I read with interest the letter by Colin Clarke and P R Rakhecha about dealing with maximum flood (December 1998, p10). I think it may have been Winston Churchill who said ‘don’t believe statistics unless you have carried out the manipulation yourself’, and the determination of floods must fall within this category.
The use of a finite period of data to predict the magnitude of an extreme flood results in the anomalous situation that the longer the period of record, the higher the calculated values are likely to be. The same is true with the prediction of a hydroelectric scheme’s absolute firm energy output, which becomes lower. However, at the time of design, the best possible predictions must be made, even if little or no data are at hand. By the time the 1957 floods arrived at the Kariba hydro power station during its construction, only 34 years of daily river stage data was available, plus 14 years of river level data taken once a week. Yet the probable maximum flood had to be predicted for the dam’s design, and it had to be adjusted after those floods and again after the 1958 floods (see The hydrological operation of the Kariba hydroelectric scheme — past, present and future, 16th icold congress, US, June 1998). Public participation would not have helped here.
I have often thought of carrying out a study to determine the accuracy of extreme flood predictions from finite data, but have had neither the time or resources, and I imagine that someone somewhere has already done this. I envisage:
A) Generating 10,000 years of data.
B) From this establishing the magnitude of the 1:10,000, 1:2000, 1:1000, etc, years events.
C) Then, assuming one is standing at, say, year 50 and then 100, 200 etc, making similar predictions (perhaps using 50, 100, 200 years, etc).
D) Making comparisons between B and C and perhaps arriving at a factor to be applied to the prediction of a 1:x event using y years of data. The results would probably only be believed by the manipulator.
I do not know what to do about land use changes yet, but I do know that this must be taken into account somehow. The sensitivity of these changes can be determined using a catchment model such as a model by W V Pitman (A mathematical model for generating monthly river flows from meteorological data in South Africa, Hydraulic Research Unit, report 2/73, University of Witwatersrand 1973. See also A comparison between automatic and trial-and-error calibration of a hydrologic simulation model, MSc thesis in water resources technology, University of Birmingham, UK 1980).
In Zimbabwe, with the paucity of data, which is getting worse and not better, flood magnitudes are predicted using a formula that depends merely on the catchment area (see Spillway design principles in Zimbabwe, 60th ICOLD executive meeting, Spain 1992). How accurate this is we will probably never know, but it is an ingenious method devised in 1974 and the best prediction tool available considering the information available. The architect of this methods, Eng TB Mitchell, recently updated the parameters (Annual river flood frequency in Zimbabwe, The Zimbabwe Engineer, October 1998).
I would be interested in the results of the experiment initiated by the authors Colin Clarke and P R Rakhecha, producing revised estimates of PMP.