SIMULATING PANDEMIC INFLUENZA RISKS OF US CITIES

摘要

We describe the spatial Agent-Based Computational Laboratory that we have developed to study the pandemic influenza risks of US cities. This research presented a series of interesting challenges, from the uncertainty surrounding the future epidemiological characteristics of a human-transmission H5N1 strain of pandemic influenza, to the need to provide timely decision-support despite modeling a country with a population of 300 million individuals. In order to provide time-sensitive policy analyses, we implemented a light-and-fast agent-based model of the spatial and temporal spread of pandemic influenza, which uses a novel compression technique to analyze large numbers of agents. We assessed the impact of parameter uncertainty and of stochastic behavior via very large numbers of simulations. To facilitate this, we developed a parallel job controller that tests combinations of disease scenarios, and a platform-independent job-submission application that harnesses the computational resources of high-performance computing environments ranging from local clusters up through TeraGrid supercomputers.