What will afternoon rush hour be like on State College's North Atherton Street in the year 2006, or 2016? How would a new strip mall or three affect the flow of fall football traffic along Route 26?

Accurately predicting the future is a tricky problem for city planners -- and for traffic engineers.

Konstadinos Goulias is trying to get a little closer to reality by looking at the bigger picture.

"In the past, engineers might have looked at how many trips are made by a given household," says Goulias, assistant professor of civil and environmental engineering. "What we're doing is to take a step back to the level of behavior. We're trying to predict traffic patterns by understanding the behavior behind them."

Goulias and graduate student Jun Ma accessed a large data set gathered by the Puget Sound Travel Panel, a sample of Seattle residents who were asked by the Puget Sound Regional Council to keep daily travel diaries over a two-day period in 1989, 1990, 1992, 1993, and 1994, listing not just when they went out and where, but why, by what route, and by what mode of transport. From patterns of activity, they hoped to discern travel behavior.

Pretty far from State College, but "Longitudinal data of this type is rare," Goulias explains. "And to make a prediction realistic, you've got to incorporate change over time -- not just day to day, but year to year."

Then there's another obstacle: How do you fit the variety of human travel behavior -- from commuters to joyriders, from Rolls-Royce-riding mustard borrowers to juggling unicyclists -- into a computer model?

Goulias expected individuality to be more of a problem than it turned out to be. Using the statistical method of cluster analysis, however, he and Ma found that they could place most people into one of four behavior patterns: Worker A ("the person who spends eight consecutive hours at work"), worker B ("those whose work time is segmented, say two hours at a time"), shopper ("those who spend most of their time at maintenance and leisure activities, as opposed to subsistence"), and inactive ("mostly the very young and the very old"). They also found convenient groupings for travel patterns: car or carpool, public transport, non-motorized, and immobile. "With four variables times four variables, we had 16 patterns to simulate," Goulias says. "These are manageable numbers."

"Once we can predict behavior," he continues, "then we can fit that information into a regional model." Equipped with GIS software to create detailed highway networks on which to simulate traffic flow, the model incorporates individual businesses and even residences, and can account for previously overlooked factors like presence of parking spaces, driveway access, and turn lanes.

Goulias and another graduate student, Jin-Hyuk Chung, tested the system on a section of Erie County, Pennsylvania that includes the Peach Street corridor, a main arterial highway. "We'll need more demographic data to really do accurate forecasting," Goulias acknowledges. Even so, when the simulation was measured against actual traffic counts, the preliminary work yielded 95 percent accuracy for some highways. "The model does less well on local roads," Goulias explains, "because we're missing some details like stop signs." To add every last stop sign, he explains, "you'd have to somehow estimate the average delay each one causes, and also the changes in route caused by people trying to avoid them."

Goulias and his students have used the system to help draft a "travel demand management plan" for State College, Pennsylvania and the surrounding region. Through the Pennsylvania Transportation Institute, they give classes and short courses on forecasting to traffic professionals around the state.

"It's important for planners and engineers to know this technology exists," Goulias says. "It can give them more confidence for making informed decisions. It can allow them to look at alternate future scenarios and set policies before developers request permits.

"The best way to resolve traffic congestion is by better land-use planning -- which has to happen before the fact."

Konstadinos G. Goulias, Ph.D., is assistant professor of civil and environmental engineering in the College of Engineering, 212 Sackett Building, University Park, PA 16802; 814-863-7926. He is also a research associate in the Pennsylvania Transportation Institute, 201 Research Office Building, University Park. Jin-Hyuk Chung and Jun Ma are Ph.D. students in the department of civil and enviromental engineering. The research reported was funded by PennDOT, the Mid-Atlantic Universities Transportation Center, and the Federal Highway Administration.