LAUSR

Abstract Smoke aerosols emitted from wildfires can transport across long distances and affect the local air quality in downwind regions. In New York State (NYS), the local air quality has significantly improved due to reductions in anthropogenic emission over the past decades. As the intensity and frequency of wildfires are continuously increasing under changing climate, smoke aerosols are predicted to become the dominant source of fine particulate matter (PM2.5) concentration in NYS in the future. In this study, smoke and non-smoke cases in NYS during the summer seasons of 2012–2019 were identified using satellite measurements and aerosol reanalysis products. Overall, smoke cases showed higher PM2.5 concentrations than non-smoke cases with average PM2.5 concentrations of 11.5 ± 5.9 μg m−3 and 6.6 ± 4.6 μg m−3, respectively. PM2.5 concentrations exceeding 20 μg m−3 mainly occurred during smoke cases. In addition, an artificial neural network (ANN) algorithm was used to estimate surface PM2.5 mass concentrations at 21 air quality monitoring sites in NYS. Results showed that, for smoke cases, the application of predictors designed as indicators of vertical transport mechanisms and smoke inflow from the fire source regions generally improved the model performance by reducing the model errors. Also, analysis of the variable correlations and variable importance indicated that synoptic subsidence, entrainment process, and turbulent mixing within PBL collectively contributed to PM2.5 concentrations for smoke cases. Machine learning techniques showed the capabilities of learning the general air quality features, characterizing the key contributors to PM2.5 concentrations, and distinguishing the vertical transport processes of smoke aerosols.