Echo State Network based on Phase Space Reconstruction: El Niño 3 Index Forecasting

Climate forecasting, especially tor the abnormal weather like El Nino and La Nina etc., is one of the most difficult tasks of time series prediction domain, because of its inherent difficulty of long-term forecasting. In this paper, we investigate whether there is a simple pattern to describe El Niño phenomenon that can be reconstructed by mathematical models and a multiple steps ahead predicting method is proposed, which is an improved echo state network (ESN) using the phase-space reconstruction (PSR). PSR is to embed a one-dimensional time series sequence in a high-dimension phase space as the input of the ESN. Specifically, the method PSR-ESN can capture seasonal characteristics of the climate data and reduce the lag of prediction as much as possible. The method is evaluated on a real Niño 3 SST index data and it outperforms the other state-of-the-art baselines from the perspective of the magnitude and the lag.