Clustering of Point Patterns Derived from Lidar Canopy Height Data

High intensity canopy height LIDAR data affords model-based estimation of tree locations. The analysis of spatial point patterns is a natural extension of this modeling capability. Identification of within-stand clusters (features) of trees deviating significantly in height from those of surrounding trees (clutter) is important for inventory and forest management purposes. We demonstrate a nonparametric profile likelihood estimation of spatial clusters using Voronoï tesselation with and without prior smoothing via a morphological closure operation on the sets of Voronoï cells considered as solutions to the clustering problem. Smoothing yields not only a more regular outline of clusters but also appears to perform significantly better when there is more than one cluster in the point pattern. Two examples derived from LIDAR canopy data collected above Douglas-fir-dominated stands on Vancouver Island (British Columbia, Canada) illustrate practical applications. The morphological closure of the Voronoï cells prior to computing the likelihood provides more appealing results with potential for practical application in forestry.