Our objective was to predict the response of Chihuahuan desert landscapes to climate change using a multiscale approach that includes vegetation-soil feedbacks, seed source and seed dispersal across the landscape. We used vegetation, soils, and climate maps for the Sevilleta National Wildlife Refuge LTER site (SNWR) in central New Mexico as input parameters in a spatially explicit, individual, plant-based simulation model (ECOTONE). The model incorporates feedbacks among vegetation and soil properties and processes that are important to perennial grass establishment and survival. We used the model to predict the locations most likely to shift from shrub- to grass-dominated vegetation due to changes in climate. Our results showed an increase in perennial grass cover throughout the SNWR. Sensitivity of shrub-dominated locations was dependent upon soil texture and distance to the nearest perennial grass seed source. Feedbacks between changes in species composition and soil organic matter modified the rate of change in dominant species. Our results demonstrate the importance of landscape-scale processes and vegetation-soil feedbacks to shifts in vegetation at ecotones.