ArticlePDF Available

Accounting for Fitness: Combining Survival and Selection when Assessing Wildlife-Habitat Relationships


Abstract and Figures

Assessing the viability of a population requires understanding of the resources used by animals to determine how those resources affect long-term popula-tion persistence. To understand the true importance of resources, one must consider both selection (where a species occurs) and fitness (reproduction and survival) associated with the use of those resources. Failure to do so may result in incorrect assessments of habitat quality and inappropriate manage-ment activities. We illustrate the importance of considering both occurrence and fitness metrics when assessing habitat requirements for the endangered greater sage-grouse in Alberta, Canada. This population is experiencing low recruitment, so we assess resource use during the brood-rearing period to identify management priorities. First, we develop logistic regression occur-rence models fitted with habitat covariates. Second, we use proportional haz-ard survival analysis to assess chick survival (fitness component) associated with habitat and climatic covariates. Sage-grouse show strong selection for sagebrush cover at both patch (smaller) and area (larger) spatial scales, and weak selection for forbs at the patch scale only. Drought conditions based on an index combining growing degree days and spring precipitation strongly reduced chick survival. While hens selected for taller grass and more sage-brush cover, only taller grass cover also enhanced chick survival. We show that sage-grouse may not recognize all ecological cues that enhance chick sur-vival. Management activities targeted at providing habitats that sage-grouse are likely to use in addition to those that enhance survival are most likely to ensure the long-term viability of this population. Our techniques account for both occurrence and fitness in habitat quality assessments and, in general, the approach should be applicable to other species or ecosystems.
Content may be subject to copyright.
ISRAEL JOURNAL OF ECOLOGY & EVOLUTION, Vol. 54, 2008, pp. 389–419
DOI: 10.1560/IJEE.54.3–4.389
*Author to whom correspondence should be addressed. E-mail:
Current address: NREL, Colorado State University and U.S. Geological Survey, 2150 Centre Avenue, Build-
ing C, Fort Collins, CO 80526-8118 USA.
Received 19 November 2007, accepted 17 July 2008.
Cameron L. aLdridge*,† and mark S. BoyCe
Department of Biological Sciences, University of Alberta, Edmonton,
Alberta T6G 2E9, Canada
Assessing the viability of a population requires understanding of the resources
used by animals to determine how those resources affect long-term popula-
tion persistence. To understand the true importance of resources, one must
consider both selection (where a species occurs) and tness (reproduction
and survival) associated with the use of those resources. Failure to do so may
result in incorrect assessments of habitat quality and inappropriate manage-
ment activities. We illustrate the importance of considering both occurrence
and tness metrics when assessing habitat requirements for the endangered
greater sage-grouse in Alberta, Canada. This population is experiencing low
recruitment, so we assess resource use during the brood-rearing period to
identify management priorities. First, we develop logistic regression occur-
rence models tted with habitat covariates. Second, we use proportional haz-
ard survival analysis to assess chick survival (tness component) associated
with habitat and climatic covariates. Sage-grouse show strong selection for
sagebrush cover at both patch (smaller) and area (larger) spatial scales, and
weak selection for forbs at the patch scale only. Drought conditions based on
an index combining growing degree days and spring precipitation strongly
reduced chick survival. While hens selected for taller grass and more sage-
brush cover, only taller grass cover also enhanced chick survival. We show
that sage-grouse may not recognize all ecological cues that enhance chick sur-
vival. Management activities targeted at providing habitats that sage-grouse
are likely to use in addition to those that enhance survival are most likely to
ensure the long-term viability of this population. Our techniques account for
both occurrence and tness in habitat quality assessments and, in general, the
approach should be applicable to other species or ecosystems.
Keywords: tness, greater sage-grouse, habitat, occurrence, persistence, sage-
brush, selection, survival
390 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Species–habitat relationships have become a priority in conservation biology (Boyce
and McDonald, 1999; Morrison, 2001; Brotons et al., 2004). Simply predicting the oc-
currence of animals across habitats is useful, but becomes much more valuable and in-
formative if occurrence (or abundance) is related to tness (Tyre et al., 2001; Breininger
and Carter, 2003; Bock and Jones, 2004; Aldridge and Boyce, 2007). Understanding
spatial variation in tness is critical to the conservation of many species of concern
(Donovan and Thompson, 2001), allowing for population viability assessment (Boyce
et al., 1994; Boyce and McDonald, 1999) and identifying appropriate management
objectives. High-quality habitats should be dened as those where animals are likely to
occur and achieve high levels of tness (reproduction and survival; Van Horne, 1983;
Morrison, 2001; Aldridge and Boyce, 2007). However, density dependence resulting in
individuals sorting themselves according to the ideal free distribution could in turn result
in higher density in selected habitats without apparent tness variation (Fretwell and Lu-
cas, 1969). Regardless, conservation of wildlife populations must make this crucial link
between resources and tness (Franklin et al., 2000; Morrison, 2001; Bock and Jones,
2004; Larson et al., 2004; Nielsen et al., 2005).
We illustrate the importance of considering both occurrence and tness metrics when
assessing habitat requirements for the endangered greater sage-grouse (Centrocercus
urophasianus; hereafter sage-grouse) in Alberta, Canada. Sage-grouse inhabit shrub-
steppe ecosystems that once covered a large portion (1.2 million km2; Schroeder et al.,
2004) of the northwestern United States and small southern portions of three western
provinces of Canada. During the last century, these ecosystems have been transformed
by agricultural activities (Connelly et al., 2004), invasion by non-native plant species
(Knick et al., 2003; Connelly et al., 2004), energy-extraction activities and develop-
ments (Braun et al., 2002; Lyon and Anderson, 2003), intense grazing pressures (Beck
and Mitchell, 2000; Hayes and Holl, 2003; Crawford et al., 2004), and climate change
(Neilson et al., 2005), resulting in direct loss of nearly half of those habitats and the
degradation and fragmentation of that which remains. All sage-grouse populations have
declined by approximately 2% per year since 1965 (Connelly et al., 2004), and low
reproductive success (Connelly and Braun, 1997; Braun, 1998; Crawford et al., 2004)
resulting from poor nesting success (Crawford and Lutz, 1985; Aldridge and Brigham,
2001; Connelly et al., 2004) and chick survival (Aldridge and Brigham, 2001; Burkepile
et al., 2002) has been identied as a potential driver of these declines. The Alberta sage-
grouse population inhabits the northern fringe of the species’ range and has declined by
66–92% since 1965 (Aldridge and Brigham, 2003).
Chick survival is one of the demographic parameters most limiting for prairie grouse
(Johnson and Braun, 1999; Aldridge and Brigham, 2002, 2003; Connelly et al., 2004;
Hagen et al., 2004) and has been identied as a priority in most conservation and recov-
ery strategies for sage-grouse throughout their range (Harris et al., 2000; Connelly et al.,
2004; Crawford et al., 2004). Thus, when identifying habitat requirements for chicks,
assessing habitat selection (occurrence) alone may result in insufcient assessments of
habitat quality (Van Horne, 1983; Morrison, 2001), potentially leading to inappropriate
management (but see Bock and Jones, 2004). The exception might occur if density de-
pendence forces sage-grouse to use sub-optimal habitats. However, sound management
strategies should assess how resources affect tness parameters such as chick survival
as well as habitat selection if sage-grouse are to persist (Aldridge, 2005; Aldridge and
Boyce, 2007).
Herein, we focus on habitats selected for brood-rearing at two spatial scales, while
simultaneously assessing how these habitats inuence chick survival for sage-grouse in
Alberta, Canada. We rst use logistic regression occurrence models to identify habitat
characteristics selected by females with broods. We then link habitat covariates to sur-
vival using a shared frailty Cox proportional hazards model to assess chick survival rela-
tive to habitat and climatic covariates. We hypothesize that sage-grouse select sagebrush
and herbaceous habitat components, as has been previously demonstrated (see Hagen
et al., 2007, for a review). Similarly, we predict that vegetation components such as
increased herbaceous cover (food) and structural cover afforded by shrubs will enhance
chick survival, whereas conditions associated with drier climate periods resulting in
reduced cover and abundance of mesic habitats containing forbs and insects (Crawford
et al., 2004) will adversely affect chick survival. However, habitat selection and survival
may not necessarily be related, particularly if sage-grouse fail to recognize ecological
factors linked to habitat quality. We then use these models to suggest minimum habi-
tat-quality thresholds that could be used by managers to maintain viable sage-grouse
The study area is located in the dry, mixed-grass prairie of southeastern Alberta, Canada
(49º24¢N, 110º42¢W, ca. 900 m elevation). Daily summer (July–August) temperatures
average 19.1 °C and annual precipitation is ca. 358 mm (AAFC–AAC 2004 unpublished
weather data). The area is characterized by many coulee draws and creeks with gentle
slopes. The dominant shrub species is silver sagebrush (Artemisia cana) and the domi-
nant forb species include pasture sage (A. frigida), several species of clover (Trifolium
spp. and Melilotus spp.), vetch (Astragalus spp.), and common dandelion (Taraxacum
ofcinale). Needle-and-thread grass (Stipa comata), june grass (Koeleria macrantha),
blue grama (Bouteloua gracilis), and western wheatgrass (Pascopyrum smithii) are the
dominant grass species (Coupland, 1961; Aldridge and Brigham, 2003).
Whereas agricultural expansion in the 1970s apparently isolated Alberta sage-grouse
from more southern populations (Schroeder et al., 2004), there has been little conver-
sion to cropland within the study region and grazing is the dominant land-use practice
(Adams et al., 2004). The landscape, however, is heavily fragmented by infrastructure
associated with oil and gas development, including roads and power lines (Braun et al.,
2002; Aldridge and Boyce, 2007). An increased frequency of extended drought condi-
tions (Aldridge and Brigham, 2002) and the introduction of West Nile virus (Naugle et
al., 2004) also adversely affect this sage-grouse population.
392 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Chick captures and relocations
Chicks of radiocollared females were captured by hand as soon as possible after hatch
by ushing the hen from her brood (May–July, 2001–2003). Chicks averaged 2.5 days
of age (range 0–8 days) at capture. From each brood we randomly selected two chicks
and attached radio transmitters to them with two sutures (similar to the technique de-
scribed by Burkepile et al. (2002; but see Aldridge, 2005). Transmitters weighed 1.6 g
and had a battery life of 10–12 weeks (BD-2G transmitters; Holohil Systems Ltd., Carp,
ON Canada). Chicks were returned to the point of capture and remotely monitored via
telemetry until the hen returned (usually within minutes).
Using standard telemetry techniques, radiomarked chicks were relocated every two
days following Aldridge and Boyce (2007). When both telemetry and ush methods
failed to detect the presence of chicks, we continued to monitor the hen every two days
to conrm brood status. Chicks were monitored through 8 weeks of age, the age at which
chicks can survive independent of the hen (Schroeder, 1997; Schroeder et al., 1999).
Habitat measurements
We assessed vegetation characteristics at one brood use location per week for each
brood tracked—typically two days after the brood was located at the site. Behaviors
were not assessed at use locations, preventing us from separating different types of
use (i.e., foraging, roosting, dispersing). We estimated the percent cover and height of
vegetation classes according to methods outlined in Aldridge and Brigham (2002; see
Table 1 for a complete list of variables). A 1-m2 quadrat was placed at the identied use
site. To identify the scale at which habitat characteristics might be selected, we took
measurements at 8 additional quadrats placed 7.5 and 15 m (two in each of the 4 cardi-
nal directions) away from the use site. The areas enclosed within the 7.5-m “patch” (the
center quadrat and the 4 quadrats 7.5 m from the center quadrat) and the 15-m “area”
(all 9 quadrats) scales were 177 and 707 m2, respectively. To obtain a potentially more
accurate estimate of percent sagebrush canopy cover (hereafter cover) we measured the
line intercept (1-cm increments) of live green sagebrush along 4–15-m line transects ra-
diating from the use site in each cardinal direction (Caneld, 1941). Measurements were
recorded separately for the rst 0–7.5 m (patch scale) and the entire 0–15 m (area scale)
transect. We recorded the same measurements at a (dependent) random location within
100–500 m of each use site, using a random azimuth and distance from the use site. From
1998–2000, Aldridge and Brigham (2002) made these same habitat assessments at a
(independent) sample of brood locations, which we use to evaluate our occurrence mod-
els. Additional variables measured only in our study from 2001–2003 included residual
grass and percent litter cover in quadrats, and we used Robel pole (Robel et al., 1970)
measurements of vertical obstruction cover at 2.5-m intervals along all 4 line-intercept
transects (Table 1).
Table 1
Explanatory habitat variables, means, and standard errors (in parentheses) of values used to assess brood occurrence and chick survival for 139
brood sites and 139 paired random locations at “patch” (177 m2) and “area” (707 m2) scales in southeastern Alberta, 2001–2003. ForbOth was
not used in survival models. When grass was absent, grass height values were considered zero. Initially, models for brood occurrence were t
with parameters above the dashed line, evaluated using an independent dataset collected from 1998–2000, and then additional parameters mea-
sured only from 2001–2003 (below dashed line) were added to the top model. No independent data were available for evaluating the nal chick
survival model
Variable 177-m2 patch scale 707-m2 area scale
code Description Brood site Random site Brood Site Random site
SBint Sagebrush cover (%) estimated using line intercept 6.12 1.76 5.12 1.94
(0.52) (0.22) (0.42) (0.22)
SB Sagebrush cover (%) estimated with 1-m2 quadrats 8.85 2.79 7.05 2.95
(0.67) (0.34) (0.49) (0.30)
Bush Cover (%) of all shrubs (including sagebrush) 11.65 4.82 10.02 5.06
estimated with 1-m2 quadrats (0.77) (0.56) (0.65) (0.53)
Gr Grass cover (%) estimated with 1-m2 quadrats 21.20 20.27 21.69 20.65
(1.15) (1.30) (1.14) (0.28)
GrHgt Mean maximum grass height (cm) 35.82 30.50 35.38 30.66
within each 1-m2 quadrat (1.20) (1.20) (13.40) (1.15)
Forb Forb cover (%) estimated with 1-m2 quadrats 8.88 8.07 8.69 8.01
(0.77) (0.72) (0.70) (0.72)
ForbOth Unpalatable (other) forb cover (%) 0.60 0.94 0.62 0.94
estimated with 1-m2 quadrats (0.10) (0.12) (0.08) (0.11)
Robel Visual obstruction reading (height in cm) 10.30 4.95 9.16 5.20
measured at 2 m from pole (0.58) (0.48) (0.46) (0.47)
Resid Residual grass cover (%) estimated using 3.61 3.62 3.63 3.65
1-m2 quadrats (0.38) (0.38) (0.37) (0.37)
Litter Estimate of the cover (%) of litter 21.20 16.86 21.22 17.27
(dead organic matter) using 1-m2 quadrats (1.02) (0.92) (0.99) (0.87)
394 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Chick survival
Date of death for a radiomarked chick was estimated as the date we failed to detect
the chick with the hen and no brooding behaviors were observed (see Aldridge and
Boyce, 2007). Chicks were recorded as having died on the date they were no longer
located with the hen.
We used a design IV approach (Erickson et al., 2001) to evaluate 4th-order (Johnson,
1980) sage-grouse brood habitat selection and chick survival. Our dependent locations
represented a random sample of unused control sites and were compared to used sites
(brood locations) for occurrence modeling using a case-control logistic regression. Sage-
grouse were not observed at any unused control sites and, given the low population density,
the proportion of control sites actually “used” by sage-grouse was low over the course of
our study (i.e., low contamination rate; Keating and Cherry, 2004). Thus, we generated a
resource selection function (RSF) contrasting used and control locations, which is propor-
tional to the probability of use (Manly et al., 2002; Keating and Cherry, 2004).
Survival analyses were based solely on used locations, comparing sage-grouse chicks
that survived (0) to those that died (1) over a particular interval. We assessed brood oc-
currence and chick survival at both measured scales (7.5-m patch and 15-m area) sur-
rounding the identied use and paired random locations. All analyses were conducted in
STATA 8.2 (STATA 2004).
Model development
A priori candidate brood occurrence models were developed using habitat data
collected from 2001–2003. These models were consistent with data collected from
1998–2000 (Aldridge and Brigham, 2002). Additional parameters (Robel, obstruction
cover; Resid, residual grass cover; and Litter, dead fallen matter) were then added in an
attempt to improve model t (Table 1).
Candidate chick survival models included all habitat variables as well as climate
covariates (Onefour Agriculture and Agri-food Canada Research Station, AAFC–AAC
2004 unpublished weather data). Small sample size limited the number of parameters we
were able to model for survival. Consequently, before testing a set of combined models
based on top models within the three groups, we chose to evaluate relative support for
candidate models within three general hypotheses describing chick survival: (1) climate,
(2) herbaceous cover and structure, and (3) sagebrush and shrub cover. We calculated
several climate variables used for survival models. Growing degree days (GDD) were
estimated as the number of degrees above 5º C for each mean daily temperature (Ball
et al., 2004), summed over the growing season (beginning 1 March and ending with the
tracking date of that year). We also developed a dryness index, which was the GDD for
that year divided by the cumulative spring precipitation since the 1 March beginning
of the growing season. We assessed all models for outliers and non-linearities (Hosmer
and Lemeshow, 1999, 2000), tested for colinearity between parameters (|r| > 0.7), and
assessed multicollinearity using variance ination factors (Menard, 1995).
Matched case-control occurrence analyses
We estimated an RSF for paired observations using a case-control logistic regression
and present coefcients for occurrence models as unstandardized linear estimates and
standard errors. This 1-to-1 matched case-control design (Hosmer and Lemeshow 2000:
223; Manly et al., 2002:150) constrains availability temporally and spatially within
similar range ecosite communities, controlling for factors that might otherwise lead to
incorrect null models or biases in habitat selection (Compton et al., 2002). We used the
Huber–White sandwich variance estimator to account for the lack of independence of
repeated habitat samples for the same brood (Pendergast et al., 1996).
Proportional hazards survival analyses
On average, chicks were relocated every 2.3 ± 0.09 days, allowing us to estimate dai-
ly survival rates using a Kaplan–Meier (KM) product limit estimator (Kaplan and Meier,
1958) with a staggered-entry design (Pollock et al., 1989; Winterstein et al., 2001). To
assess the effect of various habitat and climate covariates on chick survival, we used the
Cox proportional hazards regression model (Cox, 1972), which accommodates left and
right censoring (Andersen and Gill, 1982; Cleves et al., 2004). We used a shared frailty
model, which incorporates a latent random effect (Burnham and White, 2002) for each
brood (cluster) accounting for non-independence of chicks within broods (Cleves et al.,
2004; Wintrebert et al., 2005). We present coefcients for all survival models as hazard
ratios (exp[βi]) and standard errors.
We compared the basic KM chick survival function to the baseline cumulative survival
function without tting any covariates, but we did t a latent random effect for chicks
within broods. This method accounts for the lack of independence among siblings and
determines whether a shared frailty model is necessary. We developed Cox proportional
hazards models for each a priori candidate model using habitat (time varying) and cli-
matic (some time varying and some xed) covariates. Because we did not measure habitat
characteristics at every relocation, we carried forward habitat covariates across intervals,
assuming exposure was constant until the subsequent weekly habitat measurement loca-
tion. Independent climate variables were used for each interval (see results section).
Deaths with known “failure” times were partitioned using the Breslow estimation of
the continuous-time likelihood calculation (Cleves et al., 2004). We assessed the propor-
tional hazards assumption (Winterstein et al., 2001) for each candidate model (effects of
the covariates on survival do not change over time, except for ways in which the model
is already parameterized, Cleves et al., 2004). Models violating this assumption were
removed. We report survival estimates as means ± standard errors.
Model selection, assessment, and evaluation
We used an information-theoretic approach to model selection using Akaike’s Infor-
mation Criteria (AIC) with a correction for small sample size (AICc). We used the differ-
ences in AICc scores (Δi) to identify the best approximating occurrence or survival model
within the candidate set and AICc weights (wi) to assess the probability that a given model
was the best within the set of candidate models (Burnham and Anderson, 2002).
396 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
We used the Wald χ2 statistic (Hosmer and Lemeshow, 2000) to asses the t of each sur-
vival or occurrence model and estimated the variance explained by calculating the reduc-
tion in log-likelihood for the given model from the null model (deviance explained). For
survival models, we compared the “relative” deviance estimates between survival models
within the same set of candidate models, as outlined by Hosmer and Lemeshow (1999).
We used estimates of the receiver operating characteristic (ROC) area under the
curve (Fielding and Bell, 1997) to assess the predictive accuracy of top AICc-selected
occurrence models (Swets, 1988; Manel et al., 2001). The percent of correctly classied
(PCC) observations at the optimal cut-off was used to estimate the predictive capacity
of the top occurrence models (Nielsen et al., 2004). Predicted probabilities above the
optimal probability cut-off point (point that maximized both the sensitivity and specic-
ity curves; Swets, 1988; Nielsen et al., 2004; Liu et al., 2005) were classied as pres-
ence and those below the cutoff point were classied as absence. Prior to adding the
Robel, Resid, and Litter variables, we evaluated the top models developed with training
data (2001–2003) using an independent sample of 113 brood locations collected from
1998–2000 for 17 different broods (see Aldridge and Brigham, 2002).
To assess the t of the top combined AICc-selected chick survival models, we pre-
dicted cumulative hazard using the top model at each scale and tested for differences in
daily relative hazard for chicks that died (1) compared to those that survived (0) using
a t-test with unequal variances. Finally, we developed predictive survival curves for top
combination models to assess risk of chick mortality across the 90th percentile of the
range of availability for that parameter, while holding all other parameters at their mean
values. This allowed us to generate dose-response curves and suggest threshold levels
for the risk of chick mortality in relation to each parameter of interest based on the as-
ymptote of the curve. We could not generate similar curves for occurrence models due
to the conditional nature of the case-control analyses.
We tracked 24 broods from 2001–2003 and assessed vegetation characteristics at 139
brood sites: 42 sites from 8 broods in 2001, 15 sites from 3 broods in 2002, and 82 sites
from 13 broods in 2003. Habitat characteristics were measured at an average of 5.8 ±
0.86 sites for each brood. We captured a total of 130 chicks from 23 of the 24 tracked
broods, and radiomarked 41 chicks from 22 different broods. We obtained an average
of 11.0 (range 1–43) relocations per chick. One chick death was research related, two
chicks died from exposure (i.e., drowned in a spring rain storm), and two chicks moved
onto lands for which we could not obtain permission to access. Data on all individuals
were right censored on their last location date.
Sagebrush cover estimated by either the quadrat method (SB) or Caneld line inter-
cept method (SBint) was positively correlated at both spatial scales (r ≥ 0.87). Thus, only
one measure of sagebrush cover could be included in a given model. Grass height was
the only measure of vegetation height that was not correlated with its respective measure
of cover. All other correlated height variables were less predictive than cover estimates,
based on deviance explained in univariate models, and were not included in a priori
candidate models. Variable means at use and random locations are shown in Table 1.
Occurrence candidate models
Hypothesizing that selection for shrub cover might not be linear, we t both linear
and quadratic relationships for each shrub variable. The six different shrub component
variables were combined with six different combinations of herbaceous variables, result-
ing in 36 different a priori candidate models for sage-grouse brood occurrence (Table 2).
We present results only for occurrence models that represent the 90% condence set
(∑wi > 0.90). Additional parameters measured in 2001–2003 (visual obstruction cover
[Robel], residual grass cover [Resid], and litter ground cover [Litter]) were added to the
top model at each scale, resulting in six additional model combinations (Table 2c).
Survival candidate models
We examined seven different univariate climate models (Table 3), consisting of
various GDD and precipitation measures. The GDD model by itself violated the propor-
tional hazards assumption and was dropped from further analyses. The same six shrub
variables used for the brood occurrence analyses were used for chick survival models.
We used 13 different 1- and 2-parameter herbaceous component models (Table 4), which
we assessed both as stand-alone models and in combination with the shrub variables.
Model 12 violated the proportional hazards assumption and was dropped from our set
of candidate models.
Conditional xed-effects occurrence analyses
Tabular details for occurrence model results are shown in the Appendix. The top
brood occurrence models at both scales had weak support (wi < 0.90; Table A1), but
coefcient (βi) estimates were stable across all candidate models. When the additional
parameters were added to the top models at both spatial scales, they only marginally
increased predictive capacity and original models still had the most support (wi = 0.364
and 0.234, area and patch scales, respectively). We restricted our inferences about brood
site selection to the most parsimonious models, Model 10 and Model 28 (patch and area
scale, respectively).
Patch-scale brood occurrence
All ten highest ranked candidate models (∑wi > 0.90) contained sagebrush cover
estimated with the quadrat sampling method (SB), and the two best models included
the quadratic term. All models were highly predictive, explaining about 50% of the
variation (deviance explained) in brood occurrence (Table A1). Model #10 was the top
AICc-selected brood occurrence model and had good t (Wald χ2
4 = 43.96, p < 0.0001)
but weak support (wi = 0.16) within our set of candidate models. This model, however,
had great accuracy when predicted on both the training and testing data sets (ROCtrain =
398 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Table 2
(a) Shrub and herbaceous component models used to generate a priori candidate brood occurrence
models based on 139 brood sites and 139 paired random locations in southeastern Alberta from
2001–2003 at the patch (177 m2) and area (707 m2) scales. (b) Each of the six shrub and herba-
ceous component models were combined into 36 different initial candidate models. (c) The model
structure of the top AICc-selected model when the additional parameters were added
Shrub component variables Herbaceous component variables
A = SB g = Gr
B = SB + SB2 h = Gr + GrHgt
C = Bush i = Gr + Forb
D = Bush + Bush2 j = Gr + GrHgt + Forb
E = SBint k = Forb + GrHgt
F = SBint + SBint2 l = Gr + GrHgt + Forb + ForbOth
# Sagebrush # Bush models # Sagebrush
quadrat models intercept models
1 A + g 13 C + g 25 E + g
2 B + g 14 D + g 26 F + g
3 A + h 15 C + h 27 E + h
4 B + h 16 D + h 28 F + h
5 A + i 17 C + i 29 E + i
6 B + i 18 D + i 30 F + i
7 A + j 19 C + j 31 E + j
8 B + j 20 D + j 32 F + j
9 A + k 21 C + k 33 E + k
10 B + k 22 D + k 34 F + k
11 A + l 23 C + l 35 E + l
12 B + l 24 D + l 36 F + l
Model # Additional parameter models
Top model (Top AICc-selected model for a given scale)
1 (Top model ) + Robel
2 (Top model ) + Resid
3 (Top model ) + Litter
4 (Top model ) + Robel + Resid
5 (Top model ) + Robel + Litter
6 (Top model ) + Resid + Litter
7 (Top model ) + Robel + Resid + Litter
0.992, ROCtest = 0.841) and excellent (84.1%) and good prediction (77.0%), respectively
(Table A2).
Inferences based on this top model indicate strong positive but decreasing selection
for sagebrush cover (concave function; Table A3). Hens selected strongly for taller grass
at brood sites, and weakly for greater percent forb cover (Table A3).
Area-scale brood occurrence
Of the 12 models at the area scale within the 90% condence set (Table A1), 8 con-
tained the SBint variable as either a linear or quadratic term, and all contained the GrHgt
variable. All models explained >41.0% of the variation in brood habitat selection, with
the top model (Model 28) explaining 44.1%. Similar to that of the patch-scale model, this
model had weak support (wi = 0.18) as the top candidate model, but it had good t (Wald
4 = 56.42, p < 0.0001) and good model accuracy for both training and testing datasets
(ROCtrain = 0.900, ROCtest = 0.802, Table A2). Model 28 also had good prediction (79%)
for the training dataset and reasonable prediction on the independent testing dataset (71%,
Table A2). Inference based on Model 28 at the area scale again indicated strong positive
but decreasing selection for sagebrush cover (concave function; Table A3). Broods were
found in areas with taller grass but avoided areas with greater grass cover.
Using a basic Kaplan Meier (KM) curve, chick survival to 8 weeks (56 days) was
estimated at 0.296 ± 0.081 (Fig. 1). There were no between-year differences in survival
Table 3
Explanatory climate variables and models used to assess chick survival for 41 radiomarked chicks
from 22 different broods in southeastern Alberta, 2001–2003. Variables were generated for each
year that chicks were followed. Due to small sample sizes, a priori climate models consisted of
single parameters only
Model # Variable code Description
1a GDDb Cumulative growing degree days (above 5 ºC)
from 1 March to the chick location date
2 Sp_PPT_Cumm Cumulative growing season (since 1 March)
3 Dry_Index An overall dryness index, calculated as the GDD
(above) divided by Sp_PPT_Cumm (above)
4 Sp_PPT_Prior Total spring (April through June) precipitation for the
prior spring
5 Sp-Su_PPT_Prior Total spring and summer precipitation (April though
August) of the prior year
6 Tot_PPT_Prior Total precipitation for the prior full calendar year
7 GDD_Prior Total growing degree days (above 5 ºC) from March
through August for the prior year
aThe GDD model was dropped due to violations of the proportional hazards assumption.
bAll weather data were provided by Onefour Agriculture and Agri-food Canada Research Station,
located in the study area (Onefour, Alberta, Canada).
400 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
(log rank χ2
2 = 2.86, p = 0.24) nor between rst (n = 33) and second (n = 8; log rank χ2
1 =
2.32, p = 0.13) nesting attempts, allowing us to pool data for further survival analyses.
The baseline hazard chick survival model using the shared frailty produced lower
survival estimates to 56 days (0.123) than the KM estimate, and was outside the 95%
CI for the KM model (range 0.151 to 0.497, Fig. 1). The estimate of the frailty variance,
theta (θ = 0.96), was large and signicant at α = 0.10 (likelihood ratio χ2
1 = 1.87, p =
0.086). Therefore we t a shared frailty model for all candidate models.
Climate chick survival models
Of the six climate models tested, Model 3 (dryness index only) was the top AICc-se-
lected model. This model had only moderate support (wi = 0.34), but it had reasonable
t (Wald χ2
1 = 3.48, p = 0.06). By itself, the dryness index explained more than twice
as much variation in chick survival as any other individual climate variable (10.97%).
Climate Model 3 (Dry_Index) was selected for use in our combined models.
Shrub chick survival models
Tabular details for survival model results are shown in the Appendix. At the patch
scale, the top AICc-selected chick proportional hazards shrub model contained the SB
variable, suggesting a linear relationship with chick survival (Table A4). This model (#1)
had only moderate support (wi = 0.44), but the Akaike weight was more than double the
second best model (SBint). The model had signicant t (Wald χ2
1 = 6.13, p = 0.01),
Table 4
Candidate models used to identify the shrub and herbaceous models that best predicted sage-
grouse chick survival for 41 radiomarked chicks from 22 different broods in southeastern Alberta,
2001–2003. We did not have an independent testing dataset for candidate models containing “ad-
ditional” parameters (Resid, Robel, and Litter)
Shrub Shrub component Herbaceous Herbaceous component
Model # variables model # variables
1 SB 1 Forb
2 SB + SB2 2 Forb + Gr
3 Bush 3 Forb + Robel
4 Bush + Bush2 4 Forb + Resid
5 SBint 5 Robel
6 SBint + SBint2 6 Robel + GrHgt
7 Robel + Resid
8 Gr + GrHgt
9 Resid + GrHgt
10 Litter
11 Litter + Forb
12a Litter + Robel
13 Litter + GrHgt
aHerbaceous model #12 was dropped due to violations of the proportional hazards assumption.
explained 14.22% of the variation, and was used for combined model building at the
patch scale.
At the area scale, the top AICc-selected chick proportional hazards model contained
the quadratic for sagebrush estimated with the line intercept method (SBint + SBint2;
Table A4). This model had moderate support (wi = 0.34) and the Akaike weight was
about twice that of the next best model. This model had good t (Wald χ2
1 = 6.09, p <
0.05) and explained the most variation within the candidate set at this scale (22.56%
deviance explained, Table A4). We used shrub Model 6 (SBint + SBint2) for combined
candidate models at the area scale.
Herbaceous chick survival models
At both the patch and area scales, Model 8 (Gr + GrHgt) was the top AICc-selected
herbaceous survival model (Table A5). At the patch scale this model had weak support
Fig. 1. Kaplan Meier (KM) cumulative chick survival curves for 41 radiomarked sage-grouse
chicks from 22 different broods in southeastern Alberta, 2001–2003. The basic KM curve (solid
line) does not take into account the correlation of marked chicks within the same brood, whereas
the frailty model (dashed line) represents the baseline Cox proportional hazard survival (i.e., no
covariates) and accounts for lack of independence of siblings within the same brood. We could not
generate 95% condence intervals for the frailty model due to the conditional nature of the Cox
model on covariates within the model.
402 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
(wi = 0.30) and moderate t (Wald χ2
1 = 4.76, p = 0.09), but explained the greatest devi-
ance (18.53%) of all herbaceous models. Similarily, at the area scale, Model 8 had a poor
t (Wald χ2
1 = 3.70, p = 0.16) and weak support (wi = 0.20), but explained the greatest
deviance (14.35%; Table A5). We retained Model 8 as the herbaceous model for com-
bined survival models at both scales.
Combination chick survival models
Using the top shrub and herbaceous models for each spatial scale, and the top climate
model, we developed seven candidate models for each scale. The candidate model set
consisted of the top models from each group and all possible combinations of these mod-
els (Table A6). The patch scale combination model SB + Dry_Index failed to converge
and was removed.
Model 5, which contained a climate and herbaceous component, was the top AICc-
selected model at the patch scale (Table A7). This model had good t (Wald χ2
1 = 12.12,
p = 0.007), moderate support (wi = 0.65), and explained 42.68% of the variation in sur-
vival. Risk of chick mortality increased as the drought index increased, was strongly re-
duced with increased grass cover, but increased with grass height (Table A8). Threshold
response curves suggested a signicant reduction in risk to sage-grouse chicks if grass
cover was greater than about 20–25% (Fig. 2a). Although risk increased with increas-
ing grass height, this risk is realized only when grass height is greater than ca. 40 cm
(Fig. 2b). The model also demonstrates that the moderate-to-high dryness index values
dramatically increase the risk of chick death (Fig. 2c).
At the area scale, Model 6 was the top AICc-selected survival model (Table A7).
This model had good t (Wald χ2
1 = 16.74, p = 0.005), strong support as the top can-
didate model (wi = 0.91), and explained considerably more variation in chick survival
(58.27%) than any other model. Risk of death again increased with the dryness index,
and was positive but decreasing with sagebrush cover (Table A8), suggesting higher
chick survival in less dense sagebrush habitats. Risk of chick death was slightly re-
duced with increased grass cover but increased with grass height (Table A8). Threshold
response curves indicate that the relative risk of chick death increased with greater
sagebrush cover, and tailed off in denser sagebrush habitats (Fig. 3a). Risk was higher
above about 3% sagebrush cover (line-intercept) but was reduced if cover was greater
than ~9%. Similar to the patch-level threshold curves (Fig. 3a), risk was reduced
with increased grass cover at the area scale, but the threshold was lower (>5% cover,
Fig. 3b). Risk also increased with increasing grass height at the area scale, but only
when grass was taller than about 30–35 cm (Fig. 3c). Again, the area-level-threshold
model also illustrates that hot and dry growing seasons (high dryness index values)
reduce chick survival (Fig. 3d).
Both the patch- and area-scale models validated well on the within-sample training
dataset. The mean daily hazard was signicantly greater for chicks that died within the
56-day monitoring period compared to those that survived or were censored (patch
scale: t37.2 = 4.17, p < 0.001; area scale: t31.9 = 3.73, p < 0.001). Based on model covari-
ates, chicks that died were exposed to more hazardous or risky conditions.
Fig. 2. Threshold response curves for the top AICc-selected model (combined Model #5) at the
patch scale (177 m2) for relative risk (hazard) for sage-grouse chicks in southeastern Alberta, from
2001–2003. Responses are shown across the 90th percentile of availability for each parameter in
the model while holding the other parameters in the model at their mean values.
404 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Our results highlight the importance of accounting for tness components when as-
sessing wildlife–habitat relationships (Van Horne, 1983; Morrison, 2001; Aldridge and
Boyce, 2007). Sage-grouse may not always select for habitat characteristics (e.g., high
selection for dense sagebrush cover and tall grasses) that enhance tness measured by
chick survival (e.g., increased chick mortality in dense sagebrush and in sites with tall
[>35 cm] grasses). Thus, management efforts should strive to maintain and enhance
habitats that are likely to increase survival, in addition to those selected by the birds. For
this population, dening brood habitat requirements as those that enhance juvenile sur-
vival, and ultimately recruitment, are necessary to appropriately identify management
needs for the species (Aldridge and Brigham, 2001; Crawford et al., 2004).
Overall, we were able to explain 44–50% of sage-grouse brood habitat selection and
chick survival using only climatic and habitat covariates. Dose-response curves from
survival models allowed us to generate threshold levels for habitat variables such as
sagebrush cover, grass cover, and grass height, which will allow for enhanced chick
survival. These thresholds provide initial targets for managing sage-grouse brood-rear-
ing habitat in Alberta.
Fig. 3. Threshold response curves for the top AICc-selected model (combined Model #6) at the
area scale (707 m2) for relative risk (hazard) for sage-grouse chicks in southeastern Alberta, from
2001–2003. Responses are shown across the 90th percentile of availability for each parameter in
the model while holding the other parameters in the model at their mean values.
Similar to previous studies, we conclude that the lack of forb-rich habitats that exist
in this study likely contributed to the observed selection of sagebrush throughout the
brood-rearing period (Aldridge and Brigham, 2002). Although we detected selection
for forbs at the patch scale, a similar pattern was not evident at the area scale. Some of
the herbaceous survival models that contained forbs had reasonable deviance explained
(Table A5), yet none of the patch- or area-scale chick survival models containing forbs
were selected as the most predictive model. However, as suggested (but not assessed)
in other studies (Peterson, 1970; Schoenberg, 1982; Drut et al., 1994a; Sveum et al.,
1998; Aldridge and Brigham, 2002), the risk of chick death in our study was reduced
with greater forb cover, but the effect was weak (95% CI overlapped 1). The uniformly
low availability of forbs in southeastern Alberta may limit our ability to detect differ-
ences in selection and survival relative to forb availability. If forbs are important for
survival but abundance is low everywhere, survival rates may be uniformly low relative
to forb cover, limiting variation in survival and our ability to detect trends. More than 50
marked individuals (41 in our study) might also be required to generate robust survival
estimates (Winterstein et al., 2001).
The scale of habitat measurement appeared to play a minor role in chick survival and
habitat selection. Chick mortalities were predicted by grass cover and height at both the
patch and area scales. Taller grass at both spatial scales appeared to have negative con-
sequences for chick survival, but threshold models illustrate that habitats are not risky
until grass is taller than 35–40 cm (Figs. 2b,3c). Hens appear to recognize this, selecting
only moderately for tall grass at both scales. Conversely, patches containing grass cover
beyond 20–25% (Fig. 2a) greatly reduced the risk of chick mortality.
However, hens appear not to recognize tness ties to greater grass cover, showing
strong avoidance of dense grass cover. While dense grass cover may reduce the risk of
chick mortality, hens may be forced to make a trade-off between these less risky grass-
dominated habitats and foraging on forbs and insects in mesic habitats that are open and
thus, more risky (less grass and structural cover). The low availability of mesic forb-rich
habitats in Alberta may force hens to spend more time meeting dietary requirements, which
may put their chicks, and possibly themselves, at greater risk of predation—an ecological
trap (Delibes et al., 2001; Breininger and Carter, 2003). In Alberta, management strategies
that enhance cover of grass and increase the abundance of mesic habitats to elevate forb
abundance would enhance habitat quality and population viability for sage-grouse. Further
research is required to understand these relationships, possibly in larger populations with
more variability in forb abundance and where larger sample sizes could be obtained.
Our results suggest that precipitation and climate (dryness index) play a pivotal role
in sage-grouse chick survival. Spring precipitation has long been suggested to correlate
with sage-grouse productivity (June, 1963; Gill, 1966; Schroeder et al., 1999), but
until now quantitative studies addressing its effects have not been conducted. Warm
years with high amounts of precipitation in the growing season likely result in greater
structural growth and protective cover. This may enhance nesting success (June, 1963;
Aldridge and Brigham, 2002) and can elevate chick survival. Precipitation prevents forb
desiccation and enhances insect abundance, both of which are important food resources
406 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
for sage-grouse chicks (Klebenow and Gray, 1968; Dunn and Braun, 1986; Johnson and
Boyce, 1990; Drut et al., 1994b).
Although we cannot manage climate to benet sage-grouse populations, it is important
to recognize that weather patterns are highly variable and will affect chick survival. To
ensure that populations remain viable when subjected to stochastic events, such as extreme
weather or disease outbreaks, it would be important for managers to ensure the availability
of high-quality brood-rearing habitats that encourage sage-grouse use and maximize sur-
vival (and reproduction) when using those habitats. Ensuring these habitats are in proximity
to high-quality nesting habitats within a landscape context (Aldridge and Boyce, 2007) will
increase the probability that hens will use these habitats and successfully edge chicks.
An obvious and interesting difference in factors affecting chick survival was evident
between models at patch and area scales. While sagebrush cover was an important
component of the area-scale model, no sagebrush or shrub variables entered into the top
model at the patch scale. This lack of relationship with sagebrush cover and survival at the
patch scale was surprising, given that brood occurrence models indicated that brood hens
select strongly for moderate ranges of sagebrush cover. Previous research has shown that
sage-grouse select for sagebrush cover early in the brood-rearing cycle, prior to moving
away from sagebrush uplands (Patterson, 1952; Dunn and Braun, 1986) and into forb-rich
mesic habitats containing 14–40% forb cover (Peterson, 1970; Schoenberg, 1982; Drut et
al., 1994a). However, avoidance of dense sagebrush during brood-rearing has also been
detected in Washington (Sveum et al., 1998). While hens move their chicks into sage-
brush habitats, it appears to compromise chick survival and might be maladaptive, again
resulting in an ecological trap (Delibes et al., 2001; Donovan and Thompson, 2001; Bock
and Jones, 2004). This is signicant, given that reproduction and juvenile survival drive
population dynamics for sage-grouse (Johnson and Braun, 1999).
We strongly suggest that future studies assessing wildlife–habitat relationships con-
sider both processes that determine habitat quality for a given species: occurrence and
tness (Van Horne, 1983; Morrison, 2001; Aldridge and Boyce, 2007). Selection by
individuals for certain resources may not result in tness enhancements. Thus, man-
agement objectives developed based on occurrence information alone may result in
misguided conservation efforts, as we have demonstrated for sage-grouse in Alberta.
Whereas tness data often are more difcult and costly to gather, we encourage fur-
ther research into occurrence–tness relationships, across local and landscape scales
(Aldridge and Boyce, 2007). The techniques we used here for linking occurrence and
survival, although limited in wildlife and conservation elds, offer a proven and prom-
ising approach for accurately assessing habitat quality and developing habitat-based
population viability assessments for a variety of species (Boyce et al., 1994; Boyce and
McDonald, 1999; Aldridge and Boyce, 2007).
We thank all the landowners who allowed us to conduct our research on their lands.
This research was supported by the Alberta Conservation Association; Alberta Sustain-
able Resource Development; Alberta Sport Recreation Parks and Wildlife Foundation;
Cactus Communications (Medicine Hat, Alberta); Challenge Grants in Biodiversity;
Endangered Species Recovery Fund (World Wildlife Fund Canada and the Canadian
Wildlife Service); Esso Imperial Oil, Manyberries, Alberta; Murray Chevrolet, Medi-
cine Hat, Alberta; the North American Waterfowl Management Plan; and the University
of Alberta. C.L. Aldridge was personally supported by the Andrew Stewart Memorial
Graduate Prize, Bill Shostak Wildlife Award, Dorothy J. Killam Memorial Graduate
Prize, Edmonton Bird Club Scholarship, Izaak Walton Killam Memorial Scholarship,
John and Patricia Schlosser Environment Scholarship, Macnaughton Conservation
Scholarship, and Natural Science and Engineering Research Council Scholarship. C.
Nielsen and H. Beyer provided valuable GIS assistance. C. Johnson, J. Frair, S. Nielsen,
and M.M. Club assisted with statistical issues. T. Bush, J. Carpenter, L. Darling, C.
Dockrill, Q. Fletcher, J. Ng, M. Olsen, J. Saher, J. Sanders, D. Sharun, M. Swystun, and
M. Watters assisted with eld data collection. D. Eslinger and J. Nicholson assisted with
logistics. E. Bork, D. Coltman, M. Gillingham, S. Hannon, J. Saher, D. Neubaum, G.
Chong, W. Wetzel, C. Melcher, D. Howerter, and one anonymous reviewer improved
previous drafts of this manuscript.
Adams, B.W., Carlson, J., Milner, D., Hood, T., Cairns, B., Herzog, P. 2004. Benecial grazing
management practices for sage-grouse (Centrocercus urophasianus) and ecology of silver
sagebrush (Artemisia cana) in southeastern Alberta. Technical Report, Public Lands and Forest
Division, Alberta Sustainable Resource Development, Edmonton, Alberta, Pub. No. T/049.
Aldridge, C.L. 2005. Habitats for persistence of greater sage-grouse (Centrocercus urophasianus)
in Alberta, Canada. Ph.D. dissertation, University of Alberta, Edmonton, Alberta, Canada. 250
Aldridge, C.L., Boyce, M.S. 2007. Linking occurrence and tness to persistence: a habitat-based
approach for greater sage-grouse. Ecol. Appl. 17: 508–526.
Aldridge, C.L., Brigham, R.M. 2001. Nesting and reproductive activities of greater sage-grouse
in a declining northern fringe population. Condor 103: 537–543.
Aldridge, C.L., Brigham, R.M. 2002. Sage-grouse nesting and brood habitat use in southern
Canada. J. Wildlife Manage. 66: 433–444.
Aldridge, C.L., Brigham, R.M. 2003. Distribution, abundance, and status of the greater sage-
grouse, Centrocercus urophasianus, in Canada. Can. Field Nat. 117: 25–34.
Andersen, P.K., Gill, R.D. 1982. Cox regression-model for counting-processes—a large sample
study. Ann. Stat. 10: 1100–1120.
Ball, D.A., Frost, S.M., Gitelman, A.I. 2004. Predicting timing of downy brome (Bromus tecto-
rum) seed production using growing degree days. Weed Sci. 52: 518–524.
Beck, J.L., Mitchell, D.L. 2000. Inuences of livestock grazing on sage grouse habitat. Wildl.
Soc. Bull. 28: 993–1002.
Bock, C.E., Jones, Z.F. 2004. Avian habitat evaluation: should counting birds count? Front. Ecol.
Environ. 2: 403–410.
Boyce, M.S., McDonald, L.L. 1999. Relating populations to habitats using resource selection
functions. Trends Ecol. Evol. 14: 268–272.
408 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Boyce, M.S., Meyer, J.S. Irwin, L. 1994. Habitat-based PVA for the northern spotted owl. In:
Fletcher, D.J., Manly, B.F.J., eds. Statistics in ecology and environmental monitoring. Univer-
sity of Otago Press, Dunedin, New Zealand, pp. 63–85.
Braun, C.E. 1998. Sage grouse declines in western North America: what are the problems? Proc.
West. Assoc. of Fish and Wildlife Agencies 78: 139–156.
Braun, C.E., Oedekoven, O.O., Aldridge, C.L. 2002. Oil and gas development in western North
America: effects on sagebrush steppe avifauna with particular emphasis on sage grouse. Trans.
N. Am. Wildl. Nat. Res. 67: 337–349.
Breininger, D.R., Carter, G.M. 2003. Territory quality transitions and source-sink dynamics in a
Florida scrub-jay population. Ecol. Appl. 13: 516–529.
Brotons, L., Thuiller, W., Araujo, M.B., Hirzel, A.H. 2004. Presence-absence versus presence-
only modelling methods for predicting bird habitat suitability. Ecography 27: 437–448.
Burkepile, N.A., Connelly, J.W., Stanley, D.W., Reese, K.P. 2002. Attachment of radiotransmitters
to one-day-old sage grouse chicks. Wildl. Soc. Bull. 30: 93–96.
Burnham, K.P., Anderson, D.R. 2002. Model selection and multi-model inference: a practical
information-theoretic approach. Springer-Verlag, New York, 488 pp.
Burnham, K.P., White, G.C. 2002. Evaluation of some random effects methodology applicable to
bird ringing data. J. Appl. Stat. 29: 245–264.
Caneld, R.H. 1941. Application of the line interception method in sampling range vegetation. J.
Forest. 39: 388–394.
Cleves, M.A., Gould, W.W., Gutierrez, R.G. 2004. An introduction to survival analyses using
STATA. Stata Press, College Station, Texas, 308 pp.
Compton, B.W., Rhymer, J.M., Mccollough, M. 2002. Habitat selection by wood turtles (Clemmys
Insculpta): an application of paired logistic regression. Ecology 83: 833–843.
Connelly, J.W., Braun, C.E. 1997. Long-term changes in sage grouse Centrocercus urophasianus
populations in western North America. Wildl. Biol. 3: 229–234.
Connelly, J.W., Knick, S.T., Schroeder, M.A. Stiver, S.J. 2004. Conservation assessment of greater
sage-grouse and sagebrush habitats. West. Assoc. of Fish and Wildlife Agencies, Cheyenne,
Coupland, R.T. 1961. A reconsideration of grassland classication in the northern Great Plains of
North-America. J. Ecol. 49: 135–167.
Cox, D.R. 1972. Regression models and life-tables. J. R. Stat. Soc. B 34: 187–220.
Crawford, J.A., Lutz, R.S. 1985. Sage grouse population trends in Oregon, 1941–1983. Murrelet
66: 69–74.
Crawford, J.A., Olson, R.A., West, N.E., Mosley, J.C., Schroeder, M.A., Whitson, T.D., Miller,
R.F., Gregg, M.A., Boyd, C.S. 2004. Synthesis paper—ecology and management of sage-
grouse and sage-grouse habitat. J. Range Manage. 57: 2–19.
Delibes, M., Gaona, P., Ferreras, P. 2001. Effects of an attractive sink leading into maladaptive
habitat selection. Am. Nat. 158: 277–285.
Donovan, T.M., Thompson, F.R. 2001. Modeling the ecological trap hypothesis: a habitat and
demographic analysis for migrant songbirds. Ecol. Appl. 11: 871–882.
Drut, M.S., Crawford, J.A., Gregg, M.A. 1994a. Brood habitat use by sage grouse in Oregon.
Great Basin Nat. 54: 170–176.
Drut, M.S., Pyle, W.H., Crawford, J.A. 1994b. Technical note: diets and food selection of sage
grouse chicks in Oregon. J. Range Manage. 47: 90–93.
Dunn, P.O., Braun, C.E. 1986. Summer habitat use by adult female and juvenile sage grouse. J.
Wildl. Manage. 50: 228–235.
Erickson, W.P., McDonald, T.L., Gerow, K.G., Howlin, S., Kern, J.W. 2001. Statistical issues in
resource selection studies with radio-marked animals. In: Millspaugh, J.J., Marzluff, J.M., eds.
Radio-tracking and animal populations. Academic Press, New York, pp. 209–242.
Fielding, A.H., Bell, J.F. 1997. A review of methods for the assessment of prediction errors in
conservation presence/absence models. Environ. Conserv. 24: 38–49.
Franklin, A.B., Anderson, D.R., Gutierrez, R.J., Burnham, K.P. 2000. Climate, habitat quality,
and tness in northern spotted owl populations in northwestern California. Ecol. Monogr. 70:
Fretwell, S.D., Lucas, H.L. 1969. On territorial behaviour and other factors inuencing habitat
distribution in birds. Acta Biotheor. 19: 16–36.
Gill, R.B. 1966. Weather and sage grouse productivity. Colorado Game, Fish and Parks Dep.
Denver, Colorado, Outdoor Information Leaet 37.
Hagen, C.A., Jamison, B.E., Giesen, K.M., Riley, T.Z. 2004. Guidelines for managing lesser prai-
rie-chicken populations and their habitats. Wildl. Soc. Bull. 32: 69–82.
Hagen, C.A., Connelly, J.W., Schroeder, M.A. 2007. A meta-analysis of greater sage-grouse Cen-
trocercus urophasianus nesting and brood-rearing habitats. Wildl. Biol. 13(1): 43–50.
Harris, W.C., Lungle, K.J., Bristol, B., Dickinson, D., Eslinger, D., Fargey, P., McAdam, S.M., Po-
irier, T., Scobie, D. 2000. Canadian sage grouse recovery strategy. The Canadian sage grouse
recovery team, Edmonton, Alberta.
Hayes, G.F., Holl, K.D. 2003. Cattle grazing impacts on annual forbs and vegetation composition
of mesic grasslands in California. Conserv. Biol. 17: 1694–1702.
Hosmer, D.W., Lemeshow, S. 1999. Applied survival analysis: regression modeling of time to
event data. John Wiley & Sons, New York, 386 pp.
Hosmer, D.W., Lemeshow, S. 2000. Applied logistic regression. John Wiley & Sons, New York,
375 pp.
Johnson, D.H. 1980. The comparison of usage and availability measurements for evaluating re-
source preference. Ecology 61: 65–71.
Johnson, G.D., Boyce, M.S. 1990. Feeding trials with insects in the diet of sage grouse chicks. J.
Wildl. Manage. 54: 89–91.
Johnson, K.H., Braun, C.E. 1999. Viability and conservation of an exploited sage grouse popula-
tion. Conserv. Biol. 13:77–84.
June, J.W. 1963. Wyoming sage grouse population measurement. Proc. West. Assoc. Game and
Fish Commiss., Tucson, Arizona. 43: 206–211.
Kaplan, E.L., Meier, P. 1958. Nonparametric-estimation from incomplete observations. J. Am.
Stat. Assoc. 53: 457–481.
Keating, K.A., Cherry, S. 2004. Use and interpretation of logistic regression in habitat selection
studies. J. Wildl. Manage. 68: 774–789.
Klebenow, D.A., Gray, G.M. 1968. Food habits of juvenile sage grouse. J. Range Manage. 21:
Knick, S.T., Dobkin, D.S., Rotenberry, J.T., Schroeder, M.A., Vander Haegen, W.M., Van Riper,
C. 2003. Teetering on the edge or too late? Conservation and research issues for avifauna of
sagebrush habitats. Condor 105: 611–634.
Larson, M.A., Thompson, F.R., Millspaugh, J.J., Dijak, W.D., Shiey, S.R. 2004. Linking popula-
tion viability, habitat suitability, and landscape simulation models for conservation planning.
Ecol. Model. 180: 103–118.
Liu, C., Berry, P.M., Dawson, T.P., Pearson, R.G. 2005. Selecting thresholds of occurrence in the
prediction of species distributions. Ecography 28: 385–393.
410 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Lyon, A.G., Anderson, S.H. 2003. Potential gas development impacts on sage grouse nest initia-
tion and movement. Wildl. Soc. Bull. 31: 486–491.
Manel, S., Williams, H.C., Ormerod, S.J. 2001. Evaluating presence-absence models in ecology:
the need to account for prevalence. J. Appl. Ecol. 38: 921–931.
Manly, B.F.J., McDonald, L.L., Thomas, D.L., McDonald, T.L., Erickson, W.P. 2002. Resource
selection by animals: statistical design and analysis for eld studies. Chapman & Hall, London,
177 pp.
Menard, S. 1995. Applied logistic regression analysis. Sage Publications, Thousand Oaks, Cali-
fornia, 130 pp.
Morrison, M.L. 2001. Invited Paper: a proposed research emphasis to overcome the limits of
wildlife-habitat relationship studies. J. Wildl. Manage. 65: 613–623.
Naugle, D.E., Aldridge, C.L., Walker, B.L., Cornish, T.E., Moynahan, B.J., Holloran, M.J., Brown,
K., Johnson, G.D., Schmidtmann, E.T., Mayer, R.T., Kato, C.Y., Matchett, M.R., Christiansen,
T.J., Cook, W.E., Creekmore, T., Falise, R.D., Rinkes, E.T., Boyce, M.S. 2004. West Nile virus:
pending crisis for greater sage-grouse. Ecol. Lett. 7: 704–713.
Neilson, R.P., Lenihan, J.M., Bachet, D., Drapek R.J. 2005. Climate change implications for sage-
brush ecosystems. Trans. N. Am. Wildl. Nat. Res. 70: 145–159.
Nielsen, S.E., Munro, R.H.M., Bainbridge, E.L., Stenhouse, G.B., Boyce, M.S. 2004. Grizzly
bears and forestry II. Distribution of grizzly bear foods in clearcuts of west-central Alberta,
Canada. Forest Ecol. Manage. 199: 67–82.
Nielsen, S.E., Johnson, C.J., Heard, D.C., Boyce, M.S. 2005. Can models of presence-absence
be used to scale abundance? Two case studies considering extremes in life history. Ecography
28: 1–12.
Patterson, R.L. 1952. The sage grouse in Wyoming. Sage Books, Denver, Colorado, 341 pp.
Pendergast, J.F., Gange, S.J., Newton, M.A., Lindstrom, M.J., Palta, M., Fisher, M.R. 1996. A
survey of methods for analyzing clustered binary response data. Int. Stat. Rev. 64: 89–118.
Peterson, J.G. 1970. The food habits and summer distribution of juvenile sage grouse in central
Montana. J. Wildl. Manage. 34: 147–155.
Pollock, K.H., Winterstein, S.R., Bunck, C.M., Curtis, P.D. 1989. Survival analysis in telemetry
studies: the staggered entry design. J. Wildl. Manage. 53: 7–15.
Robel, R.J., Briggs, J.N., Dayton, A.D., Hulbert, L.D. 1970. Relationships between visual obstruc-
tion measurements and weight of grasslands vegetation. J. Range Manage. 23: 295–297.
Schoenberg, T.J. 1982. Sage grouse movements and habitat selection in North Park, Colorado.
Ph.D. thesis, Colorado State University, Fort Collins, Colorado.
Schroeder, M.A. 1997. Unusually high reproductive effort by sage grouse in a fragmented habitat
in north-central Washington. Condor 99: 933–941.
Schroeder, M.A., Young, J.R., Braun, C.E. 1999. Sage grouse (Centrocercus urophasianus). In:
Pool, A., Gill, F., eds. The birds of North America. The Birds of North America, Inc., Phila-
delphia, p. 28.
Schroeder, M.A., Aldridge, C.L., Apa, A.D., Bohne, J.R., Braun, C.E., Bunnell, S.D., Connelly,
J.W., Deibert, P.A., Gardner, S.C., Hilliard, M.A., Kobriger, G.D., Mcadam, S.M., Mccarthy,
C.W., Mccarthy, J.J., Mitchell, D.L. Rickerson, E.V., Stiver, S.J. 2004. Distribution of sage-
grouse in North America. Condor 106: 363–376.
STATA. 2004. STATA version 8.2. STATA Corporation, College Station, Texas.
Sveum, C.M., Crawford, J.A., Edge, W.D. 1998. Use and selection of brood-rearing habitat by
sage grouse in south central Washington. Great Basin Nat. 58: 344–351.
Swets, J.A. 1988. Measuring the accuracy of diagnostic systems. Science 240: 1285–1293.
Tyre, A.J., Possingham, H.P., Lindenmayer, D.B. 2001. Inferring process from pattern: can territo-
ry occupancy provide information about life history parameters? Ecol. Appl. 11: 1722–1737.
Van Horne, B. 1983. Density as a misleading indicator of habitat quality. J. Wildl. Manage. 47:
Winterstein, S.R., Pollock, K.H., Bunck, C.M. 2001. Analysis of survival data from radiotelem-
etry studies. In: Millspaugh, J.J., Marzluff, J.M., eds. Radio-tracking and animal populations.
Academic Press, New York, pp. 351–380.
Wintrebert, C.M.A., Zwinderman, A.H., Cam, E., Pradel, R., Van Houwelingen, J.C. 2005. Joint
modelling of breeding and survival in the kittiwake using frailty models. Ecol. Model. 181:
412 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Table A1
Rankings by Akaike’s information criterion corrected for small sample size (AICc) for brood
occurrence models, difference in AICc value between the ith and top-ranked model (Δi), and
Akaike weights (wi) for all models within a cumulative summed AICc weight (∑wi) of 0.90 for 139
brood locations at the patch (177 m2) and area (707 m2) scale for greater sage-grouse in Alberta,
2001–2003. All model Wald χ2 tests were signicant at p < 0.0001. Percent deviance explained
(Dev. exp.) indicates the reduction in the log-likelihood from the null model
Model Model Model % Dev.
# structure AICc Δi wi wi Wald χ2 exp.
Patch 10 SB + SB2 + Forb + 104.803 0.000 0.163 0.163 43.96 49.92
4 SB + SB2 + Gr + 104.925 0.121 0.153 0.316 43.97 49.86
3 SB + Gr + GrHgt 105.270 0.467 0.129 0.446 39.40 48.58
9 SB + Forb + GrHgt 105.400 0.597 0.121 0.567 39.41 48.51
8 SB + SB2 + Gr + 105.501 0.698 0.115 0.682 49.15 50.67
GrHgt + Forb
7 SB + Gr + GrHgt + 106.007 1.204 0.089 0.771 44.33 49.29
12 SB + SB2 + Gr + 107.574 2.770 0.041 0.812 59.42 50.73
GrHgt + Forb + ForbOth
6 SB + SB2 + Gr + Forb 107.739 2.936 0.038 0.849 43.98 48.39
11 SB + Gr + GrHgt + 108.043 3.240 0.032 0.882 55.28 49.35
Forb + ForbOth
5 SB + Gr + Forb 108.846 4.043 0.022 0.903 35.23 46.72
Area 28 SBint + SBint2 + 116.021 0.000 0.182 0.182 56.42 44.10
Gr + GrHgt
27 SBint + Gr + GrHgt 116.876 0.855 0.119 0.301 53.00 42.55
36 SBint + SBint2 + Gr + 117.212 1.192 0.100 0.401 81.46 45.73
GrHgt + Forb + ForbOth
32 SBint + SBint2 + Gr + 117.394 1.374 0.092 0.493 62.47 44.50
GrHgt + Forb
35 SBint + Gr + GrHgt + 117.395 1.374 0.092 0.584 73.14 44.50
Forb + ForbOth
31 SBint + Gr + GrHgt + 118.150 2.129 0.063 0.647 56.97 42.99
4 SB + SB2 + Gr + GrHgt 118.435 2.414 0.054 0.701 46.33 42.84
34 SBint + SBint2 + Forb + 118.504 2.484 0.053 0.754 40.43 42.81
10 SB + SB2 + Forb + 118.838 2.817 0.044 0.798 44.66 42.64
33 SBint + Forb + GrHgt 119.044 3.023 0.040 0.839 32.95 41.43
3 SB + Gr + GrHgt 119.220 3.199 0.037 0.875 47.98 41.34
9 SB + Forb + GrHgt 119.233 3.212 0.037 0.912 39318 41.33
Table A2
Comparison of top AICc-selected brood occurrence models, metrics for overall model signicance, model t, and classication accuracy for both
training (139 brood sites from 2001–2003) and testing data (113 brood sites from 1998–2000) across different scales for greater sage-grouse in
Alberta. All model Wald χ2 tests were signicant at p < 0.0001. Percent deviance explained (Dev. exp.) indicates the reduction in the log-likeli-
hood from the null model. The area under the receiver operating characteristic curves (ROC [SE]) and the percent correctly classied (PCC)
observations based on the training dataset’s optimal cutoff point were used to assess model classication accuracy
Scale Model Model % Dev. Optimal Training data Testing data
# AICc-selected model AICc Wald χ2 exp. cutoff ROC PCC ROC PCC
177-m2- 10 SB + SB2 + Forb + GrHgt 104.803 43.96 49.92 0.5012 0.992 84.17 84.09 76.99
patch scale (0.015) (0.027)
area scale 28 SBint + SBint2 + Gr + GrHgt 116.021 56.42 44.10 0.5014 0.900 79.14 0.8024 70.80
(0.017) (0.029)
414 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Table A3
Estimated coefcients (βi), standard errors (shown in parentheses), and 95% condence intervals
for top AICc-selected candidate brood occurrence models for greater sage-grouse in southeastern
Alberta. Models were developed on 139 brood sites and 139 paired random locations collected
from 2001–2003
Condence Condence
intervals intervals
Patch scale Lower Upper Area scale Lower Upper
Variable Model #10 Model #28
SB 0.460 0.296 0.623
SB2 –0.007 –0.009 –0.004
SBint 0.757 0.425 1.090
SBint2 –0.024 –0.039 –0.009
Gr –0.040 –0.062 –0.017
GrHgt 0.058 0.010 0.107 0.115 0.060 0.170
(0.025) (0.028)
Forb 0.038 –0.004 0.080
Table A4
AICc-selected shrub variable proportional hazards chick survival models and Akaike weights (wi) for all models at the 177-m2-patch and 707-m2-
area scales for 41 chicks in southeastern Alberta, 2001–2003. The Wald χ2 indicates the t of the model to the data, and K indicates the number
of model parameters estimated, which includes the covariates and the estimate of the random effect (theta). Theta is the estimate of the shared
frailty variance and the p-value for the likelihood ratio tests (LR) for the signicance of the correlation. Percent deviance explained (Dev. exp.)
indicates the reduction in the log-likelihood from the null model
Model # Shrub model Theta LR Log- Model Model Dev.
structure estimate p-value likelihood K AICc Δi AICc wi Wald χ2 χ2 p-value exp. (%)
177-m2-patch scale
1 SB <0.001 0.437 72.022 2 148.616 0.000 0.441 6.13 0.013 14.22
5 SBint 0.052 0.483 72.854 2 150.279 1.663 0.192 3.96 0.047 8.07
2 SB + SB2 <0.001 0.500 71.974 3 151.149 2.532 0.124 5.91 0.052 14.56
3 Bush 0.777 0.158 73.363 2 151.298 2.682 0.115 1.08 0.30 4.08
4 Bush + Bush2 0.805 0.169 72.522 3 152.244 3.628 0.072 2.60 0.272 10.57
6 SBint + SBint2 <0.001 0.466 72.777 3 152.755 4.138 0.056 4.25 0.120 8.65
707-m2-area scale
6 SBint + SBint2 0.348 0.342 70.795 3 148.790 0.000 0.346 6.09 0.048 22.56
4 Bush + Bush2 0.805 0.154 71.455 3 149.924 1.134 0.196 4.34 0.114 9.42
1 SB 0.059 0.475 72.676 2 150.111 1.321 0.179 4.30 0.038 18.18
2 SB + SB2 0.109 0.455 72.034 3 151.046 2.257 0.112 4.65 0.098 5.08
5 SBint 0.326 0.351 73.237 2 151.267 2.478 0.100 2.02 0.155 14.14
3 Bush 0.871 0.125 73.759 2 152.089 3.300 0.066 0.22 0.636 0.86
416 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Table A5
AICc-selected herbaceous variable proportional hazards chick survival models and Akaike weights (wi) for all models at the patch (177 m2) and
area (707 m2) scale for 41 chicks in southeastern Alberta, 2001–2003. The Wald χ2 indicates the t of the model to the data, and K indicates the
number of model parameters estimated, which includes the covariates and the estimate of the random effect (theta). Theta is the estimate of the
shared frailty variance and the p-value for the likelihood ratio tests (LR) for the signicance of the correlation. Percent deviance explained (Dev.
exp.) indicates the reduction in the log-likelihood from the null model
Model Model Theta LR Model Model Dev.
# structure estimate p-value K AICc Δi AICc wi Wald χ2 χ2 p-value exp. (%)
Patch 8 Gr + GrHgt 0.215 0.352 3 150.007 0.000 0.300 4.76 0.093 18.53
2 Forb + Gr <0.001 0.500 3 151.412 1.405 0.149 4.83 0.089 13.62
1 Forb 0.798 0.110 2 151.696 1.689 0.129 0.60 0.439 2.47
5 Robel 0.923 0.091 2 152.234 2.227 0.098 0.06 0.801 0.26
10 Litter 0.939 0.099 2 152.290 2.283 0.096 0.01 0.930 0.03
4 Forb + Resid 0.403 0.351 3 153.998 3.991 0.041 1.35 0.509 3.79
11 Litter + Forb 0.800 0.115 3 154.324 4.318 0.035 0.60 0.742 2.48
3 Forb + Robel 0.798 0.110 3 154.325 4.318 0.035 0.60 0.741 2.47
6 Robel + GrHgt 0.909 0.100 3 154.494 4.487 0.032 0.43 0.808 1.78
9 Resid + GrHgt 0.789 0.170 3 154.590 4.583 0.030 0.37 0.832 1.39
7 Robel + Resid 0.723 0.190 3 154.706 4.700 0.029 0.29 0.867 0.91
13 Litter + GrHgt 0.976 0.096 3 154.759 4.752 0.028 0.17 0.920 0.69
Area 8 Gr + GrHgt 0.296 0.307 3 151.208 0.000 0.202 3.70 0.157 14.35
1 Forb 0.779 0.117 2 151.920 0.713 0.142 0.40 0.529 1.56
5 Robel 0.878 0.099 2 151.995 0.787 0.136 0.30 0.583 1.25
2 Forb + Gr 0.034 0.483 3 152.119 0.911 0.128 4.02 0.134 11.04
10 Litter 0.995 0.083 2 152.248 1.040 0.120 0.05 0.822 0.21
4 Forb + Resid 0.201 0.444 3 154.248 3.040 0.044 1.61 0.448 2.79
6 Robel + GrHgt 0.843 0.115 3 154.289 3.081 0.043 0.62 0.734 2.62
3 Forb + Robel 0.779 0.118 3 154.439 3.231 0.040 0.50 0.777 2.01
11 Litter + Forb 0.805 0.114 3 154.490 3.282 0.039 0.45 0.800 1.80
7 Robel + Resid 0.745 0.204 3 154.584 3.376 0.037 0.39 0.822 1.41
13 Litter + GrHgt 1.016 0.083 3 154.814 3.606 0.033 0.11 0.944 0.47
9 Resid + GrHgt 0.802 0.188 3 154.817 3.610 0.033 0.14 0.933 0.45
Table A6
Overall combined candidate proportional hazards chick survival models for 41 radiomarked
chicks from 22 different broods at the patch (177 m2) and area (707 m2) scales in southeastern
Alberta, 2001–2003. The top climate (Climate), shrub (Shrub), and herbaceous (Herb) models
were used at each scale for combination models. The top within each group was also considered
as candidate models within this set. The patch-scale model with the combination of sagebrush and
the dryness index (SB + Dry _Index) would not converge on a Maximum Likelihood estimate and
was therefore not estimated
Model Patch scale Model Area scale
# combination models # combination models
1st Shrub 1- SB 1st Shrub 1- SBint + SBint2
1st Herb 2- Gr + GrHgt 1st Herb 2- Gr + GrHgt
1st Climate 3- Dry_Index 1st Climate 3- Dry_Index
4- SB + Gr + GrHgt 4- SBint + SBint2 + Gr +
5- Gr + GrHgt + Dry_Index 5- Gr + GrHgt
6- SB + Dry_Index + Gr + 6- SBint + SBint2 +
GrHgt Dry_Index + Gr +
418 C.L. ALDRIDGE AND M.S. BOYCE Isr. J. Ecol. Evol.
Table A7
AICc-selected combined proportional hazards chick survival models and Akaike weights (wi) for all models at the patch (177 m2) and area
(707 m2) scales for 41 chicks in southeastern Alberta, 2001–2003. The Wald χ2 indicates the t of the model to the data, and K indicates the
number of model parameters estimated, which includes the covariates and the estimate of the random effect (theta). Theta is the estimate of the
shared frailty variance and the p-value for the likelihood ratio tests (LR) for the signicance of the correlation is presented. Percent deviance
explained (Dev. exp.) indicates the reduction in the log-likelihood from the null model
Model # Combined Theta LR Log- Model Model Dev.
model structure estimate p-value Likelihood K AICc Δi AICc wi Wald χ2 χ2 p-value exp. (%)
177-m2-patch scale
5 Dry_Index + Gr + <0.001 0.500 67.184 4 144.474 0.000 0.650 12.12 0.007 42.68
6 SB + Dry_Index + <0.001 0.500 67.053 5 147.440 2.966 0.148 13.11 0.011 43.30
Gr + GrHgt
1 SB <0.001 0.437 72.022 2 148.616 4.142 0.082 6.13 0.013 14.22
3 Dry_Index <0.314 0.283 72.468 2 149.508 5.034 0.052 3.48 0.062 10.97
2 Gr + GrHgt 0.215 0.352 71.403 3 150.007 5.533 0.041 4.76 0.093 18.53
4 SB + Gr + GrHgt 0.001 0.500 70.362 4 150.829 6.355 0.027 8.18 0.042 25.31
707-m2-area scale
6 SBint + SBint2 + 0.314 0.283 63.377 5 140.087 0.000 0.905 16.74 0.005 58.27
Dry_Index + Gr +
1 SBint + SBint2 0.348 0.342 70.795 2 146.161 6.074 0.043 6.09 0.048 22.56
5 Dry_Index + Gr + 0.034 0.483 68.299 4 146.704 6.617 0.033 10.55 0.014 37.10
3 Dry_Index 0.779 0.117 72.468 2 149.508 9.421 0.008 3.48 0.062 10.97
4 SB + Gr + GrHgt 0.878 0.099 69.893 4 149.892 9.805 0.007 7.49 0.112 28.17
2 Gr + GrHgt 0.296 0.307 72.003 3 151.208 11.121 0.003 3.70 0.157 14.35
Table A8
Estimated hazard ratios (exponentiated coefcients—exp[βi]), standard errors (shown in parenthe-
ses), and condence intervals for top AICc-selected candidate proportional hazards chick survival
combined models for 41 chicks from 22 different broods in southeastern Alberta, 2001–2003. The
top combined model at both scales had the Dry_Index, the Gr + GrHgt herbaceous component
model, and a sagebrush component
Condence Condence
Patch scale intervals Area scale intervals
Variable model # 5 Lower Upper model # 6 Lower Upper
Dry_Index 1.441 1.123 1.850 1.707 1.256 2.321
(0.183) (0.268)
SBint 2.068 1.230 3.479
SBint2 0.941 0.898 0.985
Gr 0.932 0.882 0.985 0.953 0.894 1.017
(0.026) (0.031)
GrHgt 1.056 1.015 1.098 1.076 1.025 1.130
(0.021) (0.027)
... Presence-only methods are statistically weak, estimating niche-related objects that are found "at some unspecified point along a continuum between the fundamental and the realized niche" (Colwell & Rangel, 2009;Jiménez-Valverde et al., 2008;Peterson et al., 2011;Sober on & Peterson, 2005). Niche models that examine abundance or occupancy without corresponding measures of survey effort, habitat availability and population growth (Aldridge & Boyce, 2008;DeCesare et al., 2014), are of limited utility in an explanatory or predictive capacity (Pagel & Schurr, 2011;Schurr et al., 2012). ...
... Nevertheless, statistical analyses of the niche cannot go far without biological mechanism (McInerny & Etienne, 2012a). For example, above, I have argued that spatial data cannot estimate the niche without information on population growth, density dependence, and demography (Aldridge & Boyce, 2008;DeCesare et al., 2014;Matthiopoulos et al., 2015;Pagel & Schurr, 2011;Schurr et al., 2012). Equally, it seems foolhardy to attempt predictions of species realized niches, based purely on mechanistic models, without fitting them to distribution data (Peterson et al., 2015). ...
Full-text available
During the past century, the fundamental niche, the complete set of environments that allow an individual, population, or species to persist, has shaped ecological thinking. It is a crucial concept connecting population dynamics, spatial ecology and evolutionary theory, and a prerequisite for predictive ecological models at a time of rapid environmental change. Yet, its properties have eluded quantification, particularly for mobile, cognitively complex organisms. These difficulties are mainly a result of the separation between niche theory and field data, and the dichotomy between environmental and geographical spaces. Here, I combine recent mathematical and statistical results linking habitats to population growth, to achieve a quantitative and intuitive understanding of the fundamental niches of animals. I trace the development of niche ideas from the early steps of ecology to their use in modern statistical and conservation practice. I examine how animal mobility and behavior may blur the division between geographical and environmental space. I discuss how the central models of population and spatial ecology lead to a concise mathematical equation for the fundamental niche of animals and demonstrate how fitness parameters can be understood and directly estimated by fitting this model simultaneously to data on population growth and spatial distributions. I first illustrate these concepts theoretically for territorial species. I then fit the fundamental niche model to a dataset of house sparrow colonies to quantify how a species of selective animals can increase their fitness in heterogeneous environments. This work confirms ideas that had been anticipated in the historical niche literature. Specifically, within traditionally defined environmental spaces, habitat heterogeneity and behavioral plasticity make the fundamental niche more complex and malleable than was historically envisaged. However, once examined in higher‐dimensional environmental spaces, accounting for spatial heterogeneity, the niche is more predictable than recently suspected. This re‐evaluation quantifies how organisms might buffer themselves from change by bending the boundaries of viable environmental space, and offers a framework for designing optimal habitat interventions to protect biodiversity or obstruct invasive species. It therefore promotes the fundamental niche as a key concept for understanding animal responses to changing environments and a central tool for environmental management.
... perceptual traps for sage-grouse: non-sagebrush shrub cover for nesting, wetland and variability in heat load for brood rearing, variability in soil moisture and non-sagebrush shrub cover for summer survival, and non-irrigated agricultural disturbance for winter survival. Similarly, Aldridge and Boyce (2008) documented decreased sage-grouse brood selection for high grass cover even though more grass cover decreased chick mortality. ...
... This was suggested by Smith et al. (2018), who reported brood-rearing female sage-grouse in central Wyoming selected riskier habitats with more open herbaceous cover for foraging chicks, whereas broodless females selected less risky, more concealed habitat. In addition, Aldridge and Boyce (2008) reported brood-rearing female sage-grouse in southern Alberta may have selected riskier areas with less grass cover to maximize foraging opportunities. This was similarly suggested by Gibson et al. (2016a) in that nest-site selection by female sage-grouse may be a function of her selecting brood-rearing habitat and not necessarily for nest survival. ...
Full-text available
Maladaptive habitat selection, where animals select habitat with reduced fitness potential or avoid otherwise suitable habitat, exacerbates the threat of population decline for species vulnerable from habitat loss and fragmentation. The greater sage‐grouse (Centrocercus urophasianus) is a species of conservation concern for which research has identified scenarios where populations may be under the influence of maladaptive habitat selection. Our objective was to evaluate whether sage‐grouse selected habitat relative to habitat quality (i.e., ability to provide for higher survival or reproductive success), and to identify any habitat characteristics where they were not matching selection with costs and benefits, during the nesting, brood rearing, adult breeding, adult summer, and adult winter seasons. We measured an overall apparent adaptive relationship between habitat selection and survival for brood, adult breeding, and adult winter habitat. There was an overall apparent maladaptive relationship for nest and adult summer survival. Of 25 specific habitat characteristics that influenced sage‐grouse reproductive success or survival, 13 (52%) had an apparent adaptive selection relationship, 10 (40%) had an apparent maladaptive relationship, and 2 (8%) were either inconclusive or not strongly selected. Surprisingly, most (8 of 10) of the habitat characteristics we observed that were selected contrary to apparent costs or benefits were associated with environmental variables (i.e., topography and vegetation). Relative to possible maladaptive selection and anthropogenic disturbance, grouse selected for areas of higher mortality risk near minor roads during the breeding season and grouse did not select for non‐irrigated agricultural disturbance which had lower mortality risk. However, after accounting for the effects of habitat selection on all demographic rates that determine fitness, these apparent maladaptive selection effects were probably not biologically significant. The strongest evidence we observed for maladaptive habitat selection associated with anthropogenic land use was during summer when grouse were selecting for the edge of irrigated hayfields where there was higher mortality risk. To ensure the success of sage‐grouse conservation actions, we encourage further investigation identifying the mechanisms behind observed cases of apparent maladaptive selection or identifying any fitness benefits that grouse are gaining from selecting risky areas.
... An ultimate goal of research should be to verify that variation in habitat suitability is tied to limiting resources with fitness consequences (Morrison 2001), an aim not yet applied to MCA but increasingly targeted in studies of ecology and conservation biology (Aldridge and Boyce 2008;Losier et al. 2015;Catlin et al. 2019;Maresh Nelson et al. 2020). Despite the above caveats, maximum clique analysis provides a conservative estimate of potential population capacity of a given landscape for Indiana bobcats, and it could readily be used for other species. ...
Full-text available
Context Maximum clique analysis (MCA) can approximate landscape carrying capacity (Nk) for populations of territorial wildlife. However, MCA has not been widely adopted for wildlife applications, mainly due to computational constraints and software wildlife biologists may find difficult to use. Moreover, MCA does not incorporate uncertainty into estimates of Nk. Objectives We extended MCA by applying a vertex cover algorithm to compute Nk over a large (92,789 km²), continuous spatial scale for female bobcats (Lynx rufus) in Indiana, USA. We incorporated uncertainty by calculating confidence intervals for Nk across five thresholds of habitat suitability using 10 replicate suitability maps from bootstrapped datasets. For portions of the landscape too large to be solved with the vertex cover algorithm, we compared predictions from a linear model and a “greedy” algorithm. Results Mean estimates of Nk for female bobcats in Indiana across habitat suitability thresholds ranged from 539 (0.75 threshold) to 1200 territories (0.25 threshold). On average, each 12.5 percentile reduction in the suitability threshold increased estimates for Nk by 1.2-fold. Both the predictive and greedy algorithm produced reasonable estimates of maximum cliques for areas that were too large to compute with the vertex cover algorithm. The greedy algorithm produced smaller confidence intervals compared to the predictive approach but underestimated maximum cliques by 1.2%. Conclusions Our research demonstrates effective application of MCA to species occupying large landscapes while accounting for uncertainty. We believe our methods, coupled with availability of annotated scripts developed in R, will make MCA more broadly accessible to wildlife biologists.
... During the late brooding period, sage-grouse hens will often select for mesic habitats that are more productive (Aldridge & Boyce, 2008;Aldridge & Brigham, 2002;Connelly et al. 2011a), but topographic wetness does not always appear as a determining variable and may depend on the surrounding habitat . Here, we found preference for mesic areas with a selection for higher mean and heterogeneity of topographic wetness, higher precipitation, and higher NDVI in the BRT analysis. ...
Full-text available
Identifying and quantifying the extent to which landscape‐level habitat variables drive the spatial distribution of individuals across a region can provide fundamental insights into a species ecology and be essential to wildlife management and conservation plans. Although the preferences for habitat resources and the resources themselves are not static over time, most research at large spatial scales does not consider seasonal effects nor quantify annual temporal variability in the spatial distribution of habitat resources. In this study, we used a machine learning (boosted regression trees [BRTs]) and generalized linear mixed model (GLMM) approach to quantify seasonal habitat selection across three life stages (nest, late brood, and winter habitat) of sage‐grouse and estimated annual stability across a 13‐year dataset in south‐central Wyoming. Generalized linear mixed models had high area under the curve (AUC) values, but were not as high as the BRT models that had mean AUC values of 0.86, 0.81, and 0.87 for nest, late brood, and winter habitat, respectively. Generalized linear mixed models and BRT result provided similar results, but because of the higher validation values of the BRT models, we assessed annual variation by predicting the BRT models across years. We found significant spatial trends in the distribution of nesting habitat, with general decreases in the relative probability of use across the core of the study area and corresponding increases in selection on the periphery. The primary temporally shifting variables for the nesting BRT models were development, Normalized Difference Vegetation Index, and topographic wetness, suggesting they were shifting out of preferable ranges for these variables as habitat suitability was decreased over the course of our study. Winter habitat appeared to have similar spatial changes in probability of selection, but these changes were likely related to changes in winter precipitation and snow depth, which were the primary contributors to the winter BRT models. The annual dynamics of habitat selection are seldom addressed in large‐scale research but can have potentially dramatic influences on our identification of preferred habitats.
... 2008), resulting in an erroneous conclusion that a species relies on a suboptimal habitat type (Aldridge andBoyce 2008, Kerley et al. 2011). In addition, degraded high-quality habitats may take decades to recover (Vesk et al. 2008, DellaSala et al. 2013) before conditions allow for full use (Selwood et al. 2009, Elphick et al. 2015. ...
The San Clemente Bell’s Sparrow (Artemisiospiza belli clementeae) is a federally threatened subspecies endemic to San Clemente Island, California. Previous research suggested dependence on boxthorn (Lycium californicum) as breeding habitat and nesting substrate; however, this conclusion was based on data collected when introduced feral ungulates had severely degraded the soil and vegetation cover. Since removal of the ungulates, native vegetation has gradually increased and the San Clemente Bell’s Sparrows have expanded into areas where habitat had been unsuitable. To explore how Bell’s Sparrows use these areas, we examined reproductive metrics associated with habitat covariates gathered at 214 nest sites used by Bell’s Sparrows from 2014 to 2016. We found that nest success in boxthorn habitat, previously considered an essential habitat for Bell’s Sparrow nesting, was similar to success in alternative habitat types. Our findings contradict previous conclusions that Bell’s Sparrows were boxthorn-dependent. We believe this previously documented relationship was likely due to the lack of available alternative nesting habitat following years of feral ungulate degradation, and Bell’s Sparrows now reproduce in multiple habitat types and throughout most of San Clemente Island. Furthermore, our findings illustrate the importance of long-term monitoring and corresponding adaptive management when monitoring species in changing and recovering landscapes.
... Hazel grouse prefers large coniferous and mixed forests, but can also inhabit fragmented, smaller forest complexes. This species depends on different habitats and food resources throughout the year and could not survive and reproduce in areas which do not sufficiently provide all resources (Chalfoun and Martin 2007;Aldridge and Boyce 2008;Kajtoch et al. 2016 (Cramp and Simmons 1980). This species also requires certain plant structures to survive different phenological seasons (Ludwig and Klaus 2017;Matysek et al. 2018). ...
Full-text available
Hazel grouse habitat requirements are relatively well known in coniferous forests, and less known in mixed or deciduous forests. We studied habitat differences between sites occupied by hazel grouse Tetrastes bonasia and control plots in mixed mountain forests of the Western Carpathians in 2009 and 2010. Hazel grouse presence at sites was determined in April and May. The habitat variables (n = 21) and the proportion of tree and shrub species (n = 22) were collected both in sites of hazel grouse presence and control plots within a radius of 100 m. Greater numbers of tree species and greater proportions of deciduous trees (mainly birch Betula sp.) were found in sites where hazel grouse was present. Lower canopy cover was an important variable for hazel grouse occurrence, and sites with hazel grouse had a greater proportion of young trees (< 40 years). Sites were also characterized by a higher proportion of overgrown glades and dead woods in comparison with control plots. Sites occupied by hazel grouse were characterized by a greater number of tree species in the undergrowth (minimum of five species) in comparison with control plots. GLM models revealed that the most important environmental factors for hazel grouse occurrence in mixed mountain forests were open habitats (overgrown glades), good hiding opportunities (fallen trees and dead woods) and good conditions for foraging (trees cover in undergrowth). Poplar (Populus sp.) and willow (Salix sp.) were the most important tree species for hazel grouse occurrence. The presence of habitat structures and the vegetations richness provides good shelter or food for the hazel grouse. Extensive forest management should be proposed to increase the number of hazel grouse. Large areas covered by herbs and light-seeded tree species of low economical value for forestry are recommended to support hazel grouse population.
Full-text available
Numerous wildlife species within semi-arid shrubland ecosystems across western North America are experiencing substantial habitat loss and fragmentation. These changes in habitat are often attributed to a diverse suite of factors including prolonged and increasingly severe droughts, conifer expansion, anthropogenic development, domestic and feral livestock grazing, and invasion of exotic annual grasses which promotes increased wildfire frequency and severity. Greater sage-grouse (Centrocercus urophasianus; hereafter, sage-grouse) are considered an indicator of sagebrush ecosystem health and have experienced widespread population decline associated with habitat loss and degradation, as well as changes in predator communities. Our objectives were to model and map sage-grouse habitat selection and survival during the important brood-rearing life stage in relation to landscape-scale environmental predictors. Furthermore, we sought to understand impacts of wildfire and annual grass invasion on brood habitat, as these accelerated disturbance regimes are a primary cause of habitat loss within the Great Basin region of the USA. We used a hierarchical Bayesian modeling framework to estimate resource selection functions and survival for early and late brood-rearing stages of sage-grouse in relation to a broad suite of habitat characteristics evaluated at multiple spatial scales within the Great Basin from 2009 – 2019. Sage-grouse selected for greater perennial grass cover at higher relative elevations, closer to springs and wet meadows during both early and late brood-rearing. Terrain characteristics, including heat load and aspect, were important in survival models, as was variation in shrub height. We also found strong evidence for higher survival for both early and late broods within previously burned areas, but survival within burned areas decreased as annual grass cover (i.e. cheatgrass, Bromus tectorum) increased. This interaction effect demonstrates how invasion of annual grasses into burned areas, which has become prevalent in Great Basin sagebrush ecosystems, can lead to maladaptive habitat selection by brood-rearing greater sage-grouse. Understanding these complex relationships aids wildlife conservation and habitat management as wildfire and annual grass cycles continue to accelerate across western ecosystems.
Identifying, protecting, and restoring habitats for declining wildlife populations is foundational to conservation and recovery planning for any species at risk of decline. Resource selection analysis is a key tool to assess habitat and prescribe management actions. Yet, it can be challenging to map suitable resource conditions across a wide range of ecological contexts and use the resulting models to identify effective and universal habitat improvement actions. We developed a management-centric modeling approach that sought to balance the need to evaluate the consistency of key habitat conditions and improvement actions across multiple, distinct populations, while allowing context-specific environmental variables and spatial scales to nuance selection responses that form the basis of location-specific management prescriptions. To demonstrate this approach, we developed a set of habitat selection models for Gunnison sage-grouse (Centrocercus minimus), a threatened species under the U.S. Endangered Species Act. Conservation, species recovery, and habitat management efforts are needed in six isolated satellite populations (San Miguel, Crawford, Piñon Mesa, Dove Creek, Cerro Summit-Cimarron-Sims, and Poncha Pass) where environmental conditions differ, and the already small number of birds are declining. We used multi-scale and seasonal resource selection analyses to quantify relationships between environmental conditions and sites used by animals. All models included key habitat variables often altered through management actions to assess their differential influences across models. We found important similarities and differences among satellites, indicating that, although some rules of thumb are generally well-grounded, the consideration of population-specific environmental differences could increase the efficiency of local habitat improvement actions. Sage-grouse also had diverse responses to resource conditions at different scales, indicating that regional spatial (e.g., landscape) and local patch scale can differentially influence expected habitat improvements associated with where such management actions are implemented. Although context variables such as topography cannot be manipulated, sage-grouse associations revealed information that could guide the siting of improvement actions. This approach to balancing management objectives associated with habitat assessment may benefit spatially-structured populations with different environmental contexts and species with complex habitat needs and associations.
Full-text available
Temperature at fine spatial scales is an important driver of nest site selection for many avian species during the breeding season and can influence nest success. Sagebrush (Artemisia spp.) communities have areas with high levels of vegetation heterogeneity and high thermal variation; however, fire removes vegetation that provides protection from predators and extreme environmental conditions. To examine the influence of microclimates on Greater Sage-Grouse (Centrocercus urophasianus) nest site selection and nest success in a fire-affected landscape, we measured black bulb temperature (Tbb) and vegetation attributes (e.g., visual obstruction) at 3 spatial scales (i.e. nest bowl, microsite, and landscape) in unburned and burned areas. Nest bowls exhibited greater buffering of Tbb than both nearby microsites and the broader landscape. Notably, nest bowls were warmer in cold temperatures, and cooler in hot temperatures, than nearby microsites and the broader landscape, regardless of burn stage. Nest survival (NS) was higher for nests in unburned areas compared to nests in burned areas (unburned NS = 0.43, 95% confidence interval [CI]: 0.33–0.54; burned NS = 0.24, 95% CI: 0.10–0.46). The amount of bare ground was negatively associated with NS, but effects diminished as the amount of bare ground reached low levels. Shrub height and visual obstruction were positively associated with NS during the entire study period, whereas minimum Tbb had a weaker effect. Our findings demonstrate that thermoregulatory selection by Greater Sage-Grouse at nest sites had marginal effects on their NS. However, given that increases in vegetation structure (e.g., shrub height) provide thermal refuge and increase NS, vegetation remnants or regeneration in a post-fire landscape could be critical to Greater Sage-Grouse nesting ecology.
A literature review of empirical research on the interactions between wind-energy development and grouse (Aves:Tetraoninae) of the North American plains and shrub-steppe.
Full-text available
Modern ecological research often involves the comparison of the usage of habitat types or food items to the availability of those resources to the animal. Widely used methods of determining preference from measurements of usage and availability depend critically on the array of components that the researcher, often with a degree of arbitrariness, deems available to the animal. This paper proposes a new method, based on ranks of components by usage and by availability. A virtue of the rank procedure is that it provides comparable results whether a questionable component is included or excluded from consideration. Statistical tests of significance are given for the method. The paper also offers a hierarchical ordering of selection processes. This hierarchy resolves certain inconsistencies among studies of selection and is compatible with the analytic technique offered in this paper.
An animal is captured, fitted with a radio transmitter, and released. From the time of release, the animal's unique radio signal is monitored to determine the animal's fate at more or less regular intervals. For each animal, the investigator must know the date it was radio-marked and released, the date it was last located, and its status when last located. At each location time, the status of each animal is recorded as alive, dead, or missing. Missing animals are considered censored, meaning that the event of interest cannot be observed. This type of censoring is generally referred to as “right censoring” and is caused by such factors as radio failure, topography that inhibits signal reception, and permanent or temporary emigration. As the field of spatial analysis grows, there is a need to integrate survival and spatial analyses. As the ability to assess actual and potential habitat quality improves, the impact of an animal's movements must be modeled through the use of a mosaic of habitat types on its survival. There is also a need to develop stage-structured survival models. These models will, in many cases, mimic time-dependent models, allowing the examination of the impact on survival of an animal moving into a new age class or moving to a lower or higher quality territory.
Visual obstruction measurements were used to determine height and density of vegetation in a Kansas grassland. These visual obstruction measurements were compared with the weight of vegetation collected from each site. The weight of vegetation collected was significantly correlated with the visual obstruction measurements.
Diets and food selection by sage grouse (Centrocercus urophasianus) chicks were determined during 1989 and 1990 on 2 areas that differed in long-term grouse productivity. Chicks consumed the same foods in similar frequencies and exhibited similar dietary selection on the areas, but relative dry mass differed. Forbs and invertebrates composed 80% of the dietary mass on the area with higher grouse productivity, whereas chicks on the other area consumed primarily (65%) sagebrush (Artemisia spp. L.).
In this essay, I develop the idea that studies of wildlife-habitat relationships could do more to advance our understanding of the distribution, abundance, and fitness of wild animals. Contemporary wildlife studies are hampered by at least 2 problems: (1) a lack of progress in ecology, and (2) a confusion of concepts and terms. The plethora of meanings applied to central concepts such as habitat in ecology is inhibiting communication among scientists and between scientists and managers. Standardized, operational definitions of concepts such as habitat are essential if wildlife scientists are to measure similar habitat entities. The habitat concept certainly can be used to develop general descriptors of the distribution of animals. However, we repeatedly fail to find commonalities in defining "habitat" for most populations across space and time because we usually miss the underlying mechanisms (e.g., size and distribution of prey, forage nutrients, competitive factors) determining occupancy, survival, and fecundity. Habitat per se can provide only a limited explanation of the ecology of an animal. A major problem with focusing on habitat is that, by definition, habitat can remain the same (at least as we typically measure it) while use of niche parameters by an animal within that habitat can change. Habitat usually fails as a predictor of animal performance (e.g., fecundity, fitness) because of our unwillingness to identify constraints on exploitation of critical resources and consideration of critical limiting factors. Thus, I suggest a focus on resources: the basic, fundamental currency that allows individual animals to survive and reproduce. This approach will help us to focus our efforts and advance our understanding of how wild animals respond to variation in resource abundance. To be useful for advancing knowledge, the resource currency must have relevance to the fitness of the animal and be within the animal's perceptive abilities to measure it. Various constraints (i.e., niche parameters) act to reduce the rate of resource consumption and thus per-capita population growth. Therefore, I suggest that researchers and managers concentrate on identifying and analyzing the separate roles of critical resources and the factor(s) constraining their use. We must clearly elucidate and separate resources and constraints if we are to advance and develop reliable knowledge for wildlife management.
One climax community of Fescue Prairie and five climax and edaphic climax communities of Mixed Prairie are characterized on the basis of studies in ninety-six relatively undisturbed sites over an 18-year period in the glaciated, Canadian Great Plains. This classification represents modifications from one proposed a decade ago on the basis of response of communities to more favourable growing conditions from 1950 to 1956. The relative dominance of species is judged on the basis of calculated foliage yield. The communities are: A. Festuca scabrella Association of black soil. B. Mixed Prairie 1. Stipa-Agropyron Faciation of medium-textured soils in the dark-brown and moist portion of the brown soil zones. 2. Stipa-Bouteloua-Agropyron Faciation of medium-textured soils in the arid portion of the brown soil zone. 3. Stipa-Bouteloua Faciation, an edaphic community of well stabilized sandy soils. 4. Bouteloua-Agropyron Faciation, an edaphic climax in the more arid areas where Cretaceous shale has modified the drift. 5. Agropyron-Koeleria Faciation, an edaphic climax on fine-textured lacustrine soils. The nature of these grassland communities is similar to those which have been described in adjacent areas of the United States south of the glacial boundary. The more xeric types of Canada extend southward, while the mesic types do not. Primary and secondary plant successions towards the climax and edaphic climax communities are described.