In response to escalating climate threats, Alibaba’s research arm, DAMO Academy, has launched a breakthrough AI weather forecasting model, “Baguan.” Designed to predict conditions up to ten days in advance with hourly updates, Baguan aims to redefine precision in meteorology, helping industries adapt to climate volatility and reduce environmental impact.
Recent extreme weather incidents, such as devastating floods in Spain, heavy rainfall-caused landsides and flooding in Nepal, and a tropical storm in the Philippines that affected millions, have highlighted urgent threats posed by climate change.
According to a report titled “United in Science” by the World Meteorological Organization (WMO), the impacts of climate change and severe weather jeopardize the well-being of people and the planet. Yet artificial intelligence and machine learning can offer crucial help, as the advanced technologies “make skillful weather modeling faster, cheaper and more accessible,” especially to “lower-income countries with limited computational capacities.”
Baguan was named after the ancient Chinese practice of encompassing diverse perspectives for a comprehensive understanding. It leverages the latest advancements in AI technology to enhance the accuracy and efficiency of weather forecasting. This model is designed to predict weather conditions and provide hourly updates, with unprecedented precision from an hour to ten days in advance, and a high spatial resolution of one-by-one kilometer grids.
“Baguan represents a significant advancement in our dedication to harnessing technology for the greater good,” said Wotao Yin, Director of Decision Intelligence Lab at Alibaba DAMO Academy. “Its sophisticated technology not only helps elevate climate science but also benefits sustainable practices across diverse sectors such as renewable energy and agriculture.”
Developed using the novel Siamese Masked Autoencoders (SiamMAE) architecture and a pioneering autoregressive pre-training methodology, Baguan excels in processing and understanding complex atmospheric data. A global-regional modeling approach further enhances the model’s performance: It leverages ERA5, the European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis of the global weather from 1979 to the present, supplemented with localized weather indicators like temperature, wind speed, and solar irradiance.
Baguan’s capabilities extend beyond mere weather forecasting. In the renewable energy sector, the model’s detailed and accurate weather predictions are pivotal in optimizing energy production, contributing to more stable and efficient power management. The model’s precision was demonstrated during a sudden temperature drop in Shandong Province in China, where Baguan accurately anticipated a 20% drop in electricity demand, achieving a high accuracy—98.1% in load forecasting. This allowed for enhanced grid operation, reducing costs and improving energy distribution efficiency.
DAMO Academy’s ambition extends beyond short-term weather predictions. Built on years of research experience in mathematical modeling, time-series prediction, and explainable AI, DAMO is capable of building a high-precision weather forecast model that is designed to benefit diverse sectors and enhance adaptability across regions with diverse climate challenges.
“We will continue to enhance performance for key weather indicators such as cloud cover and precipitation, develop new technology for different climate scenario analysis, and support more applications such as civil aviation meteorological warnings, agricultural production, and sport events preparations,” added Yin.