https://doi.org/10.1177/0361198118777064
Transportation Research Record
1–11
© National Academy of Sciences:
Transportation Research Board 2018
Reprints and permissions:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0361198118777064
journals.sagepub.com/home/trr
TRR
JOURNAL OF THE TRANSPORTATION RESEARCH BOARD
Article
Travel demand models predict the number, destination, and
modal choice of trips. These characteristics depend on,
among other factors, the frequency and length of the trips.
This fact segments the demand into urban travel demand,
such as short-distance daily commuting, and long-distance
travel demand, such as overnight trips or tourism trips.
Different distance thresholds are used to distinguish between
the two demand segments, although long-distance trips gen-
erally exclude recurrent commuting trips.
Traditionally, transport modelers have paid more atten-
tion to daily short-distance urban traffic demand and its cor-
responding models, since these trips are much higher in
number than long-distance trips. Models for long-distance
trips appeared later and usually transferred parameters from
urban models. Some studies have started to highlight the
importance of the long-distance travel demand market
(sometimes mixed with the so-called intercity travel demand)
based on its contribution to the vehicle-kilometers traveled
(1), or motivated by the interest in analyzing high-speed rail
networks. The development of statewide models also
required specific modules for long-distance trips, which jus-
tified the need for their improvement. Data availability is
reported generally as the major concern for long-distance
models (2), since household travel surveys are typically
designed for urban travel demand.
This paper shows the development of a long-distance pas-
senger travel demand model for the province of Ontario,
Canada, as part of a province-wide transportation model. The
paper analyzes the limitations of survey data regarding non-
selected travel alternatives, and the acquisition of additional
datasets using location-based big data and online trip plan-
ning services.
Literature Review
Compared with short-distance or daily trips, the number of
research studies on long-distance travel demand is limited,
probably because the number of long-distance trips per
777064TRR XX X 10.1177/0361198118777064Transportation Research RecordLlorca et al
research-article 2018
1
Technical University of Munich, Munich, Germany
2
IVT, ETH Zürich, Zürich, Switzerland
3
PTV Group, Karlsruhe, Germany
Corresponding Author:
Address correspondence to Carlos Llorca: carlos.llorca@tum.de
Estimation of a Long-Distance Travel
Demand Model using Trip Surveys,
Location-Based Big Data, and Trip
Planning Services
Carlos Llorca
1
, Joseph Molloy
2
, Joanna Ji
3
, and Rolf Moeckel
1
Abstract
Long-distance trips are less frequent than short-distance urban trips, but contribute significantly to the total distance traveled,
and thus to congestion and transport-related emissions. This paper develops a long-distance travel demand model for the
province of Ontario, Canada. In this paper, long-distance demand includes non-recurrent overnight trips and daytrips longer
than 40 km, as defined by the Travel Survey for Residents in Canada (TSRC). We developed a microscopic discrete choice
model including trip generation, destination choice, and mode choice. The model was estimated using travel surveys, which
did not provide data about destination attractiveness and modal level of service. Therefore, a data collection method was
designed to obtain publicly available data from the location-based social network Foursquare and from the online trip planning
service Rome2rio. In the first case, Foursquare data characterized land uses and predominant activities of the destination
alternatives, by the number of user check-ins at different venue types (i.e., ski areas, outdoor or medical activities, etc.). In
the second case, the use of Rome2rio data described the modal alternatives for each observed trip. Combining data from
travel surveys, Foursquare, and Rome2rio, coefficients of the model were estimated econometrically. It was found that the
Foursquare data on number of check-ins at destinations was statistically significant, especially for leisure trips, and improved
the goodness of fit compared with models that only used population and employment. Additionally, Rome2rio mode-specific
variables were found to be significant for mode choice selection, making the resulting model sensitive to changes in travel
time, transit fares, or service frequencies.