OD-matricies of regular movements in Estonia
ESPON project:
Potentials of big data for integrated territorial policy development in the European growth corridors
Reference:
Aasa, A. (2019). OD-matrices of daily regular movements in Estonia [Data set]. University of Tartu, Mobility Lab. https://doi.org/10.23659/UTMOBLAB-1
Summary:
| aspect | content |
|---|---|
| Objective | To develop methodology for everyday mobility database which contains the OD-matrices of movements between territorial communities |
| Data used | OD-matrices are based on mobile positioning data. Mobile positioning data contains locations of call activities (Call Detail Records (CDR)) in network cells (location, time and random unique user ID). |
| Novelty of the approach | The mobile positioning data has high accuracy in time and space, the data allows to detect short-term differences (monthly) and seasonal fluctuations, dataset include regular movements between the place of residence and the workplace |
| Challenges | The main limitation of passive mobile positioning data is access to data, because mobile network operators are hesitant to provide their data and relatively long value chain of implementing mobile positioning data, which requires expertise from several research fields. |
| Policy implications | The resulting database would support mobility-related policymaking. |
Background & objectives
In this case study, the Partner of the Mobility Lab of University of Tartu is the Ministry of Economic Affairs and Communications of the Republic of Estonia. The objective of the Ministry is to develop high quality database of mobility and traffic data covering whole Estonia. This database would be an important data input for the Ministry in the spatial planning decision-making process to answer questions related to transport and mobility. Mobility Lab of University of Tartu is developing and implementing a methodology for everyday mobility database, which contains the OD-matrices of movements between territorial communities based on mobile positioning data. The developed database should be compatible with other mobility datasets developed by the Ministry.
Mobile positioning provides more accurate spatio-temporal information than the conventional methods, and data can be collected almost in real-time. Passive mobile positioning has two main strengths – longitudinality and extensive sample. More than 5 billion people in the world have a mobile phone and the number of its users is increasing GSMA Intelligence 2018. Mobile network operators continuously collect Call Detail Records (CDR) and datasets can cover several years. For example, the Mobility Lab of the University of Tartu has already a continuous 12-year time series of CDR data from Estonia.
The usage of mobile positioning data in mobility studies has been increasing over the years. Mobile positioning data provides spatio-temporal knowledge from multitude aspects – locations, flows, spaces – whereas both from a single individual and aggregated perspectives. Mobile positioning data reveals locations that an individual visits and important activity locations such as anchor points (places of residence and work/school) Ahas et al., 2010. This enables to examine both individual spatio-temporal mobility between activity locations Ahas et al., 2007 and individual’s daily and long-term activity spaces Järv et al., 2014. The scale up of individuals’ activity locations and movements reveals the dynamics of human presence and mobility of the whole population as well as within spatial structures. Aggregated individual mobility uncovers daily work-related commuting flows Ahas et al., 2010. Aggregated activity locations and flows unveils the urban spatio-temporal structures, land use and settlement hierarchies of a societies whereas even sheds light on how these structures change in time Ahas et al., 2015, Louail et al., 2015, Pei et al., 2014, Silm and Ahas 2010. This helps to identify and monitor hinterlands of cities, functional urban regions and growth corridors Novak et al., 2013. The mobile positioning data have been widely used in transportation and mobility studies (e.g. Calabrese et al. 2010; Isaacman et al. 2011; Yuan et al. 2012) as well as in geographical research (e.g. Blumenstock and Fratamico 2013; Silm and Ahas 2014). In Estonia, mobile positioning data has been used to map the commuting areas of municipalities by ordering the Ministry of the Interior (Ahas et. al 2010).
During the current project, Mobility Lab is developing an everyday mobility database. Study period extends from January 2016 to March 2018. Dataset will be prepared for every separate month. Based on earlier experience and preliminary analysis, the spatial granularity of the database is planned to remain on level of territorial communities. As the mobile network is very dense in Tallinn the spatial accuracy can probably be higher. Completed datasets will be in machine readable format (*.csv) and available directly from Mobility Lab web service.
Data & methods
Passive mobile positioning
The main data source for the current mobility database is passive mobile positioning. Passive mobile positioning data is automatically stored in the memory files of mobile operators for call activities or movements of handsets in the network. For current study we use database of the locations of call activities (Call Detail Records (CDR)) in network cells: the location, time and random unique ID. Passive mobile positioning data is usually collected with the precision of network cells. For the collection of passive mobile positioning data, mobile operators can aggregate anonymous geographical data from log files, ultimately not violating personal identity and privacy, and researchers can use it in surveys for scientific purposes.
Privacy concerns
Due to privacy issues, the database is anonymous for researchers and does not contain any back-traceable personal information about the user of the phone. To recognize a person, which is essential in order to analyse repeat visits and loyalty, a randomly generated unique ID number is assigned to every phone. Every respondent is given a unique ID, a numerical pseudonym that remains constant for every contract in the system, thus making it possible to analyse the digital trail of a phone’s physical presence in time and space.
The collecting, storage and processing of the data obtained complied with European Union requirements regarding the protection of personal data according to EU directives on handling personal data and the protection of privacy in the electronic communications sector.
Data
Mobile positioing data is collected and processed by Telia, Estonia’s largest mobile operator (which has ca 45% of the market share), the Mobility Lab in the Department of Geography at the University of Tartu and a spin-off company Positium LBS. The data are automatically saved and stored to the mobile operator’s memory in the form of log files.
On average, approximately 420,000 active respondents per month were noted whose home anchor points could be defined using the anchor point model (with a varying maximum of approximately ±10%). Within this study, the meaningful places for a respondent originate from the use of the anchor point model, which was developed by the Mobility Lab at the University of Tartu and Positium LBS. The model consists of eight steps that include determining the cells visited, cleaning the data and determining the anchor points Ahas et al. 2010. An anchor point is defined using the concept of actual activity locations at which people regularly stay or visit and from where they make phone calls because the anchor point model converts the locations of outbound calling activities into meaningful places (Ahas et al. 2010). Therefore, the model helps to assign locations that are meaningful to mobile phone users for every calendar month (Table 1), including the most likely home and work locations based on the respondent’s calling activities over time. This study uses home, work and multifunctional anchor points. Multifunctional anchor is defined as an everyday anchor point where the home and work-time locations are positioned at the same base station and therefore cannot be separately identified Ahas et al. 2010. Anchor points allow us to investigate meaningful locations and people’s daily activity spaces as well as more permanent moves such as changes of residence.
| Field | Description |
|---|---|
| Respondent ID | Unique numerical pseudonym of a respondent |
| Site ID | ID of an anchor point at the mobile site level |
| Timestamp | Month level |
| Location information | Longitude and latitude coordinates for a mobile site |
| Type of anchor | Possible values: home, work-time, multifunctional, secondary |
| Anchor ranking | Importance of anchor (according to days spent in specific location) |
The accuracy of spatial location information depends on the structure of the mobile tower network. The network cell level is also the minimum level of analysis. The structure of the cellular network and therefore the theoretical size of each base station are not fixed because the network changes with time due to the setup of new sites. In Estonia, more densely populated areas such as towns and the vicinities of the main roads are covered by more sites. In towns (such as Tallinn, Tartu and Pärnu), the spatial accuracy is 100–1000 m, whereas in the remainder of Estonia, the accuracy falls to between 1.5 km and 20 km. The average size of the theoretical service area in 2011 was 43 km2, and the median was 18 km2. Phones normally switch to the closest site or the one with the strongest radio coverage or the best ‘visibility’ levels. Because the network structure of sites is ambient in time, it is important to consider these modifications if the aim is to expand the time span under investigation to several years.
OD-matrix is aggregated to the level of neighbourhoods (sometimes also territorial communities). Neighbourhoods are used by municipalities and county goverments in their planning activities; these neighbourhoods usually have a unique local identity and are locally perceived as natural localities Mägi, K. et al, 2016. Current version of neighbourhoods is from year 2018 and obtained from Statistics Estonia.
OD-matrix of everyday mobility: Home - work commuting
For every unique user one home and one work location is defined per month. Everyday mobility is described as movement between home and work. Current dataset does not allow to construct real routes of movements and trips. Therefore the results are calculated as OD-matrix between neighbourhoods. Coordinates in OD-matrix are the centroids of the neighbourhoods most populated administrative unit. Estonian official coordinate system has been used L-EST97 epsg:3301.
The developed methodology and database is innovative in several aspects. The life’s pace has remarkably accelerated over the years and the 11-year periodicity of censuses is not enough to give a complete picture of the mobility of people as the start and the end of some phenomena (economic crisis, urban sprawl etc) might lay between two censuses. It means that the understanding of those processes stays in a very general level. Furthermore, the other official data sources may also give an insufficient or incorrect data about the mobility. For example, up to ¼ of home addresses registered in population registry do not correspond to the reality; traffic surveys have poor spatial coverage and do not usually give any information about the origin and destination of people.
Results
As the result of current work package are the monthly data files (csv) of origin-destination (OD) matricies.
data processing workflow:
Structure of OD-matrix:
| KANT_start | KANT_end | start_kant_id | end_kant_id | X_start | Y_start | X_end | Y_end | route_id | Population | RegularMovers |
|---|---|---|---|---|---|---|---|---|---|---|
| Aakre Valga | Aakre Valga | 1 | 1 | 627213 | 6440209 | 627213.0 | 6440209 | 1697 | 470 | 88 |
| Aakre Valga | Elva linn | 1 | 49 | 627213 | 6440209 | 642136.0 | 6456563 | 41506 | 470 | 10 |
| Aakre Valga | Elva ümbruse kant Tartu | 1 | 50 | 627213 | 6440209 | 639265.5 | 6453032 | 42353 | 470 | 1 |
| Aakre Valga | Helme Valga | 1 | 82 | 627213 | 6440209 | 611392.7 | 6429857 | 69457 | 470 | 2 |
| Aakre Valga | Hummuli Valga | 1 | 89 | 627213 | 6440209 | 619644.1 | 6418003 | 75386 | 470 | 1 |
| Aakre Valga | Jõgeveste Valga | 1 | 117 | 627213 | 6440209 | 617708.2 | 6430326 | 99102 | 470 | 3 |
Example of mapped regular movements (home - work):
Data download
Territorial communities layer from Statistics Estonia (*.shp)
OD-matricies:
Author: Anto Aasa
OD-matricies of regular movements
Last update: 2019-07-28 17:29:02
.