The impact of human land uses on ecological systems typically depends on how extensively natural conditions are modified. However, areas of intermediate-intensity human development are difficult to detect with remote-sensing data, and are therefore rarely included in models of species distribution. Similarly, most species-distribution models do not predict how future development might affect the amount and configuration of suitable habitat. We addressed both these deficiencies by integrating  species-distribution models with a spatially-explicit housing-density model to study how a large carnivore, the puma (Puma concolor), was affected by human land uses at multiple spatial scales in the western United States (U.S.). At a local scale, we studied the response of marked pumas in southern California to natural and anthropogenic features and evaluated if mortality risk varied due to the intensity of human development. At a ecoregional scale, we developed puma distribution models for the western U.S. using local spatial-regression models and harvest records collated across game-management units. At both scales, our housing-density model allowed us to examine changes in the amount and distribution of puma habitat historically (1970-2000) and in the future (2000 -2030). We also predicted areas of the western U.S. that may become future hotspots for puma-human conflict.

Results/Conclusions

We found pumas in southern California occupied much of the remaining natural landscape but rarely incorporated suburban or urban development into their home ranges. Pumas incorporated rural and exurban development into their home ranges, apparently perceiving these areas as modified habitat, rather than non-habitat. However, mortality risk increased with use of exurban areas, suggesting that exurban areas may be hotspots for puma-human conflict. At the biogeographic-scale, we found that a geographically-weighted regression (GWR) model depicted the distribution of pumas substantially better than a global, non-spatial model (AIC_cΔ = 161). This resulted from non-stationarity in puma response to some covariates across the western U.S., including the positive association with precipitation that occurred in arid regions but not mesic regions. Harvest records provided a reasonable index of the relative abundance of pumas across the western U.S., as indicated by the amount of residual variation in our model that was associated with human access (R² = 0.16). Our results suggest that local spatial-regression methods like GWR are advantageous for broad-scale species-distribution models. We identified over 15,000 km² in the western U.S. where puma-human conflict may increase in coming decades due to development. These hotspots included large urban corridors and less intensively developed regions.