Korean researchers develop AI model to predict El Niño, La Niña years in advance

South Korean researchers have developed an AI-based ocean prediction model capable of improving the accuracy of climate forecasts for phenomena such as El Niño and La Niña months to years in advance. The technology is expected to help the agriculture, fisheries, energy and disaster management sectors prepare earlier for extreme weather events including heat waves, cold snaps, droughts and heavy rainfall.

A research team led by Kang Dae-hyun at the Korea Institute of Science and Technology's Climate and Carbon Cycle Research Center announced Sunday the development of KIST-Ocean, an AI-driven global ocean prediction model.

KIST-Ocean learns from decades of accumulated three-dimensional global ocean data to forecast future ocean conditions. Using physical variables such as sea surface temperature, ocean currents and salinity, the model produces a three-dimensional picture of the global ocean at five-day intervals and can simulate changes down to a depth of 600 meters. Powered by the latest AI technology, a single GPU can generate roughly 200 days' worth of global ocean predictions in just a few seconds — a significant reduction in computation time and cost compared with conventional numerical models. That computational efficiency is also expected to prove valuable for research requiring repeated analysis of multiple climate scenarios or large-scale ensemble forecasting.

To verify how accurately KIST-Ocean reproduces real ocean physics, the team conducted a series of experiments. In a simulated wind-forcing test, the model generated ocean waves, upwelling and downwelling patterns consistent with established oceanographic theory — demonstrating that the AI goes beyond simply memorizing historical data and can in fact capture the complex physical interactions between the atmosphere and the ocean.

The researchers expect KIST-Ocean to serve as a foundational technology that extends AI's role from short-term weather forecasting to seasonal and annual climate prediction. Beyond laying the groundwork for a future AI-based Earth system model integrating the atmosphere, ocean and land surface, its speed and cost-effectiveness are expected to lower the barrier to entry for ocean and climate research and accelerate a wide range of climate-change response studies.

"Through KIST-Ocean, we have demonstrated that AI prediction can deliver outstanding efficiency and accuracy while also realistically reproducing the complex physical relationship between the atmosphere and the ocean," Kang said. "By actively leveraging AI models like this, we will be able to significantly enhance our capacity to respond to the climate crisis."

The findings were published in the latest issue of Science Advances.