Integrated models that unite local and regional data reveal larger-scale environmental relationships and improve predictions of species distributions

Abstract

The scale of environmental relationships is often inferred through the use of species distribution models. Yet such models are frequently developed at two distinct scales. Coarse-scale models typically use information-poor (e.g., presence-only) data to predict relative distributions across geographic ranges, whereas fine-scale models often use richer information (e.g., presence–absence data) to predict distributions at local to landscape scales. We unite presence–absence and presence-only data to predict occurrence of species, what we refer to as integrated distribution models. We determine if integrated models improve predictions of species distributions and identification of characteristic spatial scales of environmental relationships relative to presence–absence modeling and ensemble modeling that averages predictions from separate presence-only and presence–absence models. We apply recent advances in integrated distribution models to predict Sherman’s fox squirrel (Sciurus niger shermani) distribution in north-central Florida. Presence-only data were collected through a citizen-science program across its geographic range, while presence–absence data were collected using camera trapping surveys across 40 landscapes. Integrated models estimated environmental relationships with greater precision and identified larger characteristic scales for environmental relationships than using presence–absence data alone. In addition, integrated models tended to have greater predictive performance, which was more robust to the amount of presence–absence and presence-only data used in modeling, than presence–absence and ensemble models. Integrated distribution models hold much potential for improving our understanding of environmental relationships, the scales at which environmental relationships operate, and providing more accurate predictions of species distributions. Many avenues exist for further advancement of these modeling approaches.