Regional flood frequency analysis for Uganda.

Flood frequency analysis aims at establishing Q-T relationship. Regional flood frequency analysis is one of the most practical means of providing flood information at sites with little, unreliable or no at-site data. This is achieved by bringing together data from closely related stations, hydrologically homogeneous, to develop regional flood estimation procedures, which can be used to estimate, flood quantiles for sites with no data. A homogeneous region is an area comprising of sites from which flow characteristics are similar and hence, when flood data is combined can improve estimation of flood magnitudes at any site in the region. In this study, an attempt has been made to delineate Uganda into hydrological homogeneous regions and frequency models to estimate flood magnitudes have been developed for the delineated regions. The methodology used in this study involved data screening, delineation of homogeneous regions, testing whether the delineated regions are homogeneous, identification of underlying statistical distributions of flood flows for the delineated regions, selection of appropriate flood estimation procedures, development of regional flood frequency curves and regression models to predict mean annual floods. The data used for analysis consisted of annual maximum discharge series (AM) from 55 stations scattered around the country. Data was screened for auto-correlation (rl), Cross-correlation (r1,2), and discordancy (Di). Stations, which were found to be auto correlated and discordant, were excluded from further analysis. Delineation of Uganda into homogeneous regions involved sites, which have similar climatic conditions (i.e. mean annual rainfall was used) and geographical continuity. The coefficient of variation measure (CC), the heterogeneity measure (H) and graphical plot methods were used to assess whether the regions were homogeneous. Data from delineated regions was then used to identify underlying statistical distribution to model flood flows, ln order to assess whether a candidate distribution provides an adequate fit to regional data, the Z-statistic goodness of fit measure and the L-moment ratio diagram were used. Selection of most appropriate regional flood estimation procedure was achieved by evaluating the performance of different procedures. The performance of the procedures was observed in terms of predictive ability test indicators i.e. bias, se, rmse, and Pe. The results of the predictive ability tests were then used to select the most robust flood estimation procedure for the respective homogeneous regions. Two procedures, P3PWM and LP3MOM, were identified as the most robust flood estimation procedures to model flood flows for identified regions in Uganda. The procedures were then used to develop flood frequency curves for the regions. Lastly, linear regression between mean annual floods and catchment characteristics was carried out and models for prediction of mean annual floods developed for the different regions.