Access to this full-text is provided by IOP Publishing.
Content available from Environmental Research Letters
This content is subject to copyright. Terms and conditions apply.
Environ. Res. Lett. 18 (2023) 104028 https://doi.org/10.1088/1748-9326/acf8db
OPEN ACCESS
RECEIVED
22 June 2023
REVISED
30 August 2023
ACC EPT ED FOR PUB LICATI ON
12 September 2023
PUBLISHED
26 September 2023
Original content from
this work may be used
under the terms of the
Creative Commons
Attribution 4.0 licence.
Any further distribution
of this work must
maintain attribution to
the author(s) and the title
of the work, journal
citation and DOI.
LETTER
Effects of changing climate extremes and vegetation phenology
on wildlife associated with grasslands in the southwestern United
States
Tyler G Creech1,∗, Matthew A Williamson1,2, Steven E Sesnie3, Esther S Rubin4, Daniel R Cayan5
and Erica Fleishman6,7
1Center for Large Landscape Conservation, Bozeman, MT, United States of America
2Human-Environment Systems Program, Boise State University, Boise, ID, United States of America
3Division of Biological Sciences, U.S. Fish and Wildlife Service, Albuquerque, NM, United States of America
4Arizona Game and Fish Department, Phoenix, AZ, United States of America
5Scripps Institution of Oceanography, University of California, San Diego, La Jolla, CA, United States of America
6Department of Environmental Science and Policy, University of California, Davis, CA, United States of America
7College of Earth, Ocean, and Atmospheric Sciences, Oregon State University, Corvallis, OR, United States of America
∗Author to whom any correspondence should be addressed.
E-mail: tyler@largelandscapes.org
Keywords: climate change, climate extremes, phenology, wildlife
Supplementary material for this article is available online
Abstract
Assessments of the potential responses of animal species to climate change often rely on
correlations between long-term average temperature or precipitation and species’ occurrence or
abundance. Such assessments do not account for the potential predictive capacity of either climate
extremes and variability or the indirect effects of climate as mediated by plant phenology. By
contrast, we projected responses of wildlife in desert grasslands of the southwestern United States
to future climate means, extremes, and variability and changes in the timing and magnitude of
primary productivity. We used historical climate data and remotely sensed phenology metrics to
develop predictive models of climate-phenology relations and to project phenology given
anticipated future climate. We used wildlife survey data to develop models of wildlife-climate and
wildlife-phenology relations. Then, on the basis of the modeled relations between climate and
phenology variables, and expectations of future climate change, we projected the occurrence or
density of four species of management interest associated with these grasslands: Gambel’s Quail
(Callipepla gambelii), Scaled Quail (Callipepla squamat), Gunnison’s prairie dog (Cynomys
gunnisoni), and American pronghorn (Antilocapra americana). Our results illustrated that climate
extremes and plant phenology may contribute more to projecting wildlife responses to climate
change than climate means. Monthly climate extremes and phenology variables were influential
predictors of population measures of all four species. For three species, models that included
climate extremes as predictors outperformed models that did not include extremes. The most
important predictors, and months in which the predictors were most relevant to wildlife
occurrence or density, varied among species. Our results highlighted that spatial and temporal
variability in climate, phenology, and population measures may limit the utility of climate
averages-based bioclimatic niche models for informing wildlife management actions, and may
suggest priorities for sustained data collection and continued analysis.
1. Introduction
Assessments of species’ vulnerability to climate
change (e.g. Glick et al 2011, Pacifici et al 2015,
Staudinger et al 2015, Foden et al 2019) typic-
ally correlate wildlife occurrence or abundance
with long-term mean temperature or precipitation
(Pearson and Dawson 2003, Elith and Leathwick
© 2023 The Author(s). Published by IOP Publishing Ltd
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
2009, Garcia et al 2014). Average climate is associated
with the long-term geographic distributions of many
species, but climate extremes (e.g. droughts, floods,
heat waves), which are becoming more frequent and
intense (Meehl and Tebaldi 2004, IPCC 2012), may
be equally relevant to organisms’ body condition
and population dynamics (Zimmerman et al 2009,
Germain and Lutz 2020, Rangwala et al 2021, Stewart
et al 2021). Extremes can affect wildlife physiolo-
gically or via changes in the amount and quality of
habitat or components of habitat (Parmesan et al
2000, Maxwell et al 2019, Román-Palacios and Wiens
2020, Turner et al 2020).
Climate-induced local extinctions and popula-
tion declines more often are attributable to chan-
ging interspecific interactions, particularly those
that reduce food availability, than to exceedance of
physiological tolerances (Cahill et al 2013, Ockendon
et al 2013, Gunderson et al 2017). For example, effects
of climate change on wildlife that are mediated by
plant phenology (the timing, duration, and mag-
nitude of primary productivity and other aspects
of plants’ annual life cycles), such as changes in the
quantity and quality of food and shelter, may be
particularly strong (Asch et al 2007, McKinney et al
2012). Phenology variables can often explain or pre-
dict the density or occurrence of herbivores, omni-
vores, and insectivores (Osborne and Suárez-Seoane
2007, Mueller et al 2008, Tuanmu et al 2011, Bischof
et al 2012, Butler et al 2017).
Plant phenology can be sensitive to extreme and
fluctuating climate, such as individual storms and
precipitation variability (Knapp et al 2008, Thomey
et al 2011). Identifying the climate variables to which
phenology responds is necessary to project how wild-
life may respond to the phenology-mediated effects
of climate change. However, the outputs of many
downscaled climate models do not effectively rep-
resent spatial structure or extreme temporal vari-
ation in precipitation and temperature. By contrast,
the localized constructed analogs (LOCAs) statistical
downscaling method (Pierce et al 2014) was designed
to simulate spatial structure and temporal extremes
with good fidelity to observed historical extremes.
Therefore, LOCA outputs may be reasonable projec-
tions of future extremes, and well-suited to generate
variables needed to predict phenology.
Modeling relations between climate and phen-
ology also requires spatially extensive vegetation
data with relatively high temporal resolution, which
seldom are available from field studies. Vegetation
indices derived from remotely sensed surface-
reflectance data can serve as a proxy for some types of
field measurements and for the location and quality
of habitat for multiple taxonomic groups (Marshal
et al 2006, Kostelnick et al 2007, Osborne and Suárez-
Seoane 2007, Vi˜
na et al 2008, Stoner et al 2016,
Butler et al 2017). Vegetation indices are related
mechanistically to diet quality, body mass, and breed-
ing phenology of herbivores (Pettorelli et al 2011).
In this study, we explored the relations between
vegetation phenology, climate, and wildlife popu-
lation responses in the grasslands of the south-
western United States (‘Southwest’). We focused
on four species that are associated closely with
Southwest grasslands and are considered manage-
ment priorities: Gambel’s Quail (Callipepla gambelii),
Scaled Quail (Callipepla squamat), Gunnison’s prairie
dog (Cynomys gunnisoni), and American pronghorn
(Antilocapra americana). Our primary objectives were
to infer how these species may respond directly and
indirectly to climate change and to determine how
projected responses are influenced by including cli-
mate extremes as predictors. We first developed pre-
dictive models of vegetation phenology as a function
of climate averages, extremes, and variability. We then
developed predictive models of occurrence and dens-
ity of wildlife populations as a function of climate and
phenology variables, explicitly testing whether mod-
els that included both climate averages and extremes
explained more variance in occurrence or density of
each species than models that included only averages.
We used these models to project future phenology
and population measures, and we compared projec-
tions of population measures from models with and
without climate extremes.
2. Methods
2.1. Study area
Our study area encompassed portions of Arizona,
Colorado, New Mexico, Nevada, and Utah that
overlap the Great Basin, Mojave, Sonoran, and
Chihuahuan deserts (figure 1(A)). We limited the
study area to include only land-cover types used by
one or more of our focal species, which we iden-
tified from three data sources: maps of predicted
habitat (30 m rasters) for each species from the
U.S. Geological Survey’s Gap Analysis Project (USGS
2018); maps of Gunnison’s prairie dog colonies
provided by the Arizona Game and Fish Department
(AZGFD); and global positioning system locations of
collared American pronghorn from eight telemetry
studies conducted by AZGFD. Details are in supple-
mentary material, appendix A.
2.2. Data sources
2.2.1. Climate data
We obtained gridded daily, 6 km resolution data
for the period 1950–2015 on maximum temperat-
ure, minimum temperature, precipitation, soil mois-
ture, and potential evapotranspiration (Livneh et al
2015). We obtained LOCA projections of historical
(1950–2005) and future (2006–2100) values of max-
imum and minimum temperature and precipitation
2
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
Figure 1. Study area and locations where wildlife data used in our analyses were collected. (A) Study area as defined by the
distribution of land-cover types associated with presence of the four focal species and a detectable growing season. (B) North
American Breeding Bird Survey (BBS) routes used in analyses of Gambel’s Quail and Scaled Quail. (C) Areas in which density of
Gunnison’s prairie dog burrows was sampled. (D) Game management units (GMUs) included in analyses of American pronghorn
density.
and associated Variable Infiltration Capacity hydrolo-
gical model-derived soil moisture and potential evap-
oration (Vano et al 2020) under two emissions scen-
arios (representative concentration pathways [RCPs]
4.5 and 8.5) from ten global climate models (GCMs;
supplementary material, table A2), also at daily, 6 km
resolution. We selected GCMs that best represent
the climate, including extremes, of the Southwest
(Rupp et al 2013). Because the historical precipitation
and temperature data served as the training data for
LOCA projections (Pierce et al 2014), the historical
and projected future data were fully compatible and
comparable.
2.2.2. Vegetation phenology data
We focused on seven phenology variables that are
based on two-band enhanced vegetation index (EVI)
(Jiang et al 2008) values from Moderate Resolution
Imaging Spectroradiometer or Advanced Very High
Resolution Radiometer imagery. Our phenology vari-
ables collectively summarized the magnitude, tim-
ing, and duration of primary productivity : amplitude;
peak EVI; day of peak EVI; start, end, and length of
growing season; and cumulative EVI. Definitions are
in supplementary material, appendix A. We acquired
annual, 0.05◦(∼5 km) resolution data on land-
surface phenology from 1981 to 2016 from the
Vegetation Index and Phenology Lab, University of
Arizona (vip.arizona.edu). Phenology variables were
available for up to three distinct growing seasons
per year for each grid cell, with growing seasons
identified using a phenology algorithm that applied
thresholds for minimum magnitude and duration
of EVI increase above baseline level (Didan et al
2018). If the algorithm identified multiple growing
seasons in a particular grid cell and year, we used
3
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
phenology metric data for the growing season with
the highest peak EVI; such instances were uncom-
mon, and occurred primarily for grid cells in the
hot, North American Monsoon-influenced Mojave,
Sonoran, and Chihuahuan deserts.
2.2.3. Wildlife data
We obtained information on occurrence of Gambel’s
Quail and Scaled Quail from the North American
Breeding Bird Survey (BBS; Pardieck et al 2017),
which records bird species detected at 0.8 km (0.5 mi)
intervals along a 39.4 km (24.5 mi) route. We only
considered BBS data from 1997 to 2015 because this
was the period for which BBS, phenology, and climate
data were all available. We analyzed data from routes
along which either species had been detected during
this period (figure 1(B)), which included 45 routes
and 610 route-years for Scaled Quail, and 35 routes
and 534 route-years for Gambel’s Quail. Surveys were
conducted from late April through early July, with
most in May. We condensed the data for samples
along routes into a binary variable indicating whether
each quail species was detected anywhere along the
route in each sampling year.
We obtained information on the density of
Gunnison’s prairie dogs from annual surveys of the
Aubrey Valley population (northwestern Arizona) by
AZGFD from 2007 to 2015 (figure 1(C)). Burrow
densities, which are correlated with prairie dog dens-
ities, were estimated by counting the number of active
prairie dog burrows along 250 m transects (Biggins
et al 1993). Sampling occurred from late May through
late August, and survey dates varied among years. The
data included 20 556 observations, each representing
one 250 m sampling unit in one year.
We obtained annual estimates of American
pronghorn abundance for 32 game management
units (GMUs) in Arizona from 2008 to 2015.
Abundance estimates were based on double-observer
aerial surveys in the same general areas each year
(Magnusson et al 1978, Cook and Jacobson 1979,
Graham and Bell 1989). Surveys were conducted dur-
ing 1 June–15 September, with most surveys conduc-
ted in August. We removed GMUs for which abund-
ance estimates were missing for more than half of
the survey years. Our analysis included 25 GMUs
(figure 1(D)) and abundance estimates for 200 GMU
years. We modeled the response variable as estimated
abundance per sampled area (i.e. density).
2.3. Modeling climate-phenology relations
We modeled relations between phenology variables
(response variables) and climate variables (predictor
variables) to estimate the extent to which climate
reliably predicted phenology. We used daily histor-
ical climate data (section 2.2.1) to derive variables
describing averages, extremes, and variability of tem-
perature, precipitation, soil moisture, and poten-
tial evapotranspiration at a monthly and water-year
(1 October–30 September) resolution. Our climate
extremes variables loosely were based on those
developed by the Expert Team on Climate Change
Detection and Indices (Klein Tank et al 2009).
Coauthors of our study independently selected cli-
mate variables that they believed to be associated with
each phenology variable. We retained climate vari-
ables selected by two or more coauthors as predictors
in climate-phenology models. Each phenology vari-
able ultimately was associated with 12–32 climate pre-
dictor variables (table A5). We also included as pre-
dictors three non-climate variables that may influence
vegetation phenology (or remotely sensed estimates
of phenology): land-cover type (from the Southwest
Regional Gap Analysis Project; Lowry et al 2007), soil
type, and Level II terrestrial ecoregion (Commission
for Environmental Cooperation 1997), each calcu-
lated as the modal value within each climate grid cell.
We used the random forest machine learning
algorithm (Breiman 2001), a tree-based ensemble
learning method, to model relations between phen-
ology and climate. We selected this method because
it automatically handles nonlinear relations and cap-
tures interactions between predictors (Elith et al
2008), both of which are common in complex eco-
logical data such as ours. Preliminary analyses also
suggested that random forests outperformed other
machine learning algorithms (appendix A). One dis-
advantage of machine learning models is that their
outputs may be less interpretable than those of tra-
ditional regression models. However, we focused
on maximizing predictive performance rather than
explaining mechanistic links between climate and
phenology or identifying the individual climate vari-
ables most strongly associated with phenology. We
used data from 1981 to 2008 as the training data to fit
random forest models, and withheld data from 2009
to 2015 for assessing model performance. Additional
detail on model tuning, training, and testing is in sup-
plementary material, appendix A.
2.4. Modeling climate-wildlife and
phenology-wildlife relations
2.4.1. Predictor variables
We consulted the literature and subject-matter
experts within AZGFD to identify climate and biolo-
gical and physical variables known or hypothesized
to be associated with the spatial distribution or pop-
ulation measures of each focal species (supplement-
ary material, table A3). We also developed a set of
variables to capture the potential stresses of climate
extremes on wildlife (supplementary material, table
A4), and we included the seven annual phenology
variables described above as predictor variables for
each focal species. We included time-lagged versions
of each climate and phenology variable, with lags
of 1 and 2 years for Gambel’s Quail, Scaled Quail,
and Gunnison’s prairie dog; and lags of 1, 2, and
3 years for American pronghorn, which has a longer
4
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
Figure 2. Conceptual model of our methods for projecting wildlife responses to changes in climate and phenology.
ML =machine learning.
generation time than the other species (Hoogland
1995, Cooke et al 2018, Bird et al 2020). Full methods
are in supplementary material, appendix A.
2.4.2. Bayesian wildlife models
We used hierarchical Bayesian models to assess the
relations between climate, phenology, and popula-
tion measures (figure 2). Each model contained pre-
dictors that are relatively time-invariant (e.g. eleva-
tion, topography) and predictors that reflected cli-
mate and phenology during each observation year.
We aimed to identify the predictors that explained
annual changes in species occurrence or density
rather than those that explained decadal or longer-
term trends. Such trends likely respond in part to
deterministic changes in habitat amount as a result of
land use. We made the simplifying assumption that
land use did not appreciably change habitat amount
over the period of analysis. For quail and Gunnison’s
prairie dogs, we used a varying intercept with hier-
archical priors to account for interannual variabil-
ity and potential correlations between annual estim-
ates of occurrence or density. For American prong-
horn, we modeled the relations between predictors
and the finite rate of population change (λ) expli-
citly (see supplementary material, appendix A for
all model specifications and priors) to better reflect
annual abundance estimates that are linked to the
abundances of Antilocapra spp. in the Southwest in
the previous year (Whittaker et al 2003, Woodruff
et al 2016). Details are in supplementary material,
appendix A.
We examined the effect of climate extremes
on wildlife responses by comparing the perform-
ance of two models for each species. The first
model included variables representing climate means,
annual phenological metrics, or time-invariant bio-
physical variables (means model). The second model
included all variables in the means model and
variables representing climate extremes (extremes
model). We assessed model performance with
the leave-one-out information criterion (LOOIC)
(Vehtari et al 2017). Models with lower LOOIC val-
ues are judged to have better predictive performance,
and differences in the expected log predictive density
>4 indicate strong evidence, or low uncertainty, that
the model with the lower LOOIC value strongly out-
performs the others. We considered predictors to be
strongly associated with occurrence or density if more
than 95% of the posterior predictive mass lay to one
or the other side of 0. Given the number of variables
included in each model, we used regularizing priors to
reduce the risk of overfitting. These regularizing pri-
ors ‘shrink’ regression coefficient estimates towards
0 for uncertain or uninformative parameters while
allowing informative parameters to be estimated in
the context of the other variables in the model. As
a result, our estimates of posterior predictive density
are likely conservative; however, we did not attempt to
interpret effects of individual variables (e.g. via mar-
ginal effects or odds ratios) because our main interest
was identifying predictors that should be included in
future projections rather than interpreting any single
variable.
2.5. Assessing model performance
We assessed performance of climate-phenology mod-
els with two measures of prediction accuracy: coef-
ficient of determination (R2), the proportion of
variation in the observed phenology values that is
explained by the climate predictor variables; and
mean absolute error (MAE), the absolute average
5
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
distance between the observed phenology values
and predicted phenology values. R2of models with
different phenology response variables can be dir-
ectly compared because R2is a unitless propor-
tion, but MAE cannot because phenology vari-
ables have different units and ranges of values. We
mapped prediction errors to determine whether and
how model performance varied across the study
area.
We assessed performance of wildlife models with
graphical posterior predictive checks and Freeman-
Tukey’s measure of discrepancy. Posterior predict-
ive checking is a qualitative approach for evaluating
model fit. We determined whether our models were
plausible by drawing 1000 random draws from the
posterior predictive distribution of each model and
compared the distribution of those values to those in
the original data with respect to measures of central
tendency, variation, and skewness (Gabry et al 2019).
We calculated Freeman-Tukey’s measure of discrep-
ancy according to:
Dobserved =
iyobserved
i−µi2;
Dpredicted =
iypredicted
i−µi2
.
where the observed discrepancy (Dobserved)reflects
the difference between the observed value (yobserved
i)
and the model-based prediction (µi), and the pre-
dicted discrepancy (Dpredicted) reflects the difference
between samples drawn from the posterior predictive
distribution (ypredicted
i) and the model-based predic-
tion. The posterior predictive fit is the proportion of
sampled discrepancies that exceed the observed dis-
crepancy, and values near 0.5 indicate excellent model
fits (Gelman et al 2013).
2.6. Projecting future changes in phenology
We used the machine learning models of climate-
phenology relations, trained on historical climate
data, and the LOCA projections to project future
phenology. We generated annual projections of each
phenology variable for each combination of GCM
and RCP, and summarized results for three future
periods: 2021–2050 (2030s), 2041–2070 (2050s), and
2061–2090 (2070s). We calculated the median pre-
dicted annual phenology values for each grid cell
within each future period for each GCM and RCP.
We used climate model backcasts to make retrospect-
ive median predictions of phenology for a reference
period (1981–2010), then calculated the differences
between projected and retrospective medians (deltas)
to illustrate the range of potential changes in phen-
ology. Comparing model projections to model back-
casts rather than historical observations avoids con-
flation of biases in GCM output with potential effects
of climate change (Sofaer et al 2017).
2.7. Characterizing species responses to future
climate and phenology
We used modeled past and future values of climate
and phenology variables to project the responses of
wildlife species to future conditions. For each spe-
cies and grid cell (or GMU for American prong-
horn), we generated yearly projections of each com-
bination of GCM and RCP in four steps. First, we
multiplied projected values of influential climate and
phenology variables by the coefficients of these vari-
ables from the Bayesian wildlife models. Second, we
summed the resulting values across variables. Third,
we back-transformed to the original units of ana-
lysis (occurrence or density). Fourth, we calculated
the proportional change in the medians of the back-
transformed values (across years, RCPs, and GCMs)
between the reference period and each future period
(2030s, 2050s, and 2070s) as an index of species’
responses to changes in climate and phenology. These
indices assume that values of all biological and phys-
ical predictors, and any climate or phenology predict-
ors that we did not identify as influential, remain at
their means in the data we used to fit the models.
Therefore, the indices are not predictions of the abso-
lute change in occurrence or density.
We projected each species’ responses to future cli-
mate and phenology with the means model and the
extremes model. We focused on the projections from
the model with the lower LOOIC when interpreting
projected responses.
3. Results
3.1. Relations between climate and phenology
Predictive performance (R2) of climate-phenology
models ranged from 0.25 for amplitude to 0.65
for cumulative EVI. In ecological research, these
effects often are considered moderate to large (Cohen
1988, Møller and Jennions 2002). Predictive perform-
ance of some phenology metrics varied consider-
ably across the study area (supplementary material,
figures B1–B7), which may reflect differences in land
cover and the seasonality of precipitation and other
climate variables among ecosystems. Prediction error
for most of the phenology variables tended to be
highest along the margins of the study area, where
grassland cover types transitioned to non-grasslands.
The smallest prediction error for most phenology
variables was in grasslands in eastern New Mexico and
eastern Colorado.
Models of cumulative EVI (R2=0.653,
MAE =5.34), amplitude (R2=0.247, MAE =0.02),
and peak EVI (R2=0.453, MAE =0.03) had rel-
atively uniform and low prediction error across the
Southwest, although these models showed scattered
small pockets of high-error grid cells throughout the
study area (figures B1, B2 and B6). Models for the
remaining phenology variables had more discernable
spatial patterns of prediction error. Prediction error
6
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
for day of peak EVI (R2=0.565, MAE =25.9 d)
was relatively high in Arizona and southern Nevada
and low in the northernmost portion of the study
area (figure B3). Prediction error for length of grow-
ing season (R2=0.510, MAE =37.5 d) and end of
growing season (R2=0.443, MAE =39.1 d) was rel-
atively high in northwestern Arizona and southern
Nevada and low in the eastern and northern portions
of the study area (figures B4 and B5). Prediction
error for start of season (R2=0.416, MAE =28.2 d)
was high in southwestern Arizona but low elsewhere
(figure B7).
3.2. Relations between climate and wildlife and
phenology and wildlife
Graphical posterior predictive checks of climate-
wildlife and phenology-wildlife models (supplement-
ary material, figures A1–A8) indicated that all mod-
els plausibly fit the data. The Freeman-Tukey dis-
crepancy values for the means models were 0.498,
0.511, 0.560, and 0.761 for Gambel’s Quail, Scaled
Quail, Gunnison’s prairie dog, and American prong-
horn, respectively. Similarly, discrepancy values for
the extremes models were 0.522, 0.513, 0.559, and
0.760 for Gambel’s Quail, Scaled Quail, Gunnison’s
prairie dog, and American pronghorn, respectively.
The extremes model clearly outperformed the
means models for Scaled Quail, American prong-
horn, and Gunnison’s prairie dog (table 1). The
means model marginally (difference in expected log
predictive density <4) outperformed the extremes
model for Gambel’s Quail.
Climate and phenology variables were influential
predictors of population measures of all focal species
(table 1). The majority of the influential climate pre-
dictors were monthly extremes (minimum or max-
imum climate variables). For instance, monthly max-
ima of daily precipitation and potential evapotran-
spiration, and monthly minima of daily minimum
temperature, were strongly related to occurrence or
density of Scaled Quail, Gunnison’s prairie dog, and
American pronghorn. Consistent with differences in
life history, the most relevant months varied among
species.
Predictors related to the amount of greenness,
particularly cumulative EVI and amplitude, were
influential for all species. Predictors related to the
timing of greenness, especially start of growing sea-
son, were also influential in models of all species.
These phenology variables may indicate overall pro-
ductivity of grasses and how quickly food or shelter
becomes available after winter.
Of the 65 climate and phenology variables iden-
tified as influential in the best models across all focal
species, 38 lagged by 1–3 years (table A6). Although
relations between each species and variable are diffi-
cult to interpret, population measures generally were
most strongly associated with conditions in years
prior to the year of measurement.
Models of the density of American pronghorn,
which were evaluated at the extent of GMUs that
encompassed many grid cells, suggested that aver-
ages of climate or phenology variables over large areas
may not reflect the conditions experienced by wildlife.
Most of the influential predictors of American prong-
horn were minima, maxima, or standard deviations
rather than means or medians (table A6).
3.3. Projected changes in phenology
Projected changes of the seven phenology vari-
ables differed in magnitude and direction across the
Southwest (figure 3). Patterns were qualitatively sim-
ilar among RCPs and among future periods, but we
projected larger changes under the higher emissions
scenario and for later periods (figure 3; supplement-
ary material, figures C1–C5).
Our results suggested that the magnitude of
greenness generally will decrease or change little
over time (figure 3; supplementary material, figures
C1–C5). Amplitude was projected to decrease across
the northern half of the study area, where precipita-
tion is dominated by winter snow, and to remain rel-
atively constant in the southern half, where precipita-
tion is dominated by the North American Monsoon.
Peak EVI was projected to decrease slightly across
most of the Southwest, especially southwest Arizona
and near the borders of northern Utah and northern
Colorado.
Projections related to the timing of greenness
were less consistent (figure 3; supplementary mater-
ial, figures C1–C5). The day of peak EVI was pro-
jected to occur earlier in the southern half of the
study area and remain constant or occur slightly later
in the northern half, where green-up still may be
prolonged despite increases in winter temperatures.
The start of the growing season was projected to
occur later in the Four Corners region, earlier in the
northern portion of the study area, and at approx-
imately the same time in the eastern portion of the
study area. The end of the growing season was pro-
jected to occur earlier across most of the Southwest,
possibly indicating a trend toward development of
drought conditions earlier in the year. The length of
the growing season may decrease in the Four Corners
region.
Cumulative EVI, which is influenced by both the
magnitude and timing of greenness, was projected to
increase across much of Nevada and northern Utah,
decrease slightly across much of Arizona, and remain
relatively constant elsewhere.
3.4. Projected wildlife responses to future climate
and phenology
Projected responses of each focal species were qualit-
atively similar for different future periods and emis-
sions scenarios. Larger responses were projected for
later periods and the higher emissions scenario (sup-
plementary material, figures E1–E4).
7
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
Table 1. Climate and phenology variables strongly associated (more than 95% of the posterior predictive mass greater or less than 0) with wildlife population measures in models including and excluding climate extremes for each
focal species. Tmin, minimum temperature; P, precipitation; PET, potential evapotranspiration; EVI, enhanced vegetation index; N/A, not applicable.
Species Modela∆LOOICb∆ELPDc
Important predictor variables
Phenology Climate averages Climate extremes
Gambel’s Quail
Extremes 6.3 3.2 Cumulative EVI
Day of peak EVI None
Monthly maximum P (March, August,
October, December)
Monthly maximum PET (September)
Monthly maximum Tmin (February)
Means 0 0 Cumulative EVI
Day of peak EVI Dec-Jan P N/A
Scaled Quail
Extremes 0 0
Amplitude
Cumulative EVI
Peak EVI
Start of season
None
Monthly maximum P (February, July,
August, September)
Monthly maximum PET (June, August,
September)
Monthly maximum Tmax (May, June)
Monthly minimum Tmin (January,
November)
Means 86.4 43.2
Amplitude
Cumulative EVI
End of season
Peak EVI
Start of season
Water year P N/A
Gunnison’s prairie dog
Extremes 0 0
Amplitude
Cumulative EVI
Day of peak EVI
End of season
Peak EVI
Start of season
None
Monthly maximum P (January, March,
May, July, September)
Monthly maximum PET (June, July)
Monthly maximum Tmax (June, August)
Monthly minimum Tmin (January,
February, November, December)
Means 467.3 233.6
Amplitude
Cumulative EVI
Day of peak EVI
End of season
Length of season
Peak EVI
April–May P
June–September P
October–March P
N/A
(Continued.)
8
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
Table 1. (Continued.)
American pronghorn
Extremes 0 0
Amplitude
Length of season
Start of season
April–August P
Apr–Jun P
Monthly maximum P (January, February,
May, August, October, November)
Monthly maximum PET (May, June, July,
August)
Monthly minimum Tmin (December)
Means 4214.9 2107.5
Amplitude
Cumulative EVI
Day of peak EVI
End of season
Length of season
Start of season
April–August P
April–June P
October–April P
Water year P
N/A
aThe means models included predictors representing climate means, annual phenology metrics, or biophysical variables that did not change through time. The extremes models included all of the predictors in the means model plus
predictors representing climate extremes.
bDifference in leave-one-out information criterion; lower LOOIC values indicate models with better predictive performance.
cDifference in expected log predictive density; ∆ELPD >4 indicates strong evidence that the model with the lower LOOIC value strongly outperforms the other model.
9
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
Figure 3. Projected change (delta) in values of annual phenology variables between a historical reference period (1981–2010) and
a future period (2041–2070) under the representative concentration pathway (RCP) 8.5 emissions scenario. Delta values are
median projected changes across 10 global climate models. Units are days for day of peak enhanced vegetation index (EVI) and
start, end, and length of growing seasons; and unitless EVI values for amplitude, cumulative EVI, and peak EVI. Results for other
future periods and RCP 4.5 are included in the supplementary material (figures C1–C5).
The best model of Gambel’s Quail occurrence (the
means model) predicted negative responses to future
climate and phenology in the species’ southwestern
range and neutral or slightly positive responses else-
where. However, the extremes model, which per-
formed almost as well as the means model, predicted
10
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
Figure 4. Projected responses of Gambel’s Quail, Scaled Quail, Gunnison’s prairie dog, and American pronghorn to changes in
values of climate and phenology variables. Responses are measured as the proportional changes (deltas) in wildlife population
measures (occurrence of Gambel’s and Scaled Quail, density of Gunnison’s prairie dog burrows, and density of American
pronghorn) from the reference period (1981–2010) to the 2050s under representative concentration pathway (RCP) 8.5. Left
column: projections from models including climate extremes. Right column: projections from models excluding climate
extremes. Results for the first three species are masked to grid cells with land-cover types most commonly used by the species (see
appendix A). Results for American pronghorn are masked to game management units (GMUs). Projected responses for other
future periods and RCP 4.5 are included in supplementary material, figures E1–E4.
positive changes across nearly all of the species’ range.
Responses of Gambel’s Quail were projected by both
models to be most favorable in the species’ north-
ern and southeastern range, and least favorable in the
southwestern range (figure 4; supplementary mater-
ial, figure E1).
The best model of Scaled Quail occurrence (the
extremes model) projected almost uniformly positive
responses to future climate and phenology across the
land-cover types with which the species commonly
is associated. Responses of Scaled Quail were most
favorable in the northern range, least favorable in the
southern range, and moderately favorable in the cen-
ter (figure 4; supplementary material, figure E2).
The best model of Gunnison’s prairie dog burrow
density (the extremes model) projected larger areas of
negative response than of positive response to future
climate and phenology within the land-cover types
with which the species is associated (figure 4; sup-
plementary material, figure E3). Areas with positive
11
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
responses were mostly in the species’ southern range,
and most areas with negative responses were in the
center of its range.
The best model of American pronghorn dens-
ity (the extremes model) projected a positive effect
of changes in climate and phenology in GMUs in
southeastern Arizona, a negative effect in northwest-
ern Arizona, and both positive and negative effects
in central Arizona (figure 4; supplementary material,
figure E4).
For Gambel’s Quail and Gunnison’s prairie dog,
the means and extremes models projected different
directions (positive or negative) of response across
large portions of the study area (figure 4). The
extremes model predicted generally more favorable
responses across the species’ range than did the means
model for all species except Gunnison’s prairie dog.
4. Discussion
Species’ geographic distributions and abundances
are determined by myriad biotic and abiotic factors
(e.g. Hutchinson’s 1957 n-dimensional hypervolume
concept). Addressing wildlife responses to climate
extremes and vegetation phenology has been chal-
lenging because climate models have not accurately
captured extremes and phenology projections have
not been available. By contrast, our work not only
examines relations between climate and phenology
but capitalizes on improved climate projections and
remotely sensed phenology data to infer wildlife
responses to climate change.
We found strong evidence that climate extremes
affect wildlife species associated with Southwest
grasslands. For three of our four focal species, the
wildlife model including climate extremes variables
explained considerably more observed variation in
population responses than the model without climate
extremes, and the two models were comparable for
the fourth species. Monthly extremes of precipita-
tion were associated with population responses of all
species, although the most relevant months varied
among species. The infrequent, unpredictable pre-
cipitation in Southwest deserts can trigger pulses of
primary productivity (Noy-Meir 1973), and a single
major precipitation event may influence vegetation
growth more than a series of smaller events (Reynolds
et al 2004). Therefore, extreme precipitation dur-
ing particular life stages (e.g. during lactation for
American pronghorn) can have a strong influence on
population dynamics (Gedir et al 2015).
Minimum daily temperatures during late autumn
or early winter, which were associated with popula-
tion measures of three species, could be related to
metabolic requirements, other physiological stresses,
or predation (Kendeigh 1945, Briga and Verhulst
2015, Brodin et al 2017). Population measures of all
species were related to extremes of potential evapo-
transpiration during summer. Potential evapotran-
spiration reflects not only temperature but solar radi-
ation, wind speed, soil moisture, and vegetation cover
(Allen et al 1998).
Our results also demonstrate that phenology-
mediated effects of climate on wildlife may be sub-
stantial. Measures of the timing of greenness were
influential predictors of population responses of all
species, and sometimes more influential than the
magnitude of greenness. Again, however, remotely
sensed phenology variables do not necessarily identify
the habitat characteristics represented by these
variables.
The modest predictive performance of our
climate-phenology models likely reflected several
limitations. For example, reflectance from bare
ground in sparsely vegetated grasslands can make it
difficult to detect a phenology signal via remote sens-
ing (Huete 1988). The spatial resolution of our phen-
ology data (6 km) may have masked finer-resolution
variation in phenology. Multiple growing seasons per
year, or growth pulses in response to precipitation
events, may not be well captured by our phenology
variables. Moreover, regional variation in land-cover
types, phenology, and climate-phenology relations is
difficult to represent in a single model.
The positive predicted responses of our focal spe-
cies to future climate and phenology across substan-
tial portions of their ranges (figure 4) were unex-
pected given that these species occupy some of the
hottest and driest environments in the United States
(Gonzales et al 2018). Additionally, personal obser-
vations of AZGFD biologists suggest that Gambel’s
Quail and Scaled Quail do not respond positively
to unusually hot or dry conditions. However, pre-
vious assessments also suggested neutral or positive
responses to climate change by American pronghorn
and quail in the Southwest (Gedir et al 2015, Tanner
et al 2017).
We recognize that our projections do not account
for numerous other factors that affect the status of
wildlife populations, such as disturbance processes
(Singelton et al 2019), including the spread of non-
native invasive species (Steidl et al 2013), and urban
and exurban development (Theobald et al 2013).
Climate change may alter biotic interactions between
animal species and their competitors, pathogens, or
predators (Blois et al 2013, Bastille-Rousseau et al
2018). Furthermore, climate change eventually may
lead to transitions among grasslands, woodlands, and
shrublands (Notaro et al 2012). We believe that man-
agement actions that address well-known threats to
native species, including changes in land use, habitat
fragmentation, fire regimes, and water management,
may continue to provide substantive benefits despite
rapid changes in climate.
12
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
5. Conclusions
Bioclimatic niche models based on long-term climate
averages are now commonly used to predict wild-
life responses to climate change. Our study indicates
that this approach can be improved by incorporating
climate extremes and phenology-mediated effects of
climate, and that doing so can lead to substantially
different predictions. Our findings further suggest
that spatial variation in response to climate change
across a species’ regional range may be considerable,
especially among areas with different seasonality and
dominant forms of precipitation (e.g. snowfall versus
summer rainfall or a near-equal distribution). We
encourage consideration of these issues when anticip-
ating the effects of climate change on wildlife species
and populations.
Future research could implement the approach
demonstrated here in other ecosystems, for other
wildlife taxa, and for smaller spatial extents with
less heterogeneous climate-phenology relations that
could allow for more accurate model predictions.
Research with experimental methods better suited
to clarifying the mechanisms underlying observed
associations among climate extremes, vegetation
phenology, and wildlife responses is also needed.
Development of annual phenology metrics that bet-
ter capture multiple growing seasons and other
complex phenological patterns could lead to better
projection of wildlife responses to climate change
in the Southwest and other regions with similar
environments.
Data availability statement
The data cannot be made publicly available upon
publication due to legal restrictions preventing unres-
tricted public distribution. The data that support the
findings of this study are available upon reasonable
request from the authors.
Acknowledgments
Arizona Game and Fish Department (AZGFD)
provided Gunnison’s prairie dog and American
pronghorn survey data for this study. We thank
AZGFD employees K Bristow, J Cordova, H Hicks,
L Harding, A Munig, and A Smith for assistance with
data access and developing hypotheses about wildlife-
climate and wildlife-phenology relations. Funding for
this research was provided by the Southwest Climate
Adaptation Science Center under Grant Agreement
G17AP00098. The findings and conclusions in this
manuscript are those of the authors and do not
necessarily represent views of U.S. Fish and Wildlife
Service.
ORCID iDs
Tyler G Creech https://orcid.org/0000-0001-8049-
6680
Matthew A Williamson https://orcid.org/0000-
0002-2550-5828
Erica Fleishman https://orcid.org/0000-0003-
4435-3134
References
Allen R G, Pereira L S, Raes D and Smith M 1998 Crop
Evapotranspiration—Guidelines for Computing Crop Water
Requirements (Food and Agriculture Organization)
Asch M V, Tienderen P H V, Holleman L J M and Visser M E 2007
Predicting adaptation of phenology in response to climate
change, an insect herbivore example Glob. Change Biol.
13 1596–604
Bastille-Rousseau G, Schaefer J A, Peers M J L, Hance Ellington E,
Mumma M A, Rayl N D, Mahoney S P and Murray D L 2018
Climate change can alter predator–prey dynamics and
population viability of prey Oecologia 186 141–50
Biggins D E, Miller B J, Hanebury L R, Oakleaf B, Farmer A H,
Crete R and Dood A 1993 A technique for evaluating
black-footed ferret habitat Proc. Symp. on the Management of
Prairie Dog Complexes for the Introduction of the
Black-Footed Ferret pp 73–88
Bird J P, Martin R, Akçakaya H R, Gilroy J, Burfield I J,
Garnett S T, Symes A, Taylor J, Sekercioglu C H and
Butchart S H M 2020 Generation lengths of the world’s birds
and their implications for extinction risk Conserv. Biol.
34 1252–61
Bischof R, Loe L E, Meisingset E L, Zimmermann B, Van
Moorter B and Mysterud A 2012 A migratory northern
ungulate in the pursuit of spring: jumping or surfing the
green wave? Am. Nat. 180 407–24
Blois J L, Zarnetske P L, Fitzpatrick M C and Finnegan S 2013
Climate change and the past, present, and future of biotic
interactions Science 341 499–504
Breiman L 2001 Random forests Mach. Learn. 45 5–32
Briga M and Verhulst S 2015 Large diurnal temperature range
increases bird sensitivity to climate change Sci. Rep. 516600
Brodin A, Nilsson J-Å and Nord A 2017 Adaptive temperature
regulation in the little bird in winter: predictions from a
stochastic dynamic programming model Oecologia
185 43–54
Butler M, Sesnie S E, Timmer J M and Harris G 2017 Integrating
land surface phenology with cluster density and size
improves spatially explicit models of animal density Remote
Sens. Environ. 199 51–62
Cahill A E et al 2013 How does climate change cause extinction?
Proc. R. Soc. B280 20121890
Cohen J 1988 Statistical Power Analysis for the Behavioral Sciences
2nd edn (Lawrence Erlbaum Associates)
Commission for Environmental Cooperation 1997 Ecological
Regions of North America: Toward a Common Perspective
(Commission for Environmental Cooperation)
Cook R D and Jacobson J O 1979 A design for estimating visibility
bias in aerial surveys Biometrics 35 735–42
Cooke R S C, Gilbert T C, Riordan P and Mallon D 2018
Improving generation length estimates for the IUCN Red
List PLoS One 13 e0191770
Didan K et al 2018 Multi-sensor vegetation index and phenology
earth science data records: algorithm theoretical basis
document and user guide, version 4.1 (Vegetation Index and
Phenology Lab, University of Arizona)
Elith J and Leathwick J R 2009 Species distribution models:
ecological explanation and prediction across space and time
Annu. Rev. Ecol. Evol. Syst. 40 677–97
13
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
Elith J, Leathwick J R and Hastie T 2008 A working guide
to boosted regression trees J. Animal Ecol.
77 802–13
Foden W B et al 2019 Climate change vulnerability assessment
of species Wiley Interdiscip. Rev. Clim. Change
10 e551
Gabry J, Simpson D, Vehtari A, Betancourt M and Gelman A 2019
Visualization in Bayesian workflow J. R. Stat. Soc. A
182 389–402
Garcia R A, Cabeza M, Rahbek C and Ara´
ujo M B 2014 Multiple
dimensions of climate change and their implications for
biodiversity Science 344 1247579
Gedir J V, Cain J W, Harris G and Turnbull T T 2015 Effects of
climate change on long-term population growth of
pronghorn in an arid environment Ecosphere 6189
Gelman A, Carlin J B, Sterns H S, Dunson D B, Vehtari A and
Rubin D B 2013 Bayesian Data Analysis 3rd edn (CRC Press)
(https://doi.org/10.1201/b16018)
Germain S J and Lutz J A 2020 Climate extremes may be more
important than climate means when predicting species
range shifts Clim. Change 163 579–98
Glick P, Stein B A and Edelson N A 2011 Scanning the
Conservation Horizon: A Guide to Climate Change
Vulnerability Assessment (National Wildlife Federation)
Gonzales P et al (eds) (Southwest) 2018 Impacts, Risks, and
Adaptation in the United States: Fourth National Climate
Assessment, Volume II ed D R Reidmiller, C W Avery,
D R Easterling, K E Kunkel, K L M Lewis, T K Maycock and
B C Stewart (U.S. Global Change Research Program)
pp 1101–84
Graham A and Bell R 1989 Investigating observer bias in aerial
survey by simultaneous double-counts J. Wildl. Manage.
53 1009–16
Gunderson A R, Tsukimura B and Stillman J H 2017 Indirect
effects of global change: from physiological and behavioral
mechanisms to ecological consequences Integr. Comp. Biol.
57 48–54
Hoogland J L 1995 The Black-tailed Prairie Dog: Social Life of a
Burrowing Mammal (University of Chicago Press)
Huete A R 1988 A soil-adjusted vegetation index (SAVI) Remote
Sens. Environ. 25 295–309
Hutchinson G 1957 Concluding remarks Cold Spring Harb. Symp.
Quant. Biol. 22 415–27
Intergovernmental Panel on Climate Change (IPCC) 2012
Managing the Risks of Extreme Events and Disasters to
Advance Climate Change Adaptation: Special Report of the
Intergovernmental Panel on Climate Change ed C B Field,
V Barros, T F Stocker and Q Dahe (Cambridge University
Press) (https://doi.org/10.1017/CBO9781139177245)
Jiang Z, Huete A R, Didan K and Miura T 2008 Development of a
two-band enhanced vegetation index without a blue band
Remote Sens. Environ. 112 3833–45
Kendeigh S C 1945 Resistance to hunger in birds J. Wildl. Manage.
9217–26
Klein Tank A M G, Zwiers F W and Zhang X 2009 Guidelines on
Analysis of Extremes in a Changing Climate in Support of
Informed Decisions for Adaptation (World Meteorological
Organization) Report No. 1500
Knapp A K et al 2008 Consequences of more extreme
precipitation regimes for terrestrial ecosystems BioScience
58 811–21
Kostelnick J C, Peterson D L, Egbert S L, McNyset K M and
Cully J F 2007 Ecological niche modeling of black-tailed
prairie dog habitats in Kansas Trans. Kans. Acad. Sci.
110 187–200
Livneh B, Bohn T J, Pierce D W and Munoz-Arriola F 2015 A
spatially comprehensive, hydrometeorological data set for
Mexico, the US, and southern Canada 1950–2013 Sci. Data
2150042
Lowry J H Jr et al 2007 Land cover classification and mapping
Southwest Regional Gap Analysis Final Report ed
J S Prior-Magee et al (U.S. Geological Survey Gap Analysis
Program) pp 14–38
Magnusson W E, Caughley G J and Grigg G C 1978 A
double-survey estimate of population size from incomplete
counts J. Wildl. Manage. 42 174–6
Marshal J P, Bleich V C, Krausman P R, Reed M L and
Andrew N G 2006 Factors affecting habitat use and
distribution of desert mule deer in an arid environment
Wildl. Soc. Bull. 34 609–19
Maxwell S L, Butt N, Maron M, McAlpine C A, Chapman S,
Ullmann A, Segan D B and Watson J E M 2019 Conservation
implications of ecological responses to extreme weather and
climate events Divers. Distrib. 25 613–25
McKinney A M, CaraDonna P J, Inouye D W, Barr B,
Dertelsen C D and Waser N M 2012 Asynchronous changes
in phenology of migrating broad-tailed hummingbirds and
their early-season nectar resources Ecology 94 1987–93
Meehl G A and Tebaldi C 2004 More intense, more frequent, and
longer lasting heat waves in the 21st century Science
305 994–7
Møller A P and Jennions M D 2002 How much variance can be
explained by ecologists and evolutionary biologists?
Oecologia 132 492–500
Mueller T, Olson K A, Fuller T K, Schaller G B, Murray M G and
Leimgruber P 2008 In search of forage: predicting dynamic
habitats of Mongolian gazelles using satellite-based
estimates of vegetation productivity J. Appl. Ecol. 45 649–58
Notaro M, Mauss A and Williams J W 2012 Projected vegetation
changes for the American Southwest: combined dynamic
modeling and bioclimatic-envelope approach Ecol. Appl.
22 1365–88
Noy-Meir I 1973 Desert ecosystems: environment and producers
Annu. Rev. Ecol. Syst. 425–51
Ockendon N et al 2013 Mechanisms underpinning climatic
impacts on natural populations: altered species interactions
are more important than direct effects Glob. Change Biol.
20 2221–9
Osborne P E and Suárez-Seoane S 2007 Identifying core areas in a
species’ range using temporal suitability analysis: an
example using little bustards Tetrax tetrax L. in Spain
Biodivers. Conserv. 16 3505–18
Pacifici M et al 2015 Assessing species vulnerability to climate
change Nate Clime Change 5215–25
Pardieck K L, Ziolkowski D J Jr, Lutmerding M, Campbell K and
Hudson M-A R 2017 North American breeding bird survey
dataset 1966–2016, version 2016.0 (U.S. Geological Survey,
Patuxent Wildlife Research Center) (available at: www.pwrc.
usgs.gov/BBS/RawData/)
Parmesan C, Root T L and Willig M R 2000 Impacts of extreme
weather and climate on terrestrial biota Bull. Am. Meteorol.
Soc. 81 443–50
Pearson R G and Dawson T P 2003 Predicting the impacts of
climate change on the distribution of species: are bioclimate
envelope models useful? Glob. Ecol. Biogeogr. 12 361–71
Pettorelli N, Ryan S, Mueller T, Bunnefeld N, Jedrzejewska B,
Lima M and Kausrud K 2011 The normalized difference
vegetation index (NDVI): unforeseen successes in animal
ecology Clim. Res. 46 15–27
Pierce D W, Cayan D R and Thrasher B L 2014 Statistical
downscaling using localized constructed analogs (LOCA) J.
Hydrometeorol. 15 2558–85
Rangwala I, Moss W, Wolken J, Rondeau R, Newlon K, Guinotte J
and Travis W R 2021 Uncertainty, complexity and
constraints: how do we robustly assess biological responses
under a rapidly changing climate? Climate 9177
Reynolds J F, Kemp P R, Ogle K and Fernández R J 2004
Modifying the ‘pulse–reserve’ paradigm for deserts of North
America: precipitation pulses, soil water, and plant
responses Oecologia 141 194–210
Román-Palacios C and Wiens J J 2020 Recent responses to climate
change reveal the drivers of species extinction and survival
Proc. Natl Acad. Sci. USA 117 4211–7
Rupp D E, Abatzoglou J T, Hegewisch K C and Mote P M 2013
Evaluation of CMIP5 20th century climate simulations for
the Pacific Northwest USA J. Geophys. Res. 118 884–10,906
14
Environ. Res. Lett. 18 (2023) 104028 T G Creech et al
Singleton M P, Thode A E, Sánchez Meador A J and Iniguez J M
2019 Increasing trends in high-severity fire in the
southwestern USA from 1984 to 2015 For. Ecol. Manage.
433 709–19
Sofaer H R, Barsugli J J and Jarnevich C S 2017 Designing
ecological climate change impact assessments to reflect key
climatic drivers Glob. Change Biol. 23 2537–53
Staudinger M D, Morelli T L and Bryan A M 2015 Integrating
Climate Change into Northeast and Midwest State Wildlife
Action Plans (Northeast Climate Science Center) (available
at: necsc.umass.edu)
Steidl R J, Litt A R and Matter W J 2013 Effects of plant invasions
on wildlife in desert grasslands Wildl. Soc. Bull. 37 527–36
Stewart S B et al 2021 Climate extreme variables generated using
monthly time-series data improve predicted distributions of
plant species Ecography 44 626–39
Stoner D C, Sexton J O, Nagol J, Bernales H H and Edwards T C Jr
2016 Ungulate reproductive parameters track satellite
observations of plant phenology across latitude and
climatological regimes PLoS One 11 e0148780
Tanner E P, Papes¸ M, Elmore R D, Fuhlendorf S D and Davis C A
2017 Incorporating abundance information and guiding
variable selection for climate-based ensemble forecasting of
species’ distributional shifts PLoS One 12 e0184316
Theobald D M, Travis W R, Drummond M A, Gordon E S and
Betsill M 2013 The changing Southwest Assessment of
Climate Change in the Southwest United States: A Technical
Report Prepared for the U.S. National Climate Assessment ed
G Garfin, A Jardine, R Merideth, M Black and J Overpeck
(Island Press) pp 37–55
Thomey M L, Collins S L, Vargas R, Johnson J E, Brown R F,
Natvig D O and Friggens M T 2011 Effect of precipitation
variability on net primary production and soil respiration in
a Chihuahuan Desert grassland Glob. Change Biol.
17 1505–15
Tuanmu M N, Vina A, Roloff G J and Liu W 2011 Temporal
transferability of wildlife habitat models: implications for
habitat monitoring J. Biogeogr. 38 1510–23
Turner M G et al 2020 Climate change, ecosystems and abrupt
change: science priorities Phil. Trans. R. Soc. B375 20190105
U.S. Geological Survey [USGS] 2018 Gap analysis project species
habitat maps CONUS_2001 USGS data release (https://doi.
org/10.5066/F7V122T2)
Vano J et al 2020 Comparing downscaled LOCA and BCSD
CMIP5 climate and hydrology projections—release of
downscaled LOCA CMIP5 hydrology (available at: https://
gdo-dcp.ucllnl.org/downscaled_cmip_projections/
techmemo/LOCA_BCSD_hydrology_tech_memo.pdf) p 96
Vehtari A, Gelman A and Gabry J 2017 Practical Bayesian model
evaluation using leave-one-out cross-validation and WAIC
Stat. Comput. 27 1413–32
Vi˜
na A, Bearer S, Zhang H, Ouyang Z and Liu J 2008 Evaluating
MODIS data for mapping wildlife habitat distribution
Remote Sens. Environ. 112 2160–9
Whittaker D G, Van Dyke W A and Love S L 2003 Evaluation of
aerial line transect for estimating pronghorn antelope
abundance in low-density populations Wildl. Soc. Bull.
31 443–53 (available at: www.jstor.org/stable/3784323)
Woodruff S P, Lukacs P M, Christianson D and Waits L P 2016
Estimating Sonoran pronghorn abundance and survival
with fecal DNA and capture–recapture methods Conserv.
Biol. 30 1102–11
Zimmermann N E, Yoccoz N G, Edwards T C Jr, Meier E S,
Thuiller W, Guisan A, Schmatz D R and Pearman P B 2009
Climatic extremes improve predictions of spatial patterns of
tree species Proc. Natl Acad. Sci. USA 106 19723–8
15
Available via license: CC BY 4.0
Content may be subject to copyright.
Content uploaded by Steven E. Sesnie
Author content
All content in this area was uploaded by Steven E. Sesnie on Oct 17, 2024
Content may be subject to copyright.