| The stand structure of forest determines forest’s functions,and the distribution of trees within forest,to a certain extent,determines trees’ future growing.Analyzing forest stand structure scientifically is important for managing forestry,which is benefit to the healthy management and sustainable development of forests.Combined with modern remote sensing measurement technology,LiDAR(Light Detection and Ranging)can provide fine three-dimensional structure data of the forest.Conducting quantitative analysis of trees structure and displaying the standing trees horizontally and vertically.This study was taken in the Taizishan forest farm,a state-owned forest land,in Hubei Province,using Terrestrial Laser Scanning(TLS)scanned 9 plots,including coniferous forests,broad-leaved forests and coniferous broad-leaved mixed forests(referred to as mixed forests).Based on cloud registration,extraction terrain points and normalized vegetation height,we detected the locations of trees in the sites,fitted the DBH of the trunk,obtained the vertical structure of those sample stands.The main conclusions were as follows:1.In this study,setting 5 sites in each plot,using paired target center points to co-registration clouds points.The results of plot registration error analysis show that the maximum root-mean-square error was 3.03 cm and the minimum root-mean-square error was 0.48 cm.2.Merging the registrated points,we use Octree method to divide the whole plot points iteratively,and the merged points data were classified into ground points and non-ground points.Combined with the plane model,the RANdom Sample Consensus(RANSAC)algorithm was adopted to remove the noise points that far away from the fitting ground plane,then saving the selected ground points and using inverse distance interpolation to generate terrain elevation plot points.By comparing the real ground points with the generated terrain points,results show that the mean z-value difference between them were less than 5cm within plane plot areas,it shows that in smooth terrain area using this method can generate reliable terrain.However,for the region with deep concave,the error between the generated point cloud and the real ground points is large.3.In this study,the tree trunk recognition algorithm is constructed by clustering points in vertical layers and extracting clustering features.Using this algorithm,three modes of trunk detection were applied,including mode 1 that using the single scan data of the central sites to detect trunks,mode 2 that using the merged five-site data to detect trunks and mode 3 that detecting trunks of five scan data separately and then merged the detected trunk points.Results shows: 1)for mode1,in coniferous forests,the mean recall rate is 94.2%(the maximum recall rate is 100%,the minimum recall rate is 87.5%);the mean precision rate is 82.9%(the maximum precision rate is 93.8%,the minimum precision rate is 67.5%).In broad-leaved forests,the mean recall rate is 82.2%(the maximum recall rate is 95.2%,the minimum recall rate is 70.6%);the mean precision rate is 28.7%(the maximum precision rate is 35.0%,the minimum precision rate is 24.4%).In mixed forests,the mean recall rate is 71.1%(the maximum recall rate is 95.7%,the minimum recall rate is 44.4%);the mean precision rate is 30.9%(the maximum precision rate is 39.6%,the minimum precision rate is 20.8%).2)For mode 2,in coniferous forests,the mean recall rate is92.8%(the maximum recall rate is 100%,the minimum recall rate is 86.5%);the mean precision rate is 92.8%(the maximum precision rate is 100.0%,the minimum precision rate is 91.6%).In broad-leaved forests,the mean recall rate is 86.1%(the maximum recall rate is 95.6%,the minimum recall rate is 75.5%);the mean precision rate is 54.1%(the maximum precision rate is 68.3%,the minimum precision rate is 45.1%).In mixed forests,the mean recall rate is 59.6%(the maximum recall rate is 92.9%,the minimum recall rate is 50.0%);the mean precision rate is 47.6%(the maximum precision rate is 65.4%,the minimum precision rate is 37.5%).3)For mode 3,in coniferous forests,the mean recall rate is 96.3%(the maximum recall rate is 100%,the minimum recall rate is 93.8%);the mean precision rate is 96.7%(the maximum precision rate is 100.0%,the minimum precision rate is 93.8%).In broad-leaved forests,the mean recall rate is 84.1%(the maximum recall rate is 84.4%,the minimum recall rate is 83.8%);the mean precision rate is 68.3%(the maximum precision rate is 70.0%,the minimum precision rate is 65.9%).In mixed forests,the mean recall rate is 77.3%(the maximum recall rate is 84.8%,the minimum recall rate is 65.7%);the mean precision rate is 85.8%(the maximum precision rate is 95.8%,the minimum precision rate is 79.1%).4.Comparing the results of the three modes,it shows that the overall rate of mode 2(64.8%)is higher than that of mode 1(47.2%),and the highest recall rate of mode 3 is 83.6%.The precision rate of mode 1 is 82.5%,mode 2 is 82.8% and mode 3 is 85.9%.The recall rate and precision rate of mode 1,mode 2,mode 3 gradually increase successively.For the three types of stands,coniferous forest had the highest precision rate,followed by broad-leaved forest and followed by mixed coniferous.Coniferous forest had the highest recall rate,followed by mixed forest,and the broadleaf-forest had the smallest recall rate.5.In this study,we adopted cylindrical fitting to acquire diameter at breast height.According to the trunks coordinate-values,we selected 1.2 m-1.4 m points of stem.Using cylinder fitting to extract radius based on selected trunk points and the diameter can be calculated.The DBH of the 9 sample plots was calculated by using this method.In coniferous forests,the mean root-mean-square error was 2.0 cm,the maximum root-mean-square error was 2.9 cm,and the minimum root-mean-square error was 0.7cm.In broad-leaved forests,the mean root-mean-square error was 2.0 cm,the maximum root-mean-square error was 2.7 cm and the minimum root-mean-square error was 0.8 cm.In mixed forests,the root-mean-square error was 3.1cm,and the minimum root-mean-square was 2.2cm.6.Voxel was adopted to analysis the vertical profile coverage in our study,and it shows it has the power to display vertical structure of different plot types.For mono-canopy stands,the vertical coverage rate has an obvious single-peak characteristics,while for multi-layer stands,the vertical coverage has obvious multi-peak characteristics,which is close to the actual vertical distribution of stands and show the vertical crowns distribution.By comparing the three different stand types,in coniferous forests,the young and middle age stands of Pinus massoniana mainly contain an obvious single peak,while the mature multi-layer stands of Pinus massoniana show multi-canopy distribution characteristics.In broad-leaved pure forest,young Quercus acutissima shows an obvious single peak,and mature Quercus acutissima shows a double peak structure under the influence of vegetation regeneration.In coniferous and broad-leaved mixed forest,Pinus massoniana and Ligustrum lucidum young mixed forest presents single peak distribution,Pinus massoniana and Camellia oleifera nearly matured mixed forest shows double peak,while Pinus massoniana and Quercus acutissima the mixed forest influenced by shrub regeneration present multi-peak distribution. |
PDF Full Text Request