Low cost pollution sensors have been widely publicized, in principle offering increased information on the distribution of air pollution and a democratization of air quality measurements to amateur users. We report a laboratory study of commonly-used electrochemical sensors and quantify a number of cross-interferences with other atmospheric chemicals, some of which become significant at typical suburban air pollution concentrations. We highlight that artefact signals from co-sampled pollutants such as CO2 can be greater than the electrochemical sensor signal generated by the measurand. We subsequently tested in ambient air, over a period of three weeks, twenty identical commercial sensor packages alongside standard measurements and report on the degree of agreement between references and sensors. We then explore potential experimental approaches to improve sensor performance, enhancing outputs from qualitative to quantitative, focusing on low cost VOC photoionization sensors. Careful signal handling, for example, was seen to improve limits of detection by one order of magnitude. The quantity, magnitude and complexity of analytical interferences that must be characterised to convert a signal into a quantitative observation, with known uncertainties, make standard individual parameter regression inappropriate. We show that one potential solution to this problem is the application of supervised machine learning approaches such as boosted regression trees and Gaussian processes emulation.
Article:Smith, Katie R., Edwards, Peter M. orcid.org/0000-0002-1076-6793, Evans, Mathew J. orcid.org/0000-0003-4775-032X et al. (5 more Low cost air pollution sensors have substantial potential for atmospheric research and for the applied control of pollution in the urban environment, including more localized warnings to the public. The current generation of single-chemical gas sensors experience degrees of interference from other co-pollutants and have sensitivity to environmental factors such as temperature, wind speed and supply voltage. There are uncertainties introduced also because of sensor-to-sensor response variability, although this is less well reported. The sensitivity of Metal Oxide Sensors (MOS) to volatile organic compounds (VOCs) changed with relative humidity (RH) by up to a factor of five over the range 19-90%RH and with an uncertainty in the correction of a factor two at any given RH. The short-term (second to minute) stabilities of MOS and electrochemical CO sensor responses were reasonable. During more extended use inter-sensor quantitative comparability was degraded due to unpredictable variability in individual sensor responses (to either measurand or interference or both) drifting over timescales of several hours to days. For timescales longer than a week identical sensors showed slow, often downwards, drifts in their responses which diverged across six CO sensors by up to 30% after two weeks. The measurement derived from the median sensor within clusters of 6, 8 and up to 21 sensors was evaluated against individual sensor performance and external reference values. The clustered approach maintained the cost competitiveness of a sensor device, but the median concentration from the ensemble of sensor signals largely eliminated the randomised hour-to-day response drift seen in individual sensors and excluded the effects of small numbers of poorly performing sensors that drifted significantly over longer time periods. The results demonstrate that for individual sensors to be optimally comparable to one another, and to reference instruments, they would likely require frequent calibration. The use of a cluster median value eliminates unpredictable medium term response changes, and other longer term outlier behaviours, extending the likely period needed between calibration and making a linear interpolation between calibrations more appropriate. Through the use of sensor clusters rather than individual sensors existing low cost technologies could deliver significantly improved quality of observations.
Low-cost sensors (LCSs) are an appealing solution to the problem of spatial resolution in air quality measurement, but they currently do not have the same analytical performance as regulatory reference methods. Individual sensors can be susceptible to analytical cross-interferences; have random signal variability; and experience drift over short, medium and long timescales. To overcome some of the performance limitations of individual sensors we use a clustering approach using the instantaneous median signal from six identical electrochemical sensors to minimize the randomized drifts and inter-sensor differences. We report here on a low-power analytical device (< 200 W) that is comprised of clusters of sensors for NO 2 , O x , CO and total volatile organic compounds (VOCs) and that measures supporting parameters such as water vapour and temperature. This was tested in the field against reference monitors, collecting ambient air pollution data in Beijing, China. Comparisons were made of NO 2 and O x clustered sensor data against reference methods for calibrations derived from factory settings, in-field simple linear regression (SLR) and then against three machine learning (ML) algorithms. The parametric supervised ML algorithms, boosted regression trees (BRTs) and boosted linear regression (BLR), and the non-parametric technique, Gaussian process (GP), used all available sensor data to improve the measurement estimate of NO 2 and O x . In all cases ML produced an observational value that was closer to reference measurements than SLR alone. In combination, sensor clustering and ML generated sensor data of a quality that was close to that of regulatory measurements (using the RMSE metric) yet retained a very substantial cost and power advantage.
Supplement Table 1Components, mole fractions, and tolerances of Apel-Riemer Environmental Inc. (Broomfield, CO, USA) multicomponent standard (preparation date April 2010). Supplement Table 2Components, mole fractions, and tolerances of Apel-Riemer Environmental Inc. (Broomfield, CO, USA) multicomponent standard (preparation date summer 2014).
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.