Advancing forecasts with ecological theory

Frank Pennekamp – University of Zurich

Data-driven forecasting approaches such as machine learning have become state-of-the-art across many fields, including ecology. This is due to their ability to forecast complex dynamics even in the absence of understanding the underlying processes. In many circumstances, some process knowledge is available and including such theoretical understanding may allow for even better forecasts. Improved forecasts following the inclusion of a specific process in the forecast model represents evidence in favour of the underlying theory. Forecasting therefore is a natural benchmark for testing ecological theory. Dr. Pennekamp will use two case studies to illustrate how ecological theory can advance forecasting. First, he will show how to constrain a data-driven approach with the Metabolic theory of Ecology (MTE) to yield better forecasts in the face of temperature variation. Second, he will show how a tri-trophic food web model with flexible, data-driven functional relationships results in higher forecast skill and yields new insights into the underlying processes. Merging data-driven and process-based models hence provides exciting opportunities to increase both understanding as well as improved forecasts of ecological systems.