Co-variate Selection:
We selected a mix of six plausible remotely sensed and ground-based
variables that reflected characteristics of landscape, condition of
habitat, persistent anthropogenic pressures, and availability of major
food resources based on the review of available literature. For small
study area with few sample sites model loses power of explanation and
unwanted errors increase as the number of variables are increased in a
model. It is generally advised to use one variable per ten sites in
occupancy model. Thus, following the principles of parsimony, we
included three site co-variates and three sample covariates (See
Table 1 ). We selected termites, fruits and disturbance as sample
covariates and measured them in the field. Tree cover, terrain
ruggedness index (TRI) and enhanced vegetation index (EVI) were site
level covariates that were obtained from remotely sensed images.
Termites and fruits were selected as variables as they represent the
dominant food resource for sloth bears (Khanal & Thapa, 2014; Sukhadiya
et al. 2013; Bargali et al. 2012; Joshi et al.1997). Studies in India
have shown that sloth bears were positively associated with the
topographic ruggedness (Puri et al. 2015; Srivathsa et al. 2017) and
forest cover (Srivathsa et.al. 2017). Sloth bears have been reported to
avoid human and livestock disturbance (Babu et al. 2015; Puri et al.
2015) but they have also been reported from human dominated landscapes
with degraded habitats (Bargali et al. 2012). We combined human
disturbance, livestock disturbance and fire in our search trails as a
measure of disturbance. They are thought to prefer relatively dry
habitats as indicated by the negative relationship between habitat
occupancy and vegetation productivity (Srivathsa et al. 2017). We chose
EVI rather than normalized difference vegetation index (NDVI) to measure
vegetation productivity in our study as EVI has improved sensitivity. In
Nepal, it was found that sloth bears move to grasslands during the dry
season and prefer to remain in forests during the wet season (Joshi et
al. 1995). We used the tree cover data prepared by Hansen et al (2013)
as a proxy of habitat condition, higher cover indicating forested
habitat and lower cover indicating grassland habitat. Covariates were
first checked for collinearity (Figure 2), and then z
transformed before running occupancy models (Kirshna et al. 2008; Panthi
et al. 2017). We hypothesized that ceteris paribus 1) sloth bear
occupancy will increase with increasing termites, fruits and
heterogeneity in the terrain and, 2) sloth bear occupancy will decrease
with increasing tree cover, EVI and disturbance.