Angle count sampling reliability as ground truth for area-based LiDAR applications in forest inventories

Abstract

LiDAR-based techniques to estimate forest variables at the stand level require accurate calibration through ground truth data. One purpose of this study was to verify whether angle count samples can be used as suitable ground truth to calibrate LiDAR-based models for timber volume estimation. Volume data were acquired on the ground for 79 plots in the Latemar forest (province of Bolzano, Italian Alps). A simple linear regression model, using the sum of all of the tree canopy heights in the plot as the explanatory variable, was adopted. As angle count samples have no fixed area, three different methods to approximate their size were compared. The angle count sample area can be properly approximated by visual assessment of the tree size classes and by callipering the largest tree in the plot. The results show that angle count sampling can be an efficient solution to calibrate LiDAR-based models: they produced fair estimates at the plot level (relative root mean square error (RMSE), 26.6%) that were better than fixed-radius plot estimates with full callipering (RMSE, 29.7%). Estimate uncertainty at increasingly large forest stand areas was also calculated by means of a simulation procedure. It showed that low uncertainty (standard error of estimate = approximately 2%) could be reached at a forest compartment level (19 ha on average).