 |
Center for Nonlinear Studies |
Multi-model ensembles remain the primary tool used by the Intergovernmental Panel on Climate Change to assess uncertainty in model projections of future climate change. In this talk, we examine some of the intrinsic problems which we face when attempting to combine multiple model simulations into an integrated and objective statement about future climate and how a number of new methodologies might address some of these issues. Starting with the simplest approach of simply averaging a large number of models, we examine why the multi-model mean performs so well in replicating present-day climate but why it cannot itself be considered to be a plausible climate state, especially for fields with small-scale variability such as precipitation. We combine observationally-derived patterns of precipitation variability together with model-derived projections to produce an 'optimal' mean precipitation change estimate, which can reproduce more faithfully the distribution of precipitation change seen in the individual models. Finally, we address the issue of model interdependency: in an ensemble where models share components and parameterizations in an uncoordinated fashion, "model democracy" samples the components in a demonstrably biased fashion. We use present day simulations to define an observable space, in which the multi-model ensemble can be resampled to overcome some of these biases.
Host: Nathan Urban, nurban@lanl.gov, 665-7543