Center Identification Number:  77603       

 

Project Title:  Validating T-BEST Models with 100% APC Counts

 

Principal Investigator:        

 

Xuehao Chu, Senior Research Associate
Phone: (813) 974-9831
E-mail: xchu@cutr.usf.edu

 

Institution:                             

 

Center for Urban Transportation Research

University of South Florida

Fax: 813-974-5168

 

External Project Contact:     

 

Ike Ubaka

Public Transportation Office / Transit Planning

850-414-4532

E-Mail: ike.ubaka@dot.state.fl.us

 

 

I.  Project Objective/Problem Statement

FDOT’s Public Transit Office in recent years has invested heavily in developing a comprehensive transit demand forecasting model system. T-BEST is its third generation that makes several modeling advances. It models and forecasts transit boarding at the individual stop level. It separates direct boarding from transfer boarding for both modeling and forecasting. It explicitly treats inter-relationships in a transit network through measures of accessibility to opportunities for potential activity participation. More important, these modeling advances provide significant practical flexibility for transit service planning that has not been available before. T-BEST can be used to assess the patronage impacts of a variety of service changes, including operating strategies, schedule changes, alignment changes, system changes, and fare policy.

T-BEST’s current transit demand models are calibrated with estimated boarding volumes using a small APC sample from the Jacksonville area of Florida. The small sample of boarding data has greatly reduced the reliability of the estimation and validation of the current demand models in T-BEST. This small sample has also made it difficult to separate the morning and afternoon peaks on weekdays. The value of T-BEST can be greatly enhanced with better data on boarding volumes.

The primary objective of this research project is to use 100% counts of boarding at the individual stop level to re-calibrate the demand models in the current version of the Department’s Transit Boardings Estimation and Simulation Tool (T-BEST). As part of another NCTR research project, we now know that a number of transit agencies outside Florida in the country have a high level of penetration of automated passenger counters (APC) on their fleet. One of these agencies will be selected as the study site.

  

II.  Objectives/Tasks

The proposed scope of services for this project consists of the following tasks.

Task 1. Review T-BEST Arc Version 1.1

The research team will review the current version of T-BEST in terms of its modeling framework and its implementation. Several factors come into play in this consideration. One is lessons learned from developing the demand models for the current version of T-BEST. Another is differences in stop and route characteristics between what were included and what are desirable to be included. Some of these characteristics include modal technology, route type, special generators, etc. Some of these considerations will play a role in selecting the study area. A third consideration would be the separation of the morning and afternoon peaks for weekdays. The current version of T-BEST uses a combined weekday peak due to the small sample available.

Task 2. Select One Transit Agency

The research team will select one transit agency that has 100% or close to 100% boarding counts from APCs. This task includes several sequential components. The first is to identify several transit agencies in the nation that have archived data on 100% APC boarding counts for a period of at least six months. A starting point of this is the 11 transit agencies identified in a recent survey of APTS deployment by the Federal Transit Administration that have a high level of APC penetration on their fleet. The second is to determine the availability of desired stop and route characteristics. The third component is to determine the availability of supply data, including a stop database by route, direction, and other characteristics (e.g., special generators) and a service schedule data base. The fourth is to solicit the willingness of each agency to provide the data already available or to facilitate collecting additional data if necessary. The fourth component is whether sufficient data are available for separating the morning and afternoon peaks on weekdays. The last component is to select one agency that is willing to help and has the necessary data. When there is a tradeoff between the quality of APC data and the availability of system data, the priority will be on high quality APC data. The final selection of transit agencies will affect not only the quality of APC data but also how the boarding models are specified.

Task 3. Specify Boarding Models

The research team will specify the boarding models in terms of stop and route characteristics based on what are available with the selected agency and what are desirable to be implemented. In addition, the research team will specify separate models for the morning peak and the afternoon peak on weekdays. These specifications will serve as the guide for data collection.

Task 4. Collect Data

The research team will collect data from the selected agency about the transit system and its usage. Also census and employment data for the service area will be collected. In the case where high quality APC data are available but not system data in an electronic form, this task will manually create the necessary electronic form of the system data.

Task 5. Process Data

Once all the raw data are in place, the research team will then organize them into formats suitable for being used by T-BEST. The research team will apply T-BEST to the selected agency with the formatted data. The results of this application are a data set on the variables in the boarding models specified in Task 3 for model estimation.

Data processing in this task will be conditional on the timely and successful revision of the computational engine of T-BEST. Among other things, the Department is planning a separate research project that will revise the computational engine in the next version of T-BEST so that the morning peak and the afternoon peak are modeled and forecast separately. This revision of the computational engine is planned to start in early spring of 2005.

Task 6. Estimate Boarding Models

The research team will estimate the boarding models specified in Task 3 using the data sets from Task 5. Alternative specifications of the models will be tested. Their predictive ability will be tested as well using the out-of-sample approach.

As part of this task, the research team will also coordinate with the separate research project mentioned in Task 5, which will test and incorporate the new model into T-BEST, including separate morning and afternoon peaks on weekdays.

Task 7. Draft Final Report

The research team will prepare a draft final report to document the research efforts. CUTR will e-mail the draft final report to the FDOT Project Manager.

Task 8. Final Report

It is expected the FDOT Project Manager will provide his comments on the draft final report within two weeks of receiving the e-mailed document. After receiving comments from the FDOT Project Manager, CUTR will revise the draft final report accordingly. Once finalized, CUTR will deliver the final report to the FDOT Research Office along with a summary of the final report.

III. Deliverables

This project will have three deliverables: 1) a print version of the Final Report; 2) a CD containing an electronic version of the Final Report and an electronic version of a project summary; and 3) a PowerPoint Presentation of the project. CUTR will deliver 12 copies of the Final Report to the Research Office and 9 to the Transit Office. CUTR will also deliver the Research Office and the Transit Office one copy of the CD each.

 

IV.  Project Schedule

A project-duration of 18 months is proposed. There are several reasons for this relatively long duration. One relates to the need to identify agencies and obtain the commitment of cooperation. Another reason relates to the high degree of uncertainty in the amount of time that data collection may take. The third reason is that the task of data processing is conditional on the timely revision of the computational engine of T-BEST for separating the morning and afternoon peaks on weekdays. December 1, 2004 is the expected start.

 

 

Note:    T = Task efforts;           D = Deliverables

 

V.  Project Budget