How China's boundary data is compiled.

Chinese administrative boundaries usually get assembled the same way: three or four partial sources, names reconciled by hand, and six months later two neighboring provinces overlap by a kilometer. This site exists so that part is a download instead of a week.

How it's built

The division list and the six-digit codes come from GB/T 2260, the national standard for administrative division codes. Geometry comes from public boundary data and is matched to that list, so every polygon has a code and every code has a polygon. When the source publishes something the list doesn't have (a forest reserve, a horse ranch, a county-level city the province administers directly), it's kept and labeled. A hole in the map is worse than an unusual feature.

What gets checked

• Codes: format, uniqueness, and that every child's parent code exists one level up.
• Names: reviewed by hand for the 368 second-level units and for the irregular county names. Pinyin elsewhere, with the standard spelling for apostrophes and ü.
• Geometry: one feature per unit, closed rings, a sane bounding box for each.
• Files: each one is reopened before it goes into the zip. The shapefile with a shapefile reader, the DXF with a DXF reader, the XML against a parser.

What it isn't

Survey grade, for a start. These are the published administrative lines at a generalized scale, which is right for analysis, mapping and products, and wrong for settling a property dispute or setting out a construction site.

License, in one line

Use it for whatever you like, commercial work included, provided you aren't reselling the data itself as a dataset. The full text is on the license page.