Optimization of dual-polarized bistatic specular scatterometer for studying microwave scattering response and vegetation growth parameters retrieval of paddy crop using a machine learning algorithm

摘要

Bistatic specular (Bi-spec) scatterometer measurement system was indigenously designed at X- and C-bands in the incidence angular range from 20 degrees to 60 degrees at the interval of 10 degrees to study the scattering mechanism of vegetation growth parameters of paddy crop and their retrieval at HH and VV polarizations using machine learning algorithms. The contributions of coherent and incoherent scattering to the total reflected or scattered power was measured by the bi-spec scatterometer system to derive bi-spec scattering coefficient(sigma(0)). The effect of vegetation growth parameters such as leaf area index (LAI), plant height (PH), vegetation water content (VWC) and fresh biomass (FBm) on the sigma(0)was investigated. The values of sigma(0)(HH) at C-band were observed to be higher at HH-polarization as compared to 4-Hat X-band for all specular incidence angles due to higher penetrating ability than the X-band. An approach was made to find out the optimum parameters of the bi-spec scatterometer system by correlation analysis between the computed sigma(0)and vegetation growth parameters of paddy crop and their retrieval by the SVR model using linear, polynomial and radial kernels. The optimum parameters of the bi-spec scatterometer system for the retrieval of LAI, FBm and VWC of paddy crop were found to be HH polarization, 40 degrees angle of incidence at C-band. While, for PH retrieval, the optimum parameters were found to be VV polarization, 40 degrees angle of incidence at X band. The potential of the developed SVR model was evaluated by computing centered root mean square error (CRMSE), standard deviation (SD) and correlation coefficient (R) between estimated and observed vegetation growth parameters. The retrieval of the vegetation growth parameters of paddy crop by the developed SVR model using radial kernel provided better results in comparison to linear and polynomial kernels for LAI, FBm and VWC at C-band and PH at X-band using bi-spec scatterometer data.