R EVIEW The use of the multi-model ensemble in probabilistic climate projections BY CLAUDIA TEBALDI 1, * AND RETO KNUTTI 2 1 Institute for the Study of Society and Environment, National Center for Atmospheric Research, PO Box 3000, Boulder, CO 80304, USA 2 Institute for Atmospheric and Climate Science, Swiss Federal Institute of Technology, Universita¨tstrasse 16 (CHN N 12.1), 8092 Zu ¨rich, Switzerland Recent coordinated efforts, in which numerous climate models have been run for a common set of experiments, have produced large datasets of projections of future climate for various scenarios. Those multi-model ensembles sample initial condition, parameter as well as structural uncertainties in the model design, and they have prompted a variety of approaches to quantify uncertainty in future climate in a probabilistic way. This paper outlines the motivation for using multi-model ensembles, reviews the methodologies published so far and compares their results for regional temperature projections. The challenges in interpreting multi-model results, caused by the lack of verification of climate projections, the problem of model dependence, bias and tuning as well as the difficulty in making sense of an ‘ensemble of opportunity’, are discussed in detail. Keywords: regional climate change; probabilistic projections; multi-model ensembles; global climate models; structural uncertainty; performance-based weighting 1. Introduction (a ) Sources of model uncertainty Predictions and projections of weather and climate from time scales of days to centuries usually come from numerical models that resolve or parametrize the relevant processes. Uncertainties in constructing and applying these models are manifold, and are often grouped into initial condition, boundary condition, parameter and structural uncertainties. Initial condition uncertainty is most relevant for the shortest time scales. Weather is chaotic, and predictions are sensitive to the value of observations used to initialize numerical models (e.g. Palmer 2005). Long-term projections of climate change are typically averaged over decades and often across several ensemble members, and are thus largely insensitive to small variations in initial conditions. Even when sequences of daily output from climate models are fed into Phil. Trans. R. Soc. A (2007) 365, 2053–2075 doi:10.1098/rsta.2007.2076 Published online 14 June 2007 One contribution of 13 to a Theme Issue ‘Ensembles and probabilities: a new era in the prediction of climate change’. * Author for correspondence (tebaldi@ucar.edu). 2053 This journal is q 2007 The Royal Society