A Deep Neural Networks-based Significant Wave Height Inversion Method for GNSS-R Signals

Significant wave height (SWH), a crucial parameter in monitoring oceanic conditions, holds significant importance in the realms of wave forecasting, disaster preparedness, and risk mitigation. Compared with satellite altimetry and other technologies, the Global Navigation Satellite System-Reflectometry (GNSS-R) technology boasts distinctive advantages that set it apart. This paper has proposed a technique for extracting SWH from GNSS-R data, utilizing Deep Neural Networks (DNN). In the data preprocessing stage, model input parameters are extracted from the downloaded CYGNSS SWH data and ERA5 SWH data. After denoising and filtering the data, an appropriate spatiotemporal matching window is selected to obtain the matched CYGNSS data as the input for the model, with ERA5 data as the label. Choose DNN to handle the regression problem of SWH inversion. Finally, the SWH values inverted by the method proposed in this paper are compared and analyzed with the SWH values of ERA5 as labels. The outcomes demonstrate enhanced performance of the inversion method based on DNN, resulting in a reduced root mean square error (RMSE) and an improved correlation coefficient (CC). Specifically, the achieved RMSE falls below 0.55m, while the CC reaches 0.69. Additionally, this suggests that the presented inversion method demonstrates commendable performance.