In-situ validation of embedded physics-based calibration in low-cost particulate matter sensor for urban air quality monitoring

Abstract

Low-cost particle sensors enable dense, geospatially distributed networks that enhance the spatial and temporal resolution of urban air quality monitoring. However, field interference in complex urban systems challenges the reliability of sensor data. Robust evaluation and calibration are essential to address this issue. In this study, a low-cost sensor system was deployed near standard monitoring stations from March 1 to May 30, 2024, recording PM_2.5 concentration, PM₁₀ concentration, particle counts in six different size channels, and ambient temperature and humidity. The results revealed systematic overestimation and interactions with environmental factors in the sensor data. To address these challenges, a physics-based calibration model, leveraging sensor-reported particle size information, was developed and compared with traditional empirical and machine learning models. These calibration models were embedded into the sensor system, followed by a second field campaign from June 1 to 30. While the machine learning model achieved the best performance during the first campaign (R² > 0.90, RMSE <10 μg/m³ for PM_2.5 and PM₁₀), its generalization ability was limited. The physics-based model, however, excelled on a new dataset from the second campaign, demonstrating robust performance and strong generalization across urban conditions. These findings highlight the potential of the physics-based calibration model to improve the reliability and sustainability of urban air quality monitoring by integrating it into the embedded systems of low-cost sensors. This approach offers enhanced stability and applicability in complex urban environments, providing a more effective calibration method for urban environmental systems.