University, Fort Collins, CO 80523-1499; andnDivision of Epidemiology and
Biostatistics, University of New Mexico, Albuquerque, NM 87131
Edited by Harold A. Mooney, Stanford University, Stanford, CA, and approved September 7, 2005 (received for review July 8, 2005)
Future drought is projected to occur under warmer temperature
change-type drought, yet quantitative assessments of the triggers
pivotal uncertainties in assessing climate-change impacts. Of par-
ticular concern is regional-scale mortality of overstory trees, which
rapidly alters ecosystem type, associated ecosystem properties,
and land surface conditions for decades. Here, we quantify region-
al-scale vegetation die-off across southwestern North American
woodlands in 2002–2003 in response to drought and associated
bark beetle infestations. At an intensively studied site within the
region, we quantified that after 15 months of depleted soil water
content, >90% of the dominant, overstory tree species (Pinus
edulis, a pin ˜on) died. The die-off was reflected in changes in a
remotely sensed index of vegetation greenness (Normalized Dif-
ference Vegetation Index), not only at the intensively studied site
but also across the region, extending over 12,000 km2or more;
Notably, the recent drought was warmer than the previous sub-
continental drought of the 1950s. The limited, available observa-
tions suggest that die-off from the recent drought was more
extensive than that from the previous drought, extending into
wetter sites within the tree species’ distribution. Our results
quantify a trigger leading to rapid, drought-induced die-off of
overstory woody plants at subcontinental scale and highlight the
potential for such die-off to be more severe and extensive for
future global-change-type drought under warmer conditions.
tree mortality ? vegetation dynamics ? climate change impacts ?
woodlands ? Pinus edulis
ing temperatures (1–3), referred to here as global-change-type
drought. Protracted, subcontinental drought in the midlatitudes
is a complex response driven in part by anomalies associated with
oscillations in sea surface temperature (2–4), which can include
oscillations over periods of decades or longer, such as those
associated with the Atlantic Multidecadal Oscillation and the
Pacific Decadal Oscillation (4), and shorter periods spanning
several years, such as those associated with the El Nin ˜o Southern
Oscillation (3). Greenhouse gas forcings are expected to alter
recent protracted drought in southwestern North America,
spanning the beginning of the 2000 millennium, exhibited anom-
alous sea surface temperature patterns consistent with projec-
tions of global change response (3). Protracted drought can
trigger large-scale landscape changes through vegetation mor-
tality from water stress (5, 6), sometimes associated with bark
lobal climate change is projected to yield increases in
frequency and intensity of drought occurring under warm-
beetle infestations (5), but the potential for regional to subcon-
tinental scale vegetation die-off under global-change-type
drought remains a pivotal uncertainty in projections of climate
change impacts (1, 7, 8). Of particular concern is regional-scale
mortality of overstory trees, which rapidly alters ecosystem type,
associated ecosystem properties, and land surface conditions for
decades. The potential for this response is highlighted by a rapid
shift of a forest ecotone caused by Pinus ponderosa mortality in
response to the 1950s drought (5). The effects of drought
accompanied by warmer temperatures resulting from green-
house forcings might be expected to produce even greater effects
on vegetation change than those of periodic, protracted drought
alone (5). Yet few, if any, studies quantify rapid, regional-scale
vegetation die-off in response to drought, key environmental
conditions triggering tree mortality, such as prior soil moisture
conditions, or how anomalously high temperatures might alter
such vegetation responses. Such relationships urgently need to
be quantified to improve climate change assessments (9).
mortality of pin ˜on pine (Pinus edulis), which is sensitive to climate
variation and dominates pin ˜on-juniper woodlands, one of the most
extensive vegetation types in the western North America (10–12).
Specifically, we evaluated the impacts of the recent drought on
regional-scale mortality in the context of the potential impacts of
global-change-type drought, and (i) demonstrated that the recent
drought is not as dry as the previous drought but is warmer in
numerous respects, thereby providing a case study for global-
change-type drought; (ii) quantified site-specific conditions in soil
water content and local vegetation response (percentage of tree
mortality, and a remotely sensed vegetation index related to
photosynthetic activity and associated greenness, Normalized Dif-
dynamics; and (iii) estimated and verified regional-scale stress and
data back to 1989 and supplemental aerial surveys and field
in Arizona, New Mexico, Colorado, and Utah that were listed
with the Western Regional Climate Center (www.wrcc.dri.edu;
Western U.S. Climate Historical Summaries, September 1,
This paper was submitted directly (Track II) to the PNAS office.
Freely available online through the PNAS open access option.
Abbreviations: NDVI, Normalized Difference Vegetation Index; GAP, Gap Analysis
bTo whom correspondence should be addressed. E-mail: email@example.com.
© 2005 by The National Academy of Sciences of the USA
October 18, 2005 ?
vol. 102 ?
2004–Decmber 14, 2004) and determined annual values for total
precipitation, mean minimum temperature, and mean maximum
temperature. We then identified 22 of those stations associated
with pin ˜on-juniper woodlands based on two criteria. First,
where P. edulis was either the dominant or codominant plant
species, as defined by the United States Geological Survey Gap
Analysis Program (GAP) distribution map of dominant vegeta-
tion (scale of 1:100,000; www.gap.uidaho.edu). The GAP map is
derived from satellite imagery combined with existing data, air
and field study, and expert knowledge. Our second criterion was
that the record of monthly precipitation and temperature for the
drought intervals of interest, defined below, was relatively
complete: ?2 months missing for any year of interest, except for
two stations (Moab and El Vado) that were missing up to 4
months of data for one of the years. We compared the recent
drought with that of the 1950s by focusing on the four driest
consecutive years for each: 2000–2003 and 1953–1956, respec-
tively. We tested for differences in total precipitation, mean
minimum temperature, and mean maximum temperature be-
tween the two drought intervals by using t tests paired by station;
we applied sequential Bonferroni adjustment to determine
tablewide significance for the three t tests (13). We categorized
anomalously dry or warm years as in the extreme 10 percentiles.
We evaluated synoptic vegetation changes by using the NDVI
(1 km2? 1 pixel), calculated by using near–IR and visible
reflectance values collected by the advanced very high-
resolution radiometer located on several National Oceanic and
Atmospheric Administration satellite platforms (14). NDVI has
been widely used to estimate landscape patterns of primary
production (14) and is correlated with foliar water content and
water potential for P. edulis needles over a wide range of
conditions, including recently deceased trees (15). Weekly values
of NDVI from 1989 to 2003 were extracted by using the
GAP-delineated pin ˜on-juniper distribution of the four states.
Before extraction, the NDVI data were corrected by using
biweekly maximum value compositing to remove most cloud
contamination and were corrected to reduce effects of atmo-
spheric haze and ozone. These data were also detrended to
account for artificial value drift by examining temporal signa-
tures from invariant targets (16). The detrending provides a
conservatively large adjustment that could mask some of the
NDVI dynamics, and hence we present NDVI dynamics as a
range bounded by nondetrended and detrended estimates. For
each year, mean regional NDVI was calculated from the cor-
rected images for late May through June (Julian weeks 22–26),
a period when understory greenness was observed to have
minimal effect, allowing overstory effects of die-off to be most
We used measurements from an intensively studied site
(17, 18), to document changes in (i) precipitation, temperature,
and soil water content before and during the recent drought, and
water content was measured by using neutron attenuation at a
20-cm depth at 11 locations spaced at ?10-m intervals; mea-
surements were calibrated for local soils and generally obtained
once or more per month. Mortality at Mesita del Buey was
estimated through field surveys in 2002 and 2003 and was
compared against baseline data. Weekly NDVI values for the
Mesita del Buey site were estimated for a 3-?-3-pixel window
(after comparing results with window sizes ranging from 1 ? 1
to 9 ? 9 pixels) to jointly mitigate spatial registration issues and
land cover heterogeneity effects. For each year mean NDVI
values for Mesita del Buey were calculated for late May through
June (weeks 22–26) for intervals representing baseline (1989–
1999) and drought after extensive die-off (2002–2003).
To confirm that regional NDVI changes within the GAP-
delineated area were associated with tree mortality, we obtained
data from aerial surveys conducted by the U.S. Department of
Agriculture Forest Service and distributed by the Forest Health
Technology Enterprise Team (19), which covered ?60% of the
pin ˜on-juniper woodland distribution in the four states during
2002 and 2003. The surveys were flown 300–400 m above
ground, and areas of noticeable stand-level mortality from
drought and associated bark beetle infestation were sketch-
mapped at the 1:100,000 scale. We also estimated P. edulis
mortality for nonseedling trees (?1-m height) at a verification
site in each of the four states, based on either documented
changes in inventories that were conducted before mortality
event (Arizona, Colorado, New Mexico) or for which recent
for all stations in the four-state region (A) and only stations in or near
pin ˜on-juniper woodlands within that region (B) are shown. (C and D) Associ-
ated maximum (C) and minimum (D) temperatures for pin ˜on-juniper wood-
lands (black line: long-term mean; red line: 10th or 90th percentile, differen-
tiating driest or hottest years) are shown. Shaded bands indicate the four
consecutive driest years of the 1950s drought (1953–1956) and the recent
drought (2000–2003). Compared with the 1950s drought, the recent drought
was wetter (P ? 0.05) but warmer for maximum (P ? 0.05) and especially
minimum temperature (P ? 0.001).
Southwestern North American climate. (A and B) Annual mean
Breshears et al.
October 18, 2005 ?
vol. 102 ?
no. 42 ?
mortality was inferred (Utah); verification sites varied from 4 to
49 in number of plot or belt transects and 0.2–4.9 hectares for
total area surveyed. The New Mexico site spanned the more
intensively studied Mesita del Buey site and included wider
The recent drought spanning southwestern North America was
anomalously dry, similar to the subcontinental drought of the
1950s, but contrasted with that drought in having anomalously
high temperatures for the entire region (Fig. 1A) and for
locations only within the distributional range of P. edulis (Fig.
1B). Within the distributional range of P. edulis, total precipi-
tation during the 4-year interval for the recent drought (2000–
2003) was slightly greater than that for the 1950s drought (Fig.
1B; P ? 0.05). Yet the recent drought was warmer than the 1950s
drought with respect to maximum annual temperature (Fig. 1C;
P ? 0.05) and particularly with respect to minimum annual
temperature (June and July; P ? 0.001). At the intensively
studied site within the region located in northern New Mexico
(Mesita del Buey), mean minimum and mean maximum tem-
peratures were warmer during than before the recent drought
(Fig. 2A), a period of reduced annual precipitation spanning
2000–2003 (Fig. 2B). The timing and amount of the reductions
in precipitation in conjunction with simultaneous warmer tem-
peratures resulted in 10 consecutive months (October 1999–
August 2000) of dry soil water conditions (?15% volumetric
water content, associated with a soil water potential of less than
?2.5 MPa; ref. 18), which was later followed by an additional 15
consecutive months of dry soil water conditions (August 2001–
October 2002; Fig. 2C). Bark beetle (Ips confusus) infestation
(red) temperature (°C) (A), precipitation (mm) (B), volumetric soil water content at 20 cm (%) (C), P. edulis mortality (D), weekly NDVI at Mesita del Buey (E), and
NDVI for late May-June for Mesita del Buey (F) and the region encompassing P. edulis (G). For A–E, circles indicate means (annual for A–C and late May-June for
E–G); horizontal lines indicate predrought and drought means, with the latter being warmer, wetter, or lower in NDVI. For E, gray lines are nondetrended
estimates; for F–G, squares are nondetrended and circles are detrended estimates.
Drought-induced mortality at Mesita del Buey, northern New Mexico. Shown for predrought and drought periods are: minimum (blue) and maximum
www.pnas.org?cgi?doi?10.1073?pnas.0505734102Breshears et al.
and change in foliar water and spectral conditions were observed
subsequently during 2002 and 2003 (15). Resulting P. edulis
mortality at Mesita del Buey exceeded 90% (Fig. 2D). The tree
mortality at this site is reflected in a ?20% decrease in NDVI,
highlighted by the reduced NDVI values for 2002 and 2003 that
persisted once the P. edulis mortality began (Fig. 2 E and F).
Premortality effects of the drought are evident in the depressed
values of site NDVI for 2000, a very dry year, and then are
partially offset by increased NDVI in 2001, because of a rela-
The site-specific changes in NDVI are roughly parallel with
regional-scale changes in NDVI (Fig. 2G), which also dropped
off substantially in 2002 and to a lesser extent in 2003. The
site-specific mortality at Mesita del Buey proceeded into 2003,
2F), with NDVI remaining depressed through 2003. Conversely,
at the regional scale where mortality was less complete, the
winter 2002–2003 pulse of soil moisture appears to have resulted
in an increased herbaceous response that partially masks the
effect of the tree die-off (Fig. 2G). Nonetheless, even using the
conservatively detrended estimate, NDVI-depressed values at
the regional scale remain substantial. Indeed, reductions in
NDVI of similar magnitude to those linked directly to tree
mortality at Mesita del Buey covered much of the P. edulis
distribution, indicating the spatial extent and variation of
drought-induced mortality (Fig. 3A). Regional aerial surveys
conducted by the U.S. Forest Service for a subset of the area
within the study region confirm that there was widespread
mortality for ?12,000 km2(Fig. 3B), as reflected in our estimates
of NDVI changes. Additional field plot inventories provide
further confirmation of widespread P. edulis mortality, with
mortality from the drought and associated infestations of the
bark beetle I. confusus at the four verification areas ranging from
40% to 80% (Fig. 3B).
Our results are notable in documenting rapid, regional-scale
mortality of a dominant tree species in response to subconti-
nental drought accompanied by anomalously high temperatures.
Although the proximal cause of mortality for most of the trees
was apparently infestation by bark beetles, such outbreaks are
content in the months preceding tree mortality was sufficiently
low to have produced high plant water stress and cessation of
transpiration and photosynthesis in P. edulis (18). Importantly,
there was high mortality of as much as 90% or more at studied
high elevation sites, such as Mesita del Buey, Mesa Verde in
Colorado, and near Flagstaff, AZ, which are near the upper limit
of P. edulis distribution and where precipitation and water
availability are generally greater than at many other locations
where this species occurs. In contrast, mortality in response to
the 1950s drought in the same landscape as Mesita del Buey in
northern New Mexico was documented predominantly on drier,
mostly lower elevation sites, based on the presence or absence of
standing or downed dead pin ˜on wood (21). Most of the patchy
mortality in the 1950s was associated with trees ?100 years old,
whereas nearly complete tree mortality across many size and age
classes was observed in response to the recent drought (ref. 22
and data for Figs. 2D and 3B). Collectively, these observations
suggest that the mortality response to the recent drought was
greater in magnitude and extent than the mortality response to
the 1950s drought. The warmer temperatures associated with the
recent drought would have increased the energy load and water
stress demands on the trees and may account for the apparently
greater resulting mortality. The effects of water and temperature
stress during the recent drought could have been further exac-
erbated by (i) anomalously high precipitation in the southwest-
ern North America from about 1978–1995 that allowed rapid
greater intraspecific competition for drought-limited water and
greater susceptibility to drought, beetle infestation, and associ-
ated pathogens (22) and?or (ii) amplifying effects of warmer
temperatures and longer growing seasons on beetle growth rates
and population dynamics (23). Nonetheless, previous studies of
drought-induced die-off have highlighted the underlying impor-
tance of water stress in triggering die-off (5, 20).
The rapid, extensive regional changes in vegetation cover
through tree die-off that we have quantified for southwestern
North America have a number of important, interrelated eco-
logical implications (12, 24), including potentially large changes
in carbon stores and dynamics, of concern for carbon-related
polices and management (9). Other interrelated implications
include large changes in near-ground solar radiation (25), runoff
NDVI for the region encompassing P. edulis distribution within Arizona, New
Mexico, Colorado, and Utah, based on deviation from 2002–2003 relative to
the predrought mean (1989–1999) during the period late-May to June. (B)
Aerial survey map of pin ˜on-juniper woodlands, delineating areas that expe-
rienced noticeable levels of tree mortality (including larger, older trees),
conducted by the U.S. Forest Service (19) in four study areas throughout the
region, corroborates the NDVI and aerial survey maps and documents stand-
Regional drought-induced vegetation changes. (A) Change map for
Breshears et al.
October 18, 2005 ?
vol. 102 ?
no. 42 ?
and erosion (5), genetic structure of the dominant tree species
on the landscape (26), and land surface microclimate feedbacks
to the atmosphere (27). Additionally, future production of pin ˜on
nuts, an important food source for several species of birds and
small mammals and for local people (28), is expected to be
greatly reduced over an extensive area. Persistence of the recent
drought could lead to further mortality of P. edulis and other
plant species within its distribution. Notably, a dominant her-
baceous species within pin ˜on-juniper woodlands, Boutelua gra-
cilis, underwent a ?50% reduction in live basal cover between
1999 and 2003 near the Mesita del Buey site. Even codominant,
woody species of Juniperus monosperma, which are much more
drought tolerant than P. edulis (18), are undergoing mortality in
response to the drought, ranging from 2% to 26% at our four
field verification sites.
The cessation of drought conditions may be insufficient for
reestablishment of P. edulis and associated plant species, as
documented for landscape response of Pinus ponderosa after the
1950s drought (5). Such rapid shifts in vegetation may represent
abrupt, rapid, and persistent shifts in not only ecotones, but also
in dominant vegetation cover and associated ecosystem process
(5, 7–8). At a minimum, the spatially extensive die-off will need
to be considered in regional environmental assessments and
management decisions over the next several decades, the short-
est interval required for a P. edulis-dominated overstory struc-
ture to reestablish. More generally, an improved predictive
capability to forecast ecological responses to climate at regional
scales is needed to effectively deal with the consequences of
large-scale, long-term climate forcings (2, 8). Our results high-
light how drought-induced die-off can span across the range of
a vegetation type and challenges the current paradigm for
climate-induced vegetation dynamics, which focuses largely on
changes at the margins of a species’ range and the ecotone
boundaries within that range (1, 5, 6). Additionally, if temper-
atures continue to warm, vegetation die-off in response to future
drought may be further amplified (5, 8, 9, 12). This recent
drought episode in southwestern North America may be a
harbinger of future global-change-type drought throughout
much of North America and elsewhere, in which increased
temperature in concert with multidecadal drought patterns
associated with oceanic sea surface oscillations can drive exten-
sive and rapid changes in vegetation and associated land surface
We thank participants of the Drought Workshop in Flagstaff in 2003
and J. T. Overpeck, J. L. Betancourt, T. W. Swetnam, K. Larsen, F. J.
Barnes, J. Salazar, J. L. Martinez, T. G. Schofield, and C. B. Zou for
comments and?or technical support. D.D.B., P.M.R., and C.W.M. were
supported by Los Alamos National Laboratory Directed Research.
Land Cover Land Use Change Grant NAG5-11337. C.D.A. was sup-
ported by the Global Change Program of the U.S. Geological Survey’s
Biological Resources Discipline. R.G.B. was supported by the U.S.
Forest Service Rocky Mountain Research Station and the Los Alamos
National Laboratory Biological Resources Management Program.
O.B.M. was supported by National Institute of Environmental Health
Sciences Grant P30ES012072. Support for regional collaboration was
also provided by the National Science Foundation through the Drought-
Induced Regional Ecosystem Response Network (National Science
Foundation Grant DEB-0443526 to N.S.C.) and the Sustainability of
Semi-Arid Hydrology and Riparian Areas Science and Technology
Center (through National Science Foundation Cooperative Agreement
1. Intergovernmental Panel on Climate Change (2001) Climate Change 2001:
Synthesis Report, A Contribution of Working Groups I, II, and III to the Third
Assessment Report of the Intergovernmental Panel on Climate Change, eds.
Watson, R. T. & Core Writing Team (Cambridge Univ. Press, New York).
2. Easterling, D. R., Meehl, G. A., Parmesan, C., Changnon, S. A., Karl, T. R. &
Mearns, L. O. (2000) Science 289, 2068–2074.
3. Hoerling, M. & Kumar, A. (2004) Science 299, 691–694.
4. McCabe, G. L., Palecki, M. A. & Betancourt, J. L. (2004) Proc. Natl. Acad. Sci.
USA 101, 4136–4141.
5. Allen, C. D. & Breshears, D. D. (1998) Proc. Natl. Acad. Sci. USA 95,
6. Fensham, R. J. & Holman, J. E. (1999) J. Appl. Ecol. 36, 1035–1050.
7. Intergovernmental Panel on Climate Change (1998) The Regional Impacts of
Climate Change: An Assessment of Vulnerability, A Special Report of Intergov-
ernmental Panel on Climate Change Working Group II, Annex C, eds. Watson,
R. T., Zinyowera, M. C. & Moss, R. H. (Cambridge Univ. Press, New York).
8. Peters, D. P. C., Pielke, R. A., Sr., Bestelmeyer, B. T., Allen, C. D., Munson-
McGee, S. & Havstad, K. M. (2004) Proc. Natl. Acad. Sci. USA 101, 15130–
9. Breshears, D. D. & Allen, C. D. (2002) Global Ecol. Biogeog. 11, 1–5.
10. Ogle, K., Whitham, T. G. & Cobb, N. S. (2000) Ecology 81, 3237–3243.
11. Trotter, T. R., Cobb, N. S. & Whitham, T. G. (2002) Proc. Natl. Acad. Sci. USA
12. Brown, J. H., Whitham, T. G., Morgan Ernest, S. K. & Gehring C. A. (2001)
Science 293, 643–650.
13. Rice, W. R. (1989) Evolution 43, 223–225.
14. Wang, J., Rich, P. M. & Price, K. P. (2003) Int. J. Remote Sens. 24, 2345–2364.
15. Stimson, H. C., Breshears, D. D., Ustin, S. L. & Kefauver, S. C. (2005) Remote
Sens. Environ. 96, 108–118.
16. Kastens, J. H., Jakubauskas, M. E. & Lerner, D. E. (2003) IEEE Trans. Geosci.
Remote Sens. 41, 2590–2594.
17. Breshears, D. D., Rich, P. M., Barnes, F. J. & Campbell, K. (1997) Ecol. Appl.
18. Breshears, D. D., Meyers, O. B., Johnson, S. R., Meyer, C. W. & Martens, S. N.
(1997) J. Ecol. 85, 289–299.
19. U.S. Forest Service (2003) Forest Insect and Disease Conditions in the South-
western Region, 2002 (USDA Forest Service, Southwestern Region, Forestry
and Forest Health, Albuquerque, NM), Publication R3-03-01.
20. Waring, G. L. & Cobb, N. S. (1992) in Plant-Insect Interactions, ed. Bernays,
E. A. (CRC, Boca Raton, FL), Vol. 4, pp. 167–226.
21. Allen, C. D. (1989) Ph.D. thesis (University of California, Berkeley).
22. Swetnam, T. W. & Betancourt, J. L. (1998) J. Climate 11, 3128–3147.
23. Logan, J. A., Regniere, J. & Powell, J. A. (2003) Front. Ecol. Environ. 1,
24. Scheffer, M., Carpenter, S., Foley, J. A., Folke, C. & Walker, B. (2001) Nature
26. Mitton, J. B. & Duran, K. L. (2004) Mol. Ecol. 13, 1259–1264.
Res. Atmos. 109, D18109.
28. Floyd, M. L., ed. (2003) Ancient Pin ˜on-Juniper Woodlands: A Natural History of
Mesa Verde Country (Univ. Press of Colorado, Boulder).
www.pnas.org?cgi?doi?10.1073?pnas.0505734102Breshears et al.