Accuracy verification of low-cost CO2 concentration measuring devices for general use as a countermeasure against COVID-19 ========================================================================================================================== * Yo Ishigaki * Koji Enoki * Shinji Yokogawa ## Abstract Within the context of the COVID-19 pandemic, CO2 sensors that measure ventilation conditions and thereby reduce the risk of airborne infection, are gaining increasing attention. We investigated and verified the accuracy of 12 relatively low-cost sensor models that retail for less than $45 and are advertised as infection control measures on a major e-commerce site. Our results indicate that 25% of the tested sensors can be used to identify trends in CO2 concentration, if correctly calibrated. However, 67% of sensors did not respond to the presence of CO2, which suggests that a type of pseudo-technique is used to display the CO2 concentration. We recommend that these sensors are not suitable for infection prevention purposes. Furthermore, 58% of the investigated sensors showed significant responses to the presence of alcohol. Owing to the widespread use of alcohol in preventing the spread of infectious diseases, sensors that react to alcohol can display inaccurate values, resulting in inappropriate ventilation behavior. Therefore, we strongly recommended that these sensors not be used. Based on our results, we offer practical recommendations to the average consumer, who does not have special measuring equipment, on how to identify inaccurate CO2 sensors. ## Introduction In enclosed spaces with inadequate ventilation and air treatment, the airborne spread of SARS-CoV-2 can occur even at a distance of 6 feet or more from an infected person, especially as concentrations of very fine droplets and aerosol particles drift through the air [1]. Within this context, airborne transmission may have accounted for more than 50% of the disease transmission on the Diamond Princess, the cruise ship associated with the SARS-CoV-2 outbreak in Japan [2]. The concentration of indoor carbon dioxide (CO2) has attracted attention as a proxy for infection risk [3]. While outdoor fresh air has a constant CO2 concentration of approximately 400 ppm, human breath contains a much larger amount of CO2 at 30,000–40,000 ppm. Therefore, by measuring indoor CO2 concentration, it is possible to understand how much human breath is retained in the air. The American Society of Health Engineers (ASHE) states that there is a correlation between ventilation capacity and the risk of airborne infection from tuberculosis, measles, chicken pox, influenza, smallpox, and SARS [4]. The Wells-Riley equation (Equation 1) can be used to quantitatively evaluate the risk of airborne infection by bacteria and viruses [5-6]. ![Formula][1] In this equation, *P* is the probability that an infected person will be infected, *I* is the number of infected persons in the enclosed space, *p* is the respiratory volume of one person (m3/h), *q* is the rate of infectious droplets (/h), *t* is the time spent by the infected person in the space (h), *θ* is the time spent by the infected person in the space (h), *V* is the volume of the room (m3), and *Q* is the ventilation rate (m3/h). Thus, the Wells-Riley equation models the probability of exposure to infectious droplets emitted by an infected person and inhaled by an uninfected person through respiration as the risk quantity. As infectious droplets are accompanied by a high concentration of CO2, both of which are airborne, it is beneficial to use CO2 concentration as an indicator of the airborne infection risk. A number of researchers have suggested that the threshold ventilation rate for maintaining the effective reproduction rate of tuberculosis (the ratio of new secondary infected persons to the source of infection) below 1 corresponds to a CO2 concentration of 1,000 ppm. Specifically, Du et al. found that by improving the ventilation system in a poorly ventilated university building, where an outbreak of tuberculosis occurred (27 tuberculosis patients and 1,665 contacts), and by reducing the maximum CO2 concentration from 3204 ± 50 to 591-603 ppm, the secondary infection rate among new contacts could be reduced to zero (mean follow-up, 5.9 years). Furthermore, the incidence of tuberculosis among contacts could be reduced by 97% (95% CI: 50–99.9%) when the CO2 concentration remained below 1,000 ppm [7]. After comprehensively considering laws and regulations concerning CO2 concentration, the Occupational Hygiene and Ergonomics Subcommittee of the Japan Society for Occupational Health (JSOH) established classifications for estimating ventilation status according to CO2 concentrations (Table 1), and recommended measures according to each classification [8]. View this table: [Table 1](http://medrxiv.org/content/early/2021/07/31/2021.07.30.21261265/T1) Table 1 Ventilation criteria based on CO2 concentrations, established by The Occupational Hygiene and Ergonomics Subcommittee of the Japan Society for Occupational Health (JSOH). Like those of other countries, the Japanese Building Standard Law (revised and implemented in 2003), requires the installation of a 24-hour ventilation system in all living rooms to prevent the development of the sick building syndrome, and requires an air change rate per hour (ACPH) of 0.5 times/h or more inside houses. With an ACPH of 0.5 times/h, for example, in a 70 m2 space with a standard activity level, the CO2 concentration at saturation is 800 ppm with two people in the room, 1,000 ppm with three people in the room, and 1,200 ppm with four people in the room. While the Building Standard Law specifies the amount of ventilation, Japan’s Building Management Law (enacted in 1970) specifies CO2 concentration standards. These laws specify that in buildings used as entertainment venues, department stores, restaurants, shops, offices, and schools, where the enclosed area used for specific purposes is 3,000 m2 or more, defined as “specified buildings” (> 8,000 m2 for schools), the CO2 concentration should be 1,000 ppm or less. Against this background, the monitoring of CO2 concentration by sensors is currently being advocated to reduce the risk of COVID-19 in indoor environments. In many instances, especially in the context of a pandemics, it is recommended to maintain indoor CO2 concentrations below 800 ppm (e.g., Minnesota Department of Health, Federation of European Heating Ventilation and Air Conditioning Associations, UK Scientific Advisory Group for Emergency, US Centers for Disease Control and Prevention) [9]. In Japan, a widespread system is used, in which local government supports the cost of installing CO2 sensors in restaurants to prevent the spread of COVID-19. For example, the Tokyo Metropolitan Government, Kyoto Prefecture, and Osaka City support the purchase of infection prevention devices, including CO2 sensors, up to 100,000– 300,000 yen. In addition, Shibuya Ward in Tokyo distributes free CO2 sensors to restaurants located in the ward. In response to this sudden increase in demand, low-cost CO2 sensors have rapidly become available on e-commerce sites such as Amazon, and are advertised as being effective in preventing the spread of infectious diseases. Many of these sensors have vague descriptions of their measurement principles, with several consumers questioning the CO2 measuring ability of these sensors. In addition, there is a lack of guidelines for the use of CO2 sensors in Japan, and a lack of laws or regulations regarding their accuracy and reliability. In response to this research need, we purchased several low-cost CO2 sensors, to verify their accuracy and investigate the related measurement principles. Although several studies have already evaluated low-cost gas sensors [10, 11], including CO2 sensors, the global outbreak of COVID-19 has led to a proliferation of new products on the market for the control of infectious diseases that have not been tested for efficacy and accuracy. Therefore, we conducted this study to investigate the accuracy and reliability of these sensors and to offer advice on these sensors to the general consumer, who does not have access to specialized inspection equipment. ## Methodology In this study, we purchased and investigated 12 relatively low-cost CO2 sensors (1–12; Table 2), within a price range of 2,900–4,999 JPY (approximately 26–45 USD, at the exchange rate in July 28, 2021) from the Amazon Japan website, and installed them according to instructions. The sensors were advertised as being effective in preventing COVID-19 infections. In addition to the 12 sensors, we purchased two relatively expensive CO2 sensors, (A and B) that are widely used for industrial and experimental purposes as reference devices. Sensor A is manufactured by the T&D corporation TR-76Ui (54,780 JPY, approximately 500 USD), with a zero and 1500 ppm span calibration, while Sensor B is manufactured and factory calibrated by the C.H.C. System corporation NMA-PR-R (47,500 JPY, approximately 430 USD). Both sensors A and B are based on the non-dispersive infrared (NDIR) method. View this table: [Table 2](http://medrxiv.org/content/early/2021/07/31/2021.07.30.21261265/T2) Table 2 List of CO2 sensors used in the experiment. All 14 sensors (1 to 12, A and B) were activated simultaneously and placed in a chamber (Fig. 1). The following protocol was used for the experiment: ![Figure 1.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2021/07/31/2021.07.30.21261265/F1.medium.gif) [Figure 1.](http://medrxiv.org/content/early/2021/07/31/2021.07.30.21261265/F1) Figure 1. Image of experimental equipment and setup used for an investigation of the accuracy of 12 CO2 sensors. 1. Open the chamber lid. 2. Turn on the fan. 3. Ventilate the chamber with room air (adjusted to 25°C) for 5 min. 4. Close the chamber lid. 5. Turn off the fan. 6. Record the displayed values of each measuring instrument. However, if a value outside the measurement range is displayed, it is not recorded. 7. Turn on the fan. 8. Inject (additional) CO2 gas, (additional) room air and/or alcohol into the chamber to change the gas concentration in the chamber. 9. Wait for 5 min until the inner air and gas are evenly mixed in the chamber. 10. Turn off the fan. 11. Record the displayed values of each measuring instrument. However, if a value outside the measurement range is displayed, it is not recorded. 12. Return to step 7 (or terminate the experiment). ## Results and Discussion We compared the CO2 values of the 12 investigated sensors to those of the reference sensors (A and B; Fig. 2). The relatively expensive sensor B showed a narrower measurement range than sensor A, although both had very similar response characteristics. Sensor 10 generally displayed higher values than sensor A, while sensors 4 and 7 displayed lower values than sensor A. Although sensor 2 responded to CO2, it seemed to malfunction, as it displayed very high CO2 values. Sensors 1, 3, 5, 6, 8, 9, and 12 did not seem to respond to CO2 levels. As these seven sensors are able to display total volatile organic compound (tVOC) values, we conducted an additional experiment by applying approximately 5 ml of rubbing alcohol to the inside of the acrylic plate of the chamber. ![Figure 2.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2021/07/31/2021.07.30.21261265/F2.medium.gif) [Figure 2.](http://medrxiv.org/content/early/2021/07/31/2021.07.30.21261265/F2) Figure 2. Response characteristics of CO2 sensors. Measured values of sensor A, the reference device, are plotted on the horizontal axis, and the measured values of sensors A, B, and 1 to 12 are plotted on the vertical axis, connected by spline curves. During this experiment, rather than adding CO2 gas, 5 ml of alcohol was sprayed on the acrylic wall in the chamber (Fig. 3). To eliminate the cross- recruitment and contamination effect of alcohol, the experiment was performed independently to the previous experiment. ![Figure 3.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2021/07/31/2021.07.30.21261265/F3.medium.gif) [Figure 3.](http://medrxiv.org/content/early/2021/07/31/2021.07.30.21261265/F3) Figure 3. Response characteristics of 14 CO2 sensors to alcohol, including two reference sensors (A and B). “1. Air” on the x-axis corresponds to the sixth step (FRESH AIR) in the experimental protocol, while “2. Alcohol” refers to the twelfth step of the protocol (CONTROLLED AIR), with 5 ml of alcohol added to the chamber at the step 8 (INJECTION) instead of CO2. Our results (Fig. 3) indicate that the readings of sensors A, B, 2, 4, 6, 7, and 10 were not affected by the presence of alcohol. On the other hand, sensors 1, 3, 8, 9, 11, and 12 reacted strongly to alcohol. Furthermore, these six sensors were part of the group of seven sensors that did not respond to CO2 (Fig. 2). The remaining sensor 6 appeared to respond neither to CO2 nor to alcohol. Of the 12 relatively low-cost sensors investigated in this study, three models (sensors 4, 7, and 10; 25% of the total) can potentially be used to determine CO2 concentration levels, if correctly calibrated. However, the accuracy of the sensors was not as good as that of the more expensive reference sensors. Machine learning methods have been proposed for calibration of inexpensive sensors [12, 13]. Sensor 2 seemed to be malfunctioning (8% of the total), while eight sensor models (67%) did not respond to CO2, and seven of these (58% of the total) responded strongly to alcohol, suggesting that the sensors were equipped with different sensors to the CO2 sensing module. To investigate the sensing module, we purchased and disassembled an extra model of sensor 12 (Fig. 4). Notably, the NDIR sensor was not present. Based on the model number of the sensing module and the manufacturer’s specifications [14], we identified that the sensor was not designed for CO2 measurements, but rather for the measurement of “ammonia, hydrogen, alcohol, carbon monoxide, methane, propane, Gan, styrene, propylene glycol, alkyl phenol, toluene, ethylbenzene, xylene, formaldehyde, and other volatile organic gases, incense smoke, wood, paper smoke burning out.” This suggests that sensor 12 is designed to display the concentration of CO2 by a type of imitating algorithm based on the value of the sensing module that detects miscellaneous gases, excluding CO2. ![Figure 4.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2021/07/31/2021.07.30.21261265/F4.medium.gif) [Figure 4.](http://medrxiv.org/content/early/2021/07/31/2021.07.30.21261265/F4) Figure 4. Sensing module of Sensor 12. A cylindrical object was suspended in the air by four wire bonds, and inside it was a coil heater. Globally, several sensors display a very vaguely defined measurement called eCO2 (equivalent CO2). This is estimated from the tVOC value and has been shown to differ from the actual CO2 concentration [15]. Sensors 1, 3, 8, 9, 11, and 12, which responded strongly to alcohol (Fig. 3), all displayed tVOC values separately from the CO2 concentration. Therefore, we suspected that these six sensor models were displaying eCO2 values rather than actual CO2 concentrations. However, the sales pages for these six sensors did not indicate that they used eCO2 sensors. Based on the results of our experiments, we developed a simple method for consumers to verify the quality of sensors on offer, without needing specialized equipment. Within this context, we recommend the following: 1. Ensure that the sensor displays a value of approximately 400 ppm in fresh air outdoors (this will allow for the detection of faulty sensors such as sensor 2, with a large deviation in CO2 values). 2. Ensure that the CO2 concentration rises significantly when breathing on the sensor (this will help identify sensors that do not respond to increases in CO2, such as Sensors 1, 3, 5, 6, 8, 9, and 12). 3. Ensure that the sensor does not react erroneously to alcohol by holding a hand dipped in an alcohol disinfectant close to the sensor (this eliminates low-quality sensors that pseudo-indicate CO2 concentration from tVOC concentrations such as alcohol). ## Conclusions Twenty-five percent of the relatively low-cost sensors investigated in this study have the potential to be used for identifying trends in CO2 concentration, if correctly calibrated, although with a poor accuracy. Notably, 67% of the sensors did not respond to changes in CO2 levels, which suggests that various pseudo- masquerading techniques are being used to display CO2 concentrations. These sensors should not be used to prevent the spread of infectious diseases. In addition, 58% of the sensors reacted strongly to alcohol. As alcohol is widely used indoors to prevent the spread of COVID-19, sensors that respond to alcohol will display inflated values, which can lead to incorrect ventilation behavior. Therefore, we strongly recommended that these sensors not be used. This study was conducted using a total of 14 commercially available sensor models. However, there are many models of CO2 sensors available in the market. We suggest that future research focus on an exhaustive analysis of the sensors available in the market as a basis for a discussion on the guidelines and regulations for CO2 sensors. We are planning to use gas chromatography with an accuracy of ±0.1 ppm to simultaneously detect the concentration of alcohol and carbon dioxide for more accurate verification. With this method, we will be able to measure both organic and inorganic gases, which was not possible before. In addition, clarification of the electrical circuit, catalyst, semiconductor composition, and solid-state properties of the sensing modules (Fig. 4) will assist to pinpointing the purpose for which these sensors were developed. Furthermore, we suggest that future research focus on not only conducting exposure experiments with pure CO2 gas, but also include accurate measurements with human-derived CO2 gases, including various volatile organic compounds. ## Data Availability The raw data of the graphs are available on FigShare. [https://figshare.com/articles/figure/Accuracy\_verification\_of\_low-cost\_CO2\_concentration\_measuring\_devices\_for\_general\_use\_as\_a\_countermeasure\_against\_infectious\_diseases/15067557](https://figshare.com/articles/figure/Accuracy\_verification\_of\_low-cost\_CO2\_concentration\_measuring\_devices\_for\_general\_use\_as\_a\_countermeasure_against_infectious_diseases/15067557) ## Notes The authors declare no conflicts of interest. This paper has been previously submitted to medRxiv, a preprint server for health sciences. This study was approved by the Ethics Committee on Experiments on Human Subjects (approval number of 21005), The University of Electro-communications, located at Chofugaoka 1-5-1, Chofu, Tokyo, Japan. ## Acknowledgments This work was supported by the Research Grant Program of the KDDI Foundation. In addition, some of the data acquisition technologies in this study are based on results obtained from a project commissioned by the New Energy and Industrial Technology Development Organization (NEDO). ## Footnotes * Declaration of competing financial interests (CFI): * The authors indicate no conflicts of interest. * Received July 30, 2021. * Revision received July 30, 2021. * Accepted July 31, 2021. * © 2021, Posted by Cold Spring Harbor Laboratory This pre-print is available under a Creative Commons License (Attribution-NonCommercial-NoDerivs 4.0 International), CC BY-NC-ND 4.0, as described at [http://creativecommons.org/licenses/by-nc-nd/4.0/](http://creativecommons.org/licenses/by-nc-nd/4.0/) ## References 1. 1).CDC (Centers for Disease Control and Prevention), SARS-CoV-2 Transmission, Summary of recent changes, Updated May 7, 2021 Online: [https://www.cdc.gov/coronavirus/2019-ncov/science/science-briefs/sars-cov-2-transmission.html](https://www.cdc.gov/coronavirus/2019-ncov/science/science-briefs/sars-cov-2-transmission.html) (accessed 25 July 2021) 2. 2).Azimi P, Keshavarz Z, Laurent JGC, Stephens B, Allen JG. Mechanistic transmission modeling of COVID-19 on the Diamond Princess cruise ship demonstrates the importance of aerosol transmission. Proc Natl Acad Sci 2021; 118(8) e2015482118; DOI: 10.1073/pnas.2015482118. [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NDoicG5hcyI7czo1OiJyZXNpZCI7czoxNzoiMTE4LzgvZTIwMTU0ODIxMTgiO3M6NDoiYXRvbSI7czo1MDoiL21lZHJ4aXYvZWFybHkvMjAyMS8wNy8zMS8yMDIxLjA3LjMwLjIxMjYxMjY1LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 3. 3).Memarzadeh F. Room Ventilation and Airborne Disease Transmission, American Hospital Association, Editor: The American Society for Healthcare Engineering; 2013. 4. 4).Wells WF. Airborne contagion and air hygiene. Harvard: Harvard University Press, 1955. 5. 5).Riley EC, Murphy G, Riley RL. Airborne spread of measles in a suburban elementary school. Am J Epidemiol 1978; 107: 421–432. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1093/oxfordjournals.aje.a112560&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=665658&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2021%2F07%2F31%2F2021.07.30.21261265.atom) [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=A1978EZ85600006&link_type=ISI) 6. 6).Buonanno G, Stabile L, Morawska L. Estimation of airborne viral emission; Quanta emission rate of SARS CoV-2 for infection risk assessment. Environ Int 2020; 141: 1–8. 7. 7).Du CR, Wang SC, Yu MC, Chiu TF, Wang JY, Chuang PC, Jou R, Chan PC, Fang CT. Effect of ventilation improvement during a tuberculosis outbreak in underventilated university buildings, Indoor Air 2019: 30(3); 422–432. 8. 8).The Occupational Hygiene and Ergonomics Subcommittee of the Japan Society for Occupational Health (JSOH): Ventilation simulator for COVID-19 countermeasures (COVID-19 Taisaku-you Kanki Simulator). Online: [http://jsoh-ohe.umin.jp/covid\_simulator/ventilation\_simulator\_manual\_ver1.0.pdf](http://jsoh-ohe.umin.jp/covid\_simulator/ventilation_simulator_manual_ver1.0.pdf) (accessed 25 July 2021) 9. 9).Eykelbosh, A. Indoor CO2Sensors for COVID-19 Risk Mitigation: Current Guidance and Limitations. Vancouver, BC: National Collaborating Centre for Environmental Health. 2021 May. Online: [https://ncceh.ca/documents/field-inquiry/indoor-co2-sensors-COVID-19-risk-mitigation-current-guidance-and](https://ncceh.ca/documents/field-inquiry/indoor-co2-sensors-COVID-19-risk-mitigation-current-guidance-and) (accessed 25 July 2021). 10. 10).Lewis AC, Lee JD, Edwards PM, Shaw MD, Evans MJ, Moller SJ, Smith KR, Buckley JW, Ellis M, Gillot SR, Whited A. Evaluating the performance of low cost chemical sensors for air pollution research. Faraday Discuss 2016; 189 Online: [https://doi.org/10.1039/C5FD00201J](https://doi.org/10.1039/C5FD00201J) (accessed 25 July 2021). 11. 11).Castell N, Dauge FR, Schneider P, Vogt M, Lerner U, Fishbain B, Broday D, Bartonova A. Can commercial low-cost sensor platforms contribute to air quality monitoring and exposure estimates? Environ Int 2017; 99: 293-302, ISSN 0160-4120, 2017 Online: [https://doi.org/10.1016/j.envint.2016.12.007](https://doi.org/10.1016/j.envint.2016.12.007) (accessed 25 July 2021) 12. 12).Zimmerman N, Presto AA, Kumar SPN, Gu J, Hauryliuk A, Robinson ES, Robinson AL, Subramanian R. A machine learning calibration model using random forests to improve sensor performance for lower-cost air quality monitoring. Atmos Meas Tech, 2018; 11:291–313. Online: [https://doi.org/10.5194/amt-11-291-2018](https://doi.org/10.5194/amt-11-291-2018) (accessed 25 July 2021) 13. 13).Zaidan MA, Motlagh NH, Fung PL, Lu D, Timonen H, Kuula J, Niemi JV, Tarkoma S, Petäjä T, Kulmala M, Hussein T. Intelligent Calibration and Virtual Sensing for Integrated Low-Cost Air Quality Sensors. IEEE Sensors Journal 2020; 20(22): 13638-13652 doi: 10.1109/JSEN.2020.3010316. Online: [https://ieeexplore.ieee.org/document/9144227](https://ieeexplore.ieee.org/document/9144227) (accessed 25 July 2021) [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1109/JSEN.2020.3010316&link_type=DOI) 14. 14).TPM-300 Air Quality Module Intelligent Air Quality Monitoring Module Air Sensor Online: [https://ardubotics.eu/en/sensors/1243-tpm-300-air-quality-module-intelligent-air-quality-monitoring-module-air-sensor.html](https://ardubotics.eu/en/sensors/1243-tpm-300-air-quality-module-intelligent-air-quality-monitoring-module-air-sensor.html) (accessed 25 July 2021) 15. 15).Varzaru G, Zarnescu A, Ungurelu R, Secere M. Dismantling the confusion between the equivalent CO2and CO2concentration levels. 11th International Conference on Electronics, Computers and Artificial Intelligence (ECAI) 2019: pp. 1–4, doi: 10.1109/ECAI46879.2019.9042113. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1109/ECAI46879.2019.9042113&link_type=DOI) [1]: /embed/graphic-1.gif