Datasets:

VaidhyaMegha
/

wildfire-evac-kg

license: odbl
language:
  - en
tags:
  - knowledge-graph
  - graph-database
  - cypher
  - openstreetmap
  - road-network
  - geospatial
  - wildfire
  - emergency-management
pretty_name: Wildfire Evacuation Road Network — Paradise, California
size_categories:
  - 10K<n<100K
configs:
  - config_name: junction
    data_files: nodes/junction.csv
  - config_name: road
    data_files: nodes/road.csv
  - config_name: edge_connects
    data_files: edges/connects.csv
  - config_name: edge_on_road
    data_files: edges/on_road.csv

Wildfire Evacuation Road Network — Paradise, California

13,787 nodes. 26,831 edges. The road network of Paradise, California as a graph — 12,610 junctions and 1,177 road segments, with the connectivity between them.

Source extract: samyama-ai/wildfire-evac-kg. Graph tooling: Samyama Graph.

⚠️ Licence: ODbL 1.0 — share-alike, and it propagates

This dataset is OpenStreetMap data under the Open Database License 1.0. ODbL is stronger than the CC-BY licences most datasets carry: it is share-alike, so a derived database must itself be released under ODbL.

© OpenStreetMap contributors. Data licensed under ODbL 1.0.

If you redistribute this data, or publish a database derived from it, you must:

Attribute OpenStreetMap contributors
License the derived database under ODbL 1.0
Keep any technical restriction you add accompanied by an unrestricted version

Producing an analysis or a map from it (a "Produced Work") does not require ODbL licensing of that work, but does require attribution. Read the licence if you are unsure which side of that line you are on.

The licence is not inferred. The extract carries it in its own metadata:

"osm3s": {
  "timestamp_osm_base": "2026-05-11T16:29:15Z",
  "copyright": "The data included in this document is from www.openstreetmap.org.
                The data is made available under ODbL."
}

Provenance, and what is derived


Source	OpenStreetMap via the Overpass API
Generator	`Overpass API 0.7.62.11 87bfad18`
OSM base timestamp	2026-05-11T16:29:15Z
Area	Paradise, California, USA
Raw extract	`paradise_ca.json` — 1.6 MB, 13,787 elements, shipped unmodified

The graph form is derived by this publication, not by the upstream repository. That repository states plainly: "This repository holds the source extract, not a built graph. There is no loader, schema or .sgsnap snapshot yet." The node/edge CSVs here were produced by the obvious reading of the extract — OSM nodes become junctions, OSM ways become roads, consecutive node references along a way become connections. Nothing is invented: every row traces to an element in paradise_ca.json, which is included so you can check.

Why this place

Paradise is the town largely destroyed by the 2018 Camp Fire, in which evacuation capacity — a small number of outbound routes carrying the whole town at once — was central to what happened. The road network is therefore a real, consequential test case for evacuation modelling rather than an arbitrary extract.

The graph makes the structural questions direct:

// Roads by type — how much of the network is residential vs through-route
MATCH (r:Road) RETURN r.highway, count(*) AS n ORDER BY n DESC

// Junction degree: where the network narrows
MATCH (j:Junction)-[c:CONNECTS]-()
RETURN j.id, j.lat, j.lon, count(c) AS degree
ORDER BY degree DESC LIMIT 20

// Everything reachable from a junction within N hops — evacuation catchment
MATCH path = (start:Junction {id: $j})-[:CONNECTS*1..6]-(reach:Junction)
RETURN count(DISTINCT reach) AS reachable

Files

File	Rows	Columns
`nodes/junction.csv`	12,610	`id, osm_node_id, lat, lon`
`nodes/road.csv`	1,177	`id, osm_way_id, highway, name, surface, lanes, maxspeed, oneway, bridge, tunnel, junction_count`
`edges/connects.csv`	12,827	`src, tgt, road_id, highway` — Junction → Junction, consecutive along a road
`edges/on_road.csv`	14,004	`src, tgt` — Junction → Road
`paradise_ca.json`	—	the raw Overpass response, unmodified

Road types: residential 1,110 · secondary 30 · tertiary 24 · unclassified 13. 1,064 of 1,177 roads are named — including Skyway, the principal outbound route.

Usage

from datasets import load_dataset
junctions = load_dataset("VaidhyaMegha/wildfire-evac-kg", "junction")

Verification

0 dangling edges — every src and tgt resolves to a node in this dataset.
0 duplicate ids within either node file.
0 orphan nodes — every junction and road participates in at least one edge.
0 way node references missing from the extract — every node a road refers to is present, so no connection was dropped or invented.
Node total 13,787 equals the extract's own element count (12,610 nodes + 1,177 ways).
Road names check out against the real town — Skyway, Pearson Road, Clark Road and 1,061 others.

Limitations

A point-in-time OSM snapshot (2026-05-11) of a town that was rebuilt after 2018. It is not the network as it stood during the Camp Fire, and must not be read as such.
Not a routing graph. CONNECTS is undirected adjacency along a way; oneway is carried as a road attribute but not applied to edge direction. Do not compute drive-time or turn-restricted routes from it without handling that yourself.
No capacity, no traffic, no population. Evacuation modelling needs demand and capacity; this is the topology only.
No geometry beyond junction lat/lon — road shape between junctions is not preserved separately from the junction sequence.
OSM coverage is contributor-dependent; completeness varies.

Citation

Wildfire Evacuation Road Network — Paradise, California
(13,787 nodes, 26,831 edges).
https://huggingface.co/datasets/VaidhyaMegha/wildfire-evac-kg
Source: OpenStreetMap via the Overpass API, OSM base 2026-05-11T16:29:15Z.
© OpenStreetMap contributors, licensed under ODbL 1.0
(https://opendatacommons.org/licenses/odbl/1-0/).
Extract repository: https://github.com/samyama-ai/wildfire-evac-kg
Graph form derived from the extract for this publication.