New Efficient and Automatic Approach to Extract Dendrometric Features from Terrestrial LiDAR Point Clouds in Forest Inventories

Abstract

This work aims to automatically determine tree position, diameter at breast height (DBH) and tree height (H) of forest stands from point clouds recorded by terrestrial laser scanning (TLS) or portable terrestrial LiDAR (PTL). It introduces UALtree, a Matlab® software tool based on a combination of Individual Tree Segmentation (ITS) and Individual Tree Detection (ITD) approaches. The performance of UALtree was compared with two state-of-the-art software such as AID-FOREST and 3D FOREST. These three methods were tested in nine forest plots. Six of them were located in Sierra de María-Los Vélez (Almería, Spain), containing reforested stands of Aleppo pine. They were 25 m side square plots scanned with a TLS Faro™ Focus3D X-330. The three remaining plots were rectangular shape plots (31 m average side) located in three different poplar plantations of the province of León (Spain). They were scanned with a PTL GeoSLAM™ ZEB Horizon scanner. UALtree and AID-FOREST achieved similar tree detection rates (average F1-score of 0.9378 and 0.9339, respectively), but UALtree was up to three times faster. 3D FOREST performed significantly worse (F1-score = 0.7949). Regarding tree height estimation, both UALtree and AID-FOREST provided similar figures, with an average relative MAD (median absolute deviation) less than 4%. Again, 3D FOREST performed worse (relative MAD = 6.36%). In the case of DBH estimation, AID-FOREST performed slightly better than UALtree, presenting a lower relative MAD, although the DBH values extracted from UALtree proved to have a better fit to the observed DBH values at tree level.