Accessibility-using-Transit

Compute public-transport accessibility and equity for any city from open GTFS data - a modern, self-contained successor to my M.Sc. thesis work (public-transport-analysis, built on the POLITO/CityChrone codebase), implementing the methodology of my first-author papers:

• Assessing Transportation Accessibility Equity via Open Data - hEART 2022 (arXiv:2206.09037)
• Equity Scores for Public Transit Lines from Open Data and Accessibility Measures - TRB 2023 (arXiv:2210.00128)

Live demo: https://abadeanlou.com/accessibility/ - interactive accessibility maps for Torino, Milano, and Paris produced by this pipeline. The Torino maps live in this repository; the other cities' exports are large and are hosted server-side.

Pipeline

GTFS zip --> MongoDB (stops, trips, calendars)
         --> stop-to-stop edge list for the busiest service day
         --> hexagonal grid over the area of interest
         --> travel times (transit graph + OSRM walking legs)
         --> accessibility per hex cell:  P2P / P2POI / POI2P
         --> equity metrics (Library/equity.py): population-weighted
             Lorenz curves + Gini indices
         --> interactive Folium maps (Maps/)

• Accessibility_Calculation.ipynb - end-to-end driver notebook (Torino example; swap the GTFS zip and boundary to run any city).
• Library/ - the actual implementation (~3,700 lines): GTFS processing, grid construction, routing, accessibility kernels (Numba-optimised), and the equity module.
• Maps/ - self-contained interactive outputs.

Equity metrics

Accessibility is distributed over people, not places. Library/equity.py weights each cell's accessibility by its population and computes Lorenz curves plus the standard inequality-index family - Gini, Theil (decomposable by district), Atkinson (explicit inequality-aversion parameter), and the Palma ratio (top-10% vs bottom-40% share) - the same lens the papers use to compare how fairly transit serves a city. Pure NumPy, unit-tested standalone:

pip install numpy pytest
pytest tests -v

Equity results (published on the live demo)

Two population-weighted views, Lorenz curves and caveats on the demo page, full numbers in data/equity_results.json:

• Whole-city reach (P2P): per-hex average travel time recovered out of the map exports (scripts/harvest_maps.py, 8:00 layer), accessibility = 1 / travel time.
• Essential services: how many schools, universities, healthcare places, supermarkets and markets (OpenStreetMap) each hex reaches by transit within 60 minutes at 8:00 - a cumulative-opportunities measure computed by scripts/reachable_pois.py from current GTFS feeds (busiest weekday, 15-min walk to stops, 5-min transfers, straight-line walking x1.3 detour). A self-contained successor to the original MongoDB + OSRM pipeline: the recomputed Milano surface matches the original research export with Spearman rho = 0.81 (scripts/validate_milano_counts.py).

City	View	Gini	Theil	Atkinson (e=0.5)	Palma
Torino	P2P - whole-city reach	0.087	0.012	0.006	0.36
Milano	P2P - whole-city reach	0.107	0.018	0.009	0.40
Paris	P2P - whole-city reach	0.081	0.010	0.005	0.36
Torino	P2POI - time to amenities	0.344	0.214	0.097	1.44
Milano	P2POI - time to amenities	0.414	0.275	0.138	1.78
Paris	P2POI - time to amenities	0.234	0.088	0.042	0.77
Torino	Essential services in 60 min	0.259	0.153	0.097	0.78
Milano	Essential services in 60 min	0.287	0.200	0.134	0.90
Paris	Essential services in 60 min	0.234	0.118	0.074	0.67

The headline finding: average reach to the whole city is spread almost evenly everywhere (Gini < 0.11, expected - the average is dominated by geography every resident shares), but access to essential services is far less equal, and the ranking is consistent on every index: Milano is the most unequal of the three, Paris the most equal.

Regenerate:

python scripts/harvest_maps.py Maps/accessibility_map_<City>_P2P.html data/hexes_<City>_P2P.csv
python scripts/fetch_pois_osm.py data/hexes_<City>_P2P.csv data/pois_<City>.csv
python scripts/reachable_pois.py <gtfs.zip> data/hexes_<City>_P2P.csv data/pois_<City>.csv data/numpoi_<City>.csv
python scripts/build_equity.py    # rewrites the equity section of Maps/index.html
python scripts/build_hex_maps.py  # compact interactive maps (P2P, P2POI, essential services)

Requirements (full pipeline)

Python 3.8+, MongoDB, OSRM (see osrm/ for the Docker setup), plus pip install -r requirements.txt. The full pipeline needs a running MongoDB and OSRM instance; the equity module and its tests do not.