Analysis and Modeling of the Sulfur Dioxide Pollution Levels over the South African Mpumalanga Highveld

The Mpumalanga Highveld is located in the east and south east of Johannesburg / Pretoria. The large coalfields in that region lead to the high agglomeration of coal-fired power stations which release enormous amounts of sulfur dioxide into the atmosphere.

The research of the study focused on modeling the atmospheric parameters, which are relevant to the dispersion of the pollution, as well as on modeling near ground air pollution levels including their spatial distribution.

Information regarding the thermal atmospheric layering and the boundary layer wind field was obtained by measurements and statistic models. The thermal atmospheric layering over the Mpumalanga Highveld shows distinctive characteristics. Stable conditions are reaching high during night and unstable conditions prevail near the ground during daytime with an inconsistent layering above. The calculation of the vertical thermal layering by similarity theory based models proved to be too inaccurate. For instance, the depth of the near ground instability was seriously over- or underestimated by the models during daytime and the parameterization of the stable conditions was entirely out of the range. The calculation of the vertical course of potential air temperatures is achieved with effectual accuracy up to 100 m AGL by multiple linear regression or neural networks based on the independent four variables “potential air temperature lapse rate between 1 and 10 m AGL”, “wind speed”, “potential air temperature at 1 m AGL ” and “solar angle”. The approaches for the calculation of wind speeds and wind directions at elevated levels yield large errors when they are applied to the Mpumalanga wind field. Improvements were achieved by multivariate linear statistical approaches. The highest correlation between calculated and measured data was achieved by Neural Network calculations. The input variables were near ground wind data and various air temperature lapse rates.

Modeling of the spatial distribution of air pollution levels is based on the data of monitoring stations neural network calculations. It was possible to calculate exceedance frequencies of certain threshold pollution levels at the monitoring stations, based on various independent variables and an adequate configuration of the learning system. The model was used subsequently for a spatial interpolation of the pollution levels and calculating the number of people living in that region, who are exposed to certain pollution levels. The application or transfer of the insights to the industrial highveld regions reveals, that large areas and considerable numbers of people are exposed to pollution levels, which exceed the thresholds of ambient air quality standards more or less frequently. The area, which is affected most frequently by the highest pollution levels, is the central industrial highveld region with the cities Kendal, Leandra, Evander, Secunda, Trichardt, Bethal and Kriel. This area extends considerably further south and west than the area, which has been identified as the most problematic “Central ETH zone” by previous studies.

A meaningful modeling of the daily near ground SO2 - air pollution courses at the monitoring stations requires detailed information on the dispersion relevant atmospheric parameters, such as the wind field and thermal atmospheric layering up to the stack heights. It is possible to partially understand and explain the occurrence of some of the pollution peaks. Nevertheless, approaches to calculate the pollution levels with deterministic physical/statistical models were not successful due to the complex interrelationship of the input parameters and the lack of available data. However, it is possible to reconstruct or predict the occurrence of pollution peaks at the monitoring stations based on neural network calculations with adequate accuracy.