Guelph Sidewalks Layer

Turns the City of Guelph Sidewalks layer (~5,245 lines, 716 km built) into map-tile layers that OpenStreetMap mappers can add to the JOSM and iD editors as a reference overlay, and a weekly page listing the City sidewalks OSM has no line for, street by street.

Live layer and how to add it: https://guelph-maps.github.io/guelph-sidewalks-layer/

Part of the guelph-maps organisation, which indexes every Guelph project. Cloned from guelph-buildings-layer, itself cloned from guelph-parks-layer. This is the family's first line layer: the others match points or polygons.

Why this layer

An audit of the City's open data for OSM rated sidewalks low because "OSM has 92%". That figure divided OSM's total sidewalk length by the City's (663 / 716 km). It never asked whether they are the same sidewalks, and 42 km of OSM's are on campus and private sites the City does not inventory, which pads the ratio. Measured by location, on the first run here (2026-10-09):

• 70.5 km of the City's 715.8 km of built sidewalk has no OSM pedestrian line within 15 m and no road tag claiming one; 50.7 km has nothing within 50 m.
• It runs along streets whose roads are mapped. 100 streets have at least 0.2 km each, and the worst cluster by neighbourhood: Kortright Hills (Deerpath Dr 3.2 km, Woodland Glen Dr 2.2, Pinnacle Cres 1.8), then West Acres Dr 2.7 and Freshmeadow Way 1.9.
• The City states 1–5 m accuracy on 5,241 of its 5,245 rows, and 8.8 km of the untraced length sits 15–25 m from an OSM line, where an offset is a likelier story than a missing way. So the honest headline is ~60–75 km.

Guelph already maps sidewalks as separate ways (4,512 footway=sidewalk ways, against 237 roads with a positive sidewalk=*), so tracing them follows the local convention; there is no tagging question to settle first.

What it produces

• Vector tiles (MVT), z14–19 — interactive in iD; click a line to read its id (the City's SIDEWALKID), status, surface, width (m), installed (year, on about a third of rows) and length (m).

• Raster tiles (PNG), z15–18 — cyan lines for built sidewalks, orange for the City's proposed ones. A tracing backdrop over aerial imagery in JOSM.

• A landing page with copy-paste "add this layer" instructions for both editors.

• A gap-review page (/gaps/) that leads with streets, by untraced length, and opens each into its segments on an interactive map. Segment statuses:

  • missing — under 20% covered: add it;
  • partial — 20–70% covered: finish it;
  • far — an OSM line 15–25 m away: check the alignment before adding anything;
  • road_says_no — the OSM road says sidewalk=no where the City has one (2.7 km, most of it on Whitelaw Rd);
  • untagged — covered by a plain highway=footway: tag it footway=sidewalk;
  • off by default: on_road_tag (mapped as sidewalk=* on the road, a valid way to map it), proposed (City status PROPOSED, may not be poured), nosurface (a matched sidewalk way with no surface=*; the City's are 99% concrete, so low value).

  Its history page charts the untraced length and the counts, one point per weekly run.

All of it is published to GitHub Pages and rebuilt weekly.

What is kept

Every line, built or proposed. The slim step keeps six properties (see src/slim.py): SIDEWALKID as id, STATUS as status (built, or proposed for PROPOSED and REQUIRED), SURFTYPE lower-cased as surface, the metres out of WIDTH, the year of INSTALLDATE, and Shape__Length rounded. Condition, inspection and plow-route fields are dropped: they are maintenance records, not anything a mapper can check.

How the gap page matches

Overpass returns every road and every pedestrian way in the Guelph bbox (one query, out tags geom, ~30,000 ways). Each City sidewalk is cut into pieces of ~5 m and each piece's midpoint asks, in this order:

1. a footway=sidewalk way within 15 m — covered;
2. any other pedestrian line within 15 m (a plain footway, path, cycleway, a crossing) — covered;
3. the nearest road within 20 m with a positive sidewalk=* — covered;
4. that road saying sidewalk=no — a contradiction;
5. a pedestrian line 15–25 m away — far;
6. nothing.

A segment is matched at ≥70% covered (by length), partial at 20–70%, and a gap below that, named by what most of its uncovered length has near it. The full rule is in src/compare.py's docstring. Matching runs in UTM 17N metres with shapely's STRtree; a full compare takes ~10 s.

Each gap segment is named after the OSM road nearest most of its untraced length, within 40 m. The audit named them after the City's street centrelines instead, which is why it counted 97 streets with ≥0.2 km and this counts 100; the ranking is the same.

The first run reproduced the audit's recalibration exactly: 362 segments under 20% covered (318 missing, 23 far, 21 road_says_no), and 67.8 km untraced once road_says_no is set aside (the recalibration: 67.7).

It runs one way only. An OSM sidewalk with no City sidewalk beside it is not reported: campus, private sites and plazas are in OSM and not in the City's inventory, and that is not a gap.

The Overpass handling (mirror list, retries, refusing stale or implausibly small replies, the cache fallback, exit code 70) is the parks layer's; its README explains each piece. One tightening: a reply that carries no timestamp_osm_base at all, or one with no time zone, is refused too, rather than dated today. OSM_MIN_ELEMENTS in src/config.py is 15,000, about half the first live reply (30,494 ways on 2026-10-09).

The City's service caps a reply at 1,000 rows, so unlike its siblings the download pages through the layer, and refuses to save if the total differs from the service's own count.

There is no backfill command; the history starts with the first weekly build.

Setup

1. Install Python dependencies:

   pip install -r requirements.txt
   

   addressvault is the local link-gate package the sibling projects share; it is not on PyPI. Unlike its siblings this project needs shapely 2, numpy and pyproj: matching ~143,000 five-metre samples against ~30,000 OSM ways wants a spatial index and a metric projection, which the polygon siblings could hand-roll and a line matcher cannot cheaply.
2. On Windows, set up WSL2 + tippecanoe once — see wsl-setup.md. Elsewhere, install tippecanoe (2.49+) and the vector step runs it directly.

Usage

python run.py download   # fetch the Sidewalks layer, page by page (smart-cached)
python run.py slim       # reduce to a slim GeoJSONL
python run.py compare    # diff City sidewalks against OSM -> data/gaps.geojson
python run.py vector     # build vector (MVT) tiles via tippecanoe
python run.py raster     # build line raster (PNG) tiles
python run.py site       # render the landing page (+ the gap-review page)
python run.py publish    # force-push the site to the gh-pages branch

python run.py build      # download + slim + compare + vector + raster + site
python run.py update     # build + publish  (the scheduled entry point)

Build output lands in build/site/; that directory is what gets published. A full build on 2026-10-09: 15,394 vector tiles in ~2 min, 4,964 raster tiles in ~20 s.

Scheduling (Windows)

.\schedule-add.ps1       # registers a weekly task "kk-GuelphSidewalksLayer", Saturdays 16:30
.\schedule-remove.ps1    # unregisters it

Tests

pytest

Licence / attribution

Sidewalk data is © City of Guelph, published under the Open Government Licence – City of Guelph. The OSMF Licence Working Group approved that licence as compatible on 2024-09-09.

This project redistributes City data — as published tiles on a public URL — and the licence attaches a required attribution to that. The landing page, the gap page and the history page all carry, visibly:

Contains information licensed under the Open Government Licence – City of Guelph.

linked to the licence text. The same attribution is embedded in the vector tile metadata. This is a condition of use, not a courtesy: if you fork this, the notice travels with it.

Being able to trace from this layer is not the same as being cleared to bulk-import it. Tracing sidewalks one at a time into OSM, by a mapper checking imagery, is ordinary mapping with an approved source. A scripted import of the missing segments would be an import, and would need its own proposal under the import guidelines.