Amidst the artificial intelligence (AI) boom in China, the latest application emerges in meteorology, where AI technology has been used to enhance the accuracy of weather forecasts, offering a more effective coping mechanism for weather disasters during a summer challenged by heat waves, heavy downpours and typhoons.
In July and August this year, Shanghai Artificial Intelligence Laboratory deployed its Fengwu meteorological model on a trial operation to bolster the prediction and warning service for incoming typhoons such as Talim, Doksuri and Khanun.
The trial operation was carried out in cooperation with China's National Meteorological Center and Shanghai Meteorological Service.
For Typhoon Doksuri, by far the strongest typhoon that hit China this summer, the 24-hour prediction error of the Fengwu model is 38.7 km, considerably lower than that of the European Centre for Medium-Range Weather Forecasts (ECMWF) at 54.11 km, or that of the U.S. National Centers for Environmental Prediction (NCEP) at 54.98 km.
For Typhoon Khanun which swept parts of northeast China last week, the 48-hour prediction error of the Fengwu model is 47.5 km, while the prediction error of ECMWF is 54.5 km and that of NCEP is 63.8 km.
The Fengwu model was jointly published this April by the Shanghai Artificial Intelligence Laboratory, multiple Chinese universities and research institutions, and the Shanghai Central Meteorological Observatory.
The Fengwu is a global medium-range weather forecast model based on multi-modal and multi-task deep learning technology. It can forecast core atmospheric variables at high resolution for more than 10 days, which is more effective than traditional models.
Unlike the traditional physical models that mostly run on supercomputers, the Fengwu model only needs single graphics processing unit to generate high-precision global weather forecasts for the next 10 days in 30 seconds.
According to a researcher with the Shanghai Artificial Intelligence Laboratory, there is still much room for improvement in AI weather prediction. For example, weather prediction for a district in a city is currently achievable, and there is hope for the future that it could be extended to a street level.
According to the Fengwu research group, AI technology has improved the efficiency of weather forecasts, but weather, by its nature, is still difficult to predict accurately. The AI weather model can complement the traditional physical models in the future, and provide more accurate weather forecast information for agriculture, forestry, animal husbandry, fishing, aviation, navigation and public safety.
It will also apply AI technology to a wider range of Earth science research in the environment, astronomy, geology and other fields to support carbon neutrality, disaster prevention and reduction, and energy security.
Traditional numerical weather prediction has been proven successful in weather forecast and disaster early warning. However, with the slow growth of computing power and the growing complexity of physical models, the bottleneck of numerical prediction has become increasingly prominent.
According to the World Meteorological Organization, the effectiveness of the global medium-range weather forecast can only be improved by one day every 10 years. The data-driven AI forecasting methods are expected to realize high-precision forecasting with lower computational costs.
Pangu Weather is another model developed by China and it has been used for accurate weather forecasts. The research findings of the Pangu Weather research team from Huawei Cloud were published in the journal Nature in July, titled "Accurate medium-range global weather forecasting with 3D neural networks."
The team proposed a 3D neural network adapted to the Earth coordinate system to process complex 3D meteorological data, using a hierarchical temporal aggregation strategy to reduce the accumulation of errors.
Its weather forecast includes information about humidity, wind speed, temperature and sea level pressure, which are critical to predicting the development of weather systems, storm tracks, air quality and weather patterns.
According to Huawei, the prediction accuracy of the Pangu model from 1 hour to 7 days has exceeded the prediction accuracy of some meteorological centers in Europe and the United States in the same time span.
A comparative test report of the Pangu model and European numerical model from April to July this year, which is published by ECMWF, shows that AI weather forecast methods such as the Pangu model will break through the bottleneck of slow improvement of weather forecasting accuracy.
The Pangu model is also available online on the ECMWF website. Global weather forecasters, weather enthusiasts and the general public can view Pangu's global weather forecast for the next 10 days for free.
According to the Central Meteorological Observatory of China, the Pangu model demonstrated good performance in the path prediction of Typhoon Mawar this May and has been used in the forecast of Typhoon Doksuri this August.
According to Tian Qi, the chief scientist of the Huawei Cloud AI team, weather forecast is one of the most important scenarios in the field of scientific computing, and it is also a very complex system. At present, the primary capability of the Pangu model is to predict atmospheric evolution, thereby enhancing the existing weather forecast system.
The Central Meteorological Observatory of China said it will continue to strengthen the application of AI technology in typhoon monitoring and forecast, and work closely with universities and research institutions to offer innovative assistance for global typhoon monitoring, accurate forecasting and high-quality service provision.
READ ALSO:
