Change Detection Between Two Surveys

Two surfaces of the same area are sampled on a common grid, and the later elevation is subtracted from the earlier one. Positive values mean material was added, negative means it was removed.

The output is a signed difference raster: a map where the value of every cell is a height change in metres. From it you can derive volumes of gain and loss, and outline the areas that changed.

That is the whole arithmetic. Everything difficult about change detection is in the preparation and the interpretation.

The two surveys must sit in the same coordinate system, horizontally and vertically, or you are measuring your own registration error.

The classic failure is a vertical reference mismatch — one survey processed to ellipsoidal heights, the other levelled to benchmarks. Both look fine on their own; their difference is contaminated by a large, uniform bias.

The diagnostic is easy: look at ground you know did not change — an access road, a bench, a hardstanding. If those stable areas show a consistent non-zero difference, you have an alignment problem, not a change. Fix it before reading anything else.

Each survey carries its own uncertainty: control quality, GSD, surface roughness, vegetation, processing noise. When you subtract two uncertain surfaces, the uncertainties combine — the difference is less certain than either input.

So there is a threshold below which a difference cannot be distinguished from noise. It is called the Level of Detection, and it is the single most important number in a change report.

Below the threshold, a cell showing −4 cm and a cell showing +4 cm are telling you the same thing: nothing measurable happened here. Colouring them red and blue makes them look like erosion and deposition, and that is how difference maps mislead people.

The practical approach is to mask out everything below the threshold and report only what survives. STREAM extracts change areas from the signed difference using exactly this rule — only differences whose magnitude exceeds the level of detection are turned into polygons.

This does two things. It stops small noise from being read as real movement, and it stops noise from inflating volumes: a large area of ±3 cm scatter can add up to an impressive-looking but entirely fictional quantity.

Choose the threshold from the quality of your worst survey, not your best. And write it on the map — a difference map without its detection threshold is not a finished deliverable.

These answer different questions. Area tells you where work happened — useful for progress tracking and for spotting activity outside a permitted boundary. Volume tells you how much material moved — the number that gets paid.

Report them together. A large area with a small volume is a thin surface reshaping; a small area with a large volume is a deep cut or a tall stack. The pair describes the site far better than either alone.

Mine and quarry progress: monthly extraction volumes, waste dump growth, and whether the pit is following the plan.

Stockpile monitoring: how much came in and went out between two dates, without measuring each pile from scratch.

Earthworks certification: independent verification of claimed quantities against surveyed reality.

Erosion, landslides and landfill: slow movement that is invisible between two site visits but obvious between two surfaces.