to riparian habitats and high local cover (>80%) by shrubs
(Fig. 2d and c). The wet season model had high discrimination ability
(AUC = 0.944) (Supplementary Fig. S4a), and it did not show any
problems resulting from collinearity (VIFs < 3), outliers or extreme
values and deviations from linearity assumptions in the logit scale.
Although there was high spatial autocorrelation in the raw binary
variable, this was no longer observed in the correlogram of average
model residuals (Supplementary Fig. S1) and in the Mantel test
(p = 0.17). Temporal correlogram suggested that residuals were
correlated for resting sites monitored in the same day, but correlations
declined to close to zero afterwards (Supplementary Fig. S3a).
There were also significant correlations in the residuals for time
lags of 13 and 26 days.