Objective 2: determining whether use of parturition habitat by
searching predators tracked the phenology of the birth pulse
One strategy for predators to maximize encounters with neonates is to
shift habitat use to areas with a high probability of selection by
parturient female ungulates. To this end, we constructed a resource
selection function (RSF; Manly et al. (2007)) using GPS locations
of adult female elk in the 7 days immediately following a parturition
event with landscape and vegetative characteristics hypothesized to
influence location of parturition sites in our area based on previous
research (Johnson et al. 2000, Stewart et al. 2002, Long,
Rachlow & Kie 2008). The resource selection function for elk
parturition habitat took the form\(w(x)\ \sim\ exp(\beta_{1}\times\text{canopy\ cover}\ +\ \beta_{2}\times ln(\text{distance\ to\ open\ road})+\ \beta_{3}\times ln(\text{distance\ to\ perennial\ water\ source})\ +\ \beta_{4}\times\text{ruggedness}\ +\ \beta_{5}\times\text{shrub\ cover}\ +\ \beta_{6}\times\text{forb\ cover}\ +\ \beta_{7}\times\text{slope}\ +\ \beta_{8}\times\text{aspect}\ +\ \beta_{9}\times\text{elevation}\ +\ \beta_{10}\times\text{potential\ vegetation\ type})\),
where w (x ) is the relative probability of selection for
adult female elk during the first 7 days following parturition. Percent
shrub and forb cover variables were derived from LEMMA’s generalized
nearest neighbor model (Ohmann et al. 2011); slope, aspect, and
elevation were drawn from a digital elevation model; and other
covariates are as described above. We paired each used elk location
(coded as 1) with 10 randomly generated locations representing locations
available to but not used by elk (coded as 0) and used a generalized
linear model with a binomial error distribution and logit link to model
the relative probability of selection. Given that our objective was to
create a spatially-explicit map of the study area predicting the areas
with a high probability of selection by parturient ungulates and not
make inference on the specific resources that they selected for, we did
not conduct model selection and instead used the global model for
predictions. Each 30 x 30 m pixel in the resulting predictive map
projected onto the study area represented w (x ), or the RSF
score for parturition habitat. We then calculated the mean value of the
RSF scores across all GPS locations for each carnivore species that
exhibited active search behavior (determined from the previous analysis)
on a weekly basis from 15 April to 31 to determine whether predators
shifted habitat use toward places likely to be inhabited by ungulate
neonates. We predicted that predator use of habitat used for parturition
by elk would decline later in the season if predators were attempting to
maximize encounters with neonatal elk by shifting habitat use. Weekly
mean RSF scores represented each predator’s use of predicted parturition
habitat with higher values indicating higher use of parturition habitat,
where “use” is a measure of the investment in a set of resource units
by an animal during a sampling period (Lele et al. 2013).
We used the weekly average elk parturition RSF score at carnivore GPS
relocations as the response variable in a generalized linear mixed model
and used Julian week (i.e. the number of weeks elapsing since January 1)
as a predictor to assess how carnivore use of elk parturition habitat
changed throughout the season. We included a random intercept for animal
ID to control for differences in the mean elk parturition RSF score
available within individual predator home ranges. We fit models with
both linear and quadratic effects of Julian week as predictors. We
hypothesized that if a quadratic effect of Julian week on use of
parturition habitat tracked the phenology of the birth pulse, it would
be indicative of an effort of predators to alter habitat use to maximize
encounters with neonates immediately following parturition.
Alternatively, lack of a quadratic relationship between predator use of
parturition habitat and Julian week could indicate the predator does not
spend time in areas where neonates are likely to be immediately
following parturition, or that the predator may actively hunt older,
more mobile neonates or other age classes of prey. We used likelihood
ratio tests to compare whether the quadratic effect of Julian week was
supported over a linear effect.