Unveiling the Limits of Existing Correction Factors for a Low-Cost PM2.5 Sensor in Cold Environments and Development of a Tailored Solution

Abstract

PM_2.5 poses significant health risks and requires accurate monitoring. While the EPA’s high-cost Federal Reference Methods and Federal Equivalent Methods provide reliable data, they are often sparsely distributed, limiting community-scale assessments. Low-cost sensors like PurpleAir (PA) offer a promising alternative but require careful location-specific calibration and correction for environmental influences. Although several correction factors have been developed for use across regions and nationwide, these models often exhibit bias due to the predominance of data from temperate and warmer climates. This study was conducted to evaluate the performance of PA sensors in measuring PM_2.5 in extremely cold environments, specifically North Pole, Alaska. Data from PA sensors and a Beta Attenuation Monitoring (BAM) reference sensor were used to develop correction models. The study found that temperature and relative humidity significantly influenced PA sensor accuracy in the region. By comparing various regression models, including Ordinary Least Squares, Lasso, Ridge, and Elastic Net, an optimal model was identified that substantially reduced errors and aligned PA sensor data with BAM measurements. This research highlights the importance of localized calibration models to enhance the reliability of low-cost air quality sensors in diverse environmental conditions, particularly in cold regions.