Fine Particulate Air Pollution Estimation in Ouagadougou using Satellite Aerosol Optical Depth and meteorological Parameters

Abstract

In this paper, PM2.5 concentrations in Ouagadougou are estimated using satellite-based Aerosol Optical Depth and Meteorological Parameters. Firstly, Simple Linear Regression (SLR), Multiple Linear Regression (MLR), Decision Tree (DT), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) are developed using the available labeled data in the city. The XGBoost model outperforms all the models with a coefficient of determination (R2) of 0.87 and a root-mean-square error (RMSE) of 15.8 μg/m3. Given the outstanding performance of the supervised XGBoost model, it is upgraded by the incorporation of a semi-supervised algorithm to make use of the lots of unlabeled data in the city and allow for the extensive estimation of PM2.5. The developed semi-supervised XGBoost model has an R2 of 0.97 and an RMSE of 8.3 μg/m3. The results indicate that the estimated PM2.5 concentrations in the city are 2 to 4 times higher than the World Health Organization (WHO) 24-hour limit of 15 μg/m3 in the rainy season and 2 to 22 times higher than the WHO 24-hour limit in the dry season. The results also reveal that the average annual estimated PM2.5 concentrations are 11 to 14 times higher than the WHO average annual standard of 5 μg/m3. Finally, the results reveal higher PM2.5 concentrations in the center and industrial areas of the city compared to the other areas. There should be an improvement in traffic management in the central areas of the city and Industries should implement cleaner production methods.