geolocate-from-pixels

Inventory every visible clue first, then rank them by how much of the world each eliminates. Readable text, phone numbers, and licence plates narrow to a country; road markings, utility poles, and bollards narrow to a region; then confirm candidates against satellite and street-level imagery.

Measure the shadow azimuth against a mapped reference line and the solar elevation from an object height-to-shadow ratio, then solve with SunCalc or the NOAA Solar Calculator. Sun geometry yields two date bands per year, which you disambiguate using foliage state, snow, or dated objects in frame.

Google Earth desktop provides historical imagery timelines, Street View and Yandex Panoramas cover street-level views, and Mapillary and KartaView fill rural gaps. Sentinel-2 and Landsat date landscape changes, while Overpass queries OpenStreetMap by feature geometry.

Contradictions are findings, not failures. Cyrillic signage with right-hand-drive vehicles indicates an import market, border region, or composited image. Border regions and enclaves routinely mix plates, signage, and infrastructure, so single-clue logic breaks down there.

Shadow geometry alone produces two symmetric date bands per year and cannot resolve a single date. Near solstices, declination barely changes, so date precision degrades to weeks. Slope, non-vertical objects, and assumed object heights introduce significant elevation error.

Ornamental and introduced plants are distributed globally, so species like eucalyptus and palms appear far outside their native ranges. Vegetation should only corroborate a hypothesis formed from stronger clues like plates, poles, and signage, never form one.