Modelling nonlinear responses of a desert rodent species to environmental change with hierarchical dynamic generalized additive models

Modelling abundance fluctuations of species is a crucial first step for understanding and forecasting system dynamics under future conditions. But, especially in multivariate response data, this can be hampered by characteristics of the study system such as unknown complexity, differently formed spatial and temporal dependency, non-linear relationships, and observation characteristics such as zero-inflation. This study aimed to explore how such challenges can be addressed by using hierarchical Dynamic Generalized Additive Models (DGAM) for multivariate count responses in a Bayesian framework while modelling multi-site monthly captures for the Desert Pocket Mouse (Chaetodipus penicillatus) over 23 years from a long-term study in Arizona, USA. By fitting models of increasing complexity and developing bespoke checking functions that captured targeted ecological aspects such as spatio-temporal dependence, we show how nonlinear dynamic models can be built to improve forecasts for multivariate count-valued time series.