Evaluating the potential of remote sensing imagery in mapping ground-level fine particulate matter (PM2.5) for the Vaal Triangle Priority Area

Introduction At a global scale, air pollution is ranked fourth amongst the leading risk factors to human health, with recent estimates linking it to over 5 million premature deaths (Mannucci and Franchini, 2017; Bhanarkar et al., 2018; Health Effects Institute, 2019). In sub-Saharan Africa (SSA), particularly in the urban areas, the deterioration in air quality as a result of rapid urbanisation, population growth and industrial expansion is evident (Amegah and Agyei-Mensah, 2017; Fayiga, Ipinmoroti and Chirenje, 2018). Of great concern to public health, are the levels of fine particulate matter (PM2.5) in the cities of SSA, which are amongst the highest in the world (Fayiga, Ipinmoroti and Chirenje, 2018; Katoto et al., 2019). Long-term exposure to high levels of PM2.5 is harmful to humans as it can lead to increased severity in the symptoms of asthma and chronic obstructive pulmonary disease (Dieme et al., 2012; Feng et al., 2016). In South Africa, air pollution has become an important issue, especially in industrialised regions like the Vaal Triangle Airshed Priority Area (VTAPA) where strong economic growth has taken place (Naiker et al., 2012; Zhu et al., 2012). The VTAPA routinely experiences poor air quality as a result of strong emissions from industries, residential burning, vehicles and fugitive dust sources coupled with unfavourable meteorological conditions that have led to the accumulation of PM2.5 in high concentrations (Annegarn and Scorgie, 1997; Scorgie et al., 2003).

quality monitoring stations were placed at identified hotspots in the VTAPA, so as to assess pollution trends and to ascertain whether concentrations of PM 2.5 and other pollutants are being kept within the regulatory limits (Ngcukana, 2016;Altieri and Keen, 2019). However, similar to other urban areas across the world ( Gupta et al., 2006;Tian and Chen, 2010;Hu et al., 2014), the spatial coverage of stationary ambient monitors in the VTAPA is low. Intra-urban variability of PM 2.5 concentrations is therefore not accounted for. In order to capture the full-scale variability of PM 2.5 concentrations in the VTAPA, there is need for a vast network of monitoring stations. However, this requires large financial resources (Munir et al., 2016).
Satellite remote sensing can provide repeated observations of atmospheric pollution at large spatial scales. The monitoring of air pollutants using satellite observations is gradually gaining more attention in atmospheric pollution studies (Engel-Cox, Hoff and Haymet, 2004;Duncan et al., 2014). Advancements in regional algorithms have allowed for the large scale retrieval of PM 2.5 concentrations at fine spatial resolutions that have a reasonable agreement with ground measurements (van Donkelaar et al., 2015;van Donkelaar et al., 2016). These retrievals have been successfully used to assess long term spatial-temporal patterns of PM 2.5 in regions experiencing poor air quality such as China and Saudi Arabia (Lu et al., 2017;Munir et al., 2016;Shi et al., 2012).
In the case of South Africa, knowledge on the applicability of remote sensing to monitor air pollution levels is insufficient. A regional case study by Kneen et al. (2016) revealed that satellite technology has the potential to offer a practical and credible option to ground-based monitoring in South Africa. However, further investigation is still required in order to have more concrete evidence to advocate the use of remote sensing for Research article: Potential of remote sensing imagery in mapping PM 2.5 for the Vaal Triangle Priority Area Page 2 of 7 air quality monitoring in South Africa. This study evaluated the potential use of remote sensing imagery for air quality assessment in the VTAPA using a publicly available high resolution remotely sensed PM 2.5 concentration global dataset developed by van Donkelaar et al. (2015). This dataset begins in 1998 and ends in 2016. Spatial variations in satellite-retrieved PM 2.5 concentrations were examined, and temporal trends were explored.

Study region
The VTAPA is an industrialised region lying on the South African Highveld, a high central plateau of South Africa that forms part of the grassland biome. It stretches from southern Gauteng down to the northern section of the Free State province with an area of over 4 900 km 2 ( Figure 1). The region is host to industries including iron and steel manufacturers (ArcelorMittal steel and Davesteel), ferroalloy (Samancor Metalloys) and petrochemical companies (Sasol Chemical Industries, Natref and Omnia fertilisers). Eskom's Lethabo power station is also located in this region.
The VTAPA has a population of over 3 million people with most of its inhabitants residing in south-western Johannesburg, Soweto, Sebokeng, Sharpeville, Vereeniging, Vanderbijlpark, Sasolburg and Zamdela.

Satellite-derived data
Global satellite-derived PM 2.5 concentration data with a high spatial resolution (1 km x 1 km) covering a 10-year period from 2007 to 2016 was obtained from the Atmospheric Composition Analysis Group of Dalhousie University (http://fizz.phys.dal.ca/~atmos/ martin/). This global dataset was generated by merging satellite Based on previous research, uncertainty may exist in the global PM 2.5 data as a result of the satellite retrieval method (van Donkelaar et al., 2015). Existing studies have resolved this by applying a three-year average as an annual average (Han, Zhou and Li, 2015;Peng et al., 2016;Shisong et al., 2018). For this research, three-year moving averages were applied to the satellite retrievals from the period 2009 to 2016. A subset of the global PM 2.5 dataset for each year was extracted to cover the VTAPA study area using the Integrated Land and Water Information System (ILWIS) program (ITC, 2011).

Ground measured data
The VTAPA air quality monitoring network consists of six stations (Figure 1), which are located in Diepkloof (

Results and discussion
Comparison of temporal variations between satellite-retrieved and ground measured PM 2.5 concentrations A comparison between satellite-retrieved and ground measured PM 2.5 concentrations for the six monitoring stations in the VTAPA for the period 2009-2016 is shown in Figure 2. It can be observed that in most cases, there was an overestimation of observed PM 2.5 concentrations by the satellite retrievals, with the exception for the Kliprivier and Sebokeng stations, where ground measurements were much higher than the satellite-retrieved estimates. Both satellite retrievals and ground measurements showed that PM 2.5 concentrations for all sites, except Diepkloof, were above the annual National Ambient Air Quality Standards (NAAQS).  Satellite retrieval performance evaluation Satellite retrievals for PM 2.5 were compared with the ground measurements from all monitoring stations for the period 2009 to 2014 using the following performance evaluation metrics: root mean square error (RMSE) and correlation coefficient (R). These statistics were computed in R statistical software using the modstat function in the Open Air package. RMSE values ranged from 6 to 11 µg/m 3 , with an average of 9 µg/m 3 , indicating a significant difference between ground measured and satellite-retrieved PM 2.5 values (Table 1)    Similar spatial patterns for PM 2.5 concentrations in the VTAPA were observed throughout the period (2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016) in which the majority of PM pollution is concentrated around the centre and towards the south-western region of the VTAPA. These spatial patterns are comparable to those observed by Thomas (2008), who modelled PM 10 concentrations in the VTAPA using a dispersion model. Due to the low dispersion potential of pollutants in the VTAPA, high PM 2.5 concentrations were clustered in the central to the south-west region over the cities of Vanderbijlpark and Sasolburg, where heavy industrial (iron and steel, ferroalloy and petrochemical) activities, domestic burning and mine operations take place. Spatial distributions for the VTAPA show that high PM 2.5 concentrations were also concentrated in the northern part of the area. The main source for PM 2.5 in this area is residential combustion from the Soweto township and windblown dust from gold mine dumps.

Conclusion
This study evaluated the potential value of satellite remote sensing as a viable alternative to PM 2.5 ground-based monitoring in the VTAPA. There was a poor agreement between satellite-retrieved PM 2.5 estimates and ground-level PM 2.5 measurements. Satellite retrievals tended to overestimate PM 2.5 concentrations resulting in inflated values throughout most of the VTAPA. According to Kneen et al. (2016), due to financial constraints in South Africa, monitoring stations are placed in regions with the predicted worst air quality that is mainly urban and industrial centres. This will lead to mainly high values being used in the calibration of PM 2.5 estimates in the GWR model which in turn can result in the overestimation of satellite-retrieved PM 2.5 concentrations. Therefore, in order to improve the accuracy of the GWR model for satellite retrievals, the positioning of ground-based stations in South Africa needs to be optimised so as to have monitoring data that is more spatially representative.
The relationship between AOD and PM 2.5 is an important source of uncertainty in satellite retrieval accuracy as the AOD-PM 2.5 relationship can vary across space and countries. This could have contributed to some of the inconsistencies between ground measured and satellite-derived PM 2.5 concentrations for the VTAPA. There is a need for an independent assessment of the AOD-PM 2.5 relationship through an integrated monitoring strategy like SPARTAN (Surface PARTiculate mAtter Network) in which PM 2.5 monitoring instruments are collocated with ground-based sun photometers for AOD measurements. This will help in evaluating AOD-PM 2.5 model accuracy and enhance PM 2.5 estimates from satellite AOD retrievals.