Research On Forest Height Estimation From Polarimetric SAR Interferometry Images

Forests are the important factor providing human existence. They not only bear such ecological function as regulation of climate and water turn over in nature, they but also provide humanity with necessary natural products, such as wood, food products, fodder for domestic animal and medicinal plants. So, the monitoring and management of the forest cover on a large scale by remote sensing technology is a hot issue which is attracting a lot of research currently. Polarimetric SAR interferometry is an important branch in modern remote sensing technique. It combines two independent radar techniques, which are polarimetric SAR and interferometric SAR. Polarimetric interferometric SAR (PolInSAR) is not sensitive to shape and orientation of scatterers, but also to the spatial distribution and position of the scatterers. Based on the extraction of polarimetric information and polarimetric interferometric information in SAR images, in order to improve the accuracy of forest height estimation in PolInSAR images, forest parameter inversion models in PolInSAR, PolInSAR target decomposition (TD) and forest parameters extraction using PolInSAR are studied systematically and detailedly in this dissertation.Firstly, the scattering characteristics of target and canonical forest parameters inversion models are deeply investigated, including the random volume over ground (RVoG) and ESPRIT. Based on the theories and applications of existing forest height estimation methods, an accuracy improvement method for forest height estimation is proposed for PolInSAR images, which is combination two canonical approaches such as TLS-line fit and ESPRIT method. The proposed method is demonstrated with L-band full polarimetric interferometric data of ALOS/PALSAR of Malaysia and simulated data from PolSARProSim software. The results validate that the accuracy of the forest height estimation can be remarkable enhanced by the proposed method.Secondly, PolInSAR incoherence target decomposition based on scattering model is researched. The forest height estimation from PolInSAR image based on an adaptive model-based decomposition (AMBD) is presented in this dissertation, which described each interferometry cross correlation as a sum of contribution corresponding to single-bounce, double-bounce and volume scattering processes. The proposed method enables the retrieval not only of the forest parameters but also of the magnitude associated with each mechanism. Another advantage of the proposed method is that it makes use all of the information provided by the covariance matrix, which remains unachieved in the previous model-based decomposition methods. The validation experiment and performance validation are implemented using SIR-C/X-SAR PolInSAR images. Compared with the height forest estimation results using three-stage inversion method, it can be found that the proposed height estimation based on adaptive model-based decomposition can obtain a high accuracy.Subsequently, the forest height estimation based on the general three-layer scattering model (GTLSM) using PolInSAR images is proposed in this dissertation. In the GTLSM, the parameters related to the canopy layer can be estimated by using AMBD, while the parameters of tree trunk and ground are estimated based on solving a nonlinear optimization problem. This model provides the possibility to separate the forest model into three layers, ground layer, tree-trunk layer and canopy layer, based on polarimetric signatures and interferometric coherence diversity. The proposed GTLSM is also demonstrated with L-band SIR-C/X-SAR PolInSAR images. The results validate that the forest parameters can be retrieved directly and more accurately by the proposed GTLSM.Finally, two forest height estimation methods over sloping forest areas from PolInSAR image are proposed. The first method is based on the general model-based decomposition (GMBD). In this approach, we shall develop a general volume scattering model which can be characterized by two parameters: a degree of randomness and a mean orientation angle. The unknown parameters of forest can be obtained by using the nonlinear least square optimization method. This method enables the retrieval not only of the forest parameters but also of the magnitude associated with each mechanism. Besides, general single-bounce and double-bounce scattering models are developed to fit for the cross-polarization and off-diagonal terms by separating their independent orientation angle. The second method is based on the modified three-layer scattering model (MTLSM). The proposed model assumes that the vertical structure of forest placing on sloping terrain can be separated into three layers: canopy layer, tree trunk layer and ground layer that account for the simultaneous effect of three scattering mechanisms (volume scattering, surface scattering and double-bounce scattering component) in the sloping forest area. Furthermore, the proposed model also describes the dependence of the PolInSAR coherence on the forest height, average extinction and especially the local terrain slope. The proposed method not only enables the retrieval of the forest parameters in sloping forest areas, but also allows a more robust implementation and unambiguous estimation of the ground phase as well as canopy phase. The both methods are all applied to L-band ALOS/PALSAR of Indonesia and the results verify their effectiveness.