A statistical model is used to produce a time series of SST anomaly (SSTA) forecasts, based on a temporal EOF iteration scheme (Zhang et al. 1993). The method is a predictive form of singular spectrum analysis (SSA; Vautard and Ghil 1989), and has conceptual similarity to an analogue method. In this method, EOFs are computed using a correlation matrix derived from a specially constructed data matrix. This matrix contains data from a single spatial point that increase temporally both with row and column index (as in SSA). In the last column of the last row, a first guess of a future (unknown) value of the time series is given--perhaps the climatological mean. When EOFs are computed and the raw data are reconstructed using a truncated set of modes, the unknown value is determined in keeping with the patterns of the EOFs. This process is reiterated using the latest guess of the unknown value until the changes in that value with subsequent iterations become smaller than a prescribed amount. Then the same procedure is carried out for the next future time point, accepting the previously forecast time point as if it were observed. In essence, the method uses past temporal patterns of the variable to forecast future values.

Each of the first 6 principal components (PCS) of the tropical Pacific basin SST are forecast; these are then used to reconstruct the SSTA pattern forecasts. Using SST data spanning from 1970 (obtained from the Climate Prediction Center), 174 forecasts with various beginning and ending times have been made at lead times ranging from 1 to 24 months, respectively. For example, 1 month lead forecasts begin in January 1982 and end in June 1996, 2 month lead forecasts begin in February 1982 and end in July 1996, and so on.

The forecast skill of the model has been evaluated, as shown in Fig. 15-1 of the December 1994 issue of this Bulletin. It was demonstrated that the tropical Pacific SSTA can be forecast fairly well up to 9 months lead, especially in the eastern portion of the basin.

Figure 1 shows 3, 6 and 9 month lead tropical Pacific SSTA forecasts for August and November of 1996, and February 1997. The forecasts show below normal SSTs over the eastern part of the equatorial Pacific. A magnitude of -1EC is indicated in 3 month lead forecasts. These magnitudes decrease after August 1996, and a horseshoe-shaped pattern of positive SSTA in the northern and southern hemisphere subtropical Pacific and western equatorial Pacific surrounds the negative eastern equatorial SSTA.

The area average Nino 3 and Nino 4 SSTAs for all 174 forecasts at 3-, 6- and 9-month lead are shown in Figs. 2 and 3, respectively, along with corresponding observations (solid lines). The forecasts are for negative SSTs in the next 3 months (through August) in Nino 3, and near normal SSTs in Nino 4. Subsequent changes are in the positive direction. This implies that the Nino 3 region will become closer to normal and the Nino 4 region slightly above normal for boreal winter 1996-97.

Vautard, R. and M. Ghil, 1989: Singular spectrum analysis in nonlinear dynamics, with applications to paleoclimatic time series. Physica D Amsterdam, 35, 395­424.

Zhang, B., J. Lie and Z. Sun, 1993: A new multidimensional time series forecasting method based on the EOF iteration scheme. Advances in Atmospheric Sciences, 10, 243-247.

Figures