Spatial Relationships between COVID-19 infection rates and Air Pollution, Geo-Meteorological and Social Parameters in Dhaka, Bangladesh

Like all infectious diseases, the infection rate of COVID-19 is dependent on many variables. Geographically Weighted Regression (GWR) model and GIS was used to understand the associations between COVID-19 infection rate as a dependent variable and 17 independent variables of air pollution, geo-meteorological and social parameters using a set of temporal data from 2010-2020 (May) in Dhaka, Bangladesh. This study revealed that air pollution parameters like PM 2.5 (p<0.02), AOT (p<0.01), CO (p<0.05), water vapor (p<0.01) and O 3 (p<0.01) were highly correlated with COVID-19 infection rate while geo-meteorological parameters like DEM (p<0.01), wind pressure (p<0.01), LST (p<0.04), rainfall (p<0.01) and wind speed (p<0.03) were also similarly associated . Social parameters like population density (p<0.01), brickeld density (p<0.02), and poverty level (p<0.01) showed high coecients as the key independent variables to COVID-19 infection rate. Geographically, signicant robust relationships of these factors were found in the middle and southern parts of the city where the reported infection case was also higher. Relevant agencies can utilize these ndings to formulate new and smart rules or strategies for reducing infectious diseases like COVID-19 in Dhaka and in similar urban cities around the world.


Introduction
In December 2019, the rst reported cases of a new infectious disease were found in Wuhan, China (Guan et al. 2020). It was named as COVID-19, a new virus of the group of coronaviridae, which is also known as severe acute respiratory Syndrome Coronavirus 2 (SARS-CoV-2) (Gorbalenya 2020; Pedrosa 2020).
The typical clinical symptoms include fever, dry cough, myalgia, and pneumonia which may cause progressive respiratory failure due to alveolar damage and may eventually lead to death (Huang et al. 2020). On 11 th March 2020, the World Health Organization (WHO) has declared the COVID-19 as a public health emergency because the number of cases drastically increased outside of China (World Health Organization 2020). As of May 20, 2020, 4,900,255 con rmed cases in over 200 countries/regions and 323,341 deaths were reported worldwide (Johns Hopkins University 2020).
Like other countries, Bangladesh is also suffering from this global pandemic. Bangladesh has a population of more than 163.7 million with a population density of 1,265/km 2 making it one of the most densely populated countries in the world (Bangladesh Population 2020). Almost 37.4% of its total population lives in urban areas with a high level of air pollution (Bangladesh -Urban population 2019).
Most Bangladeshi urban areas, specially its capital Dhaka, are highly exposed to different types of air pollutants e.g. PM 2.5 , PM 10 , COx, NOx, SOx, dust, and others because of uncontrolled urbanization, industrial emissions, high tra c density and construction activities. Air pollution is considered as one of the key threats for public health (Begum et  As of May 22, 2020, according to the Institute of Epidemiology, Disease Control and Research (IEDCR), Bangladesh reported 28,521 con rmed cases with total death of 408 in the whole country. The majority of the con rmed cases were identi ed in Dhaka which is also the epicenter of this infectious disease and till now there have been 12,386 con rmed cases reported with 103 deaths (Bangladesh Covid-19 Update 2020).
Research investigating the relationships between COVID-19 infection rate and air pollution are starting to appear in the literature ( (He et al. 2020) (Bontempi 2020)). A regional study in four countries of Europe to examine the relationship between NO 2 and the number of fatalities due to COVID-19 showed that longterm exposure to NO 2 may be one of the most important contributors to fatality caused by the virus (Ogen 2020). Researchers in Italy have speculated that atmospheric pollution is a co-factor in inducing high levels of deaths due to COVID-19 (Conticini et al. 2020;Martelletti and Martelletti 2020 (Conticini et al. 2020) have suggested considering population, social habits and meteorological condition together with air pollution when analyzing COVID-19 spread and mortality rates. However, caution is recommended by (Bontempi 2020), who have shown that there is no strong evidence that COVID-19 diffusion mechanism occurs through the air using PM 10 as a carrier (Bontempi 2020) even though PM 10 was shown to have a functional linear relationship with cardiovascular disease and non-accidental mortality rate in Spain (Ortiz et al. 2017). An 11 city study was conducted in China where non-linear models were employed to investigate the relationship between non-accidental mortality and NO 2 , PM 2.5 , temperature and relative humidity as covariates (He et al. 2020). The researchers found higher effect estimates of intermediate-term NO 2 exposure on respiratory mortality compared to that of the short-term but the differences were too small to be considered statistically signi cant. (Bontempi 2020) also mentioned that the airborne diffusion of COVID-19 is affected by the local air particulate matter (PM) in northern Italy but nds that there is no signi cant relationship between PM 10 and COVID-19 mortality rates.
There have also been studies relating to COVID-19 infection rates and geo-meteorological parameters. Temperature, humidity, and rainfall may have a direct in uence to spread the COVID-19 and other infectious vector-borne diseases (Casanova et al. 2010;Qi et al. 2020). (Xie and Zhu 2020) conducted a study in 122 cities across China to establish epidemiological and experimental research on ambient temperature and COVID-19 infection. Their exposure-response curve recommended that the mean temperature and COVID-19 con rmed cases are signi cant. Surrounding temperature is an essential factor to affect the transmission and survival of coronavirus (Xie and Zhu 2020). Humidity, wind speed, and temperature are inversely associated with the infection rate of the COVID-19 in 310 areas from 116 countries. (Nazrul Islam 2020) conducted this research and found that both cold and dry environments were favorable to the spread of COVID-19. (Xie and Zhu 2020) found a negative linear relationship between the COVID-19 con rmed cases and mean temperature in 122 cities in mainland China, suggesting warmer weather will not be a crucial factor to reduce the infection case of COVID-19. On the contrary, higher mean temperature and average relative humidity are quite signi cant in enhancing the COVID-19 contamination rate in Brazil (Auler et al. 2020) although average temperature and relative humidity were not consistent in terms of geographical areas, because of spatial heterogeneity problems (Qi et al. 2020). (Bashir et al. 2020) found a positive relationship between the COVID-19 and average temperature, minimum temperature, and air quality in New York, USA. Although they found a negative relationship with rainfall. (Hu et al. 2013) assumed that the transmissibility of an infectious disease is totally dependent on the population density of an area. They found that lower population density has the lower reported cases of an infectious disease than the higher density locations. There are other results revealed and concluded that population density is one of the key factors to control the COVID-19 (Luo et al. 2020;Pedrosa 2020;Sajadi et al. 2020). Integration of environmental, climatic, biodiversity, and manmade factors are useful for analyzing and predicting vector-borne and infectious diseases (Ceccato et al. 2018).
Based on the above discussion the factors in uencing COVID-19 infection rates can classify into three categories: air pollution parameters, geo-meteorological parameters and social parameters. This study will use 17 independent variables of air pollution, geo-meteorological and social data against the COVID-19 infection rate using Geographic Information System (GIS) and an advanced statistical tool of Geographically Weighted Regression (GWR). Considering these variables and methodologies, this study aims to understand the spatial relationships between the infection rate of COVID-19 and air pollution, geo-meteorological, and social parameters in Dhaka, Bangladesh.

Study location
The study area is located in Dhaka (or Dhaka Metropolitan Area) as the administrative and nancial capital of Bangladesh ( Fig. 1). It has an area of 370 km 2 with around 20.28 million inhabitants and geographically lies between 23° 40 and 23° 54 N latitude and 90° 20 and 90° 28 E longitude (Ahmed et al. 2013;Hoque et al. 2007). Geographically Dhaka city is located in the lower reaches of the Ganges Delta on the tributary of the Meghna Ganges river system (Ahmed et al. 2014). The monthly average temperature ranges from 25 to 31℃, the mean relative humidity and evaporation range between 80 and 90% and 80-130 mm, respectively (BMD 2016). The topography of the city is relatively at and its elevation varies between 0.5 and 12 meters (Hoque et al. 2007). Therefore, this study area is considered as a hotspot to test the relationship between the COVID-19 infection rate and 17 parameters in this study.

COVID-19 data
Daily Thana (police precinct) wise COVID-19 infection rate was used from two government sources in this study (Table 1). This database was entered into a city GIS shape le for further statistical and spatial analyses. Death statistics of COVID-19 data was not possible to collect because these are not yet publicly available at the time of writing this paper (May 23, 2020).
Collection of sample points 86 random sample points were collected from different parts of the study area. During this point data selection and collection, some basic guidelines were followed considering the main objective of the study and data variation due to multi-sources such as infection rate of COVID-19, population density, urban morphology, patterns of the residential area, high tra c zone, industrial area, high-and low-income zones, land use, etc. (Fig.1). On-screen digitization process was used to extract these points from different locations of the study area with the help of high-resolution Google map and a combination of other relevant secondary data.
Air pollution parameters 7 air pollution parameters, namely particulate matter (PM 2.5 ), nitrogen dioxide (NO 2 ), Aerosol Optical Thickness (AOT), sulphur dioxide (SO 2 ), carbon monoxide (CO), water vapor and ozone (O 3 ) of the period 2010-2020 (May) were used in this study ( Table 1). The annual average of high-resolution atmospheric data from different satellites was analyzed to derive spatial maps of each parameter because satellitebased air quality mapping and monitoring in urban areas have been advent as a new avenue of atmospheric research (Engel-Cox et al. 2004). After pre and post-processing all of the temporal air pollution parameters, the digital number of each parameter was extracted for 86 sample points. Finally, seven different maps of the annual average concentration of each parameter were generated using a point interpolation method of spatial analysis technique.

Geo-meteorological parameters
The raster data of the digital elevation model or DEM (m), wind pressure (ms -1 ), rainfall (mm), land surface temperature or LST (°C), and wind speed (mph) were collected from different sources and satellite information (Table 1). These geo-recti ed data were analyzed using GIS platform and nally resampled into 30-meter resolution in order to make an alignment with land use resolution. Finally, four different maps of the annual average concentration of each parameter generated using spatial analysis technique, while the DEM had only 1-year data.

Social parameters
Both vector and raster databases were used to classify the social variables (Table 1). All the vector data, after converting into raster datasets, rasterized to 30-meter resolution keeping GCS-WGS-1984 and D-WGS-1984 geographic coordinate system and datum respectively. On-screen digitization process was used for collecting brick eld points in the study area using high-resolution Google map. Brick eld density was included in the model because brick kilns are considered one of the main sources of air pollution in Dhaka (Guttikunda et al. 2013a). Thana wise literacy data from the Bangladesh Bureau of Statistics was used to assign points into Dhaka city map digitally to perform further analysis. Geographically Weighted Regression (GWR) Geographically Weighted Regression (GWR) method was used to map the association between ambient air pollution, geo-meteorological and social data with the COVID-19 infection rate. Geographically Weighted Regression (GWR) is an approach of exploring spatial non-stationary which permits different relationships to exist at different points in space by the process of calibrating several regression models in a systematic process (Leung et al. 2000). The fundamental principle of GWR is that parameters may be estimated anywhere in the study area given a dependent variable and a set of one or more independent variables that have been measured at places whose location is known (Charlton et al. 2009;Fotheringham et al. 1998). GWR offers the potential of investigating relationships that vary over space between variables in a regression model and it is quite a handy approach that allows complex spatial variations in parameter estimates to be identi ed, mapped, and modeled (Brunsdon et al. 1996;Wheeler and Páez 2010). In the GWR model component, X is a matrix containing a set of independent or predictor variables and Y is a vector of dependent or response variables. The main calculation of this model is in equation 1.
Where Y is the dependent variable, β o…. β n. are regression coe cients, X 1… X n is the independent variables and is the residuals error. In this study, Geographically Weighted Regression model was used for air pollution, geo-meteorological, and social parameters individually and each run were done in triplicate. These main GWR calculations for the models are given below-For air pollution parameters: COVID-19 (Y) = β o + β 1 (PM 2.5 ) + β 2 (NO 2 ) + β 3 (AOT) + β 4 (SO 2 )+ β 5 (CO)+ β 6 (WV)+ β 7  All the mapping exercises and graphical presentations were done using ArcGIS and Excel software respectively. The relevant shape le (Upazila vector) was collected from the Bangladesh Local Government and Engineering Department, while the land use map of the city was processed by one of the authors of the paper.

COVID-19 infection rate
As of May 22, 2020, about 6,245 COVID-19 cases were con rmed in Dhaka city. The graph shows that this virus infection has affected all areas of the city and the increase in cases from May 06 to May 18, 2020 (Fig. 2).

Descriptive statistical analysis of all parameters
The annual average concentrations of air pollution, geo-meteorological, and social parameters showed in respectively. The mean annual rainfall and wind speed were 275 mm and 0.96 mph respectively in the study area. In the social parameters, 0-1878 per person found per square kilometer while the mean poverty and literacy rates were 70% and 74% respectively. Geographical Weighted Regression (GWR) parameters Three Geographical Weighted Regression models were executed in order to understand the statistics and spatial relationships between the COVID-19 infection rate (dependent variable) and the 17 independent variables of air pollution, geo-meteorological and social data. After running these multi-regression models, signi cant model tting results such as estimated coe cient, standard error, T, p-value, and r 2 in the 3 sets of parameters are listed (Table 3).
In the air pollution data,   (Fig. 3, Map a). The mean value was 0.39 which was mostly found in the south and middle zones of the study area. On the other hand, the mean r 2 value (0.28) of geo-meteorological parameters was found in the middle, middle eastern, and southern parts of the study area keeping minimum and maximum values 0.18 and 0.73 respectively (Fig. 3, Map c). The social parameters were important predictors for establishing a relationship with the COVID-19 infection rate. The mean r 2 value (0.27) was found in the middle, middle eastern, and north-western parts of the study area (Fig. 3, Map b). Whilst its standard deviation was 0.04, the minimum and maximum values were 0.21 and 0.70 respectively.
The higher in uence values of independent variables predict a good explanation for the dependent variable. In air pollution parameters, a number of in uence zones in the north, eastern, western, and a few in the southern parts were found (Fig. 3, Map a). Except for the middle part of the study area, the in uence statistics was quite signi cant across the study area keeping 0.13 and 0.03 in mean and standard deviation respectively. The in uence of geo-meteorological parameters was signi cantly higher in the study area; except Cantonment, Gulshan, Ramna, Kafrul, Shahbag, Bangshal, Chakbazar, Gendaria and Shyampur areas (Fig. 3, Map c). The minimum and maximum values of geo-meteorological parameters were 0.07 and 0.13 respectively while its mean value was 0.09. Social parameters were quite in uential in the north-eastern and southern parts. The minimum and maximum values of social parameters were 0.06 and 0.21 respectively, while the mean value was 0.10 ( Fig. 3, Map b). A strong in uence of social parameters was found in the southern part of the city.
The p-value is an important parameter in this study. There were some strong relationships between COVID-19 infection rates data and air pollution, geo-meteorological, and social parameters. Strong statistical evidence of p values in PM 2.5 (<0.02), AOT (<0.01), CO (<0.05), water vapor (<0.01), and O 3 (<0.01) in the air pollution parameters were found. Most of these parameters and their correlated p-values were found in the middle, south, south-western, and south-eastern parts of the study area. Interestingly, O 3 , AOT, and wind vapor were found their strong prevalence over the whole study area (Fig. 5). showed a spatial distribution of r 2 map in all of the geo-meteorological parameters in where DEM, LST, wind speed, and wind pressure were found in the south, middle, middle eastern and southern parts in the area, respectively. Rainfall had a great statistical in uence on the dependent variable, which covered 90% of the study area.
Population density (p<0.01), brick eld (p<0.02),, and poverty (p<0.01) had signi cant roles on COVID-19 infection rate in the study area. In terms of the geographical distribution of population density and poverty, about 90% of areas were covered by these factors (Fig. 7). On the other hand, brick eld was highly visible in the southern part of the study area in which the COVID-19 infection rate was higher. Land use and literacy rate were not statistically signi cant, but these had a positive coe cient with the dependent variable.

Discussion
The main objective of this study was to investigate the spatial relationships between COVID-19 infection rate and air pollution, geo-meteorological, and social parameters in the study area using Geographical Weighted Regression and spatial analysis of GIS. Different results from the 3 models and its 17 parameters of 3 thematic areas were found.

COVID-19 infection rate with social parameters
Population density is an important parameter that catalyzed the outbreak of COVID-19 infection (J and H 2020). (Dhaval 2020) showed that most of the areas having a high-density population have more con rmed COVID-19 cases. For analyzing COVID-19 data, population density information needs to be used as one of the controlling variables and it was suggested that investigation should be done at local, regional, and national levels separately (Pedrosa 2020 and other environmental parameters, the situation is alarming for Dhaka as far as COVID-19 infection rates are concerned. With the government planning to resume all economic activities in the city, the level of air pollution will increase and this will have an increased effect on COVID-19 infection rates.

Conclusion
The spatial relationships between the COVID-19 infection rate and 17 independent parameters of air pollution, geo-meteorological, and social in the study area using Graphically Weighted Regression Model and GIS platform were studied. There were no signi cant statistical relationships between the COVID-19 infection rate and air pollution, geo-meteorological, and social parameters. From air pollution parameters PM 2.5 (p<0.02), AOT (p<0.01), CO (p<0.05), water vapor (p<0.01), and O 3 (p<0.01) were highly correlated with COVID-19 infection rate while DEM (p<0.01),, wind pressure (p<0.01),, LST (p<0.04),, rainfall (p<0.01) and wind speed (p<0.03) of geo-meteorological parameters were associated with the COVID-19 attack. Moreover, population density (p<0.01),, brick eld density (p<0.02),, and poverty (p<0.01) were highly coe cient as the key independent variables to the COVID-19 infection rate in the city. In addition to this, the government of Bangladesh can utilize these ndings in order to formulate new rules and strategies for reducing infectious diseases like COVID-19. This study will be useful for all researchers who are working on understanding the latent and visible relationships between the COVID-19 and its associated variables. The methodology can be replicated to a similar country or region considering the local micro-climatic environment. Studies with more variables including ecological, meteorological and social (GDP, BMI Global Health Index) variables will be vital to model and understand the spread of COVID-19.

Declarations Con ict of interest
The authors declare that they have no con ict of interest. Figure 1 The geographical location and distribution of COVID-19 infection in Dhaka city. Per Cyan dot means the ratio of infection rate on land use map. It is observed that the middle and southern parts of the city were the most affected zones due to COVID-19. The small black dot means the sample points to collect different parameters of air pollution and geo-meteorological data.  Spatial distribution of in uence in air pollution, social and geo-meteorological parameters Figure 5 Spatial distribution of p-value in air pollution parameters Figure 6 Spatial distribution of p-value in geo-Meteorological parameters