Boris Bike usage in London during the coronavirus lockdown
Network analysis in the time of a global pandemic
Click here for update as of 9 August 2020 with the latest TfL data
Data around the extent of public travel and the transportation methods used has been closely followed during the coronavirus pandemic — for example, featuring in the UK government’s briefing each day and Citymapper’s Mobility Index. As far as I am aware — no one has looked at Boris Bike usage.
Transport for London (TfL) — through their Unified API and their open data — offer a comprehensive set of data on all journeys taken using its Santander Cycles public bike hire scheme in London, the so called ‘Boris Bikes’. In this article, I look to identify behavioural changes during the coronavirus lockdown about how people are using these bikes. All data is the author’s analysis of this TfL data.
What do we know about each journey?
In the TfL data, you can see the following details about every single completed journey:
I map the station Ids to longitude/latitude coordinates using the TfL Unified API. Whilst the data is very comprehensive, there is unfortunately a lag in availability — the latest data available is up to the end of April 2020.
How are Boris Bikes used during ‘normal periods’?
Firstly, using all journeys from 2019, we need to understand the ‘normal’ usage of Boris Bikes. The left hand chart shows a clear distinction in behaviour between weekdays and weekends. Cycle hire is bimodal on weekdays — there are two clear peaks in usage around the start and end of the working day when the bikes are used for commuting. At weekends, hire is unimodal — bikes are used for leisure journeys or exercise, with hire peaking in the mid-afternoon. Given London’s weather, it will come as no surprise that there is a lot of seasonality in the data, as seen in the right hand chart. Usage is significantly higher in the summer months (there are almost twice as many journeys in July than in December), but also the proportion of journeys taken at the weekend increases in the summer as people cycle for leisure.
How has behaviour changed during the coronavirus lockdown?
The UK government imposed a lockdown effective from Tuesday 24 March (week number 13 of the year). The below chart compares the number of journeys on a week by week basis in 2020 (dark blue) vs 2019 (light blue) with the red line showing the 2020 weekly usage as a percentage of the 2019 usage. Whilst journey frequency should ordinarily have been increasing (as the weather improved), there was a sharp drop in usage starting in week 12 (the week prior to the lockdown, when the government advised against socialising) and bottoming out in week 13, after the lockdown was imposed. Interestingly, usage has sharply rebounded and was up to 85% of 2019 levels by the end of April. This is in contrast to tube, bus and road use (as evidenced in the government daily briefings).
Has the type of usage changed since the lockdown began?
Prior to the lockdown, usage could be summarised as predominantly driven by commuting on weekdays and leisure/exercise at the weekends, with two quite distinctive regimes of hire. The left hand chart below shows no distinction during the lockdown — every day looks like the weekend. There is still a small weekday peak during morning commuting hours as some people are still travelling to work. The weekday peak in usage is later in the day (5–7pm) compared to the weekend (2–4pm). This suggests that people are cycling after work for their once-a-day exercise.
The clearest indication that a journey is purely for leisure or exercise is if you start and end at the same docking station. The right hand chart below shows the frequency of such journeys, with a 5x increase.
Given this change in usage, are people going to different places?
This data can be considered as a nice application of graph theory, with each journey representing a vertex and each docking station representing a node. The below charts map each docking station as a red dot, and each journey as a blue line (only the most popular 4,000 unique routes are displayed). The size of the docking station node corresponds to the number of journeys starting or ending at that station; this is scaled so that the most popular docking station has the same size node on both charts. A rough shape of the River Thames is shown (although it is not entirely accurate).
Before the lockdown, you can clearly see the focus around parts of central London — in particular two big concentrations of activity around King’s Cross and Waterloo, representing major hubs in the transport network. Whilst the busiest docking stations are in these areas, no individual docking station dominates, being constrained by the fact that a single docking station has a limited number of bikes/docks but also due to the higher concentration of docking stations in these areas.
By contrast, during the lockdown, there is a much greater dispersion of activity across the network. The central London transport hubs no longer dominate and usage is less concentrated at the most popular docking stations.
However — the lockdown has caused some docking stations to become extremely busy, particularly those around Hyde Park and along the river (Westminster, the South Bank). This can be seen by looking at the most popular stations and what percentage of journeys they account for.
By using the NetworkX package in Python, it is easy to calculate some graph theory metrics. For example, sizing the docking station nodes by their degree centrality (with the sizing of the nodes on the same scale between the two charts) emphasises the focus around Kings Cross and Waterloo before the lockdown, and the more uniform network usage during the lockdown. This package offers a lot of potential for more interesting analysis of the network.
Interesting next steps?
· Update the analysis as and when TfL release new data. It will be interesting to try to identify the extent to which people are returning to work through their bike hire usage
· Better graphing — it would be nice to have a map that was more accurate and more identifiably London…
· More NetworkX analysis!
Update as of 6 June 2020: TfL have released the next batch of data covering May 2020
Over the course of May 2020, the degree of lockdown was eased slightly and it was also a month of extremely nice weather in London. Usage of the Santander Cycles has continued to rise and total rides per week is now back to (and even through) 2019 levels.
Can the Santander Cycles data be used to understand behavioural changes as the lockdown has evolved? The following heatmaps show the hire frequency at different times of day (y-axis) through the course of the year (dates on the x-axis). The darker the colour, the more people are cycling at that time on that date.
The left hand chart shows the whole of 2019. Weekdays are characterised by the peak usage during commuting hours (as discussed above), whilst weekend usage is more spread out over the middle of the day. Activity is predictably greater during the summer months and the long evenings of May-July encourage evening usage.
The right hand chart shows 2020 to date. Usage was very similar to 2019 up until the lockdown was imposed. There is a clear decrease in activity during the second half of March (as discussed above) and cycle hire has since recovered over subsequent weeks. As of the end of May, there is no indication of a normal ‘9-to-5 commuting pattern’ returning, with concentrated activity at the start and end of the day. Leisure cycling still seems to dominate.
It will be interesting to monitor this data as the weeks progress, to try and observe a broader return to the office.
Update as of 9 August 2020: TfL release the next batch of data covering June and July 2020
After a long delay, TfL have finally released the latest batch of data covering the period up until 4 August. Over this period, the lockdown has continued to ease, the hospitality sector has begun to open up, the WFH guidance has been loosened and we are now Eating Out To Help Out. How have these changes been reflected in the Santander Cycles usage?
At a high level, usage has remained high and is very comparable to 2019 levels.
However, an update of the heatmap (lefthand chart) shows that leisure cycling appears to still dominate, with no real sign of a normal ‘commuting pattern’ returning.
The righthand chart makes this more explicit; this chart compares the frequency of journeys taken on a weekday between 8–9am by week for 2019 vs 2020. Historically, this hour is the peak period for Santander Cycles activity as commuters cycle to the office. Despite commuters being encouraged to cycle to work instead of using public transport, journeys during this commuting period are at less than 50% of the corresponding levels in 2019.
This latest data very much suggests that the return to the office has been slow to date.
