An agent-based model of the influence of neighbourhood design on daily trip patterns

Abstract

Post-war suburban neighbourhoods are often criticised for promoting automobile use and leading to problems such as traffic congestion, air pollution, automobile reliance, limited transit access and reduced social interactions. Newer designs, such as the neo-traditional and fused grid designs, aim to ameliorate these problems. But neighbourhood trip and traffic patterns are the collective outcome of individual decisions. Such phenomena often cannot be adequately explained by traditional aggregated methods. With consideration of personal characteristics, preferences and feedbacks between pedestrian and automobile traffic, an agent-based trip and traffic simulation model was developed and calibrated based on data from Ottawa, Ontario. Experiments show that the neo-traditional and fused grid designs generally provide more pedestrian benefits such as fewer crossings, shorter facility-access distance, less emission exposure and more social interaction opportunities, but these benefits also depend on the implementation such as the location of pedestrian-only routes. The influences are often complex. For example, elimination of pedestrian-only routes may lead to more social opportunities, but also much higher emission exposure. The study shows the importance of complex system based study of urban and neighbourhood designs, and the promise of a meso-level approach to urban and transportation simulation that can improve planning outcomes.

Introduction

The relationship between neighbourhood design and daily trip patterns is an active area of research. As the basic units of cities, the design of urban neighbourhoods directly determines the characteristics of local road networks and the availability and location of local facilities. These factors in turn influence the schedule, mode and route choices of local residents. Local traffic patterns, the collective outcome of these choices, in turn have feedback influences on trip behaviour, as people choose their schedule, route and mode to minimise travel time, reduce cost, and avoid congestion. They also attempt to reduce exposure to safety hazards (Schlossberg, Agrawal, Irvin, & Bekkouche, 2007) and automobile emissions (Bhat et al., 2009, Kaur et al., 2006, King et al., 2009), and increase their level of physical activity (Brownson, Baker, Housemann, Brennan, & Bacak, 2001). A positive social environment of social cohesion and trust, a pedestrian-friendly infrastructure with connected pedestrian paths and sidewalks, and a lower volume of traffic all contribute to the choice of walking mode (Cao et al., 2006, MacDonald, 2007), while the presence of “other people walking” influences pedestrian route choice (Schlossberg et al., 2007).

Traditionally, research on the relationship between neighbourhood design and trips focused on the statistical relationship between neighbourhood characteristics and trip characteristics (Boarnet and Crane, 2001a, Boarnet and Crane, 2001b, Cao et al., 2009, Cao et al., 2006, Cervero and Duncan, 2003, Crane, 1995, McNally and Ryan, 1992, Sen and Baht, 2009, Stead, 2001, Stone et al., 1992). Personal characteristics and preferences are often neglected in such studies, the influence of the micro-environment is also rarely considered.

Urban neighbourhoods are complex systems formed as collective outcome of human behaviour. While traditional research methods like aggregate statistical analysis and equation-based methods can be used to analyse certain aspects of such a complex system, they often fail to explain many essential characteristics such as local interactions, feedback processes and emergent phenomena. Agent-based models are suitable for modelling urban neighbourhoods, as feedback processes (such as the interaction between automobile and pedestrian traffic) and bottom-up phenomena (such as the generation of dynamic traffic patterns) can be easily and intuitively simulated. Such models also make possible the evaluation of newly proposed designs such as the fused grid design, for which no real-world aggregate data are available.

Agent-based models have been widely used to simulate movement patterns and trip behaviour in the recent years, but the daily patterns of automobile and pedestrian traffic and the interactions and feedbacks between them at the neighbourhood level are rarely considered. Existing models mostly focus on either automobile traffic patterns at metro- or national-scales (Benenson, Martens, & Birfir, 2008) or pedestrian movements in constrained building environments (Batty, 2003, Castle, 2006, Helbing et al., 2000, Lee and Lam, 2008, Schelhorn et al., 1999). While popular modelling software such as TRANSIMS (Smith, Beckman, Anson, Nagel, & Williams, 1995) and MATSim (Matsim.org, 2012) support multi-modal simulations, the majority of studies based on these software still focus on a single-mode of either automobile or pedestrian (Rieser, 2010). Rieser (2010) introduced a multi-modal agent-based model that includes car, transit and other non-car modes, but the movement of non-car modes are not explicitly simulated (i.e. such agents are teleported). Interactions and feedbacks are important for complex systems, but few studies have explicitly included such processes. The MATSim package features an iterative optimisation process based on automobile traffic feedbacks (Balmer et al., 2008, Nagel and Flötterlöd, 2009), but no pedestrian-automobile interaction is considered. Agent-based studies that consider pedestrian-automobile interactions mostly took a mechanistic view of the problem and simulate how macro-scale movement characteristics such as speed and directions influence the probability of collision (For example, see Banos et al., 2005, Lotzmann et al., 2009). Recently, a few meso-scale models have appeared. Waddell, Wang, Charlton and Olsen, (2010) introduced a sub-metro scale microsimulation model of land use and transportation, but the model assigns trip demands to the network using a static approach, with no consideration of iterative dynamics and pedestrian-automobile feedbacks. A latest study (Aschwanden, Wullschleger, Muller, & Schmitt, 2012) included detailed movement characteristics of pedestrians and automobiles in the simulation of an urban neighbourhood setting, but there is still no direct interaction between pedestrian and automobile traffic; agents’ socio-economic characteristics and trip schedules are not considered either.

As Batty (2003, p. 83) points out, behaviour in human systems is determined not only by personal preferences, intentions and desires, but also “by the environment which reflects the spatial or geometric structure in which the agents function as well as variability between agents, in terms of their intrinsic differences and the uncertainty that they have to deal with in making any response”. A neighbourhood-scale model that takes into account both personal preferences and environmental constraints, and that considers interactions, feedbacks and uncertainties has not been seen in the literature. Such a model will help improve our understanding of how neighbourhood design influences trip and traffic patterns and daily lives.