Transport Reviews

ISSN: 0144-1647 (Print) 1464-5327 (Online) Journal homepage: http://www.tandfonline.com/loi/ttrv20

Big Data and Cycling

Gustavo Romanillos, Martin Zaltz Austwick, Dick Ettema & Joost De Kruijf

To cite this article: Gustavo Romanillos, Martin Zaltz Austwick, Dick Ettema &
Joost De Kruijf (2016) Big Data and Cycling, Transport Reviews, 36:1, 114-133, DOI:
10.1080/01441647.2015.1084067

To link to this article: http://dx.doi.org/10.1080/01441647.2015.1084067

Published online: 29 Sep 2015.

Submit your article to this journal

Article views: 533

View related articles

View Crossmark data

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=ttrv20

Download by: [University of London] Date: 02 July 2016, At: 03:08

 Transport Reviews, 2016
Vol. 36, No. 1, 114 –133, http://dx.doi.org/10.1080/01441647.2015.1084067

Big Data and Cycling

GUSTAVO ROMANILLOS∗ §, MARTIN ZALTZ AUSTWICK∗∗ ,

DICK ETTEMA† AND JOOST DE KRUIJF‡
∗
tGIS Transport, Infrastructure and Territory Research Group, Complutense University of Madrid,
Profesor Aranguren S/N, Ciudad Universitaria, 28040 Madrid, Spain; ∗ ∗ Bartlett Centre for
Advanced Spatial Analysis, University College London, Gower Street WC1E 6BT, London; †Faculty
of Geosciences, Utrecht University, PO Box 80115, 3508 TC Utrecht, The Netherlands; ††Faculty of
Downloaded by [University of London] at 03:08 02 July 2016

Geosciences, Utrecht University, Heidelberglaan 2, Room 621, 3584 CS Utrecht, The Netherlands
(Received 31 July 2015; revised 7 August 2015; accepted 13 August 2015)

ABSTRACT Big Data has begun to create significant impacts in urban and transport planning.
This paper covers the explosion in data-driven research on cycling, most of which has occurred in
the last ten years. We review the techniques, objectives and findings of a growing number of
studies we have classified into three groups according to the nature of the data they are based on:
GPS data (spatio-temporal data collected using the global positioning system (GPS)), live point
data and journey data. We discuss the movement from small-scale GPS studies to the ‘Big GPS’
data sets held by fitness and leisure apps or specific cycling initiatives, the impact of Bike Share Pro-
grammes (BSP) on the availability of timely point data and the potential of historical journey data for
trend analysis and pattern recognition. We conclude by pointing towards the possible new insights
through combining these data sets with each other – and with more conventional health, socio-demo-
graphic or transport data.

1. Introduction
Big Data holds the promise to illuminate social processes that were previously
undersampled or poorly understood. For those involved in city planning,
service provision and business intelligence, it still remains central to innovation
and research. The term arose first from the large-scale collective efforts of scien-
tists at the CERN (Conseil Européen pour la Recherche Nucléaire) particle accelerator,
large-scale astronomy and genomics projects (Marx, 2013) — but for more than
five years, the potential for working with large-scale social data has been
grasped by the commercial sector (Manyika, 2011) as well as governments and
non-governmental organisations (Hall, Shadbolt, Tiropanis, O’Hara, & Davies,
2012). Despite the excitement it has generated, working definitions of the term
are problematic. The most widely adopted framework derived from Laney
(2001) refers to the ‘3Vs’ of Big Data: Volume (size), Velocity (speed of generation
or collection) and Variety (synthesising a range of sources). Later authors (Kitchin,
2014) have added additional definitions to this (including ‘Veracity’, the quality of
§
Corresponding author. Email: gustavro@ucm.es

# 2015 Taylor & Francis

 Big Data and Cycling 115

the data — as a way to preserve the alliteration of the concept), but it seems
dubious that, in the wider world of Big Data, many data sources fully qualify
under all the categories of the 3Vs, or the wider definitions. Most of the data
sources discussed in this review qualify as Big Data under the first V (Volume),
but possibly not the others — many are single source (e.g. a transport provider
or single app or web platform, disqualifying them under the variety criterion)
and few provide large velocities of data in real time.
It perhaps makes sense to view the concept of Big Data as representing an
enthusiasm for the rapid expansion of data availability. Within these technologi-
cally driven definitions, there is no focus on openness or accessibility. While the
promise of innovation and new markets may motivate engineers and computer
scientists, it is the availability of data that has empowered and excited new
actors in policy, politics and governance. New data sets have become widely
accessible which capture the detail of processes that previously were estimated,
under sampled, kept private or simply poorly understood. In part, the Open
Downloaded by [University of London] at 03:08 02 July 2016

Data movement can be thanked for its hand in not only pushing an agenda of
transparency, but encouraging service providers and government departments
to provide usable data sets and streaming APIs (application programme inter-
faces) that third parties can use to create commercializable platforms and research
outputs. The topics of data released as a result of a movement towards open gov-
ernment data (OGD) arguably have antecedents in census and administrative
data, and the transparency agenda has driven the release of largely pre-existing
data sets (see, e.g. Coleman, 2013). However, the presence of technology as a
mechanism of automation and monitoring has generated new data sets with col-
lection methods which are distinct from centrally compiled or volunteered OGD.
This is particularly true in transport, where the automated systems for ticketing or
charging create a uniquely detailed data stream; however, this data stream has sig-
nificant enough privacy issues that it is not yet available in this detailed form.
Transport and geolocated data have quite an incredible capacity to de-pseudono-
nymise and reveal new information about individuals (for example, the work
done on open data around New York taxis to ‘stalk’ celebrities or identify the
homes of people who go to strip clubs Tockar, 2014), so there is a very clear ration-
ale for caution about open data release in this sphere.
Perhaps the most notable example of this data boom is the expansion of smart
card systems for public transport in major cities (Pelletier, Trépanier, & Morency,
2011) providing journey-level information for individual users, in systems that
were previously sampled by gate counts and travel-to-work questionnaires. The
quantum leap from limited to almost complete sampling is unprecedented, and
time slices of this data are available to researchers or developers through
service providers online (for example, Transport For London, 2014). Cycling sits
in a nexus where availability of Big Data (from quantified self-data, BSP, GPS
devices and mobile tracking) intersects with societal needs around fitness, sustain-
ability and air quality, and service provision and infrastructure planning for active
transport.
This review seeks to survey the Big Data sources available to cycling research-
ers, broadly split into GPS data, live point data and journey data. These data follow
different patterns of volume and velocity, suggesting different problem domains
and generating differing analysis approaches. GPS data are collected via smart-
phone, embedded devices or specialised units. These are usually collected by indi-
vidual users within the context of a quantified lifestyle (using fitness, health and
 116 G. Romanillos et al.

leisure apps), or contributing to a specific study. While this could be shared and
acted upon in real time, in many cases, users will upload their route at the end
of a journey or at the end of the day, putting it in the category of historical data.
These provide a high level of data density. Typical GPS data are sampled every
few (three to five) seconds, generating hundreds of data points per individual
journey, and depending on the sample period, thousands per user, and hundreds
of thousands or millions in a typical GPS study (e.g. Hood, Sall, & Charlton 2011).
In the case of fitness apps and social media-driven systems, this can number tens
of millions of users and routes (Endomondo, 2013; Map My Ride, 2014). Working
with GPS data poses some challenges with respect to accuracy (Schuessler &
Axhausen, 2009a) and volume, but it has also been one of the more fruitful in
terms of the application of models which can link directly to transport planning
policy on a city or county level.
Point data refer to information collected at a particular location — an example of
this is the information provided by a docking station in a BSP (Froehlich,
Downloaded by [University of London] at 03:08 02 July 2016

Neumann, & Oliver, 2009), or the data transmitted by a traffic camera or gate
counter at a specific intersection (Rogers & Papanikolopulos, 2000). This tends
to be smaller in volume, but the increasing availability of these data is starting
to allow some extensive insights. For example, the research conducted by
O’Brien, Cheshire, and Batty (2013) analysed 38 BSP located in Europe, Asia,
the Middle East, Australia and the Americas. Furthermore, through web APIs,
BSP can provide information in real time for immediate analysis and response.
The rich spatio-temporal characteristics of this data have led to some novel appli-
cations of cluster analyses.
Journey data act at a coarser level than GPS data — providing origin and desti-
nation locations and times for individual journey, but not necessarily including
detailed information about route choice, detailed link speed and delays. A
number of bikeshare programmes (BSP) have released journey data covering a
period of months, often amounting to several million journeys, but at present,
with some exceptions such as the Capital Bike Share initiative (2015), these data
are released months after the fact, making it more amenable to long-term trend
analysis than nowcasting or rapid response. The origin – destination data sets
allow for space – time and network approaches, and researchers have used route
inference to generate the spatial richness of GPS tracks on multi-million journey
scale (Zaltz Austwick, O’Brien, Strano, & Viana, 2013), although few estimates
of the robustness of these inferences have been carried out.

2. Research focused on GPS data

Global positioning system (GPS) technology was originally developed in the 1970s,
but despite being available for civil purposes in the mid-1980s, it was only in the
1990s that it became widespread in its integration into consumer devices (Kumar
& Moore, 2002). Since then, GPS data have been collected for transport analysis
(Shen & Stopher, 2014). Initially, the technology was primarily applied to
improve aerial and maritime navigation systems, but since the late-1990s, the
largest application of GPS has been land transport. Over the last 20 years, GPS
data have been collected for evaluating system performance such as measuring
historical congestion and flow levels, analysing travel behaviour and estimating
route choice models (Rasmussen, Ingvardson, Halldórsdóttir, & Nielsen, 2013).
In the field of mobility, GPS data have also been collected in the context of
 Big Data and Cycling 117

household travel surveys, in order to complement the survey responses with

detailed trip reporting for a subset of journeys (Bricka, Sen, Paleti, & Bhat, 2012;
Doherty, Noel, Gosselin, SIROIS, & UENO, 2001; Ohmori, 2005; Shen & Stopher,
2014).
Since 2007, there has been a substantial rise in the volume of GPS data, due in
part to the smartphone ‘revolution’. In 2009, smartphones accounted for 15.4% of
the general pool of mobile phones (Li et al., 2010); by 2014, it surpassed 35%, with
over 175 billion units (eMarketer, 2014). In the USA, it rose from 44% in 2011 to
65% in 2013 (The Nielsen Company, 2014). The generalised presence of GPS tech-
nology in smartphones and the vast growth of mobile applications based on
location and tracking functionalities also fed this growth. The emergent naviga-
tion and the sport/fitness app markets (Evans, 2013; Flurry Analytics, 2014)
became apparent more recently, linking personal recorded data to online plat-
forms where people can display and manage their routes and information, and
share and compete with other people, creating different user communities.
Downloaded by [University of London] at 03:08 02 July 2016

In this section, we focus on bicycle riding GPS data collected through mobile
applications, GPS devices and online platforms specifically created for each
study, and data from big app companies, only recently available for research
and planning purposes.

2.1. GPS data collected through specific research initiatives

The first work analysing cycle mobility through GPS tracks dates from 2007
(Harvey & Krizek, 2007). In spring 2006, the research team launched an initiative
to recruit volunteers from different neighbourhoods in South Minneapolis, and
finally collected 938 trips from 51 participants (selected according to their age,
gender, home location and work location) using GPS devices in order to study
commuter cyclist behaviour, analysing chosen routes and their variations due to
existing bike facilities. The project remarked on the difficulty of cleaning GPS
data, which can contain significant positional inaccuracies; consequently, analysis
of cycling behaviour is improved by mapping the recorded GPS tracks onto street
infrastructure. Different authors (Wagner, 1997; Marchal et al., 2005 and Schuess-
ler & Axhausen, 2009a) determined diverse approaches to the map-matching
process that, with increasing complexity and sophistication, solved the main pro-
blems.
The work of Harvey and Krizek provided a descriptive approach to cyclist be-
haviour. Subsequent studies focused on developing cyclist route choice models
from larger samples of GPS routes — typically studying thousands of cyclists
and their routes. The first of these studies, conducted in Zürich (Menghini, Car-
rasco, Axhausen, & Schüssler, 2010), analysed nearly 2500 journeys from over
2400 cyclists. The sample size allowed the creation of a route choice model, but,
since this research did not collect any data associated with the cyclists or the
trips, disaggregation by individual and important features of the street network
(such as slope or traffic) were omitted in the model.
The sample analysed in Zürich was obtained from an independent GPS study
that collected raw GPS data from nearly 5000 participants carrying a GPS receiver
for up to a week, resulting in over 32 000 trips in the cities of Zürich, Winterthur
and Genève. The raw data were processed to identify different transport modes
and trips (Schuessler & Axhausen, 2009b), extracting cycle journeys for indepen-
dent analysis. Modes were detected based on the average and maximum speed
 118 G. Romanillos et al.

during the trip, or by investigating vicinity to infrastructure and stations/stops

during the trip. In the latter case, geo-data regarding stops and infrastructure
was linked to the GPS data using geographic information systems (GIS). For
instance, Stopher, Clifford, Zhang, and FitzGerald’s (2008) first extract walking
trips, followed by public transport trips. Of the remaining trips, bicycle trips
were extracted based on speed and acceleration characteristics. They comment
that GPS loggers can be configured such that they will not record when stationary
(to save the battery). However, when the respondent starts moving again, the
logger needs some time (up to a few minutes) to locate its position, potentially
leading to missing trip starts, which requires additional pre-processing. Broach,
Dill, and Gliebe (2011, 2012) developed a route choice model from GPS data col-
lected in Portland, Oregon, focusing on the journeys of regularly commuting
cyclists. This was a smaller study (with only 164 subjects and around 1500
trips), but its small scale allowed the research team to collect more detailed demo-
graphic data via questionnaire — recognising that cyclists are a heterogeneous
Downloaded by [University of London] at 03:08 02 July 2016

community whose route choices might vary significantly.

At approximately the same time, in Los Angeles, Reddy et al. (2010) had carried
out the first study using smartphones as a mechanism for collecting GPS data.
With the aim of building a platform that enriched the route sharing process, the
Biketastic project developed a mobile application for Android phone users and dis-
tributed it online for free, recruiting 450 users (Savage, 2010). The project website
allowed participants not only to visualise and manage their trips and statistics, but
also to share their routes, and visualise other cyclist’s journeys and other data. GPS
data were associated with noise level and roughness data collected through the
smartphones’ microphones and accelerometers. Volunteers could also provide
information about the route as well as uploading photos and videos of the jour-
neys — acting as a community resource, but also providing contextual data for
researchers.
Similar schemes followed in San Francisco, California (Hood et al., 2011) and
Austin, Texas (Hudson, Duthie, Rathod, Larsen, & Meyer, 2012). The first of
these used the mobile application CycleTrack, developed for the study by Charl-
ton, Schwartz, Paul, Sall, and Hood (2010) and made available for Android and
Apple iOS in an effort to broaden the volunteer base. The initiative collected the
largest sample of cycle GPS tracks to date for research purposes, with nearly one
thousand volunteers contributing data over a five-month period. Through the
app, volunteers provided data about their gender, age and travel purpose,
which were incorporated into the route choice model. Unfortunately, fewer
than one-third of these journeys were successfully mapped to the road
network for further analysis. This cleaning and map-matching processing was
improved by the research conducted shortly afterwards using the same GPS
smartphone application in Austin, Texas (Hudson et al., 2012). Although a
smaller study, they succeeded in matching a similar number of routes. In both
of these studies, the participants were recruited from the smartphone users com-
munity, raising the question of sample bias; however, comparing demographic
data from the smartphone study with information obtained from local travel
surveys did not reveal significant difference in mean age, although they did
reveal a gender bias towards males in the smartphone study. Other socio-demo-
graphic data, such as income, were not collected to avoid private concerns.
Smartphone ownership might have a skew in that regard, but it has not been
possible to test this.
 Big Data and Cycling 119

Following these pioneering studies, more recent research initiatives have

focused on smartphone GPS applications, improving the online platforms and
websites that link apps with volunteers, and providing new functionalities to
encourage people to participate. The initiative Madrid cycle track (Romanillos,
2013; 2014) engaged three hundred casual bikers, as well as cyclists for bike-mes-
senger companies. The initiative collected over 45 000 km of GPS tracks through a
free mobile application, Map My Tracks. In an effort to broaden the user base, those
without smartphones had the option of drawing their routes on an online map. In
both cases, associated information about the age and gender of participants and
the purpose of the travel was collected. It was also the first initiative to allow vol-
unteers to visualise the whole network of collected tracks on a single online map.
In the Netherlands, a similar community-focused initiative was created to gen-
erate interest in pedelecs (electric bicycles). B-Riders in Noord-Brabant in the Neth-
erlands started in September 2013 and ended December 2014, with the aim of
shifting users from car travel to pedelec use. Participants could either register
Downloaded by [University of London] at 03:08 02 July 2016

for a financial compensation — from E0.10 to E0.15 for each kilometre registered
in the morning or the evening peak hours, with a limit of E1000 for each partici-
pant, or register for a coaching programme with feedback and encouragement on
their individual behaviour, or both. To receive the financial compensation and the
feedback, participants were obliged to make use of a smartphone GPS application
developed for the programme — resulting in an unprecedented 400 000 GPS
tracks collected over the period. Bike Print (2014), which allows visualisation
and summary of the data by users (such as specific length of the trip), was devel-
oped specifically for the task, and the data were subsequently used to predict
future usage of the bike network (van de Coevering, de Leeuw, de Kruijf, &
Bussche, 2014).

2.2. Big GPS Data from ‘big app’ companies

The volume of GPS data collected by studies increased significantly when
researchers implemented GPS mobile applications. The development of associ-
ated online platforms, and advertising campaigns among the cyclist community,
served to engage larger groups of participants. However, the sample of contribu-
tors still tends to be small compared to the cycling population in the studied
locations. The growth in sports and fitness apps have opened up sampling of
huge numbers of users (Evans 2013; Flurry Analytics, 2014). In the USA, nearly
one-third of smartphones owners (46 million people) currently use health or
fitness apps (Nielsen, 2014a), aided in part by smart watches and fitness bands
(Nielsen, 2014b). These wearable devices are, however, currently mostly appeal-
ing and affordable for a limited group of wealthy young people, and even
within this group, two-thirds of users do not use these devices for more than
six months (Mitesh, Patel, MBA, & Hall, 2015). Among these fitness apps, GPS
sports tracking apps have been especially popular. In 2013, 7 of these apps sur-
passed 16 million downloads (Comstock, 2013); in 2013, the popular Endomondo
celebrated its fifth birthday and reached 20 million users in more than 200
countries (Endomondo, 2013). MapMyFitness experienced an even more rapid
expansion, surpassing 20 million members in October 2013 (Map my fitness,
2014). App developers ascribe this popularity to attractiveness of the social dimen-
sion of the service as well as the introduction of new features like training plans
(Endomondo, 2013). We are living in the era of not only Big Data, but Big Apps.
 120 G. Romanillos et al.

These apps are widely used by cyclist for tracking sport activities. Endomondo
has registered almost a billion miles of cycling activities, more than half of the
total uploaded (Endomondo, 2013). MapMyRide, one of the most popular together
with Strava, has over 20 million users (Map My Ride, 2014), who have uploaded
over 70 million routes (My fitness pal, 2014). Strava does not disclose its number
of users, but 2.5 million GPS-tracked activities are uploaded to its website every
week (Strava, 2014a) and more than 90 million rides have been collected (Alber-
gotti, 2014).
There are limited studies on these new big GPS data sets from app companies.
Cintia, Pappalardo, and Pedreschi (2013) examined GPS tracks of nearly 30 000
cyclists, collected via the Strava API and analysed training performance using
average speed, duration of ride and cyclist’s heart rate. Wamsley (2014) focused
on analysing travel times collected through Strava in order to generate pacing
strategies for a cyclist to complete a course in the fastest time possible. Other
research defined the conceptual architecture of data collection, management
Downloaded by [University of London] at 03:08 02 July 2016

and methodologies for using and analysing the data (Clarke & Steele, 2011),
including data cleaning, visualisation and trajectory clustering techniques
(Peixoto & Xie, 2013). Other work has instead focused on the use, the motivations
and the online community experience for the people who use cycling apps (Smith,
2014). Very few researchers in this field have focused on the analysis of urban
transport cycling to improve urban planning and design (Clarke & Steele, 2011)
or have developed specific tools to analyse cyclists’ routes. Researchers in Reykja-
vik (Jónasson et al., 2013) have done work in this area, using GPS data from
Garmin Connect and Strava online platforms to create heat map and analyse
cyclist route choices.
The research and planning disciplines are traditionally more interested in urban
transport cycling and require high data density, and data which are representative
of the population in their study region, to build and validate models which big
app data does not necessarily provide. This is beginning to change, as Strava is
the first of these companies to sell cycling GPS data. On May 2014, the
company launched Strava Metro, a commercial brand of the company focused
on providing data services to local authorities, research institutions and other
interested parties (Strava Metro, 2014). In 2013 (Maus, 2014), Oregon’s Depart-
ment of Transportation was the first partner to sign with Strava (Albergotti,
2014). Other urban planning authorities around the world (including London
and Glasgow in the UK, and Victoria in Australia) have followed suit (Albergotti,
2014; Sparkes, 2014). Strava have also launched Strava Labs, a high-resolution
online map that visualises the cycle flow distribution collected through the app
around the world (Strava Labs, 2014), representing over 75 million journeys and
220 billion GPS points (Mach, 2014).
Models like Strava Metro bring significant new opportunities for analysis and
understanding. First, the Street map shows a very high density of GPS tracks cover-
ing the whole metropolitan area (although still exhibiting some degree of spatial
and socio-demographic bias). The data are processed to remove users’ personal
information, but summaries of basic demographic information (gender and age
ranges) are provided, allowing demographic bias to be estimated. Additionally,
it provides not only information about the total number of cycle trips but also
the number of commuting trips — very important information for urban transport
planning. Strava Metro also provides cyclist flow information at different dates
and times — for example, via the Strava Saturday online heat map (Strava,
 Big Data and Cycling 121

2014b) — so it is possible to analyse cyclist flow for different times of the day (the
morning and the afternoon peaks), and study the evolution across the whole year,
opening up the possibility of detailed spatio-temporal and seasonal analyses.
However, Strava Metro data also present limitations. Users’ privacy concerns
mean that single route tracks are typically not accessible so it is not possible to
analyse trip length, purpose of travel or the route choice on an individual
journey level. Because these data are shared in an aggregated form, it is not poss-
ible to study the relationships between these variables; for example, the depen-
dence of route choice on the cyclist’s travel purpose. Because we only have
aggregated socio-demographic information, there is limited scope to analyse the
importance of basic factors like age or gender in route planning, journey length
or purpose. All of these analyses are likely to be important for planning, designing
and managing cycle infrastructure. The solution would be to have access to disag-
gregate data and provide single tracks, a difficult proposition when maintaining
user (and company) privacy. In order to not discourage user participation,
Downloaded by [University of London] at 03:08 02 July 2016

shortly after opening Strava Metro, the company offered members the option of
marking routes as private. These routes are then not included in Strava Metro
data set (Wehner, 2014).

3. Research focused on point data

In addition to the substantial body of research around GPS, there has been a sig-
nificant interest in analysing cycling data gathered at specific locations. Studies
have mainly explored two different data sources: point data registered at Bike
Share Programmes (BSP) stations and counts.

3.1. Exploring BSP data mines

With the exception of studies based on bike parking data provided by specific,
one-off surveys (Rietveld, 2000), bike mobility trends have not been analysed
through large point data sets gathered at BSP docking stations or parking lots
— until recently. The biggest evolution in this area came with the rapid expansion
of BSP in cities around the world. The first generation of such systems dates from
1965 (Demaio, 2009), but they remained very few and small in size till the early-
1990s, when a second generation of BSP was born. Still, these programmes grew
slowly until the mid-2000s, when a third generation of bike share (characterised
by electronic management, and hence a rich data source) became popular in
many countries. Since then, the number of such systems increased exponentially
around the world (Fishman, Washington, & Haworth, 2013). By the end of 2007,
there were about 60 cities with third-generation BSP implemented worldwide
(Demaio, 2007); according to Fishman (2015), the current number of BSP is 855,
with nearly one million bicycles in use.
A common feature of this third generation of BSP is that they record information
when a bike in undocked (hired) or docked (returned). These data were first
explored in a study in the Barcelona BSP, Bicing (Froehlich et al., 2009), covering
from August to December 2008. Three different kinds of data were gathered
from the Bicing information system by scraping the website (using an automated
programme to find and store the relevant data elements presented by the
webpage). These data were collected every two minutes and included the
station locations, the number of available bicycles and the number of vacant
 122 G. Romanillos et al.

parking slots. Bicing was launched in 2007; it had nearly 400 stations and 6000
bikes, with 150 000 subscribers. First, by applying clustering techniques, the
research identified spatio-temporal patterns, relating the use of different bike
stations to activity clusters over the course of a weekday, when more regular
BSP usage patterns were identified. Second, the research developed different pre-
dictive models to analyse the impact of several factors (such as time of the day or
the amount of historical data) in order to create tools to estimate bicycle demand
for different stations and the optimal location of future ones. The research pointed
towards the potential of this new source of data to identify not only cycling or
mobility patterns, but broader urban trends and dynamics, such as inferring
urban land uses (home, office or leisure/retail) by analysing users’ profile over
time.
A later study worked with Barcelona BSP data with more specific objectives
(Kaltenbrunner, Meza, Grivolla, Codina, & Banchs, 2010). Aware that users of
Bicing often found it difficult to find a bike to hire, or a space to leave their bike
Downloaded by [University of London] at 03:08 02 July 2016

at their destination, the researchers developed a model that could predict the
availability of bikes or docks, and could inform both users and system managers
in advance so that they could respond accordingly. Even an hour ahead, their
autoregressive– moving-average (ARMA) model was typically accurate to one
bicycle, representing a usable prediction range for cyclists. More recently, Giot
and Cherrier (2014) completed a similar predictive analysis based on Washington,
D.C. BSP data, working with a suite of research regression techniques.
There has been a range of effort to work with BSP data in real time, building new
tools for system management and to improve service. In 2009, Luo and Shen (2009)
developed an information system for the BSP of Hangzhou (China) that rep-
resented the location of the BSP stations and dynamically displayed the
availability of bikes or free parking spots. The most remarkable visualisation of
real-time BSP information is The Bike Share Map (O’Brien, 2010; 2013). Created in
2010 in order to visualise London’s BSP data, the map represents the information
of different cities around the globe since June 2013, covering at time of writing 107
BSP and visualising the availability of systems around the world. This global view
was incorporated into research based on BSP data (Cheshire & O’Brien, 2013;
O’Brien et al., 2013). The investigation collected data from 38 systems from
Europe, the Middle East, Asia, Australia and America, and the data set included
locations, capacity and current load factor of docking stations. After analysing the
data, the investigation compared and classified the BSP according to variables
such as the system’s geographical size, the variation of occupancy rates across
the day or the week, and the intensity and distribution of activity in relation to
demographics. The paper compared the geographical distribution and temporal
popularity of a range of different schemes, allowing planners to examine
schemes with elements in common in other parts of the world.
In addition to research focusing on providing useful apps and interfaces to
service providers, researchers are increasingly taking more theoretical approaches
to dock data to understand differing spatio-temporal patterns using signal proces-
sing and statistical methods. In 2012, Lathia, Ahmed, and Capra (2012) used
cluster analysis to detect ‘similar’ stations in the London system based on the
time profile of their occupation, resulting in docking stations which have
similar behaviours over the course of a day, and examining the impact of
‘casual’ users. These users pay using a credit card instead of the access keys
used by subscription users at the time of the programme’s launch — these
 Big Data and Cycling 123

casual users may be more likely to be tourists or business visitors. Similar methods
were applied by Etienne & Latifa (2012) to cluster docking stations which are
similar in their temporal patterns of occupation, focusing on the flagship Velib’
system in Paris. This covered 2.5 million trips in just one month — Velib’ is the
second largest BSP in the world. Working on the London system, Padgham
(2012) is one of the first to attempt to connect BSP activity with that of the other
parts of the public transport network, and introduced spatial interaction
model-like approaches to understanding flows between locations. Many
of these studies focused on Europe and North America. Corcoran, Li, Rohde,
Charles-Edwards, and Mateo-Babiano (2014) study Brisbane, Australia and
examine the impacts of weather and public events on city cycle use. In
Fishman, Washington, Haworth and Mazzei (2014) used data collected from
BSP trips in 2012 to visually represent the strength of the relationship between
different docking stations and how this relates to the public transport system.
Research on point data in BSP systems has yielded a raft of visualisations, apps
Downloaded by [University of London] at 03:08 02 July 2016

and analyses. Many of the more academic works have employed specialised stat-
istical techniques that are perhaps not as familiar to the policy-maker or transport
planner, and joining up the scientific expertise with services and interventions
amenable to the user, service provider or policy-maker still has a way to go.
Limited work has been done to combine it with journey data, which in itself
would yield new possibilities.

3.2. Other point data sources: Manual and automated counts

While BSP provide detailed and timely point data reporting, there are other
sources that provide large and useful point data collections, but rarely on the
same scale and level of detail. Within the scope of this review, the evolution of
counts in the last years is especially interesting.
Though manual counts cannot be considered as a source of Big Data — they just
meet the first V criterion (volume) of Laney’s (2001) classification — they are still
the most prevalent cycling data collection method (Ryus, Laustsen, Proulx,
Schneider, & Hull, 2014), producing increasingly large data sets through recent
initiatives. Many communities still successfully use conventional, lower tech
methods in order to collect point data and support an evidence base for cycling
policy. In some countries, like the USA, many cycling communities (Schneider,
Patten, & Toole, 2005) encourage volunteers to register cyclists at key locations
in precise dates through manual count methods. Among the different initiatives,
especially remarkable is the National Bicycle and Pedestrian Documentation Project
(NBPD, 2009– 2015), a programme that provides to the volunteers a methodology,
as well as training and documentation, and centralises the collection of surveys
and counts from cities all around the USA.
Apart from these massive manual counts initiatives, there is a substantial collec-
tion of cycling data through automated counts. The most common methods are
based on pneumatic tubes, inductive loops, passive infrared, automated video
counters, infrared cameras and fibre optic pressure sensors (Ryus et al., 2014).
Pneumatic and inductive are widespread, but proved to be accurate only when
detectors are properly installed, calibrated, maintained, free of external interfer-
ence, and on a dedicated bicycle lane (Nordback & Janson, 2010). Recently,
more innovative counts based on fibre optics register cyclists on mixed traffic
lanes, offering insight not only in the cycling volume but also in the speed and
 124 G. Romanillos et al.

direction. In the Netherlands, new traffic light detection loops have been
implemented to detect cyclists with high accuracy by using a new methodology
with dedicated algorithms (Rijn, 2014; Winter, 2012). This system is being
implemented extensively in some cities: Utrecht is currently adjusting 170 traffic
lights which measure motorised traffic to also detect cyclists. This cycling data
are being made available in an online open data platform (Open Data Utrecht,
2015). Such efforts could be facilitated by the technological innovators who are
working to create sensors which cost close to $50 — 1% of the cost of current
sensors (Andersen, 2015). Knock Software is one such innovator, active in Portland,
OR on a device which uses magnetic, thermal and speed detection to determine
whether a passing object is a bike, a car or a pedestrian. If this proves reliable, cov-
erage of cities could rapidly become more comprehensive, detailed and timely.
Considering that count data are at the base of many studies which examine
travel patterns, it is worthy highlighting the most important advantages and dis-
advantages in relation to other approaches. Count data register every single cyclist
Downloaded by [University of London] at 03:08 02 July 2016

at a specific location while BSP or GPS data rely on a more segregated cycling
population. However, the absence of sample bias in count data is not guaranteed
at all, and they are collected on an aggregate level such that no demographic data
are captured. According to Ryus et al. (2014), manual counting is still the most
dominant method of counting cyclists — 87% of total counts in the USA — and
still relies heavily on volunteers. That means that samples are usually registered
at a limited number of locations in a specific date or period of time, and may
have spatial bias if the count locations are not well distributed. The increasing
extension of new automated counts could allow pattern analysis across time
and, if well distributed, could reduce spatial biases.

4. Research focused on journey data from BSP

The third-generation BSP not only record information about the number of
bicycles in docking stations but also identify and register bikes (and sometimes
an identifier for their users) at the start and end dock of every journey. This
means that BSP are able to provide general mobility data through the origin – des-
tination matrices associated with users’, but also timings of these journeys (and,
by inference, duration). In addition, BSP may provide data about cyclists (age
or gender, for instance) — although this is not always the case, either because
the data are not collected (from casual, credit card users), or because that aspect
of the data is withheld for privacy reasons. Research on journey data has so far
been more limited. BSP journey data are historical; they are typically released in
large batches covering months or even years of activity. They have limited use
for nowcasting or feeding back information to users in real time. Nevertheless,
there has been significant work in visualising these data (Bargar, Gupta, Gupta,
& Ma, 2014; Wood, Slingsby, & Dykes, 2011; Zaltz Austwick et al., 2013), creating
a comparison study of different visualisation techniques with respect to these
data.
The research carried out by Borgnat et al. (2011) is one of the first analytical
approaches to these origin – destination data sets, and focusing on data from the
city of Lyon in France. The investigation analysed the data set provided by the
managing company and the City Hall, corresponding to the 13 million trips
over a two-and-half-year period. The system registered the start time and depar-
ture station, and end time and destination station, for each journey. For the first
 Big Data and Cycling 125

time, researchers could examine individual mobility, characterising different

groups according to the distance, duration or speed of their trip. While the
research carried out in Barcelona on point data (Froehlich et al., 2009) covered a
short period of time, the research conducted in Lyon allowed trend and temporal
analysis over a much longer period. The data collection began at the opening of
the system and covered expansions of the scheme, allowing the study to cover
different demand and service scenarios throughout this period, and analysed
how factors such as increasing numbers of bicycles and stations affected the
number of subscribers. The same year, Vogel, Greiser, and Mattfeld (2011) ana-
lysed similar data from Vienna’s BSP, Citibike Wien, covering around 760 000
rides from 2008 to 2009. General spatio-temporal patterns are derived from the
analysis while an integrated approach of Data Mining and Operation Research
is presented in order to develop a new trip model that anticipates bike activities
for better long-term location planning. The researchers were able to formulate
clear policy goals from their analyses.
Downloaded by [University of London] at 03:08 02 July 2016

The first multi-city analysis of origin – destination data was carried out by Zaltz
Austwick et al. (2013), which compared five cities (London, Washington DC, Min-
neapolis, Denver and Boston), using spatial network analysis methods to cluster
stations into communities (subnetworks of journeys within the wider network).
The smallest of these data sets covered 168 000 journeys (Denver) and the
largest 3.6 million (London) and allowed comparison of distance travelled and
journey time distributions between cities. The paper also used inferred routing
for visualisation purposes using Open Street Map and Routino (http://routino.
org), but did not utilise this for distance estimation or street network loading, as
there was no mechanism to validate this route choice. Bargar et al. (2014) builds
on a network analysis approach (examining data from Washington DC, Chicago
and Boston), complementing it with the spatio-temporal clustering methods
used by other researchers, and visualising both of these techniques via a web-
based map visualisation built using JavaScript libraries, integrating analysis into
a more accessible visualisation tool.
More recent work has expanded its scope beyond predicting demand or
detecting similar locations, and has focused instead on correlating cycling activi-
ties with wider policy goals around health and transport. The use of the London
BSP across the three first years of operation has been examined by Goodman and
Cheshire (2014). The study analysed the evolution in the profile of users, the
increase in the number of trips as well as variation in the proportion of trips by
registered users. This covered a period of time that included the extension of
the BSP network in 2012 and the rise of the service prices in January 2013. The
data set incorporated the gender and home postcodes of users, permitting ana-
lyses that linked geographic socio-economic factors of the residential locations,
and evaluating the demand according to the distance from homes to the start or
end stations. Defined as “trips made by two or more cyclists together in space
and time” data (Beecham & Wood, 2014, p. 1), group-cycling journeys on
London BSP were studied by analysing the trips of over 80 000 members
between September 2011 and September 2012. The research revealed some plaus-
ible patterns, like the increase of group-cycling journeys at weekends, late eve-
nings and lunchtimes, and the large proportion of group members who share
the same postal code. However, it also revealed some unexpected ones, like sets
of commuting group-cycling journeys, and some differences between group and
individual trips according to gender. This simple approach starts to connect BSP
 126 G. Romanillos et al.

work with wider interests around social behaviour, health and leisure. Faghih-
Imani, Eluru, El-Geneidy, Rabbat, and Haq (2014) studied how land use, urban
form, building environment attributes and weather impact on the bicycle flow,
by analysing the data from the Montreal BSP, BIXI, between April and August
2012. The research reports, unsurprisingly, good weather leading to high cycling
flow, but also provides interesting findings for policy-makers and urban
designers, such as the relationship between BSP usage and urban density, and
the interaction between cycling and public transport.
An underused aspect of journey data is its capability to act as a supplementary
and validating data source for the more current, accessible point data (which
through APIs, is typically updated on a minute-by-minute basis). Point data typi-
cally register only net changes — so, for example, three bikes arriving and two
bikes leaving appear the same way as one bike leaving. By using journey data
to validate the behaviour of the system, it could be used to infer expected traffic
at docking stations (and hence whether a small net change represents large or
Downloaded by [University of London] at 03:08 02 July 2016

small flows), as well as allowing spatial models for predicting flows based on
just the total ins and outs of each docking station (in GIS, interpolating a matrix
from its marginal sums is a relatively standard technique (Deming & Stephan,
1940)).
Future work on BSP will surely rely on combining different strands of data from
within the scheme, or with external data sets. If BSP utilise GPS tracking more
widely, it could open up the possibility of a linking of journey data (time-
varying origin –destination matrices), point data (station locations and statuses)
and routing data (the details of the route that users take between origin and des-
tination on the street network) — allowing inference of time-dependent BSP traffic
on the level of individual road segments. If GPS data yield route preference, and
journey data yield time-dependent demand at an origin – destination level, com-
bining both with live point data could yield a complex, timely modelling tool.
This BSP ‘nowcasting’ could allow prediction in very small time windows —
for example, docking station-level occupation and demand in ten or twenty
minutes in the future. Combining BSP data with complementary sources —
health and demographic data, for example — opens up the possibility of
linking BSP to a wider context — including transport planning, access to services
of marginalised groups, and behaviour change.

5. Conclusions
This paper reviews the recent bike mobility research based on the analysis of Big
Data collected from sources that are becoming increasingly accessible to research-
ers and policy-makers, offering a panoramic view on the growing number of
studies that, in less than ten years, have evolved as quickly as the data themselves.
Even if the achievements are remarkable, there are still important limitations that
are difficult to overcome using current data sources. By some estimates, cycling
data meet the first of Laney’s (2001) ‘3Vs’ classification of Big Data (that of
volume), given the size GPS and BSP data, and perhaps the second criterion (Vel-
ocity), since some data are available in real time (Luo & Shen, 2009; O’Brien, 2010,
2013). It is more questionable whether the other V criteria (Variety and the fourth
one added by Kitchin (2014), Veracity) are met, at least in the way that the data are
currently being used. In the context of cycling, while the data are combined with
demographic or interview data, pooling it with Big Data from other sources
 Big Data and Cycling 127

seldom occurs. As hinted, there may be scope within BSP to combine point data
(sparse, complete and real-time data) with journey data (more detailed, complete
and historical samples) and GPS data (very detailed but potentially smaller
samples, and historical) to leverage the detail of one data set against the timeliness
and sampling power of the others.
With respect to Veracity, our conclusions differ between sources; this criterion
refers to possible biases, noise or any abnormality in data, which is variable for
each of the data types. Research based on dedicated GPS data collections has typi-
cally paid attention to proper sampling procedures, so that the collected data are
by and large representative for the population studied. However, data from big
app companies rely on volunteers uploading their cycling tracks, leading to
self-selective samples. For instance, logging bike trips in Strava may be more
likely to be carried out by cycling enthusiasts who wish to show off their
cycling achievements. This would imply a lack of representativeness of the popu-
lation in terms of cycling attitude, geographical location and socio-demographic
Downloaded by [University of London] at 03:08 02 July 2016

characteristics. Groups with mobility impairments, those who are ‘afraid to

cycle’, elderly cyclists, or children may not be well represented in these accounts.
Recent studies by Buck (2013a) and Dill and McNeil (2013) demonstrate that hetero-
geneity along these lines indeed exists, suggesting that data from big app sources
will be biased. However, BSP point and journey data are representative, at least
of users of BSP. How representative this population is of wider cyclists and citizens
is, of course, open to question (see Buck et al., (2013b) for further discussion).
Indeed, there is no reason to believe that either BSP or big app data provide repre-
sentative samples of a cities’ population of cyclists or potential cyclists.
Another reason to be concerned about data veracity relates to data collection
motivation and methods. In cases in which data are collected specifically for aca-
demic purposes, it is typically enriched with contextual information (such as
socio-demographics, attitudes, spatial context or environment). When data are
collected by commercial applications, aimed at providing a service to customers
(e.g. Strava, MapMyRide), privacy policies of companies make using this contex-
tual information difficult or impossible. As a consequence, key variables to under-
standing travel behaviour, such as socio-demographics or purpose of the journey,
may be absent. However, the size of the data gathered and its continuity over time
potentially allow for analyses not possible on dedicated GPS data (e.g. spatial clus-
tering or the variation of cyclist flow distribution over time), which may deliver
useful additional insights. Similarly, BSP data are collected for management
rather than research, and lack socio-demographic context. In any case, BSP may
offer a rich database for analysing regularities in patterns of supply and
demand as well as longer term structural developments.
On a technical level, GPS accuracy is not an issue which has been completely
resolved. Dedicated GPS devices perform better than smartphones GPS
(Lindsey, Gorjestani, Hankey, & Wang, 2013), but their lack of accuracy in some
urban areas can mean analysts lack the fine detail to precisely distinguish route
choice — one of the main reasons the data are of interest. The Galileo European
Program, which is expected to be in place by 2019 (European Commission,
2014), promises improvements over the current system, but these improvements
have yet to be fully demonstrated. For users, one barrier is that, historically,
GPS apps have rapidly drained their smartphone batteries — this is significant
enough that the B-Riders scheme developed an app for an intelligent start and
end of the GPS tracking to minimise this problem.
 128 G. Romanillos et al.

Despite these caveats, there are interesting research challenges and opportu-
nities from the increasing availability of new data sets and the steady improve-
ments in their quality. The industries around sport-tracking apps have seen
increases in the number of users of GPS devices (including recent wearable
devices) (Nielsen, 2014a). If this trend continues, the volume of data will
increase with the userbase, and, through licensing schemes, so will the avail-
ability of data. Data from BSP will likely grow, due to the proliferation of BSP
around the world. Future research will have to face the challenge of bias in
its data collections, and create robust, scalable mechanisms to account for it.
We expect more GPS data to become available in a more timely fashion, not
only from app companies (some of which are already offering this service for
users, like Map My Tracks) but from the current third generation of BSP.
Some systems will soon start recording GPS tracks for every journey, which
will allow researchers to analyse bike routes and improve the existing route
choice and cycling flow distribution models, as well as analyse the real use of
Downloaded by [University of London] at 03:08 02 July 2016

existing bike infrastructure. Apart from these improvements regarding raw

location data, work is needed to enrich these data with meaningful explanatory
variables. Socio-demographic data may be approximated by linking location
data to usage patterns of specific groups. More work will also be needed on
data fusion techniques in order accommodate such approximations; however,
data providing spatial context (such as land use) are becoming increasingly
accurate and more freely available. This growth in bicycle data and its corre-
sponding availability, and joining up with data on transport, health, air
quality, demographics, route choice and leisure promises a rich period of
activity for researchers in all of these areas.
Finally, a key question remains: how will the expected advances benefit cyclists
and potential cyclists, policy-makers and BSP? And how would those benefits
create wider impacts? Will they encourage more people to cycle, or reduce conges-
tion or pollution? Many BSP users currently take advantage of real-time infor-
mation about the availability of bicycles in different docking stations so that
they can plan their journeys. In a near future, we might imagine a smart bike
route planning system, integrated in a multimodal transport system. Users will
have information about the closest available station to their destination point,
and about the best route possible for getting there, incorporating weather, traffic
and user preference — lowering barriers to cycling for less confident or experi-
enced cyclists. Cycling Apps will continue to be attractive to users of smartphones
and perhaps a new generation of wearable technology, providing information to
cyclists and reports of their peers’ performance, motivating people to cycle
longer, faster, and of course, more frequent.
For policy-makers, the range of benefits may be more diverse. GPS-based
cycling data will provide insights into cyclists’ route choice behaviour and their
preferred and disliked route characteristics, which will support the design of
cycling infrastructure networks. Coupling GPS-based cycling data with geo-
data (land use, facilities, altitudes, etc.) will greatly enhance their understanding
of cyclists’ route choice. Big Data will drive the assessment of cycling infrastruc-
ture at different levels, analysing the use of local infrastructures (such as lanes
or bike parking), identifying the main cycling routes over the course of a day, or
understanding the obstacles, delays and dangers that slow or hinder their jour-
neys. Again, a key issue here is the representativeness of the pool of GPS users.
While an initiative such as the Dutch BikePRINT project delivers useful insights
 Big Data and Cycling 129

into cycling routes and cycling densities, it relies on voluntary participants,

leaving questions about reliability of the outcomes (Coevering et al., 2014).
The recent collaboration between commercial Apps and planning institutions is
promising and will generate combined and useful information that will make new
explorations possible. As we have remarked, these new Big Data will not substi-
tute but complement other more conventional sources, since they often lack disag-
gregate data on the cyclists, which are so often necessary to understand the
contexts that influence many of their decisions. This points, then, to a future
where the fourth V – Variety – creates new innovations and insights in cycling
– as Big App data, real-time BSP feeds, and more traditional, detailed, demo-
graphic studies are brought together – and commercial, municipal, service pro-
vision and academic partners work together to create a living, user-centred
picture of the cyclable city.
Downloaded by [University of London] at 03:08 02 July 2016

Disclosure statement
No potential conflict of interest was reported by the authors

References
Albergotti, R. (2014). Strava, popular with cyclists and runners, wants to sell its data to urban planners.
Wall Street Journal. Retrieved from http://blogs.wsj.com/digits/2014/05/07/strava-popular-with-
cyclists-and-runners-wants-to-sell-its-data-to-urban-planners/
Andersen, M. (2015). This $50 device could change bike planning forever. Bike Portland ORG. Published on
January 13, 2015. Retrieved from http://bikeportland.org/2015/01/13/50-device-change-bike-
planning-forever-130891
Bargar, A., Gupta, A., Gupta, S., & Ma, D. (2014). Interactive visual analytics for multi-City bikeshare
data analysis. The 3rd international workshop on urban computing. New York City. Retrieved from
http://www2.cs.uic.edu/~urbcomp2013/urbcomp2014/papers/Bargar_Bikesharing.pdf
Beecham, R., & Wood, J. (2014). Characterising group-cycling journeys using interactive graphics.
Transportation Research Part C: Emerging Technologies, 1– 13. doi:10.1016/j.trc.2014.03.007
Bike Print. (2014). Retrieved from http://www.bikeprint.nl/index.php?lang=en
Borgnat, P., Robardet, C., Rouquier, J. B., Abry, P., Fleury, E., & Flandrin, P. (2011). Shared bicycles in a
city: A signal processing and data analysis perspective. Advances in Complex Systems, 14(3), 1 –24.
Retrieved from http://www.worldscientific.com/doi/abs/10.1142/S0219525911002950
Bricka, S., Sen, S., Paleti, R., & Bhat, C. (2012). An analysis of the factors influencing differences in
survey-reported and GPS-recorded trips. Transportation Research Part C, 21(1), 67– 88. doi:10.1016/j.
trc.2011.09.005
Broach, J., Dill, J., & Gliebe, J. (2011). Bicycle route choice model developed from revealed-preference GPS data.
Transportation Research Board 90th Annual Meeting.
Broach, J., Dill, J., & Gliebe, J. (2012). Where do cyclists ride? A route choice model developed with
revealed preference GPS data. Transportation Research Part A: Policy and Practice, 46(10), 1730– 1740.
doi:10.1016/j.tra.2012.07.005
Buck, D. (2013a). Encouraging equitable access to public bikesharing systems. ITE Journal, 83(3), 24– 27.
Retrieved from: http://faculty.washington.edu/abassok/bikeurb/resources/media/abstracts/
papers/153_Buck.pdf
Buck, D., Buehler, R., Happ, P., Rawls, B., Chung, P., & Borecki, N. (2013b). Are bikeshare users different
from regular cyclists? Transportation Research Record: Journal of the Transportation Research Board,
2387(1), 112–119.
Capital Bike Share. (2015). Retrieved from http://www.capitalbikeshare.com/
Charlton, B., Schwartz, M., Paul, M., Sall, E., & Hood, J. (2010). CycleTracks: a bicycle route choice data col-
lection application for GPS-enabled smartphones. 3rd Conference on innovations in travel modeling, a
transportation research board conference. Tempe, Arizona, USA.
Cheshire, J., & O’Brien, O. (2013). Revealing and informing transport behaviour from bicycle sharing systems.
The Geographic Information Systems Research UK (GISRUK). University of Liverpool. Retrieved
 130 G. Romanillos et al.

from http://www.geos.ed.ac.uk/~gisteac/proceedingsonline/GISRUK2013/gisruk2013_submission_
31.pdf
Cintia, P., Pappalardo, L., & Pedreschi, D. (2013). ‘Engine matters’: A first large scale data driven study on
cyclists’ performance. 2013 IEEE 13th International Conference on Data Mining Workshops (pp. 147–
153). IEEE. doi:10.1109/ICDMW.2013.41
Clarke, A., & Steele, R. (2011). How personal fitness data can be re-used by smart cities. Seventh international
conference on intelligent sensors, sensor networks and information processing (ISSNIP) (pp. 395–
400). New South Wales, Australia.
van de Coevering, P., de Leeuw, G., de Kruijf, J., & Bussche, D. (2014). Bike print. Policy renewal and inno-
vation by means of tracking technology. In Nationaal verkeerskundecongres 2014. Retrieved from http://
www.goudappel.nl/media/files/uploads/NVC2014-bikeprint-_NHTV_-_DAT_Mobility.pdf
Coleman, E. (2013). Lessons from the London Datastore. In B. Goldstein & L. Dyson (Eds.), Beyond trans-
parency - Open data and the future of civic innovation (pp. 39–50). San Francisco, CA: Code for America
Press. Retrieved from http://beyondtransparency.org/
Comstock, J. (2013). 7 fitness apps with 16 million or more downloads. Mobi Health News. Retrieved from
http://mobihealthnews.com/24958/7-fitness-apps-with-16-million-or-more-downloads/
Corcoran, J., Li, T., Rohde, D., Charles-Edwards, E., & Mateo-Babiano, D. (2014). Spatio-temporal pat-
terns of a public bicycle sharing program: The effect of weather and calendar events. Journal of Trans-
Downloaded by [University of London] at 03:08 02 July 2016

port Geography, 41, 292–305.

Demaio, P. (2007, December 30). What a year for the bike-sharing. Retrieved from http://bike-sharing.
blogspot.com/2007/12/what-year-for-bike-sharing.html/.
Demaio, P. (2009). Bike-sharing: History, impacts, models of provision, and future. Journal of Public
Transportation, 12(4), 41– 56.
Deming, W. E., & Stephan, F. F. (1940). On a least squares adjustment of a sampled frequency table
when the expected marginal totals are known. The Annals of Mathematical Statistics, 11(4), 427–444.
Dill, J., & McNeil, N. (2013). Four types of cyclists? Examination of typology for better understanding of
bicycling behavior and potential. Transportation Research Record: Journal of the Transportation Research
Board, 2387, 129–138.
Doherty, S. T., Noel, N., Gosselin, M.-L., Sirois, C., & Ueno, M. (2001). Moving beyond observed outcomes:
integrating global positioning systems and interactive computer-based travel behavior surveys. Transpor-
tation Research Board. Retrieved from http://trid.trb.org/view.aspx?id=686721
eMarketer. (2014). Smartphone users worldwide will total 1.75 Billion in 2014. eMarketer. Retrieved from
http://www.emarketer.com/Article/Smartphone-Users-Worldwide-Will-Total-175-Billion-2014/
1010536
Endomondo. (2013). Endomondo fitness app runs past 20 Million users and reaches profitability. Retrieved
from http://blog.endomondo.com/2013/10/16/endomondo-fitness-app-runs-past-20-million-users-
and-reaches-profitability/
Etienne, C., & Latifa, O. (2012). Model-based count series clustering for Bike-sharing system usage
mining, a case study with the Vélib’ system of Paris. Transportation Research-Part C, 5(3), 1 –23.
European Commission. (2014). Galileo — What do we want to achieve? Retrieved from http://ec.europa.
eu/enterprise/policies/satnav/galileo/index_en.htm
Evans, B. (2013). Mobile is eating the world. Retrieved from http://ben-evans.com/benedictevans/2013/
11/5/mobile-is-eating-the-world-autumn-2013-edition
Faghih-Imani, A., Eluru, N., El-Geneidy, A. M., Rabbat, M., & Haq, U. (2014). How land-use and urban
form impact bicycle flows: Evidence from the bicycle-sharing system (BIXI) in Montreal. Journal of
Transport Geography, 41, 306–314.
Fishman, E. (2015). Bikeshare: A review of recent literature. Transport Reviews, 1 –22. doi:10.1080/
01441647.2015.1033036
Fishman, E., Washington, S., & Haworth, N. (2013). Bike share: A synthesis of the literature. Transport
Reviews, 33(2), 148– 165. doi:10.1080/01441647.2013.775612
Fishman, E., Washington, S., Haworth, N., & Mazzei, A. (2014). Barriers to bikesharing: an analysis
from Melbourne and Brisbane. Journal of Transport Geography, 41, 325– 337. doi:10.1016/j.jtrangeo.
2014.08.005
Flurry Analytics. (2014). Health and fitness apps finally take off, fueled by fitness fanatics. Retrieved from
http://www.flurry.com/blog/flurry-insights/health-and-fitness-apps-finally-take-fueled-fitness-
fanatics#.VA2Zmfl_tik
Froehlich, J., Neumann, J., & Oliver, N. (2009). Sensing and predicting the pulse of the city through
shared bicycling. In C. Boutilier (Ed.), Twenty-First international joint conference on artificial intelligence
(IJCAI-09), Pasadena, CA, USA, July 14– 17, 2009 (pp. 1420– 1426). Menlo Park, CA: AAAI Press.
 Big Data and Cycling 131

Giot, R., & Cherrier, R. (2014). Predicting bikeshare system usage up to one day ahead. In R. Kozma, N.
Zhang, & Z. Zeng (Eds.), Computational intelligence in vehicles and transportation systems (CIVTS), 2014
IEEE symposium on (pp. 22– 29). Orlando, FL: IEEE.
Goodman, A., & Cheshire, J. (2014). Inequalities in the London bicycle sharing system revisited:
Impacts of extending the scheme to poorer areas but then doubling prices. Journal of Transport Geogra-
phy, 41, 272– 279.
Hall, W., Shadbolt, N., Tiropanis, T., O’Hara, K., & Davies, T. (2012). Open data and charities. Retrieved
from: http://eprints.soton.ac.uk/341346/
Harvey, F. J., & Krizek, K. J. (2007). Commuter bicyclist behavior and facility disruption. Transportation
Research Board. Retrieved from http://trid.trb.org/view.aspx?id=811576
Hood, J., Sall, E., & Charlton, B. (2011). A GPS-based bicycle route choice model for San Francisco, Cali-
fornia. Transportation Letters The International Journal of Transportation Research, 3(1), 63–75. doi:10.
3328/TL.2011.03.01.63-75
Hudson, J. G., Duthie, J. C., Rathod, Y. K., Larsen, K. A., & Meyer, J. L. (2012). Using smartphones to
collect bicycle travel data in Texas (No. UTCM 11-35-69). Retrieved from http://utcm.tamu.edu/
publications/final_reports/Hudson_11-35-69.pdf
Jónasson, Á., Eirı́ksson, H., Eðvarðsson, I., Helgason, K. T., Sæmundsson, T., Sigurgeirsson, D. B., &
Vilhjálmsson, H. H. (2013). Optimizing expenditure on cycling roads using cyclists’ GPS data. School of
Downloaded by [University of London] at 03:08 02 July 2016

Computer Science, Reykjavik University. Retrieved from: http://trauzti.com/files/urban-routing.

pdf
Kaltenbrunner, A., Meza, R., Grivolla, J., Codina, J., & Banchs, R. (2010). Urban cycles and mobility pat-
terns: Exploring and predicting trends in a bicycle-based public transport system. Pervasive and
Mobile Computing, 6(4), 455– 466. doi:10.1016/j.pmcj.2010.07.002
Kitchin, R. (2014). Big data should complement small data, not replace them. Impact of social sciences. The
philosophy of data science (series). Retrieved from http://blogs.lse.ac.uk/impactofsocialsciences/
2014/06/27/series-philosophy-of-data-science-rob-kitchin/
Kumar, S., & Moore, K. B. (2002). The evolution of global positioning system (GPS) technology. Journal
of science Education and Technology, 11(1), 59–80.
Laney, D. (2001). 3D data management: Controlling data volume, velocity and variety. META Group Research
Note, 6.
Lathia, N., Ahmed, S., & Capra, L. (2012) Measuring the impact of opening the London shared bicycle
scheme to casual users. Transportation Research Part C: Emerging Technologies, 22, 88– 102. doi:10.1016/
j.trc.2011.12.004.
Li, X., Ortiz, P. J., Browne, J., Franklin, D., Oliver, J. Y., Geyer, R., . . . Chong, F. T. (2010). Smartphone evol-
ution and reuse: Establishing a more sustainable model. 2010 39th International Conference on Parallel
Processing Workshops, 476 –484. doi:10.1109/ICPPW.2010.70
Lindsey, G., Gorjestani, A., Hankey, S., Wang, X., & Chen, J. (2013). Feasibility of using GPS to track bicycle
lane positioning. Minneapolis, MN. Retrieved from http://purl.umn.edu/148996
Luo, R., & Shen, Y. (2009). The design and implementation of public bike information system based on
Google maps. 2009 International Conference on Environmental Science and Information Application Tech-
nology, 156–159. doi:10.1109/ESIAT.2009.298
Mach, P. (2014). What do 220,000,000,000 GPS data points look like? Retrieved from http://engineering.
strava.com/global-heatmap/
Manyika, J. (2011). Big data: The next frontier for innovation, competition, and productivity, McKinsey &
Company. Retrieved from http://www.mckinsey.com/insights/business_technology/big_data_
the_next_frontier_for_innovation
Map my fitness. (2014). Map my fitness. About us. Retrieved September 13, 2014, from http://about.
mapmyfitness.com/about/company-history/
Map My Ride. (2014). Join over 20 million people getting fit on MapMyRide. Retrieved from http://www.
mapmyride.com/routes/
Marchal, F., Hackney, J. K., & Axhausen, K. W. (2005) Efficient map matching of large global positioning
system data sets: Tests on speed-monitoring experiment in Zürich Transporta- tion Research Record,
1935, 93–100.
Marx, V. (2013). Biology: The big challenges of big data. Nature, 498(7453), 255–260.
Maus, J. (2014). ODOT embarks on ‘big data’ project with purchase of Strava dataset. Retrieved from http://
bikeportland.org/2014/05/01/odot-embarks-on-big-data-project-with-purchase-of-strava-dataset-
105375
Menghini, G., Carrasco, N., Axhausen, K. W., & Schüssler, N. (2010). Route choice of cyclists in Zurich.
Transportation Research Part A: Policy and Practice, 44(9), 754 –765. doi:10.1016/j.tra.2010.07.008
 132 G. Romanillos et al.

Mitesh, S., Patel, M. D., MBA, M. S., & Hall, B. (2015). Wearable devices as facilitators, not drivers, of
health behavior change. The Journal of the American Medical Association, 13(5). Retrieved from http://
jama.jamanetwork.com/article.aspx?articleID=2089651
My fitness pal. (2014). Map my ride. Retrieved September 14, 2014, from https://www.myfitnesspal.
com/apps/show/184
National Bicycle and Pedestrian Documentation Project. (2009). Fact sheet and status report. Retrieved
Mars 15, 2015, from https://www.bikepeddocumentation.org/
National Bicycle and Pedestrian Documentation Project. (2015). Retrieved Mars 15, 2015, from https://
www.bikepeddocumentation.org/
Nielsen. (2014a). Hacking health: How consumers use smartphones and wearable tech to track their health.
Nielsen. Retrieved from http://www.nielsen.com/us/en/insights/news/2014/hacking-health-
how-consumers-use-smartphones-and-wearable-tech-to-track-their-health.html
Nielsen. (2014b). Tech-styles: Are consumers really interested in wearing tech on their sleeves? Retrieved from
http://www.nielsen.com/us/en/insights/news/2014/tech-styles-are-consumers-really-interested-
in-wearing-tech-on-their-sleeves.html
Nordback, K., & Janson, B. (2010). Automated bicycle counts. Transportation Research Record: Journal of
the Transportation Research Board, 2190(-1), 11 –18. doi:10.3141/2190-02
O’Brien, O. (2010). The bike share map. Retrieved from www.bikes.oobrien.com
Downloaded by [University of London] at 03:08 02 July 2016

O’Brien, O. (2013). Bike share map. Retrieved from http://oobrien.com/bikesharemap/

O’Brien, O., Cheshire, J., & Batty, M. (2013). Mining bicycle sharing data for generating insights into
sustainable transport systems. Journal of Transport Geography, 34, 262–273. doi:10.1016/j.jtrangeo.
2013.06.007
Ohmori, N. (2005). GPS mobile phone-based activity diary survey. Eastern Asia Society for Transportation
Studies, 5, 1104– 1115.
Open Data Utrecht. (2015). Retrieved Mars 15, 2015, from https://opendata.utrecht.nl/
Padgham, M. (2012). Human movement is both diffusive and directed. PLoS ONE, 7(5), e37754. doi:10.
1371/journal.pone.0037754
Peixoto, D. A., & Xie, L. (2013). Mining trajectory data. Retrieved from http://courses.cecs.anu.edu.au/
courses/CSPROJECTS/13S2/Reports/Alves_Peixoto_Douglas_Report.pdf
Pelletier, M. P., Trépanier, M., & Morency, C. (2011). Smart card data use in public transit: A literature
review. Transportation Research Part C: Emerging Technologies, 19(4), 557–568.
Rasmussen, T. K., Ingvardson, J. B., Halldórsdóttir, K., & Nielsen, O. A. (2013). Using wearable GPS
devices in travel surveys?: A case study in the Greater Copenhagen Area. Proceedings from the annual
transport conference at Aalborg University (pp. 1– 26). Aalborg, Denmark. Retrieved from http://
www.trafikdage.dk/papers_2013/188_ThomasKjaerRasmussen.pdf
Reddy, S., Shilton, K., Denisov, G., Cenizal, C., Estrin, D., & Srivastava, M. (2010). Biketastic: sensing
and mapping for better biking. In Proceedings of the SIGCHI Conference on Human Factors in Computing
Systems (pp. 1817– 1820). New York: ACM. Retrieved from http://dl.acm.org/citation.cfm?id=
1978942
Rietveld, P. (2000). The accessibility of railway stations?: The role of the bicycle in The Netherlands.
Transportation Research Part D, 5, 2– 6.
van Rijn, B. W. (2014) Van detectielussen naar fietsintensiteiten: Rijdend, afrijdend en roodrijders. Elst, IT&T.
Retrieved from http://www.it-t.nl/index.html
Rogers, S., & Papanikolopulos, N. P. (2000). Bicycle counter. Minneapolis, MN: Minnesota Department of
Transportation Office of Research & Strategic Services.
Romanillos, G. (2013). Huella ciclista de Madrid (Madrid cycle track). Retrieved from www.
huellaciclistademadrid.es
Romanillos, G. (2014). Analysing and mapping the cyclable city. A GPS-based analysis of the real and potential
bicycle use in Madrid. 15th International conference on information technology in landscape architec-
ture, Zurich, Switzerland.
Ryus, P., Laustsen, K. M., Proulx, F. R., Schneider, R. J., & Hull, T. (2014). Methods and technologies for
pedestrian and bicycle volume data collection (Vol. D). Retrieved from http://onlinepubs.trb.org/
onlinepubs/nchrp/nchrp_w205.pdf
Savage, N. (2010). Cycling through data. Communications of the ACM, 53(9), 16–17. doi:10.1145/1810891.
1810898
Schneider, R., Patten, R., & Toole, J. (2005). Case study analysis of pedestrian and bicycle data collection
in U.S. communities. Transportation Research Record: Journal of the Transportation Research Board, 1939(-
1), 77–90. doi:10.3141/1939-10
 Big Data and Cycling 133

Schuessler, N., & Axhausen, K. W. (2009a). Map-matching of GPS traces on high-resolution navigation net-
works using the Multiple Hypothesis Technique (MHT). Retrieved from http://www.baug.ethz.ch/ivt/
ivt/vpl/publications/reports/ab568.pdf
Schuessler, N., & Axhausen, K. W. (2009b). Processing raw data from global positioning systems
without additional information. Transportation Research Record: Journal of the Transportation Research
Board, 2105(-1), 28– 36. doi:10.3141/2105-04
Shen, L., & Stopher, P. R. (2014). Review of GPS travel survey and GPS data-Processing methods. Trans-
port Reviews, 34(3), 316– 334. doi:10.1080/01441647.2014.903530
Smith, W. (2014). Mobile interactive fitness technologies and the recreational experience of bicycling: A phenom-
enological exploration of the Strava community. Clemson University. Retrieved from http://tigerprints.
clemson.edu/all_theses/1951
Sparkes, M. (2014). GPS Big data: Making cities safer for cyclists. The telegraph. Retrieved from http://
www.telegraph.co.uk/technology/news/10818956/GPS-big-data-making-cities-safer-for-cyclists.
html
Stopher, P., Clifford, E., Zhang, J., & FitzGerald, C. (2008). Deducing mode and purpose from GPS data.
Institute of Transport and Logistics Studies. Retrieved from http://ws.econ.usyd.edu.au/itls/wp-
archive/itls-wp-08– 06.pdf
Strava. (2014a). Does Strava have enough data to provide a meaningful dataset? Retrieved from http://
Downloaded by [University of London] at 03:08 02 July 2016

metro.strava.com/thank-you/
Strava. (2014b). Strava Saturday heat map. Retrieved September 15, 2014, from http://www.strava.com/
saturday-heatmap#0|12|3|30.50000|-40.80000
Strava Labs. (2014). Strava Labs. Retrieved September 15, 2014, from http://labs.strava.com/heatmap/
#5/-110.69370/35.21986/blue/bike
Strava Metro. (2014). Strava Metro. Retrieved September 15, 2014, from www.metro.strava.com
Tockar, A. (2014). Riding with the stars: Passenger privacy in the NYC taxicab dataset. Retrieved from
http://research.neustar.biz/2014/09/15/riding-with-the-stars-passenger-privacy-in-the-nyc-
taxicab-dataset/.
Transport For London. (2014). Open data users. Retrieved from https://www.tfl.gov.uk/info-for/open-
data-users/
The Nielsen Company. (2014). The digital consumer report 2014. The Nielsen Company. Retrieved from
http://www.nielsen.com/us/en/insights/reports/2014/the-us-digital-consumer-report.html
Vogel, P., Greiser, T., & Mattfeld, D. C. (2011). Understanding Bike-Sharing Systems using Data Mining:
Exploring Activity Patterns. Procedia — Social and Behavioral Sciences, 20, 514–523. doi:10.1016/j.
sbspro.2011.08.058
Wagner, D. P. (1997) Lexington area travel data collection test: GPS for personal travel surveys. Final Report,
Office of Highway Policy Information and Office of Technology Applications, Federal Highway
Administration, Battelle Transport Division, Columbus, September 1997.
Wamsley, K. (2014). Optimal power-based cycling pacing strategies for Strava segments (Doctoral disser-
tation). Kutztown University of Pennsylvania, Kutztown, PA.
Wehner, M. (2014). Strava begins selling your data points, and no, you can’t opt-out. Tuaw. Retrieved from
http://www.tuaw.com/2014/05/23/strava-begins-selling-your-data-points-in-the-hopes-of-
creating/
Winter, M. (2012) Monitoren van fietsintensiteiten in Enschede. Enschede: gemeente Enschede en IT&T.
Retrieved from http://www.it-t.nl/index.html
Wood, J., Slingsby, A., & Dykes, J. (2011). Visualizing the dynamics of London’s bicycle-hire scheme.
Cartographica: The International Journal for Geographic Information and Geovisualization, 46(4), 239–251.
Zaltz Austwick, M., O’Brien, O., Strano, E., & Viana, M. (2013). The structure of spatial networks and
communities in bicycle sharing systems. PloS One, 8(9), e74685. doi:10.1371/journal.pone.0074685