The concept of "potential predictability" arose out of questions of predictability on seasonal timescales. Predictability is a feature of the system under examination and represents a measure of the "rate of separation" of initially "close" states of the system or, if the initial separation is considered to be error, of the growth of error in a perfect forecast model. Studies of predictability which follow this approach are prognostic and attempt to gain information on the growth of error in the real or model system. For seasonal prediction, the error in the seasonal mean surface air temperature, for instance, would be of interest.
Potential predictability is a diagnostic, rather than prognostic, concept. Here one asks if the "observations" of the real or a model system exhibit behaviour which suggests that the seasonal means (or some other quantity) behave in a way that indicate that they have a predictable component. In particular, the assumption is that the variation of the mean in question arises from contributions from unpredictable "weather noise" and, possibly, also from a low-frequency contribution which is at least potentially predictable. The study of the potential predictability of a system is the identification of such a component at an acceptable level of statistical confidence.
The proposed subproject will analyze the potential predictability in the surface air temperature of the coupled models in CMIP at various timescales including that of the annual, pentadal and decadal means. An attempt will be made to compare the location and nature of the potential predictability in the various models, to connect it to various coupled processes and oscillations in so far as that is possible, and to infer the effect of spin-up methodology, flux correction, and resolution on the results.
Potential predictability analysis has the potential to reveal interesting similarities and differences in model behaviour which are of direct interest to CLIVAR and to modellers in general since it tries to separate out information on long-term coupled processes from the passive response to weather noise.
The study which will identify regions of potential predict and possible connection to mechanisms as in Boer (1997) following the methods in that paper and the potential predictability methods discussed in Zwiers (1996) and von Storch and Zwiers (1998). Some "validation" may be possible using the GISST data set.
Boer, 1997: A study of atmosphere-ocean predictability on long timescales. To be submitted to Climate Dyanmics, also presented at the BMRC Syposium on Climate Prediction and Predictability.
Zwiers, 1996: Interannual variability and predictability in an ensemble of AMIP climate simulations conducted with the CCC GCM2. Climate Dynamics, 12, 825-847.
von Storch and Zwiers, 1998: Statistical analysis in climate research. To be published by Cambridge University Press.