Individual Tree Parameter Estimation From Airborne And Terrestrial Laser Scanning Point Clouds

Forest is the largest and most important terrestrial ecosystem,and it plays an important role in maintaining material and energy cycle and ecological balance in global ecosystem.Tree height and Diameter at Breast Height(DBH)are two basic parameters in forest inventory.They are important indicator for evaluating forest status and development trend.Above ground biomass(AGB)can reflect the carbon sequestration capacity and it is an important indicator for assessing forest ecological function.Traditional forest survey relies on the field measurements,which are laborintensive and time-consuming.Remote sensing technology provides an effective and promising way for forest inventory and management.The optical remote sensing image can provide spectral indices and textural information of the forest,which can be used for forest parameter inversion.However,the optical remote sensing image has a poor ability to describe the 3D structure of the forest.Light detection and ranging(Li DAR)can actively emitting laser pulses and directly obtain the 3D information of the forest structure,which offers unique advantages in measuring the vertical structure of the forest when compared with the optical remote sensing image.Li DAR provides a promising tool for forest inventory and forest parameter estimation.This study relies on different sources of Li DAR point clouds to carry out the individual tree detection research and forest parameter estimation.The abilities of airborne laser scanning(ALS),terrestrial laser scanning(TLS),and the newly developed multispectral airborne laser scanning to isolate individual trees and retrieve individual tree parameters(including tree height,DBH and AGB)are compared and analyzed.The study areas are located in a southern boreal forest of Finland and a mixed forest in Ontario,Canada.This study includes the following aspects:(1)The dissertation introduces the applications of optical remote sensing and Li DAR technology in forest inventory and ecological research,and the differences between the two observation means.Li DAR systems on different platforms for forestry applications are introduced.The advantages and limitations of Li DAR systems are summarized.The research contents and the organization structure of the dissertation are determined.(2)The dissertation introduces the researches of ALS and TLS in forestry applications.It reviews the related work of ALS and TLS in forest parameter estimations and the registration of point clouds from different sources in forest areas.The limitations of the existing methods and the development tendency are summarized.(3)Most researches have been working on stem recognition to detect individual trees from terrestrial laser scanning(TLS),the crown recognition researches are relatively fewer due to the variability of tree crown structures in the complex forest environments.This study proposes a minimum cut method to detect individual tree from the TLS point clouds.First,the stems are detected through cylinder fitting and then an objective function is designed using the similarity between the cloud points.The min cut algorithm is used to achieve the optimal solution to the objective function for crown segmentation by minimizing the similarity between objects.The proposed method obtained an average detection rate of 90.42%.Compared with the field reference,the determination coefficient R-squared values of tree height and DBH estimation were0.79 and 0.82,respectively.The root means squared error(RMSE)of tree height and DBH estimation were 1.25 m,and 2.89 cm,respectively.(4)The individual tree detection from ALS point clouds is currently based on the geometric and spatial information and the ALS are usually referred to the monochromatic wavelength laser scanner.Their performances are limited when dealing with trees with poor geometric characteristics.The newly developed multispectral airborne laser scanning makes it possible to obtain 3D information and more spectral information at the same time.This study proposes a new method to extract trees from the multispectral airborne point clouds by combining the spatial and spectral information.The results are compared with the results obtained by using spatial information alone.The experimental results demonstrated that the average detection rate by using spatial information was 82%.It increased to 88% when combined spatial and multispectral information was used.(5)The ALS and TLS point clouds are obtained from different viewpoints,thus there exists considerably different characteristics between the two point clouds.The tree crown structure is complicated with irregular natural surfaces and there are few conventional geometric features(point,line and plane)in the forest environment.These pose great challenges for forest ALS and TLS point cloud registration.This study proposes an automated method to register forest ALS and TLS point clouds through canopy density analysis,which avoids the geometric description of the point clouds in forest areas.The tree parameters(tree height,DBH,AGB)are estimated and compared before and after the registration.The abilities of ALS and TLS to detect individual trees and describe individual tree parameters are compared and analyzed.The proposed registration method obtained a good performance with a 3D distance residual of0.069 m.For the individual tree parameter estimation,TLS trends to underestimate tree height when compared with ALS.The best result for individual tree biomass estimation was obtained when ALS and TLS were combined.The RRMSE of individual tree biomass estimation based on TLS and ALS were 17.0% ～ 60.6% and 35.3% ～ 75.7%,respectively.The RRMSE reduced to 16.0% ～51.0% when registered ALS and TLS was used.