Prediction of the January-February-March Mean Maximum and Minimum Surface Temperatures for South Africa

contributed by E. Klopper

Research Group for Seasonal Climate Studies, South African Weather Bureau, South Africa

A canonical correlation analysis (CCA) forecast model (Barnett and Preisendorfer, 1987; Barnston et al., 1996) is used to predict seasonal surface-air temperature for a network of 55 stations over South Africa. The model is similar to the rainfall model developed for Southern Africa (Landman and Mason,1999). The predictor field in the temperature model consists of global scale sea-surface temperatures (SST) (Smith et al., 1996) for four consecutive, non-overlapping 3-month periods. By incorporating global scale SST at different lags evolutionary features such as a developing La Niña which takes place over a couple of months are included. The predictand field consists of 3-month mean maximum or minimum surface-air temperature at 55 locations in South Africa.

Prior to conducting CCA, the standardized predictor and predictand data are separately condensed using extended empirical orthogonal function (EEOF) analysis (Weare and Nasstrom, 1982) and empirical orthogonal function (EOF) analysis (Jackson, 1991) respectively to filter out noise and retain the essential variables. The reduced number of predictor and predictand components are then cross-correlated and used as input to the CCA procedure. Cross-validation (Ward and Folland, 1991; Barnston and Ropelewski, 1992) is performed to determine the model skill over a 39 year period. The predicted values are ranked and divided into three equi-probable categories (or terciles) from which a categorical above-normal, near-normal or below-normal prediction can be made.

In the forecast presented here, stacked sea-surface temperatures of four consecutive 3-month seasons (namely D₉₈JF₉₉, MAM₉₉, JJA₉₉ and SON₉₉) are used as predictors for JFM₂₀₀₀ mean maximum (Figure 1) and minimum (Figure 2) temperatures at 55 stations in South Africa. A categorical forecast for each station is made and those with significant skill are used to compile a seasonal forecast. On the maps A, B and N represent the predicted category, while the number shows the skill at each location. Only forecasts with correlation skill better than chance (> 33.3%) are shown.

The forecast for JFM₂₀₀₀ maximum temperature (Figure 1) are for below-normal temperature over the greater part of the country. Above-normal temperatures are confined to the west coast and adjacent interior. Seventy three percent of the stations showed significant skill. The forecast for JFM₂₀₀₀ minimum temperature is also for above-normal temperatures in the west. The forecast signal over the rest of the country is relatively poor (with only 24 out of 55 possible stations having significant skill) and mixed.

References

Barnett, T.P. and Preisendorfer, R., 1987: Origins and levels of monthly and seasonal forecast skill for United States surface air temperatures determined by canonical Correlation Analysis. Monthly Weather Review, 115, 1825-1850

Barnston, A.G. and Ropelewski, C.F., 1992: Prediction of ENSO episodes using canonical correlation analysis. J. Climate, 5, 1316-1345

Barnston, A.G., Thiao, W. and Kumar, V., 1996: Long-lead forecasts of seasonal precipitation in Africa using CCA. J. Climate, 9, 2660-2697

Jackson, J.E., 1991: A User's Guide to Principal Components. Wiley, New York, 569pp

Landman, W.A. and Mason, S.J. 1999: Operational long-lead prediction of South African rainfall, using canonical correlation analysis. Int. J. of Climatol., 19, 1073-1090

Smith, T.M., Reynolds, R.W., Livezey, R.E. and Stokes, D.C., 1996: Reconstruction of historical sea-surface temperature usinf empirical orthogonal function analysis. J. Climate, 9, 1403-1420

Ward, M.N. and C.K. Folland, 1991: Prediction of seasonal rainfall in the north Nordeate of Brazil using eigenvectors of sea-surface temperature. Int. J. Climatol., 11, 711-743

Weare, B.C. and Nasstrom, J.S. 1982: Examples of extended empirical orthogonal function analysis. Mon. Wea. Rev., 110, 481-485

Figure Captions

Figure 1: Predicted mean maximum temperature categories for January-February- March 2000 using December 1998 to November 1999 observed sea-surface temperatures as predictors. Forecast categories for 40 of the 55 possible stations are significant.

Figure 2: As in Figure 1, but for mean minimum temperature. Significant forecasts for 24 stations are depicted.