Context: Habitat loss and fragmentation are the most pressing threats to biodiversity, yet assessing their impacts across broad landscapes is challenging. Information on habitat suitability is sometimes available in the form of a resource selection function model developed from a different geographical area, but its applicability is unknown until tested. Objectives: We used the Mexican spotted owl as a case study to demonstrate how models developed from different geographic areas affect our predictions for habitat suitability, landscape resistance, and connectivity. We identified the most suitable habitats and core areas for dispersal and movement for the species. Methods We applied two multi-scale habitat selection models - a local model and a non-local model - to a broad study area in northern Arizona. We converted the models into landscape resistance surfaces and used simulations to model connectivity corridors for the species, and created composite habitat and connectivity models by averaging the local and non-local models. Results; While the local and the non-local models both performed well, the local model performed best in the part of the study area where it was built, but performed worse in areas that are beyond the extent of the data used to train it. The composite habitat model improved performances over both models in most cases. Conclusions: With rigorous testing, multi-scale habitat selection models built on empirical data from other geographical areas can be useful. Averaging predictions of multiple models can improve performance, but the effectiveness is subject to the performance of the reference models.