MichaelMedek/dach_bike_graph

DACH Bike + Rail Routing Graph

A prebuilt, ready-to-route cycling + railway graph covering Germany, Austria, and Switzerland (DACH), stored as lat/lon-tiled GeoParquet. Built from OpenStreetMap by the Bike Route Optimizer for flat-preferring, surface-aware bike routing that can also hop on a train uphill.

Everything routing needs is baked in — node elevations and full 3D edge geometry — so an application downloads this once and routes offline, with no Overpass and no elevation service at query time.

[image]

The whole network — bike roads (thin blue) and rail lines (thick purple). Shipped as `dach_graph_overview.png`.

Coverage

The exact bounding box built is recorded in meta.json (bbox = [west, south, east, north]).

What's in the graph

• —Bike network — every bike-legal way, with genuinely unrideable surfaces (mud, sand, rock, …) removed, degree-2 pass-through nodes contracted, and intersections within 25 m consolidated to shrink the graph while keeping the true road shape.
• —Railway — heavy rail plus regional light_rail and narrow_gauge (trams, funiculars, and park/miniature railways are excluded). Each station is snapped onto the nearest rail line by a rail edge, and joined to nearby bike nodes (within 200 m) by bike↔station station access edges.
• —Baked elevation — every node has an elevation_m, and every bike and rail edge's geometry is a 3D LINESTRING Z (lon lat elev …), so the elevation profile follows the real road/track.

File layout

The dataset is tiled on a 0.5° lat/lon grid so a consumer reads only the tiles a route corridor spans, not the whole country.

meta.json                     # bbox, tile_deg, consolidation tolerance, node/edge/station counts, source regions
nodes/tile_<row>_<col>.parquet
edges/tile_<row>_<col>.parquet
dach_graph_overview.png       # whole-network preview image

Schema

nodes — osmid, lat, lon, elevation_m, node_type, station_name (node_type is bike or rail; station_name is null for bike nodes and, on a rail node, may still be null for an unnamed halt).

edges — from_node, to_node, key, length_m, height_diff_m, surface, highway, mode, geometry_wkt

• —mode is one of bike, rail, station (a station edge is the bike↔station access link).
• —geometry_wkt is a 3D LINESTRING Z for bike and rail edges; null for straight station hops.
• —key is the parallel-edge index (a (from_node, to_node, key) triple is unique).

Coordinates are WGS84 (EPSG:4326), rounded to 6 decimals (~0.1 m).

Usage

from huggingface_hub import hf_hub_download, list_repo_files

repo = "MichaelMedek/dach_bike_graph"
for filename in list_repo_files(repo_id=repo, repo_type="dataset"):
    hf_hub_download(repo_id=repo, repo_type="dataset", filename=filename, local_dir="dach_graph")

Then read the tiles covering your area with pandas / pyarrow and rebuild a graph — or just use the Bike Route Optimizer, which downloads and windows this dataset automatically.

How it was built

Reproducible from OpenStreetMap: each Geofabrik region extract is read with pyrosm, the cycling + railway networks are extracted, surfaces filtered, nodes contracted and consolidated, elevation baked from the EuroDEM, and railway links added; the 31 orginal DACH sub-regions (Autria and Switzerland are furter split during pre-processing) are then merged, seam-deduplicated, and tiled.

License

Derived from OpenStreetMap, so distributed under the Open Database License (ODbL). © OpenStreetMap contributors. Elevation baked from the EuroDEM (EuroGeographics).