| Abstract |
: |
Flood Frequency Analysis (FFA) uses both flood peak series, i.e. Annual Maximum Series (AMS) and Partial Duration Series (PDS). AMS analyses number of peaks equal to number of observation years using a single best fit probability distribution, while PDS analyses peaks over a threshold value adopting Poisson distribution (PD) and negative binomial (NB) distribution counting occurrences of peaks (μ) over threshold and exponential distribution (ED), generalized logistic distribution (GLD) or Pareto distribution (GP) for magnitude. Both ED and GLD method are used to analyze at-site extreme flood information in the PDS model at Champua gauging site under Baitarani Basin, Odisha (India). Performances of AMS/PD-ED, AMS/NB-ED, AMS/PD-GLD, AMS/NB-GLD are compared in terms of uncertainty of T-year event estimator. Results show AMS/PD-ED model yields lower variance of T-year estimate than AMS/NB-ED model, while AMS/NB-GLD gives less variance in comparison to AMS/PD-GLD. Results indicate that in case of ED, flood quintiles are acceptable for μ≤8 and T≤500 within 95% confidence limit, while they are comparable for μ≤3 and T≤100 in case of GLD distribution. To summarize, NB distribution is preferred for number of flood peaks if coupled with GLD for flood exceedances and Poisson distribution in case of ED for flood exceedances values. |
