Covariates
All covariates were extracted in ArcGIS (ESRI 2014). We used a Norwegian and a Swedish vegetation map merged together to account for the cameras along the border (Northern Research Institute’s vegetation map, Norway, 30 x 30 m resolution merged with Swedish Corine land cover map Lantmäteriet, Sweden, 25 x 25 m resolution into a 25 x 25-m resolution raster, Mattisson et al., 2013, Ordiz et al., 2015), from which we extracted the proportion of agricultural land. Additionally, we obtained human density (inhabitants/km2) from Statistics Norway as a 250 m resolution raster (www.ssb.no). However, this raster only contained data for urban areas. Given that the cameras were placed at different distances from those areas, we used that data to calculate a proxy for the direct human disturbance at the camera sites. For this, we transformed the raster to a point layer, which we used to predict a planar kernel density map set with bandwidth = 2,000 and cell size = 200 m. We then extracted these two variables (proportion of agricultural land and human disturbance) to a 1-km radius circular buffer around each camera trap.
In order to assess primary productivity, we downloaded the 1-km resolution monthly EVI from the MOD13A3 V061 product (Didan, 2021) using the AppEEARS application (AppEEARS-Team, 2022). We downloaded monthly data for years 2016-2020 and deleted 11 raster cells because of low quality and missing data. We then merged the monthly EVI map layers into one final raster containing average value of overlapping cells. We masked out water to include only terrestrial cells and calculated the mean-EVI for a 10-km radius circular buffer around each camera trap. We used a 10-km buffer for primary productivity in order to focus on the wider region rather than on the micro camera site. Using an environmental productivity variable as a covariate such as EVI can help to correct for biases when considering multiple sites (Hofmeester et al., 2019).
We divided the camera trap data into two six-month periods, from October to March (winter; roughly the months with snow cover) and from April to September (summer; snow free months in most of the study area). We also created an 8-level “study area” covariate based on geographic clusters of camera traps (Figure 1).