Estimation Of The Forest Stock Volume Based On Multi-source Remote Sensing Data

Forest is an important componentof the global ecological system,and plays an important role in the global carbon cycle.Estimating the Forest stock volume and biomass by using remote sensing technology plays an important role in the spatial distribution and dynamic change monitoring of forest carbon storage.In nowadays,optical remote sensing and microwave remote sensing have been widely applied in forest parameters inversion.Although the spectral bands of high resolution remote sensing images involve less spectral information, abundant spatialinformation of the image can reflect the characteristics of objects, which makes contribution to the analysis of forest structure parameters.The Synthetic Aperture Radar(SAR) can observe the earth all day under different weatherswithout the influence caused by external environment or other factors.Furthermore,the L band signal can penetrate the tree branches for better access to the information of trunks,which is of great significance to the inversion of forest stock volume.In this paper,national forest farm Jiangle County,Sanming City,Fujian Province were selected as the study area to explore a new method forthe stock volume inversion of coniferous forest and broad-leaved forestwith the data of SAR and high resolution remote sensing images.Firstly,the textural eigenvalues were extracted from the high resolution remote sensing images,and estimation models of forest stock volumeincluding coniferous forest and broad-leaved forestwere established based on the screening eigenvalues.Secondly.the inversion of coniferous and broad-leaved forest stock volume was completed based on L band backscatter signal of SAR.Last but not least,the linear and nonlinear inversion models of forest stock volume were established by using SAR combined with high resolution remote sensing images.The main research contents and results are as follows.1.Forest stock volume inversion by applying the textural eigenvalues of high resolution remote sensing images of ZY-3.The texture featureswhich have a significant correlation with the stock volume of coniferous forest and broad-leaved forest were extractedfrom high resolution imagesbased on the gray level co-occurrence matrixto establish the model of forest stock volume inversion,.The results showed as follows.Forthe model of coniferous forest,the correlation coefficient was 0.871,and the estimation accuracy was 78.94% with the root mean square error 27.50t/hm2.While the model of broad-leaved forest,the three values were 0.702,67.84%,36.81t/hm2,respectively.The inversion accuracyof coniferous forest is better than the broad-leaved forest.2.The inversion of forest stock volume by applying the backscatter parameters extracted from the polarimetric radar data of ALOS PALSAR.The backscatter coefficients of L band were extractedin different polarization modes,and the ratio of them was figured out.Then,the correlation of different stock volume and radar parameters were analyzedin order to establish the linear regression model and exponential model respectively.According to the comparison of those two models,results showed as follows.In the aspect of coniferous forest,linear regression model effected better,the correlation coefficient was 0.783,the estimation accuracy was 78.37% with the root mean square error 28.31t/hm2.While exponential model effected better in the aspect of broad-leaved forest,the three values were 0.734,73.49%,29.11t/hm2,respectively.3.The inversion of forest stock volume by applying the combination of high resolution remote sensing images and data of SAR.The texture features of ZY-3 images and the backscatter coefficients of SAR data were combined as the independent variable.to establish the inversion models of coniferous forest and broad-leaved forest respectively.In this part,multiple stepwise regression model,partial least square model and BP artificial neural network model were mainly applied.According to the comparison of those three models results showed as follows.In the aspect of coniferous forest,multiple stepwise regression model effected best,the correlation coefficient was 0.898, and the estimation accuracy was 83.07% with the root mean square error 20.45t/hm2.While BP artificial neural network model effected best in the aspect of broad-leaved forest,the three values were 80.14%,21.42t/hm2,respectively.