Comment on "Constraining climate forecasts: The role of prior assumptions"

J. D. Annan and J. C. HargreavesFRCGC/JAMSTEC, Yokohama, Japan1. IntroductionIn an interesting and useful paper, Frame et al. [2005] (hereinafter F05) call attentionto the way in which alternative formulations of a supposedly uninformative prior cansignificantly affect the results of Bayesian estimation when observational constraints areweak. In particular, using a uniform prior in climate sensitivity S gives a substantiallydifferent result to that obtained using a uniform prior in feedback λ ∝ 1/S. The math-ematical aspects of their presentation are clear and convincing. However, we show herethat their proposed solution to this problem is inconsistent and inadequate, and it seemsimplausible that any such simplistic prescription can be considered appropriate for dealingwith what is in fact a rather fundamental issue in Bayesian estimation. Moreover, theirexploration of the problem perpetuates and highlights a widespread and fundamental con-J. D. Annan and J. C. Hargreaves, FRCGC/JAMSTEC, Yokohama, Japan (jdan-nan@jamstec.go.jp)D R A F T June 12, 2006, 4:09pm D R A F T