A multiscale morphological algorithm for improvements to canopy height models

Pixels with distinctively lower elevation values than the surrounding pixels in a canopy height model (CHM) e.g. pixels representing a pit, often lead to the underestimation of tree heights. To rectify the underestimation, this paper presents a novel multiscale CHM improvement algorithm. A multiscale Laplacian operator, a multiscale-based morphological closing operator and a multiscale median filtering operator were applied to a 1-m resolution CHM to detect and replace pit pixels. The root-mean-squared error (RMSE) and the mean absolute error (MAE) before and after the improvement were computed by comparing the CHMs with field measurements. The improvement is evident as the RMSE decreased from 0.699 m to 0.390 m and the MAE decreased from 0.364 m to 0.243 m. Furthermore, individual-tree-extraction algorithms, namely the variable-area-local maxima algorithm and the individual-tree-crown-delineation algorithm, demonstrated that the proposed algorithm increases the accuracy of the estimation of tree heights.