A LiDAR DEM That Checks Out Poorly on the Ground
A 240-acre floodplain study uses a 2019 LiDAR-derived 1-m DEM, classified to bare earth, with a stated RMSEz of 0.10 ft in open terrain and metadata citing NAVD88 (GEOID12B). You check it against 62 independent RTK shots on open pasture distributed across the site. The DEM reads higher than the shots by a mean of +0.44 ft with a standard deviation of 0.29 ft. Along a 900-ft wooded drainage ditch, the DEM surface sits 2.1 ft above RTK shots taken on the ditch bottom. Your RTK vectors were reduced using GEOID18; the geoid model difference between GEOID12B and GEOID18 at this location is 0.04 ft. The county floodplain ordinance requires the best available topography and a vertical accuracy statement with the submittal.
1. Compute the accuracy indicated by your check and decide whether the DEM can be accepted as delivered.
Read the worked answer
Combine bias and scatter to get a total error, and compare it with the claim.
RMSE = √(0.44² + 0.29²) = √(0.1936 + 0.0841) = √0.2777 ≈ 0.53 ft
At 95% confidence, 1.96 × 0.53 ≈ 1.03 ft.
The claimed RMSEz is 0.10 ft in open terrain. The check shows roughly five times that, and the bias is the dominant term: a systematic 0.44 ft offset is not random noise, and it will not average out across the site.
The DEM cannot be accepted as delivered for this purpose without explanation and correction. Two caveats should accompany that conclusion: 62 points is a modest sample, so check spatial distribution and independence, and make sure the comparison is point-to-surface at the same horizontal location, not a comparison of interpolated values against shots with their own uncertainty. Also verify that the shots are on ground that is genuinely open, not near canopy edges.
2. Diagnose the likely causes of the 0.44 ft bias in the open and the 2.1 ft discrepancy in the ditch, and state what you would do about each.
Read the worked answer
The geoid difference of 0.04 ft cannot explain 0.44 ft, so look further. A uniform offset in open terrain suggests one of: a vertical datum or geoid model misapplied in one of the two datasets, an antenna height or base station error in the RTK work, a calibration or control problem in the LiDAR block adjustment, or a DEM product that has been smoothed or hydro-flattened away from its raw classified points.
Actions: check both datasets against a published NGS benchmark with a known NAVD88 elevation; confirm the geoid model and epoch used in each; verify the RTK antenna height, base coordinates, and the receiver's geoid setting; review the LiDAR project report for the control and the classification method; and check whether the DEM is the raw bare-earth surface or a derived/hydro-enforced product.
In the ditch, the cause is different: dense canopy and steep banks mean few ground returns reach the bottom, and the bare-earth algorithm interpolates across them, producing a surface that bridges the channel. The 2.1 ft offset is a sampling failure, not a calibration error, and it will not be fixed by a site-wide correction.
Use field-surveyed breaklines and cross sections for the ditch and channel, and use the DEM only where ground returns are dense and verified.
3. Draft the vertical accuracy and data-use statement that should accompany the submittal.
Read the worked answer
The statement should let a reviewer reproduce your conclusion and see where the DEM is and is not used.
It should include:
- Source and vintage of the DEM, cell size, and classification level.
- The publisher's stated RMSEz and the geoid model cited in the metadata.
- The independent check: number and distribution of RTK points, terrain type, mean difference, standard deviation, computed RMSE, and 95% value.
- The vertical datum, realization, and geoid model used for both the DEM and the field check, including the 0.04 ft model difference and the fact that it does not account for the observed bias.
- The open-terrain bias and its treatment: corrected, not corrected, or under investigation with the publisher.
- A clear statement that channel and ditch geometry is from field survey, not from the DEM, with the reason.
- The intended use and limitations: floodplain mapping at the site scale is acceptable only after the bias is resolved; hydraulic modeling of the ditch uses surveyed sections.
End with a short, plain conclusion: the DEM as delivered does not meet its published accuracy at this site, and here is what was done instead. That is what the ordinance requires and what a reviewing engineer needs.