What happened
Staff of CMA's Shanghai Typhoon Institute and National Meteorological Centre verify typhoon forecasts each year under CMA's regulations on typhoon operations and services. Their verification of the 27 named typhoons of 2025 appeared in Meteorological Monthly in October 2026. It covers seven AI weather models, with forecasts supplied by the National Meteorological Centre. They were FengQing, FengWu, FuXi, Pangu, AIFS, FourCastNet and GraphCast. At short lead times, the 95th-percentile track errors of the AI models were generally larger than those of physics-based models. At long lead times they were smaller. In a same-sample check, ECMWF's AIFS gave the best track forecasts at every lead time. The physics-based models' 24-hour mean track errors were 12% to 55% larger than those of AIFS. FourCastNet's errors were much larger than those of the other AI models. The authors add that correcting the AI models' intensity forecasts with an adaptation model reduces their underestimation of typhoon intensity.
Sources
-
Peer-reviewed study, by the organization that used it.
YANG Mengqi, et al. 2025年西北太平洋和南海台风预报精度评定 (Accuracy Verification of Typhoon Forecasts over the Northwest Pacific and the South China Sea in 2025). Meteorological Monthly (气象) 52(10): 1257-1270, October 2026. DOI 10.7519/j.issn.1000-0526.2026.062602 (Chinese abstract).
人工智能气象模型在短预报时效的路径预报误差第95%分位数普遍高于数值预报模式，长预报时效则低于数值预报模式。
For AI-based weather prediction (AIWP) models, however, the 95th percentile track forecast errors were generally larger than those of numerical weather prediction (NWP) models at short lead times, but smaller than those of NWP models at long lead times.
Details
AI system
- Workflow
-
The National Meteorological Centre supplied the forecasts of the seven AI models. The authors verified all forecast methods against best-track data for the named typhoons of 2025.
Follow-up
- Conditions for replication
-
At short lead times, the AI models' worst errors exceed those of physics-based models. AI intensity forecasts need correction for their underestimation of typhoon intensity.
Entry details
- Sector
-
Weather forecasting / tropical cyclones.
- Purpose
-
ECMWF, the Shanghai Artificial Intelligence Laboratory, Fudan University, Huawei, NVIDIA, Google DeepMind and CMA built these AI weather models to forecast global weather from data.
Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). AIFS gave the best typhoon tracks in China's 2025 verification (AIBD-0758). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0758.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)