Posters 1

[Poster 1-03] An AI-Driven Operational ENSO Forecasting Framework for Climate-Resilient Digital Twins

*Kalpesh R. Patil¹ (1. Application Laboratory (APL), Research Institute for Earth and Information Sciences (REIS) (Japan))

We present an AI-enabled operational forecasting system for the El Niño–Southern Oscillation (ENSO), representing a significant step toward an ocean digital twin capable of delivering seasonal ENSO forecasts from 2-months to 2-years in advance. The system employs seasonally optimized convolutional neural networks (CNNs) trained on long-term ocean reanalysis datasets using a customized loss function designed to improve the prediction of extreme ENSO events. Extensive validation demonstrates that the AI system consistently outperforms state-of-the-art dynamical prediction models, particularly at long lead times, while remaining consistent with leading statistical forecasting systems. Motivated by these results, we transitioned the framework into an operational forecasting system two years ago, issuing updated Niño-3.4 index forecasts on the 20th of every month for the upcoming seasons and extending up to two years ahead. To further enhance forecast robustness, the operational framework now incorporates a multi-source sea surface temperature (SST) ensemble. Recent forecasts, initialized from the latest seasonal conditions, consistently indicate the development of extreme El Niño conditions during the coming boreal winter, in agreement with forecasts from leading international climate centers. This operational AI forecasting capability provides an important predictive component for resilient climate and ocean digital twins, supporting climate services, early warning, risk-informed decision making, and evidence-based policy for climate resilience.