Tree attribute assessment in urban greenwood using ground-based LiDAR and multiseasonal aerial photography data

Abstract

Advances in LiDAR and unmanned aerial vehicle technology have made high-resolution data available, which can be used for individual tree detection and assessing tree attributes. The accuracy of these assessments is still not clear for stands with high tree species diversity as well as leaf-off and leaf-on conditions. The aim of this study was to assess the quality of tree top detection and individual tree heights extracted from photogrammetric point clouds and canopy height models as well as ground-based LiDAR clouds in mixed and coniferous forest stands depending on the phenological stage. The study has been carried out in the Botanical Garden of the Petrozavodsk State University (Republic of Karelia, Russia). Four flight missions (in 2019–2021) using Phantom 4 Pro quadcopter were conducted in the arboretum (> 200 tree species) during periods of leafless, leaf biomass growth, full foliage and autumn leaf colouration. A single ground-based laser scanning was performed using a Leica BLK 360. Multiseasonal ultra-high resolution orthophoto mosaics (1.1–2.8 cm/pixel), photogrammetric point clouds (average density is 4200 points/m2), as well as LiDAR clouds (11 600 points/m2) were obtained. Further analysis was performed on three sites differing in tree species composition, tree density and site area. Tree tops were automatically detected from photogrammetric point clouds and their heights were estimated using R environment software. We found that most of the trees (78.9%) were correctly detected by algorithms based on photogrammetric data collected in periods of full foliage and autumn colouration. We also found that the number of false positive (FP) and false negative (FN) cases increased with decreasing in green biomass on deciduous trees. Compared with an average value, tree detection quality increased by 9.4% for coniferous trees with cone-shaped crowns (Abies sibirica, A. balsamea, A. fraseri, Picea abies, P. pungens, P. omorika, Pseudotsuga menziesii, Larix sibirica) regardless of the tree density, and tree detection quality decreased by 10% for coniferous trees with an ellipsoidal-shaped crowns (e.g. Thuja occidentalis, genus Pinus) or in cases for broad-leaved trees with high tree density. The lowest value of tree detection quality (F = 0.49) was found for the leafless period. High values (F = 0.84) obtained for periods of full foliage and autumn colouration indicates that tree detection quality was well in general. For the biomass growth period, this value (F = 0.69) also indicates a high quality of tree detection results. We also found that tree heights estimated using photogrammetric data well matched with tree heights measured on LiDAR clouds (R2 = 0.99). The highest accuracy was obtained for coniferous trees with cone-shaped crowns. We also estimated the height increments of different tree species between 2019 and 2021 based on photogrammetric point clouds. The highest annual height increment was obtained for Pinus sibirica (52 cm), and the lowest for Pseudotsuga menziesii (32 cm). Overall, our results have shown the potential to use photogrammetric and LiDAR data for tree mapping and estimating tree attributes in multi-species forest stands of arboretums or urban parks, as well as in natural forests.