Using Drone Imagery of Leaf Phenology to Identify Tree Species

Keywords: Drone Imagery, Tree Species Identification, Leaf Phenology, Forrest Management
Session Type: Virtual Poster Abstract
Day: Friday
Session Start / End Time: 2/25/2022 05:20 PM (Eastern Time (US & Canada)) - 2/25/2022 06:40 PM (Eastern Time (US & Canada))
Room: Virtual 40

Abstract

Important forest management decisions often rely on knowing what tree species are present and their relative abundance. However, manually surveying plots of forests can be time-consuming and labor-intensive. Current aerial drone technologies allow for the capture of high-resolution imagery in a fast and cost-effective manner, making it a potential alternative to manually surveying forests. In this project, we explored the effectiveness of using aerial drone imagery of autumn leaf phenology to identify 3 tree species, Green Ash, Eastern Cottonwood, and Siberian Elm. Over the course of three months in autumn, three high-resolution drone images were captured at different stages of leaf color change. From each image, the RGB bands along with DSM elevation bands were compiled into a single image so that different temporal patterns in leaf changes among tree species were reflected in different band signatures. This complied image was classified with ENVI supervised classification, with the generated classes being used to identify tree species. To assess the accuracy of these classes, random points were generated and then ground-truthed to perform a kappa error analysis. A kappa score of 67.35% was observed. These findings indicate that it is possible to identify tree species through this method, with room for improvement.