Abstract

The forecasting of high-dimensional, spatiotemporal nonlinear systems has made tremendous progress with the advent of model-free machine learning techniques [1,2]. Particularly, the forecasting of extreme events (EE), which are rare and intense phenomena arising in a lot of physical and natural complex systems [3], can have great impact in these fields. Unfortunately, a precise forecasting often relies on the precise knowledge of information on the past state of the system which is not always possible to achieve in real systems: only partial information is available for learning and forecasting in some cases. This can be due to poor temporal or spatial samplings, to physical variables difficult to measure or to access, or to noise.

© 2023 IEEE