Predicting Kenya Short Rains Using the Indian Ocean SST

The rainfall over the Eastern Africa is charaterized by the typical bimodal monsoon system. Literatures have shown that the monsoon system is closely connected with the large-scale atmospheric motion which is believed to be driven by sea surface temperature anomalies (SSTA). Therefore, we may make use of the predictability of SSTA in estimating future Easter Africa monsoon. In this study, we tried predict the Kenya short rains (Oct, Nov and Dec rainfall) based on the Indian Ocean SSTA. The Least Absolute Shrinkage and Selection Operator (LASSO) regression is used to avoid over-fitting issues. Models for different lead times are trained using a 28-year training set (2006-1979) and are tested using a 10-year test set (2007-2016). Satisfying prediciton skills are achieved at relatively long lead times (i.e., 8 and 10 months) in terms of correlation coefficient and sign accuracy. Unlike some of the previous work, the prediction models are obtained from a data-driven method. Limited predictors are selected for each model and can be used in understanding the underlying physical connection. Still, further investigation is needed since the sampling variability issue cannot be excluded due to the limited sample size.