Physics-Guided Learning of Meteorological Dynamics for Weather Forecasting and Downscaling

Yingtao Luo, Shikai Fang, Binqing Wu, Qingsong Wen, Liang Sun

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Weather Forecasting, Weather Downscaling, Spatiotemporal Modeling, Partial Differential Equations, Physics-Guided Learning

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: We propose a physics-guided learning framework to capture the nonlinear dynamics of meteorology and align deep learning models with the corrected physical mechanism to improve generalization.

Abstract: Weather forecasting is of paramount importance for a myriad of societal and scientific applications. Traditionally, numerical weather prediction (NWP) methods based on physical principles are computationally intensive and can struggle with the inherent complexity of atmospheric dynamics. Recently, deep learning techniques have shown promise in weather prediction, but the long-term generalization and physical consistency of pure data-driven approaches remain challenging. In this paper, we introduce a novel physics-guided approach for numerical weather prediction that combines the strengths of both physical mechanism and deep learning, namely PhyDL-NWP. Our method can capture the nonlinear dynamics of meteorology and align deep learning models with the underlying physical mechanism to improve generalization. Extensive experiments on real-world weather datasets show that our model can significantly improve the performance of deep learning methods in a wide range of tasks from forecasting to downscaling.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 1523