Abstract
In many urban areas there is increased interest in using travel time and its reliability to assess the level of service provided by surface transportation. This is straightforward on urban freeways with extensive systems of location-based sensors but becomes problematic off the freeways. Traffic volume and flow, collected by permanent and portable traffic counters, are arguably the measures of traffic conditions most commonly available on arterials. These could become a useful part of a travel time monitoring system if they could be converted to plausible estimates of travel time. The objective of this study is to compare several travel time estimation models given different data availability scenarios and to identify the most appropriate one to support estimation of networkwide arterial travel times. Actual link travel time data were collected by the modified license plate method from a sample of 50 arterial links in the Twin Cities, Minnesota, metropolitan area. The lengths of each link, measurements of traffic volume, and signal timing information were also obtained. Four candidate models were evaluated, including the Bureau of Public Roads function, the conical volume–delay function, the Singapore model, and the Skabardonis–Dowling model. It was found that (a) when the site-specific running time is known but the signal timing is not available, the four models produced similar mean absolute percentage errors (MAPEs) from 14.7% to 17.4% and (b) the Singapore and Skabardonis–Dowling models, using maximal site-specific information, produced MAPE values of 6.9% and 6.3%, respectively. Using the Skabardonis–Dowling model is suggested both because of its flexibility in accommodating different data availability conditions and because it tended to have the smallest MAPE values in most data availability scenarios.