Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/3087
Authors: Banerjee, Satyabrata
Issue Date: 1991
Abstract: In flood frequency analysis, a prticular model's ability to estimate flood quantiles is of great importance. A large number of techniques are in vogue for estimating the quantiles of a particular return period from given sample of data. These techniques give different results when applied to a common data set. Such results as such cannot provide any information regarding their correctness. Techniques can be best compared on the basis of their performance for estimating a quantile, if applied to a randomly drawn known population, for which the true values of quantiles are known. For comparing the performance of various distri-butions, few scattered attempts, based on simulation experi-ments have been made in the past and as such a consolidated study for various distributions on a common experimental design is not available. Such type of studies for normaliza-tion procedures are also not reported in literatures. Keeping in view the gaps in the available works, the present study has been undertaken to compare the perform-ance of twelve commonly used techniques of quantile estima-tion using flood-like synthetic data. These twelve techniques also include techniques based on normalization. For generating flood-like data, eight specific Wakeby populations with varying statistical characteristics have been considered in this study. Loss functions such as Bias, Mean Square Error, Probability of underdesign and underdesign loss etc. were obtained for different return periods, on quantiles estimated by twelve techniques. 200 samples were generated for each of the Wakeby populations for five sample lengths 20, 40, 60, 80 and 100. During the analysis of results, it was observed that some samples produced very high absurd quantile esti-mates for some of the techniques. So the same analysis was repeated after rejecting some of the samples giving absurd quantiles. The analysis reveals that Wakeby-4 parameters distribution and EV1/PWM perform very well, if Bias and MSE are used as comparison criterion respectively. While Wakeby-5 parameters distribution and One Step Power Trans-formation technique perform well, if underdesign loss is considered as criteria. The performance of Two Step Power Transformation improves considerably, if the sample giving absurd quantiles are eliminated from the analysis. This technique overesti-mates the quantiles in most of the cases, thus avoids the risk of underdesigning the flood protection measures. However this overestimation is too much in many cases. The present study had to be restricted only to 200 samples because of large computational time and memory requirements. However this study can be further extended by considering more techniques of flood frequency analysis, and by generating other synthetic samples from known popula-tions of different distributions.
Other Identifiers: M.Tech
Research Supervisor/ Guide: Goel, N. K.
metadata.dc.type: M.Tech Dessertation
Appears in Collections:MASTERS' DISSERTATIONS (Hydrology)

Files in This Item:
File Description SizeFormat 
HYD245545.pdf6.25 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.