ALIBABA DAMO Academy, the research and development arm of Alibaba Group, has officially launched “Baguan,” a groundbreaking weather forecasting model that leverages cutting-edge artificial intelligence to revolutionize weather prediction capabilities.
Named after the Chinese concept of “observing from different perspectives,” Baguan represents a significant advancement in climate science and holds promise for substantial impact on countries like the Philippines.
The forecasting model offers unprecedented accuracy in weather forecasts, providing predictions ranging from one hour to ten days ahead. The machine-learning model distinguishes itself with high spatial resolution, delivering detailed meteorological predictions down to a 1 x 1 kilometer grid, updated hourly. These capabilities make it an essential tool for applications in climate science, electricity load forecasting, renewable energy forecasting, and natural disaster prevention.
“Baguan represents a significant advancement in our dedication to harnessing technology for the greater good,” Wotao Yin, Director of Decision Intelligence Lab at Alibaba DAMO Academy said. “Its sophisticated technology not only helps elevate climate science but also benefits sustainable practices across diverse sectors such as renewable energy and agriculture.”
The technical backbone of Baguan lies in its innovative use of the Siamese Masked Autoencoders (SiamMAE) structure and a robust pre-training methodology. These innovations empower the model to uncover intricate patterns gleaned from complex dynamic atmospheric data. Through an autoregressive pre-training approach, it is able to make precise predictions across various spatio-temporal scales, from one hour to ten days in advance.
Baguan leverages ERA5, the European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis of global weather from 1979 to the present, to construct the foundational model for weather forecasting. The model is further refined with key regional meteorological indicators such as regional temperature, irradiance, and wind speed. This meticulous global-regional modeling approach not only enhances Baguan’s forecasting accuracy at the regional level but also tailors its insights to specific local conditions.
With the surging global demand for renewable energy, Baguan’s precise weather predictions have become vitally important. The model significantly enhances the reliability of renewable energy forecasts, facilitating more stable power management and supporting the expansion of green energy consumption. Baguan’s capability in weather forecasting has already been utilized in China’s power and energy sectors, supporting critical applications such as electricity load forecasting and renewable energy forecasting.
For example, during an unexpected temperature drop in Shandong province in August, Baguan accurately predicted a corresponding 20 percent drop in electricity demand one day ahead, achieving a 98.1 percent accuracy rate in day-ahead load forecasts. This precision assisted local grid operators in optimizing power dispatch, enhancing efficiency, and reducing operational costs.
“We have years of research experience in mathematical modeling, time-series forecasting, and explainable AI, which helps us in building a high-precision regional weather forecast model,” said Yin. “We will continue to enhance performance for key weather indicators such as cloud cover, extreme wind speed, and precipitation, develop new technology for different climate scenario analysis, and support more applications such as civil aviation meteorological warnings, agricultural production, and sporting events preparations.”
