Zhang et al. (2026) A physics-guided integrated global to regional method for high-resolution weather forecasting

Identification

• Journal: npj Climate and Atmospheric Science
• Year: 2026
• Date: 2026-10-01
• Authors: Jing Zhang, Yutong Dai, Yu Wang, Fang He, Pengbo Xu, Junping Yin
• DOI: 10.1038/s41612-026-01555-w

Research Groups

• Academy for Advanced Interdisciplinary Studies, Northeast Normal University, Changchun, China.
• Institute of Applied Physics and Computational Mathematics, Beijing, China.
• School of Mathematical Sciences, Key Laboratory of MEA (Ministry of Education), Shanghai Key Laboratory of PMMP, East China Normal University, Shanghai, China.
• Shanghai Zhangjiang Institute of Mathematics, Shanghai, China.

Short Summary

This study introduces AERO-ODE, a physics-guided framework that combines global dynamical information, parameterization increments, and regional terrain constraints to generate high-resolution weather forecasts. The method shows significant improvements in RMSE and ACC compared to regional and global baselines.

Objective

• Develop a physics-guided integrated global to regional method for high-resolution weather forecasting.
• Improve the accuracy of pressure-level and surface-variable forecasts over complex terrain regions.

Study Configuration

• Spatial Scale: Regional scale (3 km horizontal resolution) with two evaluation domains: eastern region ED (relatively flat) and western complex-terrain region WD.
• Temporal Scale: 72 h forecast lead time, with hourly intervals.

Methodology and Data

• Models used: AERO-ODE (AI-Enhanced Regional ODE Forecasting Framework), a physics-guided framework that combines global dynamical information, parameterization increments, and regional terrain constraints.
• Data sources: Global initial fields from NeuralGCM dynamical core, HRRR analysis field (3 km) as reference.

Main Results

• AERO-ODE shows significant improvements in RMSE and ACC compared to regional baselines WRF-ARW and YingLong-WRF, with median reductions of 29–32% and 28–33%, respectively.
• The method also outperforms global models NeuralGCM, PanGu-Weather, and IFS for most variables.

Contributions

• AERO-ODE provides a novel approach to high-resolution regional forecasting by integrating global dynamical information with regional terrain constraints.
• The method shows significant improvements in RMSE and ACC compared to existing regional and global baselines.

Funding

• This research was supported by the National Natural Science Foundation of China (Grant No. 91746201).

Citation

@article{Zhang2026physicsguided,
  author = {Zhang, Jing and Dai, Yutong and Wang, Yu and He, Fang and Xu, Pengbo and Yin, Junping},
  title = {A physics-guided integrated global to regional method for high-resolution weather forecasting},
  journal = {npj Climate and Atmospheric Science},
  year = {2026},
  doi = {10.1038/s41612-026-01555-w},
  url = {https://doi.org/10.1038/s41612-026-01555-w}
}