Why two drone surveys of the same site never line up

You flew the same quarry twice and the software says material moved that nobody touched. Here is what is actually happening.

It is a common and unsettling result: two flights, three weeks apart, and the difference map shows earth moved in places nothing happened. The instinct is that the second survey was bad. Usually both were fine, and the comparison was wrong.

Two different accuracies, and vendors blur them

RELATIVE accuracy is how right the model is against itself — distances, areas, volumes measured inside one map. It is typically one to three times your ground sample distance, and it is genuinely good.

ABSOLUTE accuracy is where that model sits on Earth. On drone GPS alone, two to six metres horizontally and worse vertically.

Every measurement inside one map depends on the first number. Every comparison BETWEEN maps depends on the second. Vendors quote the first and let you assume it covers the second.

So what is the difference map showing?

Two models, each internally fine, each floating somewhere within a few metres of the truth — and independently of each other. Subtract them and you are measuring the gap between two guesses about position. On sloped ground a horizontal shift of one metre reads as a large vertical change, which is exactly why quarry benches and embankments look like they moved.

The fix is boring and it works

Put ground control on the site. Survey it once. Use the same points on every flight. Both models are then tied to the same physical reality and the difference between them is the material that actually moved.

Semi-permanent marks — painted crosses on concrete, plates set in the ground — cost one afternoon and pay for themselves the first time somebody disputes a monthly volume. RTK or PPK on the aircraft solves the same problem without markers if your workflow supports it.

What good software should do here

It should know whether two flights share a reference, and refuse to difference them when they do not. Silently subtracting two ungeoreferenced surfaces and drawing a red-and-blue map is the single most misleading thing a mapping product can do, and most will do it without a word.

DroneGrid records whether each flight had ground control, states relative and absolute accuracy as separate numbers, and will not present a comparison it cannot stand behind. If your data cannot support the answer, it says so and tells you what was missing.