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

TL;DR

This paper introduces PhyDL-NWP, a physics-guided deep learning framework that enhances weather downscaling and forecasting by integrating physical laws, resulting in faster, more accurate, and physically consistent predictions.

Contribution

It presents a novel physics-informed deep learning approach that combines physical equations with data-driven models for improved weather prediction and downscaling.

Findings

Achieves up to 170x faster inference speed.

Improves forecasting accuracy and physical consistency.

Enables resolution-free weather modeling.

Abstract

Weather forecasting is essential but remains computationally intensive and physically incomplete in traditional numerical weather prediction (NWP) methods. Deep learning (DL) models offer efficiency and accuracy but often ignore physical laws, limiting interpretability and generalization. We propose PhyDL-NWP, a physics-guided deep learning framework that integrates physical equations with latent force parameterization into data-driven models. It predicts weather variables from arbitrary spatiotemporal coordinates, computes physical terms via automatic differentiation, and uses a physics-informed loss to align predictions with governing dynamics. PhyDL-NWP enables resolution-free downscaling by modeling weather as a continuous function and fine-tunes pre-trained models with minimal overhead, achieving up to 170x faster inference with only 55K parameters. Experiments show that PhyDL-NWP improves both forecasting performance and physical consistency.