Operational long-lead prediction of South African rainfall using canonical correlation analysis

Abstract

A statistically based technique is used to study the variability and predictability of South African summer rainfall. The country is divided into homogeneous regions on the basis of the interannual rainfall variability. Canonical variates are then used to make 3-month aggregate precipitation forecasts for October–November–December and January–February–March for South Africa from global-scale sea-surface temperatures. Four consecutive 3-month mean periods of sea-surface temperatures are used to incorporate evolutionary features as well as steady-state conditions in the global oceans. Levels and possible origins of forecast skill are investigated for up to 5-month lead-times. Modest skill (correlation >0.5) is found over mainly the central and western interiors of the country, but the skill is poor over the north-eastern regions. The most important contribution of the prediction skill comes from the equatorial Pacific Ocean, with weaker predictability from the equatorial Indian and Atlantic oceans. Sea-surface temperatures in the Atlantic and Indian oceans have important influences on the atmospheric circulation and moisture fluxes over southern Africa, and therefore provide useful predictability, at least for the October–December rainfall. When forecasting South African rainfall, it is insufficient to consider only the El Nino–Southern Oscillation (ENSO) phenomenon, because it does not occur every year and because the sea-surface temperatures of the adjacent oceans modify the ENSO forcing on South African rainfall. Unfortunately, the predictability during years not associated with the ENSO is weak. Copyright © 1999 Royal Meteorological Society