Research On Multi-source Remote Sensing Data Inversion Of Forest Key Structure Parameters

Forest ecosystem plays an important role in maintaining the balance of carbon and oxygen,water conservation,soil conservation and maintenance of biological diversity,forest structure parameters(such as tree height,DBH,canopy density,leaf area index and biomass)can reflect the growth status of forest.The development of remote sensing technology overcomes the disadvantages of traditional forest structure parameters acquisition,and can quickly obtain forest structure parameters.The airborne lidar can penetrate the forest canopy to obtain the vertical structure of forest,and the spectral information of Landsat-8 OLI remote sensing image can reflect the vegetation growth in the horizontal structure.This study takes Hulun Buir,The Inner Mongolia Autonomous Region as the study area.Based on airborne LiDAR discrete point cloud data and Landsat-8 OLI data,estimated forest leaf area index,tree height,aboveground biomass and other key structural parameters.Aimed at achieving high-precision estimates of key forest structure parameters,the main contents and results are as follows:(1)Use the difference between the information contained in multiple echo types of airborne LiDAR discrete point cloud data,after the preprocess of airborne LiDAR data.Six laser penetration indexes(LPI)were extracted from LiDAR multi-echos types data.Then the LPIs were used to estimate the forest LAI by regression with field measured LAI.It was shown that the model based on LPI derived from first echo intensity(iLPIfirst)achieved the best result(R2=0.836,MAD=0.091)among all univariate estimation models.For the multivariate models,the model involving LPI derived from first echo intensity(iLPIfirst),LPI derived from canopy echo number(nLPIcan)and LPI derived from canopy echo intensity(iLPIcan)was the best(R2=0.883,MAD=0.076).By comparing the results,it was found that the R2 from multivariate model increased by 0.047 and MAD decreased by 0.015 than that from univariate model.It was concluded that the LPI derived from LiDAR different echo types intensity data could estimate forest LAI.And the accuracy of multivariate model is better than that from the univariate model.(2)Using airborne LIDAR point cloud data,for 1,0.5,0.25,0.1,0.05 times dilute to the point cloud data.25%,50%,75%,and 95%digit height are extracted according to the height of the point cloud,the linear regression model was used to estimate the average tree height.The results show that the point cloud height percentile can be a good estimate of the forest height,In the four different density point cloud,the 75%quantile has the highest accuracy,In all models,0.5 times the point cloud has the highest accuracy(Adj.R2=0.900,RMSE=0.594).There was little difference in the estimation accuracy of tree height with different density points,It can well satisfy the error requirements of Nature Reserve,Forest Park and key ecological forest in the forest resources survey.Airborne lidar can be used well in such investigations.(3)Using airborne LiDAR point cloud energy information,According to the height of the point cloud,the point cloud were sliced,and statisticed the total energy of each point cloud layer,According to point cloud energy distribution with height of point cloud,extracted the height of the maximum point cloud energy(HDmax)and the high of canopy energy(Hc),Combining with the point cloud height percentile parameters and the laser penetration index.Used RBF neural network to establish the model and to predict the forest aboveground biomass.Analysis found there was a good correlation between the extracted parameters and forest biomass,RBF neural network can well estimate forest aboveground biomass,Modeling accuracy R2=0.86,prediction accuracy R2=0.87.The results were higher than those of multiple regression models(R2=0.76,RMSE=12.75).The results showed that airborne LiDAR point cloud data can well predict the forest biomass.(4)According to the height information of airborne LiDAR point cloud data,extracted the 75%percentile height at 0.5 times point cloud density which has the best correlation with tree height,extracted the laser penetration index iLPIfirst,nLPIcan,iLPIcan,which has the best correlation with LAI,According to the point cloud filtering DEM,extracted slope and aspect factor.According to the Landsat-8 OLI images,22 linear and nonlinear combination parameters were extracted.The airborne LiDAR parameters and Landsat-8 OLI parameters are used as the input layer of the support vector regression machine,and the output layer is the Measured biomass.The results show that compared with the LiDAR point cloud data model accuracy(Prediction accuracy,R2=0.77,RMSE=12.75 t/hm2),the precision of SVM R2 is improved by 0.05,and the RMSE is reduced by 2.43 t/hm2.Compared with support vector machine model accuracy(Prediction accuracy,R2=0.92,RMSE=7.24 t/hm2)and multiple linear regression accuracy(Prediction accuracy,R2=0.82,RMSE=7.58 t/hm2),SVM results R2 improved by 0.1 and RMSE was reduced by 0.34 t/hm2,The results show that the combined with airborne LiDAR discrete point cloud data and Landsat-8 OLI image data can be well used to estimate the forest aboveground biomass,and make up for the shortage of single sensor.