This paper is concerned with estimating the predictable variation of extratropical daily weather statistics ("stormtracks") associated with sea surface temperature (SST) changes on interannual to interdecadal scales, and its magnitude relative to the unpredictable noise. The SST-forced stormtrack signal in each winter in 1950-99 is defined as the ensemble-mean stormtrack anomaly obtained in an ensemble of atmospheric general circulation model (GCM) integrations with prescribed observed anomalous SSTs, and the noise as the standard deviation of the ensemble members from the ensemble-mean. Two sets of 12-member ensembles available from two modeling centers (NCAR and NCEP), with anomalous SSTs prescribed either globally or in the tropics alone, are used. Since the stormtrack signals cannot be derived directly from the archived GCM output, they are diagnosed from the 200 mb height signals by an empirical linear "storm track model" (STM). For two particular winters (the El Nino of JFM 1987 and the La Nina of JFM 1989), the stormtrack signals and noise are estimated directly, and more accurately, by generating additional large (60-member) ensembles with the NCEP GCM. The linear STM is shown to be remarkably successful at capturing the GCM's stormtrack signal in these two winters, and is therefore argued to be suitable for estimating the signal in the other winters.

The principal result from this analysis is that there is a significant SST-forced stormtrack signal in each winter, but one that varies substantially from case to case. The correlation of the SST-forced and observed stormtrack anomalies is high enough in the Pacific-North American (PNA) sector to be useful. Most, but not all, of this predictability is associated with tropical Pacific SST forcing. The central Pacific (Nino-4) is somewhat more important than the eastern Pacific (Nino-3) in this regard. Variations of the pattern anomaly correlation from winter to winter, and among 5-winter averages, are generally consistent with variations of the signal strength, and to that extent are identifiable a priori. The greater predictability of the Pacific storm track in the second half of the 50-year record, as well as the 50-year stormtrack trend, are thus consistent with the stronger ENSO SST forcing in the second half. None of these conclusions apply in the Euro-Atlantic sector, where the correlations of the SST-forced and observed stormtrack anomalies are much lower. Given also that they are inconsistent with the estimated signal to noise ratios, substantial GCM error in representing the regional circulation response to tropical SST forcing, rather than intrinsically low Euro-Atlantic stormtrack predictability, is argued to be behind these low correlations.