Design flood estimation for ungauged catchments in Krishna River Basin using a transformation-based approach to Regional Flood Frequency Analysis

Abstract

Estimation of flood quantiles at ungauged sites is a vital aspect of the design and planning of hydraulic structures. There are various approaches such as Conventional Index Flood (CIF), Logarithmic Index Flood (LIF), and Population Index Flood (PIF), etc, that have been established to evaluate flood quantile at ungauged sites. These conventional approaches assume that the scale and shape parameters of frequency distributions remain identical for all the sites in a homogeneous region. However, this assumption may not be valid for hydrologically similar real-world catchments. Recently, a transformation-based mathematical approach to regional frequency analysis was proposed by Basu and Srinivas (2013), which ensured the assumption of having identical scale and shape parameters across the sites in a hydrologically similar homogeneous region. The approach involves (i) identification of appropriate frequency distribution representing the homogeneous region, (ii) Mapping the flood quantile (corresponding to various non-exceedance probability) from the original space to a dimensionless space, where values of parameters of distributions at sites in the region remain identical, (iii) construction of regional growth curve in dimensionless space and, (iv) Mapping of dimensionless regional growth curve to original space by applying inverse transformation equations. This study presents an application of the approach given by Basu and Srinivas (2013) for estimating the design flood estimate at ungauged catchments in the Krishna river basin. The delineation of hydrologically similar regions is performed by using a global k-means clustering algorithm. In this study, The parameters of the inverse transformation equations are obtained by using log-linear regression model (LLRM), generalized additive model (GAM), and multivariates adaptive regression spline (MARS). Finally, a comparative analysis is performed to assess the efficacy of the regression models in estimating the parameters of transformation equations. The result revealed that the Regional Flood Frequency Analysis (RFFA) using Basu and Srinivas's (2013) approach is effective for reliable prediction of design flood estimates in Indian watersheds.