Simple is sometimes better: a test of the transferability of species distribution models

Abstract

Abstract Species distribution models (SDMs) are often empirically developed on spatially and temporally biased samples and then applied over much larger spatial scales to test ecological hypotheses or to inform management. Underlying this approach is the assumption that the statistical relationships between species observations and environmental predictors are applicable to other locations and times. However, testing and quantifying the transferability of these models to new locations and times can be a challenge for resource managers because of the technical difficulty in obtaining species observations in new locations in a dynamic environment. Here, we apply two SDMs developed in the Mid-Atlantic Bight for Atlantic sturgeon (Acipenser oxyrhynchus oxyrhynchus) to the South Atlantic Bight and use an autonomous underwater vehicle to test model predictions. We compare Atlantic sturgeon occurrence to two SDMs: one associating sturgeon occurrence with simple seascapes and one developed through coupling occurrences with environmental predictors in a generalized additive mixed model (GAMM). Our analysis showed that the seascape model was transferable across these disparate regions; however, the complex GAMM was not. The association of the imperilled Atlantic sturgeon with simple seascapes allows managers to easily integrate this remotely sensed dynamic oceanographic product into future ecosystem-based management strategies.