Sensitivity of LAIe and CHP retrieved from airborne full-waveform LiDAR data to incidence angle in discontinuous forest canopy

The aim of this study was to examine the influence of incidence angle on effective Leaf Area index (LAIe) and Canopy Height Profile (CHP) retrieved from small-footprint full-waveform laser altimetry data. LiDAR is a powerful technique for measuring forest structure, but the instrument geometry means that measurements are usually made at a range of angles, so understanding how this affects the results is important. The experiments were carried out using eight swaths of LiDAR data acquired with different incidence angles over five 15m-radius plots in a discontinuous White Cypress PineGillenbah forest in New South Wales, Australia. Two of the plots had very few trees and no understorymaking canopy cover very discontinuous, whereas the remaining three plots had a higher number of trees and some understorypresent, however still not entirely continuous canopy cover. LAIe and CHPs were calculated from raw-waveform LiDAR data using two vegetation-ground reflectance ratios: a default of 0.5 estimated for 1550nm wavelength and a dataset-adjusted reflectance ratio (Armston et al. (2013). Two discrete point methods of LAIe calculation based on ratios of (1) number of single ground points to the number of laser shots (PT1) and (2) the number of all ground returns to the number of all detected returns (PT2) were also tested. A comparison was carried out between LAIe estimates that were not corrected for the incidence angle and estimates that have been reduced by cos(I¸) term to take the influence of the incidence angle out. The LAIe and CHPs were derived for plot-aggregated (15m radius) datasets as well as in 5m grids (grid-processed) in order to assess whether the aggregation area size had an impact on LAIe and CHP repeatability across eight swaths. The plots of relative LAIe (LAIe estimates in each plot normalized by the LAIe value from the swath with the lowest, near-nadir angle) showed some angle dependence of LAIe especially for larger angles of incidence, however, the variability within the data outweighed that trend for all methods. Although correcting LAIe values for incidence angle helped to reduce the scatter in swath comparison, it still left the noise within the data higher than the variable dependence. The incidence angle was, therefore, found to affect the LAIe retrievals at plot-level, especially for angles above 15 degrees, however, this influence was outweighed by vegetation heterogeneity and noise within the data. Repeatability analysis of CHPs acquired from different angles at plot-level showed some small visual differences between the profiles especially for the plots with sparser canopy cover. However, no statistically significant difference was found between the profiles acquired from different angles. Using dataset-adjusted or a constant vegetation-ground reflectance ratio did not have any impact on CHP profiles either. Comparison of CHP profiles from plot-aggregated and grid-processed datasets yielded not statistically significantly different tests in vast majority, however quite a few p-values were on the borderline of significance, especially in the plots with sparser vegetation. This suggests that, while for plots with continuous canopy cover, using large footprint LiDAR data to retrieve vertical vegetation profiles may be sufficient when the terrain is relatively flat, for sparse and discontinuous canopy cover using small-footprint LiDAR data may be advisable.