introduction
The advent of mobile phones, in unprecedented ways, has revolutionized many aspects of modern life. In Germany, the number of registered mobile phone numbers increased from less than one million in 1992 to almost 137 million in 2018.
This means that in a country of more than 83 million people, most people own at least one cell phone and will most likely carry it with them most of the time.
The idea of using mobile phone location data for monitoring is not new. Academics and engineers in the automotive industry have been conducting so-called mobile mobility surveys.
Initially, these studies focused on using real-time traffic data to improve in-car navigation devices. Today, real-time traffic information is widely used in many advanced route planning and navigation applications.
Other disciplines are beginning to take advantage of these new possibilities to understand individual mobility behavior. For example, space planners now conduct mobility surveys for all modes of transportation, including public transit and active modes of travel (walking, cycling).
So far, only early adopters of new technologies are trying to leverage these new datasets in areas with privileged data access. The concept of space monitoring at the regional scale has not yet taken advantage of the possibilities gained from mobile phone network data to track the achievement of space policy objectives.
In Germany, research mostly relies on archived land-use statistics, topographic geographic data, and travel surveys. This data can often not be combined until three years later.
Analysis options are limited by the data. For example, inconsistent land-use classifications over time, limited or mismatched spatial resolution, or missing information about commuters from state social security systems that are not covered by statistics.
These shortcomings can now be overcome by using more timely cellular network data. With the resolution of legal issues related to data protection, the concept of integrating mobile phone network data for long-term monitoring is now feasible.
Background: Monitoring flow patterns
To plan transportation infrastructure, planners need to provide not only supply information about physical transportation networks, their capacity and usage patterns, but also about the mobility needs of local populations and economies.
Such information collection and management has evolved over the past few decades: from being the exclusive domain of traffic engineering and planning disciplines, it is now taking a more diverse interdisciplinary approach, including spatial planning disciplines and social sciences.
In this context, the way sustainable transport planning is done has led to a focus on travel behavior research. Planning practices now combine new insights into mobility demand management strategies with mobility demand research.
To support this enhanced version of sustainable transport planning, modern information systems need to integrate data on topics such as land use, transport infrastructure, demographics, and forecasting.
In the past, such information was collected from a variety of sources, sometimes involving costly and time-consuming manual labor (e.g., traffic counts, household surveys, service level assessments).
The so-called data-driven society sees new data-driven applications that generate a wealth of new information (big data) on a range of topics.
An important question is whether previously manually collected information can be replaced by equally frequent, quality, and deep automated data, such as cellular network data.
This will allow for a significant expansion of the geographical scope of transport planning studies, eliminating the limitations of the scope of traffic analysis and related monitoring activities that were previously limited by the cost and time required for data collection.
Data sources and data fusion
Telefónica Next provides aggregated and anonymized mobile phone network data, a start-end matrix based on nearly 45 million registered mobile phone numbers, 21 million customers and 8 billion network events.
When a mobile device communicates with a mobile phone base station, such as surfing the Internet or making a phone call, this database records the activity of the mobile device and anonymizes and processes it to identify the trajectory of movement and travel.
In further work, movements are classified by repeated patterns, identifying places of residence and work or paths to other destinations.
In order to responsibly carry out the statistical analysis of this mobility data, Telefónica Next has developed a specific anonymization procedure that complies with data protection regulations.
The spatial extent of mobile network traffic covers the whole of Germany, and postal codes are used as the spatial unit of traffic data. The vast spatial extent of the data provides the opportunity to explore patterns of movement between different types of urban settlement, such as traffic flows.
Travel data is provided as a comma-separated values (CSV) file with hourly time series. The data basically has five attributes, namely start area, end area, date, hour, and number of trips.
The terms start area and end area are used here to indicate the postal codes from which the trip starts and ends.
This data structure allows for aggregated queries by day, hour, and postal code, which means that travel patterns for morning and evening peaks or specific hours can be analyzed, and start and end areas of interest, as well as travel patterns on weekdays and weekends.
To detect patterns of spatial flow in data with postal codes as spatial units, and thus answer research questions about the functional relationship between start and end points, additional information about the nature and function of start and end points needs to be added.
A useful unit of representation for this purpose is the so-called RegioStaR classification (German: 'Regionalstatistische Raumtypen', English translation: 'regional statistics spatial types'), which was defined by the German Federal Ministry of Transport (BMVI Citation 2018) and used as an auxiliary data source for this study.
The new classification, published in mid-2018, aims to divide functionally homogeneous municipalities into spatial types, especially for research on transport and mobility planning.
Another requirement is to establish robust spatial types that can be updated and supplemented over time for monitoring purposes.
On this basis, trips were aggregated by RegioStaR category (1) rural areas and their subtypes (marginal rural areas and rural areas within urban areas), (2) small and medium-sized cities and their commuter areas (German: 'Regiopolitane Stadtregionen') and (3) larger metropolitan areas (German: 'Metropolitane Stadtregionen').
This allows researchers to detect typical flow patterns between these spatial types (such as the morning rush hour when commuters work downtown), or to identify situations where deviations from the assumptions of flow pattern theory are observed and need to be explained.
For this purpose, the raw mobile phone network data was linked to the RegioStaR classification of around 11,000 municipalities in Germany.
Note that phone data lacks information about transportation modes, which is an important attribute of any comprehensive travel behavior analysis. This is a weakness of mobile phone network data in the current state of development.
Theoretically, it should be possible to determine the mode of transportation by using phone location data to determine speed, that is, the time it takes to get from one location to another.
However, accuracy varies by region and mode of transportation; Distinguishing modes of transportation in dense downtown areas with many transportation options remains an open research question.
Aggregated and anonymized mobile data is stored as a table in a PostgreSQL database. Based on the details of the query execution plan, create indexes on the table to speed up the query.
Computationally intensive geoprocessing tasks, such as so-called spatial joins, are also performed in PostgreSQL using the functionality provided by its spatial database extender, PostGIS.
In addition to the disaggregated data from the BMVI, another important secondary data source is the German road network data from OpenStreetMap (OSM).
OSM data obtained from Geofabrik GmbH is preprocessed into routable shapefiles for graphical processing, such as routing or reachability analysis with ArcGIS Network Analyst.
OSM data is used to calculate the stroke length, i.e. the distance between the start and end points. Stroke length and stroke frequency are two mobility measures analyzed in the study.
The travel frequency between the start and end areas is obtained from the mobile data record, while the trip length is calculated using the Network Analyzer Extended in ArcMap 10.5.1, a commercial desktop GIS package developed by ESRI.
This software is often used by transportation planning departments and is therefore considered the easiest program for practitioners to use.
Data preparation and classification
As part of data preparation, the researchers employed basic data quality checking and aggregation procedures. The main purpose is to convert data from postal code units to spatial units relevant to the policy level.
In Germany, the most important policy levels designated by the use of space are regional plans and regional/local regional plans. The most important basic unit for monitoring the spatial policy of the relevant plans, and the objectives expressed in the text, is the municipality (according to the European Union nomenclature, LAU2 level).
Therefore, data conversion requires converting postcode data into municipalities, with no overlap in most cases. To overcome any eventual overlap, the researchers developed rules to redistribute the data according to the proportion of the area of the overlapping units.
Out of a total of 8,222 postal codes, 1,363 were found to cover more than one municipality. Only 211 of these span cities belonging to different regional classifications.
These zip codes are assigned to the regional classification of the cities with the largest overlapping areas, minimizing ambiguity.
The resulting spatial typology by zip code is then linked to the mobile traffic data table in the spatially functional PostgreSQL database. For performance reasons, indexes are required to improve query performance.
GIS-based network analysis
Frequency and spatial range constitute two key dimensions of mobility. In this sense, the researchers first measured frequency as the number of trips between the two regions, a pattern that differs at different times of the day and on different days of the week.
Spatial and temporal frequency patterns have multiple uses in urban research, for example, determining the attractiveness of an area (sometimes referred to as "gravity" from a modeling perspective).
To establish the distance between the start and end areas, travel routes were created in the street network provided by OSM, allowing the researchers to calculate the approximate length of each trip between the geometric center points of the start and end areas.
Calculating routes between 8222 (Germany ZIP Code) data points requires intensive processing. Even if not all possible combinations are practical (people can't travel long distances every day, and remote places don't go often), route calculations determine important performance considerations. Parallelization helps improve processing performance.
Discussion and future work
The organization and implementation of time-consuming and expensive specialized travel surveys can largely be replaced and supplemented by the analysis of mobile phone network data.
Users who are set to benefit from this data are transportation planners and traffic engineers, who rely heavily on traffic statistics to apply best-practice infrastructure planning and assessment procedures.
The data will also be useful to academics in disciplines related to transport and spatial planning, transport economics, academics working on local and regional policies, and planning practitioners, as well as policymakers concerned with the mobility transition, which is key to decarbonizing the transport sector.
Some studies have concluded that mobile phone network data captures flow trends better than any specialized travel survey.
This data is automatically collected by mobile phone network providers and is therefore readily available, at least for the spatial activity of a company's customer base.
As long as the sample size is large enough, the absolute number of trips can be extrapolated from the existing sample using location-specific predictions.
In this regard, the data used in the study by Telefónica cover a considerable market share, allowing for robust forecasting.
It covers the entire of Germany and its data format is easily a longitudinal data source for continuous monitoring. In contrast to existing studies focused on privileged data access in metropolitan areas, the study shows the concept of serving this continuous monitoring in spatial planning at the national, state, or regional level.
Data protection issues, which are usually covered by customer consent clauses that allow the use of anonymized data as part of a contract.
The General Data Protection Regulation (GDPR), which protects the anonymity of individuals, hinders the collection of family structure and demographic information, which is a "good option" from an analytics perspective.
Detailed information about mobility behaviour, which can be linked to social characteristics at the individual level, is likely to remain the domain of group studies and micro-census reports. For example, Germany and Switzerland have been regularly reporting on the movement of their respective populations since the seventies of the twentieth century.
Data collection is time-consuming and expensive, and that's where mobile phone network data comes in: it can supplement microcensus data frequently and in a timely manner with larger sample sizes.
Currently, there are still some weaknesses, for example in identifying the modes of transport used for travel. It is hoped that in the near future, it will be possible to improve the recognition algorithms of transportation modes by integrating additional transportation infrastructure data and information collected by sensors built into smartphones, such as accelerators or magnetometers.
conclusion
The study describes how such an analysis path can be set up and fine-tuned so that it can be performed within acceptable processing times. This technical backbone naturally requires data analysts who have been trained in such techniques. Getting this data ready for analysis is just one step.
Another step is to propose analytical concepts for spatial monitoring, enabling planners to identify, for example, mobility patterns in detail. In this context, examples are presented that require aggregation of data into spatial typology relevant to policy development in the transport sector.
Monitoring is essential to assess the effectiveness of these policies. From an analytical point of view, the new method is expected to take monitoring applications to a new level in this regard.
From an organizational perspective, datasets for such applications are currently only available from commercial vendors. Potential users will have to rely on their continuous availability in terms of data structure and price.
This risk must be assessed in comparison with official data sets provided by the Government (census data, regional statistics, geographical topography) in terms of the added value provided by the data set for monitoring purposes.
At the same time, there are issues that need to be addressed: in order for mobile network data to become the standard for transport monitoring, the mode of transport needs to be provided.
Data availability and consistency require the stability and comparability of time series over many years. There is ample evidence that these issues will be addressed in the near future, which means that the concepts and applications demonstrated by the Institute can be further developed and accepted by society.
Bibliography:
1.Ahas, R., A. Aasa, Y. Yuan, M. Raubal, Z. Smoreda, Y. Liu, C. Ziemlicki, M. Tiru, and M. Zook. 2015. “Everyday Space–Time Geographies: Using Mobile Phone-Based Sensor Data to Monitor Urban Activity in Harbin, Paris, and Tallinn.” International Journal of Geographical Information Science 29 (11): 2017–2039.
2.Andrienko, G., N. Andrienko, W. Chen, R. Maciejewski, and Y. Zhao. 2017. “Visual Analytics of Mobility and Transportation: State of the Art and Further Research Directions.” IEEE Transactions on Intelligent Transportation Systems 18 (8): 2232–2249.
3.BMVI (Federal Ministry of Transport, and Digital Infrastructure). 2018. RegioStaR Regional Statistical Spatial Typology for Mobility and Transport Research. July. Accessed May 2, 2019.
4.Böltken, F., and G. Stiens. 2002. "Settlement Structure and Territorial Categories." In: National Atlas of the Federal Republic of Germany. Villages and Cities, edited by Institut für Länderkunde Leipzig, 30.
5.Swiss Federal Statistical Office. 2017. Traffic behaviour of the population. Results of the Microcensus Mobility and Transport 2015. Accessed August, 29, 2019.