At the recent 2024 Western Technology Innovation Ecological Development Conference, the Chongqing Meteorological Bureau officially released the "Tianzi · 12h" artificial intelligence (AI) weather forecasting model. This model focuses on precise forecasting of disaster-causing precipitation in the Chengdu-Chongqing region, marking an important step forward for Chongqing in the field of weather forecasting.
The "Tianzi · 12h" model is based on Huawei's Pangu global meteorological model, integrating ground meteorological data and high-precision terrain data from the Chengdu-Chongqing area. It employs a fusion architecture of global and regional models and incorporates a three-dimensional Earth-specific transformer (3DEST) model, multi-source heterogeneous data fusion technology, and techniques for handling small samples of heavy precipitation, providing a solid foundation for improving forecasting accuracy.
Image Source Note: Image generated by AI, image authorized by service provider Midjourney
This model can forecast ground precipitation, temperature, wind, and relative humidity in the Chengdu-Chongqing region every 6 hours for the next 12 hours, with a temporal and spatial resolution of 3 hours and 3 kilometers. Additionally, it can forecast the potential height, specific humidity, temperature, and wind at 13 pressure levels in the atmosphere, with a temporal and spatial resolution of 3 hours and 25 kilometers.
Gao Yudong, director of the key laboratory for research on the integration of disaster-causing precipitation numerical models and AI forecasting technology, explained that the main forecasting timeframe of this model is between 2 to 12 hours into the future. For the first 2 hours, the forecasts rely heavily on meteorological observation data analysis, as this period is very close to potential weather disasters, allowing most weather changes to be observed. However, forecasts beyond 12 hours still depend on traditional numerical weather models, and existing AI model technologies require further research and iteration.
Zhang Yan, director of the Chongqing Meteorological Station, stated that this model can be applied in various fields, including megacity governance, low-altitude economy, and warning systems for geological disasters. For instance, in the low-altitude economy, meteorological conditions are a crucial factor affecting the safety of low-altitude flights. Since low-altitude flights occur in a highly convective and rapidly changing atmospheric environment, meteorological conditions must be closely monitored from takeoff to landing, and the AI weather forecasting model can ensure the smooth operation of drones and other low-altitude aircraft.
In terms of forecasting accuracy, the "Tianzi · 12h" AI weather forecasting model has improved the accuracy of rainfall forecasts for the first 6 hours by up to 36% compared to traditional forecasts.
