Dynamical downscaling of GloSea5 over Ethiopia

Abstract

We have implemented dynamical downscaling of the Met Office GloSea5 global seasonal forecasting system and analysed its ability to generate skilful forecasts of characteristics of the June-September rainy season in Ethiopia that are of societal relevance. The downscaling used a regional model with a resolution of 25 km, and the same atmosphere and land configuration as the global model, to produce a 3-member ensemble of seasonal hindcasts for the period 1991–2011 and a larger 15-member ensemble for four of these years comprising two anomalously dry and two anomalously wet years. The regional model was also driven by the quasi-observed ERA-Interim dataset. To provide context for the assessment of the downscaled seasonal forecasts and to show the limit for the skill of a global seasonal forecast downscaling system for the region.A mainly qualitative assessment of GloSea5 and downscaled GloSea5 forecasts demonstrated that the downscaled forecasts could be considered a faithful disaggregation of the coarse resolution GloSea5 forecasts. Forecasts of average seasonal rainfall anomalies in the three regions of Ethiopia studied were captured in three of the four years with the wet season of 2006 incorrectly forecast in all three regions, and the 21 year 3-member hindcast had a correlation of 0.65 with observations. Whilst exploring the potential for the downscaled GloSea5 to generate skillful forecasts of rainy season onset and dry spells we note that both the global and regional model have skill with onset correctly predicted as being early or late in more than 75% of cases for some regions.