Oceanic and atmospheric linkages with short rainfall season intraseasonal statistics over Equatorial Eastern Africa and their predictive potential

Abstract

Despite earlier studies over various parts of the world including equatorial Eastern Africa (EEA) showing that intraseasonal statistics of wet and dry spells have spatially coherent signals and thus greater predictability potential, no attempts have been made to identify the predictors for these intraseasonal statistics. This study therefore attempts to identify the predictors (with a 1-month lead time) for some of the subregional intraseasonal statistics of wet and dry spells (SRISS) which showed the greatest predictability potential during the short rainfall season over EEA. Correlation analysis between the SRISS and seasonal rainfall totals on one hand and the predefined predictors on the other hand were initially computed and those that were significant at 95% confidence levels retained. To identify additional potential predictors, partial correlation analyses were undertaken between SRISS and large-scale oceanic and atmospheric fields while controlling the effects of the predefined predictors retained earlier. Cross-validated multivariate linear regression (MLR) models were finally developed and their residuals assessed for independence and for normal distribution. Four large-scale oceanic and atmospheric predictors with robust physical/dynamical linkages with SRISS were identified for the first time. The cross-validated MLR models for the SRISS of wet spells and seasonal rainfall totals mainly picked two of these predictors around the Bay of Bengal. The two predictors combined accounted for 39.5% of the magnitude of the SST changes between the July–August and October–November–December periods over the Western Pole of the Indian Ocean Dipole, subsequently impacting EEA rainfall. MLR models were defined yielding cross-validated correlations between observed and predicted values of seasonal totals and number of wet days ranging from 0.60 to 0.75, depending on the subregion. MLR models could not be developed over a few of the subregions suggesting that the local factors could have masked the global and regional signals encompassed in the additional potential predictors