Study on methods of validity evaluation and time series decomposition of the online multiple air pollutants dataset

To meet the requirements of reliability and high timeliness for the quality control of online multiple monitoring dataset, the optimally statistical quality-control methods and related parameters are evaluated and screened by comparing and analyzing the Sliding Quadrant, Sliding Quadrant Gap and Sliding Variance methods. The results indicate that Sliding Quadrant Gap method with the specific sliding value (clean days 10 and pollution days 12) has an obvious effect on the detection of the abnormal data. The decomposition method of time series for online multiple air-pollutants monitoring-dataset was established in terms of the morlet wavelet analysis. Time-series components with different dominant factors influencing can be obtained by the method, providing the basis for accurately identifying the cause of air pollution. Based on the wavelet power spectrum analysis, the time series of atmospheric pollutants and the chemical species in PM_2.5 are decomposed into four bands. For most of the pollutants concentrations time-series, the contribution of the variance-sum of the decomposition bands to the total variance is over 90%, indicating that the decomposition-effect of atmospheric pollutants is better as a whole. In terms of the contribution of each band to the total variance of the original time series, the dominant factors that cause the variation of the time series of different pollutants are identified. There are obvious differences in the dominated factors that causing to the change of time series of different pollutant concentrations. For gaseous pollutants, the contributions of band 1 to band 4 to the total variance was 30.2%-49.5%, 12.5%-27.2%, 2.9%-11.1% and 15.5%-34.1% respectively. For particles, band 1 to band 4 contribute 21.0%-28.4%, 39.0%-48.7%, 21.3% and 6.6%-8.3% to the total variance. For the key chemical species in PM_2.5, band 1 to band 4 contributed 23.5%-49.6%, 17.7%-50.8%, 4.7%-23.7% and 7.8%-17.0% to the total variance, respectively.