Near-Term Forecasting of Nonstationary Dynamic Processes

Prediction of complex dynamical systems is an important task that has implications in a wide variety of technological and policy domains.  Accurate prediction of processes such as terrestrial and space weather, climate change, financial and economic markets, and pandemics can have profound policy and engineering impacts.  As a result, diverse methods of time series prediction have been developed.  We propose a time series prediction framework based on the reservoir computing formulation of machine learning that is extended to provide accurate forecasts even in cases of nonstationarity and phase changes.