[Previous Article] [Next Article]
The Research Group for Statistical Climate Studies (RGSCS) of the South African Weather Bureau issues long
range forecasts for South African rainfall for 3-month periods and temperature for 1-month periods using canonical
correlation analysis (CCA). This scheme is based on the work of Barnett and Preisendorfer (1987), and is similar to the
forecast systems later developed and implemented by Barnston and Ropelewski (1992) for ENSO prediction, and
Barnston (1994) for U.S. surface climate prediction. It includes a cross-validation design to obtain estimates of forecast
skill that are largely uninflated. Cross-validation is performed holding out 1 or 3 years at a time as the forecast targets,
using a 48 year research period (1946-93) for precipitation and a 36-year period (1960-95) for temperature.
The predictor for the CCA forecasts is the near-global SST field, spanning four consecutive 3-month periods. For
the forecasts presented here, the four predictor periods are DJF 1995-96, MAM, JJA and SON 1996. By providing
multiple predictor periods, evolutionary features such as a warming Indian Ocean or a cooling Pacific Ocean can help
determine the forecast. Roughly 100 rainfall districts are used as the predictands for the precipitation forecasts to be
shown below. This contrasts with the six region system used for forecasts for the same total area discussed in Landman
(1994). Predictions are provided only for districts having demonstrable skill for the predicted period in question;
otherwise, the climatological distribution is regarded as the best forecast.
The skill of the CCA model, determined using a 48-year research (or training) period, is expressed as a probability
associated with a categorical forecast given in real-time. A categorical system with 3 climatolog-ically equiprobable
classes (below, near, and above normal) is used, where the probabilities of the three classes would be 33.3%, 33.3%
and 33.3%, respectively, in the absence of any skill. The departure from that baseline probability is determined by the
categorical hit rate in the cross-validation trials in the research period. For example, if the hit rate is 50% (well above
the chance rate of 33.3%), then the probability associated with a real-time forecast will be indicated as 50.0%. No skill
discrimination is done as a function of which category is forecast; thus, in the above example a 50.0% probability would
be assigned to a real-time forecast regardless of which of the three categories is being forecast. The forecast anomalies
are not inflated to equalize their variance with observations.
Figure 1 shows the precipitation forecast for South Africa for JFM 1997. The forecast category (B=below, N=near,
A=above normal) is shown for each predictand region, accompanied by the associated probability. The probability is
a function of the expected skill, being the categorical hit rate for the independent forecast tests from 1985 onward. The
farther above the chance hit rate of 33.3%, the higher the forecast skill. Regions whose forecast skill has been found
to be below a minimally usable threshold are labeled "C". An asterisk is shown for regions having statistically significant
skill.
The forecasts call for favorable rainfall conditions over parts of the north central and southeastern portions of
South Africa this Jan-Feb-Mar 1997. In particular, the Eastern Northern Cape and Eastern Cape is forecast to receive
near to above normal rainfall. Below normal rainfall is predicted in the Lowveld and Eastern Northern Province, with
low to moderate expected skill. In much of the remainder of South Africa near-normal precipitation is forecast with
low to moderate skill. The summary outlook for the remainder of this rainfall season is for generally normal to favorable
conditions with the exception of the Lowveld and far northern regions that are predicted to be dry for Jan-Feb-Mar.
Further discussion of the empirical relationships between SST predictors and the rainfall in the target regions is
found in Landman (1994, 1995). Greater detail about the forecasts presented here is available in the RGSCS Bulletins
issued by the South African Weather Bureau (RGSCS, Room 5087, South African Weather Bureau, Private Bag X097,
Pretoria 0010).
Barnett, T.P. and R. Preisendorfer, 1987: Origins and levels of monthly and seasonal forecast skill for United
States surface air temperatures determined by canonical correlation analysis. Mon. Wea. Rev., 115, 1825-1850.
Barnston, A.G., 1994: Linear Statistical Short-term Climate Predictive Skill in the Northern Hemis-phere. J.
Climate, 7, 1514-1564.
Barnston, A.G. and C.F. Ropelewski, 1992: Prediction of ENSO episodes using canonical correlation analysis.
J. Climate, 5, 1316-1345.
Landman, W.A., 1994: A study of the rainfall variability of the summer rainfall regions of South Africa, as
revealed by principal component analysis. IRICP Pilot Project Final Report. Available at RGSCS, Room 5087, South
African Weather Bureau, Private Bag X097, Pretoria 0010, South Africa.
Landman, W.A., 1995: Predicting South African seasonal rainfall by means of canonical correlation analysis.
Preprints, 6th International Meeting on Statistical Climatology, June 19-23, 1996, Galway, Ireland, 479-481.
Fig. 1. Forecast for Jan-Feb-Mar 1997 on the basis of SST from Dec-Jan-Feb 1995-96 to Sep-Oct-Nov 1996,
produced by the South African Weather Bureau using CCA. Forecasts are expressed categorically (B, N, A for below,
near, or above normal) for each region, followed by the skill-based probability of occurrence: 33.3 would imply no skill,
50.0 would imply a considerable probability anomaly with respect to the no-skill level, etc. Regions with little or no
expected skill are assigned "C". Regions whose skill is statistically significant at the 0.05 level are indicated with an
asterisk. Regions with "A" forecasts are stippled; those with "B" forecasts are indicated with a heavy minus sign.