The use of robust procedures for air pollution estimation in Tomsk

To solve the problem of air pollution it is necessary to examine pollution levels in space and time. Measuring the air pollutants concentrations it is difficult to avoid different mistakes. That is why the correct estimation of the pollution level is necessary for making optimal environmental decisions. As a rule, in a sample, crude measurement errors or bad values occur, these errors and bad values garble the average value, especially the dispersion value. The paper gives a link where 10% of measurements producing bad values twice increase the dispersion estimation. Many researchers exclude bad values from data processing as they do not belong to this distribution. Other researchers exclude the drop out observations and examine them separately, as bad values can be of greater interest than the sample itself. There are a lot of recommendations on the detection and screening of bad values. The paper considers one of the methods of estimation of the pollutant fields distribution parameters. This method deals with the procedures insensitive to the data structure - the robust estimation procedures. Two most common robust estimations, Winsorized and trimmed estimators, are applied. Two variants of the statistic characteristics calculation of pollution levels are used. In the first variant the calculation was conducted using the whole data base including cases with zero pollution level. In the second variant only cases with the pollution level different from zero were used. The results showed that the value of the average pollution level when calculated by the second variant was 2-3 times higher in comparison with the first variant. Standard deviation values in both variants do not differ. In the structure of the annual course of standard deviations two maximums in May and December can be traced, they differ from other months values by 2-3 times. It has been discovered that the nitrogen dioxide concentration in May exceeds the maximum permissible concentration by 5 times, in December - by 2 times. These cases can be referred to as abnormal. The comparison of results shows that the average value, the average quadratic deviation and the confidence interval value is lower or equal when the trimmed estimator is applied rather than the Winsorized one. Thus, the usage of the robust estimations allows not only to correct the average values of the sample taking into account emissions but also to define the abnormal pollutants concentrations values.

Keywords

Authors

References