Inversion Study Of Forest Structural Parameters Based On Small Footprint LiDAR Data

Forest plays a key role in the researches on the global climate, water and carbon cycle as an important part of terrestrial ecosystem. Because of the smaller diameter, the small footprint LiDAR could have advantages that other remote sensing technologies don’t have in the inversion of forest vertical and horizontal structural parameters. This paper takes Changchun Moon Lake National Forest Park as the study area and LiDAR data as the basic data. In order to get parameters, a series of operations are done including LiDAR data denoising, the energy correcting, random selecting, the point cloud data classifying, the layer slicing and the waveform fitting and then a series of parameters are extracted. The models of tree height, leaf area index, canopy density and biomass are established with the parameters by using a linear regression, multiple linear regression, partial least squares regression and support vector regression, and the accuracies are evaluated. The main results are as follows:(1) In terms of forest height, the models including different parameters could estimate the tree height better regardless of the level of point cloud density. The models with energy parameters are better than others in all densities levels and the model with the height of medium canopy energy is the best. The model with different parameters can get better results in different point cloud density.and most models can get better results in the lower point cloud density. The accuracy of larch forest stand is higher than that of pine forest stand.(2) In terms of forest leaf area index, the OGF’s inversion model is the best of all univariate models whose fitting correlation is:Adj R2=0.790and the accuracy is:P=0.978; multi-variable inversion models are better than univariate models, for example, the combination of OGF’ and OLGF’, whose fitting correlation is:Adj R2=0.812and the accuracy is:P=0.965.The point cloud densities have different affect on the models’results, the results of OGF model are positively correlated with the point cloud density; while the results of OGF and LPI multivariate model have no strict correlation with point cloud density, and the differences among the results of models in different densities are small. All the models could better estimate forest leaf area index under different point cloud densities and could totally meet the production needs.(3) In terms of forest crown closure, the I2’s inversion model is the best of all univariate models whose fitting correlation is:Adj R2=0.810, RMSE=0.016and the accuracy is:P0.978; the combination of LPI’and I3’is the best of the multivariate inversion models whose fitting correlation is:Adj R2=0.889, RMSE=0.012and the accuracy is:P=0.972. The result shows the number ratio variable model is better than other models, multi-variable inversion models are better than univariate models. (4) In terms of forest biomass, when the number of principal components is seven, the modeling accuracy and predicting accuracy of model with PLSR are good, and Pseudo-R2are0.922and0.852respectively. While the biomass is estimated with SVR, it could get better result when the kernel is linear with C=75.5, Nu=0.01, support vector is10, and the modeling accuracy and predicting accuracy are0.847and0.830. By contrasting the results of PLSR and SVR, it’s concluded that the PLSR model is more suitable to estimate the forest biomass of this study area.