Chen Songqi's team from the School of Mathematical Sciences proposed an air quality assessment method based on pollution processes

Air pollution is an important environmental and public health problem, especially in developing countries
.
Measuring the level of pollution emission and its impact on air quality is affected by the comprehensive influence of meteorological conditions, regional transmission of pollutants and other factors, which is an urgent and challenging task
in air quality management.
The team of Professor Chen Songqi from the School of Mathematical Sciences and the Center for Statistical Science of Peking University used big data statistical algorithms to screen and model pollution processes, and proposed an air quality assessment method
based on pollution processes based on the results of the previous "static period data experiment" to effectively measure local emissions.
Recently, the research paper "Episode Based Air Quality Assessment" was published
in Atmospheric Environment, an important international journal in the field of atmospheric environment.

This paper takes the three cities of Beijing, Tangshan and Baoding located in the north of the North China Plain as the research objects, and is divided into four groups according to geographical location: southeast of Beijing, northwest of Beijing, Tangshan and Baoding
.
For the air quality data and meteorological data of the seven seasons from March 2013 to February 2020, the pollution events were selected by data-driven algorithms, and the average / of the six major air pollutants (PM_2.
5,_NO2, CO, SO₂,_O2 and PM₁₀) in the pollution process were studied by linear regression and random forest method.
The total pollution load, which is closely related
to the variables of meteorological conditions before and during the pollution process.
Compared with the traditional whole-sample method, the method defines and analyzes air quality more clearly, and evaluates air quality
from different perspectives.

Based on the geographical characteristics and meteorological reality of the North China Plain, the "pollution process" starts from the low PM 2.
5 level after the strong northerly wind system completely cleans the air, followed by the static period formed by the accumulation of local emissions, followed by regional transportation and air stagnation to bring high pollution or even serious pollution levels, and finally the cleaning process of the north wind reduces the PM_2.
5 concentration to a low value again
。 Specifically, it is determined by three key time points: (i) the start time of the pollution process t_s, (ii) the peak time t_p under continuous pollution accumulation, and (iii) the end time t_e
of the pollution process.
Based on the difference in pollution and meteorological levels in different cities, two situations
of strong cleaning and weak cleaning are defined.

Fig.
1 PM_2.
5 time series data of two types of pollution processes of strong cleaning (left) and weak cleaning (right), selected from the monitoring point of Beijing Dongsi in the autumn of 2018 and the monitoring point of Baoding Huadian Area 2 in the winter of 2014

To eliminate meteorological disturbances, the estimated average/total pollution load is seasonally adjusted
according to the 2013-2018 meteorological baseline distribution.
Based on the performance and explanatory nature of linear regression and random forests, only linear regression
is considered in the air quality assessment based on pollution processes.
For the four site clusters, the pollution process is adjusted before and after the adjustment of PM 2.
5 in autumn and winter and ozone in spring and summer, as shown
in Figure_2.

Fig.
2 Comparison of the average pollution load of PM2.
5 (left) and ozone (right) in autumn and winter (left) and ozone (right) before and after meteorological adjustment during the pollution process of the four station groups, and the boxplot depicts the distribution of the duration of the pollution process

Although the proportion of pollution processes varies from year to year, in terms of overall trends and annual changes, the results of air quality assessment based on pollution process and air quality assessment based on complete observational data (including observation data outside the pollution process) are basically the same, and there is a strong synergy in the downward trend and change magnitude
.
Taking PM_2.
5 as an example, the upper and lower figures in Figure 3 show the relative decrease
in the average pollution load in different seasons from 2014 to 2020 compared with 2013 using the assessment method proposed in this paper and using complete observations.

The results of the six pollutants showed that from 2013 to 2020, four pollutants_{, PM 2 5}.
The average pollution load of CO, SO2 and PM₁₀ showed a significant downward trend, of which SO2 decreased most significantly and continuously.

Among the 16 season-site group combinations, the average pollution loads of SO 2 and PM 2.
5 in 2020 were significantly lower than in 2013 (significance of 5%), and the decline rates ranged from 37.
9% to 91.
0% and 30.
2% to 81.
1%,
respectively_.
The average decline of NO₂ was small, with only 12 combinations showing significant declines in 2020, of which the average decline rate was only 12.
7%-61.
3%.

In sharp contrast, the seasonal average of ozone concentrations has increased significantly since 2014 compared to 2013, with growth rates of 24.
7%-63.
5% and 26.
8%-283.
8%
in spring, summer and autumn-winter 2020, respectively.
Worryingly, despite the significant reduction in_PM-SO2-CO in the three cities, the average_O3 pollution load of all seasons of pollution processes generally increased with no signs
of declining.

Fig.
3 Relative decrease in average pollution load in different seasons from 2014 to 2020 compared to 2013, based on pollution process data (top) and complete observational data (bottom).

The analysis shows that air quality assessments based on pollution processes are more sensitive
in reflecting potential changes in concentration than observations using the whole data.
This is best reflected in ozone assessments, with relative variations much larger than the use of
complete data.

Although the three cities have similar pollution-cleaning patterns due to their geographical location, through the big data statistical method, other cities can be automatically found for pollution processes and data sets can be established, so as to obtain air quality assessment results
based on pollution processes.
Although important variables may vary from city to city, the use of weather adjustments, averages, and total pollution loads can be easily scaled up.

The first author of the paper is Luo Shanshan (2020 master student at Peking University Big Data Research Center), and the other author is Zhu Yuru (2018 doctoral student, Peking University Center for Statistical Science).

The corresponding author of the article is Professor
Chen Songqi.
This research was supported
by the National Natural Science Foundation of China 92046021, 12026607 and 12071013.

For details, see Luo, S.
, Zhu, Y.
, & Chen, S.
X.
(2022).
Episode based air quality assessment.
Atmospheric Environment, 285, 119242.