Predicting tropospheric nitrogen dioxide column density in South African municipalities using socio-environmental variables and Multiscale Geographically Weighted Regression

Abstract

Atmospheric nitrogen dioxide (NO₂) pollution is a major health and social challenge in South African induced mainly by fossil fuel combustions for power generation, transportation and domestic biomass burning for indoor activities. The pollution level is moderated by various environmental and social factors, yet previous studies made use of limited factors or focussed on only industrialised regions ignoring the contributions in large parts of the country. There is a need to assess how socio-environmenral factors, which inherently exhibit variations across space, influence the pollution levels in South Africa. This study therefore aimed to predict annual tropospheric NO₂ column density using socio-environmental variables that are widely proven in the literature as sources and sinks of pollution. The environmental variables used to predict NO₂ included remotely sensed Enhanced Vegetation Index (EVI), Land Surface Temperature and Aerosol Optical Depth (AOD) while the social data, which were obtained from national household surveys, included energy sources data, settlement patterns, gender and age statistics aggregated at municipality scale. The prediction was accomplished by applying the Multiscale Geographically Weighted Regression that fine-tunes the spatial scale of each variable when building geographically localised relationships. The model returned an overall R² of 0.92, indicating good predicting performance and the significance of the socio-environmental variables in estimating NO₂ in South Africa. From the environmental variables, AOD had the most influence in increasing NO₂ pollution while vegetation represented by EVI had the opposite effect of reducing the pollution level. Among the social variables, household electricity and wood usage had the most significant contributions to pollution. Communal residential arrangements significantly reduced NO₂, while informal settlements showed the opposite effect. The female proportion was the most important demographic variable in reducing NO₂. Age groups had mixed effects on NO₂ pollution, with the mid-age group (20–29) being the most important contributor to NO₂ emission. The findings of the current study provide evidence that NO₂ pollution is explained by socio-economic variables that vary widely across space. This can be achieved reliably using the MGWR approach that produces strong models suited to each locality.

1. Introduction

Air pollution is a significant challenge affecting populations globally with more impact observed in developing countries [1]. Due to emissions from electricity generation, transportation, and residential fossil fuel burning, most of the world’s population lives in areas where air pollution levels exceed the World Health Organization’s health-based air quality limits [1]. Exposure to air pollution poses a major threat to human health and is associated with illnesses such as cancer, stroke, asthma and heart attacks [2]. In addition, air pollution is known to contribute to environmental problems, including climate change and acidification of soil and water bodies [3, 4]. Nitrogen dioxide (NO₂) is one of the main pollutants distributed across the atmosphere [5–7]. Research on the global trends of NO₂ distribution suggests that tropospheric NO₂ column density has been slightly increasing over time [8–10], although the opposite trend was observed during the lockdown forced by the coronavirus (COVID-19) pandemic that broke out in 2019 [11, 12].

The column density and distribution of NO₂ can be mediated by environmental variables that may serve as sinks or sources of atmospheric pollutants. Vegetation, for example, uses NO₂ to create amino acids, which stimulate plant growth, resulting in the removal of the compound from the atmosphere [13, 14]. A similar action by indoor plants decreases the indoor levels of NO₂ that may arise from the burning of biomass, paraffin, coal and wood [15]. Another environmental factor that affects atmospheric NO₂ level is land surface temperature (LST), which refers to the radiative temperature of the earth’s surface as a result of solar radiation [16]. The theoretical influence of LST on NO₂ is acknowledged as high NO₂ levels are observed under low-temperature conditions due to increased anthropogenic activities [17, 18]. The formation of NO₂ depends more on direct source emission compared to photochemical processes, hence NO₂ being high in winter compared to summer [18, 19]. High levels of NO₂ during winter indicate energy demand observed by the increased pollution from household combustion and other anthropogenic activities [20]. Aerosol Optical Depth (AOD), which refers to a measurement of particles suspended in the atmosphere, also plays an important role in the levels of NO₂ [21]. Aerosols are mostly produced at the surface of the earth because of various natural occurrences such as wind-borne dust, sea spray, volcanic debris and biological aerosols, as well as anthropogenic activities, industrial, agricultural-related dust, fossil fuel combustion and biomass burning [22].

Social factors such as population density can also affect atmospheric NO₂ column density [23, 24]. For example, research has shown that most air pollution is concentrated in metropolitan areas with high population density [23–25]. It is also important to acknowledge that high air pollution can occur in rural and peri-urban areas due to domestic biomass burning and cross-boundary transportation of atmospheric particles [26]. The impact of population density on air pollution is mainly driven by increased energy consumption for households, industrial activities and transportation to meet population demand [23, 27]. Gender may also play a significant role in the emission of atmospheric pollutants, with women responsible for most household activities, likely to produce higher NO₂ emissions than men. This also suggests that women and children have a higher exposure to the effects of NO₂ emission than men [28]. Literature on the influence of gender on NO₂ using remote sensing has mostly focused on the health effects of exposure to air pollution amongst men and women and the direct polluters are still unclear [29, 30]. Although the influence of age on atmospheric pollution is uncertain, epidemiological research shows that children and the elderly are highly vulnerable to the effects of air pollution and cardiorespiratory disorders because of their weakened immune systems [31]. Moreover, air pollution is associated with acute lower respiratory diseases in children less than the age of five years [32].

Developing countries tend to be the most affected by air pollution leading to their poor health and environmental conditions [28, 33–35]. Specifically, limited focus has been placed on the plight of air pollution in Africa leading to rampant socio-economic and environmental problems affecting the health of the population [34]. A recent statistic from Africa shows that, in 2019, indoor and ambient air pollution accounted for 697,000 and 394,000 deaths, respectively [35]. The recorded deaths induced by ambient air pollution were linked specifically to non-communicable diseases such as heart disease and chronic respiratory disease [35]. Industrialization, specifically in metropolitan areas of developing countries, is also contributing to air pollution and, thus health and environmental hazards [33, 36].

The correlation of NO₂ levels with socio-environmental variables is crucial to inform the required levels that benefit the environment and human health. This can be achieved by integrating spatial analysis and remote sensing that provides a quick and unbiased synoptic view of spatial variations of atmospheric pollution [8, 20, 37–40]. Zhu [37] predicted NO₂ in Chengdu, China, using Ozone Monitoring Instrument (OMI) meteorological and land use types and a Random Forest regression. The study found that NO₂ column density was higher in areas where anthropogenic activities were high. Similarly, in the Jiangsu province of China, [41] found similar results on the influence of temperature in increasing NO₂ levels in urban regions indicating the impact of factors such as industrial activities and population density on the pollution levels. Swartz [20] modelled long-term trends of atmospheric gases, including NO₂, in Mpumalanga and Limpopo, South Africa, using meteorological variables, population growth and Multiple Linear regression. The findings showed increased NO₂ column density with population growth, suggesting human-induced pollution activities in the regions. Moreover, a positive correlation between relative humidity and NO₂ was observed by the research, explaining the seasonal variation in NO₂ column density [20]. Based on a systematic review of remote sensing-based works focusing on the World Health Organization European Region, [42] found that social variables, including economic status and ethnicity, were linked to NO₂ levels. Specifically, the exposure to NO₂ was less in high-income neighbourhoods compared to lower-income neighbourhoods. Moreover, [42] found that men had less exposure to NO₂ than women.

South Africa is renowned for its coal deposits and heavy reliance on it for power generation; as a result, NO₂ represents one of the major pollutants in the country [43]. Besides electricity generation, other anthropogenically induced sources of NO₂ pollution in South Africa include transportation, fuel combustion, biomass burning and indoor air pollution [44–46]. Due to rising urbanization, waste burning has become a source of environmental and air pollution in South Africa contributing to atmospheric pollutants like NO₂ [45]. A continuous increase in population density, economic growth and urbanization necessitates a greater number of vehicles, leading further to NO₂ emissions from fuel combustion [44].

The spatial distribution of NO₂ across South Africa exhibits a high NO₂ pollution in municipalities located in the northeastern region compared to the rest of the country due to industrial activities and electricity generation in that region [47]. Variation in household energy consumption across South Africa is attributed to income inequality, geographical and social diversity and economic volatility [48, 49]. Income is a major determinant of access to electricity, with poor households having no access to electricity [49, 50]. There is a need to assess if a comprehensive list of socio-environmental variables can be used to predict atmospheric NO₂. A recent research by [47] investigated the distribution of NO₂, SO₂ and Sulphates (SO₄) concentrations across South Africa using AOD (elevated smoke and polluted dust) and wind data. The study focused on trend analysis of the pollutant observations without incorporating explanatory variables. In addition, [47] used only environmental data (wind) to qualitatively explain pollution distribution. Therefore, using various socio-environmental variables as predictors within a statistical model allows for a more comprehensive analysis and understanding of the health implications of pollutants [2, 28]. Considering that a majority of South African households across the country live below the poverty level and rely on alternative energy sources that contribute to NO₂ pollution necessitates the inclusion of social factors to predict the pollution [51]. The objective of this study was to predict annually derived tropospheric nitrogen dioxide (NO₂) column density using socio-environmental variables and Multiscale Geographically Weighted Regression at municipal scale in South Africa. The study is significant as, firstly, it represents the first national-scale assessment in South Africa and secondly, such a study provides valuable information for a rapid municipality-level expectation of atmospheric NO₂ pollution that can be exploited for decision-making at both the local and national levels.

2. Methods

2.1 Study area

The present study covers the entire South Africa (Fig 1), which has an area of 1 219 090 km² and has a coastline extending roughly 3200 km [52]. The country is divided into nine provinces subdivided into 213 local municipalities, including eight metropolitan district-level municipalities [53]. According to mid-year estimates for 2022, the country had a population of 60 604 992 with a life expectancy of 59 and 65 years for men and women, respectively [54]. Most of the local municipalities in South Africa are poverty-stricken, with rural areas being poorer than urban areas [51, 55]. It is believed that three provinces found in the north and north-central parts of the country (i.e., Mpumalanga, Limpopo and Gauteng) are the leading sources of anthropogenic NO₂ pollution in South Africa due to the power plants found in these provinces [47].

[Figure omitted. See PDF.]

2.2 Description of data

2.2.1 Environmental data.

NO₂ data acquired by the Sentinel-5P satellite sensor was downloaded from the Copernicus Open Data Hub platform (https://scihub.copernicus.eu/, (Accessed 06 February 2023). Eight spectral bands, including the ultraviolet and visible light (270–495 nm), near-infrared (675–775 nm), and shortwave IR (SWIR) (2305–2385 nm) spectrum, are acquired by Sentinel-5P equipped with TROPOMI sensor [56]. The satellite was launched on October 13, 2017, to monitor and forecast the global climate and measure atmospheric air quality factors with high spatial and temporal resolutions [56]. Moreover, data from Sentinel-5P provide continuous spatial coverage with a 3.5 x 5.5 km spatial resolution. For the present study, the daily-averaged NO₂ column density dataset for December 2018 to November 2019 was computed using the Google Earth Engine (GEE) platform and downloaded for further processing. Then, an average value was calculated for each municipality from the resultant annual NO₂ data. The mean municipality-level NO₂ distribution shows higher column density in the eastern and northeastern parts of South Africa (Fig 2).

[Figure omitted. See PDF.]

The environmental variables (EVI, LST, AOD) and NO₂ represent average values from December 2018 to November 2019. The legends of all social variables show the count of people divided by 1000.

The environmental parameters, i.e., EVI, AOD and LST, were retrieved from Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data for the period between December 2018 to November of 2019. The GEE was used to download this dataset as image collections, which are ingested from the National Aeronautics and Space Administrations (NASA) Land Processes Distributed Active Archive Center (LP DAAC) ((https://lpdaac.usgs.gov/products/mod13a2v006/, (Accessed 06 February 2023)). The MODIS EVI (MOD13A2), LST (MOD11A1) and AOD products had a spatial resolution of 1 km. MOD13A2 is a 16-day composite based on the best pixel value characterized by low cloud cover and view angle as well as the highest EVI value within the 16-day period [57]. On the other hand, the MOD11A1 product is available daily and built using the daily LST pixel values acquired using the generalized split-window technique under clear-sky conditions [58]. Finally, the AOD product (MCD19A2) was acquired from the MODIS Multi-Angle Implementation of Atmospheric Correction (MAIAC), which is generated by integrating time series analysis and a combination of pixel- and image-based processing; MAIAC provides accurate spectral reflectance that is used for cloud identification, aerosol retrievals, and Earth feature extraction [59]. Accuracy in retrieving AOD data is achieved by the post-processing stage, where several filters are employed to detect residual clouds and smooth the noise introduced from the grid from uncertainties [59]. The MODIS MAIAC AOD products are derived using the blue (0.47 μm) and green (0.55 μm) spectral bands; in this study, the blue spectral band was used as this is more sensitive to aerosol variations in the atmosphere [59].

Each of the products described above was averaged per annum using the GEE platform. Subsequently, the annually-averaged environmental parameters were spatially averaged by municipal boundaries, and the period was consistent with NO₂ column density data. A summary of statistics of the municipality-level NO₂ and the environmental variables is given in Table 1.

[Figure omitted. See PDF.]

2.2.2 Social data.

Population density data was downloaded from (https://openafrica.org/) for the year 2016. The dataset provides the geographical code, spatial extent, population count and population density of each municipality. The datasets representing municipality-level variables on household-level energy use, sex, age and dwelling types were acquired from STATISTICS SA (http://superweb.statssa.gov.za/webapi/jsf/tableView/tableView.xhtml). These datasets were collected through a national-scale community survey conducted in 2016. Such a large-scale survey is carried out by gathering household social data that are aggregated to the municipal administrative level. The 2016 datasets, therefore, represented the closest (in terms of time) to the NO₂ data at the municipality level. These data show the count of households belonging to each variable. Household energy use variables included coal, wood, electricity and paraffin for both cooking and space heating. The dwellings variable consists of stand-alone dwellings, townhouses, flats or apartments, a cluster of buildings in a complex, and informal settlements. Residence in any of these dwelling types can be linked to economic status, with well-off households affording stand-alone houses while low-income earners are forced to live in smaller and informal settlements [60]. Dwelling type can influence household energy source and demand [61] and therefore, it is relevant to explore how all dwelling types relate to NO₂ emission. The age variable included the number of people up to 69 years of age divided into 13 categories, each with a 5-year range. Maintaining the narrow-range age categories was preferred since it is unknown if such groups can influence NO₂. It is believed that analysis using such detailed categories for as long as possible does not compromise what would be obtained using a more generic category, as nearby categories would return similar results if similarities existed. The data on sex included the numbers of males and females of each municipality. All the above social data represented the year 2016. These data were, therefore, merged with the spatial data of South Africa’s municipality map that was published in 2016 [53]. The descriptive statistics of all the social variables are summarised in Table 2. In total, 35 variables, including three environmental and thirty-two social variables, were included for NO₂ estimation.

[Figure omitted. See PDF.]

2.3 Pre-processing

Since this study used a linear regression model to associate the NO₂ and socio-environmental variables, the variables had to be in a continuous numerical scale. Therefore, each social variable data was converted from count to continuous scale using the areal spatial interpolation technique [62]. The technique uses a kriging-based interpolation suitable for data collected within areal extents such as municipalities used in the current study. The approach is specifically skilled in accounting for the area of each polygon (i.e., municipality), thereby ensuring the proportional treatment of different-sized municipalities and counts of social data. The interpolated continuous data was subsequently aggregated by the municipality since the final regression analysis was done at such a scale. The plot-level median value was best compared with the original count data (i.e., Pearson correlation coefficient, r = 0.92–0.99) for each variable and, therefore, was used to derive the municipality-level social variables data. The spatial distributions of municipality-level social data used in the study are shown in Fig 2, while selected statistics of those data are summarised in Table 2. Higher concentrations of the social data values are found mostly in the eastern part of the countries, although there were variations among the variables justifying their inclusion in the study. These distributions largely coincide with the distributions of NO₂ column density in the country.

Although all the variables were converted into continuous scale, they still had different measurement units. As a result, all the response and explanatory variables were converted to a standard scale before building the regression model. In the present study, a standard score (also referred to as Z-score) was applied to each variable. The Z-score is computed by subtracting each value from the mean of all values and subsequently normalizing it by the standard deviation of all values. This process produces a value range with a mean of zero and a standard deviation of one, making the data suitable for MGWR modelling (Section 2.4). Although a linear model built on the global ordinary least square regression requires normality of data distribution, the MGWR does not need to honour a distribution pattern since the localization at different scales limits the sample sizes for each model. Standardization of variables is advantageous if the original input variables are measured in different units—as is the case in this study. The resultant parameter estimates (i.e., coefficients) derived from the scaled variables allow for a direct comparison of the influence of all explanatory variables. The outcome of the Z-score is a value range with a mean of zero and a standard deviation of one.(1)where Χ represents the value of the observed variable, μ is the mean value of the dataset and σ represents the standard deviation of the dataset [63].

2.4 Statistical analysis

The present study used a regression analysis to estimate NO₂ column density using the environmental and social variables as predictors. The Geographically Weighted Regression (GWR) is a non-stationary method that models a spatially varying relationship between a dependent variable and a set of explanatory variables [64]. Unlike the standard global linear regression that builds a single model using all the data in a study area, the GWR creates multiple models in different localities of a study area as long as sufficient variation exists among localities [64]. Thus, the method builds on Tobler’s first law of geography, which states that closer features on the Earth’s surface are more related than further ones [65]. The variation of all the data across space makes the GWR regression suitable for the present study similarly to previous studies [29, 66, 67]. The GWR is quantified as:(2)where characterizes the geographical coordinates of the point and a_κ represents the continuous function a_κ at point [64]. The GWR is considered a fixed-scale model that utilizes a single bandwidth parameter that determines how distance decay is used to weigh nearby data around each location’s coefficients [68]. This single bandwidth assumes that all relationships between the dependent and independent variables occur on an equal spatial scale [68].

The Multiscale Geographically Weighted Regression (MGWR) is an advanced GWR allowing bandwidth determination of each variable independent of other variables [69]. As such, the MGWR recognizes the variation in the influence zone that individual explanatory variables may have on the response variable. The formula for the MGWR is given as:(3)where in represents the bandwidth used to calibrate the jth conditional relationship [69]. The MGWR model is calibrated using the back fitting algorithm, which uses the expected log-likelihood method for parameter estimation [69]. In implementing the MGWR, the neighbourhood to build an optimal model for each municipality was determined using the number of neighbours. This decision is justified considering the variation in spatial size of municipalities; if distance, rather than number of neighbours, were used, there would be no guarantee of multiple members in a neighbourhood around large-sized municipalities. For each predicting socio-environmental variable, the neighbourhood size (i.e., number of neighbours) was optimized by building models iteratively and selecting the best one that returned the smallest Corrected Akaike’s Information Criterion (AICc). All analysis was performed in ArcGIS Pro 3.1.

3. Results

3.1 Local bivariate relationships between NO₂ column density and socio-environmental variables

Exploring the spatial variation in the type of relationship between each variable and NO₂ across the country is beneficial before running the MGWR. Fig 3 shows spatial variations in the significance levels and types of the bivariate relationships. The environmental variables EVI and AOD had a positive linear relationship with NO₂ in nearly all the municipalities in the western part of the country reaching up to R² of 0.83. The AOD, in particular, showed this type of relationship for a more significant part of the country than EVI. The significant relationships in the western portions of the country coincide with relatively low NO₂ column density. Notably, in the eastern part of the country, where NO₂ was relatively high, EVI and AOD had a non-significant or complex relationship with NO₂. The LST was inversely correlated with NO₂ (maximum R² = 0.79) predominantly in the northern part and some pockets in the west and south-eastern parts of the country. In the rest of the country, the LST had no significant relationship with NO₂.

[Figure omitted. See PDF.]

The relationship between energy sources for cooking and NO₂ showed clear spatial patterns, with significant influence (i.e., R² up to 0.79) observed mainly in the western part of the country (Fig 3). Moreover, a concave relationship was noted in the far western parts, while the immediate adjacent parts showed a linear positive influence on the increase of NO₂. Wood usage had the lowest significance in estimating NO₂ out of the five energy sources for cooking. The relationship between coal usage for cooking and NO₂ was markedly significant in the eastern part of the country. The relationship between energy consumption for heating purposes and NO₂ had largely similar spatial patterns of significance as the ones observed for cooking purposes. Two distinct differences can be seen with more significant influence in paraffin usage for cooking than for heating, as well as the more significance of coal for heating (maximum R² = 0.8) than for cooking (maximum R² = 0.7) in the western part of the country. The relationship between the number of households and NO₂ shows variation for each dwelling type, with a positive linear relationship being the most common type. In addition, a non-significant relationship between all dwelling types and NO₂ was observed mainly in the central parts of the country. Flats/apartments (maximum R² = 0.65), and formal (maximum R² = 0.68) and informal (maximum R² = 0.70) dwelling types had the most significant relationship with NO₂. The formal and informal dwelling type had significant but non-linear (concave) relationships with NO₂ in the far western parts of the country.

Population density exhibited a significant relationship (maximum R² = 0.82) with NO₂ for a large part of the country, with most of the relationship with NO₂ being positive linear or concave along the western and northwestern parts of the country (Fig 3). The relationship was generally insignificant in the central and eastern parts of the country for population density (Fig 3). The significance of the number of females (maximum R² = 0.65) followed the same pattern as the number of males (maximum R² = 0.77) but in a smaller number of municipalities than in the case of males. Regarding the influence of age groups, the most significant relationship with NO₂ was observed in the western half of the country in all age groups (Fig 3). Strikingly, the relationships were non-linear (concave) in the far western part of the country for all age groups.

3.2 NO₂ prediction using socio-environmental variables

Fig 4 shows the results of annual average NO₂ distributions across South Africa for the period December 2018 to November 2019. The results represent estimations achieved using all socio-environmental variables (n = 35) as predictors in MGWR. The observed NO₂ ranged between -0.72 and 5.96 mol/m² (Fig 4a) while the predicted values ranged between -1.21 and 5.30 (Fig 4b). This indicates the overall agreement between the observed and estimated NO₂. Furthermore, the MGWR predictions reflect the spatial patterns with higher and lower NO₂ in the northern and western parts of the country, respectively. The municipalities with the lowest prediction errors were concentrated mostly in the western part, with standardized residuals of -0.5 to 0.5 (Fig 4c). There were 95 and 118 municipalities with over- and under-estimated NO₂ column density, respectively. However, a global autocorrelation analysis of the errors using the Moran’s I index [70] showed a random distribution across the country (a low z-score of -0.902563; p = 0.366758), indicating the lack of spatial pattern or bias in the estimation error. On the other hand, each social and environmental variable showed significant global clustering (z-score ranging from 3.388443 to 32.678918; p < 0.001). The high correlation between the observed and predicted NO₂ column density at R² = 0.92 shows the predicting capability of the socio-environmental variables considered in the study (Fig 4d). The strongest correlations were observed mainly when standardized NO₂ values were low, with approximately 1 mol/m² or less, although even the higher column density was estimated well.

[Figure omitted. See PDF.]

The annual spatial distribution maps of standardized NO₂ vertical column density; (a) Observed standardized NO₂, (b) Predicted standardized NO₂, (c) Standardized residual map, (d) Scatterplot of the relationship between the observed and predicted standardized NO₂.

The spatial distribution of each predictor’s influence on NO₂ is shown in Fig 5. The two environmental variables, EVI and LST, have an inverse influence on NO₂ throughout the country, with higher values of the two variables resulting in lower NO₂ column density. Comparing the two environmental variables, the EVI (-0.221 to -0.230) has more impact than LST (-0.158 to -0.161). Furthermore, the EVI’s influence is higher on the eastern side of the country than in the west, while the opposite is true for the LST. AOD had a direct influence on NO₂ levels across the country (0.177 to 0.887), with higher impacts seen in the northeastern parts than in the rest of the country (Fig 5).

[Figure omitted. See PDF.]

Variations of green colour show the negative (opposite) influence of a variable on NO_2, while the yellow-to-red colour scheme shows the positive influence on NO₂.

Energy usage for cooking also directly influences NO₂ except for gas-based cooking, which showed an inverse influence on NO₂ as seen in Fig 5. Another exception is observed in wood usage for cooking which had an inverse impact of NO₂ in a pocket in the north-eastern part of the country. Apart from wood usage for cooking, the variations in the influence of each energy usage across the country were low. A comparison of the influence of the energy sources for cooking indicates that electricity has the most influence on NO₂ increase (i.e., 1.400 to 1.402), whilst gas (-0.004 to -1.010) has the lowest influence. Energy usage for heating directly influences NO₂ when gas and wood are used but has an inverse impact on electricity, paraffin and coal sources. The impacts of electricity and coal usage for heating ranked the most and least influential, respectively, on the estimation of NO₂. Like the spatial distributions of energy sources for cooking, the influence of each energy source for heating showed relatively low variations across the country.

Among the dwelling types, flat/apartments (-0.281 to -0.279), clusters in complex (-0.397 to -0.395) and formal dwellings (-0.035 to -0.032) reduced NO₂ levels throughout the country while the opposite influence was observed in townhouses and informal dwellings with coefficient values ranging between (0.150 to 0.152) and 0.220 to 0.223), respectively (Fig 5). Regarding the influence of population on NO₂ prediction, the results show that population density and the number of males affect NO₂ directly, while the number of females has the opposite effect on NO₂. Of the population data, the number of females had the most important contribution (-1.840 to -1.841) to the NO₂ level, followed by the number of males (0.210 to 0.213). The number of people by age category showed mixed levels of influence on NO₂ column density, with age groups of 5–9, 20–24, 25–29, 45–49, 55–59 and 60–64 showing a direct influence while the rest had a negative influence on NO₂. The younger and mid-age groups (up to 44) than the rest had more impacts on NO₂ levels, although the spatial variations of impacts for each social variable were relatively low.

4. Discussion

4.1 NO₂ correlation with individual socio-environmental variables

The key objective of the study was to assess the link between socio-environmental variables and atmospheric NO₂ column density distribution in South Africa. Each socio-environmental variable correlated with NO₂, with the environmental variables of EVI and AOD showing the most correlations across the country (Fig 3). The strong association of AOD with NO₂ level is not surprising considering that aerosols comprise all suspended particles, including NO₂. The inverse correlation between EVI and NO₂ is generally known; however, the present study showed an increase in NO₂ with an increase in EVI, particularly in the western part of the country. The western part is characterized by arid conditions and thus less vegetated than the east [71]; this can also be confirmed by the EVI distribution in Fig 2. This suggests that the vegetation did not have a sufficient amount to influence the NO₂ emission levels in the area. This observation is consistent with a study by [72] in India, which indicated that population density and the number of people in the younger age groups (up to age 24) also contributed to the increase of NO₂ for a large part of the country. Population density effects on air pollution especially in urban setups is linked to the extent of mobility and energy usage [24]. The younger populations’ higher influence on NO₂ is expected, given that this group is associated with commuting to schools, which contributes to transport-driven pollution [73]. A linear positive relationship was the most common correlation, followed by a concave relationship, while wood usage for cooking and heating purposes showed significant correlations with NO₂ for a limited number of municipalities. The concave relationships indicate the negative effect of a variable on NO₂ up to a certain NO₂ level, beyond which the relationship becomes positive.

4.2 Capability of socio-environmental variables to predict NO₂

The MGWR-based modelling that used all socio-environmental variables as explanatory variables returned a high R², i.e., 0.92 (Fig 4). Low errors distributed randomly, as opposed to dependence on space, across the country confirms the strength of the prediction. This is despite the clustering of values for each social and environmental variable, with high values observed mainly around major urban areas for most of the variables. The unbiased spatial distribution of the error justifies the use of GWR modelling that computes parameter estimates for each locality [64]. Furthermore, the advantage of using the improved MGWR that determines the scale of each explanatory variable independently [69] was evident in this study. This can be verified by the fact that individual factors correlated with NO₂ differently (Fig 3) and thereby have varied influence zones (spatial scales) on NO₂. Regarding the spatial distributions of NO₂, high values were predicted in the north-central or north-eastern part of the country, matching the observed map (Fig 4). The match of this distribution with that of the AOD is notable. Most of South Africa’s power plants (Fig 2) and industrial activities such as mining are found in the north-central or north-eastern parts [74, 75]. These activities emit byproducts that elevate the concentration of AOD, which includes NO_2, among other particulate matters [75, 76]. Our findings are compared to a global air pollution modelling study, which suggested a positive correlation between NO₂ and AOD is observed across the world due to similar emission sources [77]. Moreover, [77] observed this relationship in regions of high traffic density which is a variable that was overlooked in this research. The significance of AOD on NO₂ in this study is evident from the large range (0.177–0.887) of the coefficients compared to the lower and narrower ranges of EVI and LST (Fig 5).

Among energy sources and uses, electricity consumption for cooking purposes had the biggest contribution to NO₂ levels (Fig 5). The limited variation in the influence of electricity usage can indicate the homogenously high demand for it across the country, suggesting the gloomy outlook in reducing NO₂ pollution. The use of wood for cooking had a mixed effect on NO₂ levels, showing a reduction effect only in the north-central part of the country, compared to its increasing effect in the rest of the country. The increase in the number of people residing in flats/apartments and in clusters within complexes was associated with a reduction in NO₂ levels. Such dwelling types share restricted management of resources, including energy utilization; thus, their emission contribution remains limited [78]. In comparison, informal dwellings and townhouses exercise more relaxed freedoms in using energy for different purposes [79]. Although informal settlements in South Africa continue to struggle with a lack of service deliveries, many of them access electricity through illegal connections thus adding pressure to coal-powered stations [79, 80]. Because such dwellers do not pay for electricity services, there is a potential of unrestricted electricity usage in such dwelling types. Since this study found positive association between NO₂ emission levels and electricity consumption, it is critical for policymakers to consider the management of such dwelling types.

The female population group had the biggest impact on NO₂ pollution and was associated with the reduction in pollution (Fig 5). This contrasted with the impact of the male group that was associated with increasing pollution levels followed by the population density variable. The finding regarding the reducing effect of the female population group on NO₂ emission is encouraging, considering that most of the household activities (in most African contexts) are carried out by this group. Moreover, studies have suggested that women tend to be more informed about climate change and show concern for air pollution compared to their male counterparts [81, 82]. This difference in perception on air pollution could justify the observation made by this study when the influence of gender on NO₂ is concerned. The observed results suggest their wise consumption of energy for cooking and space heating since these activities heavily rest on them. This study may indicate that this strategy has been successful in the sense of greatly lessening the detrimental effects of males and population density variables with regard to their positive relationships with NO₂ (Fig 5). The feedback effect of the NO₂ pollution reduction on the female group should also be underlined, as women are generally the main victims of pollution due to their immediate exposure to domestic pollution sources [83]. Lower NO₂ pollution associated with increase in the female population is therefore welcomed as it may contribute to improvement in air quality and reduce its health effects. Although the population density proved to be the least significant contributor of pollution, its significantly high contribution in the eastern than in the western part of the country is noteworthy (Fig 3). This is attributed to the fact that the eastern part, that includes the economic hub of the country, such as the Gauteng province as well as the agriculturally rich provinces, has a higher population density than the western part. This, in turn, adds to the increased consumption of industrial products and energy resources that lead to NO₂ emissions. Ryu et al. [84] found similar results in a study conducted in South Korea of high NO₂ concentrations in the metropolitan areas compared to rural areas where population density is lower and are less industrialized. Furthermore, a worldwide investigation by [77] reported similar results showing a direct relationship between population density and NO₂, which specifically explained the variance of NO₂ mostly in Asia where population density is high.

Similarly, the high NO₂ level in the north-central or north-eastern part distinctly matched the high number of people dwelling in townhouses and clusters in complexes. Again, this is attributed to the prominence of the many dwelling types in the area to accommodate the large population participating in the region’s multitude of economic and academic activities [24]. However, it is important to take notice of other areas of the country with high concentrations of the other social variables in the eastern and western parts, in addition to the central part, but with low NO₂ levels. This suggests that the pollution in those areas can be linked to variables that were not included in the study, or the impacts of the social variables were limited by environmental variables such as the EVI and AOD. Age of people had a mixed effect on NO₂ levels, with certain groups linked to more pollution than others. The mid-age group’s (20–24 and 25–29) association with increased NO₂ levels is noteworthy since this group has one of the largest population sizes. Similar findings were reported by [85] who showed that the working-class age group (20–34) had a positive link with air pollution in 17 developing countries as a result of transit to work. Tarazkar et al. [86] also indicated that individuals in the labour force consume more energy than children and the old population, and they also release more greenhouse gases as a result of higher production activity. Given the high use of transportation expected from the mid-age group for purposes such as economic activities most likey increases the pollution, concurring with other studies that reached at the same conclusion. As a result, the pollution contributed by this age group should present a concern unless the main cause of the association is identified and addressed. The age groups 55–59 and 60–64 also contributed to NO₂ levels, though to a lesser extent than the mid-age groups. Estiri and Zagheni [87] attributed a higher energy consumption among older population groups and predicted an increasing trajectory of the effect for the future in the face of warming temperatures and climate change that place more demand on energy.

4.3 Study significance, limitations and recommendation

The findings of the present study showed high accuracies of NO₂ column density predictions could be achieved by using social and environmental variables as predictors. Although the study does not claim to represent a causality analysis, it provides strong evidence of the capability of socio-environmental variables to explain the variation in NO₂ levels. This capability can support environmental auditing programmes such as air quality monitoring and greenhouse gas emission accounting efforts. The monitoring and maintaining of clean air fit into various Sustainable Development Goals (SDGs) of the United Nations [88]. These include SDG Target 3.9.1, which focuses on the reduction in air pollution caused deaths and illnesses; SDG Target 7.1.2 which strives for access to clean energy in homes; and SDG Target 11.6.2, which promotes the reduction of the environmental impact of cities by improving air quality. One of the benefits of successfully predicting pollution using many social variables lies in the fact that all the pollution drivers can be managed given available resources and efficient management strategies. Even the environmental variables considered in the modelling exercise are largely influenced by anthropogenic activities, making them fall within the scope of social decision-making processes. Regarding the modelling technique, the MGWR approach proved highly successful in reducing spatially unbiased estimation errors. This approach is suitable when the focus area extent is large and variations due to location exist, as shown in the present study. The results revealed a varying level of influence of each socio-environmental variable across the country. Intervention efforts to reduce pollution by targeting these variables should, therefore be customized by location to invest resources efficiently.

Although the findings of the study were promising, future studies could improve on it as follows. Only three environmental variables believed to affect NO₂ column density were used in the study. Adding more environmental variables, such as other atmospheric variables and climatic data can make the prediction robust and resistant to fluctuations of certain variables. The social variables used in the study were limited to rather outdated statistical data on household-level energy consumption, population density, sex proportions, dwelling types and age distributions. Additional and up-to-date data such as energy use by age, gender, and dwelling type carry more specific information on energy consumption pattern of a community, and thus, including such data can strengthen the prediction. Furthermore, data on transport would certainly improve the modelling accuracy as such dataset remains one of the major drivers of NO₂ and other atmospheric pollutions [17, 72]. From the modelling perspective, it was opted in the study to use individual variables in the MGWR. The use of several variables in a model can suffer from collinearity among variables, resulting in an unnecessarily complex model. Such complexity can be reduced using different techniques one of which is Principal Component Analysis which combines the variables to generate a reduced number of uncorrelated explanatory variables. However, such an approach must be explored cautiously as it can result in removing information from the original variables. Another approach that can be explored to produce a robust pollution prediction model is using robust machine learning algorithms that do not require honouring statistical preconditions (e.g., data distribution) [89].

5. Conclusion

The impact of social and environmental factors on NO₂ pollution remains one of the major challenges to achieving clean air goals. This study aimed to show the link between socio-environmental variables and atmospheric NO₂ levels across South Africa using data obtained from remotely sensed sources and national population surveys. Results of the MGWR showed the explanatory power of socio-environmental variables to NO₂ variations with an overall R² of 0.92. The model’s accuracy is confirmed explicitly by the estimation errors that were not only low but also distributed randomly, indicating the unbiased prediction capability of the model. This represents a crucial success of the prediction, considering that individual explanatory factors varied with space. From the environmental variables, the AOD proved to be the major contributor to pollution while the EVI partially negated the impacts of AOD, providing the evidence to push for a greener environment. Electricity and coal usage for cooking and wood for space heating were most influential in the increase of NO₂ compared to the other energy sources and uses. Targeting these energy sources and adopting environment-friendly sources and consumption can reduce pollution levels.

Dwelling type also had a significant impact on group residences, including clusters-in-complexes and apartments, reducing the pollution amount. Given the continued population increase along with global climate warming, group residence is perhaps one of the best options since it limits the resource-consumption culture of humans. In contrast, informal settlements expectedly increased the pollution level which can be attributed to disorganized service delivery such as illegal power connections, unsustainable energy usage, and poor waste management and transport systems. A working model is therefore needed to improve informal settlements by providing platforms that reduce the impacts of the current resource usage on pollution. As expected, an increase in population density coincides with an increase in NO₂ pollution due to the intense use of energy and transport per given area. Reducing the pollution problem in densely populated areas requires a combination of intervention measures compared to what would be needed to mitigate the impacts of other factors. The impact of the female population in reducing NO₂ pollution is one of the most encouraging findings of this study. Females carry most of the burdens in society and yet proved to be efficient in resource utilization to reduce pollution. However, the positive contribution of females is undermined by males, who showed a direct correlation with the elevated pollution levels. The effect of population age groups on NO₂ pollution was mixed, with the mid-age group (20–24 and 25–29) being the main cause of increased pollution. These age groups make up a significant proportion and thus must be given greater attention when planning efforts to reduce NO₂ pollution. While the study preferred to maintain the use of several factors as predictors of NO₂ pollution, it is important to note the potential correlation among some variables that could result in overfitted models. This should be explored in future studies to reduce predictors and develop parsimonious models using only significant variables. Alternatively, an approach such as the principal component analysis that reduces predictors by extracting unique information from several variables can be explored. Lastly, the MGWR modelling yielded prediction characteristics that varied across space. Such spatial variation informs location-specific intervention measures as opposed to a generic, national-scale intervention strategy that not only may prove inefficient but also leads to wasteful expenditure of resources.

Acknowledgments

The University of Johannesburg of South Africa provided the necessary resources to conduct this research.

References

1. 1. Turner M.C., Andersen Z.J., Baccarelli A., Diver W.R., Gapstur S.M., Pope C.A III., et al. 2020. Outdoor air pollution and cancer: An overview of the current evidence and public health recommendations. CA: A Cancer Journal for Clinicians, 70(6), pp.460–479. pmid:32964460

* View Article

* PubMed/NCBI

* Google Scholar

2. 2. Lelieveld J., Evans J., Fnais M. and Pozzer A. 2015. The contribution of outdoor air pollution sources to premature mortality on a global scale. Nature, 525, pp.367–371. pmid:26381985

* View Article

* PubMed/NCBI

* Google Scholar

3. 3. Tshehla C. and Wright C.Y. 2019. 15 years after the national environmental management air quality act: Is legislation failing to reduce air pollution in South Africa? South African Journal of Science, 115(9–10), 6100.

* View Article

* Google Scholar

4. 4. Grennfelt P., Engleryd A., Forsius M., Hov Ø., Rodhe H. and Cowling E. 2020. Acid rain and air pollution: 50 years of progress in environmental science and policy. Ambio, 49, pp.849–864. pmid:31542884

* View Article

* PubMed/NCBI

* Google Scholar

5. 5. Lorente A., Boersma K.F., Eskes H.J., Veefkind J.P., Van Geffen J.H.G.M., De Zeeuw M.B., et al. 2019. Quantification of nitrogen oxides emissions from build-up of pollution over Paris with TROPOMI. Scientific Reports, 9(1), 20033. pmid:31882705

* View Article

* PubMed/NCBI

* Google Scholar

6. 6. Tran V.V., Park D. and Lee Y.C. 2020. Indoor air pollution, related human diseases, and recent trends in the control and improvement of indoor air quality. International Journal of Environmental Research and Public Health, 17(8), 2927. pmid:32340311

* View Article

* PubMed/NCBI

* Google Scholar

7. 7. Vîrghileanu M., Săvulescu I., Mihai B.A., Nistor C. and Dobre R. 2020. Nitrogen dioxide (NO2) pollution monitoring with Sentinel-5P satellite imagery over Europe during the coronavirus pandemic outbreak. Remote Sensing, 12(21), 3575.

* View Article

* Google Scholar

8. 8. Jamali S., Klingmyr D. and Tagesson T. 2020. Global-scale patterns and trends in tropospheric NO₂ concentrations, 2005–2018. Remote Sensing, 12(21), 3526.

* View Article

* Google Scholar

9. 9. Oo T.K., Arunrat N., Kongsurakan P., Sereenonchai S. and Wang C. 2021. Nitrogen dioxide (NO₂) level changes during the control of COVID-19 pandemic in Thailand. Aerosol and Air Quality Research, 21(6), 200440.

* View Article

* Google Scholar

10. 10. Sicard P., Agathokleous E., Anenberg S.C., De Marco A., Paoletti E. and Calatayud V. 2023. Trends in urban air pollution over the last two decades: a global perspective. Science of The Total Environment, 858, 160064. pmid:36356738

* View Article

* PubMed/NCBI

* Google Scholar

11. 11. Goldberg D.L., Anenberg S.C., Griffin D., McLinden C.A., Lu Z. and Streets D.G. 2020. Disentangling the impact of the COVID-19 lockdowns on urban NO₂ from natural variability. Geophysical Research Letters, 47(17), e2020GL089269. pmid:32904906

* View Article

* PubMed/NCBI

* Google Scholar

12. 12. Biswal A., Singh V., Singh S., Kesarkar A.P., Ravindra K., Sokhi R.S., et al. 2021. COVID-19 lockdown-induced changes in NO₂ levels across India observed by multi-satellite and surface observations. Atmospheric Chemistry and Physics, 21(6), pp.5235–5251.

* View Article

* Google Scholar

13. 13. Nowak D.J., Crane D.E. and Stevens J.C. 2006. Air pollution removal by urban trees and shrubs in the United States. Urban Forestry and Urban Greening, 4(3–4), pp.115–123.

* View Article

* Google Scholar

14. 14. Syafei, A.D., Irawandani, T.D., Boedisantoso, R., Assomadi, A.F., Slamet, A. and Hermana, J. 2019. The influence of environmental conditions (vegetation, temperature, equator, and elevation) on tropospheric nitrogen dioxide in urban areas in Indonesia. In IOP Conference Series: Earth and Environmental Science 303(1), 012034). IOP Publishing.

15. 15. Gubb C., Blanusa T., Griffiths A. and Pfrang C. 2022. Potted plants can remove the pollutant nitrogen dioxide indoors. Air Quality, Atmosphere and Health, 15(3), pp.479–490.

* View Article

* Google Scholar

16. 16. Khan A., Chatterjee S. and Wang Y. 2021. Urban Heat Island Modeling for Tropical Climates. Netherlands: Elsevier Science.

17. 17. Wang C., Wang T. and Wang P. 2019. The spatial—temporal variation of tropospheric NO₂ over China during 2005 to 2018. Atmosphere, 10(8), 444.

* View Article

* Google Scholar

18. 18. Voiculescu M., Constantin D.E., Condurache-Bota S., Călmuc V., Roșu A. and Dragomir Bălănică C.M. 2020. Role of meteorological parameters in the diurnal and seasonal variation of NO₂ in a Romanian urban environment. International Journal of Environmental Research and Public Health, 17(17), 6228. pmid:32867209

* View Article

* PubMed/NCBI

* Google Scholar

19. 19. Pervez S., Maruyama R., Riaz A. and Nakai S. 2021. Development of land use regression model for seasonal variation of nitrogen dioxide (NO₂) in Lahore, Pakistan. Sustainability, 13(9), 4933.

* View Article

* Google Scholar

20. 20. Swartz J.S., Van Zyl P.G., Beukes J.P., Galy-Lacaux C., Ramandh A. and Pienaar J.J. 2020. Measurement report: statistical modelling of long-term trends of atmospheric inorganic gaseous species within proximity of the pollution hotspot in South Africa. Atmospheric Chemistry and Physics, 20(17), pp.10637–10665.

* View Article

* Google Scholar

21. 21. Faniso, Z., Magimisha, E. and Malatji, T. 2021. Remote sensing of aerosol optical depth (AOD) over Pretoria, South Africa. http://researchspace.csir.co.za/dspace/handle/10204/12419

22. 22. Tomasi C. and Lupi A. 2017. Primary and secondary sources of atmospheric aerosol. Atmospheric Aerosols: Life Cycles and Effects on Air Quality and Climate, pp.1–86.

* View Article

* Google Scholar

23. 23. Lamsal L.N., Martin R.V., Parrish D.D. and Krotkov N.A. 2013. Scaling relationship for NO₂ pollution and urban population size: a satellite perspective. Environmental Science and Technology, 47(14), pp.7855–7861. pmid:23763377

* View Article

* PubMed/NCBI

* Google Scholar

24. 24. Borck R. and Schrauth P. 2021. Population density and urban air quality. Regional Science and Urban Economics, 86, 103596.

* View Article

* Google Scholar

25. 25. Esplugues A., Ballester F., Estarlich M., Llop S., Fuentes V., Mantilla E. et al. 2010. Indoor and outdoor concentrations and determinants of NO2 in a cohort of 1‐year‐old children in Valencia, Spain. Indoor Air, 20(3), pp.213–223. pmid:20408900

* View Article

* PubMed/NCBI

* Google Scholar

26. 26. Karagulian F., Belis C.A., Dora C.F.C., Prüss-Ustün A.M., Bonjour S., Adair-Rohani H. et al. 2015. Contributions to cities’ ambient particulate matter (PM): A systematic review of local source contributions at global level. Atmospheric Environment, 120, pp.475–483.

* View Article

* Google Scholar

27. 27. Sarzynski A. 2012. Bigger is not always better: a comparative analysis of cities and their air pollution impact. Urban Studies, 49(14), pp.3121–3138.

* View Article

* Google Scholar

28. 28. World Health Organization, 2022. Household air pollution. https://www.who.int/news-room/fact-sheets/detail/household-air-pollution-and-health. (Accessed on 15 September 2023)

29. 29. Li C. and Managi S. 2022. Estimating monthly global ground-level NO₂ concentrations using geographically weighted panel regression. Remote Sensing of Environment, 280, 113152.

* View Article

* Google Scholar

30. 30. Sakti A.D., Anggraini T.S., Ihsan K.T.N., et al. (2023). Multi-air pollution risk assessment in Southeast Asia region using integrated remote sensing and socio-economic data products. Science of The Total Environment, 854, 158825. pmid:36116660

* View Article

* PubMed/NCBI

* Google Scholar

31. 31. Olstrup H., Åström C. and Orru H. 2022. Daily mortality in different age groups associated with exposure to particles, nitrogen dioxide and ozone in two northern European capitals: Stockholm and Tallinn. Environments, 9(7), 83.

* View Article

* Google Scholar

32. 32. Barnes B., Mathee A., Thomas E. and Bruce N. 2009. Household energy, indoor air pollution and child respiratory health in South Africa. Journal of Energy in Southern Africa, 20(1), pp.4–13.

* View Article

* Google Scholar

33. 33. Khadija B.U. and Ibrahim M. 2019. Assessment of the Pollution extent of Sulphur Dioxide (SO2) and Nitrogen Dioxide (NO2) in Ambient air within Kano Metropolis, Kano State, Nigeria. Journal of Environmental Science, Computer Science and Engineering & Technology, 8(8), pp.396–404. http://dx.doi.org/10.24214/jecet.A.8.4.39604

* View Article

* Google Scholar

34. 34. Abera A., Friberg J., Isaxon C., Jerrett M., Malmqvist E., Sjöström C., et al. 2021. Air quality in Africa: Public health implications. Annual Review of Public Health, 42, pp.193–210. pmid:33348996

* View Article

* PubMed/NCBI

* Google Scholar

35. 35. Fisher S., Bellinger D.C., Cropper M.L., Kumar P., Binagwaho A., Koudenoukpo J.B., et al. 2021. Air pollution and development in Africa: impacts on health, the economy, and human capital. The Lancet Planetary Health, 5(10), pp.e681–e688. pmid:34627472

* View Article

* PubMed/NCBI

* Google Scholar

36. 36. Ahmad N.A., Ismail N.W., Ahmad Sidique S.F. and Mazlan N.S. 2021. Air pollution effects on adult mortality rate in developing countries. Environmental Science and Pollution Research, 28, pp.8709–8721. pmid:33068244

* View Article

* PubMed/NCBI

* Google Scholar

37. 37. Zhu Y., Zhan Y., Wang B., Li Z., Qin Y. and Zhang K. 2019. Spatiotemporally mapping of the relationship between NO₂ pollution and urbanization for a megacity in Southwest China during 2005–2016. Chemosphere, 220, pp.155–162. pmid:30583207

* View Article

* PubMed/NCBI

* Google Scholar

38. 38. Wang Q. and Su M. 2020. A preliminary assessment of the impact of COVID-19 on environment—a case study of China. Science of the Total Environment, 728, 138915. pmid:32348946

* View Article

* PubMed/NCBI

* Google Scholar

39. 39. Shikwambana L. and Kganyago M. 2021. Assessing the responses of aviation-related SO₂ and NO₂ emissions to COVID-19 lockdown regulations in South Africa. Remote Sensing, 13(20), 4156.

* View Article

* Google Scholar

40. 40. Matandirotya N.R. and Burger R. 2023. An assessment of NO₂ atmospheric air pollution over three cities in South Africa during 2020 COVID-19 pandemic. Air Quality, Atmosphere and Health, 16(2), pp.263–276. pmid:36281221

* View Article

* PubMed/NCBI

* Google Scholar

41. 41. Zhu H. and Yang L. 2023. Formation mechanism of NO₂ distribution heterogeneity at different spatial scales. Resources, Environment and Sustainability, 12, 100106.

* View Article

* Google Scholar

42. 42. Fairburn J., Schüle S.A., Dreger S., Karla Hilz L. and Bolte G. 2019. Social inequalities in exposure to ambient air pollution: a systematic review in the WHO European region. International Journal of Environmental Research and Public Health, 16(17), 3127. pmid:31466272

* View Article

* PubMed/NCBI

* Google Scholar

43. 43. Morosele I.P. and Langerman K.E. 2020. The impacts of commissioning coal-fired power stations on air quality in South Africa: insights from ambient monitoring stations. Clean Air Journal, 30(2), pp.1–11.

* View Article

* Google Scholar

44. 44. Baek J.I. and Ban Y.U. 2020. The impacts of urban air pollution emission density on air pollutant concentration based on a panel model. Sustainability, 12(20), 8401.

* View Article

* Google Scholar

45. 45. Haywood L.K., Kapwata T., Oelofse S., Breetzke G. and Wright C.Y. 2021. Waste disposal practices in low-income settlements of South Africa. International Journal of Environmental Research and Public Health, 18(15), 8176. pmid:34360468

* View Article

* PubMed/NCBI

* Google Scholar

46. 46. Kai R.F., Scholes M.C., Piketh S.J. and Scholes R.J. 2022. Analysis of the first surface nitrogen dioxide concentration observations over the South African Highveld derived from the Pandora-2s instrument. Clean Air Journal, 32(1), pp.1–11.

* View Article

* Google Scholar

47. 47. Shikwambana L., Mhangara P. and Mbatha N. 2020. Trend analysis and first-time observations of sulphur dioxide and nitrogen dioxide in South Africa using TROPOMI/Sentinel-5P data. International Journal of Applied Earth Observation and Geoinformation, 91, 102130.

* View Article

* Google Scholar

48. 48. Heunis, S. and Dekenah, M. 2014. Manual for Eskom Distribution Pre-Electrification Tool (DPET). Eskom, EOH Enerweb. https://zivahub.uct.ac.za/ndownloader/files/13347065. (Accessed on 13 August 2023)

49. 49. Harris T., Collinson M. and Wittenberg M. 2017. Aiming for a moving target: the dynamics of household electricity connections in a developing context. World Development, 97, pp.14–26.

* View Article

* Google Scholar

50. 50. Ismail Z. and Khembo P. 2015. Determinants of energy poverty in South Africa. Journal of Energy in Southern Africa, 26(3), pp.66–78.

* View Article

* Google Scholar

51. 51. Wernecke B., Langerman K.E., Howard A.I. and Wright C.Y. 2024. Fuel switching and energy stacking in low-income households in South Africa: A review with recommendations for household air pollution exposure research. Energy Research and Social Science, 109, 103415.

* View Article

* Google Scholar

52. 52. STATISTICS SA. 2010. Water Management Areas in South Africa. https://www.statssa.gov.za/publications/D04058/D04058.pdf. (Accessed on 27 August 2023)

53. 53. Municipal Demarcation Board, 2016. Municipal boundary demarcation process: a process map for the determination and re-determination of municipal boundaries. https://www.demarcation.org.za/wp-content/uploads/2021/06/MUNICIPAL-BOUNDARY-DEMARCATION-PROCESS.pdf. (Accessed on 28 August 2023)

54. 54. STATISTICS SA. 2022. Mid-year population estimates 2022. https://www.statssa.gov.za/?p=15589. (Accessed on 15 August 2022)

55. 55. Mashamaite G. and Lethoko M. 2018. Role of the South African local government in local economic development. International Journal of eBusiness and eGovernment Studies, 10(1), pp. 114–128. https://dergipark.org.tr/en/pub/ijebeg/issue/36107/535456

* View Article

* Google Scholar

56. 56. Eskes, H., van Geffen, J., Boersma, F., Eichmann, K.U., Apituley, A., Pedergnana, M., et al. 2019. Sentinel-5 precursor/TROPOMI Level 2 product user manual nitrogen dioxide. Ministry of Infrastructure and Water Management. https://sentinel.esa.int/documents/247904/2474726/Sentinel-5P-Level-2-Product-User-Manual-Nitrogen-Dioxide.pdf. (Accessed on 13 August 2023)

57. 57. Didan, K., Munoz, A.B., Solano, R. and Huete, H. 2015. MODIS vegetation index User’s Guide. Version 3.00, Collection 6. https://vip.arizona.edu/documents/MODIS/MODIS_VI_UsersGuide_June_2015_C6.pdf. (Accessed on 13 August 2023)

58. 58. Wan, Z. 2013. Collection-6 MODIS land-surface temperature products user’s guide. https://lpdaac.usgs.gov/documents/118/MOD11_User_Guide_V6.pdf.

59. 59. Lyapustin A., Wang Y., Korkin S. and Huang D. 2018. MODIS collection 6 MAIAC algorithm. Atmospheric Measurement Techniques, 11(10), pp.5741–5765.

* View Article

* Google Scholar

60. 60. Western Cape Government, 2015. Municipal human settlement demand profile: Breede Valley Local Municipality. https://www.westerncape.gov.za/assets/departments/human-settlements/docs/demand-profile-analysis/human_settlements_municipal_profile_and_analysis_breede_valley_20150605.pdf. (Accessed on 6 August 2023)

61. 61. Israel-Akinbo S., Snowball J. and Fraser G. 2018. The energy transition patterns of low-income households in South Africa: An evaluation of energy programme and policy. Journal of Energy in Southern Africa, 29(3), pp.75–85. http://dx.doi.org/10.17159/2413-3051/2017/v29i3a3310.

* View Article

* Google Scholar

62. 62. Krivoruchko K., Gribov A. and Krause E. 2011. Multivariate areal interpolation for continuous and count data. Procedia Environmental Sciences, 3, pp.14–19.

* View Article

* Google Scholar

63. 63. Mitchell, A. 2005. The ESRI Guide to GIS Analysis, vol. 2. Redlands. https://www.esri.com/en-us/esri-press/browse/the-esri-guide-to-gis-analysis-volume-2-spatial-measurements-and-statistics-second-edition. (Accessed on 24 June 2024)

64. 64. Fotheringham A.S., Charlton M.E. and Brunsdon C. 1998. Geographically weighted regression: a natural evolution of the expansion method for spatial data analysis. Environment and Planning A, 30(11), pp.1905–1927.

* View Article

* Google Scholar

65. 65. Tobler W.R. 1970. A computer movie simulating urban growth in the Detroit region. Economic Geography, 46(sup1), pp.234–240.

* View Article

* Google Scholar

66. 66. Shen Y., de Hoogh K., Schmitz O., Clinton N., Tuxen-Bettman K., Brandt J., et al. 2022. Europe-wide air pollution modeling from 2000 to 2019 using geographically weighted regression. Environment International, 168, 107485. pmid:36030744

* View Article

* PubMed/NCBI

* Google Scholar

67. 67. Zhang Y., Shi M., Chen J., Fu S. and Wang H. 2023. Spatiotemporal variations of NO₂ and its driving factors in the coastal ports of China. Science of The Total Environment, 871, 162041. pmid:36754320

* View Article

* PubMed/NCBI

* Google Scholar

68. 68. Li, Z. 2020. Multiscale Geographically Weighted Regression Computation, Inference, and Application (Doctoral dissertation). https://keep.lib.asu.edu/items/158516

69. 69. Fotheringham A.S., Yang W. and Kang W. 2017. Multiscale geographically weighted regression (MGWR). Annals of the American Association of Geographers, 107(6), pp.1247–1265.

* View Article

* Google Scholar

70. 70. Moran P.A.P. 1950. Notes on continuous stochastic phenomena. Biometrika, 37(1), pp.17–23. pmid:15420245

* View Article

* PubMed/NCBI

* Google Scholar

71. 71. Beck H.E., Zimmermann N.E., McVicar T.R., Vergopolan N., Berg A. and Wood E.F. 2018. Present and future Köppen-Geiger climate classification maps at 1-km resolution. Scientific Data, 5, 180214. pmid:30375988

* View Article

* PubMed/NCBI

* Google Scholar

72. 72. Chawala P., Shanmuga P.R. and Shiva N.S.M. 2023. Climatology and landscape determinants of AOD, SO2 and NO2 over Indo-Gangetic Plain. Environmental Research, 220, 115125. pmid:36592806

* View Article

* PubMed/NCBI

* Google Scholar

73. 73. Wolfe M., McDonald N., Arunachalam S. and Valencia A. 2017. Air pollution exposure during school commutes. Journal of Transport and Health, 5, pp.S48–S49. https://icth.confex.com/icth/2017/mediafile/Handout/Paper2073/Exposures%20Slides%20ICTH%206.27%20handout.pdf. (Accessed on 19 July 2023)

* View Article

* Google Scholar

74. 74. Utembe W., Faustman E.M., Matatiele P. and Gulumian M. 2015. Hazards identified and the need for health risk assessment in the South African mining industry. Human and Experimental Toxicology, 34(12), pp.1212–1221. pmid:26614808

* View Article

* PubMed/NCBI

* Google Scholar

75. 75. Ratshomo, K. and Nembahe, R. 2018. Directorate: energy data collection, management and analysis. South African Energy Department, Technical Report. https://www.energy.gov.za/files/media/explained/south-african-coal-sector-report.pdf. (Accessed on 06 September 2023)

76. 76. Arowosegbe O.O., Röösli M., Künzli N., Saucy A., Adebayo-Ojo T.C., Schwartz J., et al. 2022. Ensemble averaging using remote sensing data to model spatiotemporal PM10 concentrations in sparsely monitored South Africa. Environmental Pollution, 310, 119883. pmid:35932898

* View Article

* PubMed/NCBI

* Google Scholar

77. 77. Larkin A., Geddes J.A., Martin R.V., Xiao Q., Liu Y., Marshall J.D., et al. 2017. Global land use regression model for nitrogen dioxide air pollution. Environmental Science and Technology, 51(12), pp.6957–6964. pmid:28520422

* View Article

* PubMed/NCBI

* Google Scholar

78. 78. National Department of Health, 2019. Guideline for the management of domestic indoor air quality a guide for environmental health practitioners in South Africa June 2019. https://www.health.gov.za/wp-content/uploads/2022/09/DOH-Approved-Guideline-Management-of-Domestic-Indoor-Air-Quality-Final.pdf. (Accessed on 13 September 2023)

79. 79. Sarkodie S.A. and Adams S. 2020. Electricity access and income inequality in South Africa: evidence from Bayesian and NARDL analyses. Energy Strategy Reviews, 29, 100480.

* View Article

* Google Scholar

80. 80. Mujere J. 2020. Unemployment, service delivery and practices of waiting in South Africa’s informal settlements. Critical African Studies, 12(1), pp.65–78.

* View Article

* Google Scholar

81. 81. Dietz T., Kalof L. and Stern P.C. 2002. Gender, values, and environmentalism. Social Science Quarterly, 83(1), pp.353–364.

* View Article

* Google Scholar

82. 82. McCright A.M. 2010. The effects of gender on climate change knowledge and concern in the American public. Population and Environment, 32, pp.66–87.

* View Article

* Google Scholar

83. 83. Liu G., Sun B., Yu L., Chen J., Han B., Li Y. et al. 2020. The gender-based differences in vulnerability to ambient air pollution and cerebrovascular disease mortality: Evidences based on 26781 deaths. Global Heart, 15(1), 46. pmid:32923340

* View Article

* PubMed/NCBI

* Google Scholar

84. 84. Ryu J., Park C. and Jeon S.W. 2019. Mapping and statistical analysis of NO₂ concentration for local government air quality regulation. Sustainability, 11(14), 3809. http://dx.doi.org/10.3390/su11143809

* View Article

* Google Scholar

85. 85. Liddle B. and Lung S. 2010. Age-structure, urbanization, and climate change in developed countries: revisiting STIRPAT for disaggregated population and consumption-related environmental impacts. Population and Environment, 31, pp.317–343.

* View Article

* Google Scholar

86. 86. Tarazkar M.H., Dehbidi N.K., Ozturk I. and Al-Mulali U. 2021. The impact of age structure on carbon emission in the Middle East: the panel autoregressive distributed lag approach. Environmental Science and Pollution Research, 28, pp.33722–33734. pmid:32314289

* View Article

* PubMed/NCBI

* Google Scholar

87. 87. Estiri H. and Zagheni E. 2019. Age matters: Ageing and household energy demand in the United States. Energy Research and Social Science, 55, pp.62–70.

* View Article

* Google Scholar

88. 88. UN General Assembly, 2015. Transforming our world: the 2030 Agenda for Sustainable Development, 21 October 2015, A/RES/70/1, https://www.refworld.org/docid/57b6e3e44.html. (Accessed on 15 September 2023)

89. 89. Li Y., Sha Z., Tang A., Goulding K. and Liu X. 2023. The application of machine learning to air pollution research: A bibliometric analysis. Ecotoxicology and Environmental Safety, 257, 114911. pmid:37154080

* View Article

* PubMed/NCBI

* Google Scholar

Citation: Hlatshwayo SN, Tesfamichael SG, Kganyago M (2024) Predicting tropospheric nitrogen dioxide column density in South African municipalities using socio-environmental variables and Multiscale Geographically Weighted Regression. PLoS ONE 19(8): e0308484. https://doi.org/10.1371/journal.pone.0308484

About the Authors:

Sphamandla N. Hlatshwayo

Roles: Conceptualization, Formal analysis, Investigation, Methodology, Writing – original draft

Affiliation: Department of Geography, Environmental Management and Energy Studies, University of Johannesburg, Johannesburg, South Africa

Solomon G. Tesfamichael

Roles: Conceptualization, Project administration, Resources, Software, Supervision, Validation, Writing – review & editing

E-mail: [email protected]

Affiliation: Department of Geography, Environmental Management and Energy Studies, University of Johannesburg, Johannesburg, South Africa

ORICD: https://orcid.org/0000-0002-4754-5732

Mahlatse Kganyago

Roles: Supervision, Validation, Writing – review & editing

Affiliation: Department of Geography, Environmental Management and Energy Studies, University of Johannesburg, Johannesburg, South Africa

ORICD: https://orcid.org/0000-0001-9553-0378

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

[/RAW_REF_TEXT]

References

1. Turner M.C., Andersen Z.J., Baccarelli A., Diver W.R., Gapstur S.M., Pope C.A III., et al. 2020. Outdoor air pollution and cancer: An overview of the current evidence and public health recommendations. CA: A Cancer Journal for Clinicians, 70(6), pp.460–479. pmid:32964460

2. Lelieveld J., Evans J., Fnais M. and Pozzer A. 2015. The contribution of outdoor air pollution sources to premature mortality on a global scale. Nature, 525, pp.367–371. pmid:26381985

3. Tshehla C. and Wright C.Y. 2019. 15 years after the national environmental management air quality act: Is legislation failing to reduce air pollution in South Africa? South African Journal of Science, 115(9–10), 6100.

4. Grennfelt P., Engleryd A., Forsius M., Hov Ø., Rodhe H. and Cowling E. 2020. Acid rain and air pollution: 50 years of progress in environmental science and policy. Ambio, 49, pp.849–864. pmid:31542884

5. Lorente A., Boersma K.F., Eskes H.J., Veefkind J.P., Van Geffen J.H.G.M., De Zeeuw M.B., et al. 2019. Quantification of nitrogen oxides emissions from build-up of pollution over Paris with TROPOMI. Scientific Reports, 9(1), 20033. pmid:31882705

6. Tran V.V., Park D. and Lee Y.C. 2020. Indoor air pollution, related human diseases, and recent trends in the control and improvement of indoor air quality. International Journal of Environmental Research and Public Health, 17(8), 2927. pmid:32340311

7. Vîrghileanu M., Săvulescu I., Mihai B.A., Nistor C. and Dobre R. 2020. Nitrogen dioxide (NO2) pollution monitoring with Sentinel-5P satellite imagery over Europe during the coronavirus pandemic outbreak. Remote Sensing, 12(21), 3575.

8. Jamali S., Klingmyr D. and Tagesson T. 2020. Global-scale patterns and trends in tropospheric NO2 concentrations, 2005–2018. Remote Sensing, 12(21), 3526.

9. Oo T.K., Arunrat N., Kongsurakan P., Sereenonchai S. and Wang C. 2021. Nitrogen dioxide (NO2) level changes during the control of COVID-19 pandemic in Thailand. Aerosol and Air Quality Research, 21(6), 200440.

10. Sicard P., Agathokleous E., Anenberg S.C., De Marco A., Paoletti E. and Calatayud V. 2023. Trends in urban air pollution over the last two decades: a global perspective. Science of The Total Environment, 858, 160064. pmid:36356738

11. Goldberg D.L., Anenberg S.C., Griffin D., McLinden C.A., Lu Z. and Streets D.G. 2020. Disentangling the impact of the COVID-19 lockdowns on urban NO2 from natural variability. Geophysical Research Letters, 47(17), e2020GL089269. pmid:32904906

12. Biswal A., Singh V., Singh S., Kesarkar A.P., Ravindra K., Sokhi R.S., et al. 2021. COVID-19 lockdown-induced changes in NO2 levels across India observed by multi-satellite and surface observations. Atmospheric Chemistry and Physics, 21(6), pp.5235–5251.

13. Nowak D.J., Crane D.E. and Stevens J.C. 2006. Air pollution removal by urban trees and shrubs in the United States. Urban Forestry and Urban Greening, 4(3–4), pp.115–123.

14. Syafei, A.D., Irawandani, T.D., Boedisantoso, R., Assomadi, A.F., Slamet, A. and Hermana, J. 2019. The influence of environmental conditions (vegetation, temperature, equator, and elevation) on tropospheric nitrogen dioxide in urban areas in Indonesia. In IOP Conference Series: Earth and Environmental Science 303(1), 012034). IOP Publishing.

15. Gubb C., Blanusa T., Griffiths A. and Pfrang C. 2022. Potted plants can remove the pollutant nitrogen dioxide indoors. Air Quality, Atmosphere and Health, 15(3), pp.479–490.

16. Khan A., Chatterjee S. and Wang Y. 2021. Urban Heat Island Modeling for Tropical Climates. Netherlands: Elsevier Science.

17. Wang C., Wang T. and Wang P. 2019. The spatial—temporal variation of tropospheric NO2 over China during 2005 to 2018. Atmosphere, 10(8), 444.

18. Voiculescu M., Constantin D.E., Condurache-Bota S., Călmuc V., Roșu A. and Dragomir Bălănică C.M. 2020. Role of meteorological parameters in the diurnal and seasonal variation of NO2 in a Romanian urban environment. International Journal of Environmental Research and Public Health, 17(17), 6228. pmid:32867209

19. Pervez S., Maruyama R., Riaz A. and Nakai S. 2021. Development of land use regression model for seasonal variation of nitrogen dioxide (NO2) in Lahore, Pakistan. Sustainability, 13(9), 4933.

20. Swartz J.S., Van Zyl P.G., Beukes J.P., Galy-Lacaux C., Ramandh A. and Pienaar J.J. 2020. Measurement report: statistical modelling of long-term trends of atmospheric inorganic gaseous species within proximity of the pollution hotspot in South Africa. Atmospheric Chemistry and Physics, 20(17), pp.10637–10665.

21. Faniso, Z., Magimisha, E. and Malatji, T. 2021. Remote sensing of aerosol optical depth (AOD) over Pretoria, South Africa. http://researchspace.csir.co.za/dspace/handle/10204/12419

22. Tomasi C. and Lupi A. 2017. Primary and secondary sources of atmospheric aerosol. Atmospheric Aerosols: Life Cycles and Effects on Air Quality and Climate, pp.1–86.

23. Lamsal L.N., Martin R.V., Parrish D.D. and Krotkov N.A. 2013. Scaling relationship for NO2 pollution and urban population size: a satellite perspective. Environmental Science and Technology, 47(14), pp.7855–7861. pmid:23763377

24. Borck R. and Schrauth P. 2021. Population density and urban air quality. Regional Science and Urban Economics, 86, 103596.

25. Esplugues A., Ballester F., Estarlich M., Llop S., Fuentes V., Mantilla E. et al. 2010. Indoor and outdoor concentrations and determinants of NO2 in a cohort of 1‐year‐old children in Valencia, Spain. Indoor Air, 20(3), pp.213–223. pmid:20408900

26. Karagulian F., Belis C.A., Dora C.F.C., Prüss-Ustün A.M., Bonjour S., Adair-Rohani H. et al. 2015. Contributions to cities’ ambient particulate matter (PM): A systematic review of local source contributions at global level. Atmospheric Environment, 120, pp.475–483.

27. Sarzynski A. 2012. Bigger is not always better: a comparative analysis of cities and their air pollution impact. Urban Studies, 49(14), pp.3121–3138.

28. World Health Organization, 2022. Household air pollution. https://www.who.int/news-room/fact-sheets/detail/household-air-pollution-and-health. (Accessed on 15 September 2023)

29. Li C. and Managi S. 2022. Estimating monthly global ground-level NO2 concentrations using geographically weighted panel regression. Remote Sensing of Environment, 280, 113152.

30. Sakti A.D., Anggraini T.S., Ihsan K.T.N., et al. (2023). Multi-air pollution risk assessment in Southeast Asia region using integrated remote sensing and socio-economic data products. Science of The Total Environment, 854, 158825. pmid:36116660

31. Olstrup H., Åström C. and Orru H. 2022. Daily mortality in different age groups associated with exposure to particles, nitrogen dioxide and ozone in two northern European capitals: Stockholm and Tallinn. Environments, 9(7), 83.

32. Barnes B., Mathee A., Thomas E. and Bruce N. 2009. Household energy, indoor air pollution and child respiratory health in South Africa. Journal of Energy in Southern Africa, 20(1), pp.4–13.

33. Khadija B.U. and Ibrahim M. 2019. Assessment of the Pollution extent of Sulphur Dioxide (SO2) and Nitrogen Dioxide (NO2) in Ambient air within Kano Metropolis, Kano State, Nigeria. Journal of Environmental Science, Computer Science and Engineering & Technology, 8(8), pp.396–404. http://dx.doi.org/10.24214/jecet.A.8.4.39604

34. Abera A., Friberg J., Isaxon C., Jerrett M., Malmqvist E., Sjöström C., et al. 2021. Air quality in Africa: Public health implications. Annual Review of Public Health, 42, pp.193–210. pmid:33348996

35. Fisher S., Bellinger D.C., Cropper M.L., Kumar P., Binagwaho A., Koudenoukpo J.B., et al. 2021. Air pollution and development in Africa: impacts on health, the economy, and human capital. The Lancet Planetary Health, 5(10), pp.e681–e688. pmid:34627472

36. Ahmad N.A., Ismail N.W., Ahmad Sidique S.F. and Mazlan N.S. 2021. Air pollution effects on adult mortality rate in developing countries. Environmental Science and Pollution Research, 28, pp.8709–8721. pmid:33068244

37. Zhu Y., Zhan Y., Wang B., Li Z., Qin Y. and Zhang K. 2019. Spatiotemporally mapping of the relationship between NO2 pollution and urbanization for a megacity in Southwest China during 2005–2016. Chemosphere, 220, pp.155–162. pmid:30583207

38. Wang Q. and Su M. 2020. A preliminary assessment of the impact of COVID-19 on environment—a case study of China. Science of the Total Environment, 728, 138915. pmid:32348946

39. Shikwambana L. and Kganyago M. 2021. Assessing the responses of aviation-related SO2 and NO2 emissions to COVID-19 lockdown regulations in South Africa. Remote Sensing, 13(20), 4156.

40. Matandirotya N.R. and Burger R. 2023. An assessment of NO2 atmospheric air pollution over three cities in South Africa during 2020 COVID-19 pandemic. Air Quality, Atmosphere and Health, 16(2), pp.263–276. pmid:36281221

41. Zhu H. and Yang L. 2023. Formation mechanism of NO2 distribution heterogeneity at different spatial scales. Resources, Environment and Sustainability, 12, 100106.

42. Fairburn J., Schüle S.A., Dreger S., Karla Hilz L. and Bolte G. 2019. Social inequalities in exposure to ambient air pollution: a systematic review in the WHO European region. International Journal of Environmental Research and Public Health, 16(17), 3127. pmid:31466272

43. Morosele I.P. and Langerman K.E. 2020. The impacts of commissioning coal-fired power stations on air quality in South Africa: insights from ambient monitoring stations. Clean Air Journal, 30(2), pp.1–11.

44. Baek J.I. and Ban Y.U. 2020. The impacts of urban air pollution emission density on air pollutant concentration based on a panel model. Sustainability, 12(20), 8401.

45. Haywood L.K., Kapwata T., Oelofse S., Breetzke G. and Wright C.Y. 2021. Waste disposal practices in low-income settlements of South Africa. International Journal of Environmental Research and Public Health, 18(15), 8176. pmid:34360468

46. Kai R.F., Scholes M.C., Piketh S.J. and Scholes R.J. 2022. Analysis of the first surface nitrogen dioxide concentration observations over the South African Highveld derived from the Pandora-2s instrument. Clean Air Journal, 32(1), pp.1–11.

47. Shikwambana L., Mhangara P. and Mbatha N. 2020. Trend analysis and first-time observations of sulphur dioxide and nitrogen dioxide in South Africa using TROPOMI/Sentinel-5P data. International Journal of Applied Earth Observation and Geoinformation, 91, 102130.

48. Heunis, S. and Dekenah, M. 2014. Manual for Eskom Distribution Pre-Electrification Tool (DPET). Eskom, EOH Enerweb. https://zivahub.uct.ac.za/ndownloader/files/13347065. (Accessed on 13 August 2023)

49. Harris T., Collinson M. and Wittenberg M. 2017. Aiming for a moving target: the dynamics of household electricity connections in a developing context. World Development, 97, pp.14–26.

50. Ismail Z. and Khembo P. 2015. Determinants of energy poverty in South Africa. Journal of Energy in Southern Africa, 26(3), pp.66–78.

51. Wernecke B., Langerman K.E., Howard A.I. and Wright C.Y. 2024. Fuel switching and energy stacking in low-income households in South Africa: A review with recommendations for household air pollution exposure research. Energy Research and Social Science, 109, 103415.

52. STATISTICS SA. 2010. Water Management Areas in South Africa. https://www.statssa.gov.za/publications/D04058/D04058.pdf. (Accessed on 27 August 2023)

53. Municipal Demarcation Board, 2016. Municipal boundary demarcation process: a process map for the determination and re-determination of municipal boundaries. https://www.demarcation.org.za/wp-content/uploads/2021/06/MUNICIPAL-BOUNDARY-DEMARCATION-PROCESS.pdf. (Accessed on 28 August 2023)

54. STATISTICS SA. 2022. Mid-year population estimates 2022. https://www.statssa.gov.za/?p=15589. (Accessed on 15 August 2022)

55. Mashamaite G. and Lethoko M. 2018. Role of the South African local government in local economic development. International Journal of eBusiness and eGovernment Studies, 10(1), pp. 114–128. https://dergipark.org.tr/en/pub/ijebeg/issue/36107/535456

56. Eskes, H., van Geffen, J., Boersma, F., Eichmann, K.U., Apituley, A., Pedergnana, M., et al. 2019. Sentinel-5 precursor/TROPOMI Level 2 product user manual nitrogen dioxide. Ministry of Infrastructure and Water Management. https://sentinel.esa.int/documents/247904/2474726/Sentinel-5P-Level-2-Product-User-Manual-Nitrogen-Dioxide.pdf. (Accessed on 13 August 2023)

57. Didan, K., Munoz, A.B., Solano, R. and Huete, H. 2015. MODIS vegetation index User’s Guide. Version 3.00, Collection 6. https://vip.arizona.edu/documents/MODIS/MODIS_VI_UsersGuide_June_2015_C6.pdf. (Accessed on 13 August 2023)

58. Wan, Z. 2013. Collection-6 MODIS land-surface temperature products user’s guide. https://lpdaac.usgs.gov/documents/118/MOD11_User_Guide_V6.pdf.

59. Lyapustin A., Wang Y., Korkin S. and Huang D. 2018. MODIS collection 6 MAIAC algorithm. Atmospheric Measurement Techniques, 11(10), pp.5741–5765.

60. Western Cape Government, 2015. Municipal human settlement demand profile: Breede Valley Local Municipality. https://www.westerncape.gov.za/assets/departments/human-settlements/docs/demand-profile-analysis/human_settlements_municipal_profile_and_analysis_breede_valley_20150605.pdf. (Accessed on 6 August 2023)

61. Israel-Akinbo S., Snowball J. and Fraser G. 2018. The energy transition patterns of low-income households in South Africa: An evaluation of energy programme and policy. Journal of Energy in Southern Africa, 29(3), pp.75–85. http://dx.doi.org/10.17159/2413-3051/2017/v29i3a3310.

62. Krivoruchko K., Gribov A. and Krause E. 2011. Multivariate areal interpolation for continuous and count data. Procedia Environmental Sciences, 3, pp.14–19.

63. Mitchell, A. 2005. The ESRI Guide to GIS Analysis, vol. 2. Redlands. https://www.esri.com/en-us/esri-press/browse/the-esri-guide-to-gis-analysis-volume-2-spatial-measurements-and-statistics-second-edition. (Accessed on 24 June 2024)

64. Fotheringham A.S., Charlton M.E. and Brunsdon C. 1998. Geographically weighted regression: a natural evolution of the expansion method for spatial data analysis. Environment and Planning A, 30(11), pp.1905–1927.

65. Tobler W.R. 1970. A computer movie simulating urban growth in the Detroit region. Economic Geography, 46(sup1), pp.234–240.

66. Shen Y., de Hoogh K., Schmitz O., Clinton N., Tuxen-Bettman K., Brandt J., et al. 2022. Europe-wide air pollution modeling from 2000 to 2019 using geographically weighted regression. Environment International, 168, 107485. pmid:36030744

67. Zhang Y., Shi M., Chen J., Fu S. and Wang H. 2023. Spatiotemporal variations of NO2 and its driving factors in the coastal ports of China. Science of The Total Environment, 871, 162041. pmid:36754320

68. Li, Z. 2020. Multiscale Geographically Weighted Regression Computation, Inference, and Application (Doctoral dissertation). https://keep.lib.asu.edu/items/158516

69. Fotheringham A.S., Yang W. and Kang W. 2017. Multiscale geographically weighted regression (MGWR). Annals of the American Association of Geographers, 107(6), pp.1247–1265.

70. Moran P.A.P. 1950. Notes on continuous stochastic phenomena. Biometrika, 37(1), pp.17–23. pmid:15420245

71. Beck H.E., Zimmermann N.E., McVicar T.R., Vergopolan N., Berg A. and Wood E.F. 2018. Present and future Köppen-Geiger climate classification maps at 1-km resolution. Scientific Data, 5, 180214. pmid:30375988

72. Chawala P., Shanmuga P.R. and Shiva N.S.M. 2023. Climatology and landscape determinants of AOD, SO2 and NO2 over Indo-Gangetic Plain. Environmental Research, 220, 115125. pmid:36592806

73. Wolfe M., McDonald N., Arunachalam S. and Valencia A. 2017. Air pollution exposure during school commutes. Journal of Transport and Health, 5, pp.S48–S49. https://icth.confex.com/icth/2017/mediafile/Handout/Paper2073/Exposures%20Slides%20ICTH%206.27%20handout.pdf. (Accessed on 19 July 2023)

74. Utembe W., Faustman E.M., Matatiele P. and Gulumian M. 2015. Hazards identified and the need for health risk assessment in the South African mining industry. Human and Experimental Toxicology, 34(12), pp.1212–1221. pmid:26614808

75. Ratshomo, K. and Nembahe, R. 2018. Directorate: energy data collection, management and analysis. South African Energy Department, Technical Report. https://www.energy.gov.za/files/media/explained/south-african-coal-sector-report.pdf. (Accessed on 06 September 2023)

76. Arowosegbe O.O., Röösli M., Künzli N., Saucy A., Adebayo-Ojo T.C., Schwartz J., et al. 2022. Ensemble averaging using remote sensing data to model spatiotemporal PM10 concentrations in sparsely monitored South Africa. Environmental Pollution, 310, 119883. pmid:35932898

77. Larkin A., Geddes J.A., Martin R.V., Xiao Q., Liu Y., Marshall J.D., et al. 2017. Global land use regression model for nitrogen dioxide air pollution. Environmental Science and Technology, 51(12), pp.6957–6964. pmid:28520422

78. National Department of Health, 2019. Guideline for the management of domestic indoor air quality a guide for environmental health practitioners in South Africa June 2019. https://www.health.gov.za/wp-content/uploads/2022/09/DOH-Approved-Guideline-Management-of-Domestic-Indoor-Air-Quality-Final.pdf. (Accessed on 13 September 2023)

79. Sarkodie S.A. and Adams S. 2020. Electricity access and income inequality in South Africa: evidence from Bayesian and NARDL analyses. Energy Strategy Reviews, 29, 100480.

80. Mujere J. 2020. Unemployment, service delivery and practices of waiting in South Africa’s informal settlements. Critical African Studies, 12(1), pp.65–78.

81. Dietz T., Kalof L. and Stern P.C. 2002. Gender, values, and environmentalism. Social Science Quarterly, 83(1), pp.353–364.

82. McCright A.M. 2010. The effects of gender on climate change knowledge and concern in the American public. Population and Environment, 32, pp.66–87.

83. Liu G., Sun B., Yu L., Chen J., Han B., Li Y. et al. 2020. The gender-based differences in vulnerability to ambient air pollution and cerebrovascular disease mortality: Evidences based on 26781 deaths. Global Heart, 15(1), 46. pmid:32923340

84. Ryu J., Park C. and Jeon S.W. 2019. Mapping and statistical analysis of NO2 concentration for local government air quality regulation. Sustainability, 11(14), 3809. http://dx.doi.org/10.3390/su11143809

85. Liddle B. and Lung S. 2010. Age-structure, urbanization, and climate change in developed countries: revisiting STIRPAT for disaggregated population and consumption-related environmental impacts. Population and Environment, 31, pp.317–343.

86. Tarazkar M.H., Dehbidi N.K., Ozturk I. and Al-Mulali U. 2021. The impact of age structure on carbon emission in the Middle East: the panel autoregressive distributed lag approach. Environmental Science and Pollution Research, 28, pp.33722–33734. pmid:32314289

87. Estiri H. and Zagheni E. 2019. Age matters: Ageing and household energy demand in the United States. Energy Research and Social Science, 55, pp.62–70.

88. UN General Assembly, 2015. Transforming our world: the 2030 Agenda for Sustainable Development, 21 October 2015, A/RES/70/1, https://www.refworld.org/docid/57b6e3e44.html. (Accessed on 15 September 2023)

89. Li Y., Sha Z., Tang A., Goulding K. and Liu X. 2023. The application of machine learning to air pollution research: A bibliometric analysis. Ecotoxicology and Environmental Safety, 257, 114911. pmid:37154080