Ocean Swell Height Estimation From Spaceborne GNSS-R Data Using Tree Model Based Machine Learning Methods

Global navigation satellite system-reflectometry (GNSS-R), as an emerging observation method, has recently been applied to the retrieval of swell height. Existing research typically uses feature observables extracted from delay Doppler maps (DDMs) to construct empirical geophysical model functions (GMFs) for estimating swell height. However, using multiple variable parameters as model inputs poses significant challenges. For this purpose, we study the development of a multivariate regression model for swell height estimation using spaceborne GNSS-R data from the CYGNSS satellite and machine learning methods based on tree models (i.e. Random Forest (RF), Decision Tree (DT), Gradient Boosting Decision Tree (GBDT), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM)). To test the performance of the proposed model, ERA5 and Wavewatch III (WW3) data were used as reference data. The results showed that when ERA5 was used as the reference data, the proposed RF model performed the best among the five models, with root mean square error (RMSE) and correlation coefficient (CC) being better than 0.411 m and 0.83, respectively. When WW3 was used as reference, the RMSE and CC are better than 0.496 m and 0.84 m, respectively. The method proposed in this article provides a new approach for inverting global swell heights using CYGNSS data.