UAV-LiDAR and Terrestrial Laser Scanning for Automatic Extraction of Forest Inventory Parameters

The determination of the Dendrometric parameters of forest stands has a silvicultural and ecological interest for the forester, in particular for the evaluation of the dynamics of growth and productivity, and the evaluation of indicators of good ecological status. Currently, UAV-LiDAR (Unmanned Aerial Vehicle-Light Detection and Ranging) has become the new trend for measurement professionals, offering very high-resolution data collection at considerably lower survey costs. In addition, this technology has started to prove its utility in forest inventory applications namely to extract dendrometric parameters, where direct and conventional measurements are sometimes difficult. As for the TLS (Terrestrial Laser Scanning) technology, it has made it possible to obtain several abundant and refined structural information under the forest canopy. In the context of extraction of forest inventory parameters, the precision of extracting tree height for example using TLS alone, is insufficient. Hence the contribution of the combination of ALS (Aerial Laser Scanning) with TLS data to fill any information gaps that may exist. The main goal of this study is to present an approach to the automatic extraction of dendrometric parameters from UAV-LiDAR and TLS data. The proposed methodology is based on performing a TLS survey at a plot level and an ALS scan of the entire area. Our methodology is essentially made up of two steps: automatic crown delineation and automatic extraction of dendrometric parameters (position, Diameter at breast height, height, stem curve, concave and convex hull). For the first step, we compared the segmentation of the point cloud by the Watershed algorithm and by the SEGMA pipeline. Whereas the extraction of the dendrometric parameters was carried out using a set of algorithms namely RHT (Random Hough Transform) and LSR (Least Square Regression). The study focused on UAV-ALS and TLS datasets from different regions and with different densities (the Mediterranean, tropical, and coniferous forest). The validation was done using measurements carried out manually on the datasets. The results show that delineation by SEGMA gave a percentage of crown detection varying from 98 to 113% (over-segmentation) with diameters having a coefficient of determination varying from 56 to 90% depending on the area while the Watershed algorithm presented an over-segmentation of the actual crowns. Whereas the results for the DBH determination, the RHT and LSR algorithms both displayed almost 1-4 cm deviations from the reference while the height was extracted with 1-8 mm deviations.