GeoArrow (Apache Arrow)

Apache Arrow is the standard for representing tables in memory: columns lie contiguously, the schema is described explicitly, and different programs read the very same bytes without copying or re-encoding. GeoArrow is the agreement on how to store geometry inside that structure so that GeoPandas, DuckDB, Polars, deck.gl and GDAL understand it.

The format is taken when data is not displayed but computed: spatial joins over millions of features, feature preparation for a model, a dashboard on top of a columnar engine. At those volumes, parsing text GeoJSON eats more time than the analysis itself.

The output is a single .arrow file (Apache Arrow IPC, also known as Feather v2). All the layers are merged into one table, and every row has a layer column with the name of its source layer. Geometry is stored as a binary column in WKB format, and the WGS84 (EPSG:4326) coordinate system is described in the schema metadata per the GeoArrow 1.0.0 convention.

By default the file is compressed with the zstd codec; you can choose lz4 or no compression at all. There is a safety limit of 10 million features per export, so as not to eat up the server's memory.