Integration Of Lighter LiDAR And Multispectral Imagery For Estimation Of Tree Dieback Rate On Ground

Forest health is a vital issue that cannot be ignored during the whole process of forest management.Among the forest health stress factors,diseases and pests are the most important factors especially in recent years.The pine borer pests have caused large-scale disasters in Yunnan and other provinces,which result in a large number of deaths of Yunnan pine and heavy losses to forestry.In order to analyze the potential of diseases and pests’ outbreak in the future,continuous investigation and monitoring of forest diseases and insect pests is essential.The pine shoot dieback is one of the most critical indicators of the occurrence and development of most forest diseases and pests.Therefore,in this study we had proposed a method for the pine shoots dieback investigating and monitoring based on the integration of a lighter LiDAR(VLP-16)and a multispectral camera(ADC).Tree parameters and index were extracted based on integrated device combined light lidar with multispectral camera in field plots and then used for estimating the dieback rate(DBR)on both shoot and single-tree scales.The main research content and results include the following parts:(1)In order to rapidly acquire the point cloud and multispectral images in field measurement,a simple and portable integrated device that can be quickly loaded and unloaded together with a method for VLP-16 point cloud and multispectral images acquisition had been proposed in this thesis.This device and method played a large role in the outside forest survey.(2)Based on the field LiDAR scanning,point cloud noise removing and tree skeleton extraction were studied.And then a method for single tree skeleton extraction and 3D reconstruction based on point cloud was proposed for supporting the 3D modeling of a single tree and provides key information for the setting of single tree modeling rules and structural parameters.(3)Based on the structural parameters of the field investigation and the information extracted from the tree skeleton,three-dimensional models on the shoot and single tree scale were reconstructed.Shoot and single tree simulation and DBR sensitivity analysis were then performed using the improved RAPID model LiDAR point cloud simulation module respectively.Based on the analysis result,parameters most related to the DBR were proposed,including the number and intensity of the point clouds.They are used to predict the single tree DBR.Moreover,the relationship between shoot shedding and single tree structure and the DBR were analyzed and the result proved that the correlation were very close.(4)Multispectral images were classified based on support vector machines(SVM).The single tree DBR in four directions was calculated using the classification results.After verification with the field measured DBR,the result shows that the estimated error of DBR is between-0.13 and 0.29,and the root mean square error(RMSE)is 0.09.Among them,the absolute error of the estimation DBR of most single trees is between 0 and 0.15.(5)The concept of potential shoot number for single-tree scale was proposed,and a method for simulating and predicting potential shoots was also established by constructing a shoot or needle shedding parameter foliage transparency(FT).The FT combined with other point cloud parameters was used for modeling and analyzing the DBR of single trees.The result shows that the correlation between the foliage transparency and the DBR is the best among all the parameters in the single-tree scale.And After adding the foliage transparency,the random forest regression analysis model for 30 trees shows RMSE and R~2 are 0.1130 and 0.5524,respectively.(6)Integrated VLP-16 matrix,FT and multispectral image vegetation index to PCI analysis,the first seven parameters were selected,they are represent FT,dieback ratio of multispectral image classification result,corrected number of point cloud,standard of NDVI,average of point cloud intensity,average of RVI and average of EVI respectively.And then these seven parameters were used for DBR regression based on the random forest model.The result based on ten-fold cross-validation shows that the RMSE is 0.0953 and the R~2 is 0.8289.This result is better than the above two methods.In this thesis,a complete technical system for estimating the single tree DBR are proposed,including a portable integrated device based on a lighter LiDAR VLP-16 and a multispectral camera(ADC),together with a ground scanning and DBR information estimation algorithm.The method for main factors and parameters related to the DBR extraction and analysis could be used for monitoring and evaluating the health and other aspects of forest.This study provided new theories,ideas,and methods for remote sensing monitoring of forest health.