Research on Nonlinear Inversion of Vegetation Water Content Based on Multiple Ground-Based GPS-IR

It is of great significance to timely and accurately monitor the vegetation water content (VWC) for the study of plant growth status, drought assessment, and fire risk forecasting. Global Positioning System Interferometric Reflectometry (GPS-IR) is a new type of remote sensing technology. The signal-to-noise ratio recorded by the measurement receiver can be used to retrieve VWC effectively. Although studies have fully proved that the change of VWC can be effectively reflected by the normalized microwave reflectance index (NMRI), the NMRI based on a single ground-based GPS-IR cannot achieve spatial continuity. This paper makes full use of the advantages of multiple ground-based GPS-IR combined with vegetation index, weather, topography and other related elements, and a nonlinear inversion method of high spatial resolution VWC based on multiple ground-based GPS-IR is proposed. Firstly, First, various elements are acquired through Google Earth Engine, and the resolution is unified through ArcGIS; Then, the relationship model between the multiple ground-based GPS-IR and related elements is established through the least square support vector machine (LS-SVM), and the modeling effects of different model input variables are compared and analyzed, and the best modeling solution is selected; Finally, after model testing, the accurate inversion of VWC is realized. Taking a certain area of the United States as the research object, experiments show that: 1) Different fac-tors have different influences on modeling accuracy. Three vegetation indices of NDVI, GPP and LAI > topography > longitude and latitude of the station > daily average rainfall and daily average temperature. Among them, NDVI is a key element that affects modeling accuracy. 2) Using LS-SVM can effectively integrate multiple ground-based GPS-IR and related elements, and the model fitting process is relatively stable; the 16Day/500 m resolution NMRI image obtained by model inversion can better reflect regional VWC changes. 3) The model inversion error is relatively stable. The maximum error is only 0.031, and the root mean square error is only 0.040.