Flood Frequency Analysis of Partial Duration Series Using Soft Computing Techniques for Mahanadi River Basin in India

Abstract

In flood frequency analysis, the modeling based on Annual Maximum Flood (AMF) series remains the most popular approach. An alternative approach based on the “partial duration series (PDS) or peaks over threshold (POT)” has been considered in recent years, which captures more information about extreme events by fixing appropriate threshold values. The PDS approach has lot of advantages, (i) it consist more peak events by selecting the appropriate threshold hence to capture more information regarding the flood phenomena. (ii) it analyses both, the time of arrival and the magnitude of peaks, (iii) it provides extra flexibility in the demonstration of floods and a complete explanation of the flood generating process. However, the PDS approach remains underused and unpopular due to the nonexistence of general framework regarding different approaches.The first objective of the present research work is to develop a framework in the above question on selection of an appropriate threshold value using different concepts and, to verify the independency and stationarity criteria of the extreme events for the modeling of the PDS in the Mahanadi river system, India. For the analysis, daily discharge data from 22 stations with record length varying between 10 and 41 years have been used with the assumption that the whole basin is homogeneous in nature. The results confirmed that the Generalized Pareto (GP) best described the PDS in the study area and also, show that the best PDS/GP performance is found in almost all the value of λ (2, 2.5 and 3). In the second phase, the analysis is done to carry out the regional flood frequency analysis in the Mahanadi basin and to apply the developed model to the respective homogeneous region. Regionalization is the best viable way of improving flood quantile estimation. In the regional flood frequency analysis, selection of basin characteristics, morphology, land use and hydrology have significant role in finding the homogeneous regions. In this work the Mahanadi basin is divided into homogeneous regions by using fifteen effective variables initially. However, it has been observed that the whole basin is not hydro meteorologically homogeneous. Therefore, Factor analysis has been introduced in finding suitable number of variables, and nine variables are found suitable for analysis. Hierarchical (HC) and K-Means Clustering (KM) techniques are used for finding out the possible number of clusters. Here, again the Generalized Pareto (GP) distribution best described the PDS in the study area. To test the homogeneity and to identify the best-fit frequency distribution, regional L-moment algorithm is used. A unique regional flood frequency curve is developed which can estimate the flood quantiles in ungauged catchments and an index flood is also specified concerning the catchment characteristics by using the multiple linear regression approach.