Field 06 · Verification
Geolocation and imagery
Establishing where and when a photograph or video was made, using landmarks, shadows, weather records and independent imagery.
01Why location and date come first
Most disputed visual material is genuine footage of the wrong place or the wrong day. Recycled imagery, mislabelled scenes and old clips presented as current account for more errors than sophisticated manipulation, which means the two questions that matter most are where this was recorded and when.
Both questions can often be answered with public sources and patient comparison. Neither can be answered by the confidence of the person who posted the material, and a caption is a claim to be tested rather than information to be repeated.
02Fixed features as evidence
Start from what cannot move: the shape of a skyline, the profile of hills, the layout of a junction, the pattern of windows on a building, road markings, kerb design, the type of street furniture, overhead cable arrangements, signage conventions and vegetation.
Language, alphabet, number-plate formats, the side of the road traffic uses, plug sockets visible indoors and the design of emergency vehicles can narrow a country quickly. From there, satellite and street-level imagery allow a candidate location to be tested feature by feature until several independent details agree.
03Matching against independent imagery
A location claim becomes strong when multiple fixed features match a reference image that was captured independently. The discipline is to look for disconfirmation: a building that should be visible and is not, a road that curves the wrong way, a tree line that does not fit.
Reference imagery has its own date. A satellite view may be two years old, so a mismatch can reflect construction or demolition rather than a false claim. Recording the capture date of every reference used is part of the evidence, not an optional detail.
04Dating a scene
Time of day can be estimated from shadow direction and length, given a known location and date, using published solar geometry. The reasoning runs in both directions: with a known date, shadows test a claimed time; with a known time, they constrain the possible dates.
Weather is a second independent check. Official observations for a station near the scene record cloud, precipitation, wind and visibility by hour, and a clear sunlit street contradicts a claim of heavy rain. Seasonal markers add further constraint: foliage, snow cover, sports fixtures, election posters, seasonal displays and construction stages all narrow a window.
05Provenance of the file itself
The earliest available copy is more informative than the most widely shared one. Reverse image search finds prior appearances; a video can be checked frame by frame against earlier uploads; archived pages show what a caption said before it was edited.
Embedded metadata may record a device, a timestamp and coordinates, but it is frequently stripped by platforms and can be edited, so it is a lead rather than proof. A finding should rest on features in the image that agree with independent records, not on a field that anyone could have altered.
06Signs of manipulation and of ordinary editing
Not every alteration is deceptive. Cropping, colour correction, compression and re-encoding are routine. What matters is whether the change affects the meaning: a removed object, an inserted element, a composite of two scenes, a slowed or reversed sequence, audio taken from elsewhere.
Detection tools produce probabilities, not verdicts, and they misfire on heavily compressed material. Their output is one input among several and should never be the sole basis of a published claim about manipulation. Synthetic generation raises the same requirement: corroborate the event through independent records rather than argue from the pixels alone.
07Video gives more to work with than a photograph
A moving camera is an advantage, because it sees more fixed features than any single frame and it shows them in relation to one another. Working through a clip means extracting frames at the moments of widest view, noting the timecode of each one, and building a rough sketch of the geometry: what stands to the left of what, which way the street runs, where the light comes from. A candidate location then has to satisfy the whole sequence rather than one favourable angle, which eliminates coincidental matches quickly.
Sound carries evidence that images do not. Language and accent, a public announcement, a siren pattern, bells, birdsong, traffic density, the note of a passing train: each narrows a place or a time of day, and together they can contradict a caption on their own. Continuity is the other check a clip permits — shadows that jump, weather that changes between shots, a timestamp overlay inconsistent with the light — and each discontinuity is a sign that a clip has been assembled from more than one recording. Log the timecode of every observation, since a reviewer needs to reach the same moment in the file.
08Documenting the chain of reasoning
A verification is only as good as its record. Note each feature matched, the reference used with its capture date, the solar or weather source consulted, the searches performed, the contradictions found and the residual uncertainty.
Written this way, the finding can be checked by someone else and stated precisely in publication: several fixed features match a specific junction, the shadow direction is consistent with a morning recording, and the date remains unconfirmed. That sentence is more useful to a reader than a confident assertion with no method behind it.
09Limits and duties
Geolocation can establish that a scene is consistent with a place and a time. It cannot establish intention, it cannot prove that nothing happened elsewhere, and it does not confirm the account attached to the footage.
It also carries a duty of care. Publishing a precise location can endanger identifiable people, and a location that is accurate at street level is not always necessary to the public interest being served. Consider what level of detail the story requires before publishing coordinates.
| Signal | What it can establish | Limit |
|---|---|---|
| Fixed built features | A specific location | Reference imagery may be outdated |
| Shadow direction | Approximate time of day | Requires known location and date |
| Weather observations | Consistency with a claimed date | Nearest station may differ locally |
| Reverse image search | Earlier appearances | Indexing is incomplete |
| Embedded metadata | A device and possible timestamp | Stripped by platforms, editable |
| Manipulation tools | A probability of alteration | Unreliable on compressed files |
| Audio cues in video | Language, setting, time of day | Track may be replaced or added |
| Continuity across shots | Whether a clip is one recording | Ordinary editing also breaks continuity |
Checks before publishing
- Ask where and when before asking what it shows.
- Match several independent fixed features.
- Record the capture date of every reference image.
- Cross-check the claimed date against weather records.
- Find the earliest available copy of the file.
- Publish the reasoning and the remaining uncertainty.
- Note the timecode of every observation taken from a video.
Questions
Does metadata prove where a photograph was taken?
No. Embedded fields are often removed by platforms and can be edited. They are a lead to be corroborated with independent records.
Can a detection tool confirm a fake?
It returns a probability and misfires on compressed material. Treat it as one input and corroborate the event through other records.
Should an exact location always be published?
Not automatically. Precise locations can endanger identifiable people, so publish only the level of detail the public interest requires.