Evaluating the influence of neighborhood connectivity and habitat effects in dynamic occupancy species distribution models

Abstract

Exploring new approaches and methodologies to characterize species distribution dynamics, instead of solely relying on static spatial patterns, should be a priority for species distribution modelling research. Dynamic occupancy models (here, ‘dynocc models') are a promising tool to capture temporal patterns of distribution change but their spatial accuracy has been shown to be limited. In this study, we evaluated the effectiveness of incorporating neighborhood connectivity effects into the colonization and extinction functions of dynocc models. To accomplish this, we compared dynocc models accounting either for neighborhood connectivity only, for site-level habitat covariates only, or combining both neighborhood and habitat explanations in the models for species extinction and colonization. All models were evaluated for a total of 46 bird species typical of forests and shrublands using data at 1 km2 scale from two Catalan breeding bird atlases (CBBA2: 1999–2002 and CBBA3: 2015–2018). Models' predictive performance varied across species between dynocc models incorporating habitat covariates alone and those considering neighborhood connectivity alone. Among species, 68% exhibited a predominant response to habitat effects, 24% showed similar responses for habitat and connectivity effects, and 9% were mostly associated with connectivity effects. Dynocc models combining connectivity and habitat covariates achieved the best predictive performance for most species, with bigger gains for species with similar results from habitat-only and connectivity-only models. However, relative performance gains compared to dynocc models using only habitat or connectivity variables were generally modest for most species. This study shows the benefits of considering more spatially explicit formulations in dynocc models, specifically incorporating neighborhood connectivity into the extinction and colonization processes. Our work also highlights the importance of evaluating different model formulations and assessing which aspects of the model are more important depending on the study species.