Inversion Study Of Forest Key Structural Parameters Based On Airborne LiDAR Point Cloud Data

The forest plays a key role in the researches on the global climate,atmospheric circulation and water conservation as an important part of the global ecosystem.The LiDAR(Light Detection and Ranging)point cloud data could have advantages which optical remote sensing does not has in the inversion of forest vertical and horizontal structural parameters.Taking the ShangKuLi farm of Inner Mongolia Eagan Agriculture and forestry interlaced area as the study area,the study combines field measurement data and LiDAR point cloud data as modeling data.In order to establish understory DEM and estimate forest parameters,a series of parameters are extracted with LiDAR point cloud data pre-processed.The main methods and results are as follows:(1)The understory DEM(Digital Elevation Model)was established with a spline finite element interpolation method.In this study,firstly,with TIN(Triangulated Irregular Network)model established to classify L iDAR point cloud data,the terrain point was got through multiple iterations;Secondly,the fitting point and check point were extracted from terrain point with density selection method;Finally,utilizing grid interconnection of terrain point,the regular grid DEM was established with a spline finite element interpolation method.In order to compare the effects of different resolution DEM,three defferent resolution DEM were established.By comparison,the higher the DEM resolution,the higher the accuracy,the accuracy of 1m resolution DEM was the highest which was 0.034m;the accuracy of 2m resolution DEM was the worst which was 0.096m.(2)Mean Canopy Height was estimated by the average of threshold value.Average of vegetation point cloud height was extracted from vegetation point cloud after elevation normalization.Afterward,a linear regression model of Mean Canopy Height was established.Meanwhile,the accuracy was evaluated.The results showed that the threshold mean of vegetation points height was able to estimate Mean Canopy Height with the higher reliability,with 99.81%of the highest accuracy,87.09%of the lowest accuracy and 94.56%of the average accuracy.(3)Leaf area index was estimated by SVR(Support Vector Regression)establishing model.With LiDAR point cloud data after elevation normalization,5 variables which reflect the vertical structure of forest were extracted by Pearson correlation coefficient method.Afterward,the inversion model of leaf area index of 1 to 5 variables was established by support vector machine.And a comparison of the above inversion models was carried out.The results showed that the accuracy of SVR-LPI regression model(R2=0.800,A(?)R2=0.797,RMSE=0.031)was the highest in the single variable model.The accuracy of SVR model with 5 variables(R2=0.868,A(?)R2=0.856,RMSE=0.024)was the highest in the multiple variable model.(4)Forest biomass was estimated by SVR establishing model.With a number of parameters extracted from point cloud data,6 SVR regression models were established to estimate forest biomass by s-SVR and v-5VR.The results showed that the accuracy of linear kernel function were the best in the s-SVR model,R2=0.8873,RMSE=32.9517/kg.The accuracy of linear kernel function were the best in the v-SVR model,R2=0.8840,RMSE=37.9684/kg.By constrasting the results of the s-SVR model and the v-SVR model,it’s concluded that the s-SVR model was more suitable to estimate the biomass of this study area.