Long-term behavior and stability of calibration models for NO and NO₂ low-cost sensors

Abstract

Low-cost sensors are considered to exhibit great potential to complement classical air quality measurements in existing monitoring networks. However, the use of low-cost sensors poses some challenges. In this study, the behavior and performance of electrochemical sensors for NO and NO2 were determined over a longer operating period in a real-world deployment. After careful calibration of the sensors, based on co-location with reference instruments at a rural traffic site during 6 months and by using robust linear regression and random forest regression, the coefficient of determination of both types of sensors was high (R2 > 0.9), and the root mean square error (RMSE) of NO and NO2 sensors was about 6.8 and 3.5 ppb, respectively, for 10 min mean concentrations. The RMSE of the NO2 sensors, however, more than doubled when the sensors were deployed without recalibration for a 1-year period at other site types (including urban background locations), where the range and the variability of air pollutant concentrations differed from the calibration site. This indicates a significant effect of relocation of the sensors on the quality of their data. During deployment, we found that the NO2 sensors are capable of distinguishing general pollution levels, but they proved unsuitable for accurate measurements, mainly due to significant biases. In order to investigate the long-term stability of the original calibration, the sensors were reinstalled at the calibration site after deployment. Surprisingly, the coefficient of determination and the RMSE of the NO sensor remained almost unchanged after more than 1 year of operation. In contrast, the performance of the NO2 sensors clearly deteriorated as indicated by a higher RMSE (about 7.5 ppb, 10 min mean concentrations) and a lower coefficient of determination (R2 = 0.59).