Research On Soil Moisture Retrieval By Means Of Multi-source And Time-series SAR Data

Soil moisture is an essential parameter controlling the land surface process that impact water and energy exchanges at the land-atmoshere interface.The real-time and dynamic knowledge of soil moisture is significant for agriculture,ecology and hydrology.Synthetic aperture radar(SAR)was extensively applied to monitor soil surface moisture.The backscatter information is influenced by several factors,including the radar configuration parameters,soil dielectric constant,surface roughness and vegetation cover.Therefore,the influence of soil surface roughness and vegetation cover need to be effectively removed for soil soil moisture retrieval.This thesis focuses on investigating the models and algorithms for soil moisture retrieval on the basis of multi-sensor and time-series SAR data.The multi-temporal Radarsat-2,Terra SAR-X and Sentinel-1A data were applied to retrieve soil moisture content over Hebei and Ningxia study areas.The influence of surface roughness was reduced or eliminated by the combination of multi-polarization,multi-band and multi-temporal SAR data.The empirical,semi-empirical and physical models were employed to retrieve soil moisture content.The main achievements and innovations of this dissertation are as follows:(1)A multi-feature soil moisture retrieval method based on PCA(principal component analysis)dimensionality reduction technique was developed.Polarimetric Radarsat-2 data were used to extract the backscatter coefficients and polarimetric variables.The optimal input features for soil moisture retrieval were selected by means of PCA dimensionality reduction and least RMSE(root mean square error)criterion.The SVR(support vector regression)model was employed to estimate soil moisture content.The availability of this approach was validated based on four data sets over Ningxia study area.(2)The multi-band SAR data was integrated to estimate soil moisture content over agricultural areas.The integration of multi-band SAR data effectively characterized the soil surface parameters affecting the SAR signal.The experimental results demonstrated that the application of multi-band SAR data improved the soil moisture inversion results.The approach integrated the water cloud model and calibrated IEM was developed to estimate soil moisture content with multi-band SAR data as inputs over vegetation covered area.The accurate soil moisture content was obtained over corn covered agricultural area.(3)Multi-sensor SAR data was investigated to retrieve soil moisture content based on physical models over bare agricultural area.The application of multi-sensor SAR data decoupled the effect of soil moisture and surface roughness.The lookup table method was employed to estimate soil moisture content with multi-sensor SAR data as inputs.Furthermore,the distribution graphs of the cost function were introduced to evaluate uniqueness and convergence of the estimated results.The reliability of the multi-sensor SAR data for soil moisture retrieval was demonstrated.(4)The relationship between the variation of soil moisture and backscatter coefficient change was investigated based on simulation data sets.For bare soil surface with constant surface roughness,the variation of soil moisture showed good correlation with the backscatter coefficient change.The multi-temporal Radarat-2 and Sentinel-1A data were used to estimate soil moisture change over Handan study area.The variation of backscatter coefficient can effectively represent the soil moisture change.Therefore,the multi-temporal SAR data can be used to estimate soil moisture change over agricultural areas.(5)The Alpha model integrated multi-temporal Radarsat-2 and Sentinel-1A data was developed to estimate soil moisture content over agricultural areas.The simulation data sets,multi-temporal SAR data and field measurements were used to evaluate and validate the rationality of the Alpha model.The observation equations can be established based on the Alpha model and multi-temporal SAR data.Then the soil moisture content can be calculated in combination with the constraint of soil moisture.In this process,the number of the observation equations is less than the unknown parameters.In order to avoid the uncertainty of the underdetermined equations,the integration of multi-temporal Radarsat-2 and Sentinel-1A data were used to establish the observation equations,which increased the effective observation equations.Therefore,the solution of the observation equations was converted to overdetermined equation.The practicability of this method was demonstrated based on the in-situ measurements.