"The model can better grasp the process of moderate to heavy rain, and the location of heavy rain is more consistent with the real-time condition, which can offer effective tips to forecasters." the model that forecasters from Beijing Municipal Meteorological Observatory refer to is Lingxi extreme precipitation model (hereinafter referred to as the Model), a domestic artificial intelligence (AI) forecasting system independently developed by Beijing Municipal Meteorological Service (hereinafter referred to as the Service). The Model officially made its debut in this year's flood season, providing scientific support for precise forecasting and early warning services for extreme heavy precipitation in Beijing and North China.
In recent years, extreme precipitation such as short-term heavy precipitation, heavy rain and hail has increased in frequency year by year, which has posed grave threats to people's lives and property safety, and great challenges to meteorological disaster prevention and mitigation in big cities. In the wake of extreme heavy rainfall condition in Beijing last year, Beijing proposed new requirements for the research and application of AI technology in the precise forecasting of extreme weather.
In response, the Service established an AI weather model research and innovation team, pool superior resources to jointly tackle key problems, build an AI forecasting system for practical operational applications, and realize high-precision forecasting and intelligent extreme weather warning.
At present, some large circulation models of AI meteorological mid-range forecasting perform well in the mid-range forecasting of upper-air and surface meteorological factors, but still have shortcomings in terms of extreme precipitation prediction.
In order to approach this problem, the team has independently developed the Model to realize the autonomous control of computing power, algorithms and data. The Model can generate precipitation diagnosis and 6-hour cumulative precipitation forecast products by combining the circulation field.
The Model has addressed the problem of extreme unbalance of precipitation data through self-developed information balance scheme.
On July 25, typhoon Gaemi made landfall in Taiwan and Putian, Fujian, China. After landfall, its residual circulation moved north and affected inland areas of China.
"Whether it would bring extreme heavy rainfall to Beijing" was the focus of the analysis and judgment of Beijing Municipal Meteorological Observatory at that time.
"Changes in the typhoon track and intensity, even if only slightly adjusted, will exert a very big impact on the precipitation in Beijing."
In this regard, the Model has provided an important reference, forecasting that Gaemi would turn from 8 a.m. on July 28 to 8 a.m. on July 29, and the precipitation would move east in the Beijing area.
According to this conclusion, combined with the evolution trend of the circulation condition, Beijing Municipal Meteorological Observatory made the analysis that the influence of the typhoon on Beijing was significantly weakened, which provided support for the scientific and reasonable flood control work of the municipal and district levels of urban flood control departments.
The team tested the forecasting results by harnessing the real-time data in weather stations in North China. In the flood season of 2024, the Model performed well in predicting precipitation distribution and magnitude, in particular, in the forecasting of heavy precipitation.
At present, the team is promoting the integration of the Model into the field of meteorological insurance risk early warning service, providing important support for the automatic triggering of risk early warning, one-click dissemination and wider application of disaster prevention and mitigation scenarios.
Editor: LIU Shuqiao