Outlier treatment in the flood flow statistical analysis

Abstract

In the hydraulic structures design process, the reliable flood estimation is essential for designing the structure. Statistical analysis of observed flood flows, i.e. the flood frequency analysis is one way to obtain the design flood of a specified probability. The problems that can threaten the credibility of flood estimates are related, among other issues, to the treatment procedure of exceptional values (outliers) in the observed flood flow data sets. In this paper a detection of upper and lower outliers in a series of streamflow annual maxima is performed by the Grubbs and Beck statistical test on 68 hydrological stations in Serbia. The data observation period considered is from the establishment of the stations until 2012. The outlier detection results are compared to the previous results obtained for the data processing period until the year 2006.