Abstract

Coyle, A.J., Kelbe, B.E., Reed, D.W., Stewart, E.J. 1991. A temporal look at hydrological extremes. Proc. National Hydrol. Symp., Southampton, September 1991, 6.51–6.60.

Hydrology has long been a fertile area for the application and development of extreme value methods. The paper reports preliminary results from a study which, for once, is concerned less with estimating the magnitude or frequency of extremes than with indexing their temporal character. In particular, the project is examining the effect that data discretization has on the estimation of extremes.

It is rare for a hydrological analysis to use "continuous" data. Often it suffices to consider daily or monthly totals rather than attempt to treat data recorded with very great temporal detail. In many cases there is little choice; higher resolution values were either never available, have been lost, or would now be too costly to extract from the medium on which they are held. Whereas daily or monthly totals provide no obstacle to the estimation of long-term average values, it is known that the estimation of extremes is biased by discretization. For example, it is common to apply a multiplier of 1.11 or 1.14 to convert statistics of 1-day maximum rainfall to their 24-hour counterparts.

Improved logging and storage systems, and the passage of time, have led to a greater number of high resolution datasets becoming available and it is now possible to examine discretization effects more thoroughly. The paper presents analyses of a range of hydrological and meteorological variables. While one product of the research will be more comprehensive correction procedures, a greater outcome may be a better understanding of interrelationships between variables through appraisal of their temporal character. For once, statistics may complement rather than displace a more physical approach. Some preliminary results are presented.

The large data sets now available to hydrologists should be valuable in resolving a number of the scaling problems which face environmental research in the 1990s.