Potential of "passive" mobile phone dataset to construct origin-destination matrix

Abstract

Transport constitutes a strategic tool for urban policies. Travel behavior data accuracy is necessary to ensure a durable development and management of infrastructures. Among data needs, origin-destination matrices are particularly relevant for modeling purposes. But these matrices are generally issued from travel survey and/or traffic counts and have some limits. The aim of the proposed paper is to test the potential of passive mobile phone data to produce accurate origin-destination matrices. Mobile phone operators produce enormous quantity of data in order to locate in real time mobile phone for internal operational purposes. In this paper we present applications performed with a dataset (communication events + handover and LAU) collected by Orange operator from March 31 to April 11, 2009 for the whole Paris Region. It contains 1.554 million data from about 4 million devices. Data allow producing pseudo real time spatiotemporal trajectory of devices. But trajectory is not sufficient to determine trips. We need to introduce assumptions to generate origin and destination. Trip definition corresponds to a movement in space with an activity at origin and destination. An activity is stationery and need time to be performed. We therefore define a time threshold within a LA to define an origin or a destination. Paper first describes matrices construction. In order to test the accuracy of mobile phone trip matrices we confront them with matrices built from database available at conurbation level like Paris Region. Every 10 years, Paris Region transport authority (STIF) performs a household travel survey. The last one occurred in 2010. 18.000 households have been surveyed which generates 150.000 trips within Paris Region at 100 meters grid zoning level. The two datasets allow producing origin-destination matrices at LA zoning level. We use GIS data to match different zoning systems. Paper present the comparison of the different matrices produces from mobile phone on one hand and travel surveys on the other hand. Results appear promising for the comparison with household travel survey data for which we obtain rather good matching for each cell of origin-destination matrix. Paper also presents data limits and perspectives to improve matrix construction and comparison methodology.