Assessing the impact of Eskom power plant emissions on ambient air quality over KwaZamokuhle

Introduction Natural processes and anthropogenic activities release pollutants into the atmosphere which can be hazardous to the health of humans and the environment alike (Patrick et al. 2015). Although emissions occur at a local scale, it is possible for the impact to be observed globally; for example, emissions of carbon dioxide (CO2). This makes atmospheric emissions an important concern that requires attention from the local to the global scale (Pretorius 2015). Emitted pollutants are exposed to different processes in the atmosphere (turbulence caused by convection and stability, wind speed and direction, mixing height and temperature) that control their dispersion and transportation. Finally, the pollutants are transformed over time in the atmosphere to form secondary pollutants. Ultimately, the surface concentration of these trace gases and particulates determine the local air quality of an area (Hunter et al. 2002).

total emissions of criteria pollutants from South Africa's coalfired power plants are higher than all the European countries considered.
About 80% of the country's coal reserves are located in the Highveld area; the industrial epicentre of the country. This area houses 11 of the 13 Eskom coal-fired owned power plants as well as other anthropogenic activities (a large petrochemical plant, mining and steel and metal processing facilities) (SA. DEA 2011; Muthige, 2014;Language et al. 2016). The combined impact of these activities degrades local air quality hence why the region was declared as an Air-shed priority area under the National Environmental Management: Air Quality Act (NEM: AQA) (Act no. 39 of 2004).
The impacts of poor ambient air quality are mostly experienced in low-income settlements (Hersey et al., 2014). These settlements are usually associated with high levels of poverty, poor service delivery and lack of resources (Language et al. 2016). Furthermore, they often experience a double burden, which includes poor living conditions as well as exposure to poor ambient air quality levels due to indoor and outdoor air pollution from various sources (John and Sonali 2012). This is since the majority of the people in these settlements do not earn enough to afford clean energy options, so they settle for the affordable and readily available options, for instance, wood, paraffin fuel and coal (Language et al. 2016). Additionally, these settlements are usually located in areas directly impacted by major sources of emission (SA. DEA 2016).
Pressure on the industry from governmental institutions and civil community to reduce air pollution and its carbon footprint are steadily growing (ITA 2018). This includes the emission standards as set in terms of the Air Quality Act, the carbon tax bill that will be implemented at the beginning of 2019 and the alternative energy strategies implemented by the Department of Energy to control industrial emissions (SA. DEA 2016). SA. DEA (2011) acknowledges that various studies agree that the local sources, notably domestic coal burning is responsible for high concentrations of ambient air pollution but this specific contribution has not been adequately addressed nor fully been understood.
The aim of this research is to model the dispersion of emissions of sulphur dioxide (SO 2 ), nitrogen oxides (NO X ) and fine particulate matter (PM 2.5 ), originating from three Eskom power plants, and to evaluate their possible impact on air quality of KwaZamokuhle Township. Air quality assessment is enhanced by using monitoring and modelling to evaluate the most effective emissions mitigation strategies. Therefore, a combination of AERMOD dispersion modelling and ambient air quality monitoring data have been used to achieve the aforementioned aim of this study.

AERMOD
Dispersion modelling is the standard method for analysing the impact of emissions from air pollution sources (Cora and Hung 2003;US EPA 2005). The model consists of a mathematical expression of the dispersion and chemical transformation of pollutants in the atmosphere (Cora and Hung 2003). This is done by estimating downwind air pollution concentrations for a given time frame, at potential receptor sites (Mtiya 2013).
AERMOD is a steady-state plume dispersion model, which uses a Gaussian and bi-Gaussian approach (US EPA 2002;Touma et al. 2007). It is applicable to near-source emissions and up to least 50 km (US EPA 2002). The model has proved to be an effective tool in modelling the dispersion of tall stack emissions (Perry et al. 2005;Buthelezi 2010;Mtiya 2013). There are three main types of data input required in AERMOD, i.e., information on the source (including emission rates, heights and locations), meteorological data and the local topography of the area of interest. AERMET AERMET provides a meteorological pre-processor for consolidating available meteorological data into a fixed planetary boundary layer (PBL) format appropriate for use in AERMOD. This is achieved by computing distinctive surface parameters and meteorological measurements (US EPA 2016a). Wind speed and direction, ambient pressure and temperature, albedo, Bowen ratios and cloud cover were used as input data. Hourly dataset containing the aforementioned parameters was obtained from Lakes Environmental Software, Canada for the period 2015-2017 and used as input of the AERMET preprocessor. The modelled data included hourly surface and upper air meteorology data in a processed grid cell format, with 26.6484 S, 27.9256 E as the grid cell centre.

AERMAP
AERMAP allows for the processing of terrain data, including a layout of the receptor sites and sources (Touma et al. 2007). It can process several standardised data formats, which makes it possible to produce terrain base elevations for specific receptors and sources as well as a hill height scale value for each respective receptor (US EPA 2016b). To calculate the terrain height scale for the receptor location, AERMAP uses gridded data that makes it possible to calculate the divided streamline height.
The topographic data required by AERMAP was obtained from SRTM1 (Shuttle Radar Topography Mission), whereas, the land cover data was attained from (WebGIS), with a resolution of (~30m). Studies have shown that SRTM datasets have limitations-large outliers and voids exists in the dataset; the accuracy of the data decreases with an increase in elevation and slope; and it shows features like buildings and forests which can increase the error bar in studies like this (Karwel and Ewiak 2008;Gorokhovich and Voustianiouk 2006). The WGS84 projection datum was used, with the terrain data calculated based on the UTM coordinate system (35 South). The receptors' grid was set at 2000m by 2000m with 2500 points. KwaZamokuhle Township monitoring station was added as a sensitive discrete receptor in order to evaluate the simulated concentrations at the settlement.

Emission sources as model inputs
Arnot, Hendrina and Komati power plants were considered as the emission sources in this study due to their proximity to the low-income settlement in question ( Figure 1). The stack parameters used to run the model are shown in Table 1. These parameters were obtained from studies by Pretorius (2015) and Belelie (2017). The source characteristics and emission factors (calculated as an annual average from 2011-2013) used in these studies were obtained directly from Eskom. Emission rates used were converted from t.yr -1 to g.s -1 and were assumed to be constant over the study. The source characteristics and emission factors (calculated as an annual average from 2011-2013) used in these studies were obtained directly from Eskom Research, Testing and Development (RT&D).
AERMOD was run for an area of 50km x 50km using the configuration co-efficient for rural dispersion and elevated terrain. Upon completion of the model run, output plots of PM 2.5 , NO X and SO 2 maximum concentrations were obtained for 1-hour, 24-hour and annual averages for the three year period, to allow for the comparison with the ambient air quality standards. Likewise, hourly-generated concentration exceedance output files were generated, which were then related to the measured data drawn from the same location as the discrete receptor in the model.   Modelled NO X concentrations The maximum 1-hour average NO X concentration simulations for the three power plants exceeds the South African standards, while the maximums for the 24-hour and the annual averages comply with the standards ( Table 2). The model simulated maximum contributions of 263 µg.m -3 , 57.8 µg.m -3 and 6.40 µg.m -3 for the averaging periods, respectively (Figures 5 to 7). The average NO X contribution of the three power plants towards the Township was observed to be in the ranges of between 100-200 µg.m -3 , 10-20 µg.m -3 and 3-4 µg.m -3 for each averaging period, respectively.

Modelled PM 2.5 concentrations
The modelling results show that the power plants contribute very small amounts of primary particulate matter concentrations to the ambient environment in KwaZamokuhle. The results   also illustrate that at the discrete receptor at the township, maximum contributions ranging between 2-3 µg.m -3 for the hourly average, 0.5-0.6 µg.m -3 for the 24-hour average, and 0.09-0.1 µg.m -3 for the annual average are observed. Relatively small emissions were observed near the power plants for the daily and annual averages (Figure 8 and 9).
Monthly averages of modelled against monitored SO 2 , NO X and PM 2.5 seasonal signature for all the three species which is most likely due to the fact that the emission rates were kept at constant throughout the year.
The contribution of PM 2.5 is much lower compared to SO 2 and NO X . As depicted in Figure 12 the average concentrations for the three power plants combined are approximately 0.095 µg.m -3 , which is 0.2% of 48.6 µg.m -3 , measured at KwaZamokuhle Township monitoring site. The highest simulated values were observed at Komati, which had an average of 0.04 µg.m -3 , i.e., 0.08% of the average monitored PM 2.5 . Arnot contributes to an average prediction of 0.04%. The model simulated an average of 0.03 µg.m -3 at Hendrina, which translate to 0.06% of the measured concentrations.     Comparison of the temporal variations in the modelled and monitored data According to Carslaw (2015), variations of pollutions by time can be useful in revealing the likely sources. Figures 13 to 15 show the normalized diurnal and monthly patterns of the modelled contributions (power plants) in comparison to the variations in the measured data from the KwaZamokuhle Township monitoring site. The plots were generated at 95% confident interval in the mean, applying the normalised option. Normalisation is adjusting values with different scales to an estimated common one (Carslaw 2015) which is useful in comparing the patterns of the two (modelled and monitored) datasets that had distinct ranges in concentrations. The simulations showed that the model did not agree with the measurements during the nighttime in all instances, i.e., between 18:00 and 6:00 the graphs Research article: Assessing the impact of Eskom power plant emissions on ambient air quality over KwaZamokuhle Page 7 of 9 show low values. AERMOD neglects the residual plumes and it lacks the ability to approximate secondary formation of pollutants such as sulphates and nitrates produced during the day that impact on the ambient air quality during the stable night-time periods (US EPA 2005).
The diurnal plot of the modelled concentrations shows a peak at around mid-day due to the break-up of the inversion layer and downward mixing, which occurs at the same time as one of the two peaks observed for the monitoring data ( Figure 13). The early evening peaks can likely be attributed to other local sources other than the power stations. Nkosi (2018) reported that low-cost fuels (wood, coal and dung) are the main sources of energy in low-income settlements on the Highveld, with burning events dominant between 16:00 and 19:00.
Measured values show distinct seasonal patterns which can be attributed to the variation in burning events. In winter the events are less variable than in the summertime hence the peak in winter (Nkosi 2018). Modelled data does not show much variation throughout the year, with little changes as a result of meteorological conditions, for instance, a dip observed around June and July is most possibly due to stable winter conditions.
In the case of NO X (Figure 14), three peaks were observed for the monitored data, with the early morning and the evening peaks typical of low-level sources. A study by Nkosi (2018), observed that there are two common burning events i.e. morning and evening burning. The third peak observed around 10 am, which is mutual for the modelled and monitored data can be attributed to the downward mixing of tall-stack emissions. For the monthly variations, a peak is observed in the winter due to increased burning events since apart from cooking space heating is intensified. A drop is observed for the modelled data in winter due to calm conditions experienced during this time.   Low-level sources typically show a bimodal distribution with higher peaks observed in the early morning and evening, which was observed from the monitoring data. Like the other species, simulated PM 2.5 peaks around mid-day (Muthige 2014). This can be related to the breakup of inversion layers and downward mixing of the tall stack emissions. Nonetheless, unlike the previous two species, its peak coincides with a dip in the monitored concentrations. This together with the two peaks observed during the morning and evening times indicate that there are other major sources besides power plants, with domestic combustion the most probable. Nkosi (2018) substantiate the assumption by affirming that domestic burning is a major source of PM 2.5 . The monthly trend is the same as the previous species with a maximum being observed in winter for the monitored data and not much variation for the modelled.

Conclusion
In this study, the possible impact of Eskom coal-fired power plants on the ambient air quality in KwaZamokuhle Township was evaluated through modelling with AERMOD. This was investigated by comparing simulated and monitored data for the period 2015-2017. Three power plants were considered with PM 2.5 , NO X and SO 2 being analysed as the criteria pollutants. Secondary particulate formation, detailed meteorological effects were not modelled by AERMOD and the peak short-term impacts were under estimated by the model when simulating annual average concentrations and these were perceived as limitations of this study.
The results showed that the contribution of the power plants as compared to the measured data is small. It may be concluded that poor ambient air quality in the community is not solely attributable to emissions from the surrounding power plants but may be originating from other sources as well. It is evident that local sources, specifically domestic burning, may be the dominant contributor to poor air quality over KwaZamokuhle. The frequency in exceedance of the NAAQS suggests that the residents of the low-income settlement may be susceptible to health risks associated with these criteria pollutants. Temporal variations in the measured data supports this argument in that daily peaks were observed during the morning and evening-a diurnal variation pattern commonly associated with low-level sources like domestic combustion.
Therefore, existing strategies that focus predominantly on the industrial sector may not be successful in improving ambient air quality. Local sources like solid fuel burning, waste burning, dust, and vehicles, should be included in order to come up with reduction mechanisms. This is important as these sources are located at ground level and in close proximity to humans.