Deep Learning-Based Automatic Identification of Gust Fronts from Weather Radar Data

Gust fronts, which are characterized by strong winds and intense wind shear, pose a threat to both aviation and public safety. To aid forecasters in issuing timely warnings for this hazardous weather phenomenon, a deep learning-based automatic gust front identification algorithm is proposed in this study. The algorithm utilizes Mask R-CNN, a state-of-the-art instance segmentation model, trained on a large dataset of 2623 gust front samples from S-band weather radar volume scans in East China and the North China Plain between 2009 to 2016. Extensive data preprocessing and manual annotation are performed to prepare the training dataset. The optimized model achieves impressive performance on a test set of 604 samples, with a probability of detection of 93.21%, a false alarm rate of 3.60%, a missed alarm rate of 6.79%, and a critical success index of 90.08%. The algorithm demonstrates robust identification capabilities across gust fronts of varying scales, types, and parent thunderstorm systems, highlighting its operational applicability.