Developing and Verifying Origin-Destination Matrices Using Mobile Phone Data: the LLITM Case
Reza Tolouei, AECOM, Nicolae Duduta, Telefonica, Pablo Alvarez, AECOM
This paper describes the process of developing trip matrices from mobile phone data addressing certain data limitations, and the outcome of the verification tests undertaken to independently investigate the suitability of the derived trip matrices.
The Leicester and Leicestershire Integrated Transport Model (LLITM) is a highly detailed multi-modal transport model of Leicester and Leicestershire in the UK, owned by Leicestershire County Council. It includes bus, rail, car, freight and walk and cycle modes of travel, as well as a detailed parking model in urban centres. As part of a major update to LLITM to develop the LLITM 2014 Base model, highway demand matrices are being developed using a combination of Roadside Interview data (RSIs) and mobile phone data.
Mobile phone positioning data (referred to as ‘mobile data’) is a data source that is starting to be used by the transport planning community to develop ‘prior’ demand matrices. However, significant challenges remain to exist in addressing certain limitations of mobile data for use in prior matrix development. These include disaggregation into modes, vehicle types and purposes, the granularity at which the data should be interpreted, potential bias associated with the expansion of the sample mobile data records, and the potential bias associated with a possible tendency for mobile data to under-report short trips.
AECOM and Telefónica have worked together collaboratively to develop and refine the approach to produce highway demand matrices from mobile data and to address some of the limitations listed above. This paper describes in details the process adopted, how the methodology has been refined to address certain limitations of the data, and the outcome of the verification tests undertaken to independently investigate the suitability of the derived demand matrices. In particular, we discuss the methodology used to expand the mobile data, taking into account the market share variation between geographies and across different social groups. We also describe how purposes and time periods are detected for mobile phone trips, to what extent vehicle types and modes could be distinguished, the appropriate spatial resolution at which data could be processed, and how under-reporting of short trips has been addressed.
Telefónica has developed a unique mobile insight solution called Smart Steps, which collects anonymises and aggregates mobile data, to understand how segments of the population collectively behave. Smart Steps has made significant progress in the UK transport sector, across Rail, Road and Air travel. Smart Steps analysts have gained significant sector experience by providing insights on numerous projects including rail franchise bids, rail operational insights, highway traffic flows and airport catchments. Smart Steps work is now increasingly being used to underpin transport models in the UK.
The experience gained by working on numerous projects, means that Smart Steps’ teams of analysts and data scientists have a clear and definitive understanding of the strengths and limitations of mobile data. The innovative Smart Steps dataset has proven exceptional at replacing other techniques and has been validated for use in many live projects for high profile customers.
As part of the LLITM model development, Leicestershire County Council commissioned a significant programme of data collection. This included RSI data for about 155 sites (about 100 of which were surveyed in 2013/2014), local household survey data, local planning data, traffic count data, and local bus electronic ticket machine (ETM) data. Availability of these together with information from 2011 Census data provided a unique opportunity to independently test, refine, and verify the trip matrices developed from mobile phone data. In particular, the RSI surveys collected information on individuals’ mobile network operators, allowing to investigate and, if necessary, to make adjustments for biases in mobile data processing.
The verification tests undertaken included comparison of mobile phone trip-ends and their distribution with local planning data, Census population and employment data, and Census Journey to Work data, as well as comparison of trip length profiles between mobile phone data and that derived from the household interview data. In particular, the RSI data were used to estimate total trips to a number of defined cordons within the study areas, by purpose and time period; these were compared with trips derived from mobile data in terms of absolute totals and their distributions. In undertaking these, statistical analysis techniques were used to test various hypotheses, taking into account uncertainties and errors in the data.
The outcome of verification tests showed a strong correlation between mobile data matrices and various sources of data, suggesting that the final expanded matrices from mobile data are unbiased and sufficiently represent level and pattern of travel within the study area.
Association for European Transport