DATA-DRIVEN ANALYSIS OF NONLINEAR DYNAMICS BASED ON KOOPMAN OPERATOR THEORY AND ITS APPLICATION TO SHORT-TERM PRECIPITATION FORECAST

Abstract

 The data-driven approach has become a major scheme in the field of civil engineering, but as the properties of a computation model it is also required interpretability in computation process and integration with prior knowledge on natural phenomena. This research focuses on a methodology of Koopman operator analysis as a scheme of integrated data-driven and model-driven approach, and shows its effectiveness through numerical experiments. Further, the methodology is applied to a problem of short-time precipitation forecast. Considering a transient characteristic of rainfall distribution, a novel method to decompose its dynamic behavior into global advection and spatial mode is proposed and verified.