Nicholas Joshua Clark, SKM Ernest, Henry Senyondo, Juniper Simonis, Ethan P White, Glenda M Yenni, KANK Karunarathna

Forecasts of community dynamics are essential for the management of biodiversity. Theory suggests these predictions can be improved by leveraging multi-species dependencies to improve models, but empirical support for this is lacking. We test whether models that learn from multiple species, both to estimate nonlinear environmental effects and temporal dependence, improve forecasts for a semi-arid rodent community. Using Dynamic Generalized Additive Models, we analyze monthly captures for nine rodents over 25 years. We find strong evidence that multi-species dependencies improve performance, as models that captured these effects gave superior predictions. These models also provide novel insights, in our case by quantifying how changes in abundance for some species can have delayed, nonlinear effects on others while also uncovering important lagged effects of environmental drivers. We show that multivariate models are useful not only to improve ecological forecasts but also to ask targeted questions about community dynamics.