Machine Learning Gap-Filling AOD Models for High Resolution PM2.5 Predictions: Assessing Wildfire Impact on Air Quality in US Northeastern States

Abstract

Breathing in fine particulate matter with diameters less than 2.5 µm (PM2.5) has been shown to greatly increase an individual’s risk of cardiovascular and respiratory diseases. In light of the recent Canadian Wildfires' impact on air quality in the US northeastern states, this research is focused on the influence of these wildfires on PM2.5 levels. While air quality monitoring networks can provide accurate and timely PM2.5 readings, these stations are often too sparsely located to capture the spatial and temporal variabilities of PM2.5. As a solution, satellite measured aerosol optical depth (AOD) is commonly used as a source for PM2.5 pollution assessment. AOD quantifies the amount of aerosols present in the atmosphere by measuring the extinction of a ray of light as it passes through a vertical column of air. Initially, satellite AOD products were validated against ground truth data, ensuring the accuracy of the satellite measurements. Present work centers on developing machine learning gap-filling AOD models through the fusion of multisource AOD data. This approach aims to address data gaps caused by cloud cover, providing a complete hourly AOD dataset. In future work, gap filled AOD data will be combined with meteorological information and ground based PM2.5 measurements to generate precise and localized PM2.5 predictions. These insights will significantly contribute to a better understanding of wildfire impact on air quality in the US northeastern states with the hopes of more informed public health and environmental policies.