Waveform and discrete LiDAR effective LAI estimates: sensitivity analysis

This study has investigated how average effective leaf area index (LAIe) derived from full-waveform and discrete LiDAR data changes depending on the size of the grid used, over a 150 m by 80 m area of orange orchard. The full-waveform data, acquired with RIEGL LMS-Q560, were decomposed and optimized with a trust-region-reflective algorithm using a custom decomposition procedure focused on extracting denser vegetation point clouds. LiDAR effective LAI (LAIe) estimates were derived in two ways: (1) from the probability of discrete pulses reaching the ground without being intercepted (discrete point method) and (2) from raw waveform canopy height profile processing adapted to small-footprint laser altimetry (waveform method). The LAIe estimates for the orange orchard were derived for the whole site as well as in various decreasing grid cell sizes. The discrete point method provided estimates that were 5-10% higher than those of the waveform method, and this difference increased with the decreasing grid cell size. The only exception was the smallest grid (2.5 m) for which the relation was opposite. This was due to the discrete method being limited by the point density. Furthermore, percentage of vegetation cover in the test area was estimated based on aerial photography, and used to derive an average single tree effective LAI depending on the grid cell size. Consequently, to test the effects of vegetation discontinuity on LAI estimation the values of LAIe for the whole site were simulated based on a set of increasing single orange tree LAIes (from 0.2 to 5 with 0.2 increments) and known vegetation cover in the test area. This was done by predicting the LAIe of the orange tree covered area and averaging it with the LAIe of the bare soil area (LAIe=0). These ‘average’ LAIe values were compared to the ‘overall’ LAIes calculated for the whole site from summed probabilities of penetration for the orange tree area and ground area (PgapG=1). As expected, with the increasing LAIe of a single tree, the area LAIe increased as well. However, as the LAIe of single tree increased, the difference between the ‘average’ LAIe values and the ‘overall’ LAIe values increased significantly, from 5% for a single tree LAI of 0.2 to 73% for a single tree LAIe of 5.0, showing underestimation of LAIe by the latter method. The LiDAR LAIe estimates for the whole study area (simulated large-footprint laser system) differed to those computed as the mean LAIe value in a 5m by 5 m grid by 14% with the latter estimates agreeing well with simulated LAIe values of the whole area when the ‘average’ approach was used (mean single tree LAIe of 1.6). SilviLaser 2013, October 9-11, 2013 –Beijing, China 2