Domestic AI has been published in the international academic journal Nature. On July 6th, Beijing time, the journal Nature published the research results of the Huawei Cloud Pan Gu Big Model R&D team - "3D Neural Network for Accurate Medium Range Global Weather Forecasting". The paper shows that the Pangu Meteorological Model is the first AI model with accuracy surpassing traditional numerical forecasting methods, and its speed has increased by more than 10000 times compared to traditional numerical forecasting. This is the first official paper published by a Chinese technology company in recent years as the sole signatory in the journal Nature. According to the evaluation of Nature, "The Pangu meteorological model makes people re-examine the future of the meteorological forecasting model, and the opening of the model will promote the development of this field." ▲ Screenshots of papers in the international academic journal Nature since the 1920s, especially in the past 30 years, with the rapid development of computing power, Numerical weather prediction has achieved great success in daily weather forecasting, extreme disaster early warning, climate change prediction and other fields. However, with the slowing down of computing power growth and the gradual complexity of physical models, the bottleneck of traditional numerical forecasting is becoming increasingly prominent. Researchers are beginning to explore new weather forecasting paradigms, such as using deep learning methods to predict future weather. In the field where numerical methods are most widely used, such as medium to long-term forecasting, the accuracy of existing AI forecasting methods is still significantly lower than that of numerical forecasting methods, and is constrained by issues such as lack of interpretability and inaccurate extreme weather prediction. The research and development team of Pangu Big Model found that the accuracy of AI meteorological forecasting models is insufficient mainly due to two reasons: firstly, the original AI meteorological forecasting models are based on 2D neural networks, which cannot handle uneven 3D meteorological data well; Secondly, AI method lacks the constraint of Mathematical physics mechanism, so iteration error will be accumulated continuously in the process of iteration. To this end, the team creatively proposed a three-dimensional neural network adapted to the Earth coordinate system to process complex and non-uniform 3D meteorological data, and used a hierarchical time-domain aggregation strategy to reduce the number of prediction iterations, thereby reducing iteration errors. By training deep neural networks on 43 years of global weather data, the Pangu Meteorological Model surpasses traditional numerical prediction methods in terms of accuracy and speed. In May of this year, the direction of Typhoon "Marva" received widespread attention. According to the news from the National Meteorological Center of CMA, Pangu's large model performed well in the path prediction of "Mawa", and predicted that it would turn to the path in the eastern waters of Taiwan Island five days in advance. At the just concluded 19th World Meteorological Congress, the European Centre for Medium-Range Weather Forecasts also pointed out that Pangu's large meteorological model has undeniable ability in accuracy. The purely data-driven AI weather forecasting model shows the forecasting strength comparable to the numerical model. As of now, the Pangu Meteorological Model can provide global meteorological second level forecasts, with meteorological prediction results including geopotential, humidity, wind speed, temperature, sea level pressure, etc., which can be directly applied to multiple meteorological research sub scenarios. The research team will continue to explore and give full play to the application potential of AI in the meteorological field, together with the global meteorological institutions, to provide support for agriculture, forestry, animal husbandry, fishery, aviation, navigation and other industries. (New News Agency)