Statistical flood frequency analysis is subjet to many uncertainties, namely, (i) the identification of the true distribution, (ii) the effect of small size in extrapolating a large return period flood, (iii) the assignment of probabilities to the observed historical series (plotting position), (iv) the procedure for including an important historical information in the analysis. It is shown that the present practice of ignoring thise uncertainties leads to unreliable point estimation of T year flood. Selecting a single best fit distribution entirely based on data for extrapolating T year flood may not be the correct procedure due to small size of the data itself. In the frist part of the study, Bayesian regression analysis and the weithing procedure is used in arriving at a single point estimate of T year flood assumung that the data may to any one or more of four different distributions (Gumbel, LogNornal, Pearson and LP III) along with the various plotting positions (Hazen, Blom, Gringorton and Weibull). The procedure is applied on a real flood data and the results discussed. The second part of the study deals with discontinuous record, i.e. when along with small sample of continuous n years record an added historical information is available, for example, an unusual high flood occurred much earlier. The objective is to evolve a procedure for incorporating this historical information in the analysis and more importantly to estimate whether the inclusion of this information reduces the uncertainty in the estimation of T year flood. In this study, the discontinuous historical information is included in the analysis using the generalized plotting position derived on the basis of order statistics. The Bayesan regression analysis is then undertaken to check whether the inclusion estimation of historical information reduces the uncertainty in the estimation of T year flood or not. The methodology is illustrated through a real flood data and the results discussed.
Weighting Uncertainty in Flood Frequency Analysis
MANCIOLA, Piergiorgio;
1987
Abstract
Statistical flood frequency analysis is subjet to many uncertainties, namely, (i) the identification of the true distribution, (ii) the effect of small size in extrapolating a large return period flood, (iii) the assignment of probabilities to the observed historical series (plotting position), (iv) the procedure for including an important historical information in the analysis. It is shown that the present practice of ignoring thise uncertainties leads to unreliable point estimation of T year flood. Selecting a single best fit distribution entirely based on data for extrapolating T year flood may not be the correct procedure due to small size of the data itself. In the frist part of the study, Bayesian regression analysis and the weithing procedure is used in arriving at a single point estimate of T year flood assumung that the data may to any one or more of four different distributions (Gumbel, LogNornal, Pearson and LP III) along with the various plotting positions (Hazen, Blom, Gringorton and Weibull). The procedure is applied on a real flood data and the results discussed. The second part of the study deals with discontinuous record, i.e. when along with small sample of continuous n years record an added historical information is available, for example, an unusual high flood occurred much earlier. The objective is to evolve a procedure for incorporating this historical information in the analysis and more importantly to estimate whether the inclusion of this information reduces the uncertainty in the estimation of T year flood. In this study, the discontinuous historical information is included in the analysis using the generalized plotting position derived on the basis of order statistics. The Bayesan regression analysis is then undertaken to check whether the inclusion estimation of historical information reduces the uncertainty in the estimation of T year flood or not. The methodology is illustrated through a real flood data and the results discussed.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
