Hugging Face's logo

Datasets:
MichaelMedek
/
dach_bike_graph

Tasks:
Other
Modalities:
Tabular
Geospatial
Formats:
parquet
Size:
10M - 100M
Tags:
geospatial
routing
graph
cycling
railway
openstreetmap
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
from_node
int64
0
8.46M
to_node
int64
0
8.46M
key
int64
0
48
elevation_deviation_m
float64
50
1.64k
⌀
3,170,947
3,296,495
0
65
3,187,811
3,187,808
0
80
3,260,318
3,260,320
0
70
3,260,318
3,294,449
0
52
3,299,470
3,187,860
0
60
3,295,142
3,295,144
0
51
3,223,107
3,223,110
0
75.345454
3,232,909
3,232,901
0
56
3,253,445
3,195,750
0
55
3,196,180
3,257,399
0
63
3,222,847
3,196,181
0
69
3,196,382
3,196,374
0
56
3,196,382
3,196,386
0
58
3,223,116
3,223,118
0
55
3,223,116
3,299,522
0
64
3,223,030
3,263,793
0
55.555494
3,223,030
3,277,010
0
105
3,196,405
3,196,410
0
74
3,196,518
3,292,642
0
52
3,292,643
3,292,662
0
75
3,292,626
3,292,624
0
51.515387
3,196,556
3,281,658
0
72
3,217,040
3,221,822
0
58
3,217,040
3,221,823
0
62
3,217,045
3,217,048
0
54
3,222,744
3,217,047
0
66
3,296,551
3,300,096
0
61
3,222,610
3,222,607
0
52
3,270,145
3,300,345
0
51
3,270,145
3,300,346
0
58
3,280,594
3,222,984
0
96
3,223,015
3,223,026
0
63
3,223,014
3,223,016
0
55.290459
3,223,026
3,223,021
0
51
3,223,025
3,223,022
0
122
3,223,029
3,223,026
0
67.412432
3,223,035
3,257,508
0
54
3,223,075
3,223,079
0
59.078983
3,223,169
3,223,171
0
58
3,300,522
3,300,524
0
53
3,223,115
3,295,021
0
77
3,223,115
3,295,018
0
66.105756
3,281,991
3,281,993
0
68
3,281,991
3,281,994
0
91
3,281,991
3,282,005
0
71.296644
3,281,991
3,295,172
0
51
3,295,022
3,295,019
0
56
3,300,341
3,223,168
0
55
3,229,856
3,229,861
0
76
3,232,974
3,232,977
0
58
3,322,686
3,322,701
0
53
3,240,180
3,240,179
0
54.26701
3,240,181
3,294,990
0
81.473986
3,241,278
3,246,641
0
70
3,246,636
3,257,553
0
61
3,257,553
3,257,552
0
76
3,266,946
3,266,993
0
53
3,304,437
3,304,438
0
63
3,304,436
3,257,554
0
94
3,294,705
3,294,706
0
67
3,257,891
3,257,892
0
51
3,257,907
3,257,895
0
70
3,260,313
3,260,316
0
67.096489
3,260,313
3,260,317
0
54
3,260,321
3,260,323
0
64
3,295,131
3,295,129
0
51
3,263,772
3,260,326
0
108.45437
3,260,325
3,260,323
0
93
3,263,761
3,263,764
0
56
3,300,318
3,300,321
0
62
3,266,957
3,266,996
0
95.640718
3,283,232
3,283,235
0
66.670087
3,276,995
3,276,996
0
58.112221
3,277,013
3,277,036
0
58
3,277,011
3,277,010
0
51
3,277,039
3,277,037
0
52.416518
3,277,024
3,277,021
0
62
3,277,026
3,295,310
0
57.17616
3,277,026
3,295,307
0
89
3,277,075
3,277,024
0
65
3,277,037
3,277,036
0
55
3,277,092
3,293,700
0
73
3,293,699
3,298,646
0
70
3,279,043
3,279,042
0
71
3,294,523
3,294,524
0
69
3,294,524
3,294,547
0
56
3,308,029
3,280,593
0
75
3,298,111
3,305,073
0
63
3,294,525
3,294,528
0
60
3,281,992
3,294,593
0
95
3,281,992
3,294,144
0
54
3,282,005
3,294,535
0
85.336521
3,295,172
3,295,173
0
54
3,295,172
3,295,173
1
54
3,295,172
3,281,996
0
58.614753
3,295,025
3,295,006
0
86
3,295,025
3,281,991
0
56.486334
3,281,995
3,294,136
0
56
3,283,241
3,295,219
0
51.448152
3,283,242
3,283,240
0
54.854344
End of preview. Expand in Data Studio

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.

DACH bike + rail graph overview

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

Coverage

Parameter Value
West ~5.9° E
East ~17.2° E
South ~45.8° N
North ~55.1° N
Region Germany + Austria + Switzerland
Base data OpenStreetMap cycling network + railway (via Geofabrik extracts)
Elevation EuroDEM, sampled and baked into every node and edge vertex at build time

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
edge_unreliable_elevation/tile_<row>_<col>.parquet   # optional; only where offenders exist
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).

edge_unreliable_elevation (optional) — from_node, to_node, key, elevation_deviation_m: the max metres a bike edge's baked terrain deviates from its straight node-to-node line, for edges past a 50 m threshold only. A consumer may join it on (from_node, to_node, key) to deprioritize edges whose elevation profile is unreliable; routing works fine without it.

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).
Downloads last month
167
Number of rows:
33,688,038
Total file size:
1.87 GB