AI weather models rival traditional systems in convective-scale forecasts
To effectively present the uncertainty of convective-scale weather forecasts, convective-scale ensemble prediction systems have been developed at major operational centers, whose lateral boundary conditions are usually provided by global numerical weather models.
Recently, the emergence of AI weather models has provided a new approach to driving convective-scale ensemble prediction systems. AI weather models can produce forecasts for the next 7 to 10 days in just a few minutes, which is around 10,000 times faster than numerical weather models. However, the performance of using the predictions from the AI weather models as lateral boundary conditions to drive convective-scale ensemble prediction systems needs to be investigated.
Recently, researchers from the School of Atmospheric Sciences at Nanjing University, China and the Earth System Numerical Prediction Center of the China Meteorological Administration, a paper titled "Impacts of lateral boundary conditions from numerical models and data-driven networks on convective-scale ensemble forecasts" in Atmospheric and Oceanic Science Letters. This paper reported the impacts of the lateral boundary conditions provided by the AI models "Pangu-weather" and "Fuxi-weather" on convective-scale ensemble prediction systems.
The results of the ensemble prediction system using lateral boundary conditions from Pangu-weather forecasts were found to be comparable to those using the conditions from National Centers for Environmental Prediction Global Forecast System forecasts. Meanwhile, the results using lateral boundary conditions from Fuxi-weather forecasts showed slightly inferior performance compared to the others.
When the number of vertical levels of the Global Forecast System forecasts were reduced from 33 to 13 (the same as Pangu-weather and Fuxi-weather), the results were the worst among all ensemble prediction systems. This indicates that the vertical resolution of the lateral boundary conditions could significantly affect the convective-scale ensemble prediction systems.
Overall, this work shows that AI weather models have the ability to replace traditional global numerical weather models in providing the lateral boundary conditions for convective-scale ensemble prediction systems. Moreover, as the vertical resolution of AI weather models improves, it is expected that further positive impacts on convective-scale ensemble forecasts will materialize.
More information: Junjie Deng et al, Impacts of lateral boundary conditions from numerical models and data-driven networks on convective-scale ensemble forecasts, Atmospheric and Oceanic Science Letters (2025).
Provided by Chinese Academy of Sciences
