Multiseasonal Hindcasts for 1972–92

Abstract

Hindcasts for the period 1972?92 have been made using the two-tiered method. This involved forcing a global atmospheric model with sea surface temperature anomalies for the low-latitude Pacific Ocean generated with the Zebiak and Cane intermediate coupled model. Outside of this Pacific Ocean domain sea surface temperatures were specified from climatology. All hindcasts were for a duration of 12 months, with each hindcast commencing on 1 January of the individual years. An ensemble of 10 hindcasts was made for each year using different initial atmospheric conditions obtained from a long control run. In addition, a four-member ensemble simulation was made with the same global atmospheric model forced with the annually varying Global Sea Ice and Sea Surface Temperature (GISST) distribution for the period 1971?91. This provided a criterion against which the limited sea surface temperature forcing for the hindcasts could be compared. In the analysis of the results no attempt was made to remove systematic errors or to minimize other possible model deficiencies. The hindcasts reproduced the observed temporal variability of the Southern Oscillation index, with transitions between ENSO events particularly well defined. The interannual variability of the anomalous zonal wind stress over the equatorial Pacific Ocean was hindcast satisfactorily, but an index of the Pacific?North America oscillation was poorly represented in the hindcasts. This deficiency may have implications for predictability in regions influenced by this oscillation. Rainfall hindcasts are presented in some detail, particularly time series for individual model grid boxes or averages over regions. The results are presented as monthly (rather than seasonal) totals for individual years in order to provide some indication of their potential temporal limits. The highest accuracy for rainfall was achieved over the low-latitude Pacific Ocean where anomaly correlation coefficients with observations greater than 0.6 were obtained over an extended region. The quality of hindcasts over ENSO-influenced land regions varied noticeably with location, but the marked interannual variation in rainfall associated with ENSO events was quite well captured for northeast Australia. The influence of chaos on the hindcasts is illustrated by providing outputs for individual members of the ensemble generated. In general, limited systematic improvements could be identified for the GISST simulation compared to the hindcasts. This outcome is attributable to deficiencies in the model used.