Research On Methods For Estimating Diameter At Breast Height From Terrestrial Laser Scanning

Diameter at breast height(DBH)is one of the most fundamental forest structural parameters.It provides basic data for the stem volume calculations and for the construction of growth models.Terrestrial laser scanning(TLS)can acquire millimeter-level of detail from the surrounding area.This allows rapid,automatic,and periodical estimates of DBH information.Over the last two decades,there has been a growing body of studies on the use of TLS to estimate DBH,and there were many DBH estimation algorithms used in these studies.With the multitude of DBH estimation approaches available,a systematic study is needed to compare different algorithms and evaluate the ideal situations to use a specific algorithm.However,few comprehensive studies and little guidance have been offered to help researchers choose the ideal DBH estimation algorithms and preprocessing conditions for different situations.Therefore,the large-scale application of TLS in the DBH estimation still lacks a certain theoretical basis.To solve these problems,this study evaluated three representative DBH estimation algorithms:Hough-transform,linear least square circle fitting,and nonlinear least square circle fitting.They were each evaluated on their performance using two forest types of TLS data and scan modes under numerous preprocessing conditions.The sensitivity analysis was also conducted for the preprocessing parameters of the algorithms.The two forest types were natural secondary forest and plantation.The scan modes were multi-scan and single-scan scanning methods.The influences of preprocessing conditions(different thickness of the selected point cloud and gray scale threshold)on the performance of the algorithms were also investigated.There were over 400 preprocessing conditions used in the study.Under each preprocessing condition,these three algorithms were used separately to extract the DBH of all trees in the plots.The main research contents and conclusions of this study are as follows:(1)In the natural secondary forest,the algorithms were evaluated on their accuracy performance and variation law using two scan modes under numerous preprocessing conditions.The results showed that the linear least square circle fitting algorithm was most robust among the tested algorithms.It is least influenced by the preprocessing conditions and the quality of point cloud.In both multi-scan and single-scan mode of the natural secondary forest with low quality point cloud,the overall accuracy performance of this algorithm is better than the other tested algorithms.Therefore,linear least square circle fitting algorithm was most appropriate for the natural secondary forest.(2)In the plantation,the algorithms were evaluated on their accuracy performance and variation law using single-scan mode under numerous preprocessing conditions.The results showed that the nonlinear least square circle fitting algorithm obtained the highest accuracy of the tested algorithms when the quality of stem projection images is high.The DBH estimation accuracy of this algorithm is extremely high in this condition(highest relative accuracy:93.37%).Therefore,nonlinear least square circle fitting algorithm was preferable for the plantation with high data quality.(3)The sensitivity analysis of preprocessing parameters was conducted based on the accuracy performance of the algorithms under different forest types,scan modes and preprocessing conditions.The most optimal preprocessing parameters of different conditions were obtained.The algorithms can obtain highest accuracy under these preprocessing conditions.The results showed that a moderate gray scale threshold of three had the best capability of filtering noise points.It was most appropriate for the natural secondary forest with massive noise points.A small gray scale threshold of one was most appropriate for the plantation with few noise points.In the natural secondary forest,the visibility of stems is poor due to the occlusion of understory vegetation.Therefore,a large height bin of 1.34m was most appropriate for the single-scan mode.While in the multi-scan mode,more stem circumference can be recorded,thus the visibility of stems is relatively higher.A moderately large height bin of 0.24m was most appropriate for this condition.The visibility of stems is high in the plantation.Even in the single-scan mode,a small height bin can also obtain sufficient stem circumference information.Therefore,a small height bin of 0.1m was most appropriate for the single-scan mode in the plantation.