Nonstationary and nonlinear approaches for the analysis and prediction of hydroclimatic variables in eastern and southern Africa

Motivated by the lack of knowledge on the nonstationarity of hydroclimatic processes and the nonlinearity of the interation among hydroclimatic variables in Eastern Africa (EA), Central Southern Africa (CSA), Southern Africa (SA), and the Indian and Atlantic Oceans, this thesis has developed the methods of Wavelet and Hilbert empirical orthogonal functions (WEOF and HEOF) and the Wavelet and Hilbert independent component (WICA and HICA) analyses to identify the spatial, temporal and frequency variability regimes of the regional climate.; The nonlinear genetic algorithm neural network algorithm (ANN-GA) model is developed to predict the variability of hydroclimatic variables through teleconnection. The ANN-GA-disaggregation-soil moisture accounting (ANN-GA-DIS-SMA) model is developed to predict weekly streamflow from seasonal oceanic variability. The combination of ANN-GA and a statistical disaggregation model is developed to predict weekly streamflow directly from predicted seasonal rainfall.; The WEOF and HEOF have helped to extract information on nonstationary spatial, temporal and frequency patterns of the sea surface temperature (SST) of the Indian and Atlantic Oceans and the rainfall of EA, CSA and SA. This new information facilitates the accurate prediction of seasonal rainfall for the East and Southern Africa region and long term planning of agriculture and water resource management. For an 11-year validation period (1987-1997), ANN-GA accounted for 49-81% of the variance of observed EA September-November and SA summer rainfalls and 67-81% of the observed EA March-May rainfall. Using the 1984-1995 validation period for CSA rainfall, ANN-GA captured 64-81% of the rainfall variance. The ANN-GA-DIS-SMA has shown considerable skill in predicting weekly streamflow from weekly rainfall disaggregated from seasonal rainfall predicted from the seasonal SST data, and can explain 81-96% of the observed weekly streamflow variance. The ANN-GA-DIS model has shown relatively weaker skill with only 61-84% of variance explained.; The analysis of scale-based energy helped determine the effects of the El Nino southern oscillation (ENSO) on the rainfall of EA, CSA and SA. Knowledge of this effect will be useful to the countries of the region in preparing themselves for the impending droughts threat and mitigating the ENSO impact.