Inversion Of Surface Parameters And Tree Heights Using SAR Data

Tree height in forested areas is one of most important parameters in the quantification of forest carbon sinks.In order to improve the accuracy level in the height estimation in a large area,the use of synthetic aperture radar(SAR)data is explored.According to the characteristics of SAR data,interferometry,and polarimetric interferometry,numerous algorithms to estimate tree height using single-polarization,dual-polarization,and full-polarimetric SAR data are published.After studying the existing algorithms,a set of algorithms is revised.The revised algorithms one-by-one are assessed using ALOS-PALSAR and simulated SAR data.The first algorithm in the set used the single cross-polarized data because the HV(or VH)backscatter coefficient is highly related to the forest structural parameters.The empirical algorithm between the L-HV backscattering coefficient of ALOS-PALSAR and in situ height/DBH of the virgin forest stands in Zoige,China was established.The relationship was a negative logarithmic one and had low correlation coefficient,which implies the regression model using HV backscatter as independent variable was unsuitable to extract tree height in the complex alpine environments.Different polarized backscattering coefficients have equivalently different scattering centers located between the ground surface and tree canopies.Difference of the locations of the scattering centers should provide information to estimate tree height.The difference of the interferometric phase between the HH and the HV polarization channel was extracted at the same forest stands as above.The inverted tree heights were much smaller than the measured values.The sinc function based on the coherence coefficient was introduced to compensate for the underestimation.However,the improvement was limited,the dual effect of compensation coefficient and the coherence coefficient might lead to high estimates.Then,a three-step inversion algorithm using the coherent amplitude and phase optimization was proposed.The theoretic basis was the widely-used full-polarization random scattering model(RVoG)inversion algorithm.The RVoG algorithm improved the estimation accuracy of the complex scattering coefficient of volume scattering.In results,the accuracy level of tree height estimation was elevated,and the error was less than 10%.As the final component in the tree height retrieval algorithms,the estimation of signal parameters via rotational invariance technique(ESPRIT)algorithm that is based on the principle of the signal spectrum estimation was studied.Although the fullpolarization data were anticipated in the algorithm,only dual-polarization data were input into the algorithm.The averaged tree height obtained through the simulation was about 9.5% shorter than that using the fully polarized SAR data.The significance was the ESPRIT algorithm offered new potential to retrieve tree height using the dual-polarization data,and the algorithm should have much more applicable than the algorithm that requires the fully polarized data.Furthermore,to minimize the effect of terrain slope on the accuracy level in the tree height inversion,a gradient correction algorithm based on random scattering model was proposed.The simulated results showed that the algorithm successfully minimized the errors.With the available dual-polarization SAR data and ground measurements,the algorithm was applied to invert the tree height in forest stands of Zoige,and the results were satisfactory.Thus,the algorithm was validated.