A short-term forecasting model of transit demand and service
Urban transportation modeling procedures have traditionally been applied to long-term forecasting of highway traffic volumes. As a result, during the past decade, the transit industry has been keenly interested in analytical tools that would be suitable for short-term transit planning applications. In Florida, these applications include five-year forecasting of transit demand for Transit Development Plans, route-level analysis, scenario testing, and service simulation.
In response to this need, the Florida Department of Transportation funded a two-year research project beginning in 1996 to develop a short-term transit forecasting model. Operating at the level of individual routes, the model explicitly addresses the two-way interactive relationship between transit demand and service provision (hereafter, "supply"). This is important for two reasons. First, unlike roadway supply, transit supply may be changed (in response to demand) in relatively short time-frames. Then, not only is transit demand a function of supply, but supply is also a function of demand. Second, there may be several different transit service configurations that can meet the transit needs of an urban area.
A critical question is: which configuration is most cost-effective? A tool that iteratively modifies transit service attributes in response to demand patterns is needed to answer this question.
The figure shows the broad structure of the model. The model consists of two primary components. First, a set of statistical equations formulated using local demographic data and transit ridership and service data is used to estimate transit ridership and supply at the route level. In order to accommodate the two-way relationship between demand and supply, the computations proceed through several iterations until convergence is achieved. Convergence is said to have been achieved when the difference in results between two consecutive iterations is below a specified threshold value. The statistical equations provide estimates of ridership, duration of service, capacity (passengers per hour), and headway/frequency for individual routes.
Short-term transit demand and supply forecasting model structure
Second, a set of parameters is included in the model so that performance measures may be computed for each iteration. These measures include operating expense per vehicle revenue mile and hour, operating expense per passenger mile, passenger trips per vehicle revenue mile, revenue per passenger trip, and farebox recovery ratio. These statistics provide a measure of the effectiveness of the transit service configuration at each iteration of the model.
The model is linked with two Geographic Information System (GIS) software tools to provide a user-friendly and visually-powerful database environment within which a user can run the model. The two GIS packages that interface with the model are ArcView and TransCAD. Results of any iteration of the model may be displayed in either of the GIS packages. A user can click on a route of the GIS transit network, and a pop-up screen will show the service attributes, ridership, and performance measures for that route.
Any modification made to the transit network in the GIS will automatically modify the database underlying the model. These features facilitate convenient analysis and simulation of alternative service scenarios for short-term planning horizons.
The system of equations in the model is very similar to that developed by Peng et al. (1997). There are three equations. First, the demand equation predicts ridership on a route as a function of service attributes on that route, ridership on complementary and competing routes, and a series of demographic variables. Second, the supply equation predicts service attributes on a route at a given time point as a function of current ridership on the route, past ridership on the route, and a series of demographic variables. Finally, a third equation accounts for inter-route relationships by modeling ridership on competing routes as a function of service attributes of the subject and competing routes and demographic factors.
The system of equations has been estimated using transit demand and service data from Volusia County. These equations have been programmed into a spreadsheet environment so that the computations may be made conveniently and efficiently. The user, however, will interact with the model via a menu-driven graphical user interface that provides access to three different entities. The first entity is the input database that contains demographic information and route-level service attributes for the entire transit network system under analysis. The second entity is the model equation system. When the model is "run," the results of each iteration are written to an output database that can be visually displayed in ArcView or TransCAD; these GIS packages constitute the third entity.
The complete model system has been run and tested on the Volusia County transit network. The graphical user interface and GIS linkages are being refined in response to several suggestions that would make the model system better cater to the needs of the transit industry. The final product will be available to planning agencies in the state by mid-November.
For further information on this project, contact Dr. Ram M. Pendyala of the USF College of Engineering at (813) 974-1084, pendyala@eng.usf.edu, or Ike Ubaka, FDOT project manager, at (850) 414-4532.
Please direct all comments and/or questions regarding this page to webmaster@cutr.usf.edu .
Return to the CUTR Home Page