MODELING TURKISH STREAM FLOWS BY USING PROBABILITY DISTRIBUTIONS

摘要

A frequency analysis of annual instantaneous flood peaks at 117 gauging stations in Turkey (each having at least 30 years of data) was carried out in order to obtain reliable flow quantiles for several return periods, which can be used by operational hydrologists. In each case, a number of probability distributions were tested including Gumbel (G), log-Logistic, Pearson-3, log-Pearson-3, log-normal-2, log-normal-3, general extreme value, Wakeby and log-Boughton distributions. Similarly, three different parameter estimation methods were investigated: method of moments (M), maximum likelihood (ML), and probability weighted moments (PW). A number of statistical goodness-of-fit (GOF) tests including the detailed chi-squared one, Kolmogorov-Smirnov (K-S), and Cramer-von Mises (CvM) GOF tests, were used to assess their match to the sample data. The chi-squared GOF tests indicated that the log-normal-2 (M) and G (M) distributions are amongst the most appropriate models, whilst both the K-S and CvM GOF tests suggested the Wakeby (PW) distribution providing the best fit model for Turkish rivers. The study suggests that the Wakeby distribution is most appropriate for frequency analysis of annual flood peaks for Turkey. Estimated design values, obtained from the Wakeby (PW) model and corresponding to selected return periods, are presented.