Modelling tree size diversity from airborne laser scanning using canopy height models with image texture measures

Abstract

The aim of this study is to investigate the relationships between the plot-level tree size diversity and variables derived from airborne laser scanning (ALS) data, which is a type of LiDAR measurement. We conducted a study using forest stands with a range of managed and near-natural stands with a broad range of species. 33 Plots that represent the forest stand variety in the study area were sampled; within each plot four biophysical variables were measured by ground-based methods, these were height (TH), diameter at breast height (DBH), crown length (CL), and crown width (CW). The resultant tree size diversity was parameterised as Lmoments (t) statistics and compared with both point-based and grid-based laser scanning diversity variables. Point-based measures included the ratios of the Percentile means (P99/P25, P99/P50, P99/75, and P99/P90), Coefficient of variation, Skewness, Kurtosis, and Lmoments (t). The grid-based texture measures derived from the ALS Canopy Height Models (CHMs) included first-order texture, Standard Deviation of Grey Levels (SDGL), and three second-order texture measures, including Contrast, Entropy and Correlation. Furthermore, we tested the influence of scale by analysing the effect of grid cell sizes when generating CHMs from the raw point cloud ALS data. Using linear regression analysis, we show that the grid-based texture measures are superior predictors of tree height diversity than the point-based metrics. Sixty percent of the variance in the tree height diversity and 51% of the variance in the DBH Diversity were explained by the SDGL and Correlation texture measures, respectively (p<0.01). The associations between the texture features and the CL Diversity and CW Diversity were weaker compared to the TH Diversity and DBH Diversity (The highest R2 was 0.46 and 0.45, respectively, p<0.01). While the CHM calculated from a 3×3m grid cell had the strongest correlation with TH Diversity (0.60, p<0.01), the CHMs calculated from 1×1m and 2×2m cell size had the strongest association with DBH Diversity (0.51, p<0.01). Combining selected point- and grid-based variables accounted for up to 85% of the variance of tree height diversity, 68% of the variance of DBH Diversity and 52% of the variance of CL Diversity. Our study shows that the combination of laser-based height percentile ratios and texture measures derived from the ALS–CHM can be used to estimate tree size diversity across forest landscapes.