IntroductionIn 1977, the Water Resources Council (WRC) probably had little idea that the endorsement of the log Pearson III (LP lid methodology would initiate such a period of activity within the hydrology community. The following review begins with a discussion of the studies which have appeared in response to the Council's action. These early studies questioned the underlying assumptions, suggesting improvements or alternatives to the LP III methodology. The work which followed took many directions. It is impossible to compass the entire scope of this work. So, I have taken the liberty of focusing my comments on two or three major issues which have appeared in the past four years and which, in my opinion, represent some of the most exciting and fertile areas for the next quadrennium. These major areas include accomplishments in the areas of (1) distributional analysis and (2) robust parameter estimation, and what might be called developments of potential in the area of (3) regionalized and geomorphologic approaches to the flood frequency problem. While the primary focus is on American work, I have included some mention of non-American work when the contributions of American hydrologists can be seen more distinctly against a backdrop of global activity.
Underlying Assumptions: The LP III QuagmireAlthough a number of nonparametric analyses have appeared in the literature, traditional flood frequency analysis has been approached primarily as a problem in parametric statistical inference. Peak annual streamflow data are assumed to come from a parent population whose distribution function is known, is analytically expressable, and contains a finite number of parameters. A large number of peak flow distributions have been studied, for example, the normal, log normal, Gumbel and Weibull distributions. Goodness-of-fit procedures then test whether or not the data do indeed fit the assumed distribution with a specified degree of confidence. However, the lack of power of conventional goodnessof-fit tests with respect to typically skewed flood peak distributions contributed to considerable variability in the estimation of design events. In search of a unified approach to flood frequency analysis, the WRC issued Bulletin 17A in 1977 recommending the use of the three-parameter log Pearson (LP lid distribution as the parent distribution. According to many hydrologists, the choice was illadvised [Wallis, 1981]. Shortly after the appearance of Bulletin 17A, articles began to appear in the literature addressing the advisability of the use of the LP III type distribution. Such articles have continued into the current quadrennium. The WRC recommended that annual peak flows be fitted to a three-parameter LP III distribution by using conventional sample statistics of the mean and standard deviation to be estimated by the method of moments. Since sample records are not typically long enough to give a good estimate of skew, the sample skew estimate was to be replaced by a weighted combination of regional and sample skew values. The WR...
