MULTIVARIATE OUTLIER DETECTION IN A PRECIPITATION SERIES USING TWO-DIMENSIONAL CORRELATION

Abstract

The accuracy of descriptive statistics can be influenced by the existence of outliers in data sets. An observation which might not be considered as an outlier in the univariate case might be a multivariate outlier. Therefore, determination of outliers might make multivariate analysis more robust by providing an opportunity to make the required corrections prior to the modeling studies. This study presents the implementation of a two-dimensional correlation method for the determination of multivariate outliers among the observations from six precipitation stations in Turkey. The two-dimensional correlation method considers the averages of the parts of the whole series instead of the average of the whole series, and it enables determination of the location of the outlier in the compared series. The results obtained point out that an outlier analysis of hydrological variables should consider the two-directional behavior, and that the two-dimensional correlation method presented proved to be a strong alternative to be used in outlier and irregularity detection studies even with a limited number of available data.