Forest Parameter Extraction From Terrestrial Laser Scanning Data

The forests,as a principal part of the terrestrial ecosystem,play a very significant role in mitigating global climate change and maintaining regional ecology.Monitoring the condition of forests and quantifying their temporal and spatial variation are important for understanding their role in ecosystems and human activities.Timely accurate acquisition of forest parameters is important since these parameters can provide pivotal information on the current condition of forests.Traditional methods of forest parameters acquisition mainly rely on field surveys.However,these methods are characterized by a heavy workload,low efficiency,and relatively simple parameters that impose limitations on subsequent researches.Terrestrial laser scanning(TLS)technology can quickly acquire three-dimensional structural information of trees with millimeter-level precision and has great potential on forest parameters extraction.However,the obtained TLS data of forested areas have several characteristics such as huge data size and relatively complex scenes,which bring challenginges for data processing and forest parameters extraction.This paper focuses on some key issues of TLS data processing and forest parameters extraction methods.We aim to provide new ideas and technique support for further improvement of applying TLS data in forestry.The main research experiments and results of the study are as follows:(1)Based on the TLS data density and progressive strategy,a ground filtering algorithm for forested areas was developed.The surface coverage and scene structure of the forest are complex,which increase difficulty in obtaining the optimal threshold of the ground filtering algorithm and also inaccurate filtering of the trunk point near the ground and the ground point.For solving the above problems,the study realized the automatic acquisition of some complex and key thresholds based on TLS data density and improved the cloth simulation filtering algorithm with progressive strategy.TLS data of forest plots in different forest stands of the Saihanba ecological station were used in this study.We determined the grid resolution threshold and classification threshold based on the ground point density and then gradually reduced the above thresholds to progressively realize the ground filtering.The average total error was 0.38%,which decreased by 1.74% compared with the original cloth simulation filtering algorithm,and the average Kappa coefficient was 99.23%,which increased by 3.51% compared with the original cloth simulation filtering algorithm.The results showed that the algorithm proposed in this study could realize the ground filtering of the forest scene with high accuracy.The automatic acquisition of the grid resolution threshold and the classification threshold effectively improved the automatic capability.The progressive strategy effectively solved the problem that the trunk points near the ground and the ground points were difficult to be accurately distinguished.(2)Based on the local optimal strategy and machine learning algorithm,the method of extracting tree stem points based on the plot level was developed.When the tree stem points are extracted from the forest plot based on the features of the points,the number of scales is positively correlated with the extraction accuracy and negatively correlated with the efficiency of the algorithm.To solve this problem,this paper adopted the multioptimal scale method to extract the stem points based on the plot level.The method accounted for the accuracy of the results and the contribution of the scales.In this study,the TLS data of plots with different point densities and different dominant species were used.The information entropy theory was used to quantify the contribution of the predefined candidate scales and determine the multi-optimal scale set.Then the initial tree stem points were obtained by using the random forest classifier based on the twodimensional and three-dimensional geometric features of the multi-optimal scales calculated.At last,density-based spatial clustering of application with a noise algorithm was used to optimize the initial tree stem points.The results showed that compared with the traditional optimal scale method,this algorithm significantly improved the accuracy of the stem points extraction due to the increase of the number of local optimal scales.Compared with the traditional multi-random scale methods,this method took advantage of the local optimal strategy,which quantified and selected the local multi-optimal scale of points and guaranteed the scale contribution within a certain range.In this study,when the number of scales was 9,the accuracy and robustness of the algorithm were better than those of the traditional multi-random scale method.(3)Based on the TLS data of the individual tree,the main diameter at breast height(DBH)estimation algorithms(cylinder fitting algorithm,Hough transform algorithm and circle fitting algorithm)were compared in terms of the accuracy and robustness.Due to the difference in TLS data acquisition and forest structure,it is difficult to evaluate the accuracy and robustness of different DBH estimation algorithms based on existing studies.To solve this problem,the individual tree TLS data of different species and obtained by different scanning strategies were used to evaluate the above three algorithms in this study.The results showed that the cylindrical fitting algorithm had the ability to retain the geometric information in original data and obtain more comprehensive trunk feature information.Therefore,the cylindrical fitting algorithm had the highest accuracy and the best robustness.The accuracy of the results based on the Hough transform algorithm and the circle fitting algorithm was slightly different.For the tree with a regular trunk,the accuracy of the Hough transform algorithm was slightly better than that of the circle fitting algorithm.On the contrary,the result of the circle fitting algorithm was better.However,considering the results based on the Hough transform algorithm were sensitive to the threshold setting,the comprehensive performance of circle fitting algorithm was better than the Hough transform algorithm.(4)We constructed the method for estimating forest parameters based on the plot level with high accuracy and then explored the response of the growth of the larches to the stand age and nitrogen addition.The TLS data of the Saihanba larch forest plots were used in this study.Firstly,DBH of each tree in the plot was estimated by combining the density-based spatial clustering of applications with noise algorithm and weighted cylindrical fitting algorithm,and the height of each tree in the plot was calculated after clustering the points of each tree in the plot.Then based on the above results,the average DBH and tree height of each plot were calculated,and the Saihanba larch biomass model was used to estimate the biomass of each plot.Finally,the growth of DBH,tree height and biomass of each plot from 2016 to 2017 were calculated for evaluating the effects of forest age and nitrogen addition on the growth of larches.The results showed that the proposed method could estimate the forest parameters with high accuracy at the plot level.Forest structure and stand density were the main factors affecting the estimation accuracy of DBH and tree height.This work is of great significance for the application of TLS data in the forest such as forest parameters extraction and forest monitoring.