Wildfire-impact identification in the Western USA using numerical techniques.

Aerosols exhibited distinct particle size distribution (PSD) characteristics under wildfire smoke influence compared to clean conditions. To assess the impact of wildfire smoke, the k-means clustering algorithm was applied using three types of data: air pollutant concentrations (PM2.5, CO, O3, and NOx), PSD, and optical measurements. This approach demonstrated the capability of k-means in identifying wildfire influences. Wildfire-related clusters, characterized by larger particle sizes and stronger correlations between PM2.5 and CO concentrations, also exhibited increased light extinction coefficients (βext), higher absorption Ångström exponent (AAE), and elevated extinction Ångström exponent (EAE) values. Using clustering results, a Random Forest classification system (RFCS) was developed to identify smoke-influenced hours based on air pollutant concentrations. Trained on k-means clusters derived from the combination of all three data types along with time-series corresponding air pollutant concentrations and months, the RFCS was applied to classify 12-year air pollutant concentrations. During the identified smoke-influenced hours, compared to non-smoke-influenced hours, mean concentrations of PM2.5 and CO were 12 times and 2.6 times higher, respectively, with higher PM2.5/CO ratio and stronger correlations. Furthermore, βext, EAE, and AAE values were significantly higher during identified smoke-influenced hours, confirming the robust performance of the RFCS. This study proposed methods for identifying smoke-influenced hours, which are particularly useful for detecting light wildfire smoke.