LRF Modelling 200809 Physical Models AGCM ECHAM 4

LRF Modelling 2008/09 Physical Models AGCM (ECHAM 4. 5 – for GPC recognition) Asmerom Beraki Cobus Olivier NEC Verification Models (ECHAM 4. 5+GPC+Statistical) Asmerom Beraki Cobus Olivier Willem Landman Empirical Models Operational Forecasts Cobus Olivier CGCM (Feasibility study ECHAM-MOM: Implement on SA machine) Asmerom Beraki CHPC Application Model (crops) Willem Landman Noelien Somers (ARC) External Model Data (Had. GEM, Glo. Sea 4 and GPCs) Cobus Olivier LORENZ Global SST scenarios Willem Landman PC & LORENZ MM Ensemble for LRF & CC Willem Landman PC & LORENZ Aim high with the goal in mind Application Models (streamflow and malaria) Willem Landman PC & LORENZ SADC-DMC products (rainfall and temperatures) Willem Landman PC & LORENZ

Largest Exp 1 response: Exp 3 Exp 2

NCEP vs AGCM = 0. 4581 NCEP vs CGCM = 0. 3775 • AGCM better able to capture trend • Here, AGCM (in forecast mode) is therefore a better representation of reality • Should thus give better predictions of rainfall over South Africa than CGCM

• Model output statistics (MOS) applied to • AGCM ensemble mean SLP • CGCM ensemble mean SLP • Verification • 5 -year-out cross-validation • Spearman rank correlation AGCM-MOSslp – CGCM-MOSslp Only about 5% of the stations show local significant correlation differences at the 95% level Forecast skill not significantly different irrespective of the use of “correct” or “incorrect” SLP forecasts