Questioning voxel grids: Semi-continuous sampling of leaf area density using airborne waveform lidar in boreal and hemiboreal conifer and broadleaved forests

Abstract

Plant area density measurements provide spatially explicit information about the density and distribution of canopy elements. This information is needed for modeling of the forest radiation regime, climate and for other ecological applications. Terrestrial laser scanning (TLS) provides detailed information about canopy structure, but it cannot be used for monitoring large areas. Airborne laser scanning (ALS) uses similar methods to measure plant area density, but due to the larger beam footprints, the scale at which this information can be obtained is coarser than with TLS. The volumetric nature of the ALS measurement poses unique geometric challenges to plant area measurement methods, as assuming an infinitesimal beam size may lead to large errors. Further, the use of voxel grids with ALS measurements may increase errors in plant area measurements, as these grids require discrete spatial allocation of information.

In this study, we apply a spatial weighting technique to ray-traced measurements of plant area from ALS data. This spatial weighting scheme allows continuous allocation of trajectory information of ALS pulses, avoiding discontinuity introduced by voxel grids.

Our data consisted of high density ALS waveform data (over 40 points/m${}^2$) in 33 plots across two study sites in Finland and Estonia. We compared the plant area index (PAI) obtained through this new measurement method to PAI measurements from hemispheric photography (HP) and TLS, and to ALS with a voxel grid. We found PAI, measured at agrid spacing of 0.6 m, correspond best to HP and TLS measurements. Occlusion severely biased PAI at 0.2 m spacing. With increasing grid spacing, PAI estimates become increasingly biased because of clumping effects at small scales. Continuously sampled PAI measurements corresponded closer to reference values than voxel-based PAIs, indicating that a spatially weighted approach avoids bias from partitioning the volumetric ALS beams into voxels.