Identifying Nonstationarity in Ecological Time Series using Empirical Dynamical Modeling
Physical and Biological Sciences
Koret Scholarship and Independent Study
Climate change, invasive species and contemporary evolution are all factors that cause ongoing change in ecosystems, leading to the question: how can you effectively model an ecosystem when its dynamics are changing under your feet? Ecosystems undergoing changes like this are dubbed “nonstationary”, and pose a major problem for modeling and forecasting efforts. One powerful tool for nonparametric, data-driven modeling of ecosystems is SMap(a member of the suite of ecological tools Empirical Dynamical Modeling), however it assumes stationarity. We propose an extension to SMap called NSMap which gives improved predictions for nonstationary ecosystems and provides a measure of the degree of nonstationarity. The method is demonstrated on simulated data, including a system which undergoes regime shifts in dynamics during the time series, and a real time series of North Sea plankton that previous work has determined to be effected by climate change.
