LiDAR, One-Foot Contours, and a Swale That Isn't Where It Says
Your firm delivered a LiDAR-derived bare-earth DEM for a 6-acre school site. The vendor reports a fundamental vertical RMSEz of 0.28 ft against 42 checkpoints, and 0.55 ft at the 95th percentile in the vegetated portions of the site. The architect's grading plan uses 1-ft contours and sets the finished floor 1.2 ft above the bottom of a swale that crosses the site. Your own field checks with a total station and level show the DEM is about 0.8 ft different from ground at three points along the swale bottom, all in the same direction. The DEM also shows a 36-inch storm drain rim as ordinary ground surface.
1. Is this DEM adequate to support 1-ft contours for grading? Show the reasoning.
Read the worked answer
Test against the contour interval, not against an impressive-sounding RMSE.
A common standard is that vertical RMSE should be no more than about one-third of the contour interval for the mapping to be meaningful, and accuracy statements are conventionally reported at 95 percent confidence (roughly 1.96 x RMSEz for a normal distribution):
- Required for 1-ft contours: RMSEz on the order of 0.30 ft or better as a site-wide figure, with 95 percent error near 0.6 ft.
- Reported fundamental RMSEz 0.28 ft: 1.96 x 0.28 = 0.55 ft at 95 percent - marginal even in the open.
- Vegetated land cover, 95th percentile: 0.55 ft directly reported, i.e. about 1.96 x 0.28 nominally but actually derived from a weaker subset. In places the error approaches the contour interval itself.
Conclusion: the DEM is not adequate as the sole basis for 1-ft contours and certainly not for a design decision about 1.2 ft of freeboard above a swale. When the vertical uncertainty is comparable to the feature you are designing against, the survey is not measuring the design constraint - it is measuring its own noise.
Also note the distribution problem: a site-wide RMSE can be met while the specific place you care about (a narrow, vegetated swale bottom, where bare-earth classification is hardest) is the worst-performing part of the dataset.
2. The DEM is systematically 0.8 ft off at the swale, all in one direction. What are the likely causes and how do you run it down?
Read the worked answer
A consistent offset is a systematic error, and systematic errors are diagnosable. Work through:
- Datum and geoid. Is the DEM in NAVD 88 orthometric heights via a specific geoid model, or in ellipsoid heights, or in a local datum the vendor inherited? A geoid model mismatch or an unstated vertical datum produces exactly this signature.
- Control fit. Did the vendor constrain to your project benchmark, or to their own? If their control and yours disagree, the whole surface shifts.
- Bare-earth classification. In dense vegetation, ground returns are sparse and the algorithm interpolates - often low, sometimes biased. Interpolated ground under a canopy is an estimate, not a measurement.
- Feature misclassification. The storm drain rim appearing as ground is a classification failure. It tells you the classifier is making decisions you cannot rely on in exactly the areas where structures and drainage matter.
- Units. International foot versus US survey foot is only about 2 ppm and cannot produce 0.8 ft - but check anyway, because it is free to rule out.
Run it down: establish a level loop from your project benchmark to the three swale checkpoints and to several checkpoints in the open, where the DEM claims its best accuracy. If the open-ground checks agree and the swale does not, it is classification/interpolation, not datum. If everything is off by the same amount, it is datum or control.
Do not "correct" the DEM by adding 0.8 ft. Re-observe the critical areas and use the field data.
3. What do you deliver, and what do you write on the drawing and in the metadata?
Read the worked answer
Deliver a hybrid surface, and say so. The defensible product is the DEM for general terrain, augmented by field survey of everything the design depends on: the swale bottom and its breaklines, the storm drain rim and invert, tops of walls, and any hard surface that will carry runoff.
Metadata that must accompany the surface:
- Source (LiDAR acquisition, vendor, date of flight) and processing level (bare earth, classification method).
- Horizontal datum, realization, epoch, and coordinate system, in stated units.
- Vertical datum and geoid model, and whether heights are orthometric or ellipsoidal.
- Accuracy: RMSEz and the 95 percent figure, reported by land cover class, not as one site-wide number.
- The control held for the fit, and residuals at checkpoints.
- Anything field-verified or field-replaced, and what was not.
On the drawing: note that contours are derived from LiDAR with the stated accuracy, that the swale and drainage structures were field-located, and identify the datum. Add a clear note that the surface is not a boundary or legal document.
The judgment call to make explicitly: the floor is only 1.2 ft above the swale, and the swale is the single place where the data are weakest. Tell the architect in writing that the freeboard is within the data's uncertainty and must be confirmed by field survey before the floor elevation is locked. A surveyor who delivers a DEM and a contour file without that sentence has delivered a liability with a metadata header.