HDF5 (Hierarchical Data Format v5)

HDF5 is a binary container for scientific data laid out like a file system inside a single file: groups act as folders, datasets as arrays, attributes as metadata. The format is maintained by the HDF Group, it is read by Python, MATLAB, R and IDL, and it is designed so that data stays understandable without accompanying documentation.

OpenStreetMap maps are exported to .h5 when the next step is a scientific or engineering pipeline: the computation works on arrays, and one self-describing file is wanted instead of a pile of shapefiles. It is especially convenient when your own measurements already live in that same file — HDF5 happily holds heterogeneous data in one hierarchy.

Inside there is one group per OSM layer. Each group holds the longitude and latitude datasets with feature centroids and one string dataset per attribute field; coordinates are written as float64, attributes as variable-length UTF-8 strings. Metadata is written at the root and in the groups: title, source, extent, feature count, geometry type.

Honestly about the limitations: this is a container of coordinates and attributes, not a raster cube. Geometry is reduced to centroids — there are no building outlines or road lines in the file, and the exporter does not compute analytical grids (density, scores, terrain). Writing goes through h5py, and if the library is unavailable, fallbacks kick in down to gzipped JSON with the same logical structure.