Synergy Of Microwave Scattering Simulation And SVM Algorithm For Retrieval Of Biophysical Parameters In Wheat Fields

Wheat is one of the staple food crops in China.Timely,reliable and dynamic wheat growth monitoring and yield forecast prior to harvest play a role in making decision for the macroscopic regulation of wheat grain prices.Conventional methods in wheat monitoring are not only time and labor consuming,but also are destructive to sampling sites.It is hard to meet the demands of all-level management sections for information acquisition on wheat growth status and yield distribution,and further decision-making in agricultural productions.Thanks to the all-weather day-and-night observation ability and strong penetrability to vegetation and soil surfaces,synthetic aperture radar(SAR)has been becoming a promising approach to wheat growing monitoring.The study area was selected at wheat fields in the central part of North China Plain,which is a major production area of winter wheat.With in-depth analysis of wheat canopy structure,a physical model about wheat-field microwave scattering(WMSM)was proposed on the basis of microwave radiative transfer theory.A multi-output support vector machine regression(MSVR)algorithm was employed to retrieve biophysical parameters of the wheat fields at a regional scale simultaneously.Specifically,(1)On the basis of the vegetation canopy microwave scattering model,a wheat-field microwave scattering model was developed by fully analyzing the microwave signal transfer mechanisms.A series of physical parameters were measured in fieldworks synchronous or near-synchronous to radar satellite overpassing the study area.They were used as inputs of WMSM to simulate the C-band backscattering coefficients of wheat fields.Simulated results were validated by multi-source and multi-angle C-band remote sensing images(RADARSAT-2,Sentinel-1,GF-3)and a reasonable agreement was achieved with root-mean-square error(RMSE)of lower than1.8 dB for HH,VH and VV polarized data.With the physical model,the backscatter contributions of wheat canopy constituents(ear,leaf and stem)were analyzed in detail.In addition,this model was used to explore the response mechanism of canopy backscattering coefficients to different structural parameters of wheat fields.(2)With the forward theoretical model,a large number of sample datasets were simulated to train the multi-output support vector machine for an optimal parameter inversion model.With multi-source SAR data,the key structural parameters(ear length,ear diameter,ear number density,soil surface moisture)were estimated simultaneously.The estimated physical parameters of wheat fields agreed well with the measured data.The wheat ear length,ear diameter,number density and surface moisture were retrieved by the inversion scheme with RMSE of 0.75 cm,0.1 cm,119#/m~2 and 3.11%,respectively.This study confirmed the feasibility of the MSVR algorithm in key parameters retrieval of wheat field,and further provided valuable technological support for future wheat yield estimation.This thesis aims at a better understanding on the microwave scattering mechanisms of wheat-field through the simulation of wheat microwave scattering characteristics and parameter retrieval of physical parameters of wheat fields.Rapid inversion and mapping of wheat biophysical parameters can be achieved from multi-sources radar data by using multi-output support vector machine regression method.Therefore,this study illustrated a great potential of microwave remote sensing technology for crop growth monitoring.