Research On Extraction Of Forest Parameters Based On Handheld LiDAR

Sustainable forest planning and management require accurate information on forest resources.Light detection and ranging(LiDAR)technology has been applied in forest resource surveys over the past two decades.On a sample-level,ground-based LiDAR has been proven to be a method for obtaining high-quality point cloud data and extracting various tree parameters from it.However,it cannot balance cost and efficiency well,as obtaining better data means multiple instrument setups,which increases labor and time costs.Handheld LiDAR(HMLS)is a high-efficiency and convenient method for collecting forest survey data.It has started to be applied in survey practices.However,most pilot studies lack comparative studies of the applicability of handheld LiDAR in different levels of complexity environments.Therefore,this study selected three different types of sample sites with simple forest environment,low complexity environment,and medium complexity environment in Songxian County,Henan Province,and Shennongjia Forest District,Hubei Province.The handheld LiDAR data from each site was preprocessed and individual trees were segmented to verify the segmentation accuracy for different level of complexity sites.The diameter at breast height(DBH)and tree height were directly extracted from individual tree point clouds and indirect estimates of individual wood volume and stand volume were made.The statistical analysis was conducted to evaluate the accuracy of extraction of different forest parameter levels.The results are as follows:(1)In the simple forest environment and low complexity environment,the accuracy of single tree detection using handheld LiDAR can be accurately determined.The accuracy evaluation index for each type of single tree detection reaches more than 91%;the accuracy of single tree detection in moderately complex environments is slightly worse,with a comprehensive accuracy of 83%for broad-leaved mixed forests and coniferous mixed forests;Overall,a total of 1280 trees were measured in all the sample sites,of which 1058trees matched with the single trees extracted by handheld LiDAR data,achieving a comprehensive accuracy of 90.4%.The accuracy of single tree segmentation using handheld LiDAR data is relatively high,but as the complexity of forests increases,the number of misclassified and missed single trees also increases.In addition,the comprehensive accuracy of single tree segmentation for coniferous species is better than that for broad-leaved species.(2)In the simple forest environment and low complexity environment,the accuracy of DBH extraction is relatively high,with a R~2 value of 0.9 or higher in regression analysis;the accuracy slightly decreases in the moderately complex environment;the relative errors between the average DBH extracted by handheld LiDAR and the actual DBH measured in all sample sites range from 0.1%to 13.2%.In terms of tree height extraction,due to limitations in extracting crown information using handheld LiDAR,this study proposes using trunk point cloud combined with tree height curves based on Knobworth’s method to fit for tree height extraction,which can accurately extract tree heights.The relative errors between the average tree heights extracted by handheld LiDAR and the actual tree heights measured in all sample sites range from 1.2%to 11.7%,and the more accurate trunk point cloud quality,the higher the accuracy of tree height extraction.(3)The results of estimation of single wood volume and sample site volume show that the material volume can be well estimated in all types of sample sites,with a decision coefficient R~2 ranging from 0.871 to 0.956.The relative error between the average estimated material volume extracted by handheld LiDAR and the actual measured average material volume ranges from 1.3%to 8%,with the lowest being the highest.The estimation results of sample site volume also exhibit high correlation with the actual measurement,with a decision coefficient R~2 reaching 0.933.The relative error between the estimated average sample site volume and the actual measured average sample site volume for all types of sample sites ranges from 6.6%to 10.5%,and further improvement is needed in terms of accuracy.In conclusion,handheld LiDAR technology can effectively extract and estimate forest parameters at the sample site level,and can automatically process point cloud data with high efficiency.This study provides a reference for the practical application of handheld LiDAR in forestry.