On the value of information of flood frequency analysis

On the basis of Monte Carlo experiments it was noted that in the absence of information about the distribution of floods and the economic losses associated with the design of flood reduction measures, the use of the normal distribution to represent the distribution of floods is generally better than either the Gumbel, log normal, or Weibull distributions. Nothing is gained in terms of reducing expected opportunity design losses if the underlying distribution F(x) is identified over and above simply using the normal as the assumed distribution. On the other hand, without identifying F(x) but with some detailed specification of the ratio of overdesign to underdesign losses and with a limited specification of the skewness of floods, a considerable reduction in the expected opportunity design losses results.