Low-Cost PM Sensor for Indoor Air Quality Monitoring: Improving Reliability and Accessibility of In-Situ Microenvironment-Based Calibration

JALIL MOKHTARIAN MOBARAKEH, C. Victoria McCrary, Sara Jones, Perry Hystad, Molly Kile, Parichehr Salimifard, Oregon State University

     Abstract Number: 202
     Working Group: Indoor Aerosols

Abstract
Monitoring and controlling indoor air quality (IAQ) is crucial for preventing health issues caused by exposure to particulate matter (PM). While low-cost sensors (LCS) have made PM measurement for IAQ monitoring more accessible, they require regular calibration and maintenance to ensure accuracy. This has hindered the integration of LCS into IAQ monitoring programs and the use of LCS data to measure PM concentration trends. While there are various methods for laboratory calibration of LCS, on-site calibration of LCS remains under-evaluated. In this study, we conducted co-location tests in five microenvironments within a school building using LCS (PurpleAir [PA]), reference (TSI Optical Particle Sizer [OPS]), and mid-range (IC Sentinel [ICS]) sensors. A linear regression model with PM concentration, absolute humidity (AH), and temperature as variables was developed for calibration. The average Pearson correlation (R²) between LCS and the reference in each individual microenvironment was <0.50 with a large coefficient of variation (CV>30%), indicating the considerable differences between microenvironments. The overall calibration using collected data from all five microenvironments resulted in R² >0.90, showing the improvement of the model generalization using a wider range of data. ANOVA analysis indicated that PM concentration, AH, and temperature significantly affect the calibration model's predictions, with PM concentration having the highest impact, followed by AH, and then temperature (p-value<0.01). These results underscore the need for regular on-site microenvironment-based calibration to provide an appropriate range for predictors. Lack of access to a reference sensor impedes such measures. However, our results indicate acceptable proximity between ICS and OPS data; the average R² between ICS and OPS in each individual microenvironment calibration was >0.70, and the R² of overall calibration using all microenvironments’ data was >0.90. This suggests possible application of mid-cost sensors to reduce reliance on more expensive sensors for LCS calibration.