Why we use Sentinel-2 and NAIP as references
To place an old image you need a modern one of the same place to compare it with. The sharper that reference, the better the match, so the obvious move is to use the sharpest basemap going. We could not.
The licensing problem
Well-known satellite basemaps sell map displays. Their terms generally do not let you read the pixels and run analysis on them in a paid product. We read the published terms of several providers. Some prohibit creating derived content or using the imagery with machine learning. Others restrict bulk or automated fetching. For the one that looked most permissive we could not read the governing agreement, so we did not build on it.
That left sources whose licences are explicit.
What we use
- Sentinel-2 from Copernicus: 10 m, worldwide, free, commercial use allowed.
- USGS NAIP: about 1 m aerial photography of the United States, public domain.
- OpenStreetMap, for scanned maps, under the ODbL.
What 10 m costs
A 10 m reference cannot show what a 2 m photograph shows, so matches are weaker than against sub-metre imagery. In the United States NAIP closes that gap, but only for small areas. In a test, a 20 km box returned NAIP at about 20 m per pixel, coarser than Sentinel-2, and matched only 3 points. A 7 km box matched 87 of 93 points with a residual of 1.7 pixels. So the tool only offers NAIP for areas under about 12 km.
We also tried super-resolution
An open model can sharpen Sentinel-2 to 2.5 m, and it looks convincing. We compared matching against plain and sharpened Sentinel-2 at three US sites, with clean and with deliberately degraded photographs. There was no consistent gain, and sometimes it was worse. In low-texture desert both failed. We left it out.
These were small tests on modern aerial imagery as a stand-in for old photographs, so treat them as indications, not proof.
Try the tool or read the FAQ.