3.1. Different Mechanisms in Different Periods
The change in NO2 concentration was the most prominent. The average NO2 concentrations for the pre-lockdown and lockdown periods in 2020 were 42.04 and 19.75 μg/m3, compared with 55.99 and 42.84 μg/m3 for the same lunar calendar periods in 2019 and 48.87 and 35.22 μg/m3 in 2021, showing that the air pollution significantly decreased in 2020 and rebounded slightly in 2021. Similar to the abovementioned trend, for the pre-lockdown period, average PM2.5 concentrations were 96.68, 60.60, 65.98 μg/m3 and average SO2 concentrations were 9.92, 6.83, 8.29 μg/m3 in 2019, 2020 and 2021. For the lockdown period, average PM2.5 concentrations were 55.70, 38.61, 39.06 μg/m3 and average SO2 concentrations were 8.31, 7.57, 7.75 μg/m3 in 2019, 2020 and 2021. In contrast with the NO2 concentrations, the average O3 concentrations increased in 2020 and then dropped in 2021, with 41.26, 59.67, 48.28 μg/m3 for the lockdown period in each year. The average CO concentrations showed a slight decreasing trend over three years, with 0.95, 0.91, 0.78 mg/m3 for the lockdown period in each year.
As is shown in
Figure 3, it is obvious that the NO
2 and PM
2.5 concentration dropped sharply after the lockdown began in Wuhan (23 January 2020), while the O
3 concentration obviously increased. There was an inverse relationship between the amount of NO
2 and ozone [
41]. During the first week after the Spring Festival, the CO, NO
2, PM
2.5, and SO
2 concentrations significantly decreased. The difference between the NO
2 concentration in 2020 and in 2019 during the week after 30 January reached a relatively small value as the enterprises offering protective products and emergency supplies resumed work and production. In the pre-pandemic period in 2019, the NO
2 concentration rebounded slowly after the Spring Festival. However, in 2020, it kept a low level even after the Spring Festival, showing a close correspondence with the COVID-19 lockdown. Meanwhile, the concentrations of the two pollutants in 2021(the post-pandemic period) were still lower than that in 2019. The change in NO
2 concentration was most prominent, dropping by 24.9% in the pre-lockdown period, 53.9% in the lockdown period and 15.4% in the post-pandemic period in 2021, compared to the pre-pandemic period.
The meteorological conditions for different periods over the three years are listed in
Table 3, and the changes in human mobility over the same periods are illustrated in
Figure 4. The meteorological conditions changed a little. During the lockdown period, the average pressure (pressure) and wind direction (wd) in 2020 were higher than those in 2019 and 2021 while the average wind speed (ws) was lower than those in 2019 and 2021.
There are obvious trends in the variables about human mobility. The average WMI, IMI, OMI during the lockdown period in 2020 were 0.66, 0.37, and 0.37, respectively, much lower than those in 2019 (3.89, 5.43, and 3.93) and in 2021 (5.43, 4.26, and 3.48). The Spring Festival holiday usually results in the decrease in PM
2.5 and NO
2 and an increase in ozone in big megacities [
42] because many people return to their hometowns [
43,
44]. The changes in air pollutants are the comprehensive results of the holiday effect and the pandemic control policies [
45]. At the beginning of the Spring Festival migration in 2020 (10 January), the values of the WMI, IMI and OMI were higher than those in 2019. As can be seen from
Figure 4c, a large number of people left Wuhan a few days before the lockdown, which was more than the number in 2019. Even on 23 January 2020 (on which day 10:00 a.m. was Wuhan’s lockdown time), the number of people leaving Wuhan on that day was higher than that on the same day of the lunar calendar in 2019. When Wuhan undertook the COVID-19 lockdown policy, the flow of persons was reduced to zero and the IMI decreased. Different from that, indexes rebounded after the first week of the holiday of the Spring Festival in 2019 and 2021, the levels of human mobility in 2020 remained at a low level because of the COVID-19 lockdown.
In addition, the average WMI during the post-pandemic period in 2021 was 5.23, higher than 3.85 during the pre-pandemic period in 2019, while the average IMI and OMI were 3.57 and 3.44, both less than 5.03 and 4.50 in 2019. We can notice that in 2021, when the epidemic prevention and control was regular, the travel intensity of residents in the city increased, while the activities of moving into the city and moving out of the city decreased compared with pre-pandemic conditions, with most of residents traveling in the city. Notably, there was a high value of WMI on 3 April 2021, corresponding to the first day of the Tomb Sweeping Day holiday.
Different mechanisms are illustrated over four periods. The stepwise regression model of the pre-pandemic period included four explanatory variables, which explained 82.8% of the variance of the NO2 concentration. The stepwise regression model of the pre-lockdown, the lockdown and the post-pandemic period included one, one, and four explanatory variables, respectively, and each explained 54.0%, 47.6% and 63.9% of the variance of the NO2 concentrations.
The regression coefficients of the models are shown in
Table 4. The explanatory variables significantly affect the NO
2 concentrations (
p < 0.01). It is obvious that emissions from industry and households (represented by SO
2) are closely connected with the NO
2 concentration and have a positive impact on it whichever period. The wind speed (ws) had a lowering effect on the NO
2 concentration in the pre-pandemic and post-pandemic period, showing the impact of meteorological elements. The models in the pre-pandemic and post-pandemic period were more similar, while the Within-City Migration Index played a more important role in the post-pandemic period. The NO
2 concentrations in the pre-lockdown and lockdown period fluctuated largely and the Within-City Migration Index and wind speed were meaningless in the two periods.
3.2. Relative Contributions of Meteorological Conditions and Human Activities
We used a Random Forest model to fit the NO
2 concentrations during the pre-pandemic period, along with a Support Vector Machine model, a Linear model and a Stepwise Multiple Linear model to evaluate the performance. The four models all had high accuracy, the smallest cross-validation R-squared (CV-R
2) of which was 0.771. Other goodness indicators are shown in
Table 5. Overall, all the simulation results were acceptable, and the simulated concentrations agreed well with the observed data, with the correlation coefficient (COR) above 0.9 for each model.
We employed Random Forest models to operate sensitivity experiments and then calculated the relative contributions of meteorological conditions and human activities for different periods. The results are shown in
Table 6 and
Figure 5. The contributions of meteorological conditions and human activities varied from day to day, and we calculated the average NO
2 concentration and the average contributions during the whole period. In the pre-lockdown period, the NO
2 concentration simulated with human mobility in 2020 was more similar to the NO
2 concentration in the pre-pandemic period than that simulated with meteorological conditions and emissions from industry and households in 2020, which means that road traffic led to the least contribution to the change of air pollutants and the changed emissions from industry and households contributed the most. The mean values of the daily normalized contribution were 35.2%, 13.8% and 51.0% for
,
and
. During the first week of the pre-lockdown period, changes in meteorological conditions made the greatest contributions to the daily reductions in the concentrations. During the last two weeks of this period, the reduction in emissions from industry and households played the most important role in the reduction in the concentrations.
During the lockdown period, the observed NO2 concentrations were lower than those in the pre-pandemic period in 2019 except for the days between 28 January and 5 February (when the variable lunar is between 4 and 12). The average NO2 concentration simulated with meteorological conditions in 2020 was higher than the observed NO2 concentrations in the same period in 2019, indicating that the meteorological conditions during the lockdown period essentially were unfavorable to the reduction in the air pollutants. The average NO2 concentration simulated with human mobility in 2020 was the lowest, which means that road traffic dominated the reduction in the NO2 concentration during the lockdown period. In addition, the average normalized contributions of , and were, respectively, 10.0%, 73.3% and 16.7%, which emphasized the role of human mobility again. The changes in road traffic dominated changes of the concentration during the third, fourth, and last week of the lockdown period.
During the post-pandemic period, the average observed NO
2 concentration was still less than that during the same period in 2019, mostly due to emissions from industry and households. The NO
2 concentration simulated with meteorological conditions in 2021 was higher than that in 2019, which shows that the general meteorological elements in the post-pandemic period in 2021 are favorable to the increase in the NO
2 concentration. Moreover, the level of road traffic recovered to the pre-pandemic level, with the NO
2 concentration simulated with human mobility in 2021 being similar to that in 2019. According to the normalization process, the meteorological conditions controlled about 42.2% of the decrease, and the reduced emissions from industry and households controlled 40.0% of the decrease, while the level of human mobility only contributes to 17.8% of the decrease. As was shown in
Figure 5f, in the post-pandemic period, the NO
2 concentrations do not completely show a downward trend compared with the same period in 2019, and the contribution rate of variables varies greatly from day to day. When the variable lunar was −6 (5 February 2021), meteorological conditions led to an increase of 490.4% of pollutants, while human mobility led to a decrease of 569.4% of concentration. When the variable lunar was 32, 33 and 47 (15, 16, and 30 March 2021), human mobility led to an increase of 799.5%, 117.2% and 126.0% of concentration, while meteorological conditions were conducive to the reduction in pollutants.
3.3. Simulations of Different Scenarios for Road Traffic Control
The emulator demonstrates the link between NO
2 concentrations and road traffic by predicting the concentrations with different reductions in human mobility compared to the level in 2019 in three scenarios. As shown in
Figure 6, all the changes tend to approach a stable level when the reduction in human mobility continues. According to the variation rate of the NO
2 concentrations, for the three scenarios, all the change rates are very large at the beginning, but with the reduction in human mobility, the variation rate becomes gradually flat. This means the control of human mobility has a stronger effect at the beginning, with a gentler change trend later. Generally, in
Figure 6c, when human mobility within the city is controlled, the decreasing trend is almost constant whatever the basic NO
2 concentration. In
Figure 6a, bscenario (a) and (b), only when the initial basic NO
2 concentration is relatively high, the NO
2 concentration will remain decreased with the level of human mobility within the city reduced.
When the reduction in all kinds of human mobility is beyond 70%, the effect of the control policy reaches a threshold level. This means that when the proportion of the overall human mobility is less than 30% of the pre-pandemic level, the pollutant concentration will remain stable. When we control the human mobility out of and into the city and leave the mobility within the city unchanged, the threshold level is 60%, and when we only control the mobility within the city, the threshold level is 70%, which indicates that the control policy in
Figure 6b is less sustainable. Moreover, the variation rate is less than 10 when the reduction in human mobility is more than 40% both in
Figure 6a,c, while the variation rate in
Figure 6b is less than 10 when the reduction in human mobility is more than 20%. We can conclude that controlling the human mobility within the city is more effective than controlling the mobility out of and into the city, and the effect of the former policy almost equals that of taking an overall control of all kinds of human mobility.