RECENT DEVELOPMENTS OF BIG DATA IN THE DUTCH NATIONAL MODEL – Study with Mobile Phone Data



RECENT DEVELOPMENTS OF BIG DATA IN THE DUTCH NATIONAL MODEL – Study with Mobile Phone Data

Authors

Dusica Joksimovic, Dutch Ministry of Infrastructure and the Environment, Klaas Friso, DAT.Mobility, Jasper Keij, Mezuro

Description

This paper deals with the study of the use of mobile phone data in the Dutch national and regional traffic and transportation model which forecasts the long term future of mobility in general and the traffic and transport conditions in particular.

Abstract

This paper deals with the study of the use of mobile phone data in the Dutch national and regional traffic and transportation model which forecasts the long term future of mobility in general and the traffic and transport conditions in particular. This model is used for two main purposes: a) answering policy questions regarding e.g. major infrastructure investments, and b) assessing the effect of regional infrastructure projects in the exploration and planning phase.

The methodology of these models is based on the pivot point method. Growth factors are combined with base matrices that describe mobility in the base year of the model. These base matrices are estimated using a priori matrices which are derived from the synthetic model results and traffic counts, the national travel survey and roadside surveys.

The availability of mobile phone data is potentially an enormous enrichment for traffic and transport models.

The main research question of this study is: To what extent is it possible to increase the quality of the a-priori car matrix of the Dutch national model (LMS) by the enrichment with the mobile phone data? The focus is on the morning and evening peak hours, for the average work day.

In Phase 1 –Mobile phone data analysis- a comparison is made between the mobile phone data and the OViN data (National Travel Survey in the Netherlands). OViN is a survey which describes the mobility behavior of the Dutch population and is based on a sample of about 100,000 trips annually. For comparison, the mobile phone data set covers more than 1.25 billion trips per year. The main purpose of the analysis is to examine to what extent the mobility characteristics in the mobile phone data may or may not match those from the OViN. A comparison is also made between the mobile phone data and the a-priori matrix of the Dutch national model. Special attention is given to the distribution of a number of specific (difficult to model) origin-destination relations. For example, the municipality of Zoetermeer and Almere (which have special relationships with The Hague and Amsterdam respectively) and Amsterdam Schiphol airport.

In Phase 2 -Enrichment of a-priori matrix of the Dutch National Model- the distribution of the origin-destination relations in the a-priori matrix is corrected on the basis of the observed distribution according to the mobile phone data. Special attention is given to limitations because of privacy, short distance trips and distinction between car driver, car passenger and public transport. Because the zonal system of the model and the mobile phone data differ from each other, it was for this purpose necessary to aggregate both data sets so that the enrichment of the a priori O-D matrix can be performed. We have chosen the level of municipalities for the aggregation. The enriched a-priori matrices are then disaggregated to the level of zoning of the model in order to be able to perform traffic assignment.


In Phase 3 –Quality Assessment- the results of the enrichment are analyzed. Different aspects are considered: matrix totals, the comparison with traffic assignment results (using the traffic assignment model implemented in the LMS, Qblok), the matrix structure and the comparison with traffic counts.

The most significant improvement of the enrichment procedure is that by using mobile phone data it is possible to perform direct corrections in the O-D matrix at municipal level. Because destination choice is the least reliable component in the estimation of O-D matrices the usage of this big-data source is of high value.

The comparison with traffic counts does not clearly show an improvement of the assignment. The reason for this phenomena is the difference between the trip length frequencies of the a priori model and the mobile phone data. Further research has to be performed to handle this issue.

Publisher

Association for European Transport