Centimetre Accuracy: RTK, Datums and Why Coordinates Disagree

The problem nobody can settle

An aerodrome has two surveys of the same runway, taken years apart by two competent contractors. Both report centimetre-level accuracy. The thresholds disagree by tens of centimetres, and one of them puts an obstacle on the wrong side of a limit.

The usual reaction is to decide one contractor was sloppy. That is almost never what happened. Both surveys can be internally excellent and still disagree, because accuracy is a statement about a reference frame and the two numbers were never in the same one. A coordinate is not a fact about the world. It is a fact about the world plus the frame it is expressed in, and if the frame is missing the number cannot be checked.

What RTK fixes, and what it does not

RTK works by putting a second receiver on a known point close to the rover. Both see the same satellites through roughly the same atmosphere, so the errors that dominate standalone GPS positioning , namely satellite orbit and clock error, ionospheric and tropospheric delay, are common to both and cancel in the difference. What survives is small enough that the receiver can resolve the integer number of carrier wavelengths on each signal. That resolution is what buys centimetre accuracy .

So RTK fixes the error between the base and the rover. It does nothing at all about the error in the base.

This is the part that gets skipped. Every rover position is the base coordinate plus a measured vector. If the base coordinate was entered from a national network, the survey lands in that network. If the base autonomously averaged its own position for ten minutes, the survey lands wherever that average landed, and the whole dataset is rigidly shifted by that amount while every internal check still looks perfect. Two crews with identical equipment can produce two datasets that are each centimetre-consistent and metres apart. That is a base problem, not an RTK problem.

Three words that are not synonyms

Field reports use precision, accuracy and repeatability as if they meant the same thing. In practice they answer different questions, and only one of them is what an aerodrome actually needs.

A survey that reports horizontal accuracy without naming the control it was checked against has reported precision and called it accuracy. That distinction is the whole subject of measurement accuracy as a discipline, and it is why we tie every mission to surveyed control on the airfield rather than to whatever the base receiver decided about itself.

Repeatability is the one that matters most for change detection. If you are comparing this year’s pavement surface to last year’s, a stable common offset is harmless and noise is not. If you are checking an obstacle against a protection surface, the offset is exactly what will get you.

Vertical is always the worse number

Every survey specification quotes a vertical figure that is worse than its horizontal figure, and the ratio is remarkably consistent across technologies. There are two reasons, and they compound.

The geometric one. Satellites are above the receiver and never below it. The horizontal solution is constrained by satellites spread around the whole sky, while the vertical solution is constrained only from one hemisphere. That asymmetry is baked into the geometry and no amount of averaging removes it.

The atmospheric one. Tropospheric delay is very nearly a vertical error. It maps almost directly into height and only weakly into position on the ground, so the residual after correction lands mostly in the number you trust least already.

Then there is a third reason that is not about the instrument at all, and it is the one that actually causes the arguments.

The datum problem

Horizontal: which frame, and when

WGS84 is the frame ICAO requires for published aeronautical coordinates, and that requirement is why everyone assumes coordinates are comparable by default. They are not, for two reasons.

First, most national survey infrastructure is not global. National and regional frames such as the European ETRS89 family are defined to sit still relative to the local tectonic plate, so that a boundary marker keeps the same coordinate decade after decade. Global frames do the opposite. They track the actual motion of the crust. ETRS89 was defined to coincide with the global system at epoch 1989.0 and to co-move with the stable part of the Eurasian plate, which means the two have been separating ever since at the speed the plate travels. A coordinate in one is not a coordinate in the other, and the gap grows.

Second, that is why an epoch is part of the coordinate. In a global frame, a position on a moving plate changes every year by design. Quoting the frame without the epoch is like quoting a temperature without saying whether it is Celsius. The number looks complete and is not.

Vertical: ellipsoidal height is not elevation

A GNSS receiver does not measure elevation. It measures height above a mathematical ellipsoid, which is a smooth surface fitted to the whole planet and has no physical meaning at any particular airfield. Elevation as an aerodrome uses it is orthometric height, measured from the geoid, the surface that water would settle to under gravity alone.

The two differ by the geoid undulation, and converting between them requires a geoid model. Ellipsoidal height minus geoid undulation gives orthometric height. The undulation is not small, it varies across a country, and different geoid models give different answers for the same point.

So a survey can hand you a height that is centimetre-correct as an ellipsoidal height and is not the elevation your aerodrome chart uses. This is not an edge case. ICAO requires geoid undulation to be measured and reported alongside aerodrome and threshold elevation precisely because the conversion is the step that goes wrong. A vertical datum is a choice someone made, and it has to be written down.

What makes a survey actionable

Here is the difference between a deliverable you can compare next year and one you cannot.

Any row marked missing turns the survey into a picture rather than a measurement. It still looks fine. It just cannot be argued with, defended to an authority, or compared to the survey that comes after it.

What to put in the specification

When commissioning geodetic survey work, name the frame rather than the accuracy. Accuracy without a frame is unenforceable, and every contractor will meet it.

State the horizontal frame, its realization and the epoch you want results delivered in. State the vertical datum and the exact geoid model, by version. Require that the control used is named in the report, that it is on your airfield, and that it is the same control next time. Ask for horizontal and vertical accuracy as two numbers at a stated confidence level, never one combined figure. And require the raw observations, not only the processed coordinates, so a future contractor can reprocess into whatever frame is current then.

None of this is exotic. It is the ordinary content of a spatial reference declaration, and it costs nothing at tender time. It is only expensive when it is missing and someone has to reconstruct it years later from a filename.

We do this on the operational side too. A survey flown without connectivity still needs its base tied to real control, which is part of why inspecting an airfield with no internet is a workflow question rather than a hardware one. And when the output is an obstacle clearance assessment, where a penetration is decided by height above a defined surface, the frame is not paperwork. It is the answer.

Two surveys that disagree are usually not a dispute about the runway. They are a dispute about which planet-sized reference surface someone chose, recorded years apart by people who each assumed it was obvious. Write it down and the disagreement disappears.