A data-driven framework to identify and compare forest structure classes using LiDAR

As LiDAR datasets increase in availability and spatial extent, demand is growing for analytical frameworks that allow for robust comparison and interpretation among ecosystems. We utilize data-driven classification in a hierarchical design to estimate forest structure classes with parsimony, flexibility, and consistency as priorities. We use an a priori selection of six input features derived from small-footprint (32 cm), high density (17 returns/m²) airborne LiDAR: four L-moments to describe the vertical distribution of canopy structure, canopy density as a measure of vegetation coverage, and standard deviation of canopy density to characterize within-cell horizontal variability. We identify 14 statistically-separated meta-classes characterizing six ecoregions over 168,117 ha in Montana, USA. Meta-classes follow four general vertical shapes: tall and continuous, short-single strata, tall-single strata, and broken strata over short strata. Structure classes that dominate locally but are rare overall are also identified. The approach outlined here allows for intuitive comparison and assessment of forest structure from any number of landscapes and forest types without need for field training data.