Ecosystems are rapidly changing, making predictions for near-term future dynamics crucial for decision making by land managers and policy makers. Producing actionable ecological forecasts requires understanding how ecosystems will respond to changing conditions. Complex interactions among species and their environment, however, may limit the degree to which past dynamics inform the future under changing conditions. It is therefore critical to develop and assess our ability to forecast populations in novel environments.
The Portal Project has been collecting data on rodents and plants as part of a long-term experiment in southeastern Arizona for over four decades. During this time, the climate at the Portal Project has become warmer and drier, and the vegetation has gotten more dense, creating novel environmental conditions for species. Ongoing observations all us to assess how novel conditions impact our ability to forecast populations and identify the best methods to forecast changes in ecological systems.
Here, we focus on modeling and forecasting population dynamics of Merriam’s kangaroo rat (Dipodomys merriami), and built a suite of five state-space models of increasing ecological complexity in the process model: [1] random walk, [2] logistic population, [3] logistic population with environmental covariates (warm precipitation on growth rate, greenness on carrying capacity), [4] logistic population with competition covariates (population size of Dipodomys ordii on carrying capacity), and [5] logistic population with environmental and competition covariates.
Fitted parameter estimates from the models indicates that D. merriami growth rate is positively related to warm precipitation and population size is positively related to greenness and negatively related to D. ordii population size. Increasing ecological complexity in the process model improved forecast scores in both single-step and rolling-origin evaluations, indicating the importance of incorporating changing ecological and environmental conditions into predictions of population dynamics. We have incorporated these models into a iterative near-term ecological forecasting workflow, facilitating their ongoing evaluation alongside existing, phenomenological models.
