A COMPARATIVE STUDY OF REGIONAL FLOOD FREQUENCY METHODS

Abstract

Estimates of flood frequency quantiles are very important in planning and design of hydraulic structures. Hence, there is a need to seek for the most appropriate design estimator that would meet both safety and economic considerations of such structures. Over the years a lot of research geared towards achieving this goal nas been undertaken leading to a general agreement that at-site/ regional flood frequency analysis methods yield the most appropriate results Cunnane (1988) writing on the 'methods and merits of regional flood frequency analysis' concluded that the best method currently in use utilises the Wakeby distribution whose parameters are estimated from regionally averaged standardized probability weighted moments. The objective of the present study is to statisti-cally compare the performance of some of the at-site-regional flood frequency methods, cited in literature using annual Peak discharge data of some of the Indian rivers. In this study at-site regional flood frequency analysis has been carried out for 102 sites drawn from 8 different regions: 7 regions from India and 1-region from Laos using USGS, NERC and standardized probability weighted moments methods. For standardized probability weighted moments method Wakeby distribution (4-Parameter as well as 5-parameter), General Extreme value and Extreme value type distribution have been used. All the sites within any given region have been tested for homogeneity by means of Dalrymple's (1960) homogeneity test. The annual flood data of some of the sites within a designated region are used to develop the regional relationship, while the remainder (one or two) are used to test the relationship independently. An attempt has been made to compare the performance of the same methods used for at-site flood frequency analysis. The performance of various methods has been judged on the basis of average of relative deviations between computed and observed value of discharges (ADA), average of squares of relative deviations between computed and observed values of discharges (ADR) and efficiency of method