Predicting El Niño-Southern Oscillation (ENSO) events remains challenging, especially those after 2000. This study introduces a deep learning model (STPNet) that incorporates sea surface salinity (SSS) along with sea surface temperature (SST) to improve ENSO forecast skill. STPNet extends the effective ENSO forecast lead time to 24 months for 2000–2021, mitigating the spring predictability barrier. Interpretable methods reveal SST's importance for short-medium lead forecasts and SSS's role in medium-long lead forecasts. The model highlights the significance of ocean inter-basin and tropics-extratropics interactions in ENSO development.

Haoyu Wang, Shineng Hu, Cong Guan, Xiaofeng Li