Canberra has invested heavily in public transport, yet ridership remains low. Data suggest that route coverage and reliability are reasonable, but weekend frequency remains a major service gap. This analysis examines how Canberra's public transport service quality (coverage, frequency, reliability) likely falls short of demand and aims to identify which areas stand to benefit most from improvements.
For public transport to be competitive with private transport, users should not have to walk far to find a stop. Transport planners use a convention of 600m as the upper bound of acceptable distance to walk to a transit stop. In Canberra, 94% of residents live within 600m of a public transport stop, with a median distance of 289m (well within the recommended walking distance).
This choropleth map shows ACT Statistical Area Level 1s (SA1)[1] colour-coded by population density (light pink to dark red), with transport stops: bus stops (blue circles) and light rail stops (gold squares). Use the selector above to switch between districts. Hover over SA1s to show the same SA1 in the chart to the right. Hovering over SA1s will also show a tooltip with the suburb name (SA2), district (SA3), population, and density data. Hover over stops to see stop names and details. The pop-out map on the right shows where that district is located within Canberra.
Data: ABS ASGS 2021 SA1 boundaries + Census G01 (ABS downloads), Transport Canberra GTFS static feed
This scatter plot shows the relationship between SA1 population density and distance to the nearest bus stop for each SA1 in Canberra. This chart only shows SA1s with population density above 1,000 persons per km^2. This filtering was done to exclude a number of SA1s with very small populations. Use the selector to change from straight-line distance to walking distance [2]Source: Transit Capacity and Quality Service Manual.. Walking distance scaling (x1.35) was applied to the distance from the centroid of the SA1 to the nearest stop in lieu of actual walking distance data.
Data: ABS ASGS 2021 SA1 boundaries + Census G01, Transport Canberra GTFS static feed | Processing: process_sa1_distance.py, build_transit_stops.py
For public transport to offer a viable alternative to private transport, people should not have to wait too long between services. Cities with high public transport use often aim for headways of less than 5 minutes, so people do not need to plan their lives around bus timetables. Only two of Canberra's bus services reach this benchmark during peak commuting hours. While the light rail and other 'rapid' services (Routes 2-9) have scheduled headways of less than 10 minutes, many suburban feeder routes have headways of more than 15 minutes. In a relatively compact city, 15-minute headways dramatically reduce the competitiveness of public transport relative to private transport.
The system becomes much less usable on weekends, when headways for most rapid routes stretch out to 15 minutes and most suburban feeder routes have headways of more than 30 minutes.
Interactive map showing all bus and light rail routes across Canberra, colour-coded by service frequency (mean headway). Use the service period selector to view frequency for weekday peak, weekday off-peak, Saturday, or Sunday. Use the route selector to highlight individual routes. Hover over route path to highlight route in chart to the right.
Data: Transport Canberra GTFS static feed (API registration required) | Processing: download_gtfs_static.py, create_route_shapes_geojson.py
Mean headways for Canberra bus and light rail routes, with filtering by service type (peak/off-peak/weekend).
Data: Transport Canberra GTFS static feed | Processing: extract_headways_from_gtfs.py
Reliability can be just as important as frequency, particularly for lower-frequency routes. Missing a suburban feeder route because it ran 5 minutes earlier than scheduled could make you 20 minutes late to work. In this way, poor frequency compounds large headways to make public transport less competitive with private transport. The light rail is by far the most reliable service in Canberra's network, with 95% of arrivals happening within +/- 1 minute of schedule. Canberra's other rapid routes do not fare as well, with 84% of arrivals more than 3 minutes ahead of or behind schedule.
Faceted histograms showing the distribution of arrival time deviations (scheduled vs actual) for Canberra's light rail and Rapid bus routes. Each panel represents one route, with deviations binned into 1-minute intervals. The dashed vertical line at 0 represents perfect on-time performance. Data collected from ACT GTFS-Realtime API over a two-day period (23-24 December 2025).
Data: Transport Canberra GTFS-Realtime API, polled every 60 seconds over a two-day period (API registration required) | Processing: get_reliability_data.py
Stacked bar chart showing the percentage breakdown of arrival reliability for each route.
Data: Transport Canberra GTFS-Realtime API, polled every 60 seconds over a two-day period | Processing: create_ontime_percentage_data.py
Coverage, frequency, and reliability are not the only determinants of public transport use. For example, a university student is often more likely to use public transport than a high-income retiree. Looking at these demographic variables, we can identify which areas of Canberra use more or less public transport than predicted by their demographics. If the ACT government wants to increase public transport use, it should focus on areas that have strong demographic predictors of public transport use and poor service quality.
Using machine learning techniques on census data, we can identify SA1s with higher or lower public transport use than expected. Comparing this with a service quality score reveals 206 "undersupplied" SA1s covering 90,425 residents, with a high propensity to use public transport but low service quality. In the chart on the right, the 'Urban core - well served' cluster demonstrates what's possible. This cluster has similar demographics to the 'Undersupplied Urban' group but 19% higher public transport use, facilitated by higher service levels.
This chart estimates demographic propensity for PT use using census variables.
Method: A single Gradient Boosting model trained on demographic variables (occupation, age, labour force participation, education, income, population density). The target is public transport commute rate from the census, using the same train-test split across SA1s.
Findings: Demographics alone explain a reasonable share of the variation in public transport commuting (R²=0.399). The scatter shows wide dispersion around the diagonal, indicating that demographic characteristics are only a partial predictor of PT use across SA1s.
Data: ABS Census 2021 (7 tables: G01, G02, G13, G16, G17B, G43, G51) + SA1 boundaries | Processing: ml_pt_commuting_rate.py
This chart takes the public transport use propensity score from the chart on the left and compares it to public transport quality for each SA1.
Method: Using k-means clustering (k=4), SA1s are grouped based on two composite scores: (1) Demand Score - based on demographic indicators of public transport propensity, and (2) Supply Score - service quality indicators (proximity to closest stop, service frequency, departures per capita, service consistency ratios). Areas are clustered into four policy-relevant groups.
Findings: The clustering reveals 206 "Undersupplied Urban" SA1 areas (red points) serving 90,425 residents who have high demographic propensity scores but receive below-average service quality.
Data: ABS Census 2021 + Transport Canberra GTFS + SA1 boundaries | Processing: create_supply_scores.py, clustering_demand_supply.py
This analysis reveals that Canberra's public transport network is built primarily around weekday peak commuting and is not fit to replace private transport for most people. Coverage is the strongest point in the network, with the vast majority of people living within a reasonable walk from a stop. Feeder route frequency (particularly on weekends) is the clearest weak point in the network and likely the strongest barrier to broader use.
The ACT government's public transport strategy in recent years has focused on extending the light rail network to Woden. Scheduled to open in 2033, this will replace the Civic to Woden segment of the rapid 4 bus. While this will likely improve reliability for this service, this corridor is already served by the highest-frequency bus services. If the goal is to increase ridership across the network, more investment should be put into improving service frequency, particularly for feeder routes. The results of this analysis suggest that around a quarter of Canberra's population has demographic characteristics indicating higher public transport use if service quality improves. These areas should be prioritised for service improvements.
This analysis draws on data from a range of publicly available sources (see below). Some data (e.g., from the ABS) was relatively easy to work with (coming in CSV format) and mainly required hunting down the right files. The remainder of the data came from the Transport Canberra API. Signing up was easy enough, but sparse documentation and a small user community meant spending a fair bit of time getting it to work.
The reliability measure is the only dataset that is not directly replicable. This data was not published, but I was able to use the API and a Python script to collect service lateness data over a two-day period (23-24 December 2025). I acknowledge that the days leading up to Christmas may not be a representative sample, so I encourage others to collect more data over a longer period before drawing solid policy conclusions.
While others can use this script to collect new data, I should note that little time was spent optimising for the file size vs. granularity trade-off. The file size grew large quickly (47MB for 2 days). If this work were extended, more thought should be given to this (perhaps collecting data less frequently or compressing the data into summary statistics).
To test the hypothesis that improved service quality can induce greater public transport use more directly, ridership data for each service would be valuable. This could have been used to predict usage for each route based on proximity, frequency, and reliability measures. This data was previously published by the ACT Government but stopped in 2017.
Data Sources:
| Source | Description | Access Method | Update Frequency |
|---|---|---|---|
| ABS Census 2021 | Demographics, JTW, income, education (7 tables: G01, G02, G13, G16, G17B, G43, G51) | TableBuilder download (CSV) | Every 5 years |
| ABS ASGS 2021 | SA1 statistical boundaries (shapefiles, GDA2020) | Manual download (ZIP) | Annual |
| ACT GTFS Static | Bus routes, stops, timetables (GTFS format) |
Authenticated API (download_gtfs_static.py)
|
Weekly (automated) |
| ACT GTFS-Realtime | Live bus arrival predictions (Protobuf format) |
Polling API (get_reliability_data.py, 60s intervals)
|
Continuous (real-time) |
| ACT Open Data Portal | Light rail stops and route geometry (GeoJSON) | Manual download | As updated |
Coverage charts: One of the major data cleaning operations was trimming down the list of SA1s. The ACT is relatively small, but Canberra itself is even smaller. This made maps hard to read because the Canberra segment was relatively small (and Vega-Lite does not allow zoom and panning). As the ABS dataset included SA1s for all of the ACT, I manually filtered out SA1s that were outside of Canberra's urban boundaries. The populations affected are very small (11 SA1s with <50 total residents) and ACT transport does not serve these areas. To extend this project, you could explore creating the maps in a different language that allows zooming and panning. This could be a more intuitive experience for readers than drop-down menus switching between areas.
The distance to nearest stop measure was generated by a Python script that took the centre point of each SA1 and measured the straight-line distance to the nearest GTFS stop. Acknowledging that people cannot walk as the crow flies, I then scaled the distance up by x1.35 as advised in the Transit Capacity and Quality Service Manual. To extend this work further, you could improve the accuracy of this measure by using a pedestrian network graph.
Frequency charts: I extracted route shapes and stop times from the GTFS static feed, then computed mean headways by route for weekday peak/off-peak and Saturday/Sunday windows. Light rail headways were added from published timetable data because they are not in the GTFS feed.
Reliability charts: This data was collected from the GTFS feed with a Python script. One of the issues with collecting this data was the need to run a script continuously over a long period. This meant keeping my laptop running for 2 days without sleeping. This took a little while to get up and running because I had some issues with my script not saving the data properly. As the collection window was overnight (London time), troubleshooting was a multi-day affair. To extend this work, it would be good to run the script on a server so it could run for a longer, more representative period.
Machine learning charts: I merged ABS Census SA1 demographics with service-quality features (distance to stops, headways/departures, and reliability) and trained gradient boosting models to compare demographics-only, service-only, and combined predictors of PT commuting. I also used k-means clustering to group routes and SA1s into policy-relevant service types.
Coding Stack:
Generative AI (Claude, ChatGPT, Gemini, Copilot) was used in producing this website. These tools were used to troubleshoot code and help produce some of the more complicated Python scripts.