Article Text

Download PDFPDF
OP III – 5 Land use regression modelling of outdoor no2 and pm2.5 concentrations in three low-income areas of the urban western cape, south africa
  1. Apolline Saucy1,
  2. Martin Röösli1,
  3. Nino Künzli1,
  4. Ming-Yi Tsai2,
  5. Chloé Sieber1,
  6. Toyib Olaniyan3,
  7. Roslynn Baatjies3,4,
  8. Mohamed Jeebhay3,
  9. Mark Davey2,
  10. Benjamin Flückiger2,
  11. Rajen N Naidoo5,
  12. Mohammed Aqiel Dalvie3,
  13. Mahnaz Badpa1,
  14. Kees De Hoogh1
  1. 1Swiss TPH/University of Basel, Epidemiology and Public Health, Basel, Switzerland
  2. 2Swiss TPH, Epidemiology and Public Health, Basel, Switzerland
  3. 3University of Cape Town, Centre for Environmental and Occupational Health Research, School of Public Health and Family Medicine, Cape Town, South Africa
  4. 4Cape Peninsula University of Technology, Department of Environmental and Occupational Studies, Faculty of Applied Sciences, Cape Town, South Africa
  5. 5University of KwaZulu-Natal, 4) Discipline of Occupational and Environmental Health, School of Nursing and Public Health, Durban, South Africa


Background/aim Intra-urban air pollution has been associated with adverse health effects, such as cardiovascular or respiratory disorders. Land Use Regression (LUR) modelling is one method to describe small-scale spatial variation in air pollution levels based on several measurements and geographical predictors.

Methods The main goal of the study is to characterise and model the spatial distribution of air pollutants in three neighbourhoods in the Western Cape, South Africa. Weekly measurements of NO2 and PM2.5 were performed in these areas (Khayelitsha, Marconi-Beam and Masiphumulele) during 2015–2016. They were temporally adjusted to obtain seasonal means using routinely monitored pollution data in Cape Town region. We developed six LUR models (four seasonal and two annual averages) using supervised forward stepwise regression for NO2 and PM2.5. Predictor variables, like road, land use and emission data were either obtained or collected on site. The models were validated using leave-one-out-cross-validation (LOOCV) and were tested for spatial autocorrelation.

Results Measured air pollution levels were generally low. The annual mean NO2 levels were 21.5 µg/m3 and 10.0 µg/m3 for PM2.5. The NO2 annual model explained 45% of the variance (R2) in the study areas and was found to have a satisfactory internal validity (LOOCV R2=70%). The PM2.5 annual model presented lower explanatory power (R2=25%, LOOCV R2=13%). The best predictors for NO2 modelling were traffic-related variables (major roads and bus routes) and proximity to some land-use features. Smaller local sources such as open grills and waste burning sites were good predictors for PM2.5 spatial variability, together with population density. NO2 and PM2.5 mean exposure will be predicted for home and school locations of about 400 pupils at primary schools involved in an epidemiological health study.

Conclusion This research shows that land use regression modelling can be successfully applied to informal urban settings in South Africa using similar predictor variables to those performed in European and North American studies. We could also provide NO2 and PM2.5 seasonal exposure estimates and maps for the selected study areas.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.