Kumar, A., and M. P. Hoerling, 2000: Analysis of a conceptual model of seasonal climate variability and implications for seasonal prediction. Bull. Amer. Met. Soc., 81, 255-264.
A thought experiment on atmospheric interannual variability associated with El Niño is formulated and is used to investigate the seasonal predictability as it relates to the practice of generating ensemble GCM predictions. The purpose of the study is to gain insight on two important issues within seasonal climate forecasting: (i) the dependence of seasonal forecast skill on a GCM's ensemble size, and the benefits to be expected from using increasingly larger ensembles, and (ii) the merits of dynamical GCM techniques relative to empirical statistical ones for making seasonal forecasts, and the scenarios under which the former may be the superior tool.
It is first emphasized that seasonal predictability is an intrinsic property of the observed system, and is inherently limited owing to the nonzero spread of seasonally averaged atmospheric states subjected to identical SST boundary forcing. Further, such boundary forced predictability can be diagnosed from the change in the statistical distribution of the atmospheric states with respect to different SSTs. The GCM prediction problem is thus cast as one of determining this statistical distribution, and its variation with respect to SST forcing.
For a perfect GCM, the skill of the seasonal prediction based on the ensemble mean is shown to be always greater than that based on a single realization, consistent with the results of other studies. However, prediction skill for larger ensembles cannot exceed the observed system's inherent predictability. It is argued that the very necessity for larger ensembles is a testimony for the low predictability of the system.
The advantage of perfect GCM-based seasonal predictions versus ones based on empirical methods is argued to depend on the nonlinearity of the observed atmosphere to SST forcings. If such nonlinearity is high, GCM methods will in principle yield superior seasonal forecast skill. On the other hand, in the absence of nonlinearity, empirical methods trained on the instrumental record may be equally skillful.