Grassland Cover Quantification from Cropland Data Layer

We obtained land cover data for our study areas from the Cropland Data Layer for 2006–2017 [26 ]. The Cropland Data Layer combines National Land Cover Database [31 ] classifications with information on specific crop types at a 30-m spatial resolution and is ground-truthed with overall accuracy ranging from 76.5% - 92.4% for agricultural and 78% - 86.4% for nonagricultural classes [26 , 31 ]. We defined “grassland” as classes comprised of clover/wildflowers, switchgrass, grass/pasture, alfalfa, or other hay/non alfalfa in the Cropland Data Layer. We calculated the percent grassland cover at two scales around survey locations (local scale: 250-m buffer around survey points [~20 ha], landscape scale: 2,500-m buffer around survey routes [~21563 ha]). These scales represent local- and landscape-scale habitat features with respect to our focal species’ home ranges [29 ] and are similar to scales used in other multi-scale habitat analyses [32 (link)–34 ]. Because these data were not available for the period 2001–2005, we used linear regression to develop a predictive model of grassland cover at both scales around points from existing data and extrapolated values for years with missing data.
Among all years, focal area sites had a mean of 54.0% grassland cover at the local (250-m buffer) scale and a mean of 51.9% grassland cover at the landscape (2,500-m buffer) scale (95% of all stops/routes between 5.4% - 92.2% and 32.7% - 73.7%, respectively). Among all years, paired area sites had a mean of 50.1% grassland cover at the local scale and a mean of 50.0% grassland cover at the landscape scale (95% of all stops/routes between 3.6% - 92.5% and 29.4% - 72.0%, respectively). Percent grassland was generally similar among focal and paired site pairings (S1 Fig).