How AI is advancing global weather forecasting services

As climate change intensifies extreme weather events worldwide, artificial intelligence (AI) is becoming a powerful tool in weather forecasting, helping meteorologists improve forecasting efficiency and strengthen early warning systems.

Combining traditional, cutting-edge methods

Modern weather forecasting is no longer based solely on observing clouds or atmospheric conditions. It combines numerical weather prediction (NWP), AI meteorological models and the expertise of professional forecasters, said La Sa, a senior engineer at the Beijing Meteorological Bureau.

For short- to medium-range forecasts covering one to 14 days, forecasters usually compare outputs from multiple numerical models before making final assessments. For the prediction of evolving weather within three hours, they rely more on multi-source observational data from radars, satellites, high-density automatic weather stations and wind profilers, with data updated at minute- or even second-level intervals.

Unlike traditional numerical models, which simulate the atmosphere by solving complex physical equations on three-dimensional grid points, AI models learn patterns from vast amounts of historical weather data. Once trained, they can generate forecasts in minutes instead of hours of calculations.

China has developed a family of AI meteorological models for different forecasting tasks, including Fenglei for short-term nowcasting, Fengqing for global short- to medium-range forecasting, Fengshun for subseasonal and seasonal prediction, and Fengyu for space weather forecasting.

The hybrid approach has helped China achieve around 85% accuracy for 24-hour urban weather forecasts, while keeping temperature forecast errors within one to two degrees Celsius, said La.

The approach has also shown practical results in tropical cyclone forecasting.

Earlier this month, as Typhoon Bavi approached China's southeastern coast, an AI forecasting agent developed by Shanghai Meteorological Bureau continuously tracked the storm's development by learning from massive volumes of satellite, radar and historical typhoon data. Meanwhile, the Shanghai Typhoon Institute developed Haisi, another AI agent integrating NWP and AI technologies, which can forecast typhoon track, intensity and structure for up to 15 days within minutes.

Expanding AI meteorology cooperation

Recently, China has announced a series of initiatives to strengthen the use of AI in meteorological disaster prevention and expand international cooperation on early warning systems.

A range of cutting-edge AI technologies for meteorology, along with the latest achievements of MAZU, China's AI-powered meteorological early warning system, were unveiled at the 2026 World Artificial Intelligence Conference in Shanghai last week.

These include the launch of the China-Thailand Joint Laboratory for Intelligent Prediction and Early Warning of Meteorological Disasters, the world's first bilateral international laboratory dedicated to AI-driven meteorological applications. Built on the MAZU platform, it will develop AI technologies for forecasting and early warning of hazards, including typhoons, heavy rainfall, heatwaves and droughts, while supporting cross-border applications in agriculture, energy and infrastructure.

The "Djibouti 2.0" version of MAZU was also officially delivered. The upgraded system integrates monitoring, forecasting and warning functions and is designed to improve early-warning services in high-impact areas with complex conditions.

The newly launched MAZU-FengYun Satellite AI Box integrates satellite data reception, multi-source data fusion and operational applications, allowing rapid deployment and customization for different countries and regions.

China also released the Fenghe (Wind Harmony) meteorological service large language model and launched a global open-source plan aiming to promote sharing, strengthen international cooperation on disaster risk reduction and expand access to AI-enabled meteorological services.

MAZU, which stands for Multi-hazard, Alert, Zero-gap and Universal, is China's cloud-based AI early-warning solution. It aims to provide timely and convenient weather and climate services while supporting multi-hazard risk reduction across regions.

According to the China Meteorological Administration (CMA), meteorological agencies have already used the platform in more than 40 countries via cloud access. Customized versions have been deployed in seven countries, including Pakistan, Ethiopia, the Solomon Islands, Jordan, Sri Lanka, Mongolia and Djibouti, supporting local needs ranging from monitoring and forecasting to warning production and information dissemination.

China has also supported capacity building in developing countries through international training courses, scholarship programs and visiting scholar programs. Nearly 1,000 participants from more than 100 developing countries and regions have joined the early warning training programs in recent years. 

AI meteorological models already demonstrate strong capabilities in forecasting and early warning for severe weather events such as typhoons, said Pan Jinjun, chief engineer at the CMA. However, challenges such as limited samples of extreme weather events and insufficient model reliability remain, Pan said efforts will be further made to advance these models from being merely capable of making predictions to becoming reliable and ready for operational services.

Zhang Xingying, director of the CMA's Department of International Cooperation, said China will promote the global application of MAZU through country-specific implementation strategies. The CMA will also strengthen cooperation with relevant partners to provide integrated support in technology and talent development while leveraging multilateral platforms to help more developing countries access and benefit from AI-powered meteorological services.