Methodology

Pipeline Overview

flowchart LR
rawGtfs[SwissGTFSStatic] --> swissFilter[SwissBoundaryFilter]
swissFilter --> railFilter[RailRouteFilter]
railFilter --> stationDedup[StationDedup]
rawPois[OSMAndRetailPOIs] --> poiCleaning[POICleaningAndBrandNormalization]
stationDedup --> stationMeta[LogicalRailStations]
stationMeta --> tripGraph[TimeExpandedRailGraph]
tripGraph --> reachability[ModifiedDijkstra]
poiCleaning --> nearestJoin[NearestStationAssignment]
stationMeta --> nearestJoin
reachability --> storyMetrics[30MinAnd60MinStoryMetrics]
nearestJoin --> storyMetrics
poiCleaning --> densityMaps[RetailDensityMaps]
storyMetrics --> docsOutput[StaticFiguresAndDocs]
densityMaps --> docsOutput

Data Sources

SwissReach currently combines three source families:

• Swiss GTFS static timetable for the transport network
• Swiss national boundary geometry from swisstopo
• OpenStreetMap-derived POIs for schools, hospitals, IKEA, and supermarkets

The supermarket layer is exported separately and normalized in data/scripts/fetch_swiss_supermarkets.py. The insight figures and docs assets are then generated in data/scripts/export_storytelling_assets.py.

Rail Preprocessing

The preprocessing pipeline follows four steps:

1. Load the GTFS static feed and Swiss national boundary.
2. Keep only stops located inside Switzerland.
3. Filter routes and trips to rail services.
4. Collapse platform-level stops into logical stations using parent_station when available.

This design reduces noise while preserving a national view of the rail network.

Reachability Model

The reachability engine builds a time-expanded graph from GTFS stop times and computes earliest arrival times from a chosen origin station. For the insight export, the project fixes departure time at 08:00 and derives two horizons:

• 30 minutes for daily amenities
• 60 minutes for IKEA access

Key assumptions in the current model:

• static timetable, no live delays
• rail-only services
• fixed minimum transfer dwell of 3 minutes
• staying on the same train is free
• output is reachability and travel time, not full itinerary explanation

The resulting model is well suited to:

• isochrone-style maps
• comparing origins
• comparing departure times
• producing front-end friendly station-level outputs

POI Integration

SwissReach does not run full door-to-door routing to every supermarket, school, or hospital. Instead, POIs are first assigned to the nearest logical rail station in projected Swiss coordinates (EPSG:2056).

This gives two useful abstractions:

1. a count of amenities "attached" to each station
2. a simple way to aggregate reachable amenities after the timetable search is complete

The three analytical layers use this mechanism as follows:

• supermarkets_30min: number of supermarket POIs attached to stations reachable within 30 minutes
• schools_30min: number of school POIs attached to stations reachable within 30 minutes
• hospitals_30min: number of hospital POIs attached to stations reachable within 30 minutes
• ikea_60min: number of IKEA POIs attached to stations reachable within 60 minutes

Retail Dataset And Density Maps

The supermarket export script builds a normalized food-retail dataset with three layers:

• supermarket_core
• supermarket_secondary
• convenience

For the docs insights page, the Migros and Coop subsets are exported again in a minimal JSON format and visualized with spatial density plots. This density view is not a reachability metric; it is a retail geography comparison layer.

Routing Algorithm Choice

Milestone 1 does not require a more complex routing algorithm.

The present task is a feasibility study for reachability visualization, not a production journey planner. For this use case, the current modified Dijkstra approach remains appropriate because:

• the state space is already reduced to 1,938 logical rail stations
• the interaction focus is on earliest arrival and coverage
• the result can be cached and reused for repeated parameter choices

If the final project later shifts toward richer itinerary planning, likely upgrade paths include:

• path reconstruction on top of the current graph
• explicit walking edges
• RAPTOR or CSA-style public transport routing

Current Limitations

The current Milestone 1 prototype has several modeling boundaries:

• Rail-only: bus, tram, cable car, and boat access are excluded.
• No explicit walking network: transfer time is represented by a fixed dwell constant, not by station geometry or street routing.
• Nearest-station POI proxy: amenities are attached to stations rather than reached through a full last-mile walk model.
• Swiss-only clipping: border stations are kept only insofar as the stop lies inside Switzerland, so some real cross-border accessibility is understated.
• Station-level deduplication: excellent for map readability, but not detailed enough for precise platform-by-platform transfer modeling.
• OSM completeness: school, hospital, IKEA, and retail coverage depends on community-maintained POI quality.

Future Extensions

Beyond Milestone 1, the most realistic next steps are the following:

1. Add transfers.txt and, where available, pathways.txt to improve transfer realism.
2. Add a limited walking layer using nearby-stop links or an OpenStreetMap pedestrian graph.
3. Replace nearest-station amenity attachment with explicit last-mile travel for selected cities.
4. Add interactive filters for amenity type, retail brand, and time horizon in the published site.