Identifying State-Dependent Precursors to Improve ENSO Forecasting Using a Hybrid Model-Analog and Deep Learning Approach

In this work, we introduce a hybrid method that integrates deep learning with model-analog forecasting, a straightforward yet effective approach that generates forecasts from similar climate states in a repository of model simulations. The developed hybrid framework employs a fully convolutional network to estimate state-dependent weights for identifying analog states. The advantage of our method lies in its inherent transparency, offering insights into important regions through estimated weights and in the temporal evolution of the system through analog forecasting. We evaluate our approach using the Community Earth System Model Version 2 Large Ensemble to forecast the El Niño Southern Oscillation on a seasonal-to-annual time scale. We partition the 100-ensemble dataset into training (1865–1958; 70%), validation (1959–1985; 20%), and test (1986–1998; 10%) subsets. Results show a 10% improvement in forecasting sea surface temperature over the equatorial Pacific at a 12-month lead time compared to the traditional model-analog technique. Furthermore, our trained model demonstrates a 3.5% improvement when evaluated against a reanalysis dataset. This evaluation specifically uses a fair-sliding anomaly approach that refrains from utilizing future data not available at the time of the forecast. Our deep learning-based approach reveals state-dependent precursors linked to various physical processes, including the Pacific Meridional Mode and equatorial thermocline depth and zonal slope. Notably, disparities emerge in the precursors associated with El Niño and La Niña events. For instance, we find that sea surface temperature over the tropical Pacific plays a more crucial role in El Niño forecasting, while zonal wind stress over the same region exhibits greater significance in La Niña prediction. This approach has broad implications for forecasting diverse climate phenomena, including those that are challenging for traditional model-analog forecasting methods.