Abstract
One of the major concerns in Flood Frequency Analysis (FFA) is to predict floods of high magnitude for larger return periods. Magnitudes of smaller floods behave as nuisance in the estimation of larger floods. In this study, to avoid the unnecessary effect due to smaller observations, we implemented the technique of left censoring. The primary objective of this study is to see the efficacy of censoring, by comparing Regional Flood Frequency Analysis (RFFA) using Partial Linear Moments (PLM) for censored samples with RFFA using Linear Moments (LM) for uncensored samples of annual peak flows observed at ten stations of Indus Basin in Pakistan. After fulfillment of fundamental assumptions of randomness, independence, homogeneity, and stationanty, a Grubbs-Beck (GB) test for outlier detection is applied to the samples from all stations. For further analysis, Discordancy measure shows that none of the site is discordant and all ten stations would be retained for further investigation. On the basis of geographical closeness, stream hydrology and morphology of Indus basin, a single homogenous region is proposed and testified by the heterogeneity standards The best fit distribution is selected by implying the Z-statistic (goodness of fit test base on LM) and L-Moments Ratio Diagram (LMRD). Generalized Pareto Distribution (GPA) distribution under PLM while Generalized Normal Distribution (GNO) under LM is selected as the reasonable choice for design flood estimation. Monte Carlo simulation experiment is performed to check the efficacy of PLM over LM through Root Means Square Error (RMSE) and bias. These accuracy measures indicated the outperformance of censored samples under PLM to uncensored samples under LM.