Oct. 4 2006 2:00PM

Oct. 4 2006 3:00PM

At CUSPA (Center for Urban Simulation and Policy Analysis), University of Washington, we have developed a sophisticated open source simulation software, called UrbanSim, that simulates evolution in land use over a (possibly long) period of time. It has been used in several regional agencies throughout the USA, as well as in several European countries. It is implemented as a system of models simulating behavior of different types of agents, such as households and jobs, in one year increments. The system is a part of the Opus project (Open Platform for Urban Simulation) which has been developed as a collaborative platform for researchers, modelers, developers and users involved in land use, transportation and environmental planning in urbanizing regions throughout the world.

Although the nature of the simulation system, including model structure, input parameters and data, is subject to uncertainty, UrbanSim - as many other simulation systems - essentially provides
point predictions. Thus, it does not provide policy makers with information about the confidence of the predictions. As a result, uncertainty may be underestimated, and decisions may be made
that are riskier than the policy-maker thinks they are.

In this talk, I will describe UrbanSim including its model components. I will introduce a method for assessing uncertainty about quantities of interest using urban simulation models, called Bayesian melding. In a case study I will compare results from our method with a method of a simple repetition of simulation runs. It will be shown that the simple repeated runs method gives distributions of quantities of interest that are too narrow, while Bayesian melding gives well calibrated uncertainty statements.

This is joint work with Adrian Raftery (University of Washington, currently at UTIA Department of Adaptive Systems) and Paul Waddell (University of Washington).