Agent-based region growing for individual tree crown delineation from airborne laser scanning (ALS) data

Abstract

Individual tree crown delineation (ITCD) is a fundamental and vital component of individual tree-based forest inventory. Region-growing algorithms have been widely developed and applied in ITCD studies. Although individual treetops are typically designated as initial seeds, most region-growing algorithms do not consider tree competition when tree crowns touch each other and have some overlapping area. An agent-based region-growing (ABRG) algorithm was constructed for ITCD using airborne laser scanning (ALS) data considering both growth and competition mechanisms. Three region-growing algorithms were compared for both coniferous and deciduous trees. The algorithms were: (1) marker-controlled region growing with simultaneous growth (MCRG); (2) agent-based region growing with one-way competition (ABRG1W) in which taller trees limit the growth of shorter trees, but shorter trees have no impact on taller trees; and (3) agent-based region growing with two-way competition (ABRG2W) in which taller and shorter trees influence each other. Incorporating competition into the ABRG algorithm resulted in a statistically significant improvement of the accuracy of individual tree delineations for conifers and a statistically significant improvement in the accuracy of plot-level crown area for deciduous trees. For conifers, two-way competition provided more accurate results because of increased competition in coniferous plots. For deciduous plots, the accuracy of one-way competition had a strong correlation (0.92) with relative spacing (an index of competition level). This result is consistent with the one-way competition mechanism because the taller (larger) deciduous trees would limit the growth of small trees as the competition level declines. The degree of improvement in tree crown delineation accuracy by ABRG was related to the characteristics of tree height and density (i.e. distance between trees) in the plots.

Additional information

Funding

This work was supported by the Fundamental Research Funds for the Central Universities [Project #DL13BAX06].