Improving tree crown identification

By Stijn Wijdeven (the Netherlands)

Abstract
There is need for a standardized method for automated tree crown identification. In this study, two methods are analysed to identify their best practices and to develop a new, hybrid method. Method one is based on a Digital Surface Model (DSM), which determines the presence of trees based on their height compared to their neighbouring pixels. Method two is based on Object-Based Image Analysis (OBIA), which determines the presence of trees based on pixel characteristics of high resolution aerial images. Based on the best practices of both methods the hybrid method is created with the aim to improve tree crown identification accuracy. The results of the DSM-, OBIA- and hybrid methods are compared to a validation dataset by their correct number of crowns counted and crown area size. This is done for single solitary trees, for trees standing in rows and for trees as part of a group. For identifying individual tree crowns, the results show that single trees are more easily identified than trees in rows and groups. Furthermore, the performances of both the DSM-method and OBIA-method increase when using input data of a higher resolution. Based on the results it is not possible to conclude that the hybrid method is an overall improvement compared to the DSM- and OBIA-methods. Rather, it depends on what somebody wants to achieve with the crown identification method to determine what method is most suitable. The hybrid method has proven to be generally more useful for crown area estimation, but is relatively less accurate for crown count estimation.

Keywords: Tree crown identification; crown delineation; Digital Surface Models; DSM; Airborne Laser Scanning; LiDAR; OBIA; Object-Based Image Analysis; Remote Sensing; pointclouds; Boomregister; AHN.