20th Conference on Climate Variability and Change

13A.1

Scalar trend prediction in climate change

Stephen S. Leroy, Harvard Univ., Cambridge, MA; and J. A. Dykema and J. G. Anderson

The development of the field of climate signal detection and attribution arose from the need to ascribe confidence levels to global warming and the degree to which humans are responsible. At this point, the community consensus is that it is 95% certain that humans are responsible for the observed warming of the surface air over the past three decades. Climate models have been relied upon to produce fingerprints of climate change (if not the overall rate of change), but it is now incumbent upon us to develop a robust climate change predictive capability that is both accurate and precise. We will discuss how the work of Huntingford et al. (Geophys. Res. Lett., doi:10.1029/2005GL024831, 2006) on the incorporation of signal shape uncertainty into optimal fingerprinting, when put into the context of Bayesian inference and generalized to arbitrary data sets, leads directly to a method of prediction of scalar trends in climate conditioned on observed climate trends and the physics of climate change. This technique is applicable to any scalar trend in the climate system and any data set.

In ordinary optimal fingerprinting, patterns of climate change in response to greenhouse gas radiative forcing or some other forcing are treated as fingerprints that can be used to uniquely attribute change to particular types of forcing and distinguish each, at least in part, from natural variability of the climate system. One significant uncertainty in this technique is the precise form of each fingerprint, which depends to some degree on the climate model used to produce it. Huntingford et al. address this uncertainty by adding a fingerprint uncertainty covariance to the natural variability covariance, as was recommended by previous authors. We consider that data sets can be made completely general, that the normalization of signals is in fact arbitrary, and that the use of multiple models in a Bayesian framework implies marginal probabilities, and we find that the technique of Huntingford et al. is that of predicting scalar trends in the climate system with great precision. As an example, prediction of surface air warming of North America can be constrained using climate data by not only measuring historical trends in surface air temperature over North America but also by associating it with measured trends in the upper air that are correlated with North American climate change trends. The uncertainty of the correlation is fully accounted for. The result is that the precision of a prediction of the North American surface air temperature trend is reduced by a factor of two over that obtained from historical North American surface air temperature trend alone.

wrf recording  Recorded presentation

Session 13A, Detection and attribution of climate change: Part II
Thursday, 24 January 2008, 8:30 AM-9:45 AM, 215-216

Next paper

Browse or search entire meeting

AMS Home Page