Evaluation of air quality sensors for environmental epidemiology

The increase in evidence on the adverse health effects of air pollution has been possible thanks to the advances in technology for air pollution monitoring and personal exposure. In this context, air quality sensors present a huge potential for enhancing long-term personal exposure prediction. In order to prove the potential of air sensors for health research, a pilot study with patients suffering from chronic obstructive pulmonary diseases or Asthma was carried out in Stuttgart (Germany) in cooperation with the University Hospital Charité in Berlin, and an outpatient pulmonary practice in Stuttgart.

Prior to the pilot study, we first tested several sensor models for PM and NO₂ in the laboratory as well as in the field. The sensors with the best performance were selected to build two different sensor boxes for indoor and outdoor measurements. Temperature and relative humidity sensors were also included in the boxes. To avoid the overestimation of the PM readings due to hygroscopic growth, a low-cost dryer was designed and evaluated for the PM sensor of the outdoor boxes (Chacón-Mateos et al. 2022).

The measurements inside and outside the houses were carried out over 30 days. Passive sampling of NO₂ was done additionally for quality assurance of the NO₂ measurements. Participants completed spirometry, a questionnaire assessing respiratory symptoms, and a protocol of activities on a daily basis. The calibration and validation of the sensors were conducted two weeks before the start of the measurements. To perform the quality assurance, the sensor boxes for indoors were collocated in the laboratory, and the sensor boxes for outdoors in the field, together with reference-grade instruments. The PM2.5 concentrations were corrected using univariate linear regression whereas multilinear regression and machine learning algorithms were tested and applied to correct the raw data of the NO₂ sensors.

The sensor validations have shown that measuring low concentrations of PM2.5 and NO₂ has higher expanded uncertainty, but high concentrations can be measured with expanded uncertainties that fulfill the Data Quality Objectives (DQO) set by the Air Quality Directives for indicative measurements. However, we also found a high unit-to-unit variability which means that model coefficients cannot be transferred from one sensor to another. We also recommend collocating NO₂ passive samples close to the sensor boxes to determine whether or not the sensor is measuring within the expected range as the signal of some NO₂ sensors drifted unexpectedly during the measurements in the houses of the patients.

In conclusion, air sensors are not yet to be recommended to quantify the effects of low-level air pollution but they are a promising tool to increase the ubiquity of epidemiological studies for low- and middle-income countries where high air pollution is expected. Moreover, it is important to consider that in order to get actionable data with air sensors, a significant amount of time in sensor calibration should be invested.

References:

Chacón-Mateos, M., Laquai, B., Vogt, U., and Stubenrauch, C.: Evaluation of a low-cost dryer for a low-cost optical particle counter, Atmos. Meas. Tech., 15, 7395–7410, https://doi.org/10.5194/amt-15-7395-2022, 2022