Abstract
Airborne LiDAR-derived canopy height models (CHMs) have been widely applied in forestry inventory applications and have shown great advantages in obtaining forest-related parameters at different scales. Usually, first echoes are regarded as a representation of canopy surfaces during CHM generation, which may cause data pits and a lack of detail, thus resulting in negative effects for forestry applications. Therefore, we propose a canopy surface point filtering method called graph-based progressive morphological filtering (GPMF) for generating CHMs. The GPMF algorithm applies adaptive morphological filtering to exclude nonsurface points from all LiDAR points in a progressive process. The proposed method was tested in natural secondary forest stands. By comparing the performances of the GPMF method and four other CHM generation methods, namely, first-echo interpolation, Gaussian filtering, pit-filling and the highest points interpolation method, the results showed that the GPMF method produced few pits while retaining canopy details. The CHMs generated with the GPMF method were also obviously better than those generated with the other methods, as evidenced by the lowest average root mean square error (RMSE, 0.92 m) between reference points and the raster surfaces. The CHMs generated using GPMF also performed best overall in individual tree detection (average F-score = 80.57%) and tree height estimation (average RMSE = 0.21 m) among all the methods. Therefore, the proposed GPMF method can accurately represent the height of canopy surfaces and shows high potential in forestry applications.
NSTL
Elsevier