Tree Height Estimation Using Field Measurement and Low-Cost Unmanned Aerial Vehicle (UAV) at Phnom Kulen National Park of Cambodia

Abstract

Tree height estimation is one of the important parameters needed for quantifying timber resources and it is also necessary for assessing the ecological and economic value of forest stand. It has been used to calculate the individual and number of stand volumes. Tree height also uses for forest inventory, which needs to update information in order to bring it to decision making and management. Most of tree estimation with Light Detection and Ranging (LiDAR) has been achieved successfully. The limitation of cost for LiDAR data acquisition, estimation of tree height derived Canopy height model (CHM) has been applied with low-cost UAV and the result indicated acceptable accuracy [1] –[4]. This research aims to estimate, and evaluate tree height from very high resolution images of low-cost UAVs. The influence of height estimation from different flight attributes, point cloud densities, extraction methods, photogrammetry product, and point cloud classification are discussed. The tree height estimation was obtained from two extraction methods, from photogrammetry product and semiautomatic point cloud classification. Tree height extractions were divided into two groups such CHM from the point could classifications and CHM from photogrammetry products. Tree locations from field measurements were used to extract tree height with buffering distance (Ocm, 50cm, 100cm, 150cm, and 200cm). The results of height extraction from UAV data were acceptable with semi-automatic point cloud classification, but the height estimation from photogrammetry products were under estimated. The results from 50m flight possible to derive the highest $\mathrm{R}^{2}$ was 0.60 and 200m flight can reach only $\mathrm{R}^{2}$ was 0.50 as the highest. In addition, sample paired t-test, tree height estimation and ground data from 50m flight were no statistically significant difference. In addition, this proposed method is possible for only open terrains which less than 12m, due to the limitation of the design pipe meters to measure height and cashew tree are complex leaf, identify the treetops were challenging.