Downscaling GCM simulations to rainfall and stream flow

Sea-surface temperature variations of the oceans surrounding southern Africa are associated with seasonal rainfall variability, especially during austral summer when tropical influences are strongest. Because of instabilities in the association between austral rainfall over southern Africa and sea-surface temperatures of the equatorial Indian Ocean, the skilful prediction of seasonal rainfall may best be achieved using physically based models as opposed to statistical models that use sea-surface temperature features as predictors of seasonal rainfall. However, skill levels should be tested and compared in an operational environment. A retro-active forecast procedure for the December to February (DJF) season is employed over a 10-year period starting from 1987/88 in order to simulate real-time operational forecasts of several months lead. Forecasts are produced for a number of homogeneous rainfall regions over much of southern Africa, namely South Africa, Botswana, Namibia and Lesotho. The rainfall regions were defined on the basis of the interannual rainfall variability of the subcontinent. Canonical variates are subsequently used to make categorised (below-normal, near-normal and above- normal) DJF rainfall predictions for the region using evolutionary features of near-global sea- surface temperatures. These statistically based rainfall forecasts form the baseline skill level that has to be outscored by more elaborate methods involving General Circulation Models (GCMs) in order to justify the costs involved in running these more sophisticated models. The GCM used here is the COLA T30, which has a vertical resolution of 18 layers and a horizontal resolution of about 400 km over the study area. The boundary of the GCM is a predicted global sea-surface temperature field, and initial conditions are derived from NCEP reanalysis data. The sea-surface temperature fields are predicted over lead-times of up to three months using evolutionary features of near-global sea-surface temperatures. Bias-corrected GCM simulations are hence obtained for an area including most of southern Africa and adjacent oceans. These simulations, which are also conducted over the same 10-year retro-active period, are downscaled to regional level (to obtain categorized forecasts) using the perfect prognosis approach. The perfect prognosis is based on the canonical variates describing the association between observed circulation and moisture at certain standard levels and DJF rainfall, and then substituting the bias-corrected GCM forecasts into the equations. Successful forecasts are obtained when the amplitudes of large events in the global oceans (such as El Niño and La Niña episodes) are described adequately by the predicted sea-surface temperature fields. Given high skill sea-surface temperature forecasts, the scheme has the potential to provide climate forecasts that will outscore the baseline skill level substantially, suggesting that the COLA T30 GCM can be applied successfully for operational rainfall forecasts for the region. GCM simulations using persisted August sea-surface temperature anomalies instead of forecast sea-surface temperatures produced skill levels similar to those of the baseline for longer lead-times, indicating the potential for increased skill with more realistic sea-surface temperature forecasts. The GCM can thus produce forecast skill that can outperform the baseline skill substantially, provided that the forcing is done with highly skilful sea-surface temperature forecasts. The scheme is further adjusted to predict categorised naturalised stream flow into certain dams within the Vaal and upper Tugela river catchments. Retro-active forecasts are made over an 8-year period starting in 1987/88 by downscaling the GCM simulations to stream flow instead of rainfall, hence demonstrating the scheme’s operational utility, albeit over short lead-times.