Spaceborne GNSS-R Wind Speed Retrieval Using Machine Learning Methods

This paper focuses on sea surface wind speed estimation using L1B level v3.1 data of reflected GNSS signals from the Cyclone (CYGNSS) mission and European Centre for Medium-range Weather Forecast Reanalysis (ECMWF) data. Seven machine learning methods are applied retrieval, i.e., Regression trees (Binary Tree (BT), Ensembles Trees (ET), XGBoost (XGB), LightGBM (LGBM)), ANN (Artificial neural network), Stepwise Linear (SLR), Gaussian Support Vector Machine (GSVM), a comparison their performance is made. The divided into two different ranges to study suitability algorithms. A total 10 observation variables considered as input parameters importance individual or combinations thereof. results show that LGBM model performs best with an RMSE 1.419 correlation coefficient 0.849 in low interval (0–15 m/s), while ET 1.100 0.767 high (15–30 m/s). effects used retrieval models investigated metric, showing number play very significant role retrieval. It expected these will provide useful reference development advanced algorithms future.