GeoJSON, Shapefile, DXF, GML, KML and CSV formats.
Seven formats from one set of polygons: GeoJSON for web maps, ESRI Shapefile (.shp) for desktop GIS, DXF for CAD, plus GML, KML and CSV. Only the DXF is approximated. It's projected to meters and simplified to about 45 m so AutoCAD behaves; everything else is the full source geometry.
.geojsonRFC 7946, the reference version of the data.
.shpWith .shx, .dbf, .prj and .cpg. UTF-8 attributes intact.
.svgOne group per division, keyed by its code. Names sit in a layer you can hide.
.dxfR12 ASCII, projected to meters, simplified to about 45 m.
.gmlOGC GML 3.2, latitude written before longitude.
.kmlStyled so it looks reasonable when you double-click it.
.csvUTF-8 with BOM, so Excel shows the Chinese names instead of squares.
Fields on every feature
Six of the seven formats carry these columns with these names. The shapefile is the exception: dBase caps a field name at ten characters and has no array type, so five columns are abbreviated and bbox is absent. If you are joining a .dbf to the CSV, this is the mapping.
| Formats | GeoJSON · SHP · SVG · DXF · GML · KML · CSV, one zip |
|---|---|
| Coordinate system | WGS 84 · EPSG:4326 · decimal degrees |
| Geometry | MultiPolygon — one feature per division |
| Codes | 6-digit GB/T 2260, e.g. 440300 |
| Names | name_en and name_zh on every feature |
| Hierarchy | province_code → city_code → code |
| Extras | parts, center_lon, center_lat, bbox |
| License | Commercial use, one organization, no attribution required |
| Delivery | Instant download; links re-issued whenever you open them |
| Data current to |
| GeoJSON and CSV | Shapefile (.dbf) |
|---|---|
| province_code | PROV_CODE |
| province_en | PROV_EN |
| province_zh | PROV_ZH |
| center_lon | LNG |
| center_lat | LAT |
bbox has no .dbf column: dBase cannot hold an array. Read it from the GeoJSON, or compute it from the geometry.
Why the DXF is different
CAD doesn't want degrees. A line weight in millimeters against a coordinate in degrees means nothing, and the offsets snap to a grid nobody can use. So the DXF is written in meters, projected with Lambert Conformal Conic: standard parallels 25° and 47°, central meridian 105°, on WGS 84. The proj string is:
+proj=lcc +lat_1=25 +lat_2=47 +lat_0=35 +lon_0=105 +x_0=0 +y_0=0 +datum=WGS84 +units=m +no_defs
Geometry is simplified with Douglas–Peucker at roughly 45 m. You won't see it at print scale, and it keeps a national county file openable. If you need CAD at full resolution, convert from the shapefile or GeoJSON with your own projection.
Why the GML writes latitude first
Because that's the axis order EPSG:4326 actually defines. Writing longitude first under that srsName is the classic reason a GML file opens rotated in GeoServer. GDAL, PostGIS and ArcGIS all read this ordering correctly.
A feature, as delivered
{ "type": "Feature",
"id": "440300",
"properties": {
"code": "440300",
"level": "city",
"name_en": "Shenzhen City",
"name_zh": "深圳市",
"province_code": "440000",
"province_en": "Guangdong Province",
"province_zh": "广东省",
"parts": 1,
"center_lon": 114.06,
"center_lat": 22.55,
"bbox": [113.75, 22.39, 114.63, 22.87]
},
"geometry": { "type": "MultiPolygon", "coordinates": [ ... ] }
}
Two pages go further than the table above: what datum these files are in, and why a China map can look shifted, and what the layers are and how the six-digit code joins them.