Modal OD Matrices Derived From Mobile Network Data

How many people travel from where to where, when, and using which mode? This has always been the crucial question in transport planning. The increasing usage of mobile phones now gives the authors a huge empirical data source to answer this question. In the past, building an OD-matrix has been a modelling effort with many uncertainties. Especially for individual road-based transport the traditional empirical data sources have not been sufficient to construct highly reliable OD-matrices: Cross-section road counts don't tell anything about origins and destinations and questionnaire samples are too small to fill complete OD-matrices. In 2019 DB FV (German Rail long-distance passenger transport) commissioned Teralytics to generate modal OD-matrices derived from strictly anonymized mobile network data. The decision for the new data source and the contractor Teralytics was the result of pilot projects over several years and an open competitive bidding with a test case focussing on mode detection. The matrices cover all trips within Germany longer than 5 km on the modes road, rail and air. Germany is split in nearly ten thousand traffic zones. International trips are cut at the border. Mode detection works well for trips longer than 30km. The data source is a feed of strictly anonymized data from a mobile network operator with a 40% market-share in Germany. The presentation will give an insight into matrix-building at Teralytics and data usage at DB. Challenges when processing mobile network data to generate OD-matrices are anonymization, trip and mode detection as well as extrapolation to the whole population of Germany. Anonymization by number is a crucial necessity: Information may only be extracted if it is true for at least 5 people. But more than 4 people rarely travel over a long distance from the same place to the same place on the same mode in the same time interval. So, to avoid a lack of data in long-distance transport, aggregated matrices are generated on different spatial and temporal resolutions. Also trip detection in noisy mobile network data is not as simple as it might seem. You have to decide whether a stop is a beak within a trip or an end of trip and a beginning of another one. Appropriate rules for such cases are completely different for short and long journeys. For DB, modal OD-matrices are an important base for planning, validation of and arguing towards rail supply improvements and extensions. These topics become more relevant nowadays since there is a political wish for a change towards more sustainable transport. The mobile network derived OD-matrices improve the knowledge especially for extensions of the rail network to regions where DB FV have no experiences and sales data so far. For supply planning, DB operates its own travel demand model, a pivot point model with a focus on mode choice and rail assignment. This model heavily depends on reliable OD-matrix input. In order to use the mobile network derived matrices within this model, they have to be processed in several ways: completion on the fine scale, split by trip purpose, extension of the international relations from/to foreign countries. Compared to the modelled OD-matrices used so far, the mobile network derived OD-matrices are updated more frequently and should be more consistent over time and between modes. There are two further potential application areas: estimation of mode choice models and measuring changes in the transport market. Both potential applications are subject to research and hopefully the authors can tell more about it in September.


  • English

Media Info

  • Media Type: Digital/other
  • Features: Bibliography; Figures; Tables;
  • Pagination: 11p
  • Monograph Title: European Transport Conference 2020

Subject/Index Terms

Filing Info

  • Accession Number: 01766218
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Feb 16 2021 2:15PM