Assessing the Relationship between COVID-19, Air Quality, and Meteorological Variables: A Case Study of Dhaka City in Bangladesh

The novel coronavirus disease 2019 (COVID-19) has become a serious health concern worldwide for almost a year. This study investigated the effects of selected air pollutants and meteorological variables on daily COVID-19 cases in Dhaka city, Bangladesh. Air pollutants and meteorological data for Dhaka city were collected from 8 April to 16 June 2020 from multiple sources. This study implied spearman’s correlation to see the correlation between daily COVID-19 cases and different air pollutants and meteorological variables. Besides, multiple linear regression and the Generalized Additive Model (GAM) were used to investigate the association between COVID-19 cases and other variables used in this study. Due to lockdown measures, significant differences between PM 2.5 , SO 2 , NO 2 , CO, and O 3 in 2019 and 2020 were observed in Dhaka city. We used lag-0, lag-7, lag-14, and lag-21 days on daily COVID-19 cases to look at the lag effect of different air pollutants and meteorology. The LRM results showed that the daily COVID-19 cases are significantly correlated with relative humidity (lag-0 days) and pressure (lag-14 days) (p < 0.05). Additionally, the GAM model results showed a significant nonlinear association among daily COVID-19 cases and meteorology and air quality variables on different lag days. Therefore, our results suggest that an effective public health intervention measures should be implemented to slowdown the spreading of COVID-19.


INTRODUCTION
COVID-19 is an unprecedented outbreak that has a significant impact on the world's economy and public health. It is highly contagious Sohrabi et al., 2020) and has been rapidly propagating across the globe. On February 11, 2020, WHO identified the causative agents as coronavirus-2 (SARS-CoV-2) and named the disease "Corona Virus Diseases 2019 (COVID-19)" Lu et al., 2020). Until 10 March 2020, the total confirmed cases of COVID-19 were 137,658, and total deaths due to COVID-19 were 4,299 (Worldometer, 2020a). On 11 March 2020, the WHO declared the outbreak as a pandemic (WHO, 2020a). As of 28 September 2020, the WHO reported a total of 33,362,14 confirmed COVID-19 cases and a total of 1,003,198 deaths related to COVID-19 outbreaks in 215 countries and territories across the world (Worldometer, 2020a).
Bangladesh is one of the most densely populated countries in the world. It is the home of 8.9 million living in city area and a population of over 21 million in the greater Dhaka area. According to the United Nations (UN), it is one of the world's 33 megacities as of 2018 (UN, 2020). While the delineation of the Dhaka Megacity is still a matter of debate, the Dhaka City Corporation (DCC) area is considered the central urban hub. The geographical extent of DCC is located between 23.69° and 23.89°N latitudes and 90.33°and 90.44°E longitudes. The city is surrounded by six rivers: Balu and Sitalakhya on the eastern side, Turag and Buriganga on the western side, Tongi Khal to the north, and Dhaleshwari to the south. In 2011, the DCC area was divided into Dhaka North and South City Corporation. The city's climate is typically tropical monsoon with three different seasons, i.e., summer, winter, and monsoon. In Dhaka, 80% of the average annual rainfall falls during the monsoon season. Fig. 1 demonstrates study area overlaid by total COVID-19 cases by location in the city area.

Data
Four different sources of data have been used in this study. Hourly PM2.5 data for Dhaka city were collected from the Air Now website (https://www.airnow.gov/). Additional air pollutants data, including SO2, NO2, O3, and CO were collected from the Google Earth Engine (GEE) platform, collected by the Sentinel-5 mission of the Copernicus programme. We used four different meteorological variables in this study, including temperature, pressure, relative humidity, and wind speed. As such, we collected half-hourly data from Weather Underground. Daily COVID-19 cases data from 8 April to 16 June 2020 were collected from the Institute of Epidemiology, Disease Control and Research (IEDCR), Bangladesh (https://www.iedcr.gov.bd/). A detailed description of the datasets used in this study can be found in Table 1.

Methodology
Descriptive statistics of all the variables used were reported. In this study, the Spearman correlation (ρ) value has been used to look at the correlation between daily COVID-19 cases and other variables. A p-value < 0.05 were considered significant for all statistical analysis. Both the Generalized Additive Model (GAM) and Multiple Linear Regression (MLM) were used to evaluate   (Ma et al., 2020, Xie and. As the effect of meteorological variables can last for several days, and the incubation period of COVID-19 ranges from 1 to 14 day (CDC, 2020). Therefore, it was reasonable to use the lag effect of different air pollutants and meteorological variables in this study. Different studies also used the lag effect to look at the impacts of different air pollutants and meteorology on COVID-19 cases (Adhikari and Yin, 2020;Xie and Zhu, 2020;Zhu et al., 2020). Therefore, in this study, we have used lag-0, lag-7, lag-14, and lag-21 days to look at the effect of air pollutants and meteorological variables on daily COVID-19 cases in Dhaka city, Bangladesh. To implement the Generalized Additive Model (GAM), 'mgcv' package in R statistical software was used. The ArcGIS pro v2.3 was used for mapping.

Descriptive Statistics of Daily COVID-19 Cases, Meteorological Variables and Air Pollutants
Bangladesh reported the first three confirmed cases of COVID-19 in Dhaka city on 8 March 2020. Until 8 April 2020, there were 78 confirmed COVID-19 cases in Dhaka. Figs. 2(a) and 2(b) shows the variations of daily and cumulative COVID-19 cases in Dhaka from 8 April to 16 June 2020. The study shows that the daily COVID-19 cases increased gradually during the observation period, which ranged from 11 to 728, with the the mean value 200.8, Standard Deviation (SD) 124, and Coefficient Value (CV) 61.76 ( Table 2). As of 16 June, cumulative confirmed cases were 14000 in Dhaka city. Table 2 shows that from 8 April to 30 June, the PM2.5 concentration level ranged from 17.91 to 84.91 µg m -3 (Mean = 42.06 µg m -3 , SD = 14.26 µg m -3 , and CV = 33.91 µg m -3 ), the NO2 concentration level ranged from 0.000021 to 0.000179 mol m -2 (Mean = 0.000065 mol m -2 , SD = 0.000027 mol m -2 , and CV = 41.25 mol m -2 ), the SO2 concentration level ranged from -0.000281 to 0.000417 mol m -2 (Mean =0.000063 mol m -2 , SD = 0.000146 mol m -2 , and CV =234.40 mol m -2 ), the CO concentration level ranged from 0.029060 to 0.059180 mol m -2 (Mean =0.044227 mol m -2 , SD = 0.005054 mol m -2 , and CV = 11.43 mol m -2 ) the O3 concentration level ranged from 0.12079 to 0.13665 mol m -2 (Mean = 0.12834 mol m -2 , SD = 0.00436 mol m -2 , and CV = 3.40 mol m -2 ). Fig. 3 shows the daily variations of particulate matter (PM2.5) from 1 March to 30 June, and Fig. 4 shows the daily variations of air pollutants in Dhaka from 8 April to 30 June. This study found a gradual decrease in PM2.5 and CO concentration levels during the observation period. The concentration of O3 increased gradually until the last week of April and decreased gradually afterward. In contrast, the concentration of NO2 and SO2 were static. We believed that these variations in air pollutants' concentration during the observation period were caused by national lockdown measures when transportations and industrial activities were restricted. Islam et al. (2021) reported a similar finding in the major cities in Bangladesh because of reduced vehicular and industrial emissions during the national lockdown.
The study shows that temperature ranged from 24.38 to 30.73°C (Mean = 28.247°C, SD = 1.56°C,    and CV = 5.52°C), the relative humidity ranged from 56.81 to 88.81% (Mean = 71.432%, SD = 6.556%, and CV = 9.18%), the wind speed ranged from 3.25 to 29.39 km h -1 (Mean = 11.165 km h -1 , SD = 5.273 km h -1 , and CV = 47.23 km h -1 ), and the pressure ranged from 996.0 to 1011.6 hPa (Mean = 1005.6 hPa, SD = 3.45 hPa, and CV = 0.34 hPa) during the observation period ( Table 2). The daily changes of the meteorological parameters in Dhaka are shown in Fig. 5. The result shows that temperature gradually decreased until the last week of April and gradually increased afterward. Relative humidity increased gradually during the observation period. The wind speeds were found static and the pressure decreased gradually. We assumed that the changes of meteorological variables during the observation period had been seen because of seasonal variation across the country.

Improvement of Air Quality during the Lockdown
The air quality of a region depends on the concentrations of air pollutants, including PM, NOx, SOx, CO, and O3, over a specified averaging period. Due to the complete lockdown (i.e., 26 March 2020 to 30 May 2020), Dhaka's air quality was significantly improved due to the substantial reductions in air pollutants' concentration. The study shows a sharp decrease in PM2.5, NO2, SO2, CO concentrations and an increased O3 concentration immediately after the nationwide lockdown declaration in April 2020 (Table 3, Figs. 6-9). Mahato et al. (2020) also reported decreased PM2.5, NO2, SO2, and CO concentrations and increased O3 concentration due to the nationwide lockdown in the megacity Delhi (India), particularly in the areas with substantial road-transportation and industrial clusters. Such declining trends in air pollutants' concentrations were also observed in May 2020, except for NO2 and O3, while a sudden increase in NO2 and reduction in O3 concentration was observed (Table 3, Figs. 6 and 9). This trade-off relationship between NO2 and O3 supports the fact that a decrease in the levels of nitrogen oxides (NOx) subsequently increases the levels of O3 (NO + O3 = NO2 + O2) (Islam et al., 2021). These variations in NO2 and O3 concentrations indicated an increase in vehicular and industrial emissions, attributed to the consequence of lockdown relaxation and reopening of industries maintaining social distancing (Biswas et al., 2020;Hindustan Times, 2020). However, significant spatial variations in NO2 concentrations were observed in May 2020 (Fig. 6), with high intensity in the south and southeast part of the city. We assumed it might be due to the transboundary effects of local vehicular and industrial emissions from the adjoining town (i.e., Narayanganj). It should mention that Narayanganj is known as the country's most industrialized city and is also the worst COVID-19-hit district in Bangladesh (Noman et al., 2016;UNB, 2020).
Interestingly, a reduction in the air pollutants' concentrations was observed in June 2020, indicating less emission activities even after the withdrawal of complete lockdown (Table 3). This scenario can be explained by the effect of 'work from home' practices and the influence of monsoon season (i.e., June-September) in Bangladesh when rainfall helps significantly to washout atmospheric particulates and gaseous pollutant (Rahman et al., 2019). Nevertheless, the comparative analysis revealed that the concentrations of PM2.5, NO2, SO2, and CO observed in April 2020 were significantly lower than the same period in 2019 (Table 3, Figs. 6-8). On the contrary, the concentration of O3 was found higher in April 2020 than that observed in April 2019 (Table 3). Overall analysis exhibited that the concentrations of PM2.5, NO2, SO2, and CO in April 2020 were found to be decreased by about 17.31, 10.33, 47.92, and 6.07%, respectively, while O3 concentration increased by 1.84% (Table 3), reflecting the effectiveness of complete lockdown measures. The study also observed a significant reduction in SO2 concentration compared to other pollutants immediately after the complete lockdown declaration. Moreover, our study shows that the average concentrations of PM2.5, NO2, SO2, and CO were declined by about 17.11, 17.06, 10.93, and 2.15%, respectively, and O3 concentration was increased by about 3.99% in 2020 (Table 3), considering the full study period (i.e., March-June). Therefore, a substantial improvement in Dhaka's air quality has been observed due to reducing air pollutants' concentrations during the lockdown period. Table 4 represents the temporal associations between daily COVID-19 cases and air pollutants or meteorological variables in lag settings. Daily COVID-19 cases showed a significant negative correlation with PM2.5 at lag-0, with CO at lag-0 and 7, and with O3 at lag-0 to 14. However, NO2 and SO2 did not show any significant relation with daily COVID-19 cases. Bashir et al. (2020) also observed a significant negative correlation between PM2.5 and daily COVID-19 cases. In contrast, Hoang and Tran (2021) reported a significant positive association between CO and daily COVID-19 confirmed cases in Korea at lag-0 to 21. Ran et al. (2020) observed no significant association between daily COVID-19 cases and NO2, SO2, or CO in China. These regional variations in the associations between air pollutants and daily COVID-19 cases might be attributed to the differences in anthropogenic emission activities (Hoang and Tran, 2021). Moreover, it is evidence that other factors, including population density, social distancing, wearing masks, personal hygiene practices, immunity, policy interventions, etc. might influence the transmission of infection and the associations with regional air pollution.

Correlation between COVID-19 Cases, Meteorological Variables, and Air Pollutants
The study also reveals that meteorological variables have a significant correlation with the COVID-19 cases. For instance, relative humidity exhibited a significant positive correlation at lag-0, and atmospheric pressure showed a significant negative correlation at lag-0 to 21. Other studies also showed a significant positive correlation between humidity and daily COVID-19 cases in Mecca and Madina, Saudi Arabia (Alkhowailed et al., 2020), and Buenos Aires, Argentina (Bolaño-Ortiz et al., 2020). Although some studies (Alkhowailed et al., 2020;Rendana, 2020;Rosario et al., 2020) reported that the transmission of COVID-19 is negatively correlated with temperature and wind speed, our study observed no significant (p > 0.05) relationship between daily COVID-19 cases and temperature or wind speed (temperature: 24.38-30.73°C, wind speed: 3.25-29.39 km h -1 ). Xie and Zhu (2020) also observed a nonlinear relationship between temperature and COVID-19 confirmed cases at different lag days, with a mean temperature of 26.9°C. It is worth mentioning that the transmission of daily COVID-19 cases could increase during the upcoming winter season (i.e., November-February) like other coronaviruses (e.g., SARS-CoV and MERS-CoV) that had been reported to have better stability at low temperature (Chan et al., 2011). For instance, ~60% of the COVID-19 cases were observed in places with an atmospheric temperature of 5-15°C .
In addition, associations among and between air pollutants and meteorological variables are presented in Table 4. It is apparent that meteorological conditions influence air pollutants' concentration over a particular region . In our study, NO2 showed a significant positive correlation with PM2.5 at lag-0 to 21 and a significant negative correlation with CO and O3 at lag-0. However, a significantly positive association was observed between CO and O3 at lag-0. PM2.5 showed a significant negative correlation with relative humidity at lag-0 and 7 and with wind speed at lag-0 to 21. In contrast, a significant positive correlation was observed with atmospheric pressure at lag 0. NO2 showed a significant negative correlation with wind speed at lag-0 to 21 and with atmospheric pressure at lag-0. O3 represented a significant positive correlation with atmospheric pressure and a significant negative correlation with temperature and wind speed at lag-0 to 21. The temperature showed a significantly negative correlation with relative humidity and atmospheric pressure at lag-0 to 21, while other meteorological parameters showed a significantly negative correlation. Similarly, a significant negative correlation was also observed between wind speed and atmospheric pressure at lag-0 to 21. These results are found to be consistent with the results of other studies Hoang and Tran, 2021). Table 3

Association between COVID-19 Daily New Cases, Meteorological Variables, and Air Pollutants
We initially conducted a multiple linear regression model (LRM) for different lag days to assess the effects of air pollutants and meteorological variables on daily COVID-19 cases. The LRM result showed that all variables except relative humidity (lag-0 days) and atmospheric pressure (lag-14 days) exhibited insignificant results with daily COVID-19 cases. As our LRM didn't indicate any strong relation, we used Generalized Additive Model (GAM) to look at the association between COVID-19 cases and air pollutants and meteorological variables. Fig. 10 presents the exposureresponse for the mean value of different air pollutants and meteorological variables used with the log-transformed of COVID-19 cases on lag-0 days. The GAM model suggests that on lag-0 days, there was a significant nonlinear relationship between daily relative humidity and pressure (p < 0.05) with daily COVID-19 cases. This model explained about 48% of deviance. On lag-7 days, PM2.5 (p < 0.01), CO (p < 0.05), wind speed (p < 0.0001), and pressure (p < 0.001) were found to be significantly nonlinear associated with daily COVID-19 cases. The model for lag-7 days Fig. 10. Plots from GAM model in which daily COVID19 cases (lag-0) has been modeled as a smooth function of different air pollutants and meteorological variables.