Map generation and landscape analysis
We used a geotiff layer depicting land-use and land cover (LULC) of Mato Grosso in 2015 made available by the MapBiomas platform (Project MapBiomas - Collection 5.0 of Brazilian Land Cover & Use Map Series , 2020) to create the land cover map of our study landscape (Fig. 2). We modified the original LULC map by reclassifying land cover classes to reduce the number of classes to forest, savannah, cropland (primarily soybean), pasture, open-water, and urban (the savannah land cover class was excluded in subsequent analyses as the total area of this habitat type was minimal). We primarily used the raster (Hijmans, 2019),rgeos (Bivand & Rundel, 2020), and sp (Bivand et al., 2013; Pebesma & Bivand, 2005) packages in R for reading and manipulating the spatial data.
From this LULC map, we quantified the deforestation surrounding each site by measuring the percentage cover of forest, pasture, and cropland (using the landscapemetrics (Hesselbarth et al., 2019) package in R) within a 2500 m circular buffer (as shown on the map in Fig. 2) of the trapping grid at each site as the centroid. We selected a 2500 m buffer to capture the effects of deforestation and encroachment of non-forest cover surrounding each forest patch while also reducing the overlap between neighboring buffer windows. Proportions of forest and cropland cover within a 2500 m circular buffer around each trapping grid were highly correlated (r = -0.97) so we included only forest cover in evaluating the effects of deforestation in the study region.