The East African March–May Rainy Season: Associated Atmospheric Dynamics and Predictability over the 1968–97 Period

Abstract

This paper focuses on the East African March?May ?long rains.? Particularly, it investigates the atmospheric patterns associated to the March?May rainfall anomalies, then proposes a seasonal prediction model. In a preliminary step, in order to define a regional rainfall index, a new form of extended principal component analysis is performed, aimed at capturing both the spatial and intraseasonal rainfall coherence. What emerges is that although the long rains exhibit a low temporal coherence, calling for a separation between the months of March?April and May in teleconnection studies, they show a relatively strong spatial consistency over the Kenya?Uganda inland region. From composite analyses performed using NCEP?NCAR reanalyzed atmospheric data, three major signals appear for that region. Two are during March?April involving ENSO and the latitudinal location of the ITCZ, and ENSO interactions with the northern extratropical dynamics (by way of cool surges toward the Tropics and upper-ridge?trough systems). The third signal is the Asian monsoon in May. As shown using a rainfall index for Ethiopia, the interactions with the midlatitudes get stronger when considering rainfall farther to the north. The predictability study identifies four February indexes, involving several scales and several atmospheric and oceanic parameters, to serve as predictors in linear multiple regression (LMR) and linear discriminant analysis (LDA) models. The predictors, selected by a stepwise procedure, depict both regional (energy gradient and zonal wind) and remote dynamics (ENSO and 500-hPa geopotential height over the Near East): they are consistent with the signals shown in the synchronous composites. The robustness of the LMR and LDA models is demonstrated by the high linear error in probability space (LEPS) scores (44% for continuous variables and 51% for categorical variables) obtained on cross-validated results.