Modeling the uncertain impacts of climate change

Human and earth systems are extremely complex processes. The modeling of these systems to assess the effects of climate change is an activity fraught with uncertainty. System models typically involve the linking of a series of computer codes, each of which is a detailed model of some physical or social process in its own right. In such system models, the output from one process model is the input to another. Traditional methods for dealing with uncertainty are inadequate because of the sheer complexity of the modeling effort: Monte Carlo methods and the exhaustive evaluation of what if '' scenarios estimate sensitivities fail because of the heavy computational burden. More efficient methods are required for learning about system models that are constructed from a collection of computer codes. A two-tiered modeling approach is being developed to estimate the distribution of outcomes from a series of nested models. The basic strategy is to develop a simplified executive, or simplified system code (SSC), that is analogous to the more complex underlying code. An essential feature of the SSC is that it uses information abstracted from the detailed underlying process codes in a manner that preserves their essential features and interactions among them. Of course, to be useful, the SSC must be much faster to run than its complex counterpart. The success of the SSC modeling strategy depends on the methods used to extract essential features of the complex underlying codes.