Combining Spaceborne GNSS-R Data and Ensemble Machine Learning Methods for the Retrieval Vegetation Optical Depth

Spaceborne Global Navigation Satellite System Reflectometry (GNSS-R) has been widely used in land scenarios, but there is limited research focused on Vegetation Optical Depth (VOD) retrieval. For this purpose, we propose a method for inverting VOD using a combination of spaceborne GNSS-R data and an ensemble machine learning algorithm. Integrating Bistatic Radar Cross Section (BRCS), Effective Scattering Area (ESA), CYGNSS variable parameters, and surface auxiliary parameters (i.e. soil moisture, surface temperature, and roughness coefficient) to construct an Extreme Randomized Tree (ET) ensemble machine learning model. The results show that the proposed ET model has a good correlation between VOD retrieval and SMAP VOD, with retrieval accuracy comparable to the Decision Tree (DT) model, but significantly better than the retrieval results of Adaptive Boosting (AdaBoost), Support Vector Machine (SVM), and Gradient Boosting Decision Trees (GBDT) models. The Root Mean Square Error (RMSE) has improved by 178.12%, 67.87%, and 85.26%, respectively, while the Mean Absolute Percentage Error (MAPE) has improved by 87.50%, 80.00%, and 92.31%, respectively. This indicates that the ET model is feasible for high-precision and high spatiotemporal resolution VOD retrieval on land.