Search

Accurate short-term climate predictions, specifically the prediction of dominant rain belts during flood seasons, represent a significant challenge, particularly in the absence of notable external forcing signal anomalies in the previous winter. To address this challenge, we propose a novel climate index based on an Auto-Encoder (AE) Neural Network that exhibits superior performance in indicating several significant precipitation events in the Yangtze and Huaihe River Basins under varying El Niño conditions. Notably, the novel climate index rectifies the discrepancy between the 2019/2020 Niño index and the 2020 summer precipitation. In order to understand the information source of AE, the causality between AE and other indices was further analyzed by Granger Causality. The results show that AE index mainly contains the information of subtropical high index and ocean index, which is consistent with previous studies. • AE obtained by Auto-encoder can make up for the "failure" of Nino index in abnormally heavy precipitation events. • In terms of magnitude and state, AE has better description of precipitation than Nino3.4. • The main indices of AE information are subtropical high-pressure index and sea surface temperature index. [ABSTRACT FROM AUTHOR]