Evaluation of Numerical and AI Models for Forecasting Regional Flood‐Inducing Extreme Precipitation During the Meiyu Season
Year: 2026
Type: article
Abstract: Under global warming, accurate forecasting of extreme precipitation events remains a major challenge. This study presents a comprehensive evaluation of the capabilities of both numerical weather prediction (NWP) model (integrated forecasting system [IFS]) and state‐of‐the‐art AI models (AIFS, FuXi, and GraphCast) in predicting regional, flood‐inducing extreme precipitation during the 2024 Meiyu season (June 17 to July 15) in the middle and lower reaches of the Yangtze River. A set of tailored skill metrics focused on extreme events is defined to quantitatively assess the models’ performance in capturing precipitation intensity and spatial distribution for a case study of the 2024 Meiyu season. The results indicate a common underestimation of both the intensity and areal coverage of extreme rain belts across all models. However, key differences emerge: the NWP model (IFS) demonstrates stable performance and improving accuracy with shorter lead times, highlighting its strength in representing atmospheric physics. In contrast, the AI models exhibit more varied capabilities, with AIFS generally outperforming IFS in this event at short lead times, FuXi showing optimized short‐term forecasting skill, and GraphCast displaying significant forecast volatility. This comparative analysis reveals the distinct characteristics and complementary strengths of numerical and AI approaches, providing crucial insights for future development of hybrid forecasting systems that leverage the physical foundations of NWP and the pattern‐recognition prowess of AI.
Source: Advances in Meteorology
DOI: 10.1155/adme/9373465
Authors: Rong Yu, Shengxuan Ji, Panmao Zhai
Language: English
Subfield: Atmospheric Science
Field: Earth and Planetary Sciences
Domain: Physical Sciences
SDG: Climate action
Open Access status: gold
Estimated APC paid: $1,630
Awards: G1725547645, G4198602783, G473579092
OpenAlex ID: W7214769298
