Generating spike-free digital surface models using LiDAR raw point clouds: A new approach for forestry applications

Abstract

Accurately detecting single trees from LiDAR data requires generating a high-resolution Digital Surface Model (DSM) that faithfully represents the uppermost layer of the forest canopy. A high-resolution DSM raster is commonly generated by interpolating all first LiDAR returns through a Delaunay TIN. The first-return 2D surface interpolation struggles to produce a faithful representation of the canopy when there are first returns that have very similar x-y coordinates but very different z values. When triangulated together into a TIN, such constellations will form needle-shaped triangles that appear as spikes that geometrically disrupt the DSM and negatively affect treetop detection and subsequent extraction of biophysical parameters. We introduce a spike-free algorithm that considers all returns (e.g. also second and third returns) and systematically prevents spikes formation during TIN construction by ignoring any return whose insertion would result in a spike. Our algorithm takes a raw point cloud (i.e., unclassified) as input and produces a spike-free TIN as output that is then rasterized onto a corresponding pit-free DSM grid. We evaluate the new algorithm by comparing the results of treetop detection using the pit-free DSM with those achieved using a common first-return DSM. The results show that our algorithm significantly improves the accuracy of treetop detection, especially for small trees.