The Importance of Subsurface Thermal Effects to Indian Ocean Dipole Prediction Revealed by a Skillful Three-Dimensional Deep Learning Model

The Indian Ocean Dipole (IOD) is a dominant mode of interannual variability in the Indian Ocean with significant influences globally on weather, climate, and society. Despite extensive research, current numerical models and surface-based deep learning models are limited in their ability to predict IOD events beyond five to seven months. To address this gap, we introduce the novel Multi-dimensional Air–Sea Coupled Deep Learning Model (MAS-Net), which integrates three-dimensional multivariable data to deliver accurate predictions of the ocean field and the Dipole Mode Index (DMI) up to eight months in advance. Sensitivity and interpretability analyses reveal that subsurface temperature anomalies are critical in improving IOD prediction skill. By integrating subsurface dynamics into a deep learning framework, this work hence significantly advances understanding of IOD predictability and provides a robust tool for both operational forecasting and scientific exploration of the IOD. The MAS-Net framework sets a new standard for predictive modeling, offering a versatile approach that can be extended to other climate phenomena.