Predicting the seasonal evolution of southern African summer precipitation in the DePreSys3 prediction system

We assess the ability of the DePreSys3 prediction system to predict austral summer precipitation (DJF) over southern Africa, defined as the African continent south of 15°S. DePresys3 is a high resolution prediction system (at a horizontal resolution of ~60km in the atmosphere in mid-latitudes and of the quarter degree in the Ocean) and spans the long period 1959-2016. We find skill in predicting interannual precipitation variability, relative to a long-term trend; the anomaly correlation skill score over southern Africa is greater than 0.45 for the first summer (i.e. lead month 2-4), and 0.37 over Mozambique, Zimbabwe and Zambia for the second summer (i.e. lead month 14-16). The skill is related to the successful prediction of the El-Nino Southern Oscillation (ENSO), and the successful simulation of ENSO teleconnections to southern Africa. However, overall skill is sensitive to the inclusion of strong La-Nina events and also appears to change with forecast epoch. For example, the skill in predicting precipitation over Mozambique is significantly larger for the first summer in the 1990-2016 period, compared to the 1959-1985 period. The difference in skill in predicting interannual precipitation variability over southern Africa in different epochs is consistent with a change in the strength of the observed teleconnections of ENSO. After 1990, and consistent with the increased skill, the observed impact of ENSO appears to strengthen over west Mozambique, in association with changes in ENSO related atmospheric convergence anomalies. However, these apparent changes in teleconnections are not captured by the ensemble-mean predictions using DePreSys3. The changes in the ENSO teleconnection are consistent with a warming over the Indian Ocean and modulation of ENSO properties between the different epochs, but may also be associated with unpredictable atmospheric variability.