EVALUATINGGCM OUTPUT WITH IMPACT MODELS LARRY J. WILLIAMS 1 , DAIGEE SHAW 2 and ROBERT MENDELSOHN 3 1 Electric Power Research Institute, 3412 Hillview Avenue, Palo Alto, CA 94303, U.S.A. 2 Academia Sinica, Taipei, Taiwan ROC 3 Yale University, New Haven, CT 06511, U.S.A. Abstract. This study uses empirical agricultural impact models to compare the U.S. climate change predictions of 16 General Circulation Models (GCMs). The impact analysis provides a policy- relevant index by which to judge complex climate predictions. National aggregate impacts vary widely across the 16 GCMs because of varying regional and seasonal patterns of predicted climate change. Examining the predicted impacts from the full set of GCMs reveals that the seasonal detail in the GCM predictions is so noisy that it is not significantly different from a constant annual change. However, a consistent regional pattern does emerge across the set of models. Nonetheless, aggregating climate change across seasons and regions within the United States, using a national-annual climate change provides a reasonable and efficient approximation to the expected impact predicted by the 16 GCM models. 1. Introduction In order to predict the impact of increasing greenhouse gases in the atmosphere, the causal web from greenhouse gas emissions, through carbon cycles, atmospheric chemistry, radiative forcing, climate prediction, biological response, and social impacts must be examined and understood. Although each step in this process has scientific merit in its own right, the value from a policy perspective depends upon how well the research increases our understanding of the entire web. From a policy perspective, the research must help us understand what will happen to impacts as a result of changing greenhouse gas emissions. Because policy decisions on greenhouse gases must be made now in the face of uncertainty, policy makers need to know what the expected impacts will be and what is the uncertainty surrounding these estimates. Throughout this paper we use the term ‘expected’ in the probabilistic sense. It is defined as the probability of an outcome times its value summed across all outcomes. Most of our results do not depend upon any particular probability distribution function, although we find that the distribution of impacts is normally distributed. The impact in question is the change in total U.S. aggregate farm value due to a change in climate from current climate to a GCM predicted 2 CO 2 climate. The distribution results from 16 GCM predictions of 2 CO 2 climates. In this paper, we examine what climate science can tell us about the agricultural impacts of climate change. We do this by using an agricultural impact model to Correspondence may be sent to ljwillia@epri.com (650) 855-2695 voice; (650) 855-2950 fax. Climatic Change 39: 111–133, 1998. c 1998 Kluwer Academic Publishers. Printed in the Netherlands.