Quick answer: drone survey accuracy is not a property of the aircraft. It is the result of a chain — how the site was flown, how the model was tied to the ground, which coordinate system it was tied to, and what the site let the camera see. Change any link and the number changes.
So we will not quote a figure for a site we have not flown, and you should be wary of anyone who does. What we can do is set out what controls the result, so you can specify what you need.
The two questions behind drone survey accuracy
Almost every accuracy dispute we see comes from two people using one word for two different things.
Relative accuracy is the internal consistency of the model — how faithfully distances, heights and volumes within the dataset match reality when measured against each other. Measuring from a shed corner to a slab edge, calculating a stockpile volume, comparing this month’s surface to last month’s: all of that relies on relative accuracy. The model does not need to know where on Earth it sits to answer those questions.
Absolute accuracy is how closely a point in the model matches its true position in a real-world coordinate system. That matters the moment your data sits alongside someone else’s — a design drawing, a title plan, a council dataset, an earlier survey.
A dataset can be beautifully consistent internally and still sit metres from where it claims to be: useful for measuring across a site, unsafe for design set-out. Most disappointment with drone mapping accuracy comes from buying the first and assuming you bought the second.
So the first question is not “how accurate is it” but “which kind of accuracy does this job need”.
What drives relative accuracy
GSD and flight height
Ground sample distance (GSD) is the real-world size of one pixel in the imagery. Fly lower and each pixel covers less ground — finer detail, more precise feature matching, better geometry. Fly higher and you cover more ground per battery, but every measurement comes from coarser evidence.
GSD sets a floor on what is visible at all: you cannot reliably measure a feature occupying a fraction of a pixel. Build the flight plan backwards from what the deliverable has to resolve.
Image overlap and flight pattern
Photogrammetry works by seeing the same point from many positions. More overlap means more observations per point and less room for the model to drift. Crossing grids, and oblique passes where vertical faces matter, give the reconstruction geometry a single straight-line grid cannot. Thin overlap is a false economy: it saves flight time and costs you downstream.
Camera, lens and processing
Sensor size, lens quality and how well the camera’s distortion is modelled feed straight into the geometry. Motion blur and harsh contrast destroy the detail matching depends on.
Processing is the other half. Tie point filtering, camera self-calibration, outlier handling, ground classification, interpolation across gaps: these are decisions, not defaults, and the same images processed two ways give two different models. Which is why the processing report matters.
What drives absolute accuracy
Ground control points and check points
Ground control points (GCPs) are marked targets on the ground, surveyed to known coordinates and heights, that anchor the model to the real world. They are the largest single lever on absolute accuracy.
Quantity matters; distribution matters more. Control clustered at one end leaves the far end unsupported, and error grows with distance from control — most visibly in height. Spread it across the site and, on sloping ground, across the elevations.
Check points are the half most often skipped. A check point is surveyed like a GCP but deliberately withheld from processing, then compared against the finished model. Control points tell you what the model was told; check points tell you whether it listened. An accuracy claim with no check points behind it describes the software’s report card, not the ground.
RTK and PPK positioning
RTK (real-time kinematic) and PPK (post-processed kinematic) systems record a precise position for the aircraft as each image is taken, corrected against a base station or network. Done properly, they cut the ground control needed and help where placing targets is difficult or unsafe.
They do not make ground control obsolete. The correction stream, the base station’s own coordinates, the antenna offsets and the datum transformation are all places an error can enter — and a bad base coordinate shifts the whole dataset uniformly and invisibly. Independent check points catch that.
Coordinate system and datum
This is where otherwise good data gets ruined, and it is a paperwork problem rather than a technical one.
In Australia the horizontal datum is GDA2020, usually delivered projected as MGA2020 with the relevant zone. Heights are typically on AHD (Australian Height Datum), which is not the ellipsoidal height a GNSS receiver produces natively — converting needs a geoid model, and skipping it puts a systematic height shift through the dataset. Older data may still be on GDA94, and mixing the two without transforming gives an offset large enough to matter and small enough to go unnoticed.
If your drawing set runs on a local site grid instead, say so at briefing.
Site conditions
The site sets limits no operator can argue with.
- Vegetation — the camera cannot see through canopy. Ground under dense cover is interpolated, not measured.
- Water — moving, reflective surfaces do not reconstruct. Dams, creeks and wet pavement come out as noise.
- Low-contrast surfaces — fresh concrete, clean sand, uniform new roofing. No texture means nothing to match.
- Moving objects — traffic, plant, stock, people and shifting shadows sit in different places in each image.
- Wind — pushes the aircraft off its line and off-level, degrading overlap and sharpness.
- Lighting — harsh shadow, low sun angle and cloud changing mid-flight all hurt.
A competent operator raises these before the flight, not after.
How to specify the accuracy you need
Four lines in the brief prevent almost every accuracy dispute.
- State the purpose. Concept design, earthworks volumes, design set-out, comparison against last quarter’s capture. It determines everything else, and only you hold it.
- State the coordinate system and datum. GDA2020 and the MGA zone, AHD for heights, or your local site grid — plus file formats and the software the data has to open in.
- State whether you need absolute or relative accuracy. If it only has to be internally consistent, say so and save money. If it must align with a design model or another dataset, that changes the ground control plan.
- State what verification you want reported. Independent check points, reported per point rather than averaged into one flattering figure.
Then expect a processing and accuracy report with the deliverable: flight parameters and GSD, control and check point layout, coordinate system and datum, processing settings, residuals at each check point, and a plain statement of which areas are measured and which are interpolated. A bare accuracy claim with no report behind it is a marketing sentence; a report lets your engineer judge whether the data is fit for purpose.
Where photogrammetry is the wrong tool
Legal boundaries and certified levels. Title boundaries, easements, or any dimension that must be certified for a statutory approval require a licensed land surveyor. We are a CASA-licensed aerial data capture operator, not licensed land surveyors, and no aerial dataset changes that. Where boundaries are involved this is general information, not legal advice.
Dense canopy. For bare earth under thick continuous cover, LiDAR is the right instrument — a laser pulse finds gaps in foliage a photograph cannot.
A handful of high-precision points. If the job is a few levels that must be as exact as possible, a total station or GNSS rover beats a photogrammetric model. Photogrammetry’s advantage is density across an area, not the precision of any single point.
Under a roof, under water or underground. Interiors, bathymetry and buried services are outside what a camera overhead measures.
Aerial capture earns its place on dense surface data across an area, repeat capture to measure change over time, volumes and existing-conditions surfaces for design — often alongside a licensed surveyor who sets the control and the boundary.
Frequently asked questions
How accurate is a drone survey?
It depends on the flight plan, ground control, datum and site, so a credible operator will not quote a figure before seeing it. The controlling factors are GSD and flight height, image overlap, the number and spread of ground control points, RTK or PPK positioning, and how much ground the camera could see.
What is the difference between relative and absolute accuracy?
Relative accuracy is how internally consistent the model is — whether a measurement between two points on site is correct. Absolute accuracy is how closely the model’s coordinates match true positions in a real-world coordinate system. A model can be internally excellent and still sit metres from where it claims to be. Design work needs absolute accuracy; measuring across a site often needs only relative.
What are ground control points and do I need them?
Ground control points are marked targets, surveyed to known coordinates, that tie an aerial model to a real-world coordinate system. They are the main driver of absolute accuracy. If your data must align with a design, a title plan or another dataset, you need them, spread across the site rather than clustered. If you only need internal consistency, you may not.
What is GSD, or ground sample distance?
Ground sample distance is the real-world size of one pixel in the imagery, set mainly by flight height and camera. It defines the smallest detail a survey can resolve and puts a practical floor under achievable accuracy. Lower flights give finer GSD and better precision, at the cost of longer capture time.
Which coordinate system should I ask for in Australia?
Usually GDA2020 in the relevant MGA2020 zone, with heights on AHD. Confirm the zone, and whether your project runs on a local site grid instead. Nominate it before the flight, not after delivery.