Biomass Related Parameter Retrieving From Quad-Pol Images Via Target Decomposition Method

It is an efficient and effective way to use microwave remote sensing methods for biomass estimation. Most used methods at present usually only make use of either phase information in establishing relationship between biomass related parameters and interfered phase values of image pairs, or to establish relation or amplitude information to reconstruct forest characters by taking forest as a whole uniformly distributed layer through which the microwave signals have traveled. They cannot make use of amplitude and phase information together to gain a specific description of interaction between trees and electromagnetic waves, thus when coming to the case of fractional areas of trees, which corresponding to mixing pixel in an image, these methods will not function well. For the methods employing amplitude information, although there are ways to describe the interaction progress in more detail, the models generated based on theories of electromagnetics, and are complicated with too many parameters to apply. For methods using phase information only, by introducing more interfermetric image pairs, parameter represent areas not fully covered be forest, but more images are required and accurate DEM is required for calculating tree heights. The method makes fully use of both amplitude and phase information while is simple enough for applications is target decomposition. This makes target decomposition can be used to solve the problem caused by incomplete description of mixing pixel, and suitable for biomass related parameter estimation.Target decomposition is the method to analyze return signals represented in canonical scattering mechanisms, or summary of components whose values are unique, to get characters of objects. This kind of methods is a result of good cooperation of radiative transfer method and wave approach in reconstructing object properties. But main applications employing target decomposition is limited in classifying and recognition problems, and is not yet contributing to quantitive remote sensing applications.Generally, there are specific target decomposition methods for different application background. The most suitable one for forest case is Freeman-Durden decomposition, which divide the interaction of forest areas with electromagnetic waves into three kinds of scattering. In the research of this PhD thesis, existed models gained by extracting scenes of forests are studied and main kinds of interactions, which occurs between electromagnetic waves and elements in those models involving fractional areas of forests, are selected and analyzed, and are matched to components of Freeman-Durden decomposition. Then by employing existed experience models and proper approximations forwarded by the author which relate results of those interactions and biomass related parameters, the three component are expressed in detail. Finally, equations with biomass related parameters and components of Freeman-Durden decomposition are established which form the model proposed in this thesis.To make values of Freeman-Durden decomposition component more accurately gained for quantitive applications, first errors caused during decomposition by employing limited looks to estimate covariance matrix are analyzed, then data gained by simulation based on of Monte Carlo theories is used to find correction methods. Experiment shows that corrected component values are in better accordance with ground scenes.The way in this thesis to retrieve model parameters is to employ simulated data generated by setting environments as similar as possible to application scene, including sensor parameters and type of forests. In this progress, under the condition of experiments in this thesis, the parameter representing change from other kind of scattering to volume scattering appears to be zero. Thus another stricter approach has been used to solve model parameters.Finally, the proposed model is verified by comparing results got by applying it to quad-pol radar images and ground truth data. The corresponding areas of ground surveyed zone on images are gained by finding the place on optical data by GPS pad measured data, then match the optical data to radar images. For the restriction by ground truth data, averaged results are used for comparison. Although for one set of parameter can not be verified for the ground truth was not able to get, a qualitive analysis has been done for it, and its corresponding simulated data are of the same order of magnitude, this parameter can be taken as well generated from the model. Even if it cannot be accurately generated, since relation between it and other parameters which has been verified, this will not affect the performance of the model proposed. Results finally show that the model proposed can act with high accuracy for pixels whose Freeman-Durden decomposition component values can be gained directly from image data, while for those pixels whose values have to be changed due to existed Freeman-Durden decomposition progress to meet the condition of this decomposition. Thus in all, this model can generate good results and is suitable for solving mixing pixel biomass related parameter retrieving problems.