Assessing the Impacts of Human Activities on Air Quality during the COVID-19 Pandemic through Case Analysis

: COVID-19 is the latest of many pandemic affecting the world in the past few decades, and it has had a signiﬁcant impact on the global environment. Some research has analysed the effects of the pandemic on air quality; however, very few studies have employed relationship analytics. In order to analyse the potential relationship between pandemic-related information and air quality data from a more holistic and detailed point of view, we propose a methodology based on pure data analysis. Three types of data were collected, namely air quality index, pandemic-related events, and number of COVID cases. Data were collected from ﬁve cities—Wuhan, New York, Seoul, Melbourne, and Singapore—to further analyse the response of air quality index to COVID events, thus revealing how human activity inﬂuences air quality from a pandemic perspective. The results show that a potential connection does exist in most cases and provide more evidence showing that air pollution declined during the pandemic. However, the strength of this relationship may also be related to other factors, such as geography, politics, population density, and measures imposed by local authorities, etc. This study provides another perspective to assist stakeholders in improving environmental decision making.


Introduction
Coronavirus disease (COVID-19) has become a worldwide pandemic since it first appeared in Wuhan, China in December 2019 [1]. As of 3 December 2021, 263,563,622 confirmed coronavirus cases had been reported worldwide, including 5,232,562 deaths [2]. The global outbreak and spread of COVID-19 have posed a severe threat to public health and caused significant interference in worldwide economic activities [3]. The COVID-19 pandemic forced most countries to take preventive measures to stop its disease spread, with local lockdowns being one of the main preventive measures employed [4]. Air quality is also thought to strongly interact with the mortality rate of COVID-19 [5]. A study showed that when PM2.5 (fine particulate matter defined as particles with diameters ≤2.5 microns) increased by one µg/m 3 , the COVID mortality rate increased by 8% in the United States [6]. Ironically, one of the few positive effects during a pandemic may be the health "benefits" of reduced air pollution [7], and there were significant reductions in air-pollutant levels in areas in which lockdowns were imposed [8].
Some studies have explored the interaction between COVID and air quality across the globe during the pandemic [9,10]. In China, studies have been conducted to evaluate changes in air quality by analysing the concentrations of major air pollutants, such as NO 2 , SO 2 , C.O., O 3 (ozone), and PM2.5, etc. before and during the pandemic [11,12] and to quantify the effects of reduced human activities on air quality [13]. In central China, where the novel coronavirus first broke out, atmospheric concentrations of PM2.5, PM10 Hypothesis 1 (H1). The reduction in human activity during the outbreak may positively improve air quality.

Hypothesis 2 (H2).
During the pandemic lockdown, air-quality improvement may be more significant.

Hypothesis 3 (H3).
During the outbreak, changes in new cases may affect human social activities, thus indirectly affecting shifts in air quality.

Hypothesis 4 (H4).
The improvement of air quality in cities that imposed lockdowns may be more significant than in cities without restrictions.

Hypothesis 5 (H5).
The degree of variability in air quality may vary due to cultural diversity and the implementation strength of prevention measures.
Hypothesis 6 (H6). During the epidemic, city-closure policies may have an impact on air quality.
The rest of this article is organised into several parts. In Section 2, relevant data and details of analysis are given; we also introduce related methods and features involved in experiments. In Section 3, we offer visual results, as well as some descriptions, we also summarise statistical data and discuss issues here. In Section 4, the conclusion of this work and a discussion of future research are given.

Materials and Methods
We downloaded and collected three types of raw data [43] for analysis, ranging from 1 January 2019 to 30 June 2021; data from 1 July to 31 December 2019 were excluded. The weeks used in this experiment refer to the sequential weeks of the year. For example, week1 represents 1-7 January.
(1) Air quality index data Air quality index data mainly refer to PM2.5 index data, including the daily average PM2.5 value from Wuhan, New York, Seoul, Melbourne, and Singapore. Original data were downloaded from aqicn.org [44], with a total of 1074 data entries finalised.
(2) COVID-19 new-case data The raw data covering five cities worldwide reflect daily new cases during the pandemic, involving 2020 and 2021. The data were downloaded and collected from several official government websites: Hubei Provincial Health Commission [45], NYC Open Data [46], Korea Centers for Disease Control and Prevention [47], Ministry of Health of Victoria [48], and the Ministry of Health of Singapore [49]. A total of 890 data entries were imported.
(3) COVID-19-related event data These raw data include the lockdown period of cities involved and significant protests. Details and sources can be found in [43]. Event data were included as supplementary evidence, in addition to COVID new-case data, in order to complete the analysis.

Data Processing
All collected raw data were cleaned and formatted. The processes are as follows.
(1) PM2.5 weekly value process In experiments, data were processed on a weekly basis. Specifically, collected raw data were transformed from daily to weekly format. For each week, the median value of PM2.5 during the week is selected as the present week's PM2.5 data. Please refer to Table 1 for details (data for the other four cities can be found in [50]). Since not all PM2.5 values are available, there are invalid data for some weeks. In this case, we assume that pm w is the present week's PM2.5 value yet has no valid value, pm w−1 is the PM2.5 value of the previous week, pm w+1 is the value of the next week. Then, the average value of week w − 1 Atmosphere 2022, 13, 181 4 of 18 and w + 1 is applied to fill in missing data. Weekly new cases is the sum of case numbers for each week, cn i is the value of new cases on the ith day in a week, an week w's case number is cn w (each week starts on Monday and ends on Sunday in this study).  0  0  week3  187  177  157  241  1  week4  191  137  176  1472  0  week5  165  160  156  6446  0  week6  165  122  151  12,119  0  week7  149  101  156  25,767  1  week8  142  128  128  2971  0  week9  126  111  83  2099  0  week10  153  128  132  438  0  week11  149  142  128  27  0  week12  134  139  134  1  1  week13  160  63  122  1  0  week14  133  119  128  1  0  week15  129  111  134  0  1  week16  122  116  146  0  2  week17  136  109  108  0  1  week18  123  132  104  0  0  week19  128  99  82.5  6  0  week20  142  105  61  0  0  week21  143  119  67.5  0  0  week22  100  112  74  0  0  week23  121  85  67.5  0  0  week24  125  85  61  0  1  week25  86  85  73  0  0  week26  127  85  110.5  0  2 (2) PM2.5 weekly changing-rate process The changing rate of the PM2.5 index was determined based on data from the same week in 2019, 2020, and 2021. For details, please refer to Table 2 (data for the other four cities can be found in [50]). pm wy is the value of the PM2.5 on week w of the year, y; and pm wy+1 is the value of the PM2.5 on week w of the year y + 1; and pmc w represents the corresponding rate of change of the PM2.5 index on week w between those two years. (3) PM2.5 yearly changing-rate process The average changing rate of PM2.5 was calculated between every two years of 2019, 2020 and 2021. pmc y represents the corresponding rate of change of the PM2.5 index between two years. For more information, please refer to Table 3. Table 4 gives details of the relevant changing rates of four cities during lockdowns (Seoul excluded). pmc y = ∑ n i=1 pmc wi /n (4) To discover the connection between COVID pandemic data and the PM2.5 index, we generated visual plots corresponding to the outbreak and PM2.5, as well as changes in new cases, based on the final data produced from the above processes [50]. Here, we use dates to bridge the related contact. Then, we integrated the three data types in Section 2. These include the changing trend of PM2.5 and new cases in Wuhan, New York, Seoul, Melbourne, and Singapore in 2019, 2020, and 2021. Ultimately, 260 records were retained for analysis.

Methods
This section introduces the relevant methods, including visualisation methods and study procedures.

Feature Selection
The raw data involved can be found in Section 2. PM2.5 refers to the particles in the ambient air with an aerodynamic equivalent diameter ≤ 2.5 microns, which is the main characteristic of smoke. PM2.5 can be suspended in the air for a long time, and the higher its concentration in the air, the more serious the air pollution. PM2.5 particles are also a dangerous form of pollution because they are so small that they can bypass the nose and throat, enter the lungs, and possibly enter the circulatory system [51]. Therefore, the PM2.5 index is suitable to measure air-quality performance. Wuhan, New York, Seoul, Melbourne, and Singapore are representative cities for which PM2.5 and COVID new-case data were available from public sources. Therefore, in this study, PM2.5 data of those five cities were processed to reflect the air quality of cities. Additionally, we collected data concerning newly confirmed cases of COVID-19 in the five representative cities during the pandemic, along with data concerning other events, to measure impacts on air quality.

Graph Visualisation Method
In this study, line chart, bar chart, and stack area charts were applied to present the experimental results. The broken-line/bar-chart component can display multiple series of data on the chart and can be used to show details of each city's PM2.5 data. The stacking area map, on the other hand, is capable of delivering the trend lines of the PM2.5 index and changing rate of each city over time and can highlight relevant intersections between PM2.5 and the numerical values on the series axes, which makes it appropriate to assist in analysis of the variances in air-quality changes in different cities. In addition, Tableau, a powerful tool widely used in the data-analytics field, is also used for visualisation, owing to its capability of visualising complex datasets [52,53].

Cross-Sectional and Time-Series Analysis
In this study, two types of evaluation were carried out, namely cross-sectional and time-series analysis. We compared behaviours from week 1 to 26 in 2019, 2020, and 2021, respectively, in order to look into the differences during the same period across three years. In addition, relevant data were analysed following the time series of 2020, which is not the worst pandemic year but the first year that COVID spread worldwide; 2020 also might be the year in which people panicked the most due to the fear of an unknown virus.

Procedure
The collected raw data were imported as input for processing, and five factors were taken into account: date, PM2.5 index, weekly changing rate of PM2.5, yearly changing rate of PM2.5, and weekly new cases of COVID. Then, the response results of the five cities, combined with relevant social events, were compared to examine the relationships among them, i.e., whether the epidemic impacted air quality, as well as other research questions mentioned in Section 1. The workflow is as follows.
(a) Collecting relevant raw data from multiple sources. (b) Data filtering and formatting to process all collected data, deal with invalid values, etc., and then format and import the data file for graph generation. (c) Comparing visual results via observation and statistics. (d) Discussion of findings according to the hypotheses.

Results and Discussion
In the experiments, five factors were compared across five cities based on data for first six months of 2019, 2020, and 2021 for cross-sectional analysis. Time-series analysis was performed by importing relevant social events from the five representative cities in 2020. (d) Discussion of findings according to the hypotheses.

Results and Discussion
In the experiments, five factors were compared across five cities based on data for first six months of 2019, 2020, and 2021 for cross-sectional analysis. Time-series analysis was performed by importing relevant social events from the five representative cities in 2020.    (d) Discussion of findings according to the hypotheses.

Results and Discussion
In the experiments, five factors were compared across five cities based on data for first six months of 2019, 2020, and 2021 for cross-sectional analysis. Time-series analysis was performed by importing relevant social events from the five representative cities in 2020.         As seen in Figure 1, Wuhan's PM2.5 levels in 2020 was significantly lower than in 2019; only weeks 12, 18, and 22 out of 26 weeks had higher PM2.5 values in 2020 than in 2019, with lower values than in 2019 for all other weeks. During the same period, compared with 2019 PM2.5 values, in 2020, the proportion of weeks with declining PM2.5 values was 88.5%, and that of increasing week was 11.5%. Additionally, compared with same period in 2019, in 2021, the proportion of weeks with declining and increasing PM2.5 values was 84.6% and 11.5%, respectively. Comparing 2020 and 2021 data, the proportion of weeks with declining and increasing PM2.5 values was 57.7% and 38.5%, respectively. Furthermore, the yearly PM2.5 value dropped by 16.03% between 2019 and 2020 and decreased by 18.02% between 2019 and 2021. As seen in Figure 1, Wuhan's PM2.5 levels in 2020 was significantly lower than in 2019; only weeks 12, 18, and 22 out of 26 weeks had higher PM2.5 values in 2020 than in 2019, with lower values than in 2019 for all other weeks. During the same period, compared with 2019 PM2.5 values, in 2020, the proportion of weeks with declining PM2.5 values was 88.5%, and that of increasing week was 11.5%. Additionally, compared with same period in 2019, in 2021, the proportion of weeks with declining and increasing PM2.5 values was 84.6% and 11.5%, respectively. Comparing 2020 and 2021 data, the proportion of weeks with declining and increasing PM2.5 values was 57.7% and 38.5%, respectively. Furthermore, the yearly PM2.5 value dropped by 16.03% between 2019 and 2020 and decreased by 18.02% between 2019 and 2021. Regarding the yearly PM2.5 value, it was found that the PM2.5 level in 2020 was significantly lower than that in 2019 during the same period, with reductions ranging from 8.48% to 13.85%, except for Melbourne (its data shows that PM2.5 levels in 2020 were higher than those in 2019 during the same period for most weeks). However, between 2019 and 2021, PM2.5 values declined only in Asian cities (Wuhan, Seoul, and Singapore; there was a slight drop in Singapore); and between 2020 and 2021, although PM2.5 yearly values rose in all five cities, the values for Wuhan and Seoul remained relatively at the same levels. Details can be referred to in Table 3. The PM2.5 indices of Wuhan, New York, Singapore, and Seoul all exhibited similar trend during the pandemic. When the outbreak was severe or the number of new cases increased, the PM2.5 index decreased, especially during periods of city closures. In addition, in most cases, after the pandemic situation had been improved or lockdowns were lifted, the PM2.5 index increased. Although Seoul did not implement a lockdown during studied period, increasing case number had a similar impact, sometimes (weeks 2-6, 8-13, and 16-22) even more significant than in Wuhan and Singapore.
Interestingly, the PM2.5 index in Melbourne seems either not to be affected by outbreak conditions or to be correlated with the outbreak negatively. When a pandemic is severe, PM2.5 values may rise; when the pandemic improves or before it gets worse,  Regarding the yearly PM2.5 value, it was found that the PM2.5 level in 2020 was significantly lower than that in 2019 during the same period, with reductions ranging from 8.48% to 13.85%, except for Melbourne (its data shows that PM2.5 levels in 2020 were higher than those in 2019 during the same period for most weeks). However, between 2019 and 2021, PM2.5 values declined only in Asian cities (Wuhan, Seoul, and Singapore; there was a slight drop in Singapore); and between 2020 and 2021, although PM2.5 yearly values rose in all five cities, the values for Wuhan and Seoul remained relatively at the same levels. Details can be referred to in Table 3. The PM2.5 indices of Wuhan, New York, Singapore, and Seoul all exhibited similar trend during the pandemic. When the outbreak was severe or the number of new cases increased, the PM2.5 index decreased, especially during periods of city closures. In addition, in most cases, after the pandemic situation had been improved or lockdowns were lifted, the PM2.5 index increased. Although Seoul did not implement a lockdown during studied period, increasing case number had a similar impact, sometimes (weeks 2-6, 8-13, and 16-22) even more significant than in Wuhan and Singapore.
Interestingly, the PM2.5 index in Melbourne seems either not to be affected by outbreak conditions or to be correlated with the outbreak negatively. When a pandemic is severe, PM2.5 values may rise; when the pandemic improves or before it gets worse, PM2.5 values may fall. For example, after Melbourne was locked down in 2020, the PM2.5 index rose to relatively high levels. PM2.5 levels even increased compared to the previous year during the same weeks and showed a downward trend before the outbreak began and before lockdown was imposed.
The weekly changing rate is employed to indicate the magnitude of change in PM2.5. For example, Figure 6 shows the weekly changing rate in Wuhan. The green line indicates the changing rate from 2019 to 2020. It can be seen that 23 out of 26 weeks reflect reduced PM2.5 levels in Wuhan, which means that PM2.5 levels in Wuhan were lower 88.5% of the time in the first six months of 2020 compared to the same period in 2019; in other cities, the proportion of weeks with improved air pollution was 73% in New York, 69.2% in Seoul, 46.2% in Melbourne, and 73% in Singapore. In general, the changing rates were predominantly negative in all cities, except for Melbourne, from 2019 to 2020.
year during the same weeks and showed a downward trend before the outbreak began and before lockdown was imposed.
The weekly changing rate is employed to indicate the magnitude of change in PM2.5. For example, Figure 6 shows the weekly changing rate in Wuhan. The green line indicates the changing rate from 2019 to 2020. It can be seen that 23 out of 26 weeks reflect reduced PM2.5 levels in Wuhan, which means that PM2.5 levels in Wuhan were lower 88.5% of the time in the first six months of 2020 compared to the same period in 2019; in other cities, the proportion of weeks with improved air pollution was 73% in New York, 69.2% in Seoul, 46.2% in Melbourne, and 73% in Singapore. In general, the changing rates were predominantly negative in all cities, except for Melbourne, from 2019 to 2020.  It can be observed that in Wuhan, when weekly cases reached their peak in week 7, the PM2.5 level dropped to the lowest relative level during this period (Wuhan's lockdown), which was 14% lower than the average and 42.9% lower than the maximum, i.e., the air quality was relatively better during the period of lockdown. In Figure 8, New York is seen to act similarly; the lowest PM2.5 value appeared in week 14, corresponding with the greatest number of new cases, which was 53.3% lower than the average and 73% lower than the peak. Similar results can be found in Melbourne, whereas Seoul and Singapore did not exhibit a clear pattern but maintained a relatively stable decrease PM2.5 values.

Figures 7 and 8 show weekly new COVID cases and weekly PM2.5 values in Wuhan
and New York for the first six months of 2020. It can be observed that in Wuhan, when weekly cases reached their peak in week 7, the PM2.5 level dropped to the lowest relative level during this period (Wuhan's lockdown), which was 14% lower than the average and 42.9% lower than the maximum, i.e., the air quality was relatively better during the period of lockdown. In Figure 8, New York is seen to act similarly; the lowest PM2.5 value appeared in week 14, corresponding with the greatest number of new cases, which was 53.3% lower than the average and 73% lower than the peak. Similar results can be found in Melbourne, whereas Seoul and Singapore did not exhibit a clear pattern but maintained a relatively stable decrease PM2.5 values.    It can be seen from Figures 9 and 10 that PM2.5 dropped in all cities from 2019 to 2020, except for Melbourne. Wuhan had the most significant drops (negative values). In Figure 9, the blue columns representing the change from 2020 to 2021 are all positive; from 2019 to 2021, Wuhan, Seoul, and Singapore are negative, and New York and Melbourne are positive. Figure 10 is a subset of Figure 9, and it shows only four cities during lockdown periods (there was no lockdown in Seoul). Comparing data from Figures 9 and 10, the yearly changing rates from 2019 to 2020 were −17.81% and −16.03% in Wuhan, −11.72% and −8.47% in New York, and −15.17% and −9.56% in Singapore. For details, please refer to Tables 3 and 4. Cities that implemented closures tended to have greater yearly PM2.5 changing rates than those without lockdowns, with the exception of Melbourne. It can be seen from Figures 9 and 10 that PM2.5 dropped in all cities from 2019 to 2020, except for Melbourne. Wuhan had the most significant drops (negative values). In Figure 9, the blue columns representing the change from 2020 to 2021 are all positive; from 2019 to 2021, Wuhan, Seoul, and Singapore are negative, and New York and Melbourne are positive. Figure 10 is a subset of Figure 9, and it shows only four cities during lockdown periods (there was no lockdown in Seoul). Comparing data from Figures 9 and 10, the yearly changing rates from 2019 to 2020 were −17.81% and −16.03% in Wuhan, −11.72% and −8.47% in New York, and −15.17% and −9.56% in Singapore. For details, please refer to Tables 3 and 4. Cities that implemented closures tended to have greater yearly PM2.5 changing rates than those without lockdowns, with the exception of Melbourne.    Regarding Melbourne, even when it was in lockdown, that city still showed an extreme opposite reaction in comparison to other cities; its air quality was worse during the closure period than during the open period. Its "individual" behaviour may be related to the mass demonstrations and protests that occurred in response to its six lockdowns, as well as government policies and, of course, repeatedly being affected by the pandemic [37]. However, relevant data have not yet been collected, and the specific influencing factors are not within the scope of the current study. We plan to introduce relevant major events and policies in the next phase of our research to reveal how air quality was affected by the outbreak in this particular city.
As shown in Figure 11, Wuhan was in lockdown or partial lockdown for the first six months of 2020, and its PM2.5 levels changed substantially; in the second half of the year, Regarding Melbourne, even when it was in lockdown, that city still showed an extreme opposite reaction in comparison to other cities; its air quality was worse during the closure period than during the open period. Its "individual" behaviour may be related to the mass demonstrations and protests that occurred in response to its six lockdowns, as well as government policies and, of course, repeatedly being affected by the pandemic [37]. However, relevant data have not yet been collected, and the specific influencing factors are not within the scope of the current study. We plan to introduce relevant major events and policies in the next phase of our research to reveal how air quality was affected by the outbreak in this particular city.
As shown in Figure 11, Wuhan was in lockdown or partial lockdown for the first six months of 2020, and its PM2.5 levels changed substantially; in the second half of the year, due to the improvement of the pandemic situation, the closure did not continue, and PM2.5 levels increased. Singapore imposed two closures in 2020. PM2.5 levels dropped slightly after lockdowns and remained stable, not changing significantly. During the first six months in 2020 in Seoul, PM2.5 was been stable, at around 80, with slight fluctuations. Melbourne and New York implemented secondary and additional lockdowns in 2020. Melbourne was closed for the longest time of all cities in the world during this period but did not experience any significant changes in PM2.5 levels.
In Figure 12, it can be observed that: In Wuhan, the number of new cases reached its peak 3 weeks after the lockdown and went back to being under control within 4 weeks. PM2.5 reached its new low 3 weeks after the closure and reached its minimum 9 weeks after the lockdown. In New York, the number of new cases reached its peak 2 weeks after the lockdown and went back to being under control within 4 weeks. PM2.5 reached its new low 2 weeks after the lockdown and has remained stable since. In Melbourne, there were not many cases during the studied period (the first 26 weeks in 2020), and PM2.5 levels increased following lockdowns. In Singapore, the number of new cases reached its peak 1 week after the lockdown and went back to being under control within 8 weeks. PM2.5 reached its new low 5 weeks after the lockdown. In Seoul, no lockdown was imposed, and there were not many cases during the studied period. PM2.5 in all cities tended to decrease during the first 26 weeks in 2020, except for Melbourne. Additionally, for the two cities with the most cases among the five representative cities, Wuhan and New York, they both took 2-3 weeks to reach the peak number of cases and 4 weeks to get the outbreak under control after lockdown; PM2.5 levels in these two cities reached relative low points within 2-3 weeks following lockdowns.
Atmosphere 2022, 13, 181 13 of 19 due to the improvement of the pandemic situation, the closure did not continue, and PM2.5 levels increased. Singapore imposed two closures in 2020. PM2.5 levels dropped slightly after lockdowns and remained stable, not changing significantly. During the first six months in 2020 in Seoul, PM2.5 was been stable, at around 80, with slight fluctuations. Melbourne and New York implemented secondary and additional lockdowns in 2020. Melbourne was closed for the longest time of all cities in the world during this period but did not experience any significant changes in PM2.5 levels. In Figure 12, it can be observed that: In Wuhan, the number of new cases reached its peak 3 weeks after the lockdown and went back to being under control within 4 weeks. PM2.5 reached its new low 3 weeks after the closure and reached its minimum 9 weeks after the lockdown. In New York, the number of new cases reached its peak 2 weeks after the lockdown and went back to being under control within 4 weeks. PM2.5 reached its new low 2 weeks after the lockdown and has remained stable since. In Melbourne, there were not many cases during the studied period (the first 26 weeks in 2020), and PM2.5 levels increased following lockdowns. In Singapore, the number of new cases reached its peak 1 week after the lockdown and went back to being under control within 8 weeks. PM2.5 reached its new low 5 weeks after the lockdown. In Seoul, no lockdown was imposed, and there were not many cases during the studied period. PM2.5 in all cities tended to decrease during the first 26 weeks in 2020, except for Melbourne. Additionally, for the two cities with the most cases among the five representative cities, Wuhan and New York, they both took 2-3 weeks to reach the peak number of cases and 4 weeks to get the outbreak under control after lockdown; PM2.5 levels in these two cities reached relative low points within 2-3 weeks following lockdowns. As seen in Figure 13, we compared Seoul with four other representative cities. In the top-left chart, it can be seen that the blue area with negative values is larger than the red area with negative values, which implies that PM2.5 dropped more noticeably in Seoul than in Wuhan during the studied period (weeks 1-26 in 2020). Furthermore, based on the three other charts, Seoul's PM2.5 levels decreased more evidently than those of Singapore but less intensely than those of New York. The reduction in PM2.5 levels in Seoul were similar to in Melbourne during this period. As seen in Figure 13, we compared Seoul with four other representative cities. In the top-left chart, it can be seen that the blue area with negative values is larger than the red area with negative values, which implies that PM2.5 dropped more noticeably in Seoul than in Wuhan during the studied period (weeks 1-26 in 2020). Furthermore, based on the three other charts, Seoul's PM2.5 levels decreased more evidently than those of Singapore but less intensely than those of New York. The reduction in PM2.5 levels in Seoul were similar to in Melbourne during this period.
As seen in Figure 13, we compared Seoul with four other representative cities. In the top-left chart, it can be seen that the blue area with negative values is larger than the red area with negative values, which implies that PM2.5 dropped more noticeably in Seoul than in Wuhan during the studied period (weeks 1-26 in 2020). Furthermore, based on the three other charts, Seoul's PM2.5 levels decreased more evidently than those of Singapore but less intensely than those of New York. The reduction in PM2.5 levels in Seoul were similar to in Melbourne during this period. During 2020, except for Seoul, all four other representative cities implemented city closures. The combination of graphs and statistics shows that PM2.5 levels remained in the range of ±15.7% of 127 from week 39 to 46 in Wuhan, and ±11.6% of 43 from week 32 to 41 in Singapore. This suggests that PM2.5 levels in Wuhan and Singapore were essentially stable during the second half of 2020. In contrast, PM2.5 levels in New York and Melbourne fluctuated during the same period. Air quality may be affected by the implementation of lockdown policies by local governments, though this effect has not yet been studied. During 2020, except for Seoul, all four other representative cities implemented city closures. The combination of graphs and statistics shows that PM2.5 levels remained in the range of ±15.7% of 127 from week 39 to 46 in Wuhan, and ±11.6% of 43 from week 32 to 41 in Singapore. This suggests that PM2.5 levels in Wuhan and Singapore were essentially stable during the second half of 2020. In contrast, PM2.5 levels in New York and Melbourne fluctuated during the same period. Air quality may be affected by the implementation of lockdown policies by local governments, though this effect has not yet been studied.
With regards to our hypotheses, based on the generated graphs, observation, and statistics, the following can be concluded: For H1, except for Melbourne, there were significant changes in air quality in all cities during the outbreak, following a similar trend. During the pandemic, various restrictions led to a reduction in human industrial activity and traffic movement, which may improve air quality [17]. Similar trends may enhance air quality, but in our experiments, this held true for four cities, excluding Melbourne excluded. Melbourne's PM2.5 changing rate was the opposite affected in many ways. Furthermore, this represents only a general outcome of the reduction in human activity, as actual impact varies.
For H2, based on analysis of the relevant values and changing rates, the magnitude of PM2.5 change in cities in which lockdowns were imposed was greater than in cities without closures. Based on our experiments, it can be concluded that during the outbreak, air quality improved to a greater extent (more than 10%) during lockdown in most cases; it can be noticed that lockdowns caused an improvement in air quality in all cities compared to 2019 (Seoul excluded), as seen in Table 4. The reason for this outcome might be that 2020 was the year that people panicked the most due to fear of an unknown virus. Hence, lockdowns are not likely the main contributing factor to improved air quality, which could be dependent on the strictness of enforcement instead. This finding does not apply to Melbourne, which may be affected by factors such as mass demonstrations [37], although such factors remain uncertain at this stage.
For H3, when weekly new cases reached their highest points in New York, Wuhan, and Melbourne in week 14, week 7, and week 13, PM2.5 values were 53.3%, 14%, and 36.7% lower than the average, and 73%, 42.9%, and 73.8% lower than the peak, respectively. Overall, PM2.5 levels fell to relative lows when the number of new cases highest was peaking; however, a similar trend was not observed in Seoul and Singapore. Therefore, it can be concluded that changes in the number of new cases could potentially affect social activities, thus indirectly inducing changes in air quality; however, the significance of this correlation remains unknown.
For H4, although no lockdowns were implemented in Seoul, the city exhibited similar patterns in air-quality change to other cities, with significant air-quality improvements compared to 2019 (PM2.5 dropped 13.85% from 2019 to 2020). Therefore, it cannot be confirmed that air quality improves more in cities with lockdowns than those without. Many factors may be involved, and there is no solid evidence showing that cities with lockdowns have less human activity than cities without closures. Therefore, implementation strength of prevention measures may play a role, although we have not yet been able to take such factors into account, so this assumption has not yet been confirmed.
For H5, as did not discover consistent patterns among Asian cities and Western cities, it cannot be inferred whether cultural diversity and the implementation strength of prevention measures impact air pollution. Thus, this hypothesis cannot be confirmed. However, three Asian cities, particularly Wuhan and Seoul, did show greater improvements PM2.5 when comparing data between 2019 and 2020, as well as 2019 and 2021 (see Table 3), and Singapore was one of few counties that maintained opened borders in 2020 [39].
For H6, the implementation of lockdowns in Wuhan and Singapore was relatively strong, and their PM2.5 levels exhibited a steady downward trend in 2020. City closures in New York and Melbourne were relatively weak due to multiple protests. There were 17 COVID-related demonstrations in New York and 18 in Melbourne from 2020 to 2021, all for anti-lockdown and anti-mandate purposes [33,34,38]. There was also one protest over the student COVID-19 vaccine pass mandate in Seoul in 2021 [54]; however, only 70 people were involved. There were no COVID related demonstrations in Singapore and Wuhan. PM2.5 levels fluctuated in all five cities during the studied period. Among them, Melbourne was closed for the longest time [55], although PM2.5 levels for that city sometimes showed an upward trend. However, considering its many protests, Melbourne's closures might not have been as strict as in other cities. It is a well-known fact that lockdown mandates were not strictly implemented in many cities around the world [41,42]. Therefore, it can be concluded that the implementation strength of city-closure policies has an impact on air quality in most cases.
Overall, with Melbourne excluded, H4, H5, and H3 (the peak of new cases affected human activity and thus indirectly influenced air quality in New York, Wuhan, Singapore, and Seoul) cannot be proven. Other hypotheses have been confirmed in this work. The outlying behaviour of air-quality changes in Melbourne during the outbreak may have been caused by multiple factors, such as geographical, political, population density, and economic reasons, that lie beyond the scope of the current study.

Conclusions
Frequent air-quality fluctuations were reflected in changing PM2.5 values in 2020. The fear of unknown viruses may have led to a significant reduction in human activity in the early stages of the pandemic, resulting in a decrease in air pollution and, ultimately, improved air quality [13]. This study generated graphs and statistics to analyse the potential connections between COVID pandemic data and air quality. PM2.5 fluctuations were found to be more evident before and after lockdowns. PM2.5 levels were greatly affected by the outbreak in Wuhan, Seoul, and Singapore; its impact in New York was mediocre; and it was tangential in Melbourne, showing an opposite effect as compared to the four other representative cities. Our hypotheses have been partially verified, and an outbreak is likely to reduce human activity (though not in all cases) and therefore affect air quality. According to the experimental results, in most cases, worse epidemic situations, such as a greater number of infections, leads to less human activity (owing to lockdowns, etc.), resulting in air-quality improvement. In contrast, improved epidemic situations cause city closures to lifted, resulting in more human activity, leading to air-quality deterioration to some extent. However, due to many other factors, such as government prevention measurements and their implementation strength, a city's geographical location, cultural diversity, etc. that have not been investigated at this stage, the detailed indirect impact of the outbreak on air quality is uncertain, and actual reduction in air pollution varies. To the best of our knowledge, a study involving multiple cities and periods has not been carried out before. This work may be helpful for the analysis of air quality data from another perspective-the relationship between air pollution and pandemic data. Additionally, this research could potentially assist stakeholders in understanding interactions among various factors (major social events and outbreak data, etc.), offering more comprehensive evidence from which to deduce the effects of human activities on air quality. This work is the first step in analysing the relationship between the outbreak data and air quality. The limitations of this work include several aspects. Regarding the complexity of air quality data, PM2.5 was the only imported for analysis in this study, which may be an essential element, although additional factors could also contribute to the study of air quality during the pandemic. Additionally, we only collected data from five cities in Asia, North America, and Oceania, and our study lacked data from other continents (especially Europe, for which relevant open datasets are available). This study also lacks air quality data from other months/years. Political, economic, and population density factors are also important, as they relate to research on air pollution during the epidemic and have not been taken into account. Furthermore, from a visualisation perspective, this study could import spatial and temporal settings to present a better dashboard for further analytics.
We will address the above limitations in our future work, collect more air pollution factors, and import those factors into our research. We will also introduce a novel interactive visualisation method to show the relationship between COVID-19 data and air quality index data, diving deeper into the relationship and eventually providing a more comprehensive understanding of this topic based on pure data analytics.