Study On Forest Biomass Inversion Method Based On Landsat-7ETM+ And PALSAR Data

Forest biomass is an important parameter in the forest ecosystem.Among forest inversion models based on remote sensing data,using spectrum and texture of optical data can accurately get forest canopy information.While different polarization modes of SAR data have obvious advantages in obtaining vertical structure of forest,and polarization decomposition can get more information of different aspects.Therefore,combining the advantages of optical images and SAR data,a joint inversion model can be established to obtain forest biomass more accurately.This study was conducted in Fengxian and Peixian area of Xuzhou with Landsat-7 ETM+ images and ALOS PALSAR dual polarized data,and a joint inversion model was established by using support vector machine regression(SVR)on the basis of obtaining factors that significantly relative to forest biomass.By combining different factors and increasing training group numbers,we studied the influence of different factors combinations and training group numbers on the accuracy of forest biomass inversion,and got the best factors combination.(1)By analyzing the correlation between the retrieved parameters of remote sensing data and the forest biomass,the results showed that the correlation between PALSAR texture characteristics and forest biomass was the best,followed by the parameters obtained by polarization decomposition,while the correlation between Landsat-7 ETM+ images parameters and forest biomass was relatively poor.(2)According to the ground survey data of study area and forest SAR scattering mechanism,the PolSARProSIM semi empirical model suitable for the study area was established,and the forest biomass of the study area was estimated.The results showed that the trunk biomass accounted for 68-70% of the total biomass which proportion was the highest,and the branch biomass accounted for 26%-28% of the total biomass,while the proportion of tree biomass was the lowest.(3)Establishment of joint inversion model for forest biomass based on support vector machine regression.The results showed that the inversion precision of forest biomass with different factors was significantly higher than that of single factor,the accuracy of the combination of Landsat-7 ETM+ sensitivity factors,PALSAR texture features and polarization decomposition factors was the highest,which the highest value reached 0.9382.In the combinations of 2 different factors,the combination of Landsat-7 ETM+ factors and PALSAR texture feature factors had the best inversionaccuracy,followed by the combination of polarization decomposition factors and PALSAR texture feature factors.The inversion accuracy of a single factor will affect the accuracy of the joint inversion.The accuracy of joint inversion was mainly affected by the factors that had high inversion accuracy,while the factors of similarity accuracy can increase the accuracy of joint inversion with the increase of training group numbers.