Decoding ocean mysteries: China's LangYa large model set for major upgrade

QINGDAO -- In a coastal laboratory in Qingdao, East China's Shandong province, a team of scientists is putting the final touches on an advanced artificial intelligence system designed to predict the ocean's most unpredictable behavior.

Server hums echo across the research base of the Institute of Oceanology under the Chinese Academy of Sciences (IOCAS), where whiteboards are covered with algorithmic flowcharts and a red marker circles the number "2.0."

This is the home of the LangYa ocean large model, a cutting-edge AI system that researchers say will soon be able to forecast not just ocean temperature and salinity, but also typhoons, rainfall, storm surges and sea ice.

The name "LangYa" refers to an ancient terrace located just 20 kilometers away, where Chinese astronomers centuries ago observed the sky to determine the changing seasons.

FROM DATA TO DIAGNOSIS

The research and development of the LangYa series has been a journey of overcoming challenges. The hardest part was making it both fast and accurate, according to the IOCAS research team on AI in oceanography.

This is a new generation of AI large model developed specifically for marine environmental forecasting, and it marks a landmark achievement in the integration of AI and ocean science, said the team.

Scheduled for official launch in June, the upgraded 2.0 version is now being fine-tuned to incorporate forecasts of multiple marine phenomena, delivering what the team calls "critical support for marine environmental safety, resource development and disaster warning."

When the team unveiled version 1.0 in late 2024, the model could predict variables of global ocean conditions, such as temperature, salinity and currents, for the next one to seven days in a single run.

According to IOCAS, the model boosts forecasting efficiency by 10,000 times compared to conventional methods. Version 1.0 is now deployed and operational at the National Marine Environmental Forecasting Center in Beijing, a facility under the Ministry of Natural Resources that issues China's marine weather forecasts and disaster early warnings.

But temperature and salinity alone were not enough. The AI would need to understand complex ocean phenomena that cause typhoons, floods and storm surges.

"With the same batch of data, version 1.0 tells you what the sea temperature is. Version 2.0 tells you where a vortex might form and when a storm surge will arrive," said the team leader. "Version 1.0 helps a doctor read a scan, but the doctor still makes the diagnosis. Version 2.0 is the AI that reads the scan and gives you the diagnosis directly."

PROVEN PERFORMANCE

Even before its official release, version 2.0 has begun to prove its worth.

In 2025, the Sea Ice Prediction Network, an international collaboration of scientists and research institutions dedicated to improving seasonal forecasts of Arctic sea ice, released its seasonal forecast for September Arctic sea ice extent, which refers to the area of ocean covered by ice, a key indicator of climate change that also affects shipping routes and weather patterns far beyond the Arctic.

A model developed by the IOCAS team, an evolving version of LangYa 2.0, outperformed multiple international institutions, ranking first globally.

For the team, this achievement validates the core philosophy of their work. That is, the deep integration of AI and marine science is one of the key pathways to leading the world in industrial AI applications.

Their goal is to bridge the gap between basic "ocean variables," for instance, temperature and salinity, and the marine phenomena, such as typhoons and storm surges, that directly affect public safety and economic activity.

When version 2.0 launches in June, it will be able to forecast not just the state of the sea, but its behavior, according to the team.

For the millions of people living along China's coastlines, where typhoons and storm surges pose a seasonal threat, that distinction could make all the difference.