Zarr (beta)

Zarr is a format for storing large multidimensional arrays: the data is cut into chunks, and each chunk is compressed and stored separately. That is why an array is read in pieces, in parallel and straight from the cloud — S3, GCS or plain HTTP. In scientific data it is gradually displacing NetCDF and HDF5 precisely because of that cloud access.

An OpenStreetMap map is vector objects while Zarr is about arrays, so the export rasterizes the selected area into a regular grid. The format is taken when the map is needed not as a drawing but as a data layer next to climate data, satellite series or a computational model in xarray and Dask.

The archive holds a Zarr group of version two with two arrays: feature_count — how many OSM objects fell into the cell, and land_use — the code of the dominant layer in the cell. The group attributes record the coordinate system, the bounding box of the area, the grid step and the list of source layers; the data is split into 256 × 256 chunks and compressed (Blosc with Zstandard by default).

The default grid step is 0.001 degrees, about a hundred meters; over small areas it is refined automatically so that the result does not degenerate into a few cells. The grid size is limited to 10,000 × 10,000 cells. The format is in beta status: this is aggregated density and categories, not vector geometry — for that, go to GeoParquet or FlatGeobuf.