Research Article: Characterizing multicity urban traffic conditions using crowdsourced data

Date Published: March 12, 2019

Publisher: Public Library of Science

Author(s): Divya Jayakumar Nair, Flavien Gilles, Sai Chand, Neeraj Saxena, Vinayak Dixit, Yinhai Wang.


Road traffic congestion continues to manifest and propagate in cities around the world. The recent technological advancements in intelligent traveler information have a strong influence on the route choice behavior of drivers by enabling them to be more flexible in selecting their routes. Measuring traffic congestion in a city, understanding its spatial dispersion, and investigating whether the congestion patterns are stable (temporally, such as on a day-to-day basis) are critical to developing effective traffic management strategies. In this study, with the help of Google Maps API, we gather traffic speed data of 29 cities across the world over a 40-day period. We present generalized congestion and network stability metrics to compare congestion levels between these cities. We find that (a) traffic congestion is related to macroeconomic characteristics such as per capita income and population density of these cities, (b) congestion patterns are mostly stable on a day-to-day basis, and (c) the rate of spatial dispersion of congestion is smaller in congested cities, i.e. the spatial heterogeneity is less sensitive to increase in delays. This study compares the traffic conditions across global cities on a common datum using crowdsourced data which is becoming readily available for research purposes. This information can potentially assist practitioners to tailor macroscopic network congestion and reliability management policies. The comparison of different cities can also lead to benchmarking and standardization of the policies that have been used to date.

Partial Text

Growing metropolitan cities with strong economies continue to sprawl into suburbs with the desire for private transportation and on-time access to goods and services placing pressure on road networks. Traffic congestion and its repercussions are particularly alarming in developing countries as the existing transport infrastructure struggles to keep up with rapidly increasing productivity [1]. Congestion is characterized by slower speeds, increased travel times, elevated crash rates and rising emissions, placing an economic, social and environmental burden on communities. Over the last few decades, many researchers have provided varying definitions and metrics for traffic congestion that are centered on different traffic characteristics such as vehicle distance traveled [2], travel time [3], delay [4], speed [5,6], volume to capacity ratio [7], level of service [5], and travel cost [8]. Rao and Rao (2012) present an excellent review of the measuring methodologies adopted worldwide [8].

In this study, we gather traffic speed data for 29 major cities around the world for 40 days (between 9th March and 19th April 2017) using Google Maps. Due to the enormous size of these cities, only the road network surrounding the Central Business Districts (CBD) is selected for analysis. Google aggregates speed data from smartphone users to estimate speed and travel times. The data is recorded from devices which use the “Google Maps” feature or have their “My Location” feature turned on [29]. This data is processed into traffic data using different Application Programming Interface (API), developed by Google, such as distance matrix, directions, speed, and many more. These APIs provide functionality like data analysis, machine learning services (like prediction) or accessing user data (when permission to read the data is given). One of the key outputs from the Google APIs is the traffic speed data. The API calculates a representative speed value from the available crowdsourced data on a road link at any time of the day. Each link is identified by a unique string, called as the “place ID”. A road section is broken down into multiple place IDs at the locations where the road geometry or homogeneity changes (for example, merge and diverge points, intersections, etc.). The APIs provide real-time speed and free-flow speed information for a place ID or a collection of place IDs within a spatial block (based on the tile coordinate system defined by Google). Real-time speed is the speed at which the vehicles are moving on a road segment in the current traffic conditions. Free-flow speed is the typical speed at which a vehicle would be traveling on a road segment in the absence of any traffic. In this study, the link-level speed variation patterns are transformed into a generalized network level metric to compare various cities with different characteristics in terms of congestion index, the rate of spatial dispersion and network stability. Nair et al. (2019) conducted a study to get a clear understanding of the quality of crowdsourced data and how well the data matches the data from traditional sources [30]. The traffic speed data obtained from the Google API is assessed by comparing it with the loop detector traffic speed data for 3 state highways in the city of Portland, Oregon gathered from the Portland Bureau of Transportation and the floating car traffic speed data collected for selected arterial and collector roads in the city of Bandung and Cirebon, Indonesia. The analysis shows that the Google speed data reflects loop detector speed not only at the corridor level but also at a finer spatial resolution of road sections within a corridor. The RMSE value, which quantifies the deviation between the two speed profiles, is around 11 km/h for more than 90 percent of the 53 test locations. The speed comparison between google and floating car data also shows significant temporal similarity, with the difference between the two not exceeding 2 km/h. The results indicate no significant statistical difference between the data sources highlighting the potential of the crowdsourced data as a reliable and cost-effective means of traffic speed data collection.

In this section, we define a generalized congestion index to compare the traffic profiles of 29 cities. We then examine the impact of traffic congestion on these cities by analyzing the spatial heterogeneity, formation and dissipation and reliability of congestion. We also assess the existence of a stable network equilibrium condition across days.

Road traffic congestion has significant adverse impacts on the economy, environment, and society as a whole. Thus, measuring congestion and understanding its spatial and temporal patterns are critical to developing sustainable transport networks. In recent years, crowdsourced data has become increasingly popular among transport agencies as it provides easy accessibility to the way individuals travel in a road network. This study shows the utility of crowdsourced data to determine traffic conditions and, to the best of our knowledge, proposes a standard data source which facilitates comparison of traffic congestion across multiple cities of the world. This study proposed generalized congestion and stability metrics to compare and contrast speed variation patterns, congestion levels, its degree of formation and dissipation, spatial dispersion, and the stability of network equilibrium.




Leave a Reply

Your email address will not be published.