Drone Survey Accuracy vs Precision

Key takeaways

The model should not be warped, twisted, or bowed - it should reflect the real land accurately as it is.

A model can appear correct internally but still be misaligned to the real world if it's not tied to a coordinate system.

Ground control points with precise XY&Z coordinates allow the data set to use known coordinates that are very accurate

By combining ground control points and checkpoints, drone surveys achieve true accuracy and precision.

Accuracy vs precision in drone surveys - two different things, both essential

These two terms get used interchangeably in drone surveying, but they describe completely different things. Getting both right is what separates a professional survey from one that looks good but can't be trusted. Here's what each one actually means in practice.

What accuracy means

An accurate model is one that is correct relative to itself. If a road is 2.46 metres wide in the real world, measuring it in the model should return something very close - 2.45, 2.46, 2.47. Not 2.9. "The model should not be warped at the ends, dished, twisted, bowed," as Bob puts it. The land in the model should look like the land in the real world, and measurements taken from it should hold up.

This internal accuracy comes from how the drone captures data and how the photogrammetry or LiDAR processing software reconstructs the 3D model. A well-flown mission with good overlap and consistent lighting produces an accurate model. A poor one produces subtle distortions that may not be immediately obvious but will cause problems downstream.

What precision means - and why it's a separate problem

A model can be perfectly accurate internally and still be positioned in completely the wrong place. In Ireland, all surveying is referenced to two systems: ITM (Irish Transverse Mercator) for horizontal position, and the Malin Head Ordnance Datum for vertical. Without tying your model to these real-world coordinates, "the model could be off by three or four metres in this direction or that direction for all we know."

Precision means the model aligns to the real world. If you projected it over a satellite image, the roads should sit on the roads, the gates on the gates, the buildings on the buildings - with no rotation, no horizontal offset, no vertical drift.

How ground control points fix it

The standard solution is ground control points - physical markers placed on the ground before the flight, usually spray-painted crosses or small checkerboard targets. Each one is measured with a high-precision GNSS unit, which can achieve sub-centimetre accuracy on its XY and Z coordinates. Those known coordinates are then visible in the drone imagery or LiDAR data.

When the data goes into processing software, the GCPs give it anchor points it can use to de-warp, untwist, and reposition the entire dataset in 3D until it aligns with the real-world coordinate system. The result is a model that is both accurate to itself and correctly positioned on earth.

Checkpoints: verifying the result

Ground control points align the data. Checkpoints verify it. A checkpoint is another measured ground point that is kept out of the alignment process entirely - it's used afterwards to test whether the model ended up where it should be. The difference between a checkpoint's known coordinates and where the model places it gives a confidence check on the overall quality. "That allows you to verify that you've done a good job," as Bob explains, "and that the level of difference between your checkpoints and your model gives you a confidence check on exactly how good your model is." Together, GCPs and checkpoints are what make drone survey data genuinely trustworthy.

Full transcript

Show transcript

[00:00:02] I wanna talk about two other concepts which are around the subject of accuracy, one is accuracy versus precision. Accurate means that the land or the model I generate relative to itself is correct. It's not warped at the ends. It's not dished, twisted, bowed and all that. So the land looks like the land. Even in our model it is correct to it. If I perform a measurement on a piece of road the measurement is correct if I, if I know the width of a road at this point here is 2.46 centimetres, when I measure it in my model it should say 2.45 or 2.46 or 2.47 roughly speaking, but it should certainly not say 2.9 or something like that. So it is an accurate model.

[00:00:45] Now, precise. What do I mean by that? Out in the world, we have maps. An entire mapping system in Ireland we use what's called ITM and Malin. Irish Transverse Mercader is the horizontal orientation and projection of the land in Ireland and Malin, The Malin Ordnance Datum is the vertical projection system we use in Ireland. It's the data we use to measure all things. There happens to be in Malin Head a little notch on a rock, that was put there a century or two ago. And that basically is what all things in Ireland are measured from. That's our. Yeah, It's amazing. So our model could be very accurate, but it actually could be off by three or four metres in this direction, in that direction for all we know.

[00:01:45] So we always want to make sure that when we're presenting our data, not only is it accurate, that if I perform a measurement on the data it is the same as a measurement on the, in the real world, but also that my model aligns to the real world. If I was to suddenly project my model over the real world, would the road line up exactly with the road? Would the gate be exactly where the gate is? Is it at a funny angle, is it displaced horizontally? Is it displaced vertically? All sorts of things like that. It seems like it would be so difficult with the drone. It kind of is. So we do have a very, very simple way of fixing it. And we use what are called ground control points.

[00:02:30] Okay, ground control points are what they look like from our perspective is either some sort of X on the ground in spray paint or it looks like a small checkerboard pattern that we place on the ground. What we do with these is, all these are is a highly visible way of putting an XY&Z coordinate onto our mapped land. Once we place those there, we measure them with the GNSS unit, a high precision GNSS unit. This measurement can easily be below a centimetre of accuracy for this XY&Z coordinate. By measuring it with this particular unit, what we have within our data set then is known coordinates that are very accurate. We can see this in the photography or in the LiDAR, in the data we can say where that little board is, the exact centre of it is 123456789, yeah. So then once we take that into our software systems, we can match up, you see that point there .3 that's these coordinates and the software goes okay. Like that, and it does that, and so it can dewarp it, it can untwist it and it can manoeuvre the entire data set in 3D until it aligns with these known coordinates.

[00:03:50] And so we then have both an accurate data set and a precise data set. It is absolutely accurate to the real world and it is not notionally accurate. It doesn't look good. It also is correct to the outside world. We also can put what are known as checkpoints, so we can use ground control to actually manoeuvre the entire data set into the correct position orientation what not, and then we can also have what are called checkpoints and these checkpoints are purely there to see once we manoeuvre it in position, does it line up correctly? And once you do that, that allows you then to verify that you've done a good job and that the level of difference between your checkpoints and your model give you a confidence check on exactly how good your model is and how accurate and precise it is.

Drone survey FAQs

What’s the difference between accuracy and precision in a drone survey?

Accurate means the model I generate, relative to itself, is correct. It's not warped, dished, twisted, or bowed - the land looks like the land. Precision means that same model aligns with the real world. By using ground control points and checkpoints, we achieve both an accurate and a precise data set.