Winter inversions and summer smoke: A season-dependent approach to PM2.5 modeling with low-cost sensors in complex terrain

Abstract

Accurate PM2.5 monitoring in complex terrain is challenged by sparse regulatory networks and distinct seasonal pollution sources, like winter temperature inversions and summer wildfires. This study develops and evaluates a hybrid modeling framework that integrates bias-corrected low-cost sensor data to improve spatial and temporal PM2.5 mapping in Montana, USA. We first applied a local, observation-based bias correction to dense PurpleAir sensor network, substantially improving agreement with reference monitors (R2 increased from 0.750 for a generic correction to 0.838). This corrected data was used to create high-resolution smog potential (smogP) surface, a static layer representing terrain-driven pollution accumulation, which explained 59.5% of the spatial variance of mean PM2.5 during wintertime inversion events. Daily 600 m resolution PM2.5 grids were then generated using geographically weighted regression framework. Cross-validation revealed a strong seasonal dichotomy in model performance: during the winter inversion season (November–March), models incorporating the smogP layer were critical and consistently outperformed other approaches. Conversely, during the summer wildfire season (May–September), the high density of sensor data alone was often sufficient, with simpler spatial interpolation models proving as effective at capturing the diffuse nature of large smoke plumes than models constrained by static covariates like smogP or satellite aerosol optical depth (AOD). We conclude that the optimal strategy for PM2.5 mapping in complex terrain is season-dependent, and this work provides a framework for leveraging citizen-science data to enhance air quality forecasting in these challenging environments.

Citation

Swanson, Alan; Orr, Ava; Holden, Zachary A.; Bocinsky, Kyle; Landguth, Erin L. 2026. Winter inversions and summer smoke: A season-dependent approach to PM2.5 modeling with low-cost sensors in complex terrain. Science of The Total Environment. 1040: 181915.