ISSN 1000-0526
CN 11-2282/P
CN 11-2282/P
The Temperature Forecasting Method of Physics-Constrained Deep Learning Integrated with Pangu-Weather Model Forecasting Products
- Article
- Figures
- Metrics
- Preview PDF
- Reference
- Related
- Cited by
- Materials
Abstract:
Aiming at the challenge of fine-scale forecasting of 2-meter temperature (T2m) in complex terrain areas, this study selects the Guangxi region, a typical area with complex terrain, as the research object and proposes a physics-constrained deep learning forecasting model named PSD-Net, which integrates the forecast products of the Pangu-Weather model. The forecast products of Pangu-Weather (PANGU) are used as feature variables input.The generator based on the Super-Resolution Generative Adversarial Network (SRGAN) is employed to extract multi-scale features.Power-spectral-density (PSD) and Kullback–Leibler (KL) divergence terms are explicitly injected into the loss function as regularizers, enforcing strict spectral fidelity and distributional congruence between the forecast and the observed state.Comparing the T2m forecasting performance of the ECMWF, SCMOC and PANGU products in the Guangxi region in 2024, the results show that PSD-Net outperforms the comparative products at both grid and station forecasting. The grid absolute mean error is reduced by 37.6 % compared with PANGU, and the accuracy is improved by 17 percentage points. The error growth is gentle from 25 to 72 hours. This study verifies the effectiveness of the physics-constrained deep learning framework in fine-scale T2m forecasting and provides a new approach for the combination of meteorological and AI models.
Keywords:
Project Supported:
Clc Number:
Mobile website
