Using the Sun and the Shadows for Geolocation
A Bellingcat how-to (3 December 2020) showing the craft half of open-source verification with its reasoning exposed: the part glan-bellingcat-methodology explicitly assumes and does not teach.
The physical basis is that the sun’s position for a given date, time and place is computable, so shadows in a photograph are a constraint. SunCalc is the tool: it “lets users analyse the position of shadows and the sun at any given time and date, at any given location,” and dragging the sun icon shows its position at a chosen time along with “the corresponding length of its shadow” for an object of known height.
The constraint runs both ways, which is why the two words travel together:
- Chronolocation, location known and time unknown: match observed shadows against the shadows SunCalc predicts through the day.
- Geolocation, time known and location unknown: the sun’s bearing fixes which way the camera faced, and that eliminates candidate locations wholesale.
The worked example
A Portuguese video timestamped 22 November 2020, 16:31 UTC:
- Timestamp and spoken language narrow the region.
- SunCalc for that date, time and area shows “the sun was about to set at that time and to the South-West of the city.”
- Reading the sun against the visible sea and the line of the avenue gives camera orientation: “the person filming was facing South and that the avenue’s direction runs roughly East to West.”
- That orientation “excludes several locations — not only further inland, but also any areas of Greater Lisbon which are South of the Tagus River.” The elimination is the step that does the work.
- Remaining candidates are settled on distinctive detail: Avenida da República in Oeiras, with “four lanes and a row of palm trees.”
Total time: “just five to ten minutes.”
Why it matters here
Every conclusion above is stated with the evidence that forces it, and a reader can rerun each step. That is what berkeley-protocol‘s ¶25 explainability requirement asks for and what glan-bellingcat-methodology‘s Incident Assessment is built to display — here it is, in five steps over a public tool.
Set it beside llm-geolocation-test and the contrast is exact: this method reaches an answer by ruling places out from a physical constraint, while a model reaches one by recognising a scene. The first shows its working and the second, on Bellingcat’s own measurement, sometimes fabricates it.
Connections
- published by bellingcat
- the discipline: geolocation-chronolocation
- assumed but not taught by glan-bellingcat-methodology
- the explainability bar it satisfies: berkeley-protocol
Related
bellingcat · geolocation-chronolocation · glan-bellingcat-methodology · llm-geolocation-test · berkeley-protocol