High spatiotemporal resolution co-estimation of PM2.5 chemical components in the western United States

Fine particulate matter (PM₂.₅) is a complex atmospheric mixture primarily composed of sulfate (SO₄²⁻), nitrate (NO₃⁻), elemental carbon (EC), and organic carbon (OC), which exhibit distinct emission characteristics, spatiotemporal distribution patterns, and health effects. Developing long-term, continuous datasets of PM₂.₅ chemical components with high spatiotemporal resolution remains a key challenge in air quality modeling and exposure assessment. In this study, we developed a Multi-Output Extremely Randomized Trees (Multi-ET) model to simultaneously estimate four major PM₂.₅ chemical components (SO₄²⁻, NO₃⁻, EC, and OC) across the western United States by integrating daily 1 km resolution MAIAC aerosol optical depth (AOD) observations with a suite of environmental predictors. By jointly modeling multiple components and exploiting their intrinsic interrelationships, the proposed framework substantially improved computational efficiency while maintaining robust predictive performance. Random 10-fold cross-validation yielded coefficients of determination (R²) ranging from 0.70 to 0.77 across the four components. Compared with conventional single-output models, the Multi-ET model improved computational efficiency by approximately 2.9-fold. We generated a daily 1 km resolution dataset of PM₂.₅ chemical components for the western United States spanning from 2000 to 2019. Long-term trend analysis revealed decreasing concentrations of SO₄²⁻, NO₃⁻, and EC, accompanied by a slight increase in OC, reflecting changes in regional emission source structures and the implementation of air pollution control measures. The resulting high-spatiotemporal-resolution dataset provides strong support for regional air quality management, source apportionment, and component-specific health risk assessment.