3 mm of What?

Three millimeters sounds impressive.

It sounds precise. It sounds measurable. And attached to an aerial mapping system, scanner, camera, or processing workflow, it can sound like proof that one solution is better than another.

But three millimeters of what?

Resolution?

Relative accuracy?

Absolute accuracy?

Horizontal or vertical?

Measured where, across what distance, against what control, and verified how?

Those aren't technicalities.

They determine whether the number means anything at all.

The Smallest Number Doesn't Always Win

There is a natural tendency in technology to turn specifications into selling points.

Faster processing.

More points.

Higher resolution.

Smaller error.

More megapixels.

More images.

More precision.

Those numbers have value. We use them ourselves.

But a specification without context can quickly become a vanity metric.

A construction team doesn't benefit from three-millimeter accuracy simply because three millimeters sounds better than ten millimeters. It benefits from data accurate enough—and reliable enough—to answer the question the team actually needs answered.

And those questions can be very different.

What existed before excavation began?

Where was material stored at a particular point in the project?

How has drainage changed?

How much material moved?

What was visible before subsequent work covered it?

Has a surface elevation changed by several inches?

What condition existed before a dispute arose?

Each question places different demands on the data.

That is why the conversation shouldn't begin with:

“How accurate can we make it?”

It should begin with:

“What do you need to know?”

Resolution Is Not Accuracy

This is one of the easiest distinctions to lose in aerial mapping.

An image may have a very small Ground Sample Distance (GSD), meaning each pixel represents a small area on the ground.

That's useful.

It does not, by itself, tell us how accurately a feature in that image is positioned in the real world.

Likewise, a beautifully detailed point cloud can contain millions—or hundreds of millions—of points without automatically establishing that those points are correctly positioned relative to a known coordinate system.

Detail and accuracy are related to the usefulness of a dataset.

They are not the same thing.

A highly detailed model can still be poorly positioned. A less visually impressive dataset may be entirely adequate for the decision it was created to support.

The intended use matters.

Relative to What?

The word accuracy also needs a reference.

Relative accuracy describes how well features within a dataset relate to one another.

Absolute accuracy asks a different question: how closely do those features correspond to their actual positions in the real world?

That distinction matters.

If the purpose is to document general site conditions and preserve a visual record, one level of positional confidence may be appropriate.

If measurements are being compared across multiple dates, another level may be necessary.

If the question involves small elevation changes, quantities, engineering decisions, property boundaries, or other work where positional error has consequences, the requirements may change substantially.

There isn't one universally meaningful accuracy number.

There is an accuracy requirement appropriate to the intended use.

Control Matters. Verification Matters More.

Ground control can improve and constrain a photogrammetric project.

But simply saying that Ground Control Points (GCPs) were used doesn't finish the accuracy discussion either.

How were they established?

Where were they located?

How were they distributed throughout the project?

Were some points reserved as independent checkpoints rather than used to influence the model?

And what happened when the finished dataset was compared against those checkpoints?

That last question matters.

A processing report can produce impressive statistics. A model can look excellent on a screen. Neither substitutes for independently checking whether the finished product agrees with known positions that were not used to build it.

In other words:

Don't just tell us how the model was created. Show us how it was tested.

Accuracy Is Also a Design Decision

There is another side to this conversation that doesn't receive enough attention.

More accuracy isn't free.

Achieving tighter tolerances can affect flight altitude, image overlap, control requirements, field time, processing, equipment, verification, cost, and sometimes the practicality of the mission itself.

That doesn't mean accuracy should be compromised.

It means accuracy should be purposeful.

If a project requires tight tolerances, design the mission to support them and verify the result accordingly.

If the purpose is broader site documentation, don't burden the project with requirements that add cost and complexity without improving the decision the client needs to make.

The objective isn't to produce the smallest number possible.

The objective is to produce data that can be trusted for its intended purpose.

Start With the Question

This is why we believe aerial data collection should begin before the aircraft ever leaves the ground.

What are we trying to learn?

What needs to be measured?

What needs to be preserved?

What might someone need to verify six months from now?

What decision will this information support?

Once those questions are understood, the technical decisions begin to make sense.

Flight altitude.

Image resolution.

Overlap.

Ground control.

Independent checkpoints.

Processing method.

Deliverables.

An orthomosaic, Digital Surface Model (DSM), point cloud, or 3D model isn't valuable merely because it exists.

Its value comes from what someone can reliably do with it.

So, 3 mm of What?

Precision matters.

Accuracy matters.

Resolution matters.

But none of them should exist as isolated numbers in a sales pitch.

If someone tells you a system, sensor, or mapping process delivers “3 mm,” there is a perfectly reasonable question to ask next:

3 mm of what?

Then keep asking.

Relative or absolute?

Horizontal or vertical?

Under what conditions?

Across what area?

Established against what control?

Verified with what independent checkpoints?

And, finally, the question that should probably have been asked first:

What problem are we trying to solve?

Because the best aerial dataset isn't necessarily the one with the smallest number attached to it.

It's the one designed with enough accuracy, enough verification, and the right deliverables to answer the question that mattered in the first place.

THE FLYING LIZARD®

Aviation-Driven Drone Intelligence™

Where People and Data Take Flight™