From Air Quality Sensor to Sensor Network: Things We Need to Learn

Abstract:

As an essential complement to traditional regulation instruments, low-cost air quality sensors (LCAQS) can be deployed as dense monitor networks to provide timely and comprehensive snapshots of pollutant concentrations and spatial and temporal variability at various scales with minimal cost and labor. Even though the LCAQS have shown promising performances in previous studies, they still face many challenges, mostly regarding data quality. To overcome these issues, machine learning algorithms were utilized to calibrate the sensor data appropriately, mostly during post-processing. The current emergence of cloud computing has made such calibration smoother and faster. In this project, we studied data from an SCI-608 (LCAQS manufactured by Sailbri Cooper Inc.) collected from the several U.S. cites and the EPA Air Quality System (AQS) data in 2017, 2018, and 2019. As a result, we concluded our experiences and findings in 1) systematic errors in reference methods and their effects on the evaluation of LCAQS performance, 2) the improvement of LCAQS data quality by in situ calibration, with machine learning algorithms during high-pollution episodes, and 3) different pollutants show varied heterogeneity patterns in cities, which need to take into account when designing the local LCAQS network.