Regional analysis of low flows in Tanzania.

The framework is presented for regional low flow frequency analysis for predicting low flows at ungauged or gauged catchments in Tanzania. A total of 143 catchments were identified for use in this research. The 10 day annual minimum series is examined and only series without zeros are used for this study. A regional GEV estimator with probability weighted moments (PWM) parameter estimation has been found to be an efficient way of combining flow data in regional frequency analysis (Zrinji et al. 1993). As a prelude, the ability of a four parameter GEV distribution in analyzing low streamflows in Tanzania compared to Weibull-3 distribution's is assessed. Goodness of fit tests and predictive ability test for 10 day low flow produce slightly contradicting results. GEV 4 distribution is more robust than Weibull distribution but the hypothesis that annual low flow series without zero flows was largely rejected at 10% level of significance for Tanzanian rivers. The methodology used for delineating the study area into homogeneous regions make use of a coefficient or variation based test and the Hocking and Wallis (1993) approach to regionalization using 1-moments. L-moment ratio diagrams and the goodness of fit test of Hocking and Wallis are used to assess the suitability of selected distributions as regional parent distributions. The merit of the approach is based on the relative low bias in parameter estimated from L-moments as compared to conventional moments. All regions mapped out were found to be homogeneous or possibly homogeneous. However, for some regions, there are too few stations with uncorrelated 10 day low flow series without zero flows. Estimates of low flows quantiles in region seven, the largest region, are unlikely to be reliable.