Deep Learning for Weather Radar Beam Blockage Correction: Potential Impacts on QPE over Complex Terrain in Northern California

Abstract In weather radar applications, high-quality data must be collected to ensure accurate mapping and analysis of precipitation dynamics. Partial beam blockage occurs when radar signals encounter objects such as tall buildings, mountains, or trees, resulting in incomplete or distorted radar observations. Mitigating the impact of beam blockage or partial beam blockage is necessary to improve the accuracy of precipitation estimation. This study implements an advanced deep learning technique based on generative adversarial networks (GANs) to address the beam blockage issue in radar observations over complex terrain regions during large-scale winter stratiform precipitation events in Northern California. The potential impacts of blockage correction on radar quantitative precipitation estimation (QPE) are quantified using disdrometer data as ground truths. The results indicate that the beam blockage correction model effectively addresses the reflectivity continuity challenges posed by beam blockage. The beam blockage corrected radar reflectivities and QPEs agree very well with ground truths. The normalized standard error of QPE was reduced by 39.7% due to improvement on the radar rainfall rates, indicating the model’s potential value in enhancing radar-based QPE over complex terrain, which is essential for operational hydrometeorological monitoring and risk mitigation. Significance Statement The most prevalent issue encountered in radar data quality control is the presence of low-quality or even absent data. Incomplete and discontinuous data frequently have significant implications for meteorological analysis and disaster forecasting, among other applications. In such instances, beam blockage due to terrain is the most prevalent cause, particularly in the context of complex mountainous topography. Here, we apply a deep learning model to correct radar beam blockage and thus improve quantitative precipitation estimation over complex terrain. Key findings show that the model effectively recovers blocked radar data that closely match ground-truth measurements from the disdrometer and greatly reduce the QPE error introduced by the data. This improved accuracy is critical for better accurate data processing and quantitative analysis over complex terrain in the western United States.