Mateusz Loskot :: hacking on, working out, living up

OGR GeoJSON Driver

06 Nov 2007 | mloskot

Update on Nov 6th, 2007 at 3:23 pm: GeoJSON Driver Errata

During last couple of days I was trying to run away in fear of Halloween monsters, so I locked myself with laptop in my wardrobe and decided to spent this time on programming a new driver for OGR. I ended up with usable implementation of GeoJSON driver for OGR and I’d like to introduce it here and now.

Motivation was simple…to avoid zombies but not only. In the MOSS4G project, we need a way to talk to remote geospatial data services. from mobile devices. We’ve started development of WFS client but it’s not a trivial task and we don’t want to hurry with that. Also, WFS and parsing GML seems to be a heavy task for mobile solutions, but we will check it later :-). In the meantime, we have something light and fast - GeoJSON.

The GeoJSON is a new dialect based on JSON format. The JavaScript Object Notation (JSON) is a lightweight plain text format for data interchange and GeoJSON is nothing other than its specialization for geographic content. The GeoJSON format fits very well the same niches as GML, like geospatial data interchange over network. Currently, GeoJSON is supported as output format of services implemented by FeatureServer, GeoServer and CartoWeb.

It’s correct to assume that every valid GeoJSON encoded data is valid JSON content, so it’s possible to process GeoJSON data using any JSON parser you like. Internally, the OGR GeoJSON driver uses JSON-C library as a JSON parser. The JSON-C is a small and fast validating parser of JSON format implemented in portable C language by Michael Clark from Metaparadigm Pte. Ltd. The library is an Open Source Software and available under the terms of the MIT License.

The OGR GeoJSON driver is available in SVN repository of MOSS4G project in module trunk/libs/gdal/extensions/ogr. In near future, I’d like to add this driver to the official GDAL/OGR repository. For now, I’ve included patched GDAL makefiles so it should be feasible to build the driver using GDAL sources. The OGR GeoJSON driver is also available under the MIT License.

The OGR GeoJSON driver provides implementation of functions transforming GeoJSON encoded data to objects of OGR Simple Features model: Datasource, Layer, Feature, Geometry. The implementation is based on GeoJSON Specification draft, v5.0.

At the moment, the GeoJSON driver provides read-only access to all types of supported datasources (see below):

$ ogrinfo --formats
Supported Formats:
  -> "GeoJSON" (readonly)

In near future, I’m going to add ability to insert and delete features, but it will be only available for FeatureServer connections.


The OGR GeoJSON driver accepts three types of sources of data:

I should indicate that the FeatureServer is the only Web Service provider of GeoJSON data I’ve used during tests.


Currently, there is no support of multiple layer datasource. It means, that if we open a datasource the driver will always produce single OGR layer. It isn’t clear how to categorize GeoJSON objects into separate layers and there is no concept like GetCapabilities request from OGC Web Services. Another assumption is that GeoJSON is medium and server independent format, what means GeoJSON data can be stored in a file, a table in database or read from a Web Service. The FeatureServer provides HTTP request to access list of available layers but it sends response encoded in JSON but not GeoJSON format. I tried to stick to GeoJSON Specification and avoid features specific to FeatureServer (or any other server). Anyway, if this subject will be solved in the GeoJSON Specification somehow, I will implement multi-layers support.

As there are no layer objects in the GeoJSON Specification, there are also no names of layers. Because the driver generates single OGRLayer object per datasource, I enforce the driver to use pre-defined name OGRGeoJSON, like this:

ogrinfo -ro http://featureserver/data/.geojson OGRGeoJSON

When a Web Service (ie. FeatureServer datasource) is used as datasource, each request will produce new layer. This behavior conforms to stateless nature of HTTP transaction. Just to give a simple analogy:

I hope it makes sense and clarifies behavior of the OGR GeoJSON driver a little.

The GeoJSON Specification lists following types of objects that may occur in data: Point, LineString, Polygon, MultiPoint, MultiLineString, MultiPolygon, GeometryCollection, Feature, or FeatureCollection. The 3 types of the Multi* geometries are not supported at this moment, but it’s not a big deal, I just need to find 2-3 hours gap in my schedule :-).

If a top-level member of GeoJSON data is of any other type than FeatureCollection, the driver will produce a layer with only one feature. Otherwise, a layer will consists of a set of features.


The OGR GeoJSON driver maps each object of following types to new OGRFeature object: Point, LineString, Polygon, GeometryCollection, Feature.

According to the GeoJSON Specification, only the Feature must have a member with the name properties. Each member of properties is translated to OGR object of type of OGRField and added to corresponding OGRFeature object.

The GeoJSON Specification does not require all Feature objects in a collection must have the same schema of properties. If Feature objects in a set defined by FeatureCollection object have different schema of properties, then resulting schema of fields in OGRFeatureDefn is generated as union of all Feature properties.

For example, for given two features with different set of properties:

F1 = { a b c }
F2 = { b c d }

schema of resulting OGRLayer will include following attributes:

L1 = { a b c d e }

and attribute d for F1 is null and attribute a for feature F2 is also null.

I was also thinking about implementing support to request only attributes common for all features, an intersection, but I’ve put this concept away for a while. Intersecting attributes of big set of features might be a very slow process. Anyway, I’d like to ask for comments and ideas. Feel free to drop yours in the comments below.

It is possible to tell the driver to not to process attributes. Just set environment variable ATTRIBUTES_SKIP=YES. Default behavior is to preserve all attributes (an union), what is equal to setting ATTRIBUTES_SKIP=NO.

At the moment, it’s not possible to pass additional options to OGRSFDriver::Open() from the OGR API, so setting environment is just a simple workaround. This situation will probably change in future, because there is a plan to implement extension to the OGR API to handle extra options on open action. For details, check this document RFC 10: OGR Open Parameters.


Similarly to the issue with mixed-properties features, the GeoJSON Specification draft does not require that all Feature objects in a collection must have geometry of the same type. Unfortunately, OGR objects model does not allow to have geometries of different types in single layer (see GeoJSON Driver Errata). To handle this limitation I decided to use OGRGeometryCollection as a meta geometry. By default, the OGR GeoJSON driver preserves type of geometries. To avoid problems with handling mixed-geometry in GeoJSON data, then you can set another environment variable, GEOMETRY_AS_COLLECTION=YES (default is NO) to tell the driver to wrap all geometries with OGRGeometryCollection type. Again, if you see a better solution, please don’t hesitate to share it with me.

Here is sample output from ogrinfo presenting this feature in action, with default setting GEOMETRY_AS_COLLECTION=NO (default):

  name (String) = Pierwszy
  POINT (0 0)

  name (String) = Drugi
  POINT (10 10)


  name (String) = Pierwszy

  name (String) = Drugi

Here are a few examples of GeoJSON driver in action:

The plan

I’d like to thank Christopher Schmidt and Howard Butler for their help (yes, fortunately there was some WiFi signal in the closet, so I could chat with these great friends on the IRC ;-)) in understanding elements of GeoJSON and FeatureServer.

Fork me on GitHub