Scale And Landscape Heterogeneity Influence The Predictive Performance of Habitat Suitability Models

Context: Resource selection functions (RSF) are used to predict habitat selection in a wide range of taxa for various conservation and management purposes. Although such predictions often cover large geographic areas, the limitations of extrapolating beyond the original study region are rarely addressed. Objectives: Our first research objective was to demonstrate the implications of extrapolating RSF predictions across different habitats and at different spatial scales, with regard to varying landscape characteristics. For our second objective, we investigated the impact of individual variation in resource selection. Methods: We used a long-term dataset of GPS-collared lynx from two separate regions in Sweden. We built individual-level RSF models at two spatial scales for two different study regions to quantify individual and regional variation in habitat selection, and extrapolated the results from each region across the opposing study area. Results: Individual lynx selected resources differently within and between study regions, and predictions were more accurate within home ranges than between home ranges. When extrapolating across variable landscapes, encountering resource values outside the fitted range of the models led to incorrect predictions. Conclusions: Our case study highlights the importance of quantifying a variable’s value range in both the model and extrapolation area, taking into account variation in individual resource selection, and understanding the relationship with the order of selection to improve the reliability of habitat predictions. To increase the transferability of models, extrapolations should ideally be limited to areas with overlapping value ranges, to reduce the risk of misidentifying habitat suitability.