Ground control points, accuracy, and how many you need.

Drone GPS is good for finding your way home, not for survey-grade deliverables. To turn a 3D photogrammetric model into something a surveyor or insurance adjuster will sign off on, you need ground control points — physical markers placed in the field and surveyed to known coordinates. Here’s how many, and how accurate, the math says they need to be.

§ 01The accuracy curve

Drag the slider to see how horizontal and vertical RMSE drop as you add GCPs. Diminishing returns past ~10 targets — and at zero, you’re relying on consumer GPS with ~1–3 m horizontal error.

§ 02When you need them

§ 03Placement matters more than count

Five well-placed GCPs at the corners + center of a survey area outperform fifteen clustered in one quadrant. The math: bundle adjustment treats each GCP as a constraint, and constraints near the perimeter prevent the model from rotating, scaling, or shearing as a whole. Constraints in the middle stop it from sagging.

The minimum credible layout for any survey-grade work:

§ 04RTK / PPK alternatives

Modern survey drones (DJI Matrice 350, RTK-enabled Mavic 3 Enterprise) include onboard GNSS receivers that record centimeter-accurate camera positions during flight. With RTK (real-time corrections from a base station) or PPK (post-processed kinematic), you can produce survey-grade deliverables with fewer or even zero GCPs.

Even with RTK/PPK, place at least 1–2 check points to validate accuracy. The math looks fine until it doesn’t — and check points are the only honest way to know.

§ 05Drone-GPS as truth — mapplot’s geo-audit

Even with carefully-placed GCPs and RTK, projects ship with subtle CRS errors that don’t surface until someone tries to overlay your deliverable on a real cadastral layer. mapplot’s built-in geo-audit compares the EXIF GPS embedded in every photograph against the GCP CSV’s declared EPSG code, flagging adjacent-zone UTM bugs and datum mismatches before they become a callback.

The drone’s own GPS — even consumer-grade — is the closest thing to an independent ground-truth check that exists in a typical mission. Using it as a sanity layer catches mistakes that GCP RMSE alone won’t.

§ NEXTRelated field-guide pages