The use of the Akaike Information Criterion in the identification of an optimum flood frequency model.

Abstract

The results of the chi-square goodness-of-fit test were compared with those of the more objective Akaike Information Criterion (AIC) in order to identify the optimum peak-flow probability model for use in flood frequency analysis. Nine probability functions were investigated. These included 7 three-parameter distribution functions (log-normal, Pearson, log-Pearson, Fisher-Tippet, log-Fisher-Tippet, Walter Boughton and log-Walter Boughton) and 2 five-parameter density functions (Wakeby and log-Wakeby). The data used comprised the annual peak discharge from the average daily records for 60 river gauging stations within 5 major river basins of Kenya. The AIC test isolated the Wakeby and the three-parameter log-normal distributions equally but significantly from all the other distributions considered in the study. Because of the complexity of the Wakeby distribution the three-parameter log-normal distribution was considered to the best model for use in flood frequency analyses in Kenya. One of the worst fitting models was the log-Pearson distribution