• Post category:Articles
  • Post published:16.3.2023

Fluent cycling in the city? – Measuring urban cycling quality with tracking data

How fluent is cycling in your city? Are you able to cycle without interruptions, or do you have to stop frequently? The ProGIS-awarded master’s thesis introduces a method for estimating the smoothness of cyclists’ traffic flow in urban areas.


You are currently viewing Fluent cycling in the city? – Measuring urban cycling quality with tracking data
CC: Visit Finland

As one of the most environmentally friendly modes of travel, cycling plays an important role in designing sustainable urban mobility. In an ideal, cycling-friendly city, the infrastructure should allow cyclists to travel fluently: continuously at a comfortable speed, undisturbed by other road users, and without having to stop and wait. 

More and more cyclists use activity tracking applications to record, monitor and share their rides. As a result, the data collected by the application providers is unprecedented in its spatio-temporal extent and the number of active contributors. 

This data makes extensive data-driven analyses of urban cycling possible. The results can help cities to develop a better understanding of the status quo and identify effective measures to make city cycling more convenient and attractive.

Calculating a fluency measure for street segments

My graduate thesis introduces a method for estimating urban cycling fluency using mobile tracking data. The study is based on more than 50 000 real-life tracks from the ‘Sports Tracker’ application. The tracks were recorded by close to 3 700 cyclists in the Helsinki area between the years 2010 and 2012. All tracks were publicly available on the application. 

In a pre-processing step, the tracks are smoothed and cleared from outliers.  Subsequently, the tracks are mapped to the street network. Each street is divided into segments of about 25 meters. For each segment a track passes, properties related to the cyclist’s dynamics and stopping behavior on the segment are calculated. 

If a segment is passed by at least ten different cyclists, these cyclists’ properties are aggregated into segment characteristics. These characteristics describe the following aspects:  

  1. How many of the cyclists stopped on the segment? 
  2. What was the average duration of the stops? 
  3. How high was the cyclists’ acceleration on average? 
  4. How did the cyclists’ speed on this segment compare to their average speed? 

These characteristics are the building blocks of the final cycling fluency index. Each segment receives an index value between 0 and 1. High index values indicate fluent cycling conditions: relatively high speed, little acceleration, no stops with long waiting times. Low index values imply obstructions on the segment. 

Photo: Overview of the method: tracking data is processed along with street network data to obtain an estimate of cycling fluency.

A promising estimation from imperfect data

Through experiments with different secondary data, it became clear that the results describe the on-street circumstances in and around Helsinki sufficiently well. The cyclists’ stops correlated clearly with the locations of traffic lights and intersections. Moreover, the characteristics describing the cyclists’ movement correlated with the properties of single tracks that were not included in the original dataset. This suggests that the aggregated values can predict, to some extent, how fluently an individual cyclist will be able to cycle. 

However, the data can only show what has been, not what is. Additionally, deliberate choices of cyclists, such as stopping to take a picture, are difficult to detect and can distort the results. 

Furthermore, the data is biased due to participation inequality, as 10% of the cyclists in the dataset account for 65% of the tracks. Confident, enthusiastic cyclists are most likely significantly overrepresented in the data. 

Challenges like these are unavoidable when working with data recorded in an uncontrolled setting. Irregularities of individual tracks, however, are mitigated when a few tens of tracks or more are aggregated to calculate the cycling fluency index. 

Photo: Fluency of three different cycling routes between Otaniemi and Kivenlahti in Espoo. Route C is the shortest and Route A the fastest according to the dataset, but Route B offers the most fluent cycling experience.

A cyclist-centered perspective of the city

The method described here is one possibility to condense mobile tracking data into an indicator of urban cycling quality. It could support the design and monitoring of urban cycling infrastructure, as it offers a cyclist-centered perspective on the accessibility of cycling in the city. 

Experiments with the cycling fluency index as a routing criterion showed that fluency-based routing tends to produce routes resembling the fastest route. At the same time, fluency-based routing tends to include detours on popular cycling infrastructure, even though the index does not possess a notion of popularity. Thus, there is a potential for the measure to be utilized for developing advanced, data-driven routing algorithms for cyclists.

Anna Brauer works as a scientist in the Finnish Geospatial Research Institute (FGI), National Land Survey of Finland (NLS), and prepares her doctoral thesis in the University of Helsinki. Brauer got a M.Sc.(Tech.) degree from the Dresden University of Technology in 2020 and received the ProGIS master’s thesis award from her diploma thesis “Characterizing cycling traffic fluency using big mobile activity tracking data”. The work was supervised by Prof. Lars Bernard, Dresden University of Technology, Prof. Juha Oksanen (FGI, NLS) and research group leader Dr. Ville Mäkinen, (FGI, NLS). email: anna.brauer(at)helsinki.fi