Species distribution models are widely used to forecast species responses to environmental change and inform conservation. Ecologists have repeatedly called for greater investment in approaches that explicitly model processes such as physiology, dispersal, demography and biotic interactions that underpin species range dynamics. A wide array of these ‘process-explicit’ models are now available, but it is not clear which methods are best for which types of problems or scenarios. In addition, data to parameterize process-explicit models are often limited, which could reduce their ability to accurately forecast distributions. Simulation studies that test model performance under a range of scenarios and with different data availability can offer important insight into these questions. However, simulating data in a way that adequately mimics the processes driving range dynamics can be challenging. We developed novel integrated models that simulate responses of real species to environmental change by linking eco-physiological models with individual-based models. We coupled these with a sampling module that records data required to fit different model classes under different sampling regimes, assuming a fixed budget. Using this simulation system – initially built for koalas – we then tested how well different process-explicit distribution model classes (occupancy dynamics models, coupled SDM-population models, demographic distribution models) capture and predict range dynamics.
Results/Conclusions
Our integrated models captured historical changes in the distribution of koalas in response to climatic extremes (heatwaves and drought) and disturbances (fires), providing a realistic system with which to simulate individual and population-level responses over time. Performance of process-explicit models depended on the amount of data (i.e. the budget) as well as how sampling was allocated through space and time. This was particularly true for methods that rely on relatively expensive demographic data, where a limited number of sites could be sampled even for large budgets. This simulation approach is useful for gaining insight into what drives performance of different process-explicit methods, mechanisms that they do and do not capture and how to best allocate efforts to improve the reliability of range dynamics predictions under environmental change.