PR043 Africa - Djibouti

[GM4AJK]

If only there were some scientific discipline one might apply to determine the likelihood of a second researcher choosing exactly the same coordinates in a manner not driven by any particular methodology other than convenience.

Given that AI is now solving problems in that old arcane discipline, I decided to ask Claude about the odds of an independent researcher selecting the exact same position to geolocate the compound. Claude said:-

Short answer: essentially zero if two people were pointing at random, but not vanishingly small if both were deliberately trying to identify the same specific feature (e.g. that shed/building) — those are two very different questions, and it matters which one you're asking.

The coordinate precision itself
The label shows 11.388432, 43.139749 — six decimal places. At this latitude, that's a grid cell roughly 11 cm × 11 cm on the ground. That precision comes from Google's coordinate encoding, not from anything a human actually controlled by hand — nobody is clicking with sub-meter accuracy on a touchscreen or mouse.

What a human click actually resolves to
What matters more is the pixel the person actually clicked, at the zoom level shown. Judging by the size of the buildings in this crop (structures that look like ~10–30 m sheds spanning ~100–150 px), the image is covering roughly 150–200 m across ~1080 px, i.e. each pixel ≈ 0.15–0.2 m on the ground. That gives on the order of 500,000–900,000 distinct clickable positions within just this cropped view.

- If a second person clicked at random anywhere in this scene, the odds of landing on that exact pixel are on the order of 1 in 500,000 to 1,000,000 — negligible.
- If a second person clicked at random anywhere reasonable on a full Google Maps session (not narrowed to this crop first), the odds round to zero for practical purposes.

 

[FatPhil]

"is simply a convenient ground anchor within that matched scene."

*A* convenient ground ancher. But why *that one* - given that there are so many to chose from?

Do they realise that one-in-a-million (likely less, as not all points are created equal, and you selected a non-obvious one) occurances can win nobel prizes - that's a 5 sigma result!

 

[Mick West]

*A* convenient ground ancher. But why *that one* - given that there are so many to chose from?

In their paper, they include one point labeled "P01 anchor", but it is not the Flarkey Point, but a different point about 400 m to the NE.

The FP is approximately behind the black dot in the red marker (not where the marker indicates, that's the bottom point). The FP is not marked.

 

[Mick West]

A response from Dr Nandal:

I understand why the identical six-decimal coordinate raises the question. However, those six decimal places should not be interpreted as an 11 cm geolocation accuracy. They are simply the coordinate precision returned when a point is selected in Google Earth or Google Maps. Numerical precision is not the same as measurement accuracy. At this latitude, six decimal places correspond to roughly 11 cm in coordinate spacing in latitude, but that does not mean the site was localized to an 11 cm square.

This is a rather puzzling interpretation, as six decimal planes do indeed represent an 11cm geolocation within Google Earth. While GE might be a few feet off in registering the terrain with actual Lat/Lon, if you move 11cm in GE (at this location), it will tick longitude by 0.000001°, and it will do this consistently.

i.e. those coordinates uniquely specify an 11cm location in Google Earth. Saying it is "simply the coordinate precision returned when a point is selected" is meaningless, as the fact that it's the coordinate precision is precisely why there's no plausible way two people could get the same two numbers independently.