In situ drift correction for a low-cost NO2 sensor network

Abstract

Twelve low-cost NO₂ sensors (LCSs) underwent multiple bi-weekly collocations with a NO₂ reference monitor to develop a robust calibration model unique to several periods throughout a summer deployment (April 2021, to October 2021). It was found that a single calibration based on an initial two-week collocation would not hold up in the variable environmental conditions that Phoenix, Arizona experiences in the summer. Temperature, relative humidity, and ozone prove to be critical parameters that would need to be corrected for in the calibration model. Therefore, we developed a period-specific calibration that is re-trained every six weeks to better account for the conditions during that period. This calibration improved sensor performance compared to the sensor manufacturer calibration, yielding an average root-mean-square error (RMSE) of 3.8 ppb and an R² of 0.82 compared to 19.5 ppb and 0.42 when evaluated against reference NO₂ measurements. This improved calibration allowed for more accurate NO₂ measurements utilizing LCSs in a sensor network that would not be possible from the out-of-the-box calibration.