We investigate characteristic properties of the congested traffic states on a 30 km long stretch of the German freeway A5 north of Frankfurt/Main. Among the approximately 245 breakdowns of traffic flow in 165 days, we have identified five different kinds of spatio-temporal congestion patterns and their combinations. Based on an "adaptive smoothing method" for the visualization of detector data, we also discuss particular features of breakdowns such as the "boomerang effect" which is a sign of linearly unstable traffic flow. Controversial issues such as "synchronized flow" or stop-and-go waves are addressed as well. Finally, our empirical results are compared with different theoretical concepts and interpretations of congestion patterns, in particular first-and second-order macroscopic traffic models.
Summary of Previous Models and Empirical ResultsUnderstanding traffic dynamics can not only help to identify reasons for bottlenecks. It also contributes to the development of modern driver and traffic assistance systems aiming at the improvement of safety, comfort, and capacity. Progress has been made by empirical studies and theoretical modelling approaches. Apart from traffic scientists, mathematicians and physicists have also recently contributed to these fields. Because 1 of the numerous publications, our introductory overview can only be selective, so that we refer the reader to some comprehensive reviews (e.g., Gerlough and Huber, 1975;Vumbaco, 1981;Leutzbach, 1988;May, 1990;Brilon et al., 1993; Transportation Research Board, 1996; Gartner et al., 1997; Helbing, 1997a; Daganzo, 1997a;Bovy, 1998;Hall, 1999;Brilon et al., 1999; Chowdhury et al., 2000b, with a focus on cellular automata; Helbing, 2001a, containing 800 references; Nagatani, 2002).
Modeling ApproachesThe main modeling approaches can be classified as follows: Car-following models focus on the non-linear interaction and dynamics of single vehicles. They specify their acceleration mostly as a function of the distance to the vehicle ahead, the own and relative velocity (e.g., Reuschel, 1950a, b;Gazis et al., 1959Gazis et al., , 1961May and Keller, 1967;Gipps, 1981;Gibson, 1981;Bando et al., 1994Bando et al., , 1995a Krauß, 1998;Treiber et al., 2000;Brackstone and McDonald, 2000). Submicroscopic models take into account even details such as perception thresholds, changing of gears, acceleration characteristics of specific car types, reactions to brake lights and winkers (Wiedemann, 1974;Fellendorf, 1996;Ludmann et al., 1997). In favour of numerical efficiency, cellular automata describe the dynamics of vehicles in a coarse-grained way by discretizing space and time (Cremer and Ludwig, 1986;Biham et al., 1992;Nagel and Schreckenberg, 1992; Chowdhury et al., 2000b). Gas-kinetic models agglomerate over many vehicles and formulate a partial differential equation for the spatio-temporal evolution of the vehicle density and the velocity distribution. While Boltzmann-like approaches (Prigogine and Andrews, 1960;Prigogine and Herman, 1971;Paveri-Fontana, 197...