Spherical Physics-informed Neural Operator with Multi-scale Coupling for Meteorological Downscaling
Abstract
Meteorological downscaling is crucial for high-resolution regional climate forecasting and disaster early warning. While neural operators have emerged as a promising paradigm for modeling complex spatiotemporal mappings, existing frameworks often struggle with spherical manifold geometric distortions, inherent atmospheric multi-scale coupling mismatches, and lack of explicit atmospheric laws. We propose the Spherical Physics-informed Neural Operator, which utilizes a Spherical Laplacian Decomposition to partition atmospheric fields into hierarchical frequency components, maintaining exact point-wise correspondence across scales. To evaluate these representations at arbitrary locations, we introduce a localized spherical integral operator that approximates continuous kernel transforms via geometry-aware attention. Dynamical consistency is further enforced by embedding differentiable constraints into the learning process. Extensive experiments demonstrate that our framework attains superior accuracy and zero-shot generalization across various meteorological variables and unseen queries, representing a robust and interpretable solution for global-to-regional meteorological downscaling.
Versions
Related
- A Geometry-Informed Neural Operator surrogate of LISFLOOD-FP for rapid pluvial-fluvial flood forecasting: the case of Forlì, Italy
- Deep learning-based downscaling of PM2.5, NO2 and O3 over Europe across emission-reduction scenarios
- Characteristics and simulation of ozone pollution in Haikou City under combined effects of meteorological factors and pollutants
- A support vector machine classifier of PM₂.₅-based air quality index categories from co-pollutants and meteorological variables in Ahvaz, Iran
- Widespread acceleration of drought propagation from meteorological to ecological systems in the northern hemisphere under climate warming
BibTeX
@misc{ye2026spherical,
title = {{Spherical Physics-informed Neural Operator with Multi-scale Coupling for Meteorological Downscaling}},
author = {Yiqiang Ye and Yichi Wang and Haodi Zhang and Jiawei Wen and Jiahui Jiang and Zhaoyu Zhong and Jiangjian Yu and Chunxia Xiao},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.24708},
eprint = {2610.24708},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.24708},
}
RIS
TY - UNPB
TI - Spherical Physics-informed Neural Operator with Multi-scale Coupling for Meteorological Downscaling
AU - Yiqiang Ye
AU - Yichi Wang
AU - Haodi Zhang
AU - Jiawei Wen
AU - Jiahui Jiang
AU - Zhaoyu Zhong
AU - Jiangjian Yu
AU - Chunxia Xiao
PY - 2026
DA - 2026/10/17
PB - arXiv
DO - 10.48550/arxiv.2610.24708
UR - https://arxiv.org/abs/2610.24708
AB - Meteorological downscaling is crucial for high-resolution regional climate forecasting and disaster early warning. While neural operators have emerged as a promising paradigm for modeling complex spatiotemporal mappings, existing frameworks often struggle with spherical manifold geometric distortions, inherent atmospheric multi-scale coupling mismatches, and lack of explicit atmospheric laws. We propose the Spherical Physics-informed Neural Operator, which utilizes a Spherical Laplacian Decomposition to partition atmospheric fields into hierarchical frequency components, maintaining exact point-wise correspondence across scales. To evaluate these representations at arbitrary locations, we introduce a localized spherical integral operator that approximates continuous kernel transforms via geometry-aware attention. Dynamical consistency is further enforced by embedding differentiable constraints into the learning process. Extensive experiments demonstrate that our framework attains superior accuracy and zero-shot generalization across various meteorological variables and unseen queries, representing a robust and interpretable solution for global-to-regional meteorological downscaling.
ER -