Towards a universal lidar canopy height indicator

A light detection and ranging (lidar) canopy height study was conducted with 13 datasets collected using four different models of airborne laser terrain mapper (ALTM) sensors over 13 widely variable vegetation types ranging in average height from <1 m to 24 m at five sites across Canada between 2000 and 2005. The study demonstrates that the vertical standard deviation of all topographically detrended first and last laser pulse returns (LSD) is a robust estimator of canopy height (Ht) for a wide variety of vegetation types and heights and lidar survey configurations. After regressing Ht against LSD for 77 plots and transects, it was found that Ht could be predicted as a simple multiplication (M) of LSD (M = 2.5, coefficient of determination (r2) = 0.95, root mean square error (RMSE) = 1.8 m, tail probability (p) < 0.01). For forest plots only, LSD was found to better predict average tree height (r2 = 0.80, RMSE = 2.1 m, p < 0.01) than Lorey's height (r2 = 0.59, RMSE = 3.0 m, p < 0.01). A test of the LSD canopy height model was performed using stand heights (HtFRI) from an independent forest resource inventory (FRI) for four vegetation classes. Results from the raw FRI and modelled stand height comparison displayed close to a 1:1 relationship (HtFRI = 0.97HtLSD, r2 = 0.73, RMSE = 4.7 m, p < 0.01, n = 38). All plot and transect canopy heights were also compared with the localized maxima of laser pulse returns (Lmax). For individual surveys over homogeneous vegetation types, Lmax generally provides a better canopy height indicator. Across all surveys and site types, however, LSD was almost always shown to have a more consistent relationship with actual canopy height. The only observed exception was in the case of forest plot level Lorey's mean tree height. The advantages of using a multiplier of LSD to estimate canopy height are its apparent insensitivity to survey configuration and its demonstrated applicability to a range of vegetation types and height classes.