Probabilistic Rainy Season Onset Prediction over the Greater Horn of Africa based on Long-Range Multi-Model Ensemble Forecasts

Most of the socioeconomic activities in the Greater Horn of Africa (GHA) region are rain dependent, and economic sectors such as agriculture, hydroelectric power generation, and health would greatly benefit from reliable information about onset, cessation, intensity, and frequency of rainfall. 
In a seasonal climate forecast at lead times on the order of weeks or months, uncertainty about these variables is significant, making a case for probabilistic forecasting where uncertainties are communicated along with the forecast.

We present results of an evaluation of the skill of probabilistic rainy season onset forecasts over GHA, which were derived from bias-corrected, long-range, multi-model ensemble precipitation forecasts. A careful analysis of the contribution of the different forecast systems to the overall multi-model skill shows that the improvement over the best performing individual model can largely be explained by the increased ensemble size. An alternative way of increasing ensemble size by blending a single model ensemble with climatology is explored and demonstrated to yield better probabilistic forecasts than the multi-model ensemble. Both reliability and skill of the probabilistic forecasts are better for OND onset than for MAM and JJAS; for the two latter, forecasts are found to be late biased and have only minimal skill relative to climatology. While the overall level of skill is limited in our setup where predictions are made at a horizontal resolution of 0.25 degrees, we find that especially OND forecast skill increases substantially under a metric that evaluates the forecasts at coarser spatial scales.