Drone survey accuracy: GCPs, RTK and best practice

Relative and absolute accuracy

Relative accuracy concerns measurements within the model: a distance, an area, a volume. Absolute accuracy concerns the position of the model in the coordinate system: is it in the right place and at the right elevation?

A model can be internally consistent and still be offset. Without ground control points or an RTK drone, Pix4D puts the absolute error at a few metres, the accuracy of the drone’s standard GNSS. To overlay surveys, integrate a design or prepare setting-out, absolute accuracy is essential.

The starting point: ground sampling distance

The GSD is the distance on the ground covered by one pixel. It depends on the camera and the flight height: flying lower reduces the GSD but lengthens the survey. For the DJI Mavic 3 Enterprise, DJI gives GSD ≈ H/36.5 (in cm, H in metres), about 2.7 cm per pixel at 100 m.

For a correctly reconstructed and georeferenced project, Pix4D indicates an accuracy of about 1 to 2 times the GSD horizontally and 1 to 3 times the GSD vertically. These are orders of magnitude on well-textured surfaces: only field checks are conclusive.

Ground control: GCPs, RTK or both

• Ground control points (GCPs): targets visible in the photos are surveyed on the ground with GNSS RTK or a total station. The software uses them to fix the model in place.
• RTK or PPK drone: the position of each photo is corrected by GNSS. DJI specifies RTK positioning of 1 cm + 1 ppm horizontally and 1.5 cm + 1 ppm vertically for the Mavic 3 Enterprise and the Matrice 4E. This is the position of the aircraft, not the final accuracy of the model, but it reduces the number of ground points needed.
• Both: the most robust approach for topographic surveys and contractual volumes. RTK secures the geometry; a few ground points secure the elevation and prove the accuracy.

As Pix4D points out, the absolute accuracy of a project cannot exceed that of its control points or image positions.

How many control points, and where?

Three points are the technical minimum to position a model. Pix4D recommends at least five, and five to ten are usually enough even on large projects; beyond that the gain is small, except on complex terrain where extra points improve the reconstruction.

• Spread the points around the area, not exactly on the edge, and place one in the centre.
• Cover the different elevation levels: pit floor, benches, plateau.
• Avoid areas that move: stockpiles, busy haul roads.
• Use high-contrast targets about 5 to 10 times the GSD, so they stay legible in the photos.
• Survey the points to an accuracy consistent with the final accuracy required.

Verifying: check points and RMSE

Accuracy can be claimed, but above all it must be proved. Check points, surveyed on the ground but excluded from the adjustment, compare the model with reality. The root mean square error (RMSE) is calculated horizontally and vertically.

RMSE = √( Σ Δ² / n )

Δ: difference between the model and a check point; n: number of points.

The international reference, the ASPRS Positional Accuracy Standards (Edition 2, 2023), expresses accuracy through RMSE alone, without the 95 % confidence level used previously. For a fully compliant assessment it requires at least 30 check points, ground control twice as accurate as the target product, and the check points’ own error to be taken into account. On a site of a few hectares, the number of check points is agreed with the client according to how the data will be used.

What degrades accuracy

• Low-texture surfaces: water, uniform sand, smooth roofs.
• Vegetation: the ground is not visible and leaves move in the wind.
• Strong shadows and changing light during the flight.
• Moving objects: machinery, trucks, stockpiles being loaded.
• Strong wind, insufficient overlap or motion blur.

What a quality report should contain

• Coordinate system and height reference.
• Average GSD, flight height and overlaps.
• Control method: RTK, control points, number and distribution.
• Residuals on control points and RMSE on check points.
• Date, time and flight conditions.

Sources

1. DJI Enterprise, Geospatial Solutions FAQ (Mavic 3E GSD)
2. DJI Enterprise, Mavic 3 Enterprise specifications
3. DJI Enterprise, Matrice 4 Series specifications
4. Pix4D, relative and absolute accuracy of drone mapping
5. Pix4D, number, distribution and accuracy of GCPs
6. ASPRS (2023), Positional Accuracy Standards for Digital Geospatial Data, Edition 2, PE&RS

Manufacturer figures and published orders of magnitude: the accuracy of a given project is demonstrated on its own check points.

Our six-step method · Prepare a quote request