UrbanSim: Interaction and Participation in Integrated Urban Land Use, Transportation, and Environmental Modeling Alan Borning Dept. of Computer Science & Engineering University of Washington Box 352350 Seattle, Washington 98195 borning@cs.washington.edu Paul Waddell Daniel J. Evans School of Public Affairs University of Washington Box 353055 Seattle, Washington 98195 pwaddell@u.washington.edu 1. PROJECT OVERVIEW AND IMPACTS The process of planning and constructing a new light rail system or freeway, setting an urban growth boundary, changing tax policy, or modifying zoning and land use plans is often politically charged. Our goal in the UrbanSim project is to provide tools for stakeholders to be able to con- sider different scenarios, and then to evaluate these scenarios by modeling the resulting patterns of urban growth and re- development, of transportation usage, and of environmental impacts, over periods of 20–30 years. UrbanSim, combined with transportation models and macroeconomic inputs, per- forms simulations of the interactions among urban develop- ment, transportation, land use, and environmental impacts. It consists of a set of interacting component models that simulate different actors or processes within the urban en- vironment. 2. RECENT RESEARCH ACTIVITIES 2.1 Opus and UrbanSim 4 One project this past year has been collaboratively devel- oping a new software architecture and framework — Opus, the Open Platform for Urban Simulation — and rewriting UrbanSim in that framework. There were several factors that led us to take this step: a growing consensus among researchers in the urban modeling community that a com- mon, open-source platform would greatly facilitate sharing systems, the desire to make the system code more accessi- ble to domain experts, and some intractable problems with some of our previous component models (which were hard to solve due to the inaccessibility of the source code to domain experts, making rapid experimentation and testing hard). After preliminary testing and design work that began in Jan- uary 2005, we began implementing Opus and UrbanSim 4 (the latest version of the system) in March, and now have To appear in the “Project Highlights” section of the Proceedings of the 7th Annual International Conference on Digital Government Research, May 2006. a working version of both [4]. The system is written in Python, and makes heavy use of efficient matrix and array manipulation libraries (principally numarray). The imple- mentation of Opus and UrbanSim 4 contains far less code than the previous implementation, yet implements a much more modular and user-extensible system, and runs faster. It also incorporates key functional extensions such as inte- grated model estimation and visualization. Opus has been designed in collaboration with groups at the University of Toronto, Technical University of Berlin, and ETH, the Swiss Federal Institute of Technology in Zurich. The Toronto group has also been active in implementing a new open-source travel model in Opus; we plan to use that in our own work, both directly and to do baseline comparisons with an experimental activity-based travel model. 2.2 Statistical Analysis of Uncertainty Predicting the future is a risky business. There are nu- merous, complex, and interacting sources of uncertainty in urban simulations of the sort we are developing, including measurement errors, uncertainty regarding exogenous data and other input parameters, and uncertainty arising from the model structure and from the stochastic nature of the simulation. Nevertheless, citizens and governments do have to make decisions, using the best available information. At the same time, we should represent the uncertainty in our conclusions as well as possible, both for truthfulness and as important data to assist in selecting among alternatives. We are starting a new project to provide a principled sta- tistical analysis of uncertainty in UrbanSim, and to portray these results in a clear and useful way to the users of the sys- tem. We are leveraging in this work a promising technique, Bayesian melding, which combines evidence and uncertainty about the inputs and outputs of a computer model to yield distributions of quantities of policy interest. From this can be derived both best estimates and statements of uncer- tainty about these quantities. This past year we have had some initial success in employing this technique, applying it to calibrate the model using various sources of uncertainty with an application in Eugene-Springfield, Oregon. These results are reported in a journal article recently submitted to Transportation Research B: Methodology [3].