Testing and evaluating different LiDAR-derived canopy height model generation methods for tree height estimation

Abstract The LiDAR-derived Canopy height model (CHM), due to its wide applications in forestry, is important for foresters. Before a CHM can be considered a valuable, objective source of information about a canopy surface, it must be properly interpolated and preprocessed, which may sometimes be challenging, especially in case of multilayer and multispecies forest. This study tested and evaluated the impact of CHM interpolation methods on the accuracy of estimating tree height, which is one of the most important trees and stands feature. Tree heights calculated from 5 CHMs (1. raw CHM; 2. pit-free CHM; 3. spike-free CHM; 4. Smoothed CHM (with a median filter and with a Gaussian filter) were compared to heights measurements in the field. It was also tested whether applying linear regression can improve the accuracy of tree height estimations based on LiDAR-derived CHMs. The obtained results indicate that the method of generating CHMs influences the accuracy of tree height estimations. The mean differences between the means of field heights and LiDAR-derived heights (for each CHM separately and the 99th percentile) were statistically significant. The most accurate results were obtained with the spike-free CHM (RMSE calculated for all trees was 1.42 m (5.80%)). The smallest errors were observed for conifers–the RMSEs obtained for the spike-free CHM were 1.07 m (3.75%) and 1.18 m (4.57%) for spruce and pine, respectively. The use of linear regression improved the accuracy of tree height estimations from LiDAR data (especially for the CHMs filtered with Gaussian and median filters).