Retrieval Of Soil Moisture And Organic Matter Contents Considering Spectral Radiative Transfer Theory

Soil,one of essential natural resources,is closely related to the food,ecological,water security and many other aspects etc.Hyperspectral remote sensing technology provides an efficient approach for the scientific and appropriate monitoring of soil properties spatially and periodically,which contributes to the soil environmental quality assessment.In this thesis,the theory of soil radiative transport model has been studied to improve the deficiency of the hyperspectral statistical inversion algorithm on soil property information.Moreover,the inversion model for soil moisture and soil organic matter content is constructed based on the physical mechanism,which provides a new approach to handle the issues of retrieving soil environmental quality parameters.The main research contents and conclusions of this paper are as follows.（1）The soil moisture spectral reflectance model based on the advanced Hapke model（SMR-Hapke）has been presented.Firstly,the radiative transfer process of scattering on the surface of wet soil is analyzed,and the component from bulk scattering has been calculated using Hapke model considering the effects of Fresnel scattering.After that,the relationship between the soil optical parameters（absorption and scattering coefficient）and moisture content are further developed.Finally,the SMR-Hapke model applied to the soil moisture retrieval is established.To demonstrate the performance of the proposed model,Lobel-Tian laboratory soil spectral dataset was carried out for experiments.The results show that the model is capable to fit the soil spectra well at different moisture content gradients,with the MSE usually less than 3×10^-4.Moreover,the SMR-Hapke model achieved a good accuracy of moisture retrieval on the entire wavebands,with R² greater than 0.8,which demonstrates the good applicability of the proposed model.（2）The normalized difference soil moisture index model(NDSMI_{Ha pke})has been proposed.The SMR-Hapke model is further simplified to a linear form using the moisture-sensitive band on the short-wave infrared range,which reduces the complexity of calibration step significantly.The linear relationship between the converted reflectivity F and soil moisture content has been verified.Meanwhile,a novel normalized difference soil moisture spectral indexNDSMI_{Ha pke} is proposed by combining the 2190nm and 1610nm bands which are commonly used in multispectral satellite remote sensing data.The validity of theNDSMI_{Ha pke} model was verified on the Yitong soil moisture-spectral dataset,and the experimental results show that NDSMI_{Ha pke} obtains the best model performance for the soil moisture content estimation,with R²values up 0.8138 and 0.8801 on the training and test sets,respectively.Finally,the soil moisture distribution was mapped based on the Sentinel-2 multispectral data,which verified the applicability of the proposed model on larger scales.（3）The soil organic matter inversion method based on Single Scattering Albedo and Multivariate Curve Resolution（SSA-MCR）has been developed.The Hapke model is utilized to convert the nonlinear soil spectra into a linear combination of Single Scattering Albedo（SSA）based on the nonlinear mixing theory of soil spectral.Then,the Nonnegative Matrix Factorization algorithm and the idea of Multivariate Curve Resolution are combined to decompose the organic matter basis vector and mass fraction vector from the soil SSA matrix.Finally,the retrieval of soil organic matter is achieved by a fully constrained least squares algorithm.Experimental results show that the proposed SSA-MCR model obtains high accuracy with R² values up to 0.6560 and0.5518 on the training and validation sets,respectively.Two methods based on Spectral Absorption Characteristics（SAC）and Semi-supervised Deep Neural Network（Semi-DNN）are utilized as counter parts.The results show that the three methods are well consistent and the correlation coefficient is greater than 0.6.Triple Collocation algorithm is introduced to evaluate and analyze the error distribution of the three models,and the results show that the squared correlation coefficient between the SSA-MCR model and the unknown true value is 0.872,which outperforms the other two models.Finally,the relative accuracy of each model within different organic matter classes is verified using the Categorical Triple Collocation algorithm,which demonstrates the superior performance of the SSA-MCR model.The thesis has 47 figures,15 tables,and 139 references.