There are three distinct questions hiding under the phrase datum transformation, and the National Geodetic Survey publishes a different tool for each. Confusing them is common, because all three take coordinates in and give coordinates out, and all three look like they are doing the same thing.
They are not. One changes horizontal datum, one changes vertical datum, and one moves a coordinate through time within a datum. Knowing which problem you have is most of the work.
The three problems
| The question | The tool | What it does |
|---|---|---|
| My coordinates are on NAD 27 and I need NAD 83 | NADCON | Interpolates a modelled grid of horizontal shifts |
| My coordinates are on one NAD 83 realization and I need another | NADCON 5.0 | Grid shifts between realizations, including some ellipsoid height support |
| My elevations are on NGVD 29 and I need NAVD 88 | VERTCON | Interpolates a modelled grid of vertical shifts |
| My coordinates are at one epoch and I need another | HTDP | Applies plate rotation, a crustal velocity model and earthquake displacements |
| My coordinates are on NAD 83 and I need a global frame | HTDP | Performs the frame transformation as well as epoch propagation |
NADCON
NADCON exists because the difference between NAD 27 and NAD 83 is not a simple geometric transformation. Part of it is — a different ellipsoid and a different origin move everything in a smooth pattern — but the larger and more troublesome part is accumulated distortion in the NAD 27 network, which is irregular over distances of tens of kilometers and cannot be modelled by any number of rotation and translation parameters.
So NADCON does not use parameters. It uses a grid: the shift in latitude and the shift in longitude were computed at points where both NAD 27 and NAD 83 coordinates exist, gridded, and published. You interpolate that grid at your position.
The consequence for accuracy is important and frequently misunderstood. NADCON is not accurate to the precision it prints. Its accuracy is limited by the quality of the NAD 27 network in your specific area, which is to say that the answer is good where the old control was dense and good, and poor where it was sparse or strained. Treat NADCON output as a search position for a monument, not as a substitute for finding it.
NADCON 5.0 extended the same approach to transformations between NAD 83 realizations and other frames, with grids for each pair. It is the tool for the question my data is on NAD 83 (HARN) and I need NAD 83 (2011).
VERTCON
VERTCON is the vertical analogue: a gridded model of the difference between NGVD 29 and NAVD 88, interpolated at your position. It exists for the same reason — the difference is a spatially varying surface, running from roughly −0.4 m to about +1.5 m across the conterminous United States, and no constant is correct outside a small region.
Its accuracy is likewise governed by the underlying data. Where benchmarks with both NGVD 29 and NAVD 88 elevations were dense, the model is good to a couple of centimetres. Where they were sparse, it is worse. And, as always, a recovered benchmark with a published NAVD 88 elevation beats a converted NGVD 29 value every time.
HTDP
HTDP — Horizontal Time-Dependent Positioning, despite the name now handling three-dimensional positions — solves the problem the other two do not touch: the ground moves. It transforms coordinates between reference frames and between epochs, combining rigid plate rotation, a model of crustal velocity where the plate does not move rigidly, and a catalogue of coseismic displacements from significant earthquakes.
The two uses that come up most often are propagating a coordinate observed on one date to the reference epoch of a datum realization, and converting between NAD 83 and a contemporary international frame — which is where that one-to-two-meter difference caused by the NAD 83 origin offset is properly handled.
X(t₂) = X(t₁) + v × (t₂ − t₁) + Σ coseismic displacementsIn the stable plate interior HTDP will change a coordinate by very little, and it is tempting to skip it. In the tectonically active west it will change a coordinate by a decimetre or more over a couple of decades, and skipping it is a blunder. The correct habit is to run it and look at the answer rather than to guess whether it matters.
Practical discipline
- Identify the actual problem first: horizontal datum, vertical datum, or time. Do not reach for a tool until you know which.
- Record the tool and the version used, and the grid or model it applied. Transformations change between versions.
- Never apply two transformations to the same data and average them.
- Never chain transformations through an intermediate datum without checking, since each step adds its own model error.
- State the transformation on the deliverable. A converted coordinate should always be identifiable as converted.
- Where the answer matters, verify against a recovered monument with a published coordinate on the target datum.
The rule that survives all of this
A transformed coordinate is always less accurate than an observed one, because it carries both the error of the original coordinate and the error of the model. That is not an argument against transforming; sometimes the original observations are gone and a model is all that remains. It is an argument for being honest about what you produced, and for preferring an occupation over a computation whenever an occupation is available.
It also explains why the modernization effort puts so much weight on labelling. Every transformation depends on knowing exactly what you started from. A coordinate whose datum, realization and epoch are unknown cannot be transformed at all — only guessed at.