AI Reads the Ocean... KIST Develops Global Ocean Prediction Model [Reading Science]

Reproducing Atmosphere-Ocean Interactions, Accurately Predicting Even Super El Niño
Generating 200 Days of Ocean Forecasts in Seconds with a Single GPU

As the frequency of extreme weather events such as heat waves, heavy rainfall, and typhoons increases due to the climate crisis, Korean researchers have developed a technology that can rapidly and accurately predict global ocean conditions using artificial intelligence (AI). Previously, predicting ocean conditions required supercomputers to run calculations for extended periods, but this new technology enables predictions in just a few seconds with a single graphics processing unit (GPU), which is expected to significantly enhance long-term climate forecasting and the ability to respond to climate change.

The Korea Institute of Science and Technology (KIST) announced on July 24 that Dr. Kang Dae-hyun and his research team from the Climate Carbon Cycle Research Center have developed an AI-based global ocean prediction model called "KIST-Ocean." The research results were published in the latest issue of the international journal Science Advances.

Structure of the KIST-Ocean Model Inference Process. An AI trained on decades of atmospheric and oceanic data predicts global three-dimensional ocean changes up to 200 days by iteratively forecasting in 5-day intervals based on initial oceanic and atmospheric conditions. Provided by the research team.

KIST-Ocean is an AI model that predicts future ocean states by learning from decades of accumulated global three-dimensional ocean observational data. Based on key physical information of the ocean, such as water temperature, currents, and salinity, it can forecast changes up to a depth of 600 meters in 5-day intervals. In particular, a single GPU can generate up to 200 days of global ocean prediction results within seconds, drastically reducing computation time and costs compared to traditional numerical prediction models.

The research team verified how accurately the AI replicated actual physical phenomena in the ocean and found that it accurately simulated ocean waves and upwelling/downwelling in response to atmospheric changes, consistent with established ocean physics theory. Additionally, the AI successfully reproduced major development processes during the 2015 super El Niño event, such as the rise in equatorial Pacific sea surface temperatures and internal heat distribution changes, thereby confirming the reliability of AI-based ocean prediction.

KIST-Ocean's super El Niño reproduction experiment. When applying the wind conditions of 2015, sea surface temperatures in the central and eastern Pacific rose, developing a super El Niño (left), but under normal wind conditions, a La Niña response with lower sea surface temperatures appeared (right). This aligns with previous studies showing that tropical Pacific winds play a crucial role in El Niño development and demonstrates the model's ability to reproduce physical phenomena. Provided by the research team

The research team expects that this technology will not only aid short-term weather forecasting, but also serve as a foundational platform to expand the application of AI to seasonal and annual climate predictions. They further projected it would contribute to the development of integrated AI-based Earth system models, including the atmosphere, ocean, and land surface, thereby accelerating research into analyzing various climate change scenarios and responding to climate-related disasters.

Kang Dae-hyun, a researcher at KIST, stated, "KIST-Ocean has demonstrated not only high prediction accuracy and computational efficiency with AI, but also the realistic capability to reproduce the complex physical interactions between the atmosphere and ocean. We will further advance AI-based climate prediction technologies to enhance our ability to respond to the climate crisis, thus contributing to ensuring public safety and reducing social and economic damage."

This content was produced with the assistance of AI translation services.