Integrating Wavelet Empirical Orthogonal Functions and Statistical Disaggregation for Predicting Weekly Runoff for the Upper Kafue Basin in Zambia, Africa

Abstract

Wavelet empirical orthogonal function analysis, genetic algorithm driven neural networks, statistical disaggregation and hydrologic modeling were integrated into a hydrologic framework to predict weekly rainfall and runoff of the upper Kafue and Lunga Rivers in Zambia, Africa. The April–June (AMJ) seasonal variability of the Atlantic Ocean, on the basis of its AMJ sea surface temperatures, was used to predict the annual rainfall of two stations, Ndola in the upper Kafue River Basin and Solwezi in the Lunga River Basin. The predicted annual rainfall at the two stations was disaggregated to weekly totals and then used to simulate the runoff of the upper Kafue and Lunga Rivers. In the upper Kafue basin, runoff from the disaggregated weekly rainfall explained 81% of the runoff variance, compared to 88% when historical weekly rainfall data was used. For the Lunga River, 72% of the observed runoff variance was accounted for, compared to 81% when historical weekly rainfall was used. This scheme demonstrates that if a region is dominated by hydro-climatic processes whose statistics are fairly stationary, it will be possible to use disaggregated rainfall from annual rainfall predicted by a teleconnection model to predict reliable weekly basin runoff up to a year’s lead time useful for an integrated water resources management.