Detection of outliers in the sediment fingerprinting method

Selection of the suite subset of tracers, capable of discriminating sediment sources, is the first and the most important step in the sediment fingerprinting method. Selection of the suite subset is carried out by Discriminant function analysis. The presence of outliers affects the suite subset selection and prevents entering the important tracers into the model, hence reducing accurate classification percent of Discriminant function analysis. Therefore, the outliers must be detected and corrected or omitted, if enough evidences were present. In this study, different univariate and multivariate outlier detection methods were used to assess the presence of outliers in geochemical and organic elements and radionuclides of soil samples collected from Ghara aghaj watershed, Makoo township. According to four univariate outlier detection methods, no observations (samples) were outlier on a sufficient number of tracers. The [Median ± 3MAD] and box plot procedures showed better performance in outlier identification than the [Mean ± 3S] and Grubb's test methods. Also, based on multivariate outlier detection methods, namely squared Mahalanobis distance, separate box plots of squared Mahalanobis distance for each of sediment sources, principal component analysis and plot of the squared Mahalanobis distances against the quantiles of the chi-square distribution, no observations were detected as outlier. From perspectives of each of the two group methods, there was no sufficient information and demonstrable proof about true outlierness of any observation. The advantages of the approach adopted in this study are the simplicity and computability of the selected outlier detection methods with commonly used statistical softwares, and the condition that an observation is regarded as outlier if its uniqueness is confirmed with several methods.