Maximizing insights from air quality sensor networks through continuous performance evaluation
Presented by: Daniel (Dan) Peters, Environmental Defense Fund
Description: As the global use of lower-cost air quality sensors (LCS) continues to grow, data quality issues including poor accuracy, drift, and environmental interference remain a critical obstacle to a user’s ability to obtain meaningful information from sensor measurements. Even if a sensor has already been tested and calibrated in a lab or field environment, a sensor’s performance and calibration parameters may vary by region (or season) depending on numerous factors including the local characteristics of air pollution and meteorology. Due to the uncertainty and variability in LCS data quality, with no universal performance standards or QA/QC protocol, the end user is responsible for ensuring the data they obtain is robust and of appropriate quality for the intended application. Here, we present results from a network of 100 electrochemical NO2 sensors as part of the Breathe London pilot project (BL) and describe the steps that were taken to develop a detailed understanding of sensor performance over the entire duration of the 26-month measurement campaign. We discuss how this project-long understanding of sensor performance informed our interpretation of the data produced by the BL network. We validate our findings by comparing BL network results to data from an extensive network of London reference monitors, including comparisons of temporal patterns (e.g., diurnal, day-of-week, and monthly averages) as well as differences in pollution profiles between near-road and urban background monitoring sites. Through these comparisons, we show how long-term collocation intercomparisons of BL sensors at two reference sites could be used as a proxy to understand overall network performance and develop corrections. We discuss how the findings could be applied to other LCS projects, including those with less reference monitoring infrastructure, provided that some source of traceable, reliable measurements is available to perform ongoing representative evaluations of sensors.