Ground Control Is Boring and It Is the Whole Survey

Every drone survey discussion I sit in starts with the aircraft and the camera. Almost none of them start with where the ground control points are going, which is the decision that actually determines whether the deliverable is accurate.

Distribution beats quantity

Five well-distributed points beat twelve clustered near the access road. You want them at the edges of the block and at the extremes of elevation, because that is where an under-constrained bundle adjustment drifts.

Keep checkpoints out of the solve

Hold some points back and never feed them to the processing. They are the only honest measure of accuracy you will get. A report that quotes RMS error on the same points used to constrain the model is quoting how well the software fitted itself.

Plan for the second visit you do not want

• Photograph every target in place, with the site visible behind it.
• Record the observation method and the epoch, not just the coordinates.
• Fly a crossing pattern where you can — it costs ten minutes and fixes a lot of doming.

When processing is handled by a separate team, as it often is with partners such as UAV Data Processing, this documentation is the difference between a clean turnaround and a fortnight of email.