Population synthesis for long-distance travel demand simulations using mobile phone data

Analysis of long-distance travel demand has recently become more relevant. The reason is the growing share of traffic due to journeys related to remote activities, which are not part of daily life. In today’s mobile world, these journeys are responsible for almost 50 percent of traffic overall. Consequently, there is a need for reliable long-distance travel forecasting tools, such as agent-based simulation with a suitable synthetic population that also covers irregular long-distance travel demand. In addition to socio-demographic attributes, each agent requires information on the number of long-distance tours, and the purpose and duration of each tour. Accuracy of the synthetic population is crucial in order to obtain valid results from the simulations. We will show how existing data sources can be utilized to synthesize a population with these characteristics. Usually, two data sources are used to synthesize a population for agent-based simulations. First, travel surveys are performed to obtain detailed information on the persons and their travel behavior for a sample of persons. Second, official statistics (register data) provide information on the marginal totals. A fitting algorithm is then applied to create a population that matches the travel behavior reported in the survey data as well as the marginal totals of the register data. The most popular approach is the iterative proportional fitting algorithm, but also other approaches have been employed, e.g. Bayesian Networks. In case of long-distance travel behavior, these two data sources are not sufficient, because it is known that travel surveys under-report long-distance travel heavily. Therefore an additional data source is needed. We propose to add passively collected mobile phone data, which is based either on GPS or on GSM. Mobile phone data is helpful since the obtained information on long-distance travel behavior is more reliable than results from survey data. On the other hand, the available samples of phone data lack socio-demographic information. Thus, it can not fully replace the travel diary data. Potentially, phone data can be substituted by any reliable source on long-distance travel behavior.