Within the ASPRS guidelines and USGS specifications there is significant room for ambiguity in the end result caused by the user’s ability to select which points are considered when building a TIN for swath-to-swath and absolute accuracy computation. This ambiguity ultimately stems from the precision of data that a scanner is producing and the ground classification routine chosen. Current LiDAR systems available on the market, especially when mounted to slow and low flying UAS, produce extremely dense data, where the scanner’s precision is a very noticeable quality factor. A variety of methods for classifying ground and producing a final DEM exist, all possibly impacted by precision.
In short, a fuzzy layer of hard surface points leaves room for the implemented workflow to have a dramatic influence on accuracy metrics. We will illustrate 3 scenarios for an imprecise scanner that can greatly impact the absolute accuracy resulting statistics:
All Points Surface: Datasets lacking high precision will have a somewhat “fuzzy” layer of points on hard surface targets. If the entire layer of fuzz is considered to be “ground” by the user then the over and under estimation of absolute accuracy will be unpredictable. In this example, the “All Points Surface” will overestimate the accuracy of checkpoint A, and underestimate the accuracy of checkpoint B. User’s may choose to use this way of looking at their data in order to remain independent of classification algorithm or parameter selection.
Mean Surface: Creating a mean or other statistically interpolated surface that fits near the center of a dataset’s fuzz can reduce the randomness of using an all points TIN. It can also better represent the true surface by minimizing the impact of imprecision. This method is particularly useful for quantifying accuracy when the user is making products, DEM, Contours, Etc., from the interpolated surface. If this method is chosen then it is advisable to also convey a dataset’s precision to the end user as it may give a misleading impression if the full point cloud is later used to measure features that are impacted by imprecision. This is similar to the method outlined by M. Isenburg at the LiDAR for Drone 2017 Conference: https://rapidlasso.com/2017/10/29/processing-drone-lidar-from-yellowscans-surveyor-a-velodyne-puck-based-system/
Generic Ground Classified Surface: Many ground classification routines tend to work very well with lower density airborne lidar data. With point densities in the 2-8 points/m² range on the ground, common ground classifiers are sufficient for surface modeling. However, on very dense datasets the lowest elevation points tend to get selected. These points also often happen to be outliers. Comparing surveyed checkpoints to elevation values from a TIN of generic ground classification results has the potential to produce undesirable results.
