How Many Ground Control Points Do You Actually Need?

Published 2026-09-02 · UAV Imaging Inc.

Key takeaways

Placement beats count: a handful of well-distributed points outperforms a larger number clustered in the middle of the site.
GCPs and RTK/PPK solve different halves of the problem — RTK/PPK positions the camera, ground control anchors and verifies the result on the ground.
Keep at least one surveyed point out of the solution as an independent check point, otherwise you have no evidence of accuracy, only an internally consistent model.
Vertical accuracy is where thin ground control shows up first, which is why volumes and grading work justify more points than a visual orthomosaic does.

Ground control is the part of a drone survey scope that people either over-specify or skip entirely, usually without a clear reason for either. Somebody heard that five points is standard. Somebody else heard that RTK makes ground control obsolete. Both statements are close enough to true to be quoted and too vague to plan a survey with.

Here is a more useful way to decide how many ground control points a job actually needs.

What a ground control point actually does

A ground control point is a marked, visible target on the ground whose position has been surveyed independently — typically with GNSS survey equipment or a total station. It appears in multiple photos, and during processing the software is told: this pixel is at this real-world coordinate.

That does two jobs. It georeferences the model, tying it to the coordinate system and vertical datum the project actually works in. And it constrains the model, pulling out the systematic deformation that photogrammetric reconstruction is prone to when it has nothing external to check itself against.

The second job is the one people forget. A model with no ground control can be beautifully internally consistent and still be bowed, tilted or scaled slightly wrong across the site — errors that will not show up in the imagery and will show up in the numbers.

Placement matters more than count

The most common ground control mistake is not too few points. It is points in the wrong places.

Control constrains the model where the control is. A cluster of points down the middle of a site leaves the edges free to drift, and the edges are usually where the site boundary, the toe of the slope or the tie-in to existing survey lives. The pattern that works:

Perimeter first. Points near the corners and along the outer edge of the mapped area, so the extremities are anchored.
At least one in the middle. This is what catches the classic doming or bowl deformation across the interior.
Spread through the vertical range. On a site with real relief — a pit, a stockpile yard, a slope — put control at the high points and the low points, not just on the flat convenient ground near the truck.
Inside the photo block, not on its edge. A point that only appears in a few oblique frames near the boundary contributes far less than one seen well from many angles.

Rearranging the same number of points into that pattern usually improves accuracy more than adding points to a bad pattern does.

So how many?

There is no universal number, because the honest answer depends on four things: how big the site is, how much relief it has, what the deliverable is used for, and whether the flight carried RTK or PPK positioning.

How to reason about it:

Start with the geometry, not the area. The minimum useful arrangement is enough points to enclose the site plus at least one interior point. Below that you are constraining position but not shape.
Add points as the site grows or breaks up. Long linear corridors and sites split by buildings, trees or water need more, because each disconnected or elongated region needs its own anchoring.
Add points as relief increases. Flat parking lot: fewer. Open pit with 30 metres of relief: more, distributed vertically.
Add points as the tolerance tightens. A visual progress orthomosaic and a grading surface for a design do not need the same control density.
Always keep some out. See the next section.

Diminishing returns arrive quickly. Once the site is properly enclosed with good vertical spread, additional points mostly add field time, not accuracy. The survey day is often the constraint that matters most — every point is somebody walking to a location, setting a target and occupying it, and on a live industrial site that time is not free or always safe to spend.

Check points: the ones you do not use

This is the single highest-value practice in ground control and the one most often skipped.

Split the surveyed points into two sets. One set goes into the processing solution as control. The other is held out entirely — the software never sees it — and afterwards the model's value at those locations is compared against their surveyed values. Those are check points, and the differences are the only real evidence of accuracy anybody has.

Without check points, an accuracy claim rests on the processing report's internal residuals, which describe how well the model agrees with the points it was already told to fit. That is a measure of the fit, not of the truth. A model can report tight residuals on its control and still be wrong between them.

If a deliverable will be audited, disputed or used for payment, hold points out. Include the check point comparison in the report.

When RTK or PPK replaces ground control — and when it does not

RTK and PPK position the aircraft's camera to centimetre-grade at the moment of exposure, which means the photo block arrives already georeferenced without needing targets to tie it down. On many jobs that genuinely does replace the bulk of a ground control campaign, and that is a real saving in field time and site exposure.

What it does not do is remove the need for independent verification. RTK and PPK solve for where the camera was. Ground control and check points confirm where the model landed on the ground. They are different halves of the problem, and the second half is what an auditor asks about.

The practical middle ground on most RTK or PPK jobs is to drop the dense control network and keep a small number of surveyed points as check points, plus enough control to resolve the vertical.

Vertical is the specific reason to keep ground truth. GNSS heights are measured relative to an ellipsoid and have to be converted to the orthometric heights the project works in via a geoid model, and any offset in that conversion moves the entire surface up or down uniformly. It is a systematic error, it is invisible in the imagery, and a single well-surveyed ground point catches it immediately. On a stockpile inventory or a grading surface, that offset is the difference between a defensible number and a wrong one.

Match control to what the deliverable is for

Visual record or progress orthomosaic. Position needs to be roughly right, and RTK or PPK on its own is usually sufficient. A couple of check points cost almost nothing and settle arguments later.
Volumes and stockpile measurement. Vertical accuracy is the whole game, and the base surface definition matters as much as the control. Keep real ground truth in the vertical, and keep check points.
Design, grading and engineering surfaces. The tightest case. Full control plus held-out check points, in the project's coordinate system and vertical datum, with the check point results reported.
Corridors — pipeline, road, transmission. Long thin blocks are the geometry most prone to drift along their length, so control needs to be distributed down the corridor rather than concentrated at the ends.

What to ask for in the scope

If you are writing or reviewing a drone survey scope, these lines are worth having in it:

The coordinate system and vertical datum the deliverables must be in
Whether the flight will use RTK, PPK or neither
How many surveyed points, and how they will be distributed — not just the count
How many will be held out as check points
How the points themselves are surveyed, and to what accuracy
That the check point comparison will be included in the deliverable

An operator who can answer those without hesitating is telling you their accuracy claim has evidence behind it. If you want a straight answer on what your specific site needs — including the cases where less ground control is genuinely fine — get in touch.

Frequently Asked Questions

How many ground control points does a drone survey need?

There is no single number. The minimum useful arrangement encloses the mapped area and includes at least one interior point; beyond that the count rises with site size, the amount of vertical relief, how broken up the site is, and how tight the accuracy tolerance is. A flight using RTK or PPK usually needs far fewer, because the camera positions are already known. Placement drives accuracy more than count does, and returns diminish quickly once a site is properly enclosed with good vertical spread.

Where should ground control points be placed?

Around the perimeter of the mapped area first, with at least one point in the interior to catch doming or bowl deformation, and spread through the site's vertical range rather than only on convenient flat ground. Points should sit inside the photo block where they are seen well from many angles — a target near the very edge of the coverage appears in too few frames to contribute much.

Do you still need ground control points with RTK or PPK?

Usually far fewer, but not zero. RTK and PPK establish where the camera was at each exposure, which georeferences the photo block without a dense target network. They do not independently confirm where the model landed on the ground, and they do not catch a vertical offset introduced by the ellipsoid-to-orthometric height conversion, which shifts the whole surface uniformly and is invisible in the imagery. Keeping a small number of surveyed points as check points and vertical ground truth is the practical middle ground.

What is the difference between a ground control point and a check point?

Both are surveyed targets. A control point is given to the processing software so the model is fitted to it. A check point is deliberately held out — the software never sees it — and the finished model is then compared against its surveyed value. Only check points provide independent evidence of accuracy; the residuals on control points describe how well the model fits data it was already told to match.

Why does vertical accuracy need more ground control than horizontal?

Vertical is where thin control shows up first. GNSS heights are measured against an ellipsoid and must be converted to the orthometric heights a project works in through a geoid model, and any offset in that conversion raises or lowers the entire surface uniformly. That systematic shift does not appear anywhere in the imagery, but it changes every volume and every design elevation derived from the surface. One well-surveyed ground point catches it.

Book a commercial drone flightMapping, inspection, multispectral and thermal across Alberta.Request a quote →or call 587-532-9000