Origin-Destination Trip Matrix Development: Conventional Methods Vs. Use of Mobile Phone Data
Nominated for The Neil Mansfield Award
Reza Tolouei, AECOM, Stefanos Psarras, AECOM, Rawle Prince, AECOM
Performance of trip matrices derived from mobile data and conventional methods (RSI data) are compared and errors are quantified.
Conventionally, trip matrices have been derived by a complex process involving a combination of roadside interviews (RSIs) (to observe movements across defined screenlines) and the application of trip-end and gravity models (to extrapolate and infill unobserved movements), followed by matrix estimation methods to incorporate evidence from supplementary traffic counts. This method is widely accepted as the preferred approach to using synthetic matrices. More recently, mobile phone positioning data (referred to as ‘mobile data’) are being used increasingly by the transport planning community to develop ‘prior’ demand matrices as an alternative approach to RSI data or synthetic matrices.
There are a number of known strengths and weaknesses associated with both RSI data and mobile data; these have been documented in detail in a number of recent studies (for example, see Tolouei, et. al., 2016). However, there is lack of robust evidence as to whether use of mobile data results in a matrix that has an overall better or worse performance than matrices developed using alternative, conventional methods (i.e. RSI data or synthetic methods based on gravity modelling). In this study, we provide such evidence through a structured and systematic comparison of three sets of trip matrices developed using mobile data, RSI data, and a gravity modelling approach (i.e. synthetic matrices).
AECOM is undertaking a major update to the Leicester and Leicestershire Integrated Transport Model (LLITM) to develop the LLITM 2014 Base model. As part of this update, highway demand matrices have been developed using mobile data (processed by Telefonica into origin-destination trip matrices). For the purpose of the study and prior to the development of mobile data matrices, Leicestershire County Council commissioned a significant programme of data collection, including RSI data for 155 sites (about 100 of which were surveyed in 2013/2014), local planning data, household survey data, and extensive traffic count data. This provides a unique opportunity to develop trip matrices using the RSI data and compare these with the mobile data matrices, bothof which have a consistent geographical scope and represent demand pattern for the same time period. Different properties of the matrices are compared with a range of independent data, taking into account uncertainties and errors in the data, based on statistical analysis techniques.
The LLITM highway assignment model is used to assign the developed matrices onto the highway network. The assigned flows are compared with traffic counts across a range of independent long screenlines (i.e. not used in the processing of RSI data). The sampling error in the count data is quantified and comparisons are made taking into account the 95% confidence interval of the count data. This is followed by a discussion on the suitability of current UK WebTAG validation criteria, given the scale and range of errors in the count data. Whilst the above test is the key focus of the study, the following aspects of the matrices are also compared with independent observed data:
• correlation of trip-ends with estimates based on the National Trip-End Model (NTEM);
• consistency of trip rates with estimates based on National Travel Survey (NTS) data; and
• comparison of trip length distributions with estimates from the NTS.
The results reported in this paper suggest that, overall, the outcome of using the mobile phone data, when systematically refined using independent data sources to address various known limitations, does not look to be either biased or less accurate than the conventional method, using RSI data. In the areas of the model where no RSI data or other similar observed data are available, use of mobile data could even result in a more consistent estimate of trips, benefiting from a significantly larger sample size.
Tolouei, R., Álvarez, P., Duduta, N. (2015). “Developing and Verifying Origin-Destination Matrices using Mobile Phone Data: The LLITM Case”, proceedings of the European Transport Conference, Frankfurt, 2015.
Association for European Transport