Abstract:
Many observed annual flood series exhibit reverse curvatures when plotted on lognormal probability paper. None of the existing frequency distributions fits such a series. The physical rationale for the mixed-distribution concept is developed for fitting such shapes. Outliers and inliers greatly distort the fitted distribution parameter. Statistics for their detection at various significance levels are developed from extensive Monte Carlo experiments. A versatile flood frequency methodology with objective detection and modification of any outliers/inliers is developed and computerized. The use of mixed distribution and modification of outliers/inliers are shown to significantly reduce the uncertainty in high flood estimates.
