ABSTRACT
Flood frequency analysis is an established method for determining critical design discharges for small to moderate sized hydraulic structures. In water resources planning and other water related design projects, engineers require flood estimates at particular site locations; this includes the design of structures like spillways, bridges, culverts, water supply systems and other diversion structures. Flood volume estimates are also required for flood plain zoning and the design of flood control structures such as levees and barrages. The major problem usually encountered in many aspects of water resources engineering is that of estimating the return period of rare events, such as extreme floods or precipitation for a site or group of sites (Hosking, 1994; Cunane, 1989). With adequate records, statistical methods will show that floods of certain magnitudes may, on the average, be expected annually, every 10 years, every 100 years and so on. Frequency analysis involves the definition and selection of the type of hydrological event and extreme characteristics to be studied, the selection of
an appropriate extreme value model and probability distribution to describe the data, the estimation of the parameters of the distribution, and lastly, the calculation of extreme values or risks estimated for the given problem. When hydrologists assign a return period to an extreme flood event, the accuracy of estimation depends on the length of the record; it may be necessary to have a record that covers more than 30 years. The records of observed flood events should be comprised of independent series; the maximum event in each year is extracted since it is unlikely that the maximum flow in one year will be affected by that of the previous year. Therefore a selection of this type is known to constitute an annual series of data. Existence of extremely large values (outliers) in data series causes most statistical flood frequency methods to underestimate the occurrence of very large floods. If these outliers are omitted from the analysis, the resulting quantiles are more accurate, but the sample can no longer be regarded as random and unbiased. Cunane (1989) observed that the effect of these outliers is negligible if appropriate methods for parameter estimation are applied. The challenges being faced
today are centred on determining the most appropriate form of model, the “underlying distribution” of floods, and then estimating the parameters of this underlying distribution (Ware and Lad, 2004).
