DenMerD: A Feature Propagation Enhanced Approach to Beam Blockage Correction in Weather Radar

Liu, Qi; Sun, Jiawei; Liu, Xiaodong

Abstract

In the realm of meteorological research, extensive global radar networks serve to detect and provide early warnings for a diverse array of weather phenomena. However, the inherently discontinuous nature of radar observations often results in the presence of small-scale data inconsistencies or voids. These data anomalies can be primarily attributed to obstructions leading to gaps and clutter stemming from atmospheric refraction, equipment malfunctions, electromagnetic interference, among other factors. Conventional techniques for rectifying radar echo blockage involve interpolation and supplementation of missing data using proximate radial information, but such methods fail to effectively exploit the underlying patterns within radar data and exhibit considerable limitations. The present study approaches the issue of weather radar echo blockage correction from the perspective of image restoration and proposes a sophisticated deep learning network called DenMerD incorporating densely connected and merge-distribution (MD) transition units. The model improves key metrics, such as CSI and POD in meteorological domains, by 30.7% and 30.1% compared to existing models. It also excels in SSIM metrics, showcasing the effectiveness of this design in enhancing feature propagation and beam blockage accuracy.

Citation

Liu, Q., Sun, J., & Liu, X. (in press). DenMerD: A Feature Propagation Enhanced Approach to Beam Blockage Correction in Weather Radar. Journal on Artificial Intelligence,

• Journal Article Type: Article
• Acceptance Date: Oct 11, 2023
• Deposit Date: Dec 11, 2023
• Journal: Journal on Artificial Intelligence
• Print ISSN: 2579-0021
• Electronic ISSN: 2579-003X
• Publisher: Tech Science Press
• Peer Reviewed: Peer Reviewed
• Keywords: Beam blockage correction, weather radar, image restoration, deep learning
• Publisher URL: http://techscience.com/journal/jai
• Related Public URLs: https://doi.org/10.21203/rs.3.rs-3573950/v1