Modelling species distributions using remote sensing predictors: Comparing Dynamic Habitat Index and LULC

Abstract: This study compares the predictive capacity of the Dynamic Habitat Index (DHI) - a remote sensing (RS)-based measure of habitat productivity and variability - against traditional land-use/land-cover (LULC) metrics in species distribution modelling (SDM) applications. RS and LULC-based SDMs were built using distribution data for eleven bird, amphibian, and mammal species in ^Ile-de-France. Predictor variables were derived from Sentinel-2 RS data and LULC classifications, with the latter incorporating Euclidean distance to habitat types. Ensemble SDMs were built using nine algorithms and evaluated with the Continuous Boyce Index (CBI) and a calibrated AUC. Habitat suitability scores and their binary transformations were assessed using niche overlap indices (Schoener, Warren, and Spearman rank correlation coefficient). Both RS and LULC approaches exhibited similar predictive accuracy overall. After binarisation however, the resulting niche maps diverged significantly. While LULC-based models exhibited spatial constraints (habitat suitability decreased as distance from recorded occurrences increased), RS-based models, which used continuous data, were not affected by geographic bias or distance effects. These results underscore the need to account for spatial biases in LULC-based SDMs. The DHI may offer a more spatially neutral alternative, making it a promising predictor for modelling species niches at regional scales.