Seasonal Predictions of African Rainfall and Their Verifications

Abstract:

The ability of coupled climate models from the national multi-model ensemble (NMME) dataset to reproduce the interannual variability of precipitation in Sub-Saharan Africa and associated teleconnections is examined. The analysis is for the period 1982-2010. The satellite based Global Precipitation Climatology Project (GPCP) data is used as proxy for observed rainfall and to validate the models. We examine rainfall patterns during the four main seasons in DJF, MAM, JAS, and OND. All models have spatially averaged values of standard deviation lower than that observed. Models are able to reproduce to some extent the main features of the precipitation variability maximum, but with deficiencies in the amplitudes and locations. The areas of highest variability are generally depicted, but there are significant differences among the models. Teleconnections in the models are investigated by first conducting an EOF in the precipitation anomaly fields and then perform a regression of the first or second EOF time series onto the global SST. For the OND and DJF seasons, the majority of the models show an ENSO teleconnection and consistent with the observations. The models’ raw simulations are examined and corrected using canonical correlation analysis (CCA). A new ensemble based on models corrected forecasts is then formed and the results are presented. Retrospective forecasts over a 10-year period are performed and the forecasts verifications presented.