Wave Height Estimation From X-Band Marine Radar Data Using SWHFormer Method

This paper presents an innovative deep learning-based approach to measuring sea significant wave height (SWH) from shore-based X-band radar data. Initially, the first image from each minute of the radar sequence acquired by the X-band radar is extracted and sub-images covering the ocean area are intercepted for analysis. Subsequently, these extracted radar sub-images are fed into a SWH regression network based on the Vision Transformer (ViT) model, named SWHFormer, for training the model and conducting real-time SWH measurements. The radar data used in the validation experiment were collected at Guadalupe Dunes, CA, USA. Concurrent hourly SWH data provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) are employed as ground truth. Comparative analysis with several existing SWH estimation methods reveals that the proposed SWHFormer-based method achieves a relatively superior SWH estimation performance.