CHAPTER 6 83
6.2 Dispersion modelling results validation
Figure 6.1-13. Modelled PM2.5 and PM10 concentrations from individual identified sources simulated at Zamdela monitoring station.
Figure 6.1-14 illustrates the PM contributions of the identified sources to concentrations at the Wedela monitoring station. As anticipated, dust emissions contribute significantly at over 40% in the simulations, due to Wedela being surrounded by mine tailings and bare areas. Vehicles have the second-highest PM contribution, as vehicle ownership is on the rise in South Africa and individuals prefer to use their cars over public transport.
Figure 6.1-14. Modelled PM2.5 and PM10 concentrations from individual identified sources simulated at Wedela monitoring station.
however, in this study, variable emission factors were based on hours and seasons to emulate reality.
The measured and modelled concentrations were compared for the highest 95th percentile hourly, daily, and annual averages (Table 6.2-1). The comparison was made at the 95th percentile since the model generally overpredicts along the plume centreline (Hertwig et al., 2018). The ratio between measured and monitored PM is also presented, considering the range of uncertainty recommended by the US EPA of >0.5 <2.0 (modelled to monitored ratio).
The simulated ground level concentrations compared relatively well with measured ambient PM throughout the averaging periods with the ratios falling within the recommended range. The lowest ratio of 0.75 (10 μg.m-3 - modelled and 13 μg.m-3 - monitored) and the highest was 1.86 (99 μg.m-3 - modelled and 53 μg.m-3 - monitored) were estimated over Wedela for PM2.5.
The general model bias is to underpredict the long-term averaging periods, i.e., annual averages, which was evident for this study. There is also a suggestion of overprediction of the 24-hour averages, which for Zamdela could be attributed to the threshold assumed for the monitoring sensors. There is also a possibility that the emissions’ zones of impact differ between monitored and modelled data due to the upper air meteorological parameters that were acquired from an alternative source owing to the unavailability of the data for the study areas.
Table 6.2-1. Comparison of measured and predicted PM averages within Zamdela and Wedela.
Hourly Daily Annual
Zamdela
PM2.5
Monitored 51053 μg.m-3 140 μg.m-3 54 μg.m-3 Modelled 870 μg.m-3 207 μg.m-3 64 μg.m-3
Ratio 1.71 1.48 1.19
PM10
Monitored 801 μg.m-3 206 μg.m-3 101 μg.m-3 Modelled 1220 μg.m-3 279 μg.m-3 94 μg.m-3
Ratio 1.52 1.35 0.93
Wedela
PM2.5
Monitored 53 μg.m-3 41 μg.m-3 13 μg.m-3
Modelled 99 μg.m-3 48 μg.m-3 10 μg.m-3
Ratio 1.86 1.16 0.75
PM10
Monitored 192 μg.m-3 97 μg.m-3 38 μg.m-3 Modelled 219 μg.m-3 104 μg.m-3 36 μg.m-3
Ratio 1.14 1.08 0.94
The monitored and modelled PM concentrations results were compared to one another in Figure 6.2-1 to Figure 6.2-4 using scatter plots. The analysis showed a positive correlation between the
modelled and monitored concentrations for both townships. A weak correlation is seen between modelled and monitored PM2.5 for Wedela. This could be because only primary emissions were considered, and secondary particulates were excluded (Figure 6.2-1).
The model simulated the lower values well compared to higher concentrations, where the model overestimated as it continuously shows higher than the monitored values (Figure 6.2-1 to Figure 6.2-4). This however is ideal, as the expected performance of the model is to overestimate rather than underestimate, and consequently, the simulations would be conservative, thus ensuring compliance with ambient air quality regulatory frameworks and policy standards (DEA, 2014).
The overall performance of the model is positive; thus, it can be argued that the model can make a valuable contribution to air quality predictions for policy and regulatory purposes.
Figure 6.2-1. Scatter plot of monitored against modelled daily average PM2.5 concentrations for Zamdela.
Figure 6.2-2. Scatter plot of monitored against modelled daily average PM10 concentrations for Zamdela.
Figure 6.2-3. Scatter plot of monitored against modelled daily average PM2.5 concentrations for Wedela.
Figure 6.2-4. Scatter plot of monitored against modelled daily average PM10 concentrations for Wedela.
According to Carslaw (2015), variations of pollution by time can be useful in revealing likely sources. The plots were generated at a 95% confidence interval in the mean, applying the normalised option. Normalisation requires adjusting values with different scales to an estimated common one (Carslaw, 2015), which is useful in comparing the patterns of datasets that had distinct concentration ranges.
Figure 6.2-5 show the temporal variation of the modelled results in comparison to the measured data from the Zamdela monitoring station. The diurnal plots for both PM sizes show an agreement between the monitored and modelled results, with both peaking during the early morning (06:00–
09:00) and early evening (17:00–20:00), a pattern generally associated with domestic fuel burning events and vehicle emissions (Nkosi, 2018). The same is seen throughout the week with the weekends showing relatively lower variations suggesting the slowing down of events as people are generally resting. The monthly plots show peaks during the winter season, which are aligned to the monitored maxima, but failed to show the spring peaks. The peaks can be attributed to increased fuel burning events for space heating and can occur when there is little to no rainfall, resulting in easier entrainment of dust into the atmosphere. The failure to show the September peak demonstrates that the simulated values prove to be more sensitive to time variables in source activities, and unsuccessfully simulated wind impact.
Figure 6.2-5. Temporal variation plots for normalised PM2.5 and PM10 concentrations respectively comparing measured and simulated data over Zamdela.
Figure 6.2-6 shows the temporal variation at the Wedela site over the study period. Unlike at Zamdela, the diurnal patterns for PM2.5 and PM10 in Wedela are distinct, however, the variations between the monitored and modelled concentrations for both particulate sizes show agreement.
PM2.5 depicts a bimodal pattern that is associated with domestic fuel burning, waste burning and traffic events. On the contrary, PM10 reaches a maximum during the afternoon around 14:00 which can be attributed to the breaking of the inversion layer leading to downward mixing of particles.
This is also the time when the wind impact is at its peak, therefore the peaks are suggested to be a result of dust sources. The variation is uniform thought the week with low measurements observed during the weekends as activity slows down. The monthly variations showed little agreement. For both PM sizes, monitored observations recorded the highest concentrations during the windy periods (August and September). This is the time when fugitive dust emissions are at their peak. The simulated values proved to be more sensitive to variations in source activities than the measured data for the diurnal profile and failed to simulate wind impact. This is a general AERMOD bias as it underpredicts the impact of winds.
Figure 6.2-6. Temporal variation plots for normalised PM2.5 and PM10 concentrations respectively comparing measured and simulated data over Wedela.