Creating better digital surface models from LiDAR points

In order to generate a Digital Terrain Model (DTM) that represents the bare terrain, LiDAR ground returns are typically Delaunay triangulated into a Triangular Irregular Network (TIN) that is then rasterized onto a grid at a user-specified resolution. In a similar same manner all of the first returns are usually interpolated in order to generate a Digital Surface Model (DSM) that represents the elevation of vegetation and objects. A Canopy Height Model (CHM) can be generated (1) either by subtracting the DTM raster from the DSM raster or (2) by replacing the elevation of each first return with its height-above-ground before interpolating them into a raster. This standard way of DSM and CHM generation has two drawbacks: (1) using only first returns means that not all LiDAR information is used meaning that some detail is missing and (2) using all first returns practically guarantees the formation of needle-shaped triangles in vegetated areas that appear as spikes in the TIN. These spikes represent extreme irregularities in the triangulated surface that result in "data pits" in the corresponding raster. These "pits" limit the utility of the DSM or CHM especially in forestry application, for example, to extract individual trees. We describe an algorithm that can generate significantly better Digital Elevation Models (DEMs) or Canopy Height Models (CHMs) from LiDAR point clouds compared to the "standard" approaches by constructing a spike-free TIN from all relevant returns taking the survey's pulse spacing into account. A working prototype of this algorithm was implemented and will soon be available as an integral LiDAR processing module of the LAStools software suite.