TR-59 : Regional flood frequency analysis for Godavari basin Subzone (3F)

Abstract

The most significant development in regional flood frequency analysis was the bringing out of a manual by U.S. Geological Survey in 1960, which was followed by a number of studies including U.K. Flood Studies and other typical studies covering general probability considerations, use of historical information, criteria for regional homogeneity etc. Some important developments during the last fifteen years include use of GEV (General Extreme Value) and Wake by distributions, and use of PWM (Probability Weighted Moments) approach for parameter estimation. The application of PWM technique of parameter estimation provides efficient and reliable flood estimates even for situations where historical records are extremely short. Most of the regional flood frequency analysis studies carried out for some typical regions in India are based on conventional methods such as U.S.G.S. Method, regression based methods and Chow's Method etc. In a few studies, conducted at National Institute of Hydrology and at some academic organisations, attempts have been made to study the applications of new approaches, such as Wakeby (PWM), GEV(PWM) and Power transformation techniques, etc for regional flood frequency analysis of some of the typical regions in India for which the conventional methods have been already applied.

In this study, flood frequency analysis using peak flood series data of hydrometeorologically homogeneous region of Godavari basin (Sub zone 3f) involving application of EVI (PWM) and GEV (PWM) methods based on : i) at site data, ii) at site and regional data combined and iii) regional data alone is described and discussed. Homogeneity of the region has been tested using U.S.G.S Homogeneity test and co-efficient of variation based Homogeneity test. The annual peak flows for 16 to 26' years for twelve sites are considered in two parts. Ten sites data for parameter estimation and two sites data as independent test data. Descriptive ability of the various frequency methods considered for the study have been tested based on some goodness of fit criteria.

In order to evaluate the predictive ability criteria, synthetic flood series have been generated using the regional EV1 and GEV distributions parameters, derived from the historical records. Generated data sets have been considered for ten sites for a specific record length (same as the record lengths of historical data for respective gauging sites) and two independent sites, considering one at a time, of a variable record length (1,5,10,20,24,30 or 40). EV1 (PWM) method has been applied to the generated data of different sample sizes for an independent site considering. i) at site data, ii) at site and regional data, obtained from the generated data of the ten gauging sites, and iii) regional data alone involving the relationship between the mean annual peak flood and catchment area and the regional parameters for the concerned distribution. The computations have been repeated for second independent site on the same lines. Similarly, GEV (PWM) method was also applied to the generated data and computations have been made for the two independent gauging sites taking different sample sizes. The performance of different methods have been evaluated based on the predictive ability criteria, viz. bias, co-efficient of variation and root mean square error computed from the generated samples of different sizes by considering 1000 replications of the computation procedure for each sample length. The results obtained from EV1 (PWM) and GEV (PWM) with generated data of two different populations have been compared for the different methods. It is seen that the method based on GEV(PWM) approach using at site and regional data in a

combined form, provides estimate of flood peaks for different recurrence intervals with computationally less bias and, comparable root mean square error and co-efficient of variation for the two independent catchments. The study, thus, establishes the applicability of GEV(PWM) approach for regional flood frequency analysis considering at site and regional data in the combined form of the the Godavari Basin sub-zone (3f) region.