Soil Moisture And Crop Biophysical Parameters Estimation From Time Series Of PolSAR Imageries

Accurately obtaining the knowledge of geophysical and biophysical parameters of surface and crop using remote sensing technique and a well understanding of climate change and crop status are of vital importance to ensure food security and agricultural productivity.Polarimetric synthetic aperture radar(PolSAR)is sensitive to the soil permittivity and has the advantage of collecting data with high spatial resolution owing to the active generation and the penetration ability of microwave.It is a powerful tool which could provide multi-temporal,multi-incidence,and multi-frequency information for soil moisture(SM)and vegetation parameters estimation.Currently,the inversion of soil moisture and crop parameters remains uncertainty due to the rough surface and vegetation canopy.How to decouple the scattering of soil and vegetation and improve the parameter inversion accuracy by incorporating the multiple PolSAR observations is known to be the frontier and hotspots.In this context,two works including soil moisture and crop biophysical parameters estimation are implemented by using time-series and multi-incidence PolSAR observations.For SM estimation,in order to decouple the scattering between soil and vegetation canopy and circumvent the problem of undetermined inversion of retrieval algorithms,we construct a multiple polarimetric decomposition(PD)model and modify the SM retrieval method by introducing the multi-incidence time-series observations.For crop biophysical parameter estimation,the time series of crop phenology(Biologische Bundesanstalt,Bundessortenamt,and CHemische,BBCH)and height parameters are estimated by means of dynamic model and particle filtering(PF)algorithm.The study of the thesis includes:1)the multi-temporal and/or multi-incidence polarimetric decomposition(PD)modeling:the framework is constructed by combining the X-Bragg model,extended double Fresnel scattering model,and generalized volume scattering model(GVSM).Compared with traditional PD models,it considers both the depolarization of dihedral scattering and the diverse vegetation contribution.Benefitting the introducing of multi-incidence and/or multi-temporal observations,the ill-posed inversion problem could be circumvented.The results show that over the selected agricultural fields the proposed retrieval framework provides an inversion accuracy of RMSE<6.0%.Specifically,the stem permittivity(SP)which is retrieved synchronously with SM also shows a linear relationship with crop biomass and plant water content.2)analyzing the influence factors of SM estimation:it is known that in the proposed multiple decomposition model the surface depolarization angle should be derived from the circular polarization correlation before the parameter calculating,Besides,in the dihedral scattering model the co-pol phase difference is empirically set to 0.In this work,the influence of surface depolarization angel and co-pol phase difference on the SM estimation are also investigated.different values of surface depolarization angel and co-pol phase difference are introduced in the proposed multi-temporal PD SM retrieval algorithms.It shows that the SM retrieval performance could be improved by using a non-zero co-pol phase difference and the surface depolarization angel which is calculated from PolSAR data obtained on the least vegetated date.3)the extended alpha approximation SM retrieval method:the original time series SM retrieval method is extended by incorporating multi-incidence data,to enlarge the observation space,and a PD approach,to separate in a better way the response of the ground from that of the vegetation.In addition,we also include a modification in the way the bound conditions of the soil dielectric constant are defined.The soil dielectric bound conditions are derived from the Dubois model and the priori knowledge of soil roughness.Results show that compared with the original method,the retrieval accuracy is improved when the PD theorem and multi-incidence observations are included in the approach.It is also found that the SM estimation at VV(RMSE=4.64cm³/cm³)channel is better than at the HH(RMSE=7.46 cm³/cm³)channel.4)the crop biophysical parameter estimation by means of the dynamic model and the particle filtering:in this work the BBCH and the height evolution models are constructed by fitting the site observations with the date of sowing(Do S)over rice fields.Whereas the SAR observation models are obtained by fitting the SENTINEL-1A/B backscattering coefficients of VV(₍(1(1)),VH(₍₁₎and the power ratio(₍(1)/₍(1(1))with the site observations(i.e.BBCH and height).Both the evolution and observation models are embedded into the particle filtering algorithm for the biophysical parameter estimation.The results show that the accuracy of BBCH and crop height estimates is RMSE=3.7 and RMSE=4.1,respectively,at the beginning of the growth stage(BBCH<40).However,underestimation is found due to the saturation of SENTINEL-1 data in the middle and late stage of rice phenology.It should also be noted that this study has a great potential to provide priori vegetation information for decoupling the surface and vegetation scattering on SM retrieval.