How accurate is a drone survey? GSD, RTK and PPK, explained fast
A drone survey flown with RTK or PPK positioning and proper ground control is typically accurate to 20–50mm horizontally and 30–80mm vertically — good enough for planning drawings, earthworks volumes and most engineering work. The same drone flown on standard GPS with no ground control is accurate to a few metres. Both produce a sharp, convincing map. Only one of them is a survey.
The difference comes down to three pieces of jargon: GSD, RTK/PPK and ground control. Here’s what each means, and which numbers to ask for before you accept a quote.
GSD: the size of one pixel on the ground
Ground sample distance is how much real-world ground each pixel in the photos covers. It’s set by the camera and the flying height: a typical 20-megapixel survey drone at 80–100m captures a GSD of around 2–3cm per pixel; fly at 50m and it tightens to roughly 1.5cm.
GSD is the floor under everything else, because no amount of processing recovers detail smaller than a pixel. The working rule of thumb: best-case horizontal accuracy is about 1–2× the GSD, and vertical about 2–3× — assuming the positioning is right, which is the next part.
Standard GPS, RTK and PPK
Every survey drone tags its photos with a position. How good that position is decides whether the finished model sits where it claims to.
Standard GNSS — the receiver in a consumer drone — is accurate to roughly 1–5m. The model will look internally consistent but can sit metres from its true position and elevation.
RTK (real-time kinematic) streams corrections to the drone during the flight, from a local base station or a network service, bringing photo positions down to 2–3cm. The catch: it needs an uninterrupted correction link for the whole flight.
PPK (post-processed kinematic) logs the raw satellite data and applies the corrections after landing. Same end accuracy as RTK, but nothing is lost if the link drops mid-flight, which is why many pilots prefer it on awkward sites. In the finished data the two are indistinguishable.
Ground control points and checkpoints
GCPs are marked targets placed across the site and measured with survey-grade GNSS before the flight. The processing software uses them to pin the model to known coordinates. Checkpoints are measured the same way but withheld from processing — they exist purely to test the result. The difference between where the model says each checkpoint is and where it actually is becomes the survey’s measured accuracy, usually reported as an RMSE table.
That checkpoint report is what separates a quoted accuracy from a real one. A pilot doing survey work weekly will produce it without being asked. If the answer to “how do you verify accuracy?” is “don’t worry, it’s very accurate” — worry.
Relative vs absolute accuracy
Worth thirty seconds, because it changes what you should pay for. Relative accuracy is whether distances, areas and volumes within the model are right. Absolute accuracy is whether the model sits in the right place on the national grid. Stockpile volumes only need the first. A planning topo needs both — in the UK that means coordinates on OS National Grid (OSGB36) and heights against Ordnance Datum Newlyn.
The accuracy you actually need
| Job | Sensible spec |
|---|---|
| Planning topographical survey | ±30–50mm, OSGB36 + Ordnance Datum Newlyn, checkpoint report |
| Earthworks / stockpile volumes | 1–2% repeatable volume accuracy; relative is enough |
| Roof or building condition | Sub-centimetre GSD for readable detail; grid accuracy barely matters |
| Construction progress vs design | ±50mm; looser if it’s a visual record only |
| 3D model for design work | Set by what the model feeds — ask the architect or engineer first |
Overspeccing is the most common way to overpay. If the question is “what state is the flashing in?”, the answer is photography, not RTK — and the quote should reflect that.
Four things to ask before accepting a quote
- Stated accuracy, with how it’s verified. “±40mm against five checkpoints” is an answer; an adjective isn’t.
- Coordinate system and datum. OSGB36 and Ordnance Datum Newlyn for anything going near a planning application or an engineer.
- GSD. Tells you whether the imagery will show the detail you care about.
- Deliverable formats. DXF/DWG for CAD, orthomosaic as GeoTIFF, point cloud as LAS/LAZ — named formats your architect or engineer can actually open.
If you’re choosing between camera-based survey and laser scanning, Photogrammetry vs LiDAR covers that decision; for what these specs cost, see UK drone survey prices.
For the work itself: land and topographic drone surveys cover levels, contours and planning topos; photogrammetry and 3D mapping covers full models and point clouds. One request through our form reaches up to 4 CAA-certified, insured pilots covering your postcode — each quotes against the spec you give, so the accuracy conversation happens before the price, not after.