Comparing two drone surveys: progress, and why flights don't line up

One drone survey tells you what a site looks like. Two tell you what happened, which is the question a progress claim, a monthly inventory and an earthworks dispute all come down to. The comparison is simple arithmetic between two surfaces. Everything that goes wrong with it goes wrong before the arithmetic, in whether the two surfaces are on the same ground.

What the comparison is

Each flight gives an elevation model: a grid of heights, one per pixel. Line the two grids up and subtract the earlier from the later. A positive difference means the ground is higher now (fill, placed material); a negative difference means it is lower (cut, material removed). Multiply each difference by the pixel area and sum inside a boundary, keeping positive and negative separate, and you have the volume placed, the volume removed and the net. Draw the differences as colours and you have a difference map.

Which surface to compare

Usually the surface model (DSM) of each flight, because the things that change on a working site, the stockpiles, the pads, the excavations, are what the surface model records. Compare terrain models when the question is the ground itself and the site has vegetation or parked equipment that would read as change. Never compare a surface model to a terrain model: the difference is the height of everything standing on the ground. (The three files, and which is which.)

Why two flights do not line up

Two surveys of the same site, each fine on its own, routinely sit at different heights as a whole. The reasons, in order of how often they happen:

1. Each flight positioned itself from its own GPS. The drone's receiver is good to a metre or so, and it is wrong by a different amount each day. Two such flights can differ by a few decimetres to a couple of metres, mostly vertically. The shapes are right; the offset is nearly constant across the site.
2. Different control. Flight one used ground control, flight two used RTK against a network, and the two datums differ by the geoid separation or a local site adjustment: often tens of centimetres, sometimes tens of metres.
3. Ellipsoid against orthometric heights. One flight's heights are above the ellipsoid, the other's above sea level. The difference is a constant of 20 to 40 m in most of North America. Dramatic, and easy to spot.
4. A scale difference. A flight without control can be a percent large or small, which shows up as a difference that grows across the site rather than a constant. This one cannot be levelled away; it needs control.

The cure is the same control every time: ground control points left in place, or RTK or PPK on the drone, with a few checkpoints to prove each flight landed on the same datum. (What each gives you.) Flights already flown without it are not lost, though.

Levelling two flights on unchanged ground

If some ground in both flights was not touched between them, a haul road, a hardstanding, the top of a bund, that ground should show zero change. The average difference over it is the offset between the flights, and subtracting that constant from one flight levels them. The spread of the differences over the same ground (how far they disagree after levelling) is the noise floor of the comparison, which is worth knowing before you believe a small number.

Clarity does this under Compare: it warns when two flights stand more than about half a metre apart as a whole, lets someone with surveyor access mark ground that did not change (or suggest it), levels the pair on that ground, and prints the shift applied and the spread beside every figure measured between them, so the volumes carry their own caveat. The levelled figures are change between the two flights, not heights against the world.

The deadband

Two surveys of ground nobody touched never agree exactly. Photogrammetric noise, grass that grew, a different sun, give two or three centimetres of random difference across the site. Summed over a hectare that is tens of cubic metres of imaginary cut and the same of imaginary fill. A deadband says: count nothing smaller than this as movement. Set it to about the noise seen on unchanged ground, 3 to 5 cm on a good pair of flights, and the difference map goes quiet everywhere except where something really happened.

From a map of change to a quantity

• Volume moved in an area: draw a boundary around the pad, the cell or the stockpile and read cut, fill and net between the two flights. This is the monthly progress figure.
• Stock drawn down or added: measure the same pile against the earlier flight. The difference is what left or arrived, and it does not depend on a base. (Stockpile measurement.)
• Against the design, each month: cut and fill remaining to design, with the earlier month beside it, shows the rate. (Cut and fill methods.)
• Patches, not totals: grouping the changed ground into separate areas with their own volumes turns a site total into a list of jobs: this pad, that pit, those three piles. Clarity's "find what changed" does that grouping.
• Along a line: a cross-section with both flights drawn on it shows a batter, a drain or a haul road's change along its length, which a plan view cannot.

A routine that keeps comparisons honest

1. Same control every flight: targets left in place, or RTK/PPK with checkpoints.
2. Same flight plan: same height, overlap and roughly the same time of day, so the noise is alike.
3. Fly when the site is still: no trucks in the stock, no loader on the pile.
4. Check the offset on unchanged ground before reading any number.
5. Set the deadband to the noise you saw, and write it next to the figures.
6. Keep each month's measurement with the flight it was made on, so the series can be re-read later.

How those comparisons become a monthly progress report is in construction progress monitoring with drones. Clarity puts every flight of a site on one timeline and keeps the measurements with their flight; the comparison tools are on the progress page.