Seaf - A prototype of an expert system for at- site frequency analysis of hydrological annual maxima

Abstract

The usually short-sized samples of data recorded at a site and the large uncertainties involved in parameter and quantile estimation are some of the shortcomings encountered in at-site frequency analysis of hydrologic annual maxima. Another drawback is the subjectivity entailed by the arbitrary selection of a candidate probability distribution to model the sample of hydrological annual maxima. Conventional goodness-of-fit tests are not designed to discriminate among candidate models and are not powerful enough to provide the necessary objective backing to such a decision-making process and may possibly lead a novice hydrologist to inadequate choices. In this paper, we describe t he authors experience in employing the technology of artificial intelligence and fuzzy-logic theory to Guild a computer expert system that emulates the reasoning principles used by a human expert to select one or more candidate probability distribution models for at-site hydrologic frequency analysis. The expert system has been applied to 20 relatively large samples of annual maxima of daily rainfall and daily streamflow recorded at gauging stations in the Brazilian southeast. In order to check the system performance, the same samples have been submitted to a panel of actual experts in frequency analysis. The comparison of the results provides evidence that the computer system performs at an expert level and may be utilized to help an inexpert to select one or more appropriate candidate distributions for at-site hydrologic frequency analysis.