The Effects of Pandemic Restrictions on Public Health—Improvements in Urban Air Quality

The present study aims to provide evidence on the effects of pandemic curtailment measures on public health, targeting the changes in breathable air quality, within urban areas. The analyzed period covers the full impact of lockdowns in Europe in 2020. We used everyday data for each analyzed pollutant, NO2, SO2, CO, PM2.5 and PM10, from urban monitoring stations that provided real-time concentrations (provided by Copernicus Atmosphere Monitoring Service, Environmental Protection Agency repository and European Environment Agency map services) and satellite data (provided by NASA Orbiting Carbon Observatory 2). In the present study, the urban air quality was computed using a composite index that was further analyzed in comparison with pandemic restrictions. Descriptive statistics, charts and maps were used to visualize the data that covered the analyzed countries. Our results show that air pollution was reduced by 12% after lockdowns in European urban areas, with a 0.76 correlation between air pollution and pandemic restrictions. All air pollutants registered significant drops.


Introduction
As an adult inhales approximative 11,000 L/day of air [1], it is highly important that the quality of the air is reasonably high and consists of a normal state of the usual components of terrestrial atmosphere (nitrogen, oxygen and a tiny percent of argon, carbon dioxide, neon, helium and hydrogen) [2]. Anthropogenic or natural events can adjust these concentrations or add new components to the air structure, such as carbon dioxide (CO 2 ), carbon monoxide (CO), nitrogen dioxide (NO 2 ), particle matter (PM), volatile organic compounds (VOCs), ozone (O 3 ), sulfur dioxide (SO 2 ), methane (CH 4 ) or other gases that can have a negative impact on human or environmental health [3,4].
The COVID-19 lockdown period imposed severe restrictions regarding industrial activity and human mobility; thus, it is the best opportunity to test if the reduction of these types of activities had a significant impact over the quality of air, thus improving the public health in urban areas.
The severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) first emerged in China in December 2019, but has spread rapidly since, quickly becoming a global problem, being declared a pandemic by the World Health Organization (WHO) on 11 March 2020. The governments of the affected countries have imposed many restrictions to human interactions in order to stop the virus transmission. Symptoms are extremely heterogenous, although the virus primarily affects the respiratory system [5]. COVID-19 is primarily transmitted through physical contact or oral transmission [6,7]. Other studies reported surface contamination up to 72 h on plastic materials [8]. A study showed that protection is increased when physical distancing exceeds 1 m and eye and face protection are worn, with a significant difference for N95 or similar respiratory masks in comparison with normal The present study provides data on 28 European countries over a period of 10 months. The countries were selected from the European continent, taking into consideration the data availability level for each state, thus selecting the ones which had over 70% data availability. Data on air pollutants were extracted from urban monitoring stations implemented in Europe, measured as real-time concentrations from the Environmental Protection Agency repository (link available in the data availability statement). Also, satellite data were obtained by accessing the NASA Orbiting Carbon Observatory 2 product, through the Copernicus Atmosphere Monitoring Service. Furthermore, meteorologic conditions such as wind, temperature or precipitation were obtained from the ERA5 satellite through the Copernicus Atmosphere Monitoring Service (link also available in the data availability statement).
Restrictions in the COVID-19 pandemic were obtained from the European Centre for Disease Prevention and Control repository, for each day starting from 15 of March 2020, while data on the impact of NACE rev. 2 activities on air quality and the impact of the main activities of manufacturing and industry over the national value added were obtained from Eurostat. Air quality missing data were linearly extrapolated using previous values. Data completeness statistics were concluded at 96% (4% of the data was estimated). Table 1 presents the sources for the data integrated in the present analysis.

Air Pollution Index (API) and Air Quality Index (AQI)
All data points regarding pollutants included in the air quality index were obtained in the US EPA standard, measured as the daily median for each country considered. Air quality data were normalized by applying the US EPA air quality index methodology, in order to homogenize all the analyzed pollutants and compute the air quality index for each country and for the entire area. Thus, for obtaining the air pollution score for each pollutant, Equation (1) was applied: where I P is the air pollution score for each pollutant, C P is the period average truncated concentration for each pollutant (as PM 2.5 and CO have to be truncated to 1 decimal place, and PM 10 , SO 2 and NO 2 have to be truncated to integer), BP Hi is the concentration breakpoint first to exceed the C P , BP Lo is the concentration breakpoint first below the C P , and I Hi and I Lo are the AQI values corresponding to BP Hi and BP Lo , respectively. The breakpoints for each pollutant and the corresponding AQI values are available in the US EPA air quality index methodology (link available in the data availability statement). After calculating the I P , the score was normalized using linear normalization. An air pollution index for each country and for each period was created using the average of the air pollution score; the index used PM, SO 2 , NO 2 and CO data. Equal weights were attributed in order to compute the air pollution index.

of 23
The air pollution index (API) was composed as follows: API (X) = avg [I p ]; (2) where X represents each country taken in consideration and I P is the air pollution score for each pollutant considered. An air quality index was created using the reciprocal of the air pollution index. The air quality index (AQI) was composed as follows: The meteorology data include total precipitation (m), near surface wind speed (m s −1 ), near surface air temperature ( • K) and boundary layer height (m), which were obtained from the Copernicus Climate database using ERA5 analysis of meteorological data.

Restriction Index (RES)
The pandemic restriction data refer to the collection of data representing each restriction type that occurred each day in a specific country and that was aggregated using specific weights. In order to scale the weights with actual emissions data, we used the air emissions accounts by NACE Rev. 2 activity from the Eurostat database [14]. For example, the highest sulfur oxides (SO 2 equivalent) emission is attributed, in this exact order, to transportation and storage, electricity, gas, steam and air conditioning supply and manufacturing. Transportation, storage, public administration, defense, and compulsory social security present the highest emissions of PM 2.5 , while agriculture, forestry, fishing, transportation, storage, manufacturing and construction are related to the highest emissions of PM 10. The definitions for each activity contained in NACE Rev. 2 are available at the link presented in the data availability statement. Therefore, transportation contains activities related to air, water and land transportation, while storage reflects all the activities related to warehousing and support activities for transportation and postal and courier activities. As the repository does not allow us to break down activity classes into more specific activities and transportation and storage are the main activity sources for SO 2 and PM 2.5 , as well as in the top five sources for PM 10 , we assume that air, water and land transportation are the main sources for these pollutants, while we do not have any evidence that a specific restriction affected the storage activity, other than the transportation process related to the storage activity. Thus, all restrictions related to transportation must have the maximum weight, although, in the EU, "road transport has the largest share of modes of transport" [15]. Similarly, regarding the main sources for PM 2.5 , the "public administration and defence; compulsory social security" class contains activities related to general public administration activities, foreign affairs, defense activities, justice and public order and compulsory social security activities. As the repository does not allow us to break down activity classes into more specific activities, we assume that all the restrictions related to the general public administration activities must have the maximum weight, while any restrictions related to defense or compulsory social security activities have not occurred. The class "electricity, gas, steam and air conditioning supply" covers activities related to electric power and gas generation, manufacture and distribution, and the production, collection and distribution of steam and air conditioning supply. Because the repository does not allow us to break down the activity class, we must assume that activities related to electric power and gas generation, manufacture and distribution, respectively, are the most pollutant activities, as there were not any restrictions to be found in relation to steam and air conditioning supply activities, and thus they do not affect the present research.
The maximum weight was attributed to: closing high schools, closing primary schools, closing sectors, stay home orders, teleworking, closing workplaces and closing daycares. Closing public activities, entertainment venues, hotels and other accommodations, restaurants and cafes, public transportation and restricting outdoor activities over 1000 were aggregated with a weight of 2, while closing gyms, sport centers and non-essential shops were aggregated with a minimum weight. In the aggregation process we used weights from 1 to 3.
A restriction index was created using summarization of the elevated restrictions in the selected periods, considering that the aggregation process of the mentioned restrictions was implemented for each day from the analyzed period.

Descriptive Statistics and Hypothesis Testing
As shown in Table 2, data were computed into bimonthly averages, considering that the period of incubation for COVID-19 symptoms could extend to 14 days [14], with a median of 5.2 days [5], and that governments usually took into consideration a two-week lockdown period. The ARX approach relates to the bimonthly periods. Moreover, visual maps were generated for each air pollutant, in three main periods:  Descriptive statistics refers to the central tendency indicators, coefficient of variation and correlation among the variables and also the relative modifications of each pollutant concentration, analyzed before and after the COVID-19 pandemic lockdown in Europe. Air pollution and restrictions were indexed for each analyzed period.
The research design is based on secondary data analysis of a time series.
To pursue our alternative hypothesis, that pandemic restrictions significantly affected air quality in urban areas, we ran a t-test with paired groups. For further forecasting, we used an ARX (autoregressive model with exogenous variable), enhancing a lag of 4 periods. We have also extended the regression to foster the best scenario. Therefore, in order to summarize the curtailment restriction implications on public health, we have introduced a representative diagram as shown in Figure 1.
fected air quality in urban areas, we ran a t-test with paired groups. For further forecasting, we used an ARX (autoregressive model with exogenous variable), enhancing a lag of 4 periods. We have also extended the regression to foster the best scenario. Therefore, in order to summarize the curtailment restriction implications on public health, we have introduced a representative diagram as shown in Figure 1.

Urban Air Pollution and Restriction Index Values
where APIUE is the air pollution index for the entirety of Europe and RESUE is the restriction index for Europe.  Figure 2 shows the air pollution index (API) and the restriction index (RES) for each analyzed period, for the entirety of Europe, composed as follows:

Urban Air Pollution and Restriction Index Values
where API UE is the air pollution index for the entirety of Europe and RES UE is the restriction index for Europe.  The European air pollution index is measured on a numerical scale from 0.19 to 0.25 units, where 0.25 is the maximum value and 0.19 is the minimum value, and the restriction index is measured in points, from 0 to 25 points, summarizing the restrictions for each period analyzed.
In Figure 2, the air pollution index, which is an aggregate index of NO2, CO, SO2, PM10 and PM2.5, had a decreasing pattern in urban areas as a result of energy reduction, traffic-related combustion reduction and industry breaks. Since energy dropped by 25% every week when full lockdown was active, and with an average decrease of 3.8% from spring 2019 to spring 2020 on the basis of the drop in traffic-related activities and partial or full industries breaks, air pollution settled on a decreasing trend, consisting of a total decrease of 12% in the analyzed period, which is the most significant air quality improvement since the industrial revolution [13]. As can be seen from Figure 2, air pollution in Europe hit the maximum in the first period, when the index registered was 0.246, and presented a decreasing pattern after the restrictions were put in place. The period with the highest restrictions was 28 March-10 April 2020, when the restriction index reached 22.3 The European air pollution index is measured on a numerical scale from 0.19 to 0.25 units, where 0.25 is the maximum value and 0.19 is the minimum value, and the restriction index is measured in points, from 0 to 25 points, summarizing the restrictions for each period analyzed.
In Figure 2, the air pollution index, which is an aggregate index of NO 2 , CO, SO 2 , PM 10 and PM 2.5 , had a decreasing pattern in urban areas as a result of energy reduction, trafficrelated combustion reduction and industry breaks. Since energy dropped by 25% every week when full lockdown was active, and with an average decrease of 3.8% from spring 2019 to spring 2020 on the basis of the drop in traffic-related activities and partial or full industries breaks, air pollution settled on a decreasing trend, consisting of a total decrease of 12% in the analyzed period, which is the most significant air quality improvement since the industrial revolution [13]. As can be seen from Figure 2, air pollution in Europe hit the maximum in the first period, when the index registered was 0.246, and presented a decreasing pattern after the restrictions were put in place. The period with the highest restrictions was 28 March-10 April 2020, when the restriction index reached 22.3 points. Since lockdown, both the air pollution and restriction index had a decreasing trend, figuring the causal relationship of a self-determined model. Figure 3 presents the main air pollution index changes (API change ) from March to October, for each country considered, composed as follows: where API 1 (X) is the air pollution index for a country X for the last period considered and API 0 (X) is the air pollution index for a country X for the first period considered. The change refers to the last two weeks and the first two weeks of the analyzed period.  When pollution is related to restrictions, the biggest impact was found in Croatia (1 unit of increase in restrictions generated a 0.50 decrease in pollution), followed by Bulgaria (0.25 decrease in air pollution for 1 unit increase in restriction index), Austria (0.21 decrease in air pollution for 1 unit increase in restriction index), Iceland and Lithuania (0.18 decrease, respectively, in air pollution for 1 unit increase in restriction index). Figure 4 presents the air pollution index (colored scale) and restriction index (blackgrey scale) in each country analyzed, from the European Union, in three periods: 1 March-30 May, 1 June-30 August and 1 September-30 October, in order to observe the seasonal changes. When pollution is related to restrictions, the biggest impact was found in Croatia (1 unit of increase in restrictions generated a 0.50 decrease in pollution), followed by Bulgaria (0.25 decrease in air pollution for 1 unit increase in restriction index), Austria (0.21 decrease in air pollution for 1 unit increase in restriction index), Iceland and Lithuania (0.18 decrease, respectively, in air pollution for 1 unit increase in restriction index). Figure 4 presents the air pollution index (colored scale) and restriction index (blackgrey scale) in each country analyzed, from the European Union, in three periods: 1 March-30 May, 1 June-30 August and 1 September-30 October, in order to observe the seasonal changes. decrease in air pollution for 1 unit increase in restriction index), Iceland and Lithuania (0.18 decrease, respectively, in air pollution for 1 unit increase in restriction index). Figure 4 presents the air pollution index (colored scale) and restriction index (blackgrey scale) in each country analyzed, from the European Union, in three periods: 1 March-30 May, 1 June-30 August and 1 September-30 October, in order to observe the seasonal changes. The values were composed as follows: The values were composed as follows: where API period y (X) represents the air pollution index for each of the three periods plotted (1 March-30 May, 1 June-30 August and 1 September-30 October) and X represents each country taken into consideration. All data points regarding pollutants were obtained in the US EPA standard, measured as the daily median for each country considered, normalized and computed as shown in Section 2 of the paper. Figure 4 clearly shows the decreasing pattern from the first period to the second, with an average decrease of 8.2% in the air pollution index, while from the second period to the third the decrease was around 3.2%. COVID-19 restrictions had the biggest positive impact regarding air pollution in Ukraine, Portugal, Bosnia and Herzegovina, Hungary, Italy and Austria, although Bosnia and Herzegovina, Hungary and Italy remained the most air polluted countries in Europe in late 2020. In contrast, Bulgaria, Romania and Iceland are the top cleanest countries regarding tropospheric air emissions. Before and after maps, presented in Figure 4, expose, in addition to the most air polluted countries within Europe, the most positively affected countries, with the most visible impact on Ukraine. If we relate the changes to restrictions put in place over the specified period of time, Ukraine, Bosnia and Herzegovina and Austria were in the top positively affected countries considering the restrictions put in place.
The results show that nitrogen dioxide had the biggest decrease from the first period to the last (from March to October), representing approx. 20.69% of the initial value, while particle matter 2.5 and carbon monoxide followed with an average decrease of 13.64% and 11.79%, respectively, from the first period to the last.
Changes in breathable air parameters are presented below, in Figure 5, as relative values from 1 March-30 May to 1 September-30 October, for each country and species analyzed: where avg 0 [P] represents the average of the daily median of the P pollutant for the period 1 March-30 May and avg 1 [P] represents the average of the daily median of the P pollutant for the period 1 September-30 October. analyzed: Changes in pollutant P = × 100 − 100% (9) where avg0 [P] represents the average of the daily median of the P pollutant for the period 1 March-30 May and avg1 [P] represents the average of the daily median of the P pollutant for the period 1 September-30 October. All species that were studied in the present paper suffered individual drops, or a slightly stable concentration from March to October.
From Figure 5 we can observe that the PM10 concentration dropped in the analyzed period for all the countries, except for Cyprus, Estonia, Iceland and Denmark, where the concentrations of PM10 in urban air had an increasing pattern from March to October 2020. Moreover, Estonia and Iceland had increasing patterns in both SO2 and PM2.5, while Cyprus' concentration of SO2 decreased over the analyzed period. North Macedonia, Bosnia All species that were studied in the present paper suffered individual drops, or a slightly stable concentration from March to October.
From Figure 5 we can observe that the PM 10 concentration dropped in the analyzed period for all the countries, except for Cyprus, Estonia, Iceland and Denmark, where the concentrations of PM 10 in urban air had an increasing pattern from March to October 2020. Moreover, Estonia and Iceland had increasing patterns in both SO 2 and PM 2.5 , while Cyprus' concentration of SO 2 decreased over the analyzed period. North Macedonia, Bosnia and Herzegovina, Serbia, Italy and Ukraine registered significant drops in all the analyzed pollutants, except for Bosnia and Herzegovina's changes in CO concentrations. On the other side, Denmark and Finland had increasing patterns in PM 10 and PM 2.5 concentrations, but drops in SO 2 , NO 2 and CO. While NO 2 had decreasing trends for all of the countries, CO and SO 2 registered drops or maintained the same concentrations over the analyzed period, except for in Estonia, France and Croatia, where significant or small increases for SO 2 were registered. PM 10 and PM 2.5 registered decreasing trends for the intercontinental countries, while they showed growth for the coastal, peninsular or insular ones.
The correlation of the air pollution index with the changes in industrial production and manufacture is highly important in order to understand the impact of each industry and manufacture economic activity as a source for air pollution. Thus, Table 3 presents correlations among the changes in the main industries and manufacture activities from 2019 to 2020 and the changes in air pollution during the same period. Table 3. Pearson correlation between changes in production in the main industries and manufacture activities and changes in urban air pollution.

p-Value
Pearson Correlation Industry and Manufacture Economic Activities It can be seen from Table 3 that the manufacture of textiles, motor vehicles, trailers and semi-trailers, and rubber and plastic products along with activities of households as employers and undifferentiated goods-and services-producing activities of households for their own use are the activities most correlated with air pollution among industrial and manufacturing activities.

Meteorological Analysis
Wind speed in spring 2020 exceeded the values from the same period in 2019, while the values from summer suffered a regression from 2019 to 2020, especially in the Iberian Peninsula. In September wind speed all over Europe presented a higher value in 2019 than 2020, with an exception in the Iberian Peninsula, while in the rest of autumn the upper trend from 2019 remained present, as shown in Figure 6. Wind speed was found to be negatively associated with the concentration of air pollutants, as shown in [16], with a higher air pollution being expected in 2020 than in 2019. The results are opposite to this statement, and thus it is reasonable to argue that the lowered value of air pollutants in 2020 had its source in other causes than the state of the wind. Mean temperature over Europe registered lower values in 2020 than 2019 in spring, summer and autumn, with an exception in July, when the Balkan region registered higher values in 2020 than 2019, as shown in Figure 7. While cold weather tends to foster the accumulation of PM and CO and make pollutants more visible, warm weather increases the amount of O3 concentration, along with PM2.5, PM10 and CO, when associated with natural forest fires. Moreover, CO tends to decrease when temperatures are high. The lower value of near-surface temperatures justifies, in part, the higher visibility of PM2.5, PM10 and CO. The average of the boundary layer height was lower in the months of April-August Mean temperature over Europe registered lower values in 2020 than 2019 in spring, summer and autumn, with an exception in July, when the Balkan region registered higher values in 2020 than 2019, as shown in Figure 7. While cold weather tends to foster the accumulation of PM and CO and make pollutants more visible, warm weather increases the amount of O 3 concentration, along with PM 2.5 , PM 10 and CO, when associated with natural forest fires. Moreover, CO tends to decrease when temperatures are high. The lower value of near-surface temperatures justifies, in part, the higher visibility of PM 2.5 , PM 10 and CO.
summer and autumn, with an exception in July, when the Balkan region registered higher values in 2020 than 2019, as shown in Figure 7. While cold weather tends to foster the accumulation of PM and CO and make pollutants more visible, warm weather increases the amount of O3 concentration, along with PM2.5, PM10 and CO, when associated with natural forest fires. Moreover, CO tends to decrease when temperatures are high. The lower value of near-surface temperatures justifies, in part, the higher visibility of PM2.5, PM10 and CO. The average of the boundary layer height was lower in the months of April-August in 2019 than in April-August in 2020, while presenting comparable values in autumn, with an exception for Scandinavia and the Baltics, where the boundary layer height was lower in 2019 than 2020, as shown in Figure 8. Boundary layer height and the ventilation coefficient are negatively correlated with the concentration of air pollutants, as shown in [17] for PM2.5 and black carbon. Total precipitations of 2020 exceeded the values from 2019 in June and July, and were exceeded by the ones from 2019 in April and May, as shown in Figure 9. Total precipitations are negatively associated with the concentration of NO2, CO and PM when controlling for wind speed, as shown in [16], while they are positively correlated with SO2. Total precipitations of 2020 exceeded the values from 2019 in June and July, and were exceeded by the ones from 2019 in April and May, as shown in Figure 9. Total precipitations are negatively associated with the concentration of NO 2 , CO and PM when controlling for wind speed, as shown in [16], while they are positively correlated with SO 2 . Total precipitations of 2020 exceeded the values from 2019 in June and July, and were exceeded by the ones from 2019 in April and May, as shown in Figure 9. Total precipitations are negatively associated with the concentration of NO2, CO and PM when controlling for wind speed, as shown in [16], while they are positively correlated with SO2. Figure 9. Mean of total precipitation over Europe in three periods, 2019 and 2020.

Individual Air Pollutants Analysis
In Figure 10, the mass concentration of carbon monoxide over Europe on three different dates is provided, covering the period before and after COVID-19 lockdown. The concentrations presented are averages of hourly data, as provided by the Copernicus Atmosphere Monitoring Station (CAMS).

Individual Air Pollutants Analysis
In Figure 10, the mass concentration of carbon monoxide over Europe on three different dates is provided, covering the period before and after COVID-19 lockdown. The concentrations presented are averages of hourly data, as provided by the Copernicus Atmosphere Monitoring Station (CAMS). Although the wind speed was more intense in April and May 2020 over Europe than in the same period of 2019, in July-September 2020 the intensity was set under the value of the same period of 2019, and the concentration of CO maintained a downward slope, in comparison with the same period of 2019, when the concentration was rising in the autumn of 2019. In [16], wind speed is negatively associated with the concentration of CO, when controlling for temperature, precipitation and relative humidity, with values ranging between −0.164 and −0.120, while temperature is negatively correlated when controlling for wind speed and air pressure (−0.225 and −0.272) and positively correlated when controlling for precipitation and relative humidity (0.220 and 0.197). Though mean temperatures in the late spring, summer and first period of autumn of 2019 were higher than in the same period of 2020, the concentration of CO in 2020 was not exceeded on the basis of colder weather, as expected. The boundary layer height was lower in April-July 2019 than in the same period of 2020; thus, it is expected for the concentration of CO to decrease from 2019 to 2020. Total precipitations of May and April 2019 surpass the amount from the same period of 2020, but in June-July the situation was reversed; thus, it is expected, by this condition, for the CO concentration to be lower in May and April of 2020 compared with the same months of 2019, and the opposite is expected in the June-July period.
Our analysis shows that carbon monoxide was cleaned up at the ground-level layer in Europe from March to October, with an approximately 12% decrease based on a reduction of activities that engage fossil fuel combustion, which is the main source of CO, up to 95%, as reported in [18]. Although fossil fuel combustion may primarily consist of outdoor activities, such as traffic, industry or power plants, some indoor activities may continue Although the wind speed was more intense in April and May 2020 over Europe than in the same period of 2019, in July-September 2020 the intensity was set under the value of the same period of 2019, and the concentration of CO maintained a downward slope, in comparison with the same period of 2019, when the concentration was rising in the autumn of 2019. In [16], wind speed is negatively associated with the concentration of CO, when controlling for temperature, precipitation and relative humidity, with values ranging between −0.164 and −0.120, while temperature is negatively correlated when controlling for wind speed and air pressure (−0.225 and −0.272) and positively correlated when controlling for precipitation and relative humidity (0.220 and 0.197). Though mean temperatures in the late spring, summer and first period of autumn of 2019 were higher than in the same period of 2020, the concentration of CO in 2020 was not exceeded on the basis of colder weather, as expected. The boundary layer height was lower in April-July 2019 than in the same period of 2020; thus, it is expected for the concentration of CO to decrease from 2019 to 2020. Total precipitations of May and April 2019 surpass the amount from the same period of 2020, but in June-July the situation was reversed; thus, it is expected, by this condition, for the CO concentration to be lower in May and April of 2020 compared with the same months of 2019, and the opposite is expected in the June-July period.
Our analysis shows that carbon monoxide was cleaned up at the ground-level layer in Europe from March to October, with an approximately 12% decrease based on a reduction of activities that engage fossil fuel combustion, which is the main source of CO, up to 95%, as reported in [18]. Although fossil fuel combustion may primarily consist of outdoor activities, such as traffic, industry or power plants, some indoor activities may continue emitting the gas. For example, heaters engaged in cooking or heating and smoking can produce CO [19], although indoor pollution does not reach the industrial levels [13]. In addition to its positive effects, such as enhancing plant growth in labs, mediating efficacy in pulmonary hypertension and acute liver failure when associated with NO, health benefits in animals in different pathologies, potential anti-inflammatory effect, anti-apoptotic effect, positive effects on blocking CO proliferation and anti-aggregatory properties, carbon monoxide can cause serious poisoning due to its complex bond with hemoglobin, thus reducing the capacity of O 2 absorption [19]. Short-term health effects of CO may include headache, dizziness, weakness, nausea, vomiting and even loss of consciousness and death from poisoning [19]. Moreover, visual perception, audition, sensorimotor performance and vigilance are weakened after a short exposure to CO. Long exposure to CO can lead to the development of long-term neurological symptoms, changes in memory, sleep changes, vision, smell and direction changes, anxiety, psychomotor disfunction and balance problems. Moreover, subtle effects on the brain after prolonged low exposure were obtained in [20]. It may cause more than 50% of fatal poisonings [21]. From the first period to the third one, carbon monoxide showed a 12% drop in outdoor concentrations on the groundlevel as measured by outdoor stationary monitoring stations along most important cities in Europe. Figure 10 presents the most cleaned areas of tropospheric CO above Europe, showing a pattern of industrial carbon monoxide, with a low impact of household sources on outdoor air pollution. Figure 10 shows that ground-level carbon monoxide does not exceed the 9 ppm per 8 h exposure (approx. 10,310 µg/m 3 ) in the US EPA standard or 10 mg/m 3 (10,000 µg/m 3 ) in EU standards.
In Figure 11, the mass concentration of nitrogen dioxide over Europe on three different dates is provided, covering the period before and after COVID-19 lockdown from spring 2020. The concentrations presented are averages of hourly data, as provided by the Copernicus Atmosphere Monitoring Station (CAMS).

Int. J. Environ. Res. Public Health 2022, 19, x FOR PEER REVIEW
important cities in Europe. Figure 10 presents the most cleaned areas of troposphe above Europe, showing a pattern of industrial carbon monoxide, with a low imp household sources on outdoor air pollution. Figure 10 shows that ground-level c monoxide does not exceed the 9 ppm per 8 h exposure (approx. 10,310 μg/m 3 ) in t EPA standard or 10 mg/m 3 (10,000 μg/m 3 ) in EU standards.
In Figure 11, the mass concentration of nitrogen dioxide over Europe on three ent dates is provided, covering the period before and after COVID-19 lockdown spring 2020. The concentrations presented are averages of hourly data, as provided Copernicus Atmosphere Monitoring Station (CAMS). The mass concentration of nitrogen dioxide registered a major regression of 2 from March to June 2020 compared to the same period of the previous year, over E as a result of a higher value of wind speed in April and May 2020 and boundary height in late spring and summer 2020, compared to 2019, in addition to the transpor and industrial activities reduction starting from March 2020. Moreover, in [22], a conducted in the Lombardia Region in Italy, based on a multivariate regression, concluded that wind speed negatively affects the NO2 concentrations with an appro tively 6% reduction. In a study conducted in Bangladesh, [23], wind speed was nega corelated with nitrogen dioxide with a correlation of −0.37, while temperature and h ity presented nonsignificant correlations. Although total precipitation was lower spring of 2020, compared to 2019, the trend was reversed in October, thus improvi The mass concentration of nitrogen dioxide registered a major regression of 20.69% from March to June 2020 compared to the same period of the previous year, over Europe, as a result of a higher value of wind speed in April and May 2020 and boundary layer height in late spring and summer 2020, compared to 2019, in addition to the transportation and industrial activities reduction starting from March 2020. Moreover, in [22], a study conducted in the Lombardia Region in Italy, based on a multivariate regression, it was concluded that wind speed negatively affects the NO 2 concentrations with an approximatively 6% reduction. In a study conducted in Bangladesh, [23], wind speed was negatively corelated with nitrogen dioxide with a correlation of −0.37, while temperature and humidity presented nonsignificant correlations. Although total precipitation was lower in late spring of 2020, compared to 2019, the trend was reversed in October, thus improving the air quality in comparison with the same period of 2019 by negatively correlating with NO 2 concentration [16]. In [22], it was found that both temperature and precipitation have no significant effect on NO 2 pollutant concentrations. Moreover, in [24], boundary layer height had the most significant influence over NO 2 concentration (18 ± 6%) followed by the surface wind speed (12 ± 5%).
Results show that nitrogen dioxide, the most important traffic-related pollutant, suffered a visible drop from the first period to the third one, based on a significant reduction of human mobility, expressed also by an oil demand reduction of 5%, by restricting aviation and terrestrial mobility (accounting for 60% of global oil demand) [13]. Thus, NO 2 recorded an approximately 21% reduction in terrestrial concentration, with a prolonged positive effect of the lockdown period and mobility restrictions. Health effects caused from exposure to NO 2 include respiratory system irritation and disease, coughing, wheezing, dyspnea, bronchospasm and eyes, throat and nose symptoms with limited exposure. NO 2 exposure can lead to pulmonary edema when a person is exposed to large concentrations (>0.2 ppm) or chronic lung disease when long-term exposure is involved. Other than during fossil fuel combustion from transportation and aviation, nitrogen is also released during synthetic fertilization, deforestation and biomass burning and from natural soils or oceans [25], of which only oceans possess a significant impact over coastal urban areas [26]. Although the biggest percentage of nitrogen dioxide emissions comes from the Haber-Bosch fertilization process [25,26], urban emissions are dominated by traffic-related activities [27]. As restrictions forced a human mobility reduction of up to 23 points in March 2020, in Europe, all traffic-related pollutants dropped along urban areas across Europe, with the most significant impact on nitrogen dioxide. While annual nitrogen dioxide levels were elevated in developed countries such as Italy, Germany and France before lockdown, surpassing the annual limit of Europe, after lockdown the general level of NO 2 stabilized around 15 µg/m 3 , below the US EPA annual limit of 99.73 µg/m 3 and EU annual limit of 40 µg/m 3 , as shown in Figure 11.
In Figures 12 and 13 the mass concentrations of PM 10 and PM 2.5 over Europe in three different periods are provided, covering the period before and after COVID-19 lockdown. The concentrations presented are averages of hourly data as provided by the Copernicus Atmosphere Monitoring Station (CAMS).
As a result of the negative association with temperature, the values of PM 10 and PM 2.5 from 2020 are below the ones from 2019, while the mean temperature from 2019 exceeded the near-surface air temperature at the tropospheric level from 2020. In October both PM 10 and PM 2.5 registered higher values in 2019 than the rest of the year, as a result of a lower mean temperature, but the spike was not present in 2020, when the decreasing trend was maintained. In contrast, in [22] it was found that both temperature and precipitation have no significant effect on PM 10 pollutant concentrations. The values were higher in winter and lower in summer, in both periods. Furthermore, in [22], it was concluded that wind speed negatively affects the PM 10 concentrations with values ranging from −13.89 to −6.97. Moreover, in [16], wind speed is negatively correlated with PM 2.5 when controlling for temperature, relative humidity, air pressure and precipitation with values ranging between −0.167 and −0.055. In a study conducted in Bangladesh, [23], wind speed is negatively corelated with PM 2.5 with a value of −0.58. In the same study, temperature is negatively associated with PM 10 and PM 2.5 when controlling for wind speed and air pressure, with values ranging between −0.406 and −0.251, while being positively associated when controlling for precipitation (0.255 and 0.238). Total precipitation is negatively associated with both PM 10 and PM 2.5 when controlling for wind speed, temperature, relative humidity and air pressure with values between −0.605 and −0.323. NO2 stabilized around 15 μg/m 3 , below the US EPA annual limit of 99.73 μg/m 3 and EU annual limit of 40 μg/m 3 , as shown in Figure 11.
In Figures 12 and 13 the mass concentrations of PM10 and PM2.5 over Europe in three different periods are provided, covering the period before and after COVID-19 lockdown. The concentrations presented are averages of hourly data as provided by the Copernicus Atmosphere Monitoring Station (CAMS).  As a result of the negative association with temperature, the values of PM10 and PM2.5 from 2020 are below the ones from 2019, while the mean temperature from 2019 exceeded the near-surface air temperature at the tropospheric level from 2020. In October both PM10 annual limit of 40 μg/m 3 , as shown in Figure 11.
In Figures 12 and 13 the mass concentrations of PM10 and PM2.5 over Europe in three different periods are provided, covering the period before and after COVID-19 lockdown. The concentrations presented are averages of hourly data as provided by the Copernicus Atmosphere Monitoring Station (CAMS).  As a result of the negative association with temperature, the values of PM10 and PM2.5 from 2020 are below the ones from 2019, while the mean temperature from 2019 exceeded the near-surface air temperature at the tropospheric level from 2020. In October both PM10 Particle matter is a pollutant caused by fossil fuel combustion, industrial and agricultural activities, in direct form, or within the process of transformation of NO x , SO 2 or other chemical compounds [28], being directly linked to the traffic level in urban areas [29]. Natural sources, such as volcanoes, dust storms, forest fires, living vegetation or sea spray count for a small percentage in urban areas, with the exception of sea spray, which is the largest source for coastal urban PM 10 [30]. Health effects are related to dyspnea (on short-term exposure) and cardiovascular disease and infant mortality on long-term exposure to PM 2.5 . Heath effects of long exposure to PM 10 and PM 2.5 include respiratory diseases, afflictions of the immune system, cancer, diabetes or other toxic effects and short exposure impact may consist of chest discomfort, chest pain, coughing and wheezing [19]. Along with these negative effects, PM can work as an adjuvant for allergic sensitization [31]. Restrictions on human mobility, over the first period, generated an almost 14% PM 2.5 reduction, based on the reduction of the combustion of coal, oil and gasoline by almost 3.8% in the first trimester [13] and the reduction of the transformation of other chemical pollutants, due to the reduction of already mentioned gasses together with the reduction in the erosion of pavements and abrasion of multivehicle components as a result of transportation restrictions. PM 10 registered a 10% decrease in urban areas, mostly based on suspension of soil reduction when agricultural and industrial activities or construction and transportation dropped, while ocean spray remained at the same markers, within its annual cycle, in the analyzed period. Thus, Figures 12 and 13 show the decreasing trend of both PM 10 and PM 2.5 in terrestrial areas within the European Union. PM 10 did not exceed the NAAQS limit for 24 h in all the analyzed period, all the concentrations being under 150 µg/m 3 , with Bosnia and Herzegovina and North Macedonia having the highest concentrations in the first days of March. Furthermore, PM 10 did not exceed the EU annual standards of 40 µg/m 3 in the whole period, except for Bosnia and Herzegovina, North Macedonia and Serbia. All the PM 2.5 concentrations exceeded the 24 h EPA standard of 12 µg/m 3 and the annual standard of 35 µg/m 3 in the analyzed period, except for Bulgaria. Moreover, PM 2.5 exceeded the EU standards of 25 µg/m 3 for most of the analyzed countries in the analyzed period.
In Figure 14, the mass concentration of sulfur dioxide over Europe on three different dates is provided, covering the period before and after COVID-19 lockdown from spring 2020. The concentrations presented are averages of hourly data as provided by the Copernicus Atmosphere Monitoring Station (CAMS). Sulphur dioxide maintained the decreasing trend from spring to autumn 2020, in o position with 2019, when the concentration presented a spike in autumn. Total precipit tions were higher in October 2020 than in October 2019, so the concentration of SO2 shou be lower in 2020 than in 2019 (in the same period, as they are negatively associated wi values ranging between −0.378 and −0.215 [16]). Among the health effects of SO2, irritatio to the respiratory system, penetration deep into the lungs, the introduction of respirato disease, bronchospasm/pulmonary edema, eye symptoms and cardiovascular disea were found in [32], while skin redness was mentioned in [33].
Sulfur dioxide suffered a reduction of 8.38% in ground-level emissions. The ma source of sulfur dioxide, such as power plants and other fossil fuel sources, along with o refineries and natural gas refineries, reported a reduction in activity based on general e ergy demand reduction, and sulfur dioxide followed the same trend, while other sourc still settled at the same scale, such as biomass burning [34] or smelters, which are some the largest sources of SO2 [35]. Moreover, individual facilities are not stringently consi ered as hotspots of sulfur dioxide emissions, as a recent study on the Middle East expose [35]. Sulfur dioxide did not exceed the EU limit of 125 μg/m 3 in the analyzed period, shown in Figure 14, nor the US EPA annual standard of 0.3 ppm. Table 4 presents the correlation coefficients between the restriction index constructe as an aggregated index of austerity and the analyzed pollutants. Thus, between the r striction index and air quality we observe a very strong relationship with a correlatio coefficient of 0.76 (with a significance value of 0.000). All correlations present a strong an direct relationship among the variables. Sulphur dioxide maintained the decreasing trend from spring to autumn 2020, in opposition with 2019, when the concentration presented a spike in autumn. Total precipitations were higher in October 2020 than in October 2019, so the concentration of SO 2 should be lower in 2020 than in 2019 (in the same period, as they are negatively associated with values ranging between −0.378 and −0.215 [16]). Among the health effects of SO 2 , irritation to the respiratory system, penetration deep into the lungs, the introduction of respiratory disease, bronchospasm/pulmonary edema, eye symptoms and cardiovascular disease were found in [32], while skin redness was mentioned in [33].

ARX Model
Sulfur dioxide suffered a reduction of 8.38% in ground-level emissions. The main source of sulfur dioxide, such as power plants and other fossil fuel sources, along with oil refineries and natural gas refineries, reported a reduction in activity based on general energy demand reduction, and sulfur dioxide followed the same trend, while other sources still settled at the same scale, such as biomass burning [34] or smelters, which are some of the largest sources of SO 2 [35]. Moreover, individual facilities are not stringently considered as hotspots of sulfur dioxide emissions, as a recent study on the Middle East exposed [35]. Sulfur dioxide did not exceed the EU limit of 125 µg/m 3 in the analyzed period, as shown in Figure 14, nor the US EPA annual standard of 0.3 ppm. Table 4 presents the correlation coefficients between the restriction index constructed as an aggregated index of austerity and the analyzed pollutants. Thus, between the restriction index and air quality we observe a very strong relationship with a correlation coefficient of 0.76 (with a significance value of 0.000). All correlations present a strong and direct relationship among the variables. In validating the alternative hypothesis that confirms the significant influence of human mobility restrictions over urban air quality, we used a paired t-test that concluded with a p-value of 0.00 under the considered significance level of 0.05. We assumed the normal distribution of sample means by the central limit theorem, taking into consideration that we used over 300,000 data points.

ARX Model
To understand the behavior of air pollution, we constructed an autoregression with exogenous variable (ARX). The best fitted model is constructed upon the last four periods (x t−1 , x t−2 , x t−3, x t−4 ) and the last pollution index (y t−1 ), as shown by Equation (10). The function took the following form: where α and β are coefficients; y is the dependent variable, air pollution; x is the independent variable, the restriction index; and ε are the residuals. We used the autoregressive function to forecast the air pollution index in the next four months, considering the restrictions to be at their maximum value. We took into consideration a value of restriction of 22.8, being the maximum value of the restriction index in the analyzed period. The forecast presumes the same behavior of air pollution based on the last historical data on air pollution and restriction. Figure 15  normal distribution of sample means by the central limit theorem, taking into consi tion that we used over 300,000 data points.
To understand the behavior of air pollution, we constructed an autoregression exogenous variable (ARX). The best fitted model is constructed upon the last four pe (xt−1, xt−2, xt−3, xt−4) and the last pollution index (yt−1), as shown by Equation (10). The fun took the following form: where α and β are coefficients; y is the dependent variable, air pollution; x is the pendent variable, the restriction index; and ε are the residuals. We used the autoregressive function to forecast the air pollution index in the four months, considering the restrictions to be at their maximum value. We took into sideration a value of restriction of 22.8, being the maximum value of the restriction i in the analyzed period. The forecast presumes the same behavior of air pollution b on the last historical data on air pollution and restriction. Figure 15 presents the beh of air pollution, under an ARX model, taking into consideration the last period fo pollution index and the last four periods of the restriction index. The left scale o graphic shows the air pollution index value for each of the periods analyzed, with a imum of 0.16 and a maximum of 0.25. The horizontal axis shows the period consid (from 0 to 20). It is clearly shown in Figure 15 that urban air pollution may be red below historical values, if human mobility restriction is maintained at the maximum els.

Discussion
The present study aimed to argue the impact of COVID-19 restrictions over air lution dynamics, such as the air pollution index in urban areas and changes in tr spheric air pollutants or pollution hotspots during or after the lockdown period. Re show a direct, significant relationship between restrictions and air pollution indicato

Discussion
The present study aimed to argue the impact of COVID-19 restrictions over air pollution dynamics, such as the air pollution index in urban areas and changes in tropospheric air pollutants or pollution hotspots during or after the lockdown period. Results show a direct, significant relationship between restrictions and air pollution indicators.
Overall, air pollution dropped by 12%, on average, with a local maximum in Ukraine (27% drop in air pollution), Croatia (with a 29% drop from March to October), Bulgaria, Lithuania and Portugal.
The results are supported by other findings, as in [36], where CO, NO 2 , PM 2.5 and PM 10 decrease and air quality improvement were associated with the lockdown period. Moreover, decreased NO 2 concentrations were reported in Italy, France and Poland [37][38][39]. Also, due to several measures taken by the government, there were reported decreases in CO, PM 2.5 and PM 10 in the US, Italy and Poland [40]. NO 2 reduction was also reported in Italy, China, India and the US [41] due to oil demand reduction. The same sources mention clean waters where water traffic was abundant before lockdown. A drop of 47-55% in NO 2 and CO emissions was reported in European countries in [42]. Reduction in NO 2 emissions in Italy, China, the US, Spain and France were obtained by satellite images in [43]. Moreover, Ref. [44] reported a reduction of 40% in NO 2 , with oil and coal as the most significant NO 2 sources. Furthermore, as traffic was reduced by almost 70% in the UK during lockdown, there was a reported 38.3% drop in NO 2 and 16.5% drop in PM 2.5 [45]. Similar results on CO and NO 2 were found in Italy [46]. As discussed in the present paper, general energy demand reduction contributes to air quality improvement. An interesting distinction was found between PM 2.5 and PM 10 changes, as ocean spray still remains an important source of PM 10 , with some studies supporting the same findings [47]. NO 2 , SO 2 and PM reduction in India was also reported in [48]. Moreover, an improvement in air quality index was found in northern China [49], with a decrease in SO 2 , PM 2.5 , PM 10 , NO 2 and CO (a drop of 6.76%, 5.93%, 13.66%, 24.67% and 4.58%, respectively). These results are consistent with our findings. A significant reduction of PM 2.5 was also found in India, as shown in [50], while particle matter, NO 2 and CO registered a drop during lockdown in the megacity Delhi, the actual most polluted city worldwide, with an approximate 40-50% increase in air quality [51], and in [44] particle matter dropped by 10%.
Other studies reported serious causalities among air pollutants and virus transmission, as in [36] it is stated that nitrogen dioxide is a promotor of COVID-19 transmission. For example, a significant increase in incidence rates is related to 1 µg/m 3 growth in NO 2 , and a smaller increase is related to 1 µg/m 3 growth in PM 2.5 [52]. In contrast, in [53], a weak association between O 3 and PM 2.5 with COVID-19 incidence rate was reported. Similar results were found in [54], where PM 10 and NO 2 were also significantly related to the risk of COVID-19 diagnosis.
In 2017, air pollution caused almost 1,200,000 deaths in India, and the same in China [44], with a total of 7 million deaths worldwide [55] in association with respiratory or pulmonary diseases; thus, ground-level air clean-up comes as a necessity in the global context. Moreover, COVID-19 had a 1.4% mortality rate [56], while air pollution registered a 7.6% mortality rate in in 2016 [57]. Furthermore, in 2019 a total of 441,998 premature deaths were associated with tropospheric air pollution across Europe, of which 307,000 were associated with PM 2.5 , 40,400 with NO 2 and 16,800 with O 3 [58]. According to the European Centre for Disease Prevention and Control, in Europe a total of 1,111,693 deaths were attributed to COVID-19 disease in 2020. Our data show that approximatively 42,000 premature deaths were avoided due to a reduction in PM 2.5 , while 8000 premature deaths were avoided due to a reduction in NO 2 , far beyond the number of deaths reported in relation with COVID-19 disease. In addition, an increased trend in waste fires, including biomass, was mentioned in [44] based on people's incapacity to properly handle the waste.
The present study was conducted on a continental region, including 28 countries, on a time perspective of 10 months, with daily data on each air pollutant and restriction being handled, with a total of over 300,000 data points that were imputed for missing values, normalized and aggregated to construct air quality and restriction indexes. Limitations of the study include the lack of data on methane and CO 2 , which is necessary to pursue for a more exhaustive approach, and the lack of a quantitative meteorological analysis and the implied uncertainties related to it.
Future directions may concentrate on widening the geographical area upon which the present study was conducted by including the global area and pursuing a global forecasting model of tropospheric air pollution on behalf of the intensity of human activity and mobility.

Conclusions
Before pandemic restrictions were put in place, ground-level air pollution was a serious problem regarding human health, consisting of a 7.6% mortality rate in 2016 [57], while in the lockdown period air pollution dropped by 12%. All pollutants registered declines, such as SO 2 , which registered an 8.58% decrease; CO, which registered an 11.94% decrease; PM 2.5 , which dropped by 13.64%; and PM 10 , which dropped by 10.63%. NO 2 presented the most significant drop of around 20.69%. All measurements were taken from urban monitoring stations in major cities across Europe.
The main causes of air quality improvement consist of energy demand reduction and transportation, industrial and agricultural activities reduction, although ocean spray remains an important source of PM 10 , explaining the small reduction of PM 10 in comparison with PM 2.5 . Other studies showed similar results in the same or different regions.
To observe the future behavior of air pollution, we constructed an autoregression with exogenous variable and found that the influence of restrictions is prolonged for four periods on best fit regression, with a forecast of another 3% drop in air pollution if restrictions maintain a maximum value.

Conflicts of Interest:
The authors declare no conflict of interest.

Symbol
Description NO 2 Nitrogen dioxide SO 2 Sulphur dioxide CO Carbon monoxide PM 2.5 Particle matter under 2.5 microns diameter PM 10 Particle matter under 10 microns diameter NASA The National Aeronautics and Space Administration COVID- 19 Coronavirus The concentration breakpoint first to exceed the C P BP Lo The concentration breakpoint first below the C P I Hi The AQI values corresponding to BP Hi I Lo The AQI values corresponding to BP Lo ARX Autoregressive model with exogenous input API change Air pollution index changes API period y (X) The air pollution index for each of the three periods: