Exascale Hybrid Numerical-AI Ensembles for Operational Flood-Season Forecasting in East Asia: 15-km Decadal Hindcasts and 1-km High-Resolution Capability

Mengxuan Chen, Yunpu Xu, Qiuyan Sun, Han Zhang, Jiayi Lai, Zheng Zhou, Juepeng Zheng, Hongsong Meng, Nan Wei, Jinxiao Zhang, Xiongchuan Tan, Haodong Bian, Yinan Cai, Ge Yang, Fang Wang, Yunyun Liu, Conghui He, Runmin Dong, Lanning Wang, Yutong Lu, Yongjiu Dai, Haohuan Fu*.

• Seasonal forecasting of summer rainfall in East Asia remains a grand challenge, as predictability at 3 to 6 month lead times is constrained by the spring predictability barrier, weak large-scale signals, and localized nonlinear convective extremes. We address this challenge with CAPES, which integrates a kilometer-resolution coupled regional model with atmosphere, land, and ocean components and a data-driven AI seasonal forecasting system. At 15 km resolution, the fused workflow combines 174 numerical members from varying start times, physics schemes, and parameter perturbations with 1,600 AI members generated from initial and physical perturbations. Using the full LineShine system, CAPES completes ten annual 1,774-member hindcasts for 2016 to 2025 within 14.6 hours, improving the mean prediction score from ECMWF’s 71.8 to 75.9 and delivering a major gain in operational forecasting capability. The 1-km configuration further enables fine-scale typhoon simulation and establishes the feasibility of kilometer-scale fused ensemble forecasting on a one-week timescale.
• The development of CRESM was primarily carried out by Han Zhang, and the HPC optimization and deployment of CRESM were primarily carried out by Yunpu Xu.

SeasonBench-EA: A Multi-Source Benchmark for Seasonal Prediction and Numerical Model Post-Processing in East Asia

Mengxuan Chen, Guowen Li, Ziheng Zou, Fang Wang, Jinxiao Zhang, Runmin Dong, Juepeng Zheng, Haohuan Fu* (2025)

• SeasonBench-EA is a multi-resolution, multi-source benchmark dataset for seasonal prediction in East Asia. It integrates ERA5 reanalysis data, covering key atmospheric variables and boundary conditions, together with ensemble seasonal forecasts from leading operational centers. SeasonBench-EA enables two core tasks: (1) machine learning–based prediction using reanalysis inputs, and (2) post-processing of ensemble forecasts. Beyond standard deterministic and probabilistic metrics, it also includes a hindcast evaluation framework to assess long-term predictive skill and robustness.

An Interpretable Weather Forecasting Model With Separately-Learned Dynamics and Physics Neural Networks

Mengxuan Chen, Jinxiao Zhang, Runmin Dong, Yidan Xu, Haoyuan Liang, Juepeng Zheng, Lanning Wang, Haohuan Fu* (2025)

• A lightweight weather forecast model with graph network and multi-layer perceptron is designed to depict dynamics and physics separately. The design of the graph follows the Arakawa C-grid, with the wind speeds embedded as edge features to simulate the large-scale dynamics. Correlations between graph parameters and atmospheric processes highlight model’s interpretability, paving way for more reliable predictions.

ResU‐Deep: Improving the Trigger Function of Deep Convection in Tropical Regions With Deep Learning

Mengxuan Chen, Haohuan Fu*, Tao Zhang, Lanning Wang* (2023)

• A location-aware and deep-learning-based deep convection trigger function is proposed to improve the diurnal cycle simulation in tropics. Results show that terrain information, temporal dependence of convection, and water vapor content are essential for predicting convection. Also, including information from neighboring atmospheric columns can improve the performance of the deep convection trigger function.