Frequently asked questions

No. It is an experimental version. Results can be wrong, some images will not match at all, and a match can look convincing and still be off. It fits a simple transform (scale, rotation and shear), so it cannot correct paper distortion, a tilted camera or a change of map projection. Check every result on the map before you rely on it. The tool tells you when to be careful, but it cannot catch everything.

Georeferencing is working out where an image sits on the Earth, so a map program can draw it in the right place at the right scale. The result is usually a small file of numbers, such as a world file, that goes alongside the image.

No. Matching runs in your browser on your own graphics card. The page does fetch reference imagery for the area you choose, so the location you ask about is requested from the imagery providers, but the image itself stays on your computer.

Aerial and satellite photographs of places with roads, buildings, field boundaries or coastlines usually work best, because those features survive for decades. Scanned map sheets can be matched against a rendered map. Images of open desert, sea, forest or snow have few features to match, and often fail.

Roughly, yes. You can search for a place, draw a box or type coordinates. If the file has a GPS tag, the tool reads it. The area does not have to be exact, but it has to contain the image.

Sentinel-2 satellite imagery at 10 m worldwide, USGS NAIP aerial imagery at about 1 m for the United States, a rendered OpenStreetMap for scanned maps, or a reference image you provide yourself. Sharper commercial basemaps are not used, because their licences do not allow this kind of automated use.

The tool reports how many points matched, how many were consistent, and the residual error. Treat those as the guide. In our tests on clean aerial images the residual was one to two pixels of the reference, which is roughly 10 m against Sentinel-2. Old, grainy images of areas that have changed a lot do worse, and some fail. This is a prototype, so check every result.

The common causes are an area with too little detail, an image taken long before a lot of change, a search area that is too large, or too little overlap with the reference. Try a smaller area, a different reference, or a different part of the image. In the United States, NAIP is sharper than Sentinel-2 but only helps on areas under about 12 km across.

A world file (.wld, or the format-specific .jgw, .pgw or .tfw), a QGIS Georeferencer .points file so you can inspect and adjust the fit, and a GeoJSON footprint.

CORONA was the original motivation. Whether a particular scene matches depends on the scene. Imagery of towns, roads and irrigated land often gives enough to match. Very large frames may need to be cut into smaller pieces first.

A recent Chrome, Edge or other Chromium browser with WebGPU is fastest. Other browsers fall back to a slower method.

Yes, for now. A paid version is planned for the future, with features such as larger maps, batches, sharper reference imagery and transforms for scans with paper distortion. A free, limited version will stay open. Everything runs on your own machine, so the free version does not depend on a server bill, and nothing that works today will move behind the paid version.