Glossary
GCP (Ground Control Point)
A marked point on the ground whose real-world position and elevation have been surveyed, used to tie a drone photogrammetry model to actual coordinates rather than to the camera’s own estimate of where it was.
Definition
A photogrammetry model is built by matching features across hundreds of overlapping photos and solving for where the camera must have been when each one was taken. That process reconstructs the shape of the ground very well on its own, but the result floats: the model can be internally consistent while sitting at the wrong absolute position, the wrong elevation, or on a slight tilt. A ground control point pins it down. Somebody lays a high-contrast target on the ground, surveys its position and elevation to a known accuracy, and the processing software is told that this identifiable pixel corresponds to those coordinates. With enough of them spread around the edges and through the interior of the site, the model is pulled onto real-world coordinates and the drift and tilt come out of it. The companion term is a checkpoint, which is the same kind of surveyed target deliberately held out of the solution: because the model was never fitted to it, the gap between its surveyed position and where the finished model places it is an independent measurement of how accurate the model actually is.
In an Alberta Context
On an Alberta farm the live question is almost never whether ground control is good, it is whether this particular job still needs it now that RTK-equipped aircraft are common. An RTK drone records a corrected position for every photo as it flies, which removes most of the original reason for laying targets, and plenty of mapping flights are now flown without any. The useful nuance is that position from the aircraft is strongest horizontally and weakest vertically, and elevation is exactly the axis a drainage or tile plan depends on. So the practical pattern goes like this. General field mapping, crop monitoring, acreage and boundary work: RTK on its own is normally sufficient and laying targets is effort spent for no gain. Elevation work where water is going to be moved: a few surveyed points are cheap insurance. Work that has to agree with an existing benchmark, a legal survey, last year’s model, or a contractor’s grade control: control points are how two datasets are made to line up. Anything heading toward a professional survey deliverable belongs with a land surveyor in the first place.
Why It Matters
Because an accurate map and an accurate-looking map are indistinguishable by eye. A photogrammetry model with no control still renders a clean contoured surface with numbers on it, and nothing about how it looks reveals that the whole thing is tilted or sitting low. The error surfaces later, when a run is cut to a grade that does not drain, when this year’s elevation model will not line up with last year’s, or when a contractor’s GPS and the plan disagree in the field and nobody can say which one to believe. The distinction underneath all of that is between relative and absolute accuracy: a model with no control can measure distances and slopes across itself perfectly well while being wrong about where it sits on the earth, and which of those two failures matters depends entirely on what the map is for. Control points are also what make an accuracy claim checkable, because a held-out checkpoint turns a phrase like centimetre grade from a description into a measurement. On a survey that a real excavation is going to be built from, that is the thing worth asking about, and asking costs nothing.
Frequently Asked Questions
+What is a ground control point in drone mapping?
It is a marked target placed on the ground before a mapping flight, whose position and elevation have been surveyed to a known accuracy. During processing the software is told that a specific identifiable pixel in the imagery corresponds to those coordinates, which anchors the reconstructed model to real-world position instead of leaving it floating on the camera’s own estimate of where it was.
+Do I still need ground control points if the drone has RTK?
For most farm mapping, no. An RTK aircraft records a corrected position for each photo, which covers general field mapping, crop monitoring, and acreage or boundary work without any targets on the ground. The case for adding control is elevation work, because position from the aircraft is strongest horizontally and weakest vertically, and vertical is the axis a drainage or tile plan lives or dies on.
+What is the difference between a control point and a checkpoint?
Both are surveyed targets. A control point is fed into the solution so the model is fitted to it. A checkpoint is deliberately withheld, so the difference between its surveyed position and where the finished model puts it is an independent measure of accuracy. An accuracy figure quoted from checkpoints is a stronger claim than one quoted from the control points the model was already fitted to.
+How many ground control points does a field survey need?
There is no single number, because it depends on the size and shape of the area, the terrain, and the accuracy the deliverable has to hit. The principle is more useful than a count: the points need to be spread around the perimeter and through the interior rather than clustered, because control at the edges is what keeps tilt out of the middle, and some should be held back as checkpoints so the result can be verified rather than assumed.
+Why does elevation accuracy matter more than horizontal on a drainage job?
Because drainage is decided by small vertical differences over long horizontal distances. Being off by a metre horizontally rarely changes where water goes; being off vertically, or being tilted across the field, changes the direction of flow in the design. A model that is uniformly low in elevation but internally consistent will still produce a plan that looks correct on screen and drains the wrong way in the ground.
+Can a map be accurate relative to itself but still wrong?
Yes, and that is the failure mode ground control exists to prevent. A model with no control can measure distances, areas, and slopes across itself correctly while sitting at the wrong absolute position or elevation. That is fine for comparing one part of a field to another, and a problem the moment the map has to agree with a benchmark, a legal survey, a previous year’s model, or a contractor’s grade control.