Assessment of Predictive Accuracy for Selected Regional Flood Frequency Distribution Estimation Methods. A case of upper Omo-Gibe River Basin, Ethiopia

Abstract

Floods are one of the most common and destructive natural disasters, causing substantial loss of life and property all over the world. Assessment of predictive accuracy for regional flood frequency distribution estimation method has been the backbone of water resources project planning, design of any structures, and the economic analysis of flood control projects. The goal of this study was to test the predictive fit of probability distributions to yearly maximum flood data and to determine which distribution and estimation method provide the best match for the Upper Omo-Gibe River Basin. Using a river basin as a case study, the performance of nine probability distributions, three fitting tests, evaluation processes, and selection procedures was examined. To achieve this, data from eleven stream gauged sites, three hydrological homogeneous sub-regions were defined and delineated based on L-moment homogeneity tests, namely Region-A, Region-B, and Region-C. Delineation of homogeneous regions was accomplished using ArcGIS10.4.1. Discordancy of regional data of the L-moment statistics was identified using Matlab2018a. The R programming language was used in conjunction with a collection of the most recent computer statistical programs in an integrated development environment. The most relevant distribution models were identified using maximum likelihood estimation, goodness-of-fit tests-based analysis, and information criteria-based selection techniques. The performances of the distributions were evaluated using Kolmogorov Smirnov, Anderson-Darling, and Chi Squared goodness-of-tests. After three goodness of fit tests were carried out, the results showed that the lognormal and gamma distribution models were the best-fit functions for Region-A. Because they had the lowest Akaike Information Criterion (AIC) values of - 46.251 and -45.802, and Bayesian Information Criterion (BIC) values corresponding to - 43.320 and -42.870, respectively. Similarly, for both Region-B and Region-C the lognormal and gamma functions were the best-fit distribution functions and identified as suitable distributions for analyzing accurate annual maximum flows in the basin. Based on best-fit distributions for the three regions, regional flood frequency curves were constructed. In this study, the flood magnitude is estimated for 2, 5, 10, 15, 20, 25, 50, 75, 100, 200, 500, and 1000 years return period, and their respective extreme event for Region-A became 427.65, 547.24, 626.42, 671.1, 702.37, 726.46, 800.6, 800.68, 843.82, 874.35, 917.31, and 947.75. 1044.59. The derived flood frequency curves at a given return period suggested that how important engineering decisions and actions such as design and operation of the water resources project have to be undertaken carefully