When Aerial Imagery Holds Up as Evidence, and When It Does Not

Most aerial captures are bought to answer a question this month. A volume for the year end, a progress record for the draw, a condition look at a roof before the shingles are ordered. Then eighteen months later something goes wrong, and the capture turns into the only existing record of what the site looked like on a date that now matters to somebody else: an adjuster, a lawyer, an auditor, a regulator, or the party on the other side of a change order.

At that point the question stops being about accuracy and starts being about credibility. Can anybody establish that this image is of this site, on that date, unaltered, captured by a method somebody can describe? That is a records question rather than a flying question, and it is decided almost entirely by choices made before and immediately after the flight rather than by anything done at the time of the dispute.

The imagery is only half of it

A picture of a site proves very little on its own. A screenshot of an orthomosaic in a slide deck is a picture of a site. What makes a capture carry weight is the package around it:

• The original data. The raw frames as they came off the aircraft, unmodified.
• The metadata. The embedded capture timestamp, camera and lens parameters and the positional record written by the aircraft at the moment of exposure.
• The method record. Who flew, what aircraft and sensor, the flight parameters, the ground control and the coordinate system and vertical datum the output was produced on.
• The processing record. Which software, which version, which settings, and the accuracy report the processing produced.
• The custody record. Where the data has lived, who has held it and what has been done to it since.

None of that is exotic. All of it is routine output of a normal commercial capture. The failure is almost never that the record did not exist. It is that nobody kept it, because at the time the only thing anybody wanted was a number and a nice image.

Keep the originals, not just the deliverable

The deliverable is a derived product. An orthomosaic, a surface and a point cloud are all outputs of a processing run that made choices along the way, and a bare-earth surface in particular is the result of a filtering decision rather than a direct measurement. That is fine for using the data. It is a problem if the output is ever challenged, because a derived product cannot be checked without the inputs it was derived from.

What to retain, in order of how often it turns out to matter:

• The raw images, as captured, with metadata intact.
• The flight log from the aircraft.
• The base station observation file and the base coordinate, where the work used RTK or PPK positioning.
• The ground control coordinates with the named datum and geoid they were held on, and the independent check point results.
• The processing report, including the residuals and the stated accuracy.
• The delivered products themselves, in open formats.

That is the same list a buyer needs in order to keep a measurement series comparable across a change of vendor, which is covered in switching drone contractors. Whether you or the contractor holds it, and for how long, is a term in the agreement rather than an assumption, and it belongs in the conversation about data ownership and retention before the flight rather than after.

The quiet way a good capture becomes worthless

Metadata loss is the single most common self-inflicted wound, and it almost always happens through normal, well-intentioned office handling:

• Images opened in a photo editor and re-saved, which can rewrite or discard the original tags.
• Frames pasted into a document or a slide deck, which keeps the pixels and discards everything else.
• Images sent through a messaging app or a consumer photo service, which commonly recompresses and strips.
• Files renamed or reorganised in a way that breaks the link to the flight log they belong to.
• A cropped, brightened, stitched or straightened version saved over the original.

The rule that prevents all of it is simple: the original set is read-only and is never the working copy. Do presentation work, markup, cropping and colour adjustment on copies, and keep the copies clearly labelled as copies. A capture whose original frames still carry their timestamps and positions can always be re-derived. One whose originals were overwritten cannot be repaired at any price.

Write the method down while it is still true

A method record produced at the time of the flight and a method record reconstructed a year later are not the same document, even if they contain the same sentences. The first is a record. The second is somebody's recollection, and it is open to the obvious question of how they remember.

The record does not need to be long. It needs to state the site and the area captured, the date and the local start and end times, the operator and the certificate the flight was conducted under, the aircraft and sensor, the flight altitude, overlap and resulting ground sample distance, the positioning method and ground control used, the weather and light conditions at the time, and the coordinate system and vertical datum of the output. Anything unusual about the day belongs in it too, because conditions that get written down read as transparency and conditions that get discovered later read as something else. A capture flown in smoke, at a low winter sun angle, in high wind or over partial snow cover is still a legitimate dataset, but it is a dataset with properties, and those properties belong in the file. The same logic applies to conditions that limit the sensor, as with thermal work flown in the wrong conditions.

If the capture is likely to be read by somebody outside the organisation, state the limits in the record as well. What was not captured, what the surface represents, and what the accuracy figure does and does not cover. The distinction between absolute and relative accuracy is exactly the kind of thing an opposing reader will test.

Chain of custody, in plain terms

Chain of custody is the ability to say where the file has been without relying on anybody's memory. For commercial aerial work the practical version is modest:

• One authoritative location. The original set lives in one place in the organisation, not on a crew member's laptop and not in an email thread.
• Write-once handling of originals. Access to read is broad. Access to overwrite is narrow or absent.
• A dated transfer record. Who received a copy, when, and in what form. A simple log is sufficient.
• A named custodian. One person responsible for the data still being there and still being intact in five years.
• Verifiable integrity. A file checksum recorded when the data is received makes it possible to show later that the file is bit for bit the one delivered. This is a minute of work on receipt and is not reproducible after the fact.

A municipality or a facility owner who already has a records management policy should put aerial captures inside it rather than invent a parallel system. The data is a record, and it should be handled like the other records of the same sensitivity. Where the imagery covers land the organisation does not own, access and release decisions have a privacy dimension as well, which is the subject of neighbour consent and notification.

Annotation, markup and the derived image problem

Most imagery that reaches a decision maker has been annotated: an arrow at the defect, a polygon around the area in question, a measurement callout. That is useful and it is not a problem, provided the annotation is a layer rather than a modification. Keep the markup separate from the image, so the unmarked original remains available, and label the annotated version as an interpretation rather than as the capture.

Two practices worth adopting. Caption every annotated image with the source, the capture date and the fact that markup has been added. And keep the person who did the interpreting identifiable, because a finding that nobody will put a name to is weaker than the image it came from. What separates a usable inspection deliverable from one built to look thorough is covered in how to tell a good drone inspection report from a bad one.

What a capture can and cannot settle

Being honest about the boundary is what makes the rest of it credible. A dated aerial capture is strong evidence of the visible state of a surface on a date. It can show extent, position, surface condition at the resolution flown, stored material, drainage, access, whether something was present and whether something had been built.

It does not establish a property boundary, which remains the work of an Alberta Land Surveyor, and the division of responsibility is set out in who signs off drone survey data. It says nothing about structural capacity, subsurface conditions, pipe condition or what is below water. It records a state rather than a cause, so it shows that a slope moved and not why. And it covers one date, which is why a single capture is far weaker than a series. The broader list of limits is in what a drone survey cannot tell you.

The strongest evidentiary position is also the cheapest one to be in: a baseline capture flown before anything is in dispute. An owner with last spring's capture can show what changed. An owner without it is arguing from memory, and that is a materially worse place to argue from.

The short version

Four decisions, all made before or immediately after the flight, do most of the work. Agree in writing who holds the raw data and for how long. Keep the originals unmodified and separate from every working copy. Get the method record at the time of capture rather than on request later. And name somebody who is responsible for the data still being intact when it is finally needed.

This page describes general records practice for commercial aerial work, not legal advice. If a capture is likely to enter a claim, a regulatory file or a proceeding, the handling requirements should be confirmed with your own counsel or adjuster, ideally before the flight rather than after.