Mathematical Sciences: Methods for Developing and Evaluating Computer Models Used in Integrated Assessment

点击翻译按钮获取中文摘要

9634297 Berk In this project, the investigators will develop methodological tools essential for computer models used in integrated assessments. Three kinds of methodological contributions are involved: 1) a flux coupler that links output from one model to input from another, 2) methods to assess uncertainties in coupled computer models, and 3) statistical diagnostics for coupled computer models. The flux coupler transforms model output in one set of units, and for particular spatial and temporal scales, into model input in new units and spatial temporal scales. The tools for assessing uncertainty capitalize on recent advances in resampling procedures to represent sampling distributions for key model parameters and outputs. The diagnostic techniques exploit new ways to link multiple inputs to multiple outputs in a nonlinear fashion. As such, they provide a yardstick with which to judge the performance of coupled computer models. The laboratory for developing and testing the methodological tools is a coupled computer model linking local weather, local air quality, and human outdoor water use for the Los Angeles Basin. The interactions between humans and the larger physical and biological environment are often simulated with large-scale computer models. In this project, the investigators focus on how human water use affects local climate and air quality and then, in turn, is affected by local climate and air quality. Outdoor water use, for instance, has impacts on local temperature and the meteorological conditions that contribute to urban smog. But the computer models used in such simulations need to be carefully evaluated. The key products of this project are new tools for developing and assessing the computer models used to understand human-environmental interactions. The new tools include: 1) a means to allow computer models to more effectively transfer data between them, 2) methods that allow the statistical uncertainties in the computer results to be better described, and 3) techniques for analyzing more thoroughly how well the computer models are using the information provided to them. These tools all speak to the more general issue of improving large-scale computer simulations in a wide variety of applications.