Identifying leading species using tree crown metrics derived from very high spatial resolution imagery in a boreal forest environment

Abstract

Utilizing the spatial information inherent in panchromatic very high spatial resolution (VHSR) imagery, we explored the use of tree crown metrics for identifying leading species over four study sites in the Yukon Territory, Canada. Image segmentation was used to delineate homogeneous forest stands, followed by a tree crown delineation algorithm that identified individual tree crowns within each stand. Leading species in the study area included white spruce, black spruce, lodgepole pine, and trembling aspen. Nonparametric multivariate statistical tests indicated that some tree crown metrics generalized at the stand level have significant utility for discriminating leading species. Based on this result, a classification tree was generated using the crown metrics and independent calibration and validation datasets. The classification tree accurately identified leading species in 72.5% of the stands used for validation (n = 212), with the accuracy for individual species ranging from 43.9% to 100.0%. Most errors resulted from confusion between white spruce and the three other, less common, leading species. This study demonstrates the capacity of the spatial information content of panchromatic VHSR imagery to generate a series of crown metrics for discriminating among four common tree species of the Yukon Territory, a location with spatially and temporally limited forest monitoring practices.