As a consequence of the severe drought that occurs from 1970 to 2000, forecasting rainfall over the Sahel at seasonal time scales became a priority regarding to impacts in terms of food, health and security managements since the 1970’s. Improving the prediction of the West African Monsoon (WAM) system and its impacts on health, water resources and food security is one of the main aims of the AMMA (African Monsoon Multidisciplinary Analysis) project. In parallel, within the European DEMETER and ENSEMBLES projects framework, a well-validated European coupled multi-model ensemble forecast system for reliable seasonal to interannual prediction has been developed, and maintained at ECMWF. Using the model forecasts performed in ENSEMBLES and DEMETER projects in order to highlight their performance and their reliability over West Africa appeared to be essential, from the first cross AMMA/ENSEMBLES meeting that took place in Bamako, Mali.
Within this framework, we propose to highlight the performance of the different models participating in the ENSEMBLES and DEMETER projects in capturing the mean features of the WAM, as well as of the multi-model hindcasts, during the rainy season (from July to September) over the [1991-2001] common period. Different reference observation data sets are used (CRU, GPCP, NCEP....) to validate the model forecasts. This task is achieved using common validation tools, namely typical indices and maps to determine the model biases (mean/variability) and determinist/probabilistic scores (Correlation/Brier/ROC) to highlight the forecasts performance. A perfect model approach is also used to determine the potential predictability (PP) of the main impact variables (this analysis provides an idea about the upper limit of forecasting probabilities).
Preliminary results highlight common precipitation biases as simulated by coarse resolution GCM used in climate studies, namely overestimation of rainfall over the high mountains (Ethiopia plateau) and underestimation over the low ones (Cameroon mounts, Senegal coast). These biases are relatively similar for both the mean and the variance. Other common (well-known) biases are depicted for relevant SST areas, namely a warm bias over the Gulf of Guinea and the Eastern Pacific region (ENSO). The rainfall potential predictability is relatively high over the Guinea Coast whereas it is relatively weak over the Sahel. Moreover, the model rainfall forecasts seem to be more skilful over the Guinea Coast and the tropical Atlantic Ocean than the Sahel.
Similarities and differences between DEMETER and ENSEMBLES hindcasts systems over sub-Saharan Africa will then be given in the conclusion. A discussion will also be done about the limiting selected common period (1991-2001). Namely, is 10 year model climate enough to have significant statistical results (mean, variance, etc) over the Sahel? Strategies and methods to improve the model forecast performance will then be highlighted as perspectives.Presentation link:
Guideline (few comments slide by slide):