MicrosoftThe Start team recently announced that they have developed a new data-driven AIWeather prediction modelsThe model is able to accurately predict weather conditions for the next 30 days. This research result not only achieved a significant improvement in forecasting accuracy, but also realized a huge breakthrough in computational efficiency.
Paper address:https://arxiv.org/pdf/2403.15598
The new AI weather prediction model is the result of years of research by Microsoft's Start team, which combines five different AI models and three deep learning architectures. By training with massive amounts of weather data from the past decades, the model is able to learn and simulate complex weather patterns of change.
In predicting the weather forecast for the coming week, the new AI model's temperature accuracy is 171 TP3T better than that of the current leader, the European Center for Medium-Range Weather Forecasts (ECMWF.) Even when predicting the weather for the next four weeks, the new model's accuracy is 41 TP3T better than that of the ECMWF's model.
In contrast to the currently commonly used NWP/ECMWF model, which needs to rely on thesupercomputers to perform a large number of calculations, Microsoft's AI weather prediction model offers a significant advantage in terms of computational efficiency. The model is able to run on a single NVIDIA V100 graphics processing unit and produce 100 30-day forecasts in less than two hours, significantly reducing the need for computing resources.
This research result from Microsoft has not only made significant progress in the accuracy of weather forecasting, but also revolutionized the computational efficiency. This will hopefully drive the development of weather prediction services and bring positive impacts to many fields such as agricultural production, transportation, and energy management. With the continuous advancement of AI technology, future weather prediction will be more accurate and efficient.