Resource Selection Functions
To determine individual-level habitat selection, we used a logistic
regression to estimate resource selection functions (RSFs) for each
individual within the annual home ranges (Manly et al. 2002,
Bastille-Rousseau and Wittemyer 2019). The transient individuals that
were previously excluded from home range size analysis were included in
the RSFs, using KDE annual home range estimations (instead of AKDE).
Twelve thousand random locations were generated within each of these
home ranges. Each random location was also assigned a random date and
time (Bastille-Rousseau et al. 2015), and the previously described
temporal periods were applied to each used and random point so that each
point was categorized based on its season and diel period.
Bobcat and coyote RSFs were estimated using the package ‘IndRSA’ in
Program R (Bastille-Rousseau and Wittemyer 2019). ‘IndRSA’ estimates an
individual-level RSF for each individual and a population average in a
second step (Murtaugh 2007). K-fold cross-validations were performed for
each output, and those with a k-fold value less than 0.2 were excluded
from the results. Landcover categories, human modification, distance to
water, and distance to road covariates were extracted for each used and
random point. Landcover categories included the dummy variables of
forested (reference category), agricultural, exurban, and other. The
continuous variables of human modification, distance to water, and
distance to road covariates were scaled so they could be compared to
ease interpretation (Schielzeth 2010). Models for each permutation of
species and temporal period were estimated in this manner.