Estimation Of Forest Canopy Biomass Based On GF-3 Fully Polarized SAR Data

Forest canopy which is defined as the aggregate of all crowns in a forest stand plays a vital role in maintaining biodiversity,maintaining regional and global ecosystem balance.Accurate estimation of forest canopy biomass is of great significance for research on forest ecosystem productivity,monitoring forest environment changes and guiding forest management.However,traditional canopy biomass estimation methods have problems such as difficulty in measurement,time-consuming and labor-intensive.Synthetic Aperture Radar(SAR)has great potential and advantages in forest canopy biomass estimation.Based on GF-3 full polarization SAR data,this study taked the Wangyedian forest farm in Chifeng as the experimental area,and 25 sample plots were measured.Freeman three-component decomposition,Freeman two-component decomposition and Yamaguchi three-component decomposition method are used to obtain the polarization decomposition components,and compare and analyze the decomposition results of different decomposition methods on the forest region,and construct the volume-to-ground scattering ratio parameters under the polarization decomposition methods respectively.Then the correlation between 15 variables of SAR backscattering coefficient,polarization decomposition and volume-to-ground scattering ration parameter and canopy biomass are explored by the correlation analysis method.The stepwise regression method is used to screen the variables,and the forest canopy biomass regression model is established.The accuracy of the model is evaluated by using the Leave-One-Out Cross-Validation(LOOCV)method.This study draws the following conclusions.(1)Different polarization decomposition methods have different decomposition effects on forest regions.The results of Yamaguchi three-component decomposition are the best,Freeman two-component decomposition takes the second place,Freeman three-component decomposition has the worst results,and the double-bounce scattering component and surface scattering component result have some negative values.To a certain extent,adding adaptive parameters to the volume scattering model can improve the polarization decomposition effect,limit the range of adaptive parameters in the volume scattering model,and perform orthogonal transformation and unitary transformation on the coherent matrix can solve the problem that the double-bounce scattering component and the surface scattering component have negative values,and good decomposition results are obtained.(2)The backscattering coefficient and polarization decomposition components of SAR data have a significant correlation with canopy biomass,which can reflect the forest canopy structure parameter information.The correlation of polarization decomposition components and canopy biomass is higher,which shows that the decomposition components have greater potential in canopy biomass estimation.(3)The optimal model selection parameters are the volume-to-ground scattering ratio parameters constructed by Freeman two-component decomposition and Yamaguchi three-component decomposition.The R2 of the model is 0.721,the RMSE is 5.496 t/hm2,and R2 of the LOOCV result is 0.70,the fitted RMSE is 6.313t/hm2.The test results show that the prediction error of the model was relatively low(ME=-0.336 t/hm2,MAE=5.168t/hm2,M%E=3.662%,MA%E=22.475%).The relationship between the true value and the predicted value is close to the 1:1 line,and there is no obvious saturation point.The polarization decomposition components can accurately estimate the forest canopy biomass.