Estimation of Likely Impact of Climate Variability on Runoff Coefficients fromLimpopo Basin using Artificial Neural Nefwork (ANN)

Abstract

Forecasting future response behaviour of a semi-arid catchment in terms of runoff coefficient being trivial, an attempt has been made io apply an Artificial Neural Network (ANN) model to-forecast the run off coefficients (ROC) for the Limpopo catchment system in Botswana. ROCs computed from l97l to lfl00' h-v the water balance technique have been used to develop the optimal network architecture with appropriate choice of the size of input vectors, number of hidden layers and number of neurons in the hidden layers, training algorithms and transfer functions for the network. Based on its performance in terms of reproducibility of the water balance run off coefficients, the network was used to forecast the runoff coefficients up to 20 1 6. For the decades between I 971 - I 980, 1981-1990 and 1991-2000 the average runoff coefficients were found to be 0.40 to 0.4.l and0.47 respectively' The average forecast runoff coefficients for the decade 2001-2010 and the period 2011 - 2016 were found to marginally increase to likely values of 0.48 and 0.50 respectively. This may therefore need an appropriate watershed management strategy to conserve soils and run off from the basin.