Incorporating Low-Cost Sensor Measurements into High-Resolution PM2.5 Modeling at a Large Spatial Scale

Abstract

Low-cost air quality sensors are promising supplements to regulatory monitors for PM_2.5 exposure assessment. However, little has been done to incorporate the low-cost sensor measurements in large-scale PM_2.5 exposure modeling. We conducted spatially varying calibration and developed a downweighting strategy to optimize the use of low-cost sensor data in PM_2.5 estimation. In California, PurpleAir low-cost sensors were paired with air quality system (AQS) regulatory stations, and calibration of the sensors was performed by geographically weighted regression. The calibrated PurpleAir measurements were then given lower weights according to their residual errors and fused with AQS measurements into a random forest model to generate 1 km daily PM_2.5 estimates. The calibration reduced PurpleAir's systematic bias to ∼0 μg/m³ and residual errors by 36%. Increased sensor bias was found to be associated with higher temperature and humidity, as well as longer operating time. The weighted prediction model outperformed the AQS-based prediction model with an improved random cross-validation (CV) R² of 0.86, an improved spatial CV R² of 0.81, and a lower prediction error. The temporal CV R² did not improve due to the temporal discontinuity of PurpleAir. The inclusion of PurpleAir data allowed the predictions to better reflect PM_2.5 spatial details and hotspots.