Assessment of tree detection and height estimation by AHN-2 point cloud data

By Liliana González Garcia (Mexico)

Abstract
This research presents an approach for detecting individual trees in forests and estimating their height by using the Airborne Laser Scanning (ALS) data, free of cost from the Dutch elevation dataset AHN (Actueel Hoogtebestand Nederland). The analysis was performed in a plot of coniferous forest in The Netherlands. The followed algorithm for treetop detection, use a canopy height model (CHM) as input. A common approach to create a CHM directly from the point cloud could give as a result a CHM with irregularities or the structure of the tree can be loose, which could affect the accuracy in the tree detection. To avoid this, the “pit-free CHM” was followed, having good visual results and also in terms of the treetop detection. In order to detect the treetops and their height, the variable window technique was applied to a pit-free Canopy Height Model (CHM). Afterward, the individual tree detections derived from the LiDAR point cloud was compared to a reference dataset. In order to know the best combination of parameters as input for the variable window technique a sensitivity analysis was performed, having as output is a list with combinations of parameters with their corresponding overall accuracy which aims to optimize the treetop detection algorithm. The best accuracy estimated in the study area was 80%. In terms of height accuracy, the maximum tree height was estimated with the largest relative percentage error (-10%), while the minimum tree height was overestimated with a relative percentage error of 2.8%. These values could be affected since the creation of the CHM, the approximation estimated for the growing seasonality or even the ability of the LiDAR scanner to capture the treetops. However, findings showed that the point cloud provided by the AHN-2 with the presented methodology gives reliable treetop detections with their corresponding heights.

Keywords: Actueel Hoogtebestand Nederland; AHN; LiDAR; forest; treetop heights; variable window size