Matching the forecast horizon with the relevant spatial and temporal processes and data sources
Peter B. Adler, Ethan P. White, Michael H. Cortez
Abstract
Most phenomenological, statistical models used to generate ecological forecasts take either a time‐series approach, based on long‐term data from one location, or a space‐for‐time approach, based on data describing spatial patterns across environmental gradients. However, the magnitude and even the sign of environment–response relationships detected using these two approaches often differs, leading to contrasting predictions about responses to future environmental change. Here we consider how the forecast horizon determines whether more accurate predictions come from the time‐series approach, the space‐for‐time approach or a combination of the two. As proof of concept, we use simulated case studies to show that forecasts for short and long forecast horizons need to focus on different ecological processes, which are reflected in different kinds of data. First, we simulated population or community dynamics under stationary temperature using two simple, mechanistic models. Second, we fit statistical models to the simulated data using a time‐series approach, a space‐for‐time approach or a weighted average. We then forecast the response to a temperature increase using the statistical models, and compared these forecasts to temperature effects simulated by the mechanistic models. We found that the time‐series approach made accurate short‐term predictions because it captured initial conditions and effects of fast processes such as birth and death. The space‐for‐time approach made more accurate long‐term predictions because it better captured the influence of slower processes such as evolutionary and ecological selection. The weighted average made accurate predictions at all time scales, including intermediate time‐scales where the other two approaches performed poorly. A weighted average of time‐series and space‐for‐time approaches shows promise, but making this weighted model operational will require new research to predict the rate at which slow processes begin to influence dynamics.