Estimating PM_2.5 in Fort Collins, Colorado using a robust drive-by correction of low-cost air quality monitors...

We characterized PM_2.5 concentrations in Fort Collins, Colorado, using eight low-cost Personal Air Monitors (PAMs) deployed on school buses and maintenance vehicles between June 22, 2022, and August 25, 2023. The PAMs routinely passed four reference-grade monitoring stations, and we considered them co-located with reference monitors when within a specified distance. We developed several corrections for the PAMs using these co-located data. Beyond typical cross-validation techniques, we evaluated model transferability across space by systematically leaving out one reference monitor, training models on data from the remaining sites, and testing performance at the excluded location. Although the random forest (RF) correction model outperformed others in cross-validation, simpler models often demonstrated superior transferability to left-out sites. The RF model transferred better to locations impacted by local sources, while linear models showed stronger transferability to sites influenced by regional sources. We generated representative PM_2.5 concentrations for the study area based on measurements corrected with different models. We observed high concentrations of PM_2.5 near the former downtown airport and at major intersections that were robust to the correction model used; however, the RF model highlighted certain hotspots that were not present when using other corrections. Other researchers can use our framework to evaluate calibration model transferability and determine what air pollution insights they can robustly claim based on their chosen approach. SYNOPSIS: We introduce a novel drive-by correction technique for mobile low-cost sensors and present results from testing it in Fort Collins, Colorado.