Inversion Of Gnss-r Sea Surface Wind Field Based On Machine Learning Tree Model

Ocean surface wind field is an important dynamic parameter of the ocean,which is closely related to almost all ocean activities.Global Navigation Satellite System(GNSS)can not only be used for traditional navigation and positioning,but also for remote sensing applications using its forward scattering signals.This technology is called GNSS-Reflectometry(GNSS-R).GNSS-R technology has attracted wide attention due to its low cost,all-weather,high spatial and temporal resolution,wide coverage and other advantages.After decades of development,GNSS-R technology has been applied in ocean height,sea surface wind field,sea surface oil spill,soil moisture and other aspects.In this paper,GNSS-R technology combined with machine learning tree model algorithm is used to retrieve sea surface wind field.In this paper,the original sample set is obtained by space-time matching of TDS-1(TechDemoSat-1)satellite and ECMWF(European Center for Medium-Range Weather Foresting)analysis data.After pretreatment of the original sample set,the training set and verification set suitable for machine learning tree model learning are obtained.The training set is used to train the tree model,and the verification set is mainly used to test the inversion accuracy of the learner.Five commonly used tree model decision trees,random forests,GBDT,LightGBM and XGBoost are used to retrieve sea surface wind field based on sliding windows.For the inversion of wind speed,the inversion error of integrated tree model is less than 2m/s,which meets the inversion requirement.Because the wind direction sensitive physical parameters may not be extracted,the inversion error of wind direction is larger,about 50 degrees.On this basis,random forest model and LightGBM model with better wind speed inversion effect are selected to fuse the model.Averaging,CV-Averaging and Stacking model fusion methods are used to further improve the accuracy of wind speed inversion.Then,further research and analysis based on random forest and LightGBM are carried out,which has important guiding significance for improving the accuracy of sea surface wind field inversion.It mainly includes the following three aspects:(1)With the equator as the center,the inversion accuracy of the tree model with latitude changes on both sides;(2)For the inversion of wind fields in different months,the differences in inversion accuracy between the northern and southern hemispheres;(3)Comparison and analysis of wind field inversion results for different time periods of day and night.In this paper,the sea surface wind field inversion is realized based on the machine learning tree model algorithm.Under the premise of extracting the sensitive physical parameters of the wind field,good results can be obtained.