Every drone survey company quotes an accuracy figure. Fewer can show you how they measured it. I trained as an engineer in the RAF, where “it should be fine” was never an acceptable answer, and I feel the same way about survey data: if it has not been checked, it is not known.
A well-planned drone topographic survey with ground control can achieve horizontal and vertical accuracy in the range of 15–30 mm RMSE on open, hard or short-grass surfaces. That is comparable to a traditional GNSS survey for most design purposes, and it is achieved with far denser data. But accuracy figures are easy to quote and harder to interpret. This article explains what drives accuracy, what the numbers actually mean, and how to write an accuracy requirement that you can hold a contractor to.
Accuracy, precision and resolution are different things
- Accuracy is how close a measured position is to its true position in the chosen coordinate system.
- Precision is how consistent repeated measurements are with each other. A survey can be very precise and still consistently wrong, for example if the whole model is shifted by a control error.
- Resolution is the level of detail captured: ground sample distance for imagery, point density for LiDAR. High resolution does not guarantee high accuracy.
- Absolute accuracy is position relative to the national grid and datum. Relative accuracy is the consistency of measurements within the model, such as the height difference between two nearby points.
For design work, both matter. Relative accuracy governs whether falls and gradients are right. Absolute accuracy governs whether the survey fits with other datasets, setting-out and neighbouring surveys.
Ground sample distance is not accuracy
Ground sample distance (GSD) is the ground size of one image pixel. It depends on camera sensor, lens and flying height. A 45 MP full-frame camera with a 35 mm lens flown at 90 m gives a GSD of about 11 mm. A compact enterprise drone camera at the same height gives roughly 25 mm.
GSD sets a practical limit on achievable accuracy: as a rule of thumb, a well-controlled photogrammetry survey achieves horizontal accuracy of 1–2× GSD and vertical accuracy of 1.5–3× GSD. But a 10 mm GSD survey with poor control can still be wrong by 100 mm. Treat any quote that presents GSD as the accuracy figure with caution.
RTK, PPK and why they are not enough on their own
Survey drones carry a GNSS receiver that records each image position. With real-time kinematic (RTK) positioning, corrections from a base station or a network service are applied in flight. With post-processed kinematic (PPK), raw GNSS data from the aircraft and a base station are processed together after the flight. Both can position the camera to a few centimetres.
RTK and PPK dramatically reduce the number of ground control points needed, but they do not remove the need for independent verification. Camera calibration errors, timing offsets, a wrongly entered base coordinate or a datum mismatch can all introduce systematic errors that the aircraft's own positioning cannot reveal. In our experience, RTK-only surveys without ground control can show vertical biases of 30–80 mm that go unnoticed unless checked.
Ground control points and checkpoints
Ground control points (GCPs) are marked targets on the ground whose coordinates are measured with survey-grade GNSS or total station. They are used in processing to fix the model to the national grid and datum. Checkpoints look identical but are withheld from processing. Because the model has not been adjusted to fit them, comparing the model to checkpoints is an honest test of accuracy.
- Distribute GCPs around the perimeter and through the interior of the site, including at the highest and lowest points.
- Place checkpoints independently of GCPs and across the range of surface types, not only on the easiest hard surfaces.
- As a guide, small sites need a minimum of 5 GCPs and 5 checkpoints; for larger sites, we typically use one GCP per 3–5 ha with RTK, and at least 20 checkpoints where the survey supports a design or contract.
- Coordinate control with a method at least three times more accurate than the target survey accuracy.
“An accuracy figure without checkpoints is a sales claim, not a survey result.”
RMSE: the number that matters
Root mean square error (RMSE) summarises the differences between the model and the checkpoints. For vertical accuracy, you take each checkpoint's height difference, square it, average the squares and take the square root. RMSE penalises large errors more than a simple average, and because it is calculated from squared values, errors in opposite directions do not cancel out.
| Checkpoint | Surveyed height (m) | Model height (m) | Difference (mm) |
|---|---|---|---|
| CP01 | 84.312 | 84.325 | +13 |
| CP02 | 83.907 | 83.889 | −18 |
| CP03 | 85.140 | 85.162 | +22 |
| CP04 | 86.021 | 86.012 | −9 |
| CP05 | 84.776 | 84.791 | +15 |
| CP06 | 83.455 | 83.431 | −24 |
| CP07 | 85.609 | 85.614 | +5 |
| CP08 | 84.230 | 84.246 | +16 |
In this example the mean difference is +2.5 mm, showing no significant bias, and the vertical RMSE is about 16 mm. Always ask for the mean error as well as the RMSE: a mean far from zero suggests a systematic shift, which is often correctable once found.
What does “±20 mm” really mean?
When a survey is described as accurate to “±20 mm”, it rarely means that every point is within 20 mm. It usually means the RMSE against checkpoints is 20 mm. If errors are normally distributed, about 68% of points will fall within ±20 mm and about 95% within roughly ±39 mm (1.96 × RMSE for vertical accuracy). A few points will fall outside that.
So there is a meaningful difference between a specification that says “20 mm RMSE” and one that says “all points within ±20 mm”. The second is roughly twice as demanding and may not be achievable on soft or vegetated ground with any airborne method. Neither is wrong, but you should know which one you are asking for.
| Stated RMSE | ≈68% of points within | ≈95% of points within |
|---|---|---|
| 15 mm | ±15 mm | ±29 mm |
| 20 mm | ±20 mm | ±39 mm |
| 30 mm | ±30 mm | ±59 mm |
| 50 mm | ±50 mm | ±98 mm |
Surface type changes everything
Accuracy on tarmac is not accuracy in a hay meadow. Photogrammetry measures the top of whatever it sees, and LiDAR returns from grass and crops are scattered through the vegetation. On long grass, both methods may sit tens of millimetres above true ground, and the error is a bias rather than random noise. A good contractor will report checkpoint results by surface type, and will specify where supplementary ground survey is needed.
Coordinate systems and datums
In Great Britain, most design work uses Ordnance Survey National Grid coordinates (OSGB36) and Ordnance Datum Newlyn (ODN) heights. GNSS measures in ETRS89, so a transformation is needed: OSTN15 for coordinates and OSGM15 for heights are the current standard. Northern Ireland uses Irish Grid, and some island sites use local datums.
Two other points trip up projects. First, National Grid distances differ from ground distances by a scale factor that varies across the country, typically by a few hundred parts per million; on large sites this matters for setting-out, so agree whether you want grid or ground coordinates. Second, if the site has an existing survey or local control, tie the new survey to it or explain the differences, so that datasets fit together.
How to specify accuracy in a tender
Industry guidance, such as the RICS guidance note on measured surveys of land, buildings and utilities, sets out accuracy bands and specification wording that many UK clients reference. Whether or not you adopt it formally, a clear specification should cover the following.
- Required horizontal and vertical accuracy, stated as RMSE or as a percentage of points within a tolerance, and the confidence level.
- How accuracy will be verified: number and distribution of independent checkpoints, and whether they include soft surfaces.
- Coordinate system, datum and transformation (for example OSGB36 via OSTN15, ODN via OSGM15), and whether a local grid or scale factor applies.
- Deliverables and scale: for example 1:200 or 1:500 CAD, DTM grid resolution, contour interval, orthomosaic GSD.
- Treatment of vegetated areas, water and areas where airborne methods cannot achieve the specification.
- A written survey report including control observations, checkpoint residuals, mean error and RMSE.
Typical accuracy by method
| Method | Typical vertical RMSE | Notes |
|---|---|---|
| Drone photogrammetry, RTK + GCPs | 15–30 mm | Depends on GSD and surface |
| Drone LiDAR, RTK + GCPs | 25–50 mm | Reaches ground under vegetation |
| Drone, RTK only, no GCPs | 30–80 mm | Risk of undetected bias |
| GNSS rover walked survey | 15–25 mm | Sparse points, slow on large sites |
| Total station | 3–10 mm | Best for hard detail and small areas |
| Environment Agency national LiDAR | ≈150 mm or worse | Useful for context, not design |
Drone surveys are not a replacement for a total station where millimetre accuracy is essential, such as structural monitoring or setting out foundations. They are an excellent way to capture complete, dense and verifiable ground models over large areas, quickly and without people walking hazardous ground. The key is to specify what you need, and to insist on seeing the checkpoint evidence.
Related services
