Effect of Regionalization on Flood Frequency Analysis

Abstract

Watersheds in a region can be classified into homogeneous groups by using the regionalization methods discussed in the previous chapters. Regionalization is, however, only a prelude to flood frequency analysis and several aspects related to flood estimation remain to be examined. Perhaps the most important of these is related to quantifying the improvement in flood quantile estimation brought about by regionalization of watersheds into hydrologically homogeneous groups. In general, the error in estimation of flood quantiles depends on the method used for regional flood frequency analysis (RFFA). An inefficient method may have errors which could nullify any gains obtained by regionalization. Hence it is preferable to investigate more than one RFFA method in order to assess their relative advantages and to quantify the improvement in flood estimation brought about by regionalization. In this chapter performance of two commonly used methods of RFFA is investigated. The first method is based on the regional L-moment algorithm, whereas the second method is based on regional regression analysis. Another important aspect is related to verifying whether a single distribution is acceptable for estimating flood quantiles in all the regions formed for flood frequency analysis. Bulletin 17 of the U.S. Water Resources Council (1976, 1977, 1981) recommends fitting log-Pearson type III (LP3) distribution to annual maximum streamflows. The stipulation that LP3 distribution must be used over a large area such as the continental United States presumes that it is the preferred distribution. In order to test the validity of such an assumption, the performance of LP3 can be compared with that of several other frequency distributions in fitting peak flow data of different homogeneous regions. If LP3 distribution is preferred to other distributions in such a comparative analysis then it has some justification behind using it. Also, the analysis would be useful to conclude whether any particular frequency distribution is preferred to fit peak flow data in all the regions. Further, in the past decade, simple scaling methods have been developed in flood frequency analysis. It is found that within hydrologically homogeneous regions