China develops AI system to detect dust storms, wildfire smoke and aerosol pollution 360 times faster

Chinese researchers have developed what they describe as the world's first AI-driven Global Aerosol-Meteorology Forecasting System (AI-GAMFS), an artificial intelligence model capable of predicting aerosol pollution events such as dust storms, wildfire smoke and urban haze up to five days in advance.

The breakthrough, published in Nature, could significantly improve early warning systems for air pollution, environmental hazards and disaster preparedness. Unlike conventional forecasting systems that rely on computationally intensive atmospheric chemistry models, the AI-powered system can generate a complete five-day global aerosol forecast in around one minute. According to the researchers, this makes it approximately 360 times faster than traditional numerical forecasting while maintaining a high level of accuracy.

What is the new weather forecasting AI system developed by Chinese scientists

The newly developed system, known as AI-GAMFS, combines two advanced artificial intelligence architectures, a Vision Transformer and a U-Net neural network, to learn the complex relationships between weather patterns and airborne particles.A a research team led by scientists from the Chinese Academy of Meteorological Sciences (CAMS) under the China Meteorological Administration (CMA), together with researchers from several Chinese and international institutions, published the world's first Artificial Intelligence-driven Global Aerosol-Meteorology Forecasting System (AI-GAMFS) in the journal Nature.

Rather than solving millions of mathematical equations every time a forecast is produced, the AI learns these relationships from historical observations. Researchers trained the model using 42 years of global atmospheric data from NASA's Modern-Era RetrospectiveAnalysis for Research and Applications, Version 2 (MERRA-2). It is then initialised with real-time meteorological information from the Global Earth Observing System Forward Processing (GEOS-FP), allowing it to predict both weather conditions and aerosol movement simultaneously.The system generates forecasts every three hours at a spatial resolution of roughly 50 kilometres, estimating aerosol optical depth, a key indicator of how many airborne particles are present as well as concentrations of dust, sulphates, black carbon, organic carbon and sea salt across the globe.

Detecting dust storms, wildfire smoke and pollution

To test the system, researchers compared its forecasts with independent observations from the Aerosol Robotic Network (AERONET) and the Chinese Aerosol Remote Sensing Network (CARSNET).

They also benchmarked its performance against leading operational forecasting systems, including the Copernicus Atmosphere Monitoring Service (CAMS).The study found that AI-GAMFS consistently produced more accurate predictions of aerosol optical depth and dust events. It successfully tracked major environmental events, including Saharan dust storms, wildfire smoke from Central Africa, and pollution episodes overChina and the United States, while providing improved forecasts of surface aerosol concentrations.These improvements could enable authorities to issue earlier warnings before hazardous air pollution reaches populated areas, giving people more time to take protective measures.

Why detecting storms is important for satellites, GPS and power grids

Traditional aerosol forecasting models simulate atmospheric chemistry in remarkable detail but require enormous computing power, often taking four to six hours to produce a five-day forecast using supercomputers.AI-GAMFS performs the same task in less than one minute on a single graphics processing unit (GPU).

This dramatic reduction in processing time could allow forecasting agencies to update predictions more frequently while using far fewer computational resources, making advanced air-quality forecasting more accessible worldwide.

How artificial intelligence could transform future space weather monitoring

The researchers believe AI-GAMFS demonstrates that artificial intelligence can complement traditional atmospheric models rather than replace them. Faster forecasts could improve responses to dust storms, wildfire smoke, volcanic ash and severe pollution episodes while supporting aviation, agriculture, climate monitoring and public health planning.As extreme weather events and wildfire activity become more frequent in many parts of the world, rapid and accurate aerosol forecasting may become an increasingly important tool for protecting both people and the environment.