Presentation Information
[AAS08-P24]A Comparison of Central Tendency Measures for Estimating Annual Averages of Atmospheric Pollutant Concentrations based on data from Russian EANET sites
*Ekaterina Zhigacheva1, Vladimir Ivanov1, Ekaterina Pozdnyakova1, Sergey Gromov1,2 (1.Institute of Global Climate and Ecology, Moscow, 2.Institute of Geography, Russian Academy of Sciences, Moscow)
Keywords:
Air pollution,trend evaluation,central tendencies,monitoring data
Monitoring of minor atmospheric impurities in the air began at Russian EANET sites in the early 2000s, providing over 20 years of data on various inorganic pollutants. The first essential step in estimating interannual trends of pollution levels is correctly calculating annual averages. The arithmetic mean is commonly used as the measure of central tendency. However, research has shown that component concentrations in atmospheric air are often characterized by a multimodal lognormal distribution (Izrael, 1984). Thus, the geometric mean was proposed for using in Russian national reports as a more appropriate central tendency. Further observations revealed that the obtained data frequently contain extremely low or high values that might significantly affect the geometric mean. Lately it was proposed to use an alternative averaging method - median, a more robust statistic. (Bruce, 2018; Rosenberg, 1994). Currently, national reports of the Integrated Background Monitoring Network have started to use and publish annual median values for presenting mean atmospheric air concentrations. This approach helps to reduce the impact of extreme values and, in some cases, enables the identification of more significant trends. Some data processing algorithms, such as using the calendar year instead of the geophysical year, have also been revised recently.
In this work, the data on component concentrations in air obtained from three Russian EANET sites (Irkutsk, Listvyanka, Primorskaya) were reviewed and recalculated using three different measures of central tendency: arithmetic mean, geometric mean, and median. While all three types of averages correlated strongly, the arithmetic mean differed from the geometric mean and median, generally trending toward higher values. The difference between the median and geometric mean was smaller. The median overall showed better performance as it is less influenced by outliers or values below the detection limit, which are not uncommon in atmospheric chemistry.
All calculated annual data were assessed for the presence of significant trends. The method used to calculate averages does not substantially affect the detection of significant trends. In most cases, if a statistically significant trend is identified, its presence is confirmed by any averaging method. However, trends calculated using the geometric mean are often less pronounced. All statistically significant trends identified at the EANET sites for the period 2004-2024 were decreasing, indicating a downward trend in air pollutant concentrations.
This study was carried out in the framework of the Research Project (reg. 125031703847-1) «Development of methods and technologies for monitoring of environmental pollution under the influence of transboundary pollutants transport and acid deposition (UNECE: EMEP, ICP IM; EANET». Evaluation of method and data consistency was conducted as part of the State Assignment of the Institute of Geography RAS Project No. FMWS-2024-0009 No.1023032700199-9.
Reference
Bruce P., Bruce A. Practical statistics for data scientists BHV-Petersburg; 2018.
Izrael YA, Rovinskii FY, Antonovskii MY, Buchstaber VM, Zelenyuk EA, Cherkhanov YP. On the statistical foundation of the atmosphere regional pollution components in background region. Doklady Akademii Nauk, 1984 (Vol. 276, No. 2, pp. 334-337). Russian Academy of Sciences.
Rosenberg GS, Shitikov VK, Brusilovsky PM. Ecological Forecasting (Functional Predictors of Time-Series). Togliatti: IEVB RAS. 1994.
In this work, the data on component concentrations in air obtained from three Russian EANET sites (Irkutsk, Listvyanka, Primorskaya) were reviewed and recalculated using three different measures of central tendency: arithmetic mean, geometric mean, and median. While all three types of averages correlated strongly, the arithmetic mean differed from the geometric mean and median, generally trending toward higher values. The difference between the median and geometric mean was smaller. The median overall showed better performance as it is less influenced by outliers or values below the detection limit, which are not uncommon in atmospheric chemistry.
All calculated annual data were assessed for the presence of significant trends. The method used to calculate averages does not substantially affect the detection of significant trends. In most cases, if a statistically significant trend is identified, its presence is confirmed by any averaging method. However, trends calculated using the geometric mean are often less pronounced. All statistically significant trends identified at the EANET sites for the period 2004-2024 were decreasing, indicating a downward trend in air pollutant concentrations.
This study was carried out in the framework of the Research Project (reg. 125031703847-1) «Development of methods and technologies for monitoring of environmental pollution under the influence of transboundary pollutants transport and acid deposition (UNECE: EMEP, ICP IM; EANET». Evaluation of method and data consistency was conducted as part of the State Assignment of the Institute of Geography RAS Project No. FMWS-2024-0009 No.1023032700199-9.
Reference
Bruce P., Bruce A. Practical statistics for data scientists BHV-Petersburg; 2018.
Izrael YA, Rovinskii FY, Antonovskii MY, Buchstaber VM, Zelenyuk EA, Cherkhanov YP. On the statistical foundation of the atmosphere regional pollution components in background region. Doklady Akademii Nauk, 1984 (Vol. 276, No. 2, pp. 334-337). Russian Academy of Sciences.
Rosenberg GS, Shitikov VK, Brusilovsky PM. Ecological Forecasting (Functional Predictors of Time-Series). Togliatti: IEVB RAS. 1994.