Estimate of PM2.5 concentration in New York State during fire seasons of 2016 – 2019 using machine learning

Abstract:

Wildfires have significant impacts on air quality by emitting enormous amount of fine particulate matters (i.e. PM_2.5, with radius less than 2.5 µm), so-called smoke aerosols, and trace gases to the atmosphere. Under favorable conditions, these smoke aerosols can transport over long distance and affect local air quality in downwind regions. Episodic transport of smoke aerosols from wildfires in the Western States and Canada has been reported across New York State (NYS). While the anthropogenic emissions in NYS continue to decrease due to various emission controls, the contributions of such transported smoke aerosols in PM_2.5 could become increasingly more significant. The objective of this study is to use machine learning techniques with satellite aerosol measurements and model products to estimate surface PM_2.5 concentrations in NYS during fire seasons of 2016 – 2019. Machine learning algorithms, which are widely used in air quality studies, are capable to objectively detect the nonlinear correlation between given variables. Indicators of vertical mixings and aloft aerosol layers are incorporated into machine learning trainings and lead to improvement in the accuracy of PM_2.5 estimate. Understandings of the linkages between transported smoke aerosols and surface PM_2.5 are beneficial to improve air quality monitoring and forecast.