Automated estimation of tree number in Pinus elliottii Engelm. and Eucalyptus sp. plantations using UAV imagery

Abstract

Forestry activities require accurate inventories to support sustainable forest management; however, conventional field-based inventories are costly and difficult to implement over large areas. This study aimed to define and evaluate a methodology to estimate the number of trees in commercial plantations of Pinus elliottii and Eucalyptus sp. using images acquired by an Unmanned Aerial Vehicle (UAV) in the central region of Rio Grande do Sul, Brazil. The study was conducted in three plantations (P1 – Pinus, 17 ha; P2 and P3 – Eucalyptus, totaling 33 ha). Circular sampling plots of 400 m² were established for the conventional forest inventory. Orthomosaics were generated from UAV flights using a DJI Mavic 3M at 120 m altitude, with 80% forward and 70% side overlap. Circular and rectangular sampling units with the same area were delineated on the images at the same sampling locations. Image processing was performed in Agisoft Metashape, and spatial analysis was conducted in QGIS. Tree counts obtained by the three methods (field inventory, circular plots on imagery, and rectangular plots on imagery) were compared using the Shapiro–Wilk normality test and, when appropriate, Student’s t-test or the Mann–Whitney U test. Statistically significant differences were observed between field-based counts and UAV-derived estimates in all study areas, whereas no significant differences were found between the two image-based methods. Discrepancies were smaller in Pinus plantations and more pronounced in Eucalyptus stands, indicating underestimation related to crown architecture, stand density, crown overlap, and image resolution and illumination constraints. Under the evaluated conditions, UAV-based approaches were not efficient for sample-based estimation of tree numbers, highlighting the need for species-specific calibration, lower-altitude flights, and improved image processing strategies.