HyMeshAI: Deep learning enabled three-dimensional adaptive mesh generator for high-resolution atmospheric simulations

March 6, 2026

The paper titled “HyMeshAI: Deep learning–enabled three-dimensional adaptive mesh generator for high-resolution atmospheric simulations” was recently published in the Journal of Computational Physics. The study was led by Pu Gan, with co-authors Jinxi Li, Fangxin Fang, Xiaofei Wu, Jiang Zhu, Zifa Wang, Mingming Zhu, and Xun Zou. This work supports CW3E’s Strategic Plan (Atmospheric Rivers and Extreme Precipitation Research, Prediction, and Applications) and advances AI applications and model development in high-mountain regions.

Adaptive mesh refinement (AMR) enables seamless multi-scale simulations in numerical weather prediction (NWP) models, yet its three-dimensional implementation remains challenging because dynamically evolving meshes require frequent reconstruction and topology updates not supported by traditional NWP frameworks. Here we develop HyMeshAI, a hybrid deep-learning framework that combines CNN-based mesh-density prediction with ANN-driven nodal positioning to achieve end-to-end dynamic mesh generation within an AMR atmospheric model (Fig. 1). HyMeshAI retains the computational efficiency of adaptive refinement while addressing key challenges associated with dynamic meshes. Unlike most data-driven AI approaches that rely on fixed-dimension feature matrices, HyMeshAI distills AMR into two static descriptors: mesh-point cardinality and the spatial probability distribution of mesh generation. Performance is evaluated using idealized advection and rising-bubble experiments. In the advection test, HyMeshAI reduces mesh counts by 20-40% and improves mesh quality while maintaining comparable accuracy. In the rising-bubble case, it reproduces buoyancy-driven convection and resolves fine-scale structures, including Kelvin–Helmholtz vortices, with a 30-40% improvement in mesh quality.

Figure 1. A visual description of HyMeshAI (graphical abstract from Gan et al. 2026).

Citation:

Gan, P., Li, J., Fang, F., Wu, X., Zhu, J., Wang, Z., Zhu, M., & Zou, X. (2026). HyMeshAI: Deep learning enabled three-dimensional adaptive mesh generator for high-resolution atmospheric simulations. Journal of Computational Physics, 554, 114760. https://doi.org/10.1016/j.jcp.2026.114760