Content uploaded by María Henar Salas-Olmedo
Author content
All content in this area was uploaded by María Henar Salas-Olmedo on Mar 17, 2018
Content may be subject to copyright.
Contents lists available at ScienceDirect
Transportation Research Part A
journal homepage: www.elsevier.com/locate/tra
Exploring the potential of mobile phone records and online route
planners for dynamic accessibility analysis
Pedro García-Albertos
a
, Miguel Picornell
a
, María Henar Salas-Olmedo
b
,
Javier Gutiérrez
b,⁎
a
KINEO Mobility Analytics S.L, Diego de León, 47, 28006 Madrid, Spain
b
Departamento de Geografía Humana, Universidad Complutense de Madrid, C/Profesor Aranguren, s/n, 28040 Madrid, Spain
ARTICLE INFO
Keywords:
Dynamic accessibility
Time-sensitive accessibility
Big Data
Mobile phone records
Online route planners
Madrid
ABSTRACT
Big Data sources offer new possibilities for urban mobility and accessibility studies. As people
carry out their activities in a city, they leave behind a digital fingerprint that can be used to
analyze the population’s daily mobility patterns and determine the exact times of travel between
points of origin and destination at different times of the day. These data present high spatial and
temporal resolution, and enable accurate and dynamic analysis of accessibility. The objective of
this study was to conduct a dynamic analysis of urban accessibility considering its two main
components: travel times and the attractiveness of destinations. To this end, we calculated travel
times between transport zones using the Google Maps API and constructed origin and destination
(OD) travel matrices from mobile phone records. Several scenarios were generated to analyze
dynamic accessibility and the separate influence of its two components. We also conducted a
cluster analysis to characterize transport zones according to their accessibility in each of the
scenarios and times of day considered. Our results indicate that these new sources of geolocated
data show considerable potential for use in time-sensitive accessibility studies, since they yield
more accurate and realistic information than static or partially dynamic analyses. Such in-
formation could help politicians take better decisions concerning transport and land use.
1. Introduction
Accessibility can be defined as the ease with which any land use activity can be reached from a location using a given trans-
portation system (Burns and Golob, 1976). Accessibility has been recognized as a key concept in transport planning. Since the 1990s,
attempts have been made to widen the scope of urban transportation planning from transport network performance to broader
environmental and social dimensions. In particular, there is a need to highlight accessibility rather than placing traditional emphasis
on increasing personal mobility (Coppola & Papa, 2013). Accessibility assessments are data intensive tasks, as they require a detailed
knowledge of the location of amenities, the performance of transport networks, and the behavior of individuals. Accessibility
measures can be used as a social indicator if they show the availability of social and economic opportunities for individuals (or groups
of individuals), i.e. the level of access to resources essential for human existence such as jobs, food, and health and social services,
together with the potential for social interaction with family and friends (Geurs and Van Wee, 2004).
Accessibility focuses on the importance of reaching desired destinations, such as shops, school, or work (Forth et al., 2013). This
https://doi.org/10.1016/j.tra.2018.02.008
⁎
Corresponding author.
E-mail addresses: pedro.garcia@kineo-analytics.com (P. García-Albertos), miguel.picornell@kineo-analytics.com (M. Picornell),
msalas01@ucm.es (M.H. Salas-Olmedo), javiergutierrez@ghis.ucm.es (J. Gutiérrez).
Transportation Research Part A xxx (xxxx) xxx–xxx
0965-8564/ © 2018 Elsevier Ltd. All rights reserved.
Please cite this article as: García-Albertos, P., Transportation Research Part A (2018), https://doi.org/10.1016/j.tra.2018.02.008
raises two questions. How can we determine the desired destinations? And how can we measure the ease of access to those desti-
nations in terms of travel times? These questions become more complex when one takes into account the temporal variation in the
two components of accessibility: both desired destinations and travel times alike vary according to the time of day, particularly
within urban areas. New data sources offer new avenues for answering both questions.
Traditionally, accessibility studies have considered estimated mean travel times to measure ease of access, and have used a proxy
to measure the attractiveness of destinations, such as employment or population, but have not considered temporal variations in these
two variables. Recently, several new sources of data have emerged, often related to the concept of ‘Big Data’. Examples include
geolocated mobile phone data, GPS data, and smart card data from public transport and social network data. The integration of these
new data sources into current models is a trending research area that is expected to yield an increase in the quality, amount, and
frequency of the information available, and therefore to enhance the efficiency and effectiveness of transport and urban planning
policies.
This paper presents an innovative method to extract valuable information from new data sources for use in accessibility studies.
First, the attractiveness of the desired destinations is measured using phone data records. OD (origin-destination) travel matrices are
constructed according to time slots and the number of trips with a destination in each transport zone is calculated. This yields a proxy
for the desired destinations. Next, travel times by car are obtained according to time slot using the Google Maps Application
Programming Interface (API). Then, time-sensitive accessibility indicators are constructed which take into account both sources of
temporal variation: travel times and destination attractiveness. Finally, a cluster analysis was performed in order to synthesize the
main accessibility characteristics of transport zones.
The remainder of this paper is structured as follows. Section 2presents a literature review on the use of new data sources in time-
sensitive urban accessibility, particularly by private transport. Section 3reports the methodology and data. Section 4contains an
analysis of the results and Section 5presents the main conclusions.
2. Dynamic accessibility indicators
Accessibility is a measure of people’s opportunity or ease of access to reach the activities they wish to engage in. It has been
analyzed using various indicators that encompass different aspects of this multifaceted concept (Handy and Niemeier, 1997;
Reggiani, 1998). Accessibility indicators are typically composed of two basic components; the cost of travel (determined by the
transport networks and the spatial distribution of travelers and opportunities) and the quality/quantity of opportunities (Páez et al.,
2012). Most urban accessibility studies have conducted a static analysis of these two components (for example, Gutierrez and Gomez,
1999; Martín et al., 2010; Chiarazzo et al., 2014). However, transport network conditions (e.g. traffic congestion or frequency of
public transport) and the attractiveness of destinations vary throughout the day. New Big Data sources provide new possibilities for
conducting dynamic analyses of accessibility (Geurs et al., 2016), since they enable the study of temporal variation in the two
components of accessibility throughout the day.
To date, very few accessibility studies have used new data sources dynamically. Of these, most have only conducted dynamic
analyses of travel times, while the opportunities available remained static. Several studies have used General Transit Feed
Specification (GTFS) files to calculate public transport travel times at different times of the day, considering temporal variation in
frequencies, in order to carry out time-sensitive accessibility analyses. Thus, Farber et al. (2014) obtained transit travel times from
each census block to its nearest supermarkets at different times of day in order to investigate food deserts and their changing shape
based on the time of day considered. Fransen et al. (2015) calculated levels of accessibility to key destinations at regular time
intervals in order to identify public transport gaps. Farber and Fu (2017) calculated the shortest path transit travel time between sets
of origins and destinations in a city, at all times of day, to investigate how transit travel times were impacted by service cuts and
expansions and the consequences of this for job accessibility.
Travel times by car can also be calculated considering dynamic traffic conditions to obtain more accurate accessibility mea-
surements (Li et al., 2011). Accessibility can be calculated dynamically using floating car data collected from mobile global posi-
tioning systems embedded in moving vehicles (Li et al., 2011; Møller-Jensen et al., 2012) or from data recorded by embedded loop
detectors and GPS (Owen and Levinson, 2015). Recently, accessibility studies have begun to incorporate new data sources generated
by companies that use Big Data analytics and location technologies to obtain trafficand speed data. Thus, Sweet et al. (2015) studied
the impact of congestion on accessibility using historical traffic service data provided by Inrix. Moya-Gómez and García-Palomares
(2015, 2017) used TomTom “Speed Profiles”data to create dynamic maps, revealing the impact of congestion on daily accessibility.
Accessibility varies throughout the day depending not only on variations in car or public transport travel times, but also on
changes in destination attractiveness. Temporal variation in this second component has been ignored in most dynamic studies of
accessibility. One exception was the study conducted by Boisjoly and El-Geneidy (2016). They calculated transit accessibility to jobs,
considering fluctuations in job availability (based on mobility survey data) and transit service (using GTFS data) throughout the day.
More recently, Moya-Gómez et al. (2017) analyzed urban accessibility dynamically using two Big Data sources: TomTom Speed
Profiles data to consider the performance of the transport network, and Twitter data to determine destination attractiveness. This
approach enabled them to obtain temporal accessibility profiles for each zone, identifying the influence of traffic congestion and the
effect of changes in population location (as a proxy of destination attractiveness).
The availability of information from mobile phone records opens up new opportunities for dynamic accessibility analysis. This
data source has been used to analyze mobility patterns and build travel matrices (see, for example, Caceres et al., 2007, Alexander
et al., 2015, Calabrese et al., 2013, Iqbal et al., 2014, Tool et al., 2015, Graells-Garrido and García, 2015), but not to feed accessibility
models. On the other hand, the availability of online travel planners allows travel times by car to be calculated according to time
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
2
slots. The best known of these route planners, used by millions of people around the world, is Google Maps, which stores car travel
time data collected by Google from smartphones. However, this route planner has scarcely been used in scientific research. Some
exceptions are Gu et al. (2010), Wang and Xu (2011) and Pilkington et al. (2017) in their papers on accessibility to health care
facilities, but they nevertheless adopt a static perspective. The only example of a dynamic approach to calculating travel times by
using the Google Maps API is a very recent study by Dumbliauskas et al. (2017), although this focuses on the procedure for data
extraction and not on accessibility analysis.
In the present study, we used new data sources to calculate the two components of accessibility dynamically. We employed the
Google Maps Directions API to calculate private transport travel times between transport zones. To study daily variations in desti-
nation attractiveness, we constructed OD matrices of travel between transport zones for each hour of the day using data from mobile
phone activity records. The underlying idea was that the number of trips ending in each transport zone was representative of the
degree of attractiveness of destinations at each hour of the day. Mobile phone activity records present clear advantages over Twitter
for this purpose. As can be seen in Lenormand et al. (2014), the mobile phone data sample is typically much larger than the
geolocated Twitter user sample, even when the Twitter data was collected for more than a year. Although both data sources give
similar results for recurrent mobility, the temporal granularity of mobile phone data allows leisure and other non-recurrent trips to be
studied. Moreover, some sampling bias is present in the Twitter user pool, as shown in Mislove et al. (2011). Due to its high
penetration rates, mobile phone data is less prone to these biases. Also, the data comes from one of the main mobile network
operators in Spain, which targets different segments of the population through a set of virtual network operators, thereby reducing
possible biases associated with mobile phone ownership.
3. Data and methodology
In the present study, we analyzed temporal variation in private vehicle accessibility in the municipality of Madrid (Spain) on an
average weekday in November 2014, considering the resident population aged over 16 years old. We used the transport zones defined
by the Madrid transport authority (Consorcio Regional de Transportes), which consist of 584 transport zones (see Fig. 5). The
municipality of Madrid has about 3 million inhabitants and is sufficiently large and varied to accommodate very diverse areas with
varying degrees of density and a range of land uses. In general, there is a higher concentration of activity and jobs in the center and
north of the city, whereas the south is predominantly residential. This configuration creates flows from the periphery to the center
and north in the morning, and in the opposite direction in the evening.
We used geolocated mobile phone data to identify the attractiveness of destinations and Google Maps API data to estimate car
travel times:
–The mobile phone data used for this study consisted of a set of anonymized call detail records provided by one of the main mobile
network operators in Spain, with a market share of around one third of the Spanish population. These data have traditionally been
stored for billing purposes, collecting information every time a mobile phone device is used to make or receive a phone call, send a
SMS, or connect to the Internet. This yields a temporal granularity of approximately one record per half an hour/per hour, which
is extremely useful for mobility studies. Information is stored about the time and the antenna to which the device is connected,
providing an indication of the geographical location of the device at a given moment. The spatial resolution is usually associated
with antenna coverage area, providing a location accuracy of a few hundred of meters in urban areas.
Mobile phone data present several other advantages over surveys: samples are usually orders of magnitude larger, data are
collected passively, which eliminates vague or incomplete answers, and data are already being stored by mobile network operators,
rendering the cost of data collection much lower than that of a survey with an equivalent sample size. However, information on the
characteristics of the trips provided by mobile phone records is not as complete as that provided by mobility surveys.
–OD travel times were calculated using the Google Maps Directions API. Travel times for each OD relation were calculated ac-
cording to time slot, to study the effect of peak hour congestion on travel time. When several route options were given, the
shortest route (in time) was chosen. The Google Maps API returns travel distance and time for each origin-destination (OD) pair at
a specified departure time. Google Maps stores historical data from mobile phones on the average travel time on a particular road
section at a specific time on specific days. When Android smartphone users turn on their devices and enable their GPS, the
smartphones send back anonymous bits of location data to the Google database. This data allows the company to identify
movement trajectories, make further analysis and provide anonymized aggregated information (speeds and travel times) to a
wider audience (Dumbliauskas et al., 2017).
To analyze temporal variation in accessibility, we considered three time slots: morning peak (between 9:00 and 10:00 h), off-peak
(between 12:00 and 13:00) and evening peak (between 19:00 and 20:00). The first reflects the end of peak time in the morning, with
a predominance of travel to work and educational institutions. The second presents much more fluid traffic conditions, with a wider
range of reasons for travel, such as shopping, errands, or visits to the doctor. The third, timed to coincide with peak time in the
evening, presents high levels of congestion and mostly consists of return journeys home or trips for shopping and leisure.
The method consisted of a three-step approach:
–First, dynamic OD travel matrices were constructed from geolocated mobile phone data. The procedure followed to extract these
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
3
OD matrices was based on the existing literature (Calabrese et al., (2013), Alexander et al., (2015) and Çolak et al., (2015)
amongst others). First, non-valid users were filtered out from the sample. Non-valid users were those who did not produce mobile
phone registers regularly enough for mobility information to be extracted from them. In particular, two time thresholds between
registers were defined: 8 h during the nighttime period (defined as from 8 pm to 7 am) and 4 h during the rest of the day. If mobile
phone users presented gaps without mobile phone registers longer than these time thresholds on the day of study, they were
removed from the valid sample. After the non-valid users were removed, the effective mobile phone sample remained at 10–15%
(depending on the region) of the Spanish Census population, which is still higher than the size of the sample of most surveys. Call
detail records were then analyzed to determine the location of the activities performed by the users during the day. This was done
using a clustering approach, following an adaptation of the SPD algorithm described in Gatica-Perez et al. (2010). Activities were
identified when mobile phone users remained during a certain amount of time (set to 30 min) within the same area. Trips were
defined as displacements between consecutive activities. However, the start time, end time and duration of the trips needed to be
corrected, as there is typically a gap between the real start time of an activity and the first register associated with that activity.
Trip duration was estimated under constant speed using linear trajectory analysis, as it is not always possible to determine mode
and route from mobile phone data in urban areas. Once trip duration had been estimated, start time was estimated through a
probability function based on a mobility survey. Sample information was then expanded to the total population by applying an
expansion factor based on the relationship between sample and total population at residential census tract level. Residential
location was estimated from mobile phone data following the methodology proposed by Picornell et al. (2015).
1
The number of trips with a destination in each transport zone expresses the desired destinations in each time slot and is therefore a
good proxy of the changing attractiveness of each zone. From a social perspective, it is important to note that temporal variation
in desired destinations refers not only to changes in the direction of these (journeys to places of work predominate in the morning,
whereas in the evening, trips for shopping, leisure or to return home predominate), but also in the intensity of the flow. Some
activities such as shopping are not accessible outside certain times. Furthermore, within the times when such activities are
accessible, individuals’“degree of desire”to access them changes over time. For example, shopping is usually done in the evening,
although shops and shopping centers are also open in the morning. This temporal variation can be captured through OD travel
matrices. In our study, we considered all journeys, including return journeys home, considering that these were also desired
destinations and that activities can also be performed at home.
–Second, dynamic OD travel time matrices for private transport were obtained from the Google Maps Directions API. Travel times
change during the day depending on historical traffic conditions, dynamically affecting accessibility. This temporal variation in
travel times is specific to each pair of transport zones. Some pairs present the worst congestion at peak time in the morning,
whereas others present it in the evening peak time. When choosing the coordinates to define a journey between two zones, we
used the zone centroids weighted with the resident population in each of their associated census blocks.
–Finally, a time-sensitive accessibility indicator was calculated using as input travel times between transport zones and the number
of trips with a destination in each zone. In this study, we used a location-based accessibility indicator (contour indicator) to
measure accessibility in Madrid according to time slot. The contour indicator is the simplest and most easily understood acces-
sibility measure (Curl et al., 2011). This enabled us show more clearly the role of temporal variation in travel times and mass of
destinations in accessibility values. The indicator is a cumulative measure that consists of calculating the number of opportunities
that can be reached from a node within a certain travel time (Gutierrez and Gomez, 1999), according to:
∑
=AWft·(
)
ijij
s.t.:
=⎧
⎨
⎩<
f
tif t travel time limit
otherwise
() 1
0
ij ij
where
A
i
is the accessibility of node i,
w
j
is the attraction mass of destination j,
and t
ij
is the travel time between i and j.
In our study, the attraction mass of each zone j was measured as the total number of trips with a destination in that transport zone
at that time of the day. Trip destinations should reflect where the desired opportunities were located in each of the time slots. The
travel time between i and j transport zones was calculated using the Google Maps API for this time slot, plus 10 min estimated for
access time walking from the origin of the trip to the parking space, the time to find a parking space and the access time walking from
1
The following formula was used in this study to identify frequent locations: : minimum_frequency = α·sample_days [1]. A location is considered as a frequent
location if the number of appearances in that location during a specific time period (sample days) is greater than a minimum threshold (minimum frequency). Home
location was estimated as the most frequent location between 8 p.m. and 7 a.m. during working days. The Alpha parameter (‘α’) was set to 0.2 in accordance with
previous validated studies (Picornell et al., 2015).
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
4
the parking space to the destination. We are aware that the time to find a parking space varies between different parts of the city.
However, since no data is available on this subject, we decided to use a fixed value for the whole city.
To facilitate comparisons between scenarios, we considered a single time limit of 30 min. This limit has frequently been used in
other cumulative accessibility studies in urban areas (see, for example, Gutierrez and Gomez, 1999; Martín et al., 2010). A lower time
limit would only encompass short trips, while with a higher time limit, most destinations could be reached from a wide range of
transport zones, rendering it difficult to accurately record variation in accessibility in the different transport zones.
The results obtained represent the variation in accessibility throughout the day. However, this variation depends on variations in
the two variables considered to calculate accessibility: travel times and the mass factor. To identify the influence of each of these two
components on variation in accessibility, we generated four scenarios for analysis (Fig. 1), in line with the proposal suggested by
Moya-Gómez et al. (2017):
(1) Reference scenario. Accessibility was calculated based on mean travel times and destination attractiveness throughout the day.
This was a static scenario taken as a reference to measure changes throughout the day based on temporal variations in congestion
and destination attractiveness within the metropolitan area.
(2) Dynamic congestion scenario. Accessibility was calculated for the three time slots considering variation in congestion levels,
while destination attractiveness remained static (mean spatial distribution of the destinations). This made it possible to isolate the
effect of variation in congestion levels on accessibility.
(3) Dynamic attractiveness scenario. Accessibility was calculated for the three time slots, but in this case only the variation in the
mass factor was considered dynamic, whereas congestion remained static (mean travel times). This made it possible to isolate the
effect of variation in destination attractiveness on accessibility.
(4) Dynamic accessibility scenario. Accessibility was calculated considering variation in congestion and destination attractiveness.
4. Results
4.1. Variations in travel times and destination attractiveness according to time of day
It should be noted that the daily mean for travel times and OD travel matrices includes night-time, when congestion and the
number of trips are very low. Therefore, the mean travel times in the three time slots considered were higher than the daily mean,
particularly in the morning and evening peak times. As an example, Fig. 2 shows the variation in travel times between the central
Travel times
Daily mean
Varies depending on time
of day
Destination
attractiveness
Daily mean
Scenario 1: Reference
scenario
Scenario 2: Dynamic
congeson scenario
Varies
depending on
time of day
Scenario 3: Dynamic
aracveness scenario
Scenario 4: Dynamic
accessibility scenario
Fig. 1. Analysis scenarios.
Fig. 2. Variation in mean travel times (in seconds) at different times of day with respect to the daily mean.
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
5
business district (AZCA district) and the airport at different times of day. It was therefore to be expected that the dynamic congestion
scenario would present a widespread loss of accessibility, especially in the morning and evening peak times.
However, congestion exerts a disparate effect on the different OD relationships (Fig. 3). Some relationships were strongly affected
by congestion, whereas others were much less affected. It was therefore to be expected that the dynamic congestion scenario would
generate a wide range of variations in accessibility depending on transport zone and the time of day.
The effect of variations in destination attractiveness on accessibility was exactly the opposite. In the three time slots considered,
not only did the direction of travel change (destination transport zones), but the total number of trips increased (and thus the
attractiveness of destinations) with respect to the daily mean, suggesting a stronger desire during these slots to reach destinations and
carry out activities than in the daily mean. For most of the relationships, the weight of destinations increased substantially with
regard to the mean, especially in the morning and evening peak times (Fig. 4). It was therefore to be expected that the dynamic
attractiveness scenario would present a widespread increase in accessibility with regard to the reference scenario, particularly in the
Fig. 3. Variation in travel times in the three slots considered with respect to mean daily times: frequency distribution of the different OD relationships and mean
values.
Fig. 4. Variation in destination attractiveness in the three slots considered with respect to mean daily attractiveness: frequency distribution of the different OD
relationships and mean values.
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
6
morning and evening peak times, albeit with significant differences according to transport zone.
4.2. Comparison between scenarios
First, we present the spatial distribution of accessibility in the reference scenario, i.e. considering a mean value for travel times
and destination attractiveness alike (Fig. 5). This figure shows the typical center-periphery pattern, since it is possible to access more
opportunities from central locations than from peripheral ones. However, some central transport zones presented lower accessibility
than others nearby because they had a low concentration of destinations (e.g., parks).
2
Conversely, some transport zones in the
periphery presented greater accessibility than others nearby because they contained areas of economic activity (such as offices or
shopping centers).
The dynamic accessibility scenario (scenario 4) showed variation in accessibility according to time of day (Table 1). As expected,
highest mean accessibility was recorded during the off-peak period (with an increase of 12.7% with respect to the daily mean) and the
lowest during the morning peak time (−21.1%). These accessibility values depended on congestion and destination attractiveness.
Scenario 2 (dynamic congestion scenario) enabled us to identify the impact of temporal variation in travel times by fixing the mass
factor. As can be seen in Table 2, congestion made a negative contribution to accessibility in the three time slots considered, which
was especially marked in the morning peak time (−36.4%). Scenario 3 (dynamic attractiveness scenario) showed the impact of
temporal variation in the mass factor by fixing travel times. In all three time slots considered, destination attractiveness made a
positive contribution to accessibility, which was particularly pronounced in the evening peak time (+48.9%) (Table 3). According to
the daily rhythms of the city obtained from the OD travel matrices, the population made more journeys at these times of day
(returning home and trips for leisure) than in the daily mean (which was heavily influenced by night-time), suggesting an increase in
destination attractiveness and therefore in dynamic accessibility.
The maps of differences between the dynamic accessibility scenario and the reference scenario in different time slots (Fig. 6a)
show that temporal variation in accessibility exerted a very disparate effect on different transport zones. Most of the peripheral
transport zones presented heavy losses in the morning peak time, whereas the central and northern zones maintained levels of
accessibility similar to the daily mean. In the off-peak morning period, most transport zones presented similar levels of accessibility to
the daily mean. In the evening peak time, many southern zones showed a substantial improvement in levels of accessibility, whereas
those in the north presented levels below the daily mean.
Fig. 5. Accessibility in scenario 1 (reference scenario).
2
In large parks, centroids are located deep inside the park, so travel times estimated by Google Maps include a large amount of walking time that distorts total travel
times compared to those in surrounding areas, making the accessibility of these parks lower than that of the surrounding transport zones.
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
7
Table 1
Available opportunities in scenario 4 and differences between scenarios 4 (dynamic accessibility) and 1 (reference) according to transport zones.
Scenario 1 Count Minimum Maximum Sum Mean Standard deviation CV
584 430 416,025 161,976,018 277,356 75,004 27.0
Scenario 4 Count: Minimum: Maximum: Sum: Mean: Standard deviation: CV
Morning 584 636 456,672 127,763,552 218,773 91,509 41.8
Off-peak 584 544 507,794 182,609,046 312,687 91,147 29.1
Evening 584 0 553,201 168,171,065 287,964 117,414 40.8
Differences Count: Minimum: Maximum: Sum: Mean: Standard deviation: CV
Morning —206 40,647 −34212466 −58583 16,505 14.8
Off-peak —114 91,769 20,633,028 35,331 16,143 2.1
Evening —−430 137,176 6,195,047 10,608 42,410 13.8
Differences% Count: Minimum: Maximum: Sum: Mean: Standard deviation: CV
Morning —47.9 9.8 −21.1 −21.1 22.0 54.8
Off-peak —26.5 22.1 12.7 12.7 21.5 7.8
Evening —−100.0 33.0 3.8 3.8 56.5 51.1
Table 2
Available opportunities in scenario 2 and differences between scenarios 2 (dynamic congestion) and 1 (reference) according to transport zones.
Scenario 1 Count Minimum Maximum Sum Mean Standard deviation CV
584 430 416,025 161,976,018 277,356 75,004 27.0
Scenario 2 Count: Minimum: Maximum: Sum: Mean: Standard deviation: CV
Morning 584 430 351,795 103,049,115 176,454 64,072 36.3
Off-peak 584 430 413,125 149,709,074 256,351 72,241 28.2
Evening 584 0 372,978 113,416,107 194,206 76,976 39.6
Differences Count: Minimum: Maximum: Sum: Mean: Standard deviation: CV
Morning —0−64230 −58926903 −100902 −10932 9.3
Off-peak —0−2900 −12266944 −21005 −2763 1.2
Evening —−430 −43047 −48559911 −83150 1972 12.6
Differences% Count: Minimum: Maximum: Sum: Mean: Standard deviation: CV
Morning —0.0 −15.4 −36.4 −36.4 −14.6 34.4
Off-peak —0.0 −0.7 −7.6 −7.6 −3.7 4.4
Evening —−100.0 −10.3 −30.0 −30.0 2.6 46.7
Table 3
Available opportunities in scenario 3 and differences between scenarios 3 (dynamic attractiveness) and 1 (reference) according to transport zones.
Scenario 1 Count Minimum Maximum Sum Mean Standard deviation CV
584 430 416,025 161,976,018 277,356 75,004 27.0
Scenario 3 Count: Minimum: Maximum: Sum: Mean: Standard deviation: CV
Morning 584 636 540,474 209,727,359 359,122 104,357 29.1
Off-peak 584 544 511,481 198,205,787 339,393 94,857 27.9
Evening 584 627 617,649 241,229,180 413,064 109,778 26.6
Differences Count: Minimum: Maximum: Sum: Mean: Standard deviation: CV
Morning —206 124,449 47,751,341 81,766 29,353 2.1
Off-peak —114 95,456 36,229,769 62,037 19,853 0.9
Evening —197 201,624 79,253,162 135,708 34,774 −0.4
Differences% Count: Minimum: Maximum: Sum: Mean: Standard deviation: CV
Morning —47.9 29.9 29.5 29.5 39.1 7.8
Off-peak —26.5 22.9 22.4 22.4 26.5 3.3
Evening —45.8 48.5 48.9 48.9 46.4 −1.5
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
8
Fig. 6. Changes in accessibility throughout the day according to scenario: (a) dynamic accessibility scenario; (b) dynamic congestion scenario; (c) dynamic destination
attractiveness scenario.
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
9
The explanation for these temporal changes in dynamic accessibility resides in the combined influence of variation in congestion
(Fig. 6b) and destination attractiveness (Fig. 6c):
–In the morning peak time, accessibility losses were higher in the periphery than in the center and northern center, since journeys
from the periphery to work in the center and northern center predominated. Thus, peripheral zones and in particular southern
peripheral zones presented high levels of congestion impeding access to the center and northern center. Conversely, central zones
presented easier access (less congestion) to peripheral zones. The center and northern center presented a slight drop in acces-
sibility due to the combination of a weak influence of congestion and a high concentration of destinations. In contrast, the
southern periphery evidenced a sharp drop in accessibility due to the combined action of high congestion and a low concentration
of destinations.
–In the off-peak morning period, most transport zones maintained levels of accessibility similar to the daily mean. During this
period, there was a widespread drop in congestion, with values similar to the daily mean, although levels of congestion in the
historic center remained fairly high. In contrast to the high concentration of workplace destinations in the morning peak time,
destinations were now much more widely dispersed and differences with respect to the daily mean were negligible in all transport
zones. Values similar to the daily mean in the dynamic congestion and dynamic destination attractiveness scenarios gave rise to
values similar to the daily mean in the dynamic accessibility scenario.
–In the evening peak time, many southern zones showed a substantial improvement in levels of accessibility, whereas those in the
north presented levels below the daily mean. The south now presented more destinations and a lower loss of accessibility due to
congestion than the north, since many of the journeys that occurred at this time were return journeys home from work. These
journeys now flowed from the north (where many jobs and activities were located) to the south (of a more residential nature),
causing north-south congestion in the city, whereas there was much less congestion in the south-north direction.
Inequalities in accessibility between different transport zones can be measured using the coefficient of variation. In the dynamic
accessibility scenario, the coefficient of variation was higher than in the reference scenario at all times, but particularly in the
morning peak time (54.8%) and evening peak time (51.1%) (Table 1), indicating that differences in accessibility between transport
zones increased. In the reference scenario, changes in accessibility at these times of day tended to favor zones with high accessibility
and penalize zones with low accessibility, thus accentuating differences between transport zones.
Once again, the explanation for changes in the dynamic accessibility scenario resides in changes in the components: dynamic
congestion and the dynamic attractiveness of destinations. In the dynamic congestion scenario (Table 2), the coefficient of variation
was higher than in the reference scenario in all three time slots, and especially in the morning and evening peak times, indicating that
congestion heightened inequalities in accessibility between transport zones. In fact, zones presenting the greatest accessibility losses
were located in the southern periphery (morning peak time) and north (evening peak time), with low mean daily levels of acces-
sibility. However, increased congestion did not imply a parallel increase in inequality in accessibility. A substantial increase in
congestion in the morning was accompanied by a lower increase in accessibility inequality than a somewhat smaller increase in
congestion in the evening, but more heavily concentrated in the northern transport zones. In the dynamic destination attractiveness
scenario, the percentage changes in the coefficient of variation were weak and indicated that at peak time in the morning, desired
destinations tended to be concentrated in zones with high accessibility (predominance of travel to work), while in the evening
(predominance of shopping, leisure and return home journeys), destinations were more dispersed and the coefficient of variation
barely changed. Therefore, changes in inequalities in accessibility in the dynamic accessibility scenario seemed to be due more to
variations in congestion than to changes in destination attractiveness.
4.3. Type of transport zone according to dynamic accessibility
We concluded the study with a cluster analysis to classify transport zones according to the accessibility values obtained for the
reference scenario (scenario 1) and the percentage differences between values in the dynamic congestion scenario (scenario 2) and
dynamic destination attractiveness scenario (scenario 3) with respect to the reference scenario, for each of the time slots considered.
3
To avoid duplication, this cluster analysis did not include scenario 4 (dynamic accessibility), since the influence of its components
were incorporated separately (scenarios 2 and 3). Six groups of transport zones were identified, characterized by the following
accessibility profiles (Figs. 7 and 8):
–Group 4 (orange). City center transport zones: these obtained high accessibility values in the reference scenario, constituted very
attractive destinations throughout the day and were less affected by congestion than the mean, an effect that increased as the day
progressed.
–Group 6 (brown). Historic city center transport zones: these obtained accessibility values that were slightly higher than the mean
in the reference scenario, and were especially affected by congestion during the morning off-peak time.
–Group 3 (green). Accessibility above the mean in the reference scenario, with particularly heavy congestion at peak time in the
morning and high attractiveness in the evening. The zones belonging to this group were predominantly located in the southern
3
The cluster analysis was performed in ArcGIS 10.4, using a k-means algorithm, with no spatial constraint. Although statistically, the optimum number of clusters
was 3, we selected a solution with 6 clusters that better described the dynamic accessibility characteristics of the transport zones.
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
10
center.
–Group 1 (blue). Southern peripheral zones presenting characteristics similar to those of the zones in Group 3. These zones were
particularly affected by congestion at peak time in the morning, but showed less accessibility than those in group 3 in the
reference scenario, and less destination attractiveness in the three time slots considered.
–Group 2 (red). Very peripheral areas with very low accessibility in the reference scenario and very low attractiveness as desti-
nations in all time slots.
–Group 5 (purple). Northern central zones with average accessibility in the reference scenario, attractive as destinations at peak
time in the morning and strongly affected by congestion at peak time in the evening.
–This classification synthesizes center-periphery and north-south differences between transport zones and makes it possible to
characterize them according to their different accessibility profiles.
5. Conclusions
Traditionally, accessibility has been analyzed statically, without considering the changes that occur throughout the day. New data
sources make it possible to address accessibility dynamically, considering different time slots. This is especially important in urban
accessibility studies, since private and public transport travel times in cities undergo large variations throughout the day due to
variations in congestion and frequencies, respectively. In addition, the attractiveness of destinations also presents considerable
temporal variation because the desirability of reaching one or another zone in the city changes substantially throughout the day, also
affecting traffic conditions.
Most of the time-sensitive accessibility studies published in recent years have been only partially dynamic, since they only
considered temporal variation in travel times. The present study employed an original approach to accessibility, considering temporal
variation in both its components using new data sources: Google Maps Directions API to calculate matrices for private transport travel
times between transport zones, and mobile phone records to construct OD travel matrices. The number of journeys to a destination in
each transport zone represented their time-sensitive desirability and was used as a proxy of the changing attractiveness of these
destinations.
A comparison between the dynamic accessibility and reference scenarios (mean values over 24 h) revealed substantial losses of
accessibility at peak time in the morning, considerable improvements at off-peak time in the morning and values near the mean at
peak time in the evening. The construction of partially dynamic accessibility scenarios revealed the influence of each of the com-
ponents. Thus, very steep losses of accessibility were identified due to the effect of congestion at peak times in the morning and
evening, which exerted a very disparate effect on the different transport zones depending on whether they were residential areas or
areas of activity. In contrast, destination attractiveness at the same times of day behaved in the opposite manner, whereby
Fig. 7. Transport zones according to clusters.
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
11
accessibility improved in general and particularly in the principal travel destination zones: the center at peak time in the morning and
the south at peak time in the evening.
Another interesting finding was that congestion substantially increased differences in accessibility between transport zones at
peak times, since the zones most affected by congestion were peripheral zones with low accessibility. The southern periphery showed
a steep loss of accessibility due to congestion at peak time in the morning, as did the northern periphery at peak time in the evening.
In contrast, changes in destination desirability only exerted a weak effect on differences between transport zones: differences in-
creased at peak time in the morning when destinations were concentrated in the center and north, and decreased slightly at peak time
in the evening, when destinations were more scattered. The cluster analysis identified type of transport zone according to accessi-
bility, synthesizing center-periphery and north-south differences in the municipality of Madrid.
The results of our analysis contribute to a better understanding of the problems and causes of low accessibility by car according to
transport zones and time slots, differentiating between problems related to the performance of the road network (dynamic congestion
scenario) and those related to land use (dynamic destination attractiveness scenario). These results should not lead decision-makers
to take measures to increase capacity in order to mitigate car accessibility problems, since it is well known that increasing capacity
attracts more traffic and generates more congestion and greater external costs in the long run. The value of this type of analysis for
managers and planners is that it offers insights into the effects of possible actions to be taken in transport and land use policies. Thus,
with respect to the implementation of traffic management measures (for example, an urban road toll for accessing the city center), it
would be possible to identify a priori which transport zones would be most affected by the measure, since we have an OD matrix
obtained from mobile phone data. The implementation of a road pricing scheme would lead to an increase in the demand for public
transport, which should react by increasing frequencies, thereby improving public transport accessibility within the city. Regarding
policies on land use, our model suggests that both new residential developments in areas with low car accessibility values during the
E2_E1_M: Differences between dynamic congeson scenario and reference scenario (morning)
E2_E1_OP: Differences between dynamic congeson scenario and reference scenario (off-peak)
E2_E1_E: Differences between dynamic congeson scenario and reference scenario (evening)
E3_E1_M: Differences between dynamic aracveness scenario and reference scenario (morning)
E3_E1_OP: Differences between dynamic aracveness scenario and reference scenario (off-peak)
E3_E1_E: Differences between dynamic aracveness scenario and reference scenario (evening)
Scen_1_30Min: Reference scenario
Fig. 8. Accessibility characteristics of the clusters. E2_E1_M: differences between dynamic congestion scenario and reference scenario (morning). E2_E1_OP: differ-
ences between dynamic congestion scenario and reference scenario (off-peak). E2_E1_E: differences between dynamic congestion scenario and reference scenario
(evening). E3_E1_M: differences between dynamic attractiveness scenario and reference scenario (morning). E3_E1_OP: differences between dynamic attractiveness
scenario and reference scenario (off-peak). E3_E1_E: differences between dynamic attractiveness scenario and reference scenario (evening). Scen_1_30Min: Reference
scenario.
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
12
morning peak and more offices in areas with low car accessibility values during the evening peak would aggravate congestion and
low accessibility problems in those areas. Policies aimed at reducing spatial mismatch in these transport zones could therefore be
appropriate, particularly Transit Oriented Developments (TOD), where mixed land uses and new transport infrastructures favor the
use of public transport and non-motorized modes.
These examples highlight the need to consider accessibility by public transport too. Due to space constraints, this paper is limited
to an analysis of private transport accessibility. In future research, we aim to conduct a dynamic study of public transport accessibility
using GTFS files to calculate public transport travel times, and data from mobile phone records to analyze variation in the attrac-
tiveness of transport zones. An analysis of the results obtained for public and private transport will provide a better understanding of
accessibility problems and enable politicians to take better decisions. In addition, the results of analyzing dynamic accessibility by
public transport could be combined with social indicators (e.g. income level by transport zone) to identify those groups which
experience the worst conditions of accessibility, paying particular attention to the most disadvantaged social groups (who are most
dependent on public transport). Another future line of research would be to conduct a time-sensitive analysis of public transport
network vulnerability to disruption, since this can affect public transport users very differently according to the time of day.
Acknowledgments
The authors gratefully acknowledge funding from the ICT Theme of the European Union's Seventh Framework Program (INSIGHT
project-Innovative Policy Modeling and Governance Tools for Sustainable Post-Crisis Urban Development, GA 611307), the SESAR
Joint Undertaking under grant agreement No. 699260 under European Union’s Horizon 2020 research and innovation programme,
the Spanish Ministry of Economy and Competitiveness and the European Regional Development Fund (TRA2015-65283-R and FPDI
2013/17001), and the Madrid Regional Government (SOCIALBIGDATA-CM, S2015/HUM-3427).
References
Alexander, L., Jiang, S., Murga, M., González, M.C., 2015. Origin–destination trips by purpose and time of day inferred from mobile phone data. Transport. Res. Part C:
Emerging Technol. 58, 240–250.
Boisjoly, G., El-Geneidy, A., 2016. Daily fluctuations in transit and job availability: a comparative assessment of time-sensitive accessibility measures. J. Transport
Geogr. 52, 73–81.
Burns, L.D., Golob, T.F., 1976. The role of accessibility in basic transportation choice behavior. Transportation 5 (2), 175–198.
Caceres, N., Wideberg, J.P., Benitez, F.G., 2007. Deriving origin–destination data from a mobile phone network. IET Intell. Transport Syst. 1 (1), 15–26.
Calabrese, F., Diao, M., Di Lorenzo, G., Ferreira, J., Ratti, C., 2013. Understanding individual mobility patterns from urban sensing data: a mobile phone trace example.
Transport. Res. Part C: Emerging Technol. 26, 301–313.
Chiarazzo, V., dell’Olio, L., Ibeas, Á., Ottomanelli, M., 2014. Modeling the effects of environmental impacts and accessibility on real estate prices in industrial cities.
Proc.-Soc. Behavioral Sci. 111, 460–469.
Çolak, S., Alexander, L.P., González, M., Guatimosim, A., 2015. Analyzing Cell Phone Location Data For Urban Travel: Current Methods, Limitations and Oportunities,
Transportation Research Board 94th Annual Meeting, 2015.
Coppola, P., Papa, E., 2013. Accessibility Planning tools for sustainable and integrated Land Use/Transport (LUT) development: an application to Rome. Proc.-Soc.
Behavioral Sci. 87, 133–146.
Curl, A., Nelson, J.D., Anable, J., 2011. Does accessibility planning address what matters? A review of current practice and practitioner perspectives. Res. Transport.
Business Manage. 2, 3–11.
Dumbliauskas, V., Grigonis, V., Barauskas, A., 2017. Application of Google-based data for travel time analysis: Kaunas city case study. PROMET-Traffic Transport. 29
(6), 613–621.
Farber, S., Morang, M.Z., Widener, M.J., 2014. Temporal variability in transit-based accessibility to supermarkets. Appl. Geogr. 53, 149–159.
Farber, S., Fu, L., 2017. Dynamic public transit accessibility using travel time cubes: comparing the effects of infrastructure (dis) investments over time. Comput.,
Environ. Urban Syst. 62, 30–40.
Gatica-Perez, D., Montoliu, R., Raul Montoliu, 2010. Discovering human places of interest from multimodal mobile phone data. In: Proceedings of the 9th International
Conference on Mobile and Ubiquitous Multimedia, Limassol, Cyprus, 2010.
Geurs, K.T., Van Wee, B., 2004. Accessibility evaluation of land-use and transport strategies: review and research directions. J. Transport Geogr. 12 (2), 127–140.
Geurs, K.T., Patuelli, R., Dentinho, T.P. (Eds.), 2016. Accessibility, Equity and Efficiency: Challenges for Transport and Public Services. Edward Elgar Publishing.
Graells-Garrido, E., García, J., (2015, December). Visual exploration of urban dynamics using mobile data. In: International Conference on Ubiquitous Computing and
Ambient Intelligence, Springer International Publishing. pp. 480–491.
Gu, W., Wang, X., McGregor, S.E., 2010. Optimization of preventive health care facility locations. Int. J. Health Geogr. 9 (1), 17.
Gutierrez, J., Gomez, G., 1999. The impact of orbital motorways on intra-metropolitan accessibility: the case of Madrid's M-40. J. Transport Geogr. 7 (1), 1–15.
Foth, N., Manaugh, K., El-Geneidy, A.M., 2013. Towards equitable transit: examining transit accessibility and social need in Toronto, Canada, 1996–2006. J. Transport
Geogr. 29, 1–10.
Fransen, K., Neutens, T., Farber, S., De Maeyer, P., Deruyter, G., Witlox, F., 2015. Identifying public transport gaps using time-dependent accessibility levels. J.
Transport Geogr. 48, 176–187.
Handy, S.L., Niemeier, D.A., 1997. Measuring accessibility: an exploration of issues and alternatives. Environ. Plann. A 29 (7), 1175–1194.
Iqbal, M.S., Choudhury, C.F., Wang, P., González, M.C., 2014. Development of origin–destination matrices using mobile phone call data. Transport. Res. Part C:
Emerging Technol. 40, 63–74.
Lenormand, M., Picornell, M., Garcia Cantu Ros, O., Tugores, A., Louail, T., Herranz, R., Barthelemy, M., Frias-Martinez, E., Ramasco, J., 2014. Cross-checking
different sources of mobility information. PLoS ONE 9.
Li, Q., Zhang, T., Wang, H., Zeng, Z., 2011. Dynamic accessibility mapping using floating car data: a network-constrained density estimation approach. J. Transport
Geogr. 19 (3), 379–393.
Martín, J.C., García-Palomares, J.C., Gutiérrez, J., y Román, C., 2010. Impacts of orbital motorways in metropolitan areas on accessibility: efficiency and equity. A case
study in Madrid. Transport and Land Use, 3(1), 67–84.
Mislove, A., Lehmann, S., Ahn, Y., Onnela, J., & Rosenquist, J., 2011. Understanding the demographics of twitter users. In: Proceedings of the Fifth International AAAI
Conference on Weblogs and Social Media. p. 11.
Møller-Jensen, L., Kofie, R.Y., Allotey, A.N., 2012. Measuring accessibility and congestion in Accra. Norsk Geografisk Tidsskrift-Norwegian J. Geogr. 66 (1), 52–60.
Moya-Gómez, B., García-Palomares, J.C., 2015. Working with the daily variation in infrastructure performance. The cases of Madrid and Barcelona. Eur. Transport Res.
Rev. 7 (2), 1–13. http://dx.doi.org/10.1007/s12544-015-0168-2. Article 20.
Moya-Gómez, B., García-Palomares, J.C., 2017. The daily dynamic potential accessibility by car in London on Wednesdays. J. Maps 13 (1), 31–39.
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
13
Moya-Gómez, B., Salas-Olmedo, M.H., García-Palomares, J.C., Gutiérrez, J., 2017. Dynamic accessibility using big data: the role of the changing conditions of network
congestion and destination attractiveness. Networks Spatial Econom. http://dx.doi.org/10.1007/s11067-017-9348-z.
Owen, A., Levinson, D.M., 2015. Modeling the commute mode share of transit using continuous accessibility to jobs. Transport. Res. Part A: Policy Practice 74,
110–122. http://dx.doi.org/10.1016/j.tra.2015.02.002.
Páez, A., Scott, D.M., Morency, C., 2012. Measuring accessibility: positive and normative implementations of various accessibility indicators. J. Transport Geogr. 25,
141–153.
Picornell, M., Ruiz, T., Lenormand, M., Ramasco, J., Dubernet, T., Frias-Martinez, E., 2015. Exploring the potential of phone call data to characterize the relationship
between social network and travel behavior. Transportation 42, 647–668.
Pilkington, H., Prunet, C., Blondel, B., Charreire, H., Combier, E., Le Vaillant, M., Zeitlin, J., 2017. Travel time to hospital for childbirth: comparing calculated versus
reported travel times in France. Maternal Child Health J. 1–10.
Reggiani, A., 1998. Accessibility, trade and location behaviour: an introduction. In: Reggiani, A. (Ed.), Accessibility, Trade and Location Behaviour. (Aldershot:
Ashgate), pp. 1–16.
Sweet, M.N., Harrison, C.J., Kanaroglou, P.S., 2015. Gridlock in the Greater Toronto Area: its geography and intensity during key periods. Appl. Geogr. 58, 167–178.
Toole, J.L., Colak, S., Sturt, B., Alexander, L.P., Evsukoff, A., González, M.C., 2015. The path most traveled: travel demand estimation using big data resources.
Transport. Res. Part C: Emerging Technol. 58, 162–177.
Wang, F., Xu, Y., 2011. Estimating O-D travel time matrix by Google Maps API: implementation, advantages, and implications. Ann. GIS 17 (4), 199–209.
P. García-Albertos et al. Transportation Research Part A xxx (xxxx) xxx–xxx
14
A preview of this full-text is provided by Springer Nature.
Content available from Networks and Spatial Economics
This content is subject to copyright. Terms and conditions apply.
