Dam safely : an evaluation of some procedures for design flood estimation

Cluckie (1990) states that "when a site has about 40 or 50 years of record it is unlikely to contribute significantly to the improvement of a regionally based statistical estimate by the availability of more data". In the present paper, the authors conclude that "databases with more than 300 station-years can be considered in a quite good condition in terms of quantity of information and that, consequently, the addition of more records may not materially affect the outcome of any future analysis". Coupled with the encouragement given to downweight or exclude "outliers", there would seem to be an inference that there is sometimes an overabundance of information for flood frequency estimation. While this may be true in lesser contexts, the conclusion seems misplaced in a paper which discusses dam safety procedures. The authors' use of a pooling system based on distance from the subject site has similarities with that put forward for rainfall frequency estimation by Reed & Stewart (1989) but does not make any explicit allowance for interdependence in the gauged annual maxima. Both Schaefer (1990) and Buishand (1990) appear to accept Hosking & Wallis's conclusion that intersite dependence adds uncertainty but no bias to regional flood estimates (Hosking & Wallis, 1988). The example which follows demonstrates that the GEV-PWM regional procedure can be highly sensitive to the occurrence of a widespread extreme event. In these circumstances it appears advisable for a regional flood frequency analysis to exploit all the available information, and to make some allowance for inter-site dependence. This finding does not sit comfortably with the conclusions of Cluckie & Pessoa (1990) and Hosking & Wallis (1988).