Why your map is 2–6 m off, and when that matters

Relative and absolute accuracy are different things. Confusing them is how people end up disappointed by a perfectly good model.

A drone map with no ground control points is typically 2 to 6 metres away from its true position on Earth. The same map can measure the distance between two points on it to a few centimetres. Both statements are true at once, and the difference between them decides whether the map is useful to you.

Relative accuracy

Relative accuracy is how well the model measures within itself: the width of a road, the area of a pad, the height of a pile above its own base. It is governed mostly by ground sample distance — the size of one pixel on the ground — and is usually in the range of one to three times GSD.

Fly a 2 cm GSD and measurements within the model land in the 2 to 6 cm range. Stockpile volumes, cut and fill, progress comparisons and area take-offs are all relative measurements. They do not care where the model sits on the globe.

Absolute accuracy

Absolute accuracy is where the whole model sits in the world. Without ground control it comes from the drone's onboard GNSS, stamped into each photo's EXIF. Consumer-grade GNSS is good to a few metres, so the finished model is good to a few metres — as a block. The entire map is shifted together, which is why internal measurements stay sharp.

This is the crucial part: the error is a translation of the whole model, not noise sprinkled through it. A map 4 m out of position can still measure a pile to 5 cm.

When 2–6 m is fine

When it is not

The fix, and its cost

Ground control points solve absolute accuracy. You place surveyed targets, measure them properly, and the reconstruction is tied to them. That converts absolute error from metres to centimetres.

It also means someone walks the site with survey equipment before the flight, which is work you cannot skip or automate away. We do not fly and we do not place control. If your job needs GCPs, tell us and we will discuss it honestly rather than pretend processing can substitute for them.