China's AI model advances effective forecast time of global mid-term weather to 11.25 days

A Chinese artificial intelligence lab has successfully advanced the effective forecast time of global mid-term weather to 11.25 days for the first time. It managed to achieve weather forecasting ranging from severe convective weather that occurs within minutes to 10-year ocean climate prediction.

The weather forecast model is called Fengwu, a machine-learning model developed by the Shanghai Artificial Intelligence Laboratory.

Unlike the traditional physical models that mostly run on supercomputers, Fengwu uses artificial intelligence to analyze the elements provided by atmospheric data assimilation, such as wind speed, temperature, and humidity to predict future weather. Artificial intelligence can utilize past meteorological elements, such as temperature, to make more precise forecasts, explained Ouyang Wanli, a scientist from the Shanghai lab, to the Global Times.

Currently, Fengwu has achieved a breakthrough in global medium-term meteorological forecasting, advancing the effective forecast time to 11.25 days for the first time. Moreover, the model has been upgraded and expanded to include modules for severe convective weather forecasting, global medium-term meteorological forecasting, and ocean climate forecasting, the Global Times learned from the Shanghai lab.

It can now provide forecasts ranging from minute-level severe convective weather to decade-long ocean climate predictions, covering precipitation, temperature, wind, solar radiation, ocean currents, sea temperature, and other climate and meteorological elements.

The forecast of 11.25 days has made Fengwu surpass some of the international equivalents in predicting weather.

In November 2023, Google DeepMind released an AI that delivers 10-day weather forecasts with unprecedented accuracy and speed.

Moreover, Fengwu has realized weather forecasts from "doorstep" to "the heart of the ocean." It can provide accurate weather forecasts of severe convection weather within one kilometer, and offer accurate weather and ocean activity forecasts for different altitudes and depths of the sea.

In recent years, the world has suffered from increasingly frequent climate disasters. More common extreme weather-related disasters have highlighted an urgency for weather forecasts to become more precise.

So far, the Shanghai Artificial Intelligence Laboratory has cooperated with both national and Shanghai's municipal meteorological bureaus to construct severe convective weather forecast models, allowing for precise radar precipitation forecasts every five minutes for the next two hours.

Due to its own limitations and the uncertainty of weather, the traditional model of weather forecast still cannot meet the diverse and growing needs of today's users; whilst data-driven AI methods provide very useful tools to bridge this gap, Dai Kan, deputy head of China's National Meteorological Center, told the Global Times in a previous interview.

In addition to Fengwu, the Shanghai Academy of Artificial Intelligence for Science (SAIS) and Fudan University unveiled the upgraded model, known as FuXi 2.0, last month. It is the first global large weather model for weather routing, Xinhua News Agency quoted a press release of the university as saying.

Fuxi 2.0 has made progress in terms of mid-term weather forecasting models and sub-seasonal models, targeting industries such as new energy, aviation, and maritime transportation.

The Chinese government has already started exploring various ways of using artificial intelligence in weather forecasting.

In May this year, the China Meteorological Administration (CMA) launched a pilot program for AI weather forecasts.

AI models participating in the program will use real-time observational data provided by the CMA as input fields to produce weather forecasts for the coming 15 days. The demonstration products include forecasts for high-altitude meteorological elements, surface meteorological elements, typhoon paths, and forecasts for hazardous weather processes.