PATTERN-ANALYSIS OF SST-FORCED VARIABILITY IN ENSEMBLE GCM SIMULATIONS - EXAMPLES OVER EUROPE AND THE TROPICAL PACIFIC

An ensemble of atmospheric general circulation model (GCM) simulations with prescribed sea surface temperature (SST) generates a rich datase t. The main aim here is to advocate and demonstrate an approach to ski ll and reproducibility based on spatial anomaly patterns, Benefits and applications of this type of analysis include the efficient extractio n of the model's forced variability, guidance on systematic errors in the model's response to SST forcing, clues to physical mechanisms, and a basis for model output statistics for seasonal forecasting. Some of the possible statistical techniques are illustrated, though the aim i s not to provide an exhaustive comparison of the different spatial ana lysis techniques available. The examples are taken from an ensemble of three GCM integrations forced with observed SST through 1979-88. Bore al summer examples are given for the tropical Pacific and Europe, prov iding a contrast of a high and a low skill situation, respectively. Fo r model verification, a coupled pattern singular value decomposition a nalysis is performed between model and observed fields over the analys is domain. Over Europe, a model rainfall pattern is identified that sp ecifies the contrasting rainfall anomalies associated with blocked and westerly summers observed through the period 1979-88, though statisti cal significance for the result cannot be proven using this small samp le size. In the central and western tropical Pacific (CWTP), the leadi ng model (rainfall) and observed (outgoing longwave radiation) modes h ave near-perfect temporal agreement, but the model's spatial pattern l acks weight near Indonesia, which may be useful information for model developers, Significant reproducibility of climate anomalies among ens emble members indicates potential seasonal forecast skill, because the similar atmospheric anomalies must derive from a common response to t he anomalous SST forcing. To study reproducibility, the cross-covarian ce among all ensemble members is used to define a model base pattern ( referred to as the forced pattern) that maximizes temporal covariance among ensemble members. The close relationship with the principal comp onents of the ensemble mean anomaly is demonstrated. Monte Carlo tests show that the covariances among ensemble members associated with the CWTP and European forced patterns are highly statistically significant . It is suggested that this approach is an efficient way to identify s tatistically significant reproducibility.