Flood variables frequency analysis using parametric and non-parametric methods

The common methods in the flood frequency analysis are used to examine the flood peak variable while each flood event consists of three naturally random variables: flood’s peak, volume and duration. Furthermore, it was essential to assume the variable of interest follows a specific parametric distribution chosen from an appropriate distributions’ family. With the aim to overcome these limitations a wide range of the well-known parametric distributions functions and non-parametric methods based on kernel density estimation and orthonormal series approximation is considered to estimate the distribution of flood variables. The investigated variables is extracted from annual maximum flood series at Ahvaz hydrometric station. In order to compare between the fitted parametric and non-parametric distributions to the data, we used the statistical criterions such as Akaike Information Criteria, Bayesian Information Criteria and the Root mean square error. We then used the chi-square goodness of fit test to examine that the acceptability of the chosen distribution in the previous stage. The results showed that the flood peak follows the parametric log Pearson type III distribution function while the non-parametric orthogonal series approximation were the best fit for the flood volume and flood duration. The estimated distributions based on the orthogonal series expansion were able to capture the graphical features of the data and the corresponding fitted densities reproduced the same unimodality or multi-modality that the original histogram of data illustrates. This is not the case when one uses the traditional methods to analyze the frequency of flood.