How accurate is drone photogrammetry?

"How accurate is it?" is the first question most clients ask about a drone survey, and the honest answer is that it depends on what you mean by accurate and what you intend to do with the output. A photogrammetric survey can produce an orthomosaic and a point cloud that match a site closely enough for earthworks planning, design feasibility and progress records. It is not a substitute for a setting-out survey. This post explains the difference and what drives the numbers.

What "accuracy" actually means

There are two separate ideas hiding behind the word, and most confusion comes from mixing them up.

Absolute accuracy is how closely a point in your data sits to its true position in a real coordinate system: in the UK, usually OSGB36 grid and Ordnance Datum Newlyn. If your survey says a manhole cover is at a particular easting, northing and level, absolute accuracy is the size of the error in that statement.

Relative accuracy is how well the data represents distances and differences within itself. If two buildings are 40 metres apart on site, does the model say 40 metres? If a bund has dropped half a metre since last month, does the data show half a metre? Relative accuracy can be good while absolute accuracy is poor, for example when a model is internally consistent but sitting a metre off in the wrong place.

Accuracy also splits by axis. Horizontal accuracy is generally better than vertical accuracy in photogrammetry. This is a property of the geometry: height is derived from the intersection of rays from overlapping images taken from broadly similar altitudes, so the vertical component is the weaker one. A common rule of thumb in the industry is that vertical error is roughly two to three times the horizontal error for the same dataset. Treat that as an expectation, not a specification.

Finally, an orthomosaic and a digital surface model are not the same product. The orthomosaic is a corrected plan-view image. The surface model is the height data. They have different error characteristics and a client should be told which one a figure refers to.

What governs accuracy

Ground sample distance

Ground sample distance, or GSD, is the real-world size of one pixel. It is set by the camera sensor, the lens focal length and the height you fly. Fly lower and GSD gets smaller, meaning more detail. GSD puts a floor under everything else: you cannot measure to a precision much finer than you can see. A feature smaller than a few pixels is not reliably identifiable, let alone measurable.

Flight height and overlap

Lower flights mean better GSD but more images, longer flight time and more opportunity for drift in the processing. Overlap between images, both along the flight line and between adjacent lines, gives the software the common points it needs to solve camera positions. Too little overlap produces gaps, holes and a weak solution. High overlap is standard for survey-grade work. For vertical surfaces and anything complex, oblique imagery at an angle is added, because nadir-only flights model vertical faces badly.

Ground control

Ground control points are targets on site whose coordinates have been measured independently, usually by GNSS or total station. They are what tie the model to the real world and give you absolute accuracy. Their number and distribution matter more than people expect: control spread across the site and around its edges constrains the model far better than a cluster in the middle. Some points are held back as check points so you can quantify the error rather than assert it.

Without ground control you still get a model, and relative accuracy can be fine, but absolute positioning rests entirely on the aircraft's own positioning, which is a much weaker constraint.

RTK and PPK

RTK and PPK are methods of getting a far more precise position for each photo, either corrected in real time from a base station or network, or corrected afterwards against logged observations. They dramatically improve absolute accuracy compared with a standard GNSS receiver, and they reduce the number of ground control points needed. They do not remove the case for ground control entirely. A small number of independent check points is still the only way to prove what you delivered, and vertical datum transformation remains a place where errors creep in.

Camera calibration and site conditions

Lens distortion has to be modelled. Self-calibration during processing handles this reasonably well on a well-flown job with good geometry, but a poor flight plan can bake distortion into the surface as a bowl or dome. Conditions matter too. Wind, rolling shutter effects on fast flights, deep shadow, low sun, wet reflective surfaces, dense vegetation and large expanses of uniform texture such as still water, fresh tarmac or snow all degrade the result. Vegetation is a particular limit: photogrammetry measures the top of what it can see, so a grassed or wooded area gives you a surface model, not bare ground.

What drone photogrammetry is fit for

Used properly, it is well suited to:

• Orthomosaic site plans and base mapping for feasibility and design
• Contours and elevation data over sites that would be slow to walk
• Indicative volume estimates for stockpiles and cut and fill
• Dated progress records that can be compared over time
• Existing-conditions capture before design work starts
• Context drawings and overlays for planning submissions

Our drone site mapping and aerial land surveys cover that ground: orthomosaic maps, indicative volume estimates and site plans that can be overlaid on Google Maps.

When you need something else

Be clear about the limits. Photogrammetry is the wrong tool when:

• You need setting out. Positioning kerbs, foundations, services or structures on site requires a ground survey with instrument-measured points and a surveyor's certification.
• You need levels under tree canopy or thick vegetation. That is a job for ground survey or lidar.
• You need buried services. Nothing flown can see below the surface.
• You need legally defined boundaries. That is a land registry and legal question, not a photogrammetric one.
• You need a chartered structural or condition opinion. A survey flight gives imagery and measurement, not professional judgement of that kind.
• You need the inside of a building. Interiors need a different method.

A good deliverable states the control used, the check point residuals and the datum. If a supplier quotes an accuracy figure without saying how it was verified, ask how. Accuracy that has not been checked against independent points is an estimate.

Next step

Tell us the site, the area and what the output has to support, and we will say what accuracy is realistic and whether a drone survey is the right method. See drone site mapping and aerial land survey, then get in touch for a quote.