A vertical datum is the reference surface that an elevation is measured from. An elevation on its own is meaningless; it is always a height above something. A geoid model is the mathematical surface that lets a satellite-derived height be converted into a height above that datum. Together they decide whether the elevations in your deliverable agree with the elevations in everything else you own.
Why there are two heights
A GNSS receiver, including the one in an RTK drone or on a survey rover, does not natively measure height above sea level. It measures ellipsoidal height, which is the distance above a smooth mathematical ellipsoid fitted to the whole earth. In Canada that reference frame is NAD83(CSRS). The ellipsoid is convenient for satellites and useless for water, because it does not follow gravity.
What engineering actually needs is orthometric height, a height above a gravity-based surface that water will run downhill on. That surface is the geoid, and the national realisation of it is CGVD2013, computed through the geoid model CGG2013a. The conversion between the two heights is the geoid separation, and across Alberta it is on the order of tens of metres and varies from place to place.
The practical consequence: a drone survey processed to ellipsoidal height and handed to a designer expecting orthometric height is not slightly wrong. It is wrong by the geoid separation at that site, which is an enormous number in grading terms, and it is wrong in a way that looks internally consistent.
The older datum is still in service
Many Alberta drawings, benchmarks, legal surveys and as-builts are on CGVD28, the earlier Canadian vertical datum. CGVD2013 is the current official datum, but CGVD28 has not disappeared from the files that matter on a site.
The two do not agree. The offset between them varies geographically, and it is comfortably large enough to matter against a grading or freeboard tolerance. The number for a specific site comes from the published conversion tools, not from a rule of thumb, and the rule to work by is simple: find out which datum the existing site data is on and match it, rather than delivering the current datum because it is current.
What a datum mismatch looks like in the deliverable
This is the useful part, because the symptom is distinctive. A datum or geoid mismatch produces a smooth, nearly uniform vertical offset across the entire site. The surface has the right shape, the slopes and the relative dimensions are correct, the volumes between two of your own surfaces are correct, and everything is displaced up or down by about the same amount.
If a cut and fill map says the whole site is uniformly underbuilt by a similar amount everywhere, including ground nobody has touched, the problem is almost never accuracy. It is a frame problem: the two surfaces are expressed against different reference surfaces. A genuine accuracy problem looks nothing like this, because real error varies across a site and is noisy rather than uniform. The same diagnosis applied to a design comparison is set out in comparing an as-built against a design surface.
It follows that a mismatch is usually recoverable. The data was measured correctly and can be re-expressed in the right datum, provided the record says what it was measured in. A dataset with no stated datum is the genuinely expensive case.
What to put in the scope
Four lines prevent nearly all of this:
- The horizontal coordinate system, which is a separate decision and is covered in which coordinate system your survey should be in.
- The vertical datum by name, CGVD2013 or CGVD28, matched to the existing site data rather than chosen fresh.
- The geoid model used for the conversion, named in the deliverable metadata.
- The control that ties it, meaning which benchmark or published station the heights are held on, and whether an independent check point was observed. Ground control is the mechanism, and the count and placement question is covered in GCP and how many ground control points you need.
On a repeating measurement program, those four lines are fixed once and reused every year, because a mid-program datum change produces an apparent step in the data that looks like material movement and is not.