How much information can be obtained from a survey? Challenges faced and lessons learned from using on-board and household travel surveys for validating a dynamic transit assignment model
Corresponding Author: Andisheh Ranjbari, Puget Sound Regional Council
Presented By: Andisheh Ranjbari, University of Arizona
Abstract
Supported by the SHRP2 C10 Implementation Assistance Program, the Metropolitan Transportation Commission (MTC), San Francisco County Transportation Authority (SFCTA) and Puget Sound Regional Council (Puget Sound Regional Council (PSRC) are developing an applications-ready person-based dynamic transit passenger assignment component into their regional transportation planning models. The underlying open-source software is based on the Fast-Trips software, which uses the trip-based hyperpath algorithm to find and evaluate paths, and then simulates transit passenger route-finding and user experiences. Key to this network model being a useful improvement, is the development of variables and associated parameters within the route choice model that are sensitive to policy variables of current interest as well as a validation and model performance that make it a plausible tool for representing individuals route choices. This presentation discusses the observed data needs (and realities) necessary to effectively evaluate dynamic transit model performance.
The development team’s applications-ready approach leveraged initial route choice parameters derived from the existing SFCTA estimated mode choice models, and calibrated these parameters to meet the performance targets. To support this approach the team used observed data from on-board surveys (OBS) 16 agencies conducted in the Bay Area between 2012 and 2014 complemented with a trip GPS sample from the 2013 California Household Travel Survey (CHTS) and automated passenger counter (APC) data. The surveys were converted into the Fast-Trips demand (“dyno-demand”) and route (“dyno-path”) data standards in order to run the model with the demand from each survey and compare the outputs with observed data, and APC data was used to validate the assignment of regional demand.
Due to differences and gaps in the observed data, the conversions from observed data to both dyno-demand and dyno-path formats required substantial data wrangling and assumptions. This talk will focus on the data needs, challenges faced and lessons learned that can improve on-board and household surveys’ ability to support calibration and validation of dynamic transit passenger assignment. There were certain observed data the authors found very useful, and there were missing, unclear, and overly aggregate data that caused substantial data wrangling and introduced inaccuracies because of required assumptions. Moreover, the authors will share their experience in extracting valuable information from these observed datasets, and through linking them with other existing data which might not seem very obvious at the beginning, but could be obtained by spending more time and effort.