REGIONAL FLOOD FREQUENCY ESTIMATION IN INDIA

Abstract

Estimation of magnitudes of likely occurrence of floods is of great importance for design of various types ofhydraulic structures. Floods ofdifferent return periods are also required for taking up some of the non-structural measures of flood management. As per the Bureau of Indian Standards hydrological design criteria, frequency based floods find their applications in estimation of design floods for almost all the types of hydraulic structures viz. small size dams, barrages, weirs, road and railway bridges, cross drainage structures, flood control structures etc., excluding large and intermediate size dams. For design of large and intermediate size dams probable maximum flood (PMF) and standard project flood (SPF) are adopted, respectively. However, in these two cases also flood frequency analysis is invariably performed for assessing the return periods of PMF and SPF. Whenever, rainfall or river flow records are not available at or near the site of interest, it is difficult for hydrologists or engineers to derive reliable design flood estimates directly. In such a situation, regional flood frequency relationships developed for the region are one of the alternative methods, which may be adopted for estimation of design floods especially for small catchments. As the studies on flood frequency estimation in India are limited, scattered and mostly based on the conventional techniques; hence, there is an urgent need for making systematic efforts for developing a reliable and convenient regional flood frequency estimation procedure based on the state of art technique for gauged and ungauged catchments. Further, the soft computing techniques offer real advantages over conventional modeling, including the ability to handle large amounts of noisy data from dynamic and nonlinear systems, especially when the underlying hydrological relationships are not fully understood. These techniques viz. Artificial Neural Networks (ANN) and Fuzzy Logic (FL) have been applied for solving some of the hydrological problems such as development of stage-discharge relationship, flood forecasting, rainfall-runoff modeling, estimation of precipitation and evaporation, ground water modeling, water quality modeling etc. However, applications of ANNs in regional flood frequency estimation are limited and use of Fuzzy Logic in regional flood frequency estimation remains to be investigated. Whereas, some of the recent studies show that the fuzzy modeling is more versatile and improved alternative to ANNs. In this study, regional flood frequency relationships have been developed for 17 hydrometeorologically homogeneous categorized Subzones of India using the Lmoments approach. The applicability of soft computing techniques viz. Artificial Neural Networks (ANN) and Fuzzy Inference System (FIS) in regional flood frequency estimation has also been investigated. The L-moments form basis of an elegant mathematical theory and can be used to facilitate the estimation process in regional frequency analysis. The L-moment based methods are demonstrably superior to those that have been used previously, and are now being adopted by many organizations worldwide. For carrying out the regional flood frequency estimation study, screening of the annual maximum peak flood data has been carried out for assessing the suitability of the data for regional flood frequency analysis by the Lmoments based Discordancy (Dj) statistic test. The regional homogeneity of the 17 Subzones has been tested employing the L-moments based heterogeneity measure (H) by carrying out 500 simulations using the four parameter Kappa distribution. For carrying out regional flood frequency analysis studies based on the L-moments approach twelve frequency distributions viz. Extreme Value (EV1), General Extreme li Value (GEV), Logistic (LOS), Generalized Logistic (GLO), Normal (NOR), Generalized Normal (GNO), Exponential (EXP), Uniform (UNF), Generalized Pareto (GPA), Pearson Type-Ill (PE3), Kappa (KAP) and five parameter Wakeby (WAK) have been used. Based on the L-moment ratio diagram as well as Zdist -statistic criteria robust frequency distributions have been identified for the 17 Subzones of India. The 17 Subzones covertotal 25,89,342 km2 area, which constitutes about79% of the geographical area of India. The annual maximum peak flood data and catchment areas of 261 streamflow gauging sites of the 17 Subzones of India were collected for carrying out the study. Outof these, the data of 196 streamflow gauging sites and their catchment areas have been used for regional flood frequency estimation. The data of remaining 65 streamflow gauging sites have been excluded as per the data screening and regional homogeneity testing procedures. The record length for these streamflow gauging sites varies from 5 to 38 years. The catchment areas of the streamflow gauging sites range from 6 km2 to 2,297 km2 and their mean annual peak floods vary from 12.8 m3/s to 1687.3 m3/s. Out of the 17 Subzones, PE3 has been identified as the robust distribution for 7 Subzones, GNO for 3 Subzones, GEV for 3 Subzones, GPA for 3 Subzones and GLO for 1 Subzone of India. The regional flood frequency relationships have been developed based on the respective robust identified frequency distributions for estimation of floods of various return periods for gauged catchments for the 17 Subzones. For estimation of floods of various return periods for ungauged catchments, the regional relationships have been developed betweenmean annual peak floods and catchments areas of the gauged catchments of the 17 Subzones using the Levenbergiii Marquardt (LM) iteration procedure. The performance of this technique has been evaluated based on the statistical performance indices viz. Efficiency (EFF), Correlation Coefficient (CORR), Root Mean Square Error (RMSE) and Mean Average Error (MAE). The regional relationships developed between mean annual peak floods and catchments areas for the 17 Subzones have been coupled with the respective L-moments based robust identified regional flood frequency relationships developed for gauged catchments for each of the Subzones. The regional flood frequency relationships have also been developed for estimation of floods of various return periods for gauged and ungauged catchments for 4 Subzones out of the 17 Subzones using ANN and FIS techniques. Performances of ANN, FIS and L-moments in regional flood frequency estimation have been compared based on the statistical performance criteria viz. EFF, CORR, RMSE and MAE. The regional flood frequency relationships developed in the present study based on L-moments provide a convenient method for estimation of floods of various return periods for gauged and ungauged catchments of the 17 Subzones of India for the practitioners. The applicability of ANN and FIS in regional flood frequency estimation is explored and comparison of ANN, FIS and L-moments establishes the potential of FIS in regional flood frequency estimation.