Helsinki City Bikes
Station activity, trip flows and network resilience in April 2021.
Overview
Helsinki City Bikes models stations as nodes and observed April 2021 journeys as directed connections: 258,514 cleaned trips, 351 stations and 32,815 directed connections. It asks where station activity concentrates, which stations gain or lose bikes, and how much of the network stays connected when stations are removed.
What I built
- An audit of the cleaned April 2021 cohort against the research repository's graph aggregates.
- A trip graph with weighted directed connections between origin and destination stations.
- A station map of activity, trip flows and net arrivals.
- A resilience simulation comparing random, degree-targeted and betweenness-targeted station removal.
- A comparison against 100 degree-preserving rewired networks.
Results
- Connectivity falls below 10% after 316 random removals (mean), 294 degree-targeted and 284 betweenness-targeted.
- In all four tested statistics, the observed value lies outside the entire range of the 100 degree-preserving rewirings.
- Net arrivals differ by station: Kaivopuisto +213, Pasilan asema −548.
Station map
Switch the view, filter by day type or clock hour, and search or click a station to see its arrivals, departures and busiest routes.
Where station events concentrate
Departures by clock hour · click a bar to filter the map
Route layer requires interactive evidence.
Station positions on OpenFreeMap Dark, using OpenStreetMap data. Basemap requires WebGL and an internet connection. In light mode the basemap colours are inverted.
Interactive charts need JavaScript and the evidence file.
Analysis
Station activity, network structure, communities and resilience, computed from the same April 2021 trip graph.
Ten most active stations
Activity by station
Arrivals Departures
Net arrivals at the same stations
More departures More arrivals
Net arrivals by time of day
Weekday morning 07:00–10:00 (3 hours) · evening 15:00–19:00 (4 hours).
| Weekday window | Arrivals | Departures | Net events | Net / weekday / hour |
|---|
Concentration of trips in station pairs
Horizontal = share of ranked pairs · vertical = cumulative share of trips
K-core depth
Each bar is one k-core level: stations that survive when every station with fewer than k links is removed.
Long shoulder (CCDF)
Busy vs in-between
Five Louvain communities
Trips between communities
Rows are the origin community, columns the destination. Shading follows the trip count.
Community stability
How stable the five are
| Community | Stations | Trips within | Share of all trips |
|---|
Removal connectivity
Trip exposure
Largest component / 351 stations Incident trips / 258,514 trips
Null-model comparison
The observed value lies outside the entire range of 100 rewirings in all four tested statistics.
Limits
- One month cannot establish seasonal patterns or causal effects.
- Connectivity under node removal is not predicted trip loss.
- Arrivals and departures do not measure station inventory or capacity.