Decomposition Methods for the Estimation of Bare Soil Moisture Using Fully Polarimetric Sar Data

Soil moisture plays a critical role in the surface energy balance at the soil-atmosphere interface [4]. Using radar backscattering to retrieve soil moisture is a method of major concern due to its sensitivity to soil moisture [14] [15].In the past decades, many empirical, semi-empirical and physical models that relate the measured 0 to volumetric soil moisture mv have been carried out[14][15]. Integral Equation Model (IEM) [2] as an electromagnetic wave scattering and emission model is widely used to predict target’s backscattering coefficients due to its larger applicable region than Kirchhoff model [3] and SPM [3]. The recently proposed Advanced Integral Equation Model (AIEM) [6] [7] is used in this paper because of its several major modifications and higher accuracy. With the increasing exploration on fully polarimetric SAR data, target polarimetric decomposition of fully polarimetric SAR data has gained much attention in respect that it can generate an average or dominant scattering mechanism for the purpose of either classification or inversion [10]. Shi, et al. [9] only certified that employing Cloude decomposition has an improvement for estimating bare surface soil moisture. However, few studies have compared the effectiveness of applying different decomposition methods to moisture retrieval. In this paper we apply two different target polarimetric decomposition methods based on filtered fully polarimetric SAR data. According to the comparison of AIEM simulation results, we have validated the effectiveness of applying target polarimetric decomposition compensation to fully polarimetric SAR data for soil moisture estimation in bare soil case. Finally we conclude that the result of Cloude decomposition matches the theoretical results better. In the first step, we simulate the radar backscattering coefficients by inputting synchronous in-situ ground parameters to AIEM. When isotropic surface is assumed, it can be reduced to general form: 0 , , qp r RMS C AIEM h L , where RMS h and C L denote the root mean square(RMS) of surface height and the * Thanks to the R&D Special Fund for Public Welfare Industry of China (Meteorology): (GYHY200806022), the National Natural Science Foundation of China (40771148), and the High-Tech Research and Development Program of China (2009AA12Z128 and 2008AA121806-04) for funding. correlation length respectively; r represents dielectric constant of bare soil which is calculated by ( , , , ) r f mv sc T freq , Where mv , sc , T and freq stand for soil moisture, characteristic and temperature and microwave frequency separately. In the second step, to reduce/restrain the speckle noise, we need to filter SAR data before any manipulation. J. S. Lee polarimetric Refined Filter [5] with 7 7 window is applied on 3 T to reduce the speckle noise and then the filtered 3 T is converted to 3 C according [5]. In the next step, Cloude decomposition [8] is performed on the filtered 3 T matrix to extract the first component that represents the dominant surface scattering. Then the decomposed 3 T is converted to 3 C to retrieve backscattering coefficients. For the Freeman decomposition [1], it is applied on filtered 3 C matrix to get the surface scattering coefficients. Finally after geometric correction, the comparison between AIEM predictions versus filtered polarimetric data, Cloude decomposed and Freeman decomposed results can be conducted respectively. For the experimental sites in this paper [11], (Fig. 1), the RMSE of 0 hh and 0 vv are 1.96 and 1.25 dB after filtering (Fig. 2(a)). After Freeman decomposition (Fig. 2(b)), the RMSE of 0 hh is reduced to 1.64 dB however the RMSE of 0 vv slightly increases to 1.35dB. On the other hand, the results of Cloude decomposition (fig. 2(c)) are significantly improved at both polarizations with the RMSE of 0 hh and 0 vv are 1.45 and 1.14 dB respectively. In summary both Freeman and Cloude decomposition methods are performed on AIRSAR L-band fully polarimetric data. The results of Freeman decomposition has an improved accuracy of 0 hh yet slightly degrades on 0 vv , but lowers overall error. The accuracy of backscattering coefficients is significantly improved by Cloude decomposition, which can link a target distributed pixel to its equivalent “pure target” that matches the single scattering of AIEM. Freeman decomposition assumes that cross-polarized returns * HV HV S S are contributed only by volume scattering V f , actually double-bounce D f and surface scatter S f components also have slight contribution. Furthermore, if * Re 0 HH VV S S , the surface scatter is considered as dominant and the parameter is fixed with 1 [5], which may lead to imprecise results in the quantitative analysis. Through comparison with AIEM predictions, Cloude decomposition demonstrates much more accuracy for quantitatively retrieving soil moisture.