If you need to generate either a bare-earth DTM or a first-return DSM, you'll first need to determine the appropriate parameters for creating raster surfaces from the LAS dataset:
- Cell size (resolution of the raster)
- Interpolation algorithm
- Void fill options
You should test key areas of interest to identify the optimum settings to create your derived DTM and DSM. Identify a few important sample areas within your project, generate test rasters using different settings, and then evaluate before processing the full dataset. Depending on your project terrain and land cover, suggested test areas should include the following:
- An urban or suburban site
- Any water boundaries (lakes, rivers, coastlines)
- Heavily forested areas
- An area of steep terrain
If any areas were identified during the QC stage with pulse density below the project specification, these areas should also be tested.
If your project extent results in any concave regions along the boundary, these should also be tested to ensure void filling is not attempted outside your project. This may be prevented with a proper polygon that defines the usable extents of your lidar data.
Raster resolution for the DSM can very likely be set to a nominal value appropriate for the full project, but the resolution of the DTM demands greater care. Since datasets with different pixel sizes can be handled in ArcGIS, you could consider splitting your project into multiple regions with different pixel sizes based on land cover, because the density of ground points will normally be lower in forested areas. With lower point density, the DTM surface should be built with a lower resolution (larger cell size) to ensure a statistically valid sampling of the lidar values in each cell.
Recommendations for required parameters
Cell size—DTM
It is expected that you have a desired cell size (spatial resolution) already known, but as a general guideline, for the bare-earth DTM, it is recommended that your cell size be based on the ground point density of your lidar data, with a nominal sampling of approximately 3 by 3 lidar points (which equates to 9 samples per cell, or per pixel) in the DTM. The recommended minimum is 2 by 2, yielding 4 lidar points per cell. See above for creation of the ground point density raster.
Keep in mind that the ground point density will vary across a project. Although the DTM should ideally have multiple lidar points per cell, a compromise will typically be required unless your organization is willing to create DTM tiles at different resolutions in different areas. You may want to consider publishing the ground point density raster (for example, as an image service) as an indication of the spatially variable sampling that exists in your source data, especially if you choose to create the DTM with variable resolution across the project.
Cell size—DSM
For the first-return DSM, some organizations use the same value as for the DTM for data consistency. If the highest possible resolution is desired, the DSM cell size may be as small as one lidar point per cell.
Interpolation algorithm
If you do not have feature classes (breaklines, hydrology, and so forth) to constrain the DTM, the typical recommendation for the interpolation algorithm is to use Binning and cell assignment type Average for the DTM. In the case of adding feature constraints for the DTM (described below), it will be necessary to use one of the Triangulation options, typically with Natural Neighbor void filling.
The exception to this is in areas where the ground point density may be very sparse, for example, under heavy tree canopy. If use of binning results in many void areas, it is recommended to either increase the pixel size or switch to triangulation.
For the DSM, presuming it will be used for visibility analysis, it is typical to set cell aggregation type to Maximum, but keep in mind it is important to remove any noise (for example, check for outlier points such as bird strikes) prior to this step.
Void fill method
The recommended selection is typically Natural Neighbor, using a polygon to define the usable data extents (see above). This clip polygon is necessary to prevent the interpolation from filling outside the areas of valid data. Choosing None may be appropriate if your users want control over the void filling process; this will create a DTM surface with voids (NoData) in any cells not containing lidar points. In this case, note that the Elevation Void Fill raster function can be applied to fill those voids on the fly.
Tools to create raster surfaces
After selecting appropriate test areas, it is recommended to create sample outputs and evaluate the quality of each based on your project needs. When the best parameters have been determined, you can proceed with creating output surfaces for the full project area.
Before creating sample outputs, review Create derived rasters to identify which workflow is applicable to your data and organization.
You can use geoprocessing (GP) tools built into ArcGIS to create raster outputs, for example, LAS Dataset to Raster or Terrain to Raster (as appropriate, based on options in Create derived rasters), but note that these tools will create one single output file. Since some lidar collections are very large, creating a single file may not be scalable. Additional GP tools have been created to manage large datasets by outputting rasters as multiple tiles: LAS Dataset To Tiled Rasters and Terrain To Tiled Rasters. These tools can be downloaded from 3D Samples.
These tools are ideal for creating test outputs, since they can accept a feature class as input, to define areas to process. You can create multiple area of interest (AOI) polygons in a single feature class and use this as input to the tool to create test rasters with various parameters. If you are using the built-in GP tools (LAS Dataset to Raster or Terrain to Raster), small sample areas can be generated by using the environment setting for Processing Extent to limit the processed area.