In recent years, global extreme weather events have occurred frequently, from heat waves in Europe to mountain fires in North America, from rainstorm in East Asia to drought in Africa. Climate change is affecting human life at an unprecedented rate. Although traditional numerical weather forecasting has made significant progress in the past, it still faces bottlenecks such as insufficient forecasting accuracy and high computational resource consumption when facing complex nonlinear systems. The rise of AI models has brought about a revolution in weather forecasting.
Huawei Cloud's Pangu Meteorological Model has improved the prediction accuracy of temperature, wind speed, precipitation and other factors by about 15% within a forecast window of 1 hour to 7 days, and the inference speed has increased by more than 1000 times. In terms of typhoon path prediction, it performed particularly well, accurately predicting the landing location of the super typhoon 5 days in advance. Google DeepMind's GraphCast models the complex relationships between global atmospheric variables through graph neural networks, surpassing the best traditional model of the European Centre for Medium Range Weather Forecasts on over 90% of testing indicators, and completing a global 10 day forecast in just 1 minute.
The core advantage of AI models lies in their ability to capture nonlinear relationships. The essence of atmospheric circulation is a chaotic system. Traditional models rely on discretizing physical equations for solution, while AI models learn atmospheric motion laws from massive historical data to more accurately characterize small-scale weather processes. At present, AI weather forecasting is evolving towards integration with traditional models. The China Meteorological Administration has partnered with multiple technology companies to incorporate AI models into the business forecasting chain.
Looking ahead to the future, AI weather models are expected to achieve kilometer level and minute level precision forecasting, which will have a profound impact on disaster prevention and reduction, precision agriculture, renewable energy generation prediction, and other fields.
【 Reference sources 】 Huawei Cloud Disk Ancient Meteorological Model Paper, Google DeepMind GraphCast Paper, China Meteorological Administration AI Application Report.