Jing Wang, Janet Sprintall, Haoyu Wang, Xiaofeng Li
Abstract The El Niño–Southern Oscillation (ENSO) predictability has notably declined since the 1998–2013 global warming hiatus, challenging conventional sea surface temperature (SST)‐based forecast models. We identify sea surface salinity (SSS) as a critical yet underappreciated driver for restoring ENSO forecast skill post‐hiatus. Through an interpretable deep learning framework (STPNet), we show that incorporating SSS sustains forecast skill above 0.8 beyond 20‐month leads, outperforming SST‐only predictions after 2014. Attribution analysis reveals that Indo‐Pacific SSS anomalies promote interbasin heat redistribution and preserve ocean memory critical for ENSO evolution. The Indonesian Throughflow functions as a salinity‐sensitive conduit, where salinity‐mediated geostrophic transport compensates for thermally induced weakening, sustaining Indo‐Pacific connectivity. These findings reveal a new thermohaline pathway influencing ENSO forecasts and show that explainable AI can boost long‐lead forecast skill while uncovering key ocean–climate interactions, offering a salinity‐informed strategy to improve future predictions under climate change.