T
RAVEL-DEMAND modeling is evolving away from
relying on crude data aggregated at large urban zones
toward using highly disaggregated approaches where
individual travelers are modeled by interacting with fine-grained
spatial settings represented by parcel data. And the conceptual
framework that supports travel-demand modeling is moving away
from understanding travel as a series of trips toward a view of
people interacting with their surroundings as a series of activities.
This evolution in travel-demand modeling has given rise to analysis that combines highly detailed travel data collected through
global positioning systems (GPS) with techniques in geographic
information systems (GIS). Collecting travel-behavior data by
GPS offers several important advantages over conventional trip
diary surveys: GPS data can be collected over much longer
periods of time than the typical two-day diary; they do not rely
on the memory and estimates of a survey respondent; and they
provide linkages among complex trips, tours, and daily travel patterns. The most important advantage of GPS data is that they allow
us to address the dynamic properties of travel behavior by capturing detailed spatial, temporal, and attribute conditions throughout
the full length of the traveling experience. Unlike conventional
travel diaries that provide no information between origins and destinations, GPS data offer insights into the traveler’s choices and
decisions while en route. However, despite these advantages,GPS data present significant challenges that hinder their widespread adoption for travel behavior studies. The volume of data
is massive, and converting points of data into a meaningful
model of highly complex travel—with trip-chains of multiple
activities and purposes—makes for a cumbersome database
design.
This paper investigates driving behavior based on GPS data
collected by the University of Michigan Transportation Research
Institute (UMTRI). The database contains driving data for 78
drivers living in the Detroit metropolitan region in 2004, with
automobile use tracked on a day-to-day basis for four weeks,
with geographic positions captured every second by GPS. We
combine the GPS data with geocoded street addresses of business
establishments, land-use polygons, aerial photographs, census
data, and road attributes. The paper has two main objectives.
The first is to explain methodological challenges of converting
an enormous set of geocoded data points into a meaningful database that describes the complexity of trips and tours. The second
objective is to describe in a detailed manner the driving characteristics of a single driver over the course of a month of driving, to
illustrate the kinds of valuable lessons that transportation analysts
can learn from GPS data. We find that common travel patterns are
more complex than generally understood from traditional travel
surveys and that transportation engineers and planners can
benefit from GPS data used as a new technology for travel study.
Why Use GPS Data for Travel Behavior Research?
Transportation engineers and planners in the United States generally collect data on daily travel patterns using self-reported
written diaries and telephone surveys. These conventional travel
surveys have the advantage of being fairly straightforward to
administer and the data collected are easy to manage. Selfreported surveys are particularly useful for travel behavior
studies because a person can describe the exact nature of the
purpose for taking a trip, such as to go shopping, to visit friends,
or to eat a meal at a restaurant.
But conventional travel surveys have several serious limitations for travel behavior research. First, the self-reporting of
data is known to be unreliable. People typically underreport
short trips, and underestimate trip durations and misrepresent the
time that a trip starts and ends. Trip destination locations arereported inconsistently, such as listing the nearest main intersection when a street address is unknown. A second disadvantage is
that self-reported surveys are collected over very short time
periods, typically over two days. Third, these self-reported
surveys fail to capture important spatial information about trips
because they collect data on individual trips by aggregating
them to traffic analysis zones (TAZs) for analysis and modeling.
Points of origin and destination for each trip are coded to a
TAZ, so that travel behavior at small scales within a TAZ is
lost. Furthermore, this method of limiting data collection to trip
end locations leaves no information about travel behavior
between origin and destination, so that the actual route traveled
between TAZs is unknown. Transportation planners must resort
to such methods as route choice modeling or shortest path networks to ascertain the route between TAZs.