Predicted Calving and Post-calving Season Resource Use of the Porcupine Caribou Herd During 2012-2018 With Future Projections for the 2030s, 2040s, and 2050s

This dataset contains rasters and polygon shapefiles related to predicted resource use of the Porcupine Caribou Herd (PCH) during the calving (26 May-10 June) and post-calving (11-30 June) seasons in Alaska and the Yukon Territory. Resource selection was analyzed for each season using random forest models, which compared female caribou GPS collar locations (2012-2018) to available locations within the study area. The models assessed the influence of annual variation in spring phenology (dates of snowmelt, onset of vegetation greenness, and the 50% maximum NDVI value) on caribou resource use, while also accounting for static landcover and topographic variables. For each year GPS collar data were collected (2012-2018), we provide rasters of predicted relative probabilities of use (ranging from 0 to 1) during the calving and post-calving seasons given year-specific phenological conditions. We also provide annual polygons depicting "suitable" habitat, which were delineated from the rasters based on probability of use thresholds that minimized the mean class error (i.e., the average of the used and available location error rates; calving threshold: 0.534, post-calving threshold: 0.522). In addition to annual rasters and polygons, we provide an average raster and polygon across the study period (2012-2018) for each season.
We then projected future PCH calving and post-calving resource use based on predicted climate-driven shifts in spring phenology. We first used random forest regression to model phenology dates as a function of temperature, precipitation, and topographic covariates. We then obtained decadal averages of climate projections from the Coupled Model Inter-Comparison Project 5 for Representative Concentration Pathway 8.5 for the 2030s, 2040s, and 2050s. We used the climate projections to predict future dates of spring snowmelt and vegetation phenology across the study area, which were then applied to our resource selection models to project future PCH calving and post-calving resource use. For each decade in the 2030s, 2040s and 2050s, we provide a raster of the average predicted probability of use and a polygon depicting average suitable habitat. Decadal averages were used to reduce annual variation in spatial predictions so trends in suitable habitat could be more easily compared across time periods. Probabilities of use could not be estimated for pixels with missing phenology data, and thus, those pixels are not assigned probability values.
We also provide a shapefile of the study area boundary, which defined habitat availability for resource selection models and the extent of our spatial predictions. The study area was delineated from a 95% contour of an annually-weighted kernel density estimate of caribou locations during the combined calving and post-calving seasons and buffered by the mean daily caribou movement rate in the location dataset (13 km).