Research On Pine Tree Defoliation Estimation By Using Single-scan Terrestrial Laser Scanning Data

Defoliation is an index that quantitatively reflects the severity of forest leaf damage.By calculating the proportion of leaf loss to total leaf volume,the evaluation of forest damage degree can be achieved,which can provide data support for the estimation of forest disaster degree.It is one of the important bases for formulating and implementing forest management measures.Limited to the resolution of remote sensing data,most of the current research on this topic were conducted on large scale defoliation classification,which means that there are more methods required in single-tree scale and deriving more specific defoliation estimation.Aiming at this gap,this research developed algorithms for defoliation estimation by using terrestrial laser scanning data and presented the validation and accuracy.The research area was in Jianping,Liaoning Province.48 sample plots,including 1098 individual trees,were set respectively from slightly,moderately and severely defoliated forest area,and the TLS data and field-measured defoliation were obtained.The whole research conducted with three main contents:(1)60 features were constructed and extracted from the terrestrial laser scanning point cloud data from three different aspects,i.e.the morphological properties of single trees,the distribution characteristics and the density characteristics of point cloud of the crowns.By using a Random Forest method,features were selected and defoliation were classified into different severity levels.(2)Accoding to the principle of point density calculation,an equiangular voxelizaiton method were presented to dividing the space into appropriate units.To show its advantages,the importance that the point density calculated by such units played in defoliation classification was compared with the tranditional cubic voxel units.(3)The parameters used in point density calculation were further optimized to summarize a simple linear relation between point density and defoliation.Based on such relation,an intergrated framworks to estimate defoliation using TLS data were concluded.The results showed that(1)Features such as point density,penetrate ratio and gap fraction obtained higher importance within the 60 features.When classify trees and plots into silght,moderate and severe defoliation classes,the overall accuracy at individual tree and plot scale were 80%and 94%,respectively.(2)The voxelization study verified the advantages of the innovative equiangular voxelization method in the aspect of the equilibrium point cloud density.It also showed the capacity of avoiding the influence of the distance from the scanner on the calculation of features due to the uneven density distribution.At this aspect,its performance were more robust than the traditional cubic voxelization method.(3)Through the further optimization of the point density features calculated by voxels from the new voxelization method,the point density features showed a simple linear relation with defoliation with 0.89 R2 on a single-tree scale and 0.83 R2 on a plot scale,respectively,and both derived 12%RMSE.The innovations of the whole research includ:(1)This is the first time that defoliation estimation is achieved by using TLS data,which has shown great application potential in forest health monitoring and management.(2)It has innovative construction of point cloud density,point cloud spatial distribution pattern,penetrability and other characteristic parameter database,and they were confirmed to have played an important role in defoliation estimation.(3)The innovative equiangular voxel structure provides a new idea for the data management and parameter calculation for single-scan TLS data.This whole research is a demonstration that the precise estimation of defoliation can be achieved by using the single-scan TLS technology,and the method proposed in this study has a high applicability,which is of great significance for pest monitoring and damage measurements of pinus tabulata.