The controller says RTK FIX. The processing report says the imagery has a ground sample distance of 1.2 centimetres. Those are encouraging numbers. Neither one, by itself, proves that every feature in the finished drone map is in the correct real-world position to within 1.2 centimetres.

That gap matters whenever an orthomosaic, surface model or point cloud will support a measurement, design decision, progress claim or change analysis. “Centimetre-level” can describe several different things: the position written into each photograph, the size of one image pixel on the ground, the fit of the photogrammetric model, or the independently tested accuracy of the delivered map. They are not interchangeable.

What RTK FIX actually tells you

Real-time kinematic positioning can improve the camera-position information recorded during flight. In a fixed solution, the system has resolved the carrier-phase ambiguities needed for high-precision positioning under its operating conditions. That is valuable input to photogrammetry, but it is still input—not a certificate for the final product.

DJI provides a useful example of how carefully an accuracy claim needs to be scoped. Its current Terra support page states that, for imagery from the Phantom 4 RTK and Zenmuse P1 in RTK FIX status and without ground control points, reconstruction accuracy can be expressed as a formula involving centimetres plus a multiple of ground sample distance. The same page immediately says actual accuracy can be affected by RTK positioning precision, lighting, camera lenses and route planning. That is a product-and-workflow-specific statement, not a universal guarantee for every RTK drone, site or processing method.

The practical reading is: RTK FIX supports a stronger starting position for the photographs. The map still has to survive checks that are independent of the data used to build it.

Why a tiny GSD is not the same as accuracy

Ground sample distance, or GSD, describes how much ground one image pixel represents under the assumed geometry. A lower GSD can reveal smaller objects and give the software more image detail. It does not tell you, on its own, whether the whole model is correctly located, scaled or shaped.

The US Geological Survey’s guidelines for calibrating uncrewed-aircraft imagery make the distinction directly: ground-control accuracy and tie-point quality contribute to geometric accuracy regardless of imagery GSD. The report warns against assigning higher map accuracy simply because the pixel size is smaller.

A simple thought experiment shows why. A detailed image can be shifted as a whole, tilted slightly, stretched at an edge or referenced to the wrong height system. The pixels remain small. Their absolute positions can still be wrong.

Control points and check points do different jobs

A ground control point is a clearly identifiable point with known coordinates that participates in orienting or optimising the photogrammetric model. A check point is withheld from that adjustment and used to test the resulting model against an independent measured position.

DJI Terra’s current documentation describes the distinction in operational terms: GCPs are used to optimise aerial triangulation, while check points compare the triangulated result with actual measurements to check absolute accuracy. The same support material identifies three important influences on the result: marking error, coordinate-measurement error, and the number and distribution of GCPs.

This independence is the heart of credible validation. A point that helped force the model into place can report how well the model fits that input, but it is not as persuasive as a separate point the model did not use. Low residuals at control points are not automatically low errors everywhere else.

Distribution matters too. Manufacturer guidance for Terra recommends spreading points across the mapping area and warns against a straight-line layout, edge placement and ambiguous targets. Its layout guidance is software-specific, but the underlying reader question is universal: do the checks cover the centre, edges, elevation range and difficult parts of the site, or only the easiest patch?

A checkpoint report must say what was measured

An accuracy statement should identify at least:

• whether it reports horizontal, vertical or three-dimensional error;
• the number and distribution of independent check points;
• how those points were surveyed and their own uncertainty;
• the coordinate reference system, datum and height type;
• the statistic used, such as RMSE, and the units;
• the flight, camera, RTK and processing conditions; and
• any excluded terrain, vegetation, water, reflective surface or edge area.

Formal standards go further. The USGS summary of the 2024 ASPRS positional-accuracy update notes that product-accuracy assessment now requires inclusion of survey checkpoint accuracy and raises the minimum number of checkpoints from 20 to 30. Those requirements belong to standards-based accuracy reporting; they should not be casually converted into a claim that every small informal mapping job is certified after 30 points. The lesson is narrower: a defensible product claim depends on independent observations, their quality and a declared method.

The datum can defeat a precise-looking number

Coordinates only make sense inside a reference system. A workflow can be internally precise and still disagree with another dataset if the horizontal datum, projection, vertical datum, geoid model, units or epoch do not match.

The US National Geodetic Survey defines a geodetic datum or reference frame as the coordinate system and reference surface used to establish known locations. Its overview distinguishes positions relative to an ellipsoid from elevations tied to a vertical datum. That US example is not a worldwide prescription, but it illustrates why “height” needs a definition. NGS separately explains that deriving an orthometric height from GNSS requires an ellipsoid height, a compatible geometric datum and an appropriate geoid model. A centimetre figure without that context can be precise but unusable.

Before comparing a drone map with design CAD, a previous survey or a public basemap, verify that all datasets use compatible coordinate and height references. Do not “fix” a constant offset by visually dragging layers until they look right; first identify whether the difference comes from reference systems, transformations, control, time or the data itself.

Five questions before accepting “centimetre accurate”

1. What does the number describe? Photo positions, GSD, model residuals or independent product accuracy?
2. What evidence is independent? Were check points kept out of the adjustment?
3. Where are the checks? Do they cover the full area and elevation range, including edges and difficult surfaces?
4. What reference system is used? Record projection, horizontal datum, vertical datum or height type, units and any transformation.
5. Is the claim fit for the decision? A visually convincing orthomosaic may be adequate for communication yet insufficient for boundary, engineering or deformation work.

What you can honestly conclude

An RTK-fixed flight can reduce one important source of uncertainty. Fine GSD can improve visible detail. Well-surveyed ground control can strengthen orientation. Independent, well-distributed check points can test the delivered product. Correct reference systems make the coordinates interpretable. Each is useful; none substitutes for all the others.

The responsible conclusion is therefore not “RTK maps are inaccurate.” It is that accuracy belongs to a defined product, method, place and test—not to a green FIX label alone. The strongest map is not the one with the smallest number on the processing screen. It is the one whose accuracy claim another competent person can understand, reproduce and challenge.