A point cloud records where the surface is. A confidence layer records how far each point has been checked. GroundMap grades every point of a fused model into one of five classes. A pass is any separate capture of the same ground: another flight at a new heading or angle, or an earlier survey aligned to the same frame.
The five classes
- 3 passes agreeAfter alignment, both other passes measure the surface height within 15 cm of the reference pass.
- 2 passes agreeAfter alignment, two passes measure the surface height within 15 cm of each other.
- Single passMeasured, not cross-checked: one pass reached it, or it lies below the top surface, such as under canopy.
- Change detectedPasses differ by more than 15 cm: real change, or disagreement between surveys. Change maps test it separately.
- ModelledFilled by geometric modelling where no pass measured, such as walls. Flagged, never counted as confirmed.
Agreement is tested between passes, after alignment. Only the first two classes are cross-checked; the third is measured but not cross-checked; the fourth marks a difference; the fifth was never measured.
Tolerances are set per scan type: 15 cm between airborne surveys, 10 and 30 mm between terrestrial scans.
Evidence: laser-measured, image-measured, modelled
Fusion Spectra adds points measured from overlapping imagery where the laser has none, each with its own confidence; they are measured, never modelled.
- Laser-measuredA LiDAR return. Graded by how many aligned passes agree on the surface.
- Image-measuredMeasured from overlapping images by photogrammetry and stereo, with its own confidence from the image geometry.
- ModelledA surface no pass measured, filled by geometric modelling. Flagged in every deliverable, never counted as measured.
What each class does and does not say
Agreement is not absolute accuracy. Aligned passes can agree closely and still share an offset from true height. Absolute accuracy is a separate test against independent survey control, reported to the data’s specification; the USGS quality levels, for example, state it as RMSEz. How GroundMap measures and states accuracy.
Single pass is not a fault. It is measured data with nothing to check it against: ground under canopy, or a surface only one pass reached.
Change detected is a flag, not proof of change. It marks a difference beyond 15 cm: real change, or disagreement between surveys. Change maps and volumes test it against each pair’s measured scatter. The newest pass describes the site today; the older pass is its history.
Modelled is never measured. It completes surfaces no pass reached, such as walls. Modelled points carry the LAS synthetic flag, which the ASPRS specification defines as a point created by a technique other than direct observation, so any LAS software can exclude them.
The practice has precedent in public elevation data. The Copernicus DEM, derived from TanDEM-X radar data, is distributed with a height error mask giving each pixel’s height error as a standard deviation, and an edit mask marking every pixel modified during editing.
Using the classes
- Volumes
- GroundMap computes volumes from measured points only; modelled points are excluded by their flag. Where a volume boundary crosses Change detected, GroundMap takes the newest pass for the current state and reports the difference as change only where the change map confirms it.
- Design
- Design levels are set on ground where passes agree; on Single pass ground GroundMap states that survey’s accuracy or adds a pass or check survey before construction.
- Reporting
- Every delivery states the share of the area in each class and the 15 cm tolerance, with the classes in the data.
GroundMap Fusion grades every point in these five classes.