Ground control and datum: why point clouds land in the wrong place

Ask three questions of any capture proposal: what datum, zone, units and geoid model the delivery will be in; whether independent check points will be held out of the adjustment; and what the report will say about voids. The accuracy figure on the cover page answers none of them.

Two different kinds of accurate

Relative accuracy is how well the points agree with each other. Fly a parking lot, and if the surface comes out flat and the striping lines up between flight lines, relative accuracy is good. That is mostly a function of the sensor, the inertial unit and the processing.

Absolute accuracy is whether that internally consistent cloud sits in the right place on the earth, in the coordinate system your project uses. That is a function of control.

A dataset can be beautifully self consistent and two feet low. Every measurement inside it is correct. Everything about where it sits is wrong. Tie your design to it and the error propagates into everything downstream, and nothing in the data will ever reveal it to you.

What ground control is

Ground control points are physical targets placed on the site, whose coordinates are then observed with survey grade equipment. In processing, the collected data is adjusted to fit those known positions.

Check points are the half that gets skipped. A check point is a control point deliberately held out of the adjustment. After processing, the dataset is compared against it, and because the solution never saw it, the difference is a real measure of error rather than a restatement of the fit.

A processing report showing only how well the data fit the points it was adjusted to is telling you almost nothing. Ask for check point residuals every time.

Datum, in plain language

A datum is the agreed reference frame that turns a measurement into a coordinate. There are two of them and they are independent of each other.

Horizontal

Where things are in plan. In Florida that is usually a State Plane zone, North or East, on NAD83, in US survey feet or international feet. Those last two differ by two parts per million, which sounds like nothing until you carry it down a long corridor.

Vertical

How high things are. In practice NAVD88, though older projects may sit on NGVD29, and the offset between them on this coast is not small. Meanwhile GNSS natively measures ellipsoid height, which is not an elevation anyone designs to. Converting ellipsoid height to orthometric height requires a geoid model, and using a different geoid model than your project used produces a consistent, invisible vertical offset across the whole dataset.

That is the most common way good data ends up unusable. Nobody made a mistake. Two parties assumed different frames and never said so.

How much control a site needs

There is no fixed answer and anyone who gives you one without seeing the site is guessing. It depends on the size and shape of the site, since a long corridor needs control distributed down its length rather than clustered at one end. It depends on tolerance, because design level topography on flat ground needs more than a stockpile volume does. It depends on terrain and cover, since targets have to be visible from the air and canopy limits where they can go. And it depends on whether your crew has already set and observed control, which is often enough on its own and has the side benefit of keeping the data in your frame by construction.

What a control plan should tell you before the flight

• How many control points, roughly where, and how they will be observed
• How many check points will be held out of the solution
• Horizontal and vertical datum, zone, units, and geoid model
• What residuals are acceptable, and what happens if they are not
• Whether existing project control will be held

What should arrive with the data

A short processing report, and it is not a formality. It should state the datum and geoid used, list control and check points with residuals, describe the classification approach, and note anything unusual: voids over water, sparse ground returns under the heaviest cover, an area that could not be flown.

Voids in particular should be disclosed rather than quietly interpolated across. A surface that smoothly bridges a gap where there was no data is not a better deliverable. It is the same deliverable with the uncertainty hidden, which is worse, because now you cannot see it either.