Field-based calibration and operation of low-cost sensors for particulate matter by linear and nonlinear methods

The increasing awareness of air pollution's detrimental effects has driven the demand for affordable air quality monitoring solutions, particularly low-cost fine particulate matter (PM_2.5) sensors. However, these sensors often suffer from low data accuracy and require rigorous calibration, especially in real-world settings. This study evaluates the field calibration of low-cost PM_2.5 sensors under low ambient concentration conditions, utilizing both linear and nonlinear regression methods. The research was conducted in Sydney, Australia, where data were collected from both low-cost Hibou sensors and a research-grade DustTrak monitor. Our analysis compares calibration performance across various time resolutions, meteorological factors, and traffic conditions. The results indicate that nonlinear models significantly outperform linear models, achieving an R² of 0.93 at 20-min resolution, surpassing the U.S. EPA's calibration standards. Additionally, our findings suggest that temperature, wind speed, and heavy vehicle density are the most influential factors in calibration accuracy. After comparing the corrected measurement data with WHO standards, it was observed that PM_2.5 concentrations at the bus stop measurement site ranged from 7 to 76 μg/m³, with 24 % of the data exceeding the WHO 24-h standard. This finding highlights that traffic-generated PM_2.5 pollution remains a significant concern in Sydney. The study concludes that nonlinear calibration methods are more effective for low-cost PM_2.5 sensor deployment in urban environments, though further exploration is needed to enhance the interpretability and computational efficiency of deep learning models.