edoc

Air pollution exposure assessment in sparsely monitored settings; applying machine-learning methods with remote sensing data in South Africa.

Arowosegbe, Oluwaseyi Olalekan. Air pollution exposure assessment in sparsely monitored settings; applying machine-learning methods with remote sensing data in South Africa. 2022, Doctoral Thesis, University of Basel, Faculty of Science.

[img]
Preview
PDF
4Mb

Official URL: https://edoc.unibas.ch/90091/

Downloads: Statistics Overview

Abstract

Air pollution is one of the leading environmental risk factors to human health – Both short and long-term exposure to air pollution impact human health accounting for over 4 million deaths. Although the risk of exposure to air pollution has been quantified in different settings and countries of the world. The majority of these studies are from high-income countries with historical air pollutant measurement data and corresponding health outcomes data to conduct such epidemiological studies. Air pollution exposure levels in these high-income settings are lower than the exposure levels in low-income countries. The exposure level in sub-Saharan Africa (SSA) countries has continued to increase due to rapid industrialization and urbanization. In addition, the underlying susceptibility profile of SSA population is different from the profiles of the population in high-income settings. However, a major limitation to conducting epidemiological studies to quantify the exposure-response relationship between air pollution and adverse health outcomes in SSA is the paucity of historical air pollution measurement data to inform such epidemiological studies.
South Africa an SSA country with some air quality monitoring stations especially in areas classified as air pollution priority areas have historical particulate matter less than or equal to 10 micrometres in aerodynamic diameter (PM10 μg/m3) measurement data. PM10 is one of the most monitored criteria for air pollutants in South Africa. The availability of satellite-derived aerosol optical depth (AOD) at high spatial and temporal resolutions provides information about how particles in the atmosphere can prevent sunlight from reaching the ground. This satellite product has been used as a proxy variable to explain ground-level air pollution levels in different settings.
This thesis main objective was to use satellite-derived AOD to bridge the gap in ground-monitored PM10 across four provinces of South Africa (Gauteng, Mpumalanga, KwaZulu-Natal and Western Cape). We collected PM10 ground monitor measurement data from the South Africa Weather Services across the four provinces for the years 2010 – 2017. Due to the gaps in the daily PM10 across the sites and years. In study I, we compared methods for imputing daily ground-level PM10 data at sites across the four provinces for the years 2010 – 2017 using random forest (RF) models.
The reliability of air pollution exposure models depends on how well the models capture the spatial and temporal variation of air pollution. Thus, study II explored the spatial and temporal variations in ground monitor PM10 across the four provinces for the years 2010 – 2017. To explore the feasibility of using satellite-derived AOD and other spatial and temporal predictor variables, Study III used an ensemble machine-learning framework of RF, extreme gradient boosting (XGBoost) and support vector regression (SVR) to calibrate daily ground-level PM10 at 1 × 1 km spatial resolution across the four provinces for the year 2016.
In conclusion, we developed a spatiotemporal model to predict daily PM10 concentrations across four provinces of South Africa at 1 × 1 km spatial resolution for 2016. This model is the first attempt to use a satellite-derived product to fill the gap in ground monitor air pollution data in SSA.
Advisors:de Hoogh, Kees and Röösli, Martin and Adar, Sara
Faculties and Departments:09 Associated Institutions > Swiss Tropical and Public Health Institute (Swiss TPH) > Department of Epidemiology and Public Health (EPH) > Environmental Exposures and Health Systems Research > Physical Hazards and Health (Röösli)
UniBasel Contributors:de Hoogh, Kees and Röösli, Martin
Item Type:Thesis
Thesis Subtype:Doctoral Thesis
Thesis no:14817
Thesis status:Complete
Number of Pages:ix, 62
Language:English
Identification Number:
  • urn: urn:nbn:ch:bel-bau-diss148174
edoc DOI:
Last Modified:29 Oct 2022 04:30
Deposited On:28 Oct 2022 10:27

Repository Staff Only: item control page