Drone survey accuracy: ground control, checkpoints and how to report it

What actually controls drone mapping accuracy

• Ground sample distance (GSD). The ground size of one pixel, set by camera, lens and flying height. You cannot reliably pick a point more precisely than the pixel that shows it, so GSD puts a floor under achievable accuracy.
• Image geometry. Forward and side overlap, oblique images over vertical relief, and consistent flying height decide how well the photogrammetric solution holds together.
• Georeferencing. Ground control points (GCPs), RTK or PPK camera positions, or a combination, tie the model to real coordinates.
• The surface itself. Photogrammetry measures what the camera sees. Over tall grass, crops or brush, the “ground” surface is the top of the vegetation; no amount of control fixes that. For bare earth under vegetation you need lidar.

Ground control points versus checkpoints

Both are surveyed targets on the ground, but they do opposite jobs:

• Ground control points are used in the processing to position and scale the model.
• Checkpoints are withheld from processing and used only afterwards to test the result.

Accuracy measured on the GCPs themselves (the residuals) only shows how well the model fits its own control. It is not an accuracy statement. Only independent checkpoints tell you how accurate the product is.

RTK, PPK and fewer targets

With RTK or PPK, each camera position is known to a few centimeters, so the block can be georeferenced with far fewer ground targets. In practice most survey-grade workflows still set a small number of GCPs to catch a vertical offset or a datum mistake, and always set checkpoints.

Two things to verify on every RTK/PPK job: that the base station coordinates are in the project’s datum and epoch, and that the camera’s lever-arm and timing offsets are applied. Either error shifts the whole dataset uniformly, and only checkpoints reveal it.

How many GCPs, and where

There is no single required count; it depends on site size and shape, terrain relief and whether the camera positions are RTK/PPK. The layout matters more than the number:

• Put control around the perimeter of the site and some in the interior; never leave the edges unconstrained.
• Avoid placing them in a straight line, which lets the model tilt about that line.
• Add control at high and low points on sites with significant relief, since vertical error grows where the terrain changes.
• For long corridors, place pairs of targets along the length at regular intervals.
• Use targets that are clearly visible at the planned GSD, surveyed to a higher accuracy than the product you are testing.

Checkpoints follow the same logic: spread across the site, in open terrain, and never reused as control.

Reporting accuracy under the ASPRS standard

The ASPRS Positional Accuracy Standards for Digital Geospatial Data, Edition 2 (Version 2, 2024) is the reference US reviewers expect, and it now includes an addendum specific to UAS mapping. The points that matter for a drone project:

• RMSE only. Accuracy is reported as RMSE: RMSE_H horizontal, RMSE_V vertical. The old 95% confidence figures are gone.
• At least 30 checkpoints for a horizontal or vertical accuracy test (up from 20 in Edition 1), rising with project area to a maximum of 120. If fewer are used, the statement must say so.
• Checkpoint error counts. The survey accuracy of the checkpoints themselves is combined into the reported product accuracy, so control and checkpoints should be surveyed to at least twice the accuracy of the product (Edition 1 asked for four and three times).
• Vegetated areas are reported separately as vegetated vertical accuracy; for photogrammetry, remember that the surface in vegetation is the vegetation.

The standard does not prescribe a fixed number or spacing of ground control points; that remains a design decision for the surveyor, proven by the checkpoint test. A product is stated as meeting, for example, “a 5 cm RMSE_H horizontal positional accuracy class”, with vertical accuracy reported the same way.

A sample accuracy statement

A defensible accuracy report names the checkpoints, the comparison and the result, for example:

Tested 0.06 ft RMSEH horizontal and 0.09 ft RMSEV non-vegetated vertical accuracy, against 30 independent checkpoints surveyed by RTK GNSS, not used in processing.

The figures above are illustrative, not a promise of what any particular flight will achieve. What matters is the form: measured values, the number and source of checkpoints, and the fact that they were independent.