Deep learning model predicts Southern Indian Ocean Dipole seven months in advance

The South Indian Ocean Dipole (SIOD), a seesaw of warm and cold water across the southern Indian Ocean, can reach beyond the equator, modulating the East Asian monsoon and influencing rainfall in China. Traditional climate models have traditionally been able to predict SIOD only about two to three months in advance.

Recently, a Chinese research team used deep learning to extend that to seven months, and their results were published in a journal. Atmospheric and Oceanic Science Letters. They developed a deep learning model that uses sea surface temperature and ocean heat content anomalies at depths of 300 meters as input. By combining multi-step time convolution with an attention mechanism, the model automatically learns important features of how ocean temperature changes. The model successfully extends the effective prediction of the SIOD peak season (January to March) up to 7 months in advance, outperforming several traditional dynamic prediction systems.

“We found that the model relied on different key signals depending on the forecast lead time,” explains corresponding author Dr. Meng Xu. For short-term forecasts, the model focuses on local air-sea interaction signals in the southern Indian Ocean, such as feedbacks between wind fields and sea surface temperatures. For long-term forecasts, the focus shifts to the central-eastern Pacific near the equator, which is closely associated with El Niño-Southern Oscillation (ENSO). This means that the model captures changes in physical factors. Short-term forecasts rely on local ocean memory, while long-term forecasts exploit remote ENSO teleconnections via atmospheric bridges.

This study also revealed a significant asymmetry between positive and negative SIOD events. For positive SIOD events, long-term forecasts are mainly related to La Niña. In the case of negative SIOD events, besides El Niño, there is an additional intermediate-term signal from the South Atlantic, which can stimulate eastward-propagating atmospheric Rossby waves and ultimately affect the southern Indian Ocean.

“This study not only shows the great potential of deep learning in climate prediction, but more importantly, by using attention analysis and sensitivity experiments, we can understand the physical basis behind the model’s predictions,” says Dr. Meng Xu.

This research provides a new avenue for combining artificial intelligence and physical processes to further improve climate prediction capabilities.