Why ground model from point cloud in Global Mapper is inaccurate?

Question

I have a LAS file that I assumed was based on LiDAR, but it is actually based on ADS (airborne digital sensor) imagery, according to the body that captured the data.

I load the file in to Global Mapper and generate a point cloud which is totally unclassified. Using the auto-generate ground points feature, about 60% of the points become ground. However, when I generate an elevation grid, the areas that should be flat (in this case, fairways on a golf course) remain quite bumpy.

Is my point cloud inaccurate because it's not LiDAR?

Below are images of the area in question.

Answer 1

Is my point cloud inaccurate because it's not LiDAR?

Not exactly. All types of point clouds (from laser scanning, stereo images, etc) can have varying degree of accuracy, resolution, type of information, etc.

---

You can try the following to generate a better output:

1. Check for outliers.

   First thing, check if the point cloud has any type of outliers. If it has, it can compromise all subsequent processing steps in the analysis. See Editing LiDAR point cloud to remove noise/outliers present below and above ground?.

2. Set the ground filtering algorithm/tool's parameters accordingly.

   It is clear from the DTM (which is a TIN representing the terrain surface) image that it was generated using non-ground points to triangulate as well (see the bigger triangles which are exactly where the trees are). Global Mapper's Lidar Automatic Ground Classification has some parameters which are user-defined. Understanding what each one does and testing/setting them accordingly is essential to avoid labeling errors (classifying non-ground points as ground).

   For example, given the scene is significantly flat, and the point cloud is of low to medium resolution:

   • Set the Maximum Height Delta parameter to a value smaller than the height of the tallest trees (e.g. 70% of average height from trees). This should eliminate the presence of major trees in the DTM.
   • Set the Base Bin Size parameter based on the point cloud's point spacing (at least 2 times the average point spacing) AND make sure it will also be greater than the largest tree crown diameter in the scene. This will make sure every bin has at least one actual ground point to compare with. This is also valid for the same parameter when generating the DTM/DEM (see here).
   • Start with the default value for Minimum Height Departure from Local Mean, but also test greater values.

   
   This will help minimizing labeling errors, but will generalize the DTM as well. You can make tests until you find the best trade off.

3. Another method for ground point classification.

   Take a look in additional options for ground point cloud classification. For example, Determining bare earth DEM from unclassified LAS file?. Likewise in step 1, in every tool/algorithm, make sure all parameters were set accordingly, so be sure the best of each tool/algorithm was tested.

4. Generalize the DTM by making a DEM (raster surface).

   A Digital Elevation Model (DEM) is a gridded raster representation of the DTM (see What is the difference between DEM, DSM and DTM?), i.e., a form of generalizing the DTM.

   Depending on your needs about the terrain information (some uses of it need more accuracy than others), you can generate a generalized DEM with a large pixel size (the same reasoning explained for ground classification in #2) so to capture more than one ground point per pixel and avoid interpolation.

   In Global Mapper, assuming the point cloud is already correctly classified, Create Elevation Grid from 3D Vector Data (with LiDAR data), select the Grid Method as Binning (Minimum Value - DTM) or Binning (Average Value).

   _{Considering the final goal to visualize terrain information and considering previous results obtained with the TIN-DTM, do not select Binning (Maximum Value - DSM) as the presence of non-ground points (outliers) might shift the DEM upwards; and also do not select Triangulation (Grid TIN of Points) as it will produce a raster surface with pixel size smaller than most of the triangles in the intermediate TIN and will yield similar results to the TIN-DTM you already have, see here}.

   Don't forget to filter by class code 2 (ground) in Filter LiDAR points ... as well.

   Last, you can also try smoothing/denoising the output DEM, in the most possible way. See some GIS SE posts about DEM generalization/smoothing:

   • What raster smoothing/generalization tools are available?
   • Smoothing DEM using GRASS?
   • How to de-noise a DSM?
   • Cleaning blocks pattern from urban LIDAR (blocks uplifted from streets)

---

Here is a YouTube tutorial video from Blue Marble Webinars with more info about making DTM and DEM in Global Mapper.