Empirical dynamics for ecosystem forecastingMathematical Biology
|Speaker:||Stephan Munch, NOAA Fisheries and UC Santa Cruz|
|Start time:||Mon, Jan 30 2023, 4:10PM|
Ecosystems are complex, containing hundreds of species with evolving traits and context dependent interactions. Because of this complexity, models are critical in guiding decision making in ecosystem management. At the same time, ecosystems are sparsely observed – we rarely have data on all of the relevant species, making it difficult to develop and validate ecosystem models. Empirical dynamic modeling (EDM) helps circumvent this difficulty by using time-delay embedding to compensate for unobserved species and other state variables. In many cases EDM has improved predictions compared to mechanistic models. But climate change poses challenges to EDM and other data-driven approaches since they rely on past behavior to predict the future. Here I will discuss several recent extensions of EDM incorporating minimal mechanistic constraints that broaden the time horizon over which we can make useful predictions and potentially serve as early warning signals of critical transitions.