How to extract GIS data from a website
The short answer: find out what kind of service the map is using, because that decides everything. Vector services hand you real features. Image and tile services hand you pictures. No amount of cleverness turns one into the other.
Most "how do I scrape this map" questions are really this question, asked before knowing which situation you are in. So start there.
First, work out what you are looking at
Open the map, press F12, go to the Network tab, and reload the page. Then look at what the map is asking for.
| What you see in the requests | What it is | What you can get |
| --- | --- | --- |
| GetMap, SERVICE=WMS | WMS | Images. Sometimes features via WFS on the same server |
| GetFeature, SERVICE=WFS | WFS | Real features |
| /FeatureServer/0/query | Esri FeatureServer | Real features |
| /MapServer/export | Esri MapServer | Images, sometimes features |
| /collections/.../items | OGC API Features | Real features |
| .png or .jpg at /12/4021/2578.png | XYZ tiles | Pictures of data |
| .pbf or .mvt | Vector tiles | Geometry, but chopped up per tile |
| A single .geojson or .json | A static file | The whole thing, immediately |
The good outcome is any row that says "real features". The awkward one is tiles.
The easy case: a service that serves features
If the map is backed by WFS, a FeatureServer, or OGC API Features, you can just ask for the data.
For WFS:
https://example.com/wfs?SERVICE=WFS&VERSION=2.0.0&REQUEST=GetFeature
&TYPENAMES=parcels&OUTPUTFORMAT=application/json&COUNT=1000
For Esri, see getting data out of an ArcGIS web map.
In both cases the practical advice is the same: ask for a bounding box rather than the whole dataset, page through it rather than demanding 50,000 features at once, and check the count before you start.
The easy case you might miss: a file sitting right there
A surprising number of web maps are a static GeoJSON file and a bit of JavaScript. If
you see a single request pulling data.geojson or layers.json, that is the whole
dataset, and you can just save it. Right-click the request, Open in new tab, and save
the page.
Same for KML. Same for a .zip that turns out to be a shapefile.
Downloading GeoJSON or KML from a web map covers
this in more detail.
The hard case: tiles
If the map is drawing raster tiles, there is no underlying data to extract. Each tile is a 256x256 picture. Stitching a few hundred together gets you a big picture, not a dataset. You cannot recover the parcel boundaries from a photograph of the parcel boundaries, and attempting it with automated tracing produces a worse dataset than whoever published it already has.
Vector tiles (.pbf, .mvt) are more hopeful, because they contain real geometry. But
that geometry has been simplified for display at each zoom level, clipped at every tile
boundary, and often stripped of most of its attributes. You can extract it, and the
result is a mangled, tile-shredded approximation. It is fine for a background layer and
wrong for anything you would measure.
When you hit tiles, the better move is to stop and look for the source. Search the organisation's open data portal for the layer name. Very often the same data is published properly as a download, and you have been trying to reverse-engineer something that was available all along.
The one that surprises people
Look at the capabilities document, not just the layer the viewer showed you:
https://example.com/wms?SERVICE=WMS&REQUEST=GetCapabilities&VERSION=1.3.0
Viewers are curated. Services frequently publish far more than the map exposes, and it is all equally public. What GetCapabilities tells you is the guide for reading it.
Doing all of this automatically
This entire process, working out what service is behind a map, cleaning up its URL, reading its capabilities, and checking whether it will load in a browser, is what Geo Hound does while you browse. It recognises OGC, OGC API, Esri REST, tiles, and linked files, and saves each with its metadata already fetched.
It will not turn a raster tile into a dataset, because nothing can.
Where the limits actually are
Technical: you can get what the browser can get. If a service requires a session, an API key, or a signed URL, Geo Hound and DevTools alike only see it while you are logged in, and it will stop working outside that session.
Legal: this is the one that matters. Public does not mean unlicensed. Most government data carries an open licence requiring attribution, but plenty is more restrictive, and some viewers show data the host does not have redistribution rights over. The terms are usually on the portal.
The reasonable test: are you using data the way its publisher would recognise as reasonable? Downloading a suburb's parcels to answer a question is what the service is for. Scraping the national dataset overnight to resell it is not, and it is also how services end up behind logins for everybody else.