| Soil moisture retrieval using Global Navigation Satellite System Reflectometry(GNSS-R)technology has emerged as a focal point of research in recent years.However,given the complex surface conditions that influence GNSS-R soil moisture retrieval performance,existing methods often rely on empirical selection of environmental factors for modeling and inversion,while research on the varying impact of different environmental factors on soil moisture remains limited.Consequently,this study proposes a GNSS-R-based method combining multiple factors for soil moisture inversion,ultimately producing a regional high-precision 9km soil moisture product and incorporating an interpretability approach to explain the constructed inversion model.The primary content and research findings of this paper are as follows:(1)The method initially compares five algorithms–multiple linear regression(MLR),genetic algorithm-optimized BP neural network(GA-BP),extreme gradient boosting(XGBoost),random forest(RF),and light gradient boosting machine(LGBM)–selected from three aspects:linear models,neural networks,and machine learning models.The XGBoost algorithm,with superior comprehensive performance,is ultimately chosen as the GNSS-R-based soil moisture inversion model.(2)Twelve common environmental factors are selected,divided into four categories–topography,vegetation,land cover,and climate–and used to establish 20different factor combination schemes.Through experimental analysis,the optimal combination of factors influencing GNSS-R soil moisture inversion is determined to be:Lon+Lat+Doy+BRCS+Aspect+Slope+Elevation+PPT+Tmean+Land＿Cover.(3)The optimal inversion model is employed to combine the best factor combination with Cyclone Global Navigation Satellite System(CYGNSS)GNSS-R observation data,generating a regional high-precision 9 km soil moisture product.The results indicate that the highest accuracy for soil moisture inversion is achieved through a combination of topography(excluding Hillshade),climate,and land cover factors,with prediction accuracy RMSE,MAE,and r values of 0.0462cm~3/cm~3,0.0364cm~3/cm~3,and 0.9009,respectively.Among the considered environmental factors,topography has the most significant influence on soil moisture inversion,followed by precipitation and temperature,while NDVI and Hillshade suppress model prediction performance.This study provides a basis for selecting environmental factors when establishing GNSS-R soil moisture inversion models.(4)As the machine learning algorithms employed in this research exhibit a black-box characteristic,they pose difficulties in directly interpreting the decision-making process during model prediction.Therefore,this study innovatively introduces the interpretability method-SHAP to explain the model,aiming to provide a clearer understanding of how different factors impact the final soil moisture prediction outcome. |
