The causes of PM2.5 is dynamic and systematic. However, many studies approach the PM2.5 problem by focusing only on either socioeconomic factors or geo-meteorological factors in isolation such data insufficiency might undermine the effort to control PM2.5. We propose a LSTM-XGBoost model composing both socioeconomic and geo-meteorological factors together to improve the PM2.5 monitoring system. We forecast the weekly PM2.5 concentrations in five regions in Taiwan based on machine learning training data. The results indicate that overall small trucks usage should be reduced by 80% while maintaining semi-trucks and passenger cars at current level. In addition, coal and IPP Gas power have no impact on PM2.5 concentrations in central Taiwan while usage in passenger cars, small tracks and tractor trailers should be reduced by 80% in central Taiwan. Overall, central Taiwan and Chiayi regions have the highest PM2.5 projections at XGBoost output of 68.5 and 59.1 level. Finally, our model indicates that the use of fossil fuel based small tracks and tractor trailers should be reduced by 80% to maintain a reasonable level of PM2.5.
|Journal||Aerosol and Air Quality Research|
|State||Published - Oct 2022|
- Air pollution
- Machine learning