Density‐dependent growth and dispersal can accurately forecast near‐term range shifts in a dominant dryland tree species

Abstract

Forecasting how species will shift their distribution and abundance in response to global change is a pressing challenge facing ecologists. Over broad scales, extrinsic environmental factors (e.g. climate) are often recognized as the primary driver of species range limits. Yet, range limits are the culmination of a complex set of scale‐dependent mechanisms that ultimately drive a population to shift in space, and the degree to which each of these factors must be captured to accurately forecast near‐term species range shifts is unclear. Using a hierarchical Bayesian spatiotemporal modelling approach, we tested the extent to which external drivers (climate and topography) and intrinsic population dynamics (density‐dependent growth and dispersal) could predict observed species range expansions in one of the most widespread vegetation types in the western US, Pinyon‐Juniper woodlands. We built and trained a hierarchical Bayesian spatiotemporal model using 31 years of remotely sensed tree cover data along a historically expanding range margin. We tested a suite of models with varying environmental covariates and evaluated forecast performance on a 5‐year holdout period. We also evaluated model transferability and forecast performance in new locations. We found that the addition of climatic and topographic covariates to our base population model did not result in higher forecast accuracy. In sample, all models resulted in normalized root mean square error (NRMSE) of 0.1, for a 5‐year holdout period. Additionally, the base model emerged with the highest forecast accuracy in new locations, and performance was markedly similar to the original, in sample location, by the last 5 years of a 35‐year holdout period (NRMSE 0.17–0.19). Synthesis. We found that the inclusion of external drivers such as climate conditions or topography generally did not improve forecast accuracy and that at multidecadal time scales, intrinsic population processes (density‐dependent growth and dispersal dynamics) can accurately predict shifting abundances along a historical range margin. Our results suggest that accurate near‐term forecasts of changing plant distributions and abundances may be possible using comparatively simple ecological models.