New approaches for modeling the regional pollution in Europe

Generally, official statistical reports provide information on the pollution extent over a region using the average records from all the observation sites. In the outliers' presence, the average is not a good choice. Therefore, in this article, we propose two alternatives for replacing the average series by most significant regional series, obtained by two selection procedures. The first algorithm chooses the candidates to be utilized for the regional estimation of pollution by a data segmentation that provides the most representative value for a given time interval. Since the number of segments to be used should be prior introduced, the second algorithm proposes a version of the selection procedure based on the k-means algorithm. The performances of these methods are verified on three groups of series (carbon oxides, sulfur oxides, and nitrogen oxides) recorded in the EEA33 countries during a period of 28 years. Both algorithms give better results than the average series, in terms of mean standard errors (MSE) and mean absolute errors (MAE).