An experimental approach combining internet of things (IoT) hardware and software development was used to measure air pollutants and air quality metrics. The researchers used AMS to nonintrusively monitor air quality through daily indoor life. AMS provided visuals and recorded trends that could indirectly indicate the relationship between lifestyle and observed pollutant values.

In this work, the researchers used AMS to collect air quality data indoors at home (in Cupertino, California; South Lake Tahoe, California; and Hyderabad, Telangana, India) as well as during local commute in the United States and during an inflight journey from the United States to India. These locations were chosen to monitor and evaluate the performance of AMS in environments associated with distinct indoor air quality profiles and climates. The device, similar to a thermostat placed in a room, is completely noninvasive. To initiate readings, it is powered on with a USB cable (or a wall socket) and AMS software is launched by the user. The studies were conducted, intermittently, between January 2020 and January 2021.

After initial calibration, to ensure the continued accuracy of the measurements made by AMS on an ongoing basis, the sensors were recalibrated every 3 months. Measurements were repeated over similar activities in several different timeframes to better analyze and predict possible relationships between activities and associated pollutants. To control for bias, measurements were also carried out across several recalibrations.

AMS comprises 3 distinct modules that collect, analyze and store, and display data. As shown in Figure 1, these 3 interconnected modules include the data collection system (DCS), data analysis and storage engine (DASE), and user interface (UI). The DCS is a sensor system that collects the user’s environmental data in user-defined intervals. The data gathered are then analyzed and stored by the system’s DASE, and the air quality metrics obtained are displayed by the system’s web-based UI.

Materials used to build the DCS and the software used to create DASE and UI are discussed in the following sections.

In this section, the methods involved in calibrating the DCS device, collecting data from a setting, and posting data to DASE will be discussed. Moreover, the analyzed data can be viewed on UI screens for the given set up.

The data analysis flowchart is depicted in Figure 8. Analysis of the collected data from the DCS is automatically executed in the background and is transparent to the user. Upon receiving a request from the UI, or through the background user alert process, the database is queried to retrieve data. The data are validated against thresholds predetermined by environmental safety limits [52,53]. If the user is registered to receive alerts and the values exceed safe thresholds, an alert is sent to the user. Additionally, the values and corresponding qualitative metrics of each pollutant are determined. Finally, AQI is computed and the data are returned.

The authors considered including an adaptive threshold for each analyte to account for spatial, temporal, and environmental variations. However, AMS is recalibrated in each new usage location and is intended for indoor use. Thus, environmental variations are small and no substantial error between AMS and COTS sensors was observed when analyzing the data. Nonetheless, it is well-known that MQ sensors exhibit high temperature coefficients and sensitivity to humidity [54]. Drift of these sensors is low and ongoing recalibration every 3 months is more than sufficient to address such drift [54]. Considering these factors, the ranges at which the AQI considered are absolute, but the unit itself is adjusted to account for location. Further, the sample rate is considered sufficient for the application. In the data that follow, it does not appear that the sensors are undersampled, so temporal variation is not considered in this embodiment. Nonetheless, one of the goals for the future work is to automate the calibration process and also allow customization of the calibration interval according to the user’s choice.

If a user opts to be alerted when pollutants are above the optimal threshold, DASE sends an email and SMS text message alerts. This could be invaluable in preventing accidents or calling for emergency services in the case of an NG leakage or similar emergency.

The purpose of this work was to measure various pollutants and other air quality metrics that affect individual environmental atmosomes. The following section presents the results of a variety of air quality metrics in selected environments and relates them with the conditions in their atmosome.

Figures 15 and 16 show the quantity and value of readings of relative humidity and temperature, respectively, taken at 3 different indoor locations: a home in Cupertino, California (shown in blue); a home in Hyderabad, Telangana, India (shown in orange); and an airplane economy cabin during a nonstop flight of 17 hours (shown in green). The low humidity readings from the airplane correlate with the dryness and discomfort often experienced by airplane passengers and align with the United States Centers for Disease Control and Prevention’s (CDC) air travel yellow book [56].

Figure 17 shows a graph of CO₂ measurements taken in the same environments as above. The readings in the home in the United States showed much higher indoor CO₂ levels than the readings in a more polluted area in India. Further analysis has revealed, however, that the closed windows and doors throughout the day during winter in the United States reduced ventilation and increased CO₂ concentration. Studies show that higher CO₂ exposure can cause drowsiness [57]. These data highlight the importance of ventilation during the winter. The readings on the lengthy airplane journey confirmed that the DCS readings and published values by the airline were within the range of each other. Figure 18 illustrates CO₂ readings in 2 rooms and shows the role AMS had in enabling the researcher to take corrective actions to improve indoor air quality.

Figure 19 depicts the variation in PM_2.5 during the course of a typical home activity (cooking food). The onset of cooking is associated with a sharp, transient increase in the levels of PM_2.5 followed by a sustained period of unhealthy PM_2.5 levels. In this particular scenario, the user responded to the elevated concentration of PM_2.5 by activating the exhaust fan, deactivating the stove, and opening all windows for improved cross-ventilation.

This set of mitigating actions lowered the PM_2.5 levels but unhealthy concentrations (defined by the horizontal orange and purple lines in Figure 19) persisted even after cooking ceased. These observational data highlight the power and utility of AMS in identifying and mitigating substantially unhealthy levels of indoor air quality. A longitudinal analysis of such data has the potential to offer rich insights into the interaction between indoor air pollutants and respiratory health outcomes, and aid data-driven health policy research.

Raw data stored in DASE were downloaded in .csv format using its REST API for advanced users. The data were graphed in a python notebook using matplotlib. As shown in Figure 20, there is a substantial difference between Indian and American household air pollution levels in terms of tVOC measurements. The difference in results could be attributed partly to the substantially higher levels of air pollution in India compared with those in the United States. The current results are consistent with previous results suggesting vehicle exhaust as one of the leading sources of VOC-related pollution in India [58].

Figure 21 displays a strong correlation between tVOC and ground-level O₃ measurements. For the most part, spikes in tVOC result in increases in O₃ levels that cause nitrogen oxides to react with tVOC in the presence of sunlight to create ground-level O₃.

Figure 22 displays tVOC readings recorded during various daily activities such as house cleaning, cooking at night, and commuting by car. The tVOC values were 1549, 2008, and 2868 ppb, respectively, for the aforesaid activities, well above the accepted moderate range of 220-660 ppb (presented in Figure 8), and demonstrate the high levels of tVOC during innocuous common activities [59]. The results also show that tVOC readings impact the indoor AQI value.

Studies carried out in 2003 by the California Air Resources Board show that cleaning products alone account for the release of 7.4 t of VOCs per day and that their various health effects include asthma attacks and eczema [60]. AMS can furnish necessary information to take precautions in such situations (eg, a spike in tVOC levels). Homes with infants or expecting parents often undergo more cleaning than typical. AMS, however, suggests that a presumably “well-cleaned” house may, in fact, pose a heightened risk of childhood asthma to infants [60], delays in child language development [61], or prenatal exposure to the fetuses that may impact their postnatal growth [62].

In summary, these results show that IAQ may be at unhealthy levels while conducting typical daily activities. AMS is able to track air quality through its sensors and indicate the impact of harmful levels of indoor air pollutants such as O₃, tVOC, CO₂ levels, and the presence of PM_2.5 inside homes.

This work has aimed to determine the performance of a low-cost AMS in indoor spaces in terms of gathering sensor data for a variety of air pollutants, transferring data reliably to an analysis engine in the cloud, displaying air quality monitoring results to the user, and sending alerts when the pollutants exceed safe thresholds. This system also provided a way for users to choose to share their IAQ data anonymously for further health research, and for users and researchers to retrieve data using REST APIs for further analysis and data analysis. This can be a foundation for building a public IAQ database across geographical regions.

The results establish that AMS is effective in analyzing multipollutant data streams in multiple settings, displaying AQI value, hourly visual and quantitative data of 18 pollutants, average exposure statistics of various pollutants, and the user’s quarterly pollutant exposure graphs. The dashboard also includes tips for each pollutant that display the GOOD, MODERATE, or POOR thresholds, enabling the user to better interpret the values and graphs shown in the UI and receive recommendations for keeping various pollutants in check with simple mitigation. The individual pollutant pages give the user further insights into the pollutant and show novel research in which AMS sensors and data collection have been employed. The system is also effective in sending alerts when pollutants exceed safe thresholds. The following sections discuss the findings of the work, which are corroborated by previous work and studies of a similar nature.

Particulate matter refers to mixtures of microscopic solid and liquid particles suspended in the air. There are 2 types of particulate matter that are most relevant to air pollution: PM₁₀ and PM_2.5. PM₁₀ consists of particles that are between 2.5 and 10 μm (diameter); some examples of these include dust, pollen, and particles of mold. PM_2.5 consists of fine particles that are 2.5 μm in diameter or less; some examples of these include fuel combustion, cigarette smoke, and aerosols. Particulate matter is a health risk because it is small enough to be inhaled and deposits itself in the airways of the human respiratory system. Smaller particles can lodge themselves deep in the lungs or enter the bloodstream. Even short-term exposure to PM_2.5 has been associated with worsening respiratory diseases and can lead to emergency care. Long-term exposure (ie, months to years) has been linked to premature death, especially in people with chronic conditions, and leads to reduced lung function in children [63]. The EPA set the maximum 24-hour exposure limit to PM_2.5 to 35 µg/m³ and annual exposure limit to 12 µg/m³ [64].

Quite interestingly, results of this work clearly indicate that when cooking is involved, the level of PM_2.5 increases well beyond recommended exposure limits and these levels remain high even after cooking has ceased. While it was observed that ventilation could mitigate levels to some extent, it is imperative that cooking methodologies are modified to reduce the emission of PM_2.5. A study that focused on emission of PM_2.5 from cooking in homes found that the levels were raised 20-40-fold to 160 µg/m³ in the kitchen and 10-fold in the nearby living room to 60 µg/m³ [65], clearly corroborating the results of this work. Measuring particulate matter in indoor air can direct researchers toward finding easily implementable corrective actions to reduce exposure levels.

CO₂ is considered to be a dominant air pollutant. Moderately high concentrations of CO₂ in indoor air can lead to drowsiness, fatigue, and headaches. Increasing amounts can cause dizziness and nausea [66]. In the experimental data, CO₂ levels increased dramatically in poorly ventilated rooms as shown in the temporal data in Figure 18. In both the rooms, air conditions were initially unhealthy and then improved drastically. In room 1, conditions became unhealthy again after some hours due to a room heater that did not circulate air. Taking advice furnished by AMS and adding a fan, along with regular ventilation, improved conditions again. These results corroborate the findings of a previous work in the Texas elementary schools that monitored CO₂ levels in 120 randomly selected classrooms in 2 school districts [67]. The simple process of improving ventilation and air circulation can reduce high concentrations of CO₂ and improve comfort.

The experimental results show that even seemingly routine daily activities can induce significant tVOC release. A variety of household items, such as candles, cooking fumes, room fresheners, cosmetics, cleaning products, and paints, emit VOCs. These are organic chemicals that are usually in gaseous form at room temperature and are photo-chemically active. Short-term exposure to VOCs can cause optic or respiratory irritation, headaches, memory lapses, and dizziness. Long-term exposure can cause nausea, fatigue, organ damage, and cancer [68]. Results shown in Figure 22 based on an AMS report of activities, such as cleaning, cooking, and commuting, corroborate the findings of previous work [69-72].

Readings taken while cleaning, cooking, and commuting, as shown in Figure 22, highlighted a correlation between activities that lead to elevated VOC levels in the presence of sunlight and elevated O₃ levels. In support, studies in Los Angeles have found as many VOCs being emitted from household products as from vehicle exhaust pipes, which then react in the presence of sunlight to produce ground-level O₃ [73]. These results match the EPA’s report that mentions the adverse effects of synthetic chemicals and emissions caused by cars, power plants, and other industrial setups that react in the presence of sunlight to increase ozone levels [74]. The results of this work show that the same levels of VOCs do not elevate O₃ as much at night, as there is no sunlight to facilitate such a chemical reaction.

Ground-level O₃ can trigger a variety of health problems, including chest pain, coughing, throat irritation, and airway inflammation. It can also worsen bronchitis, emphysema, and asthma [74]. This can lead to the need for increased medical care [75]. In 2016, 90% of noncompliance to the national ambient air quality standards in the United States was due to O₃. Both short- and long-term exposures to O₃ at concentrations below the current regulatory standards are associated with increased mortality from respiratory and cardiovascular diseases [76].

The DCS can measure both events induced by high temperature (eg, the formation of ground-level O₃ and low-temperature triggers [77]) and brown adipose tissue metabolism [78]. The direct impact of temperature reflected in a user’s heart rate data stream during their sleep was characterized. An increasing trend in ambient temperature even in the 68-80°F (20-27°C) range and most definitely beyond (>27°C; Figure 16) could be unhealthy. This observation is supported by a previous study highlighted in Figure 23, which shows data collected for 1 user over 424 nights using an iPhone with the Sleep Cycle app and a Garmin Fenix 5 Watch [79]. Clear increases in the user’s circulation flow (ie, reflected by resting heart rate) were observed and appear to be correlated with increases in ambient temperature.

Lower humidity levels can cause skin dryness, corneal dryness, dry nasal passages, and sinusitis [80], while higher relative humidity levels can promote the growth of mold, bacteria, and viruses. According to previous studies, low humidity levels lead to increased aqueous tear evaporation [81,82]. The low humidity values in the flight shown in Figure 15 corroborated with the researcher’s discomfort due to dry eye symptoms during the flight.

Previous studies have shown that several air pollutants, including PM_2.5, PM_10, tVOC, CO, CO₂, and more, can be present at much higher quantities in indoor air than in the outdoors [83]. There has been an increasing concern among the scientific community about the connection between IAQ and the impact of personal health [84].

The study results are significant in several ways. Data access in AMS is both simple and comprehensive. Real-time AMS data can be obtained and analysis performed from any device through a progressive web application. The website also includes historic data logs and recommendations on how to improve the user’s current air quality. Because it uses a cloud-based API, it opens the possibility to integrate new types of analytical graphs or recommendations into the system without the users needing to update their device. With all the air quality measurements collected and posted in tandem, there is an increased knowledge about the interconnections and impacts of pollutants on an individual, community, or specific demographic. Other researchers can also tap into the system to download data and include their own analyses for their unique studies.

Because the DCS unit’s calibration is not performed on a large scale, the calibration training set is limited when developing data correction algorithms. Another limitation is that the data were collected only by the researcher, and not external participants, which indicates that the data set is small and does not provide insight into various indoor pollutants affecting various living situations. This limits the development of training sets for learning algorithms to generate optimal individual weights for each of the pollutants to calculate AQI and to generate customized optimal recommendations for managing the pollutants to maintain healthy indoor air quality. While the DCS unit can be USB powered and portable when providing 13 data streams, due to increased power consumption for the comprehensive 22 data streams model, it requires to be plugged into a wall socket, thereby limiting its portability.

Future research could implement AMS on a large scale to safeguard public and personal health. Users can anonymously share their data to help researchers draw connections between indoor air pollution and public health and to further research in the development of new modeling techniques for public health, low-cost sensor data correction algorithms, and sensor modeling. The larger vision of AMS is to build a public database for indoor air, like EPA’s Air Quality System database [11,12] for ambient air. The REST APIs in the system can be used to access the open database populated by users willing to share their data.

AMS could include dynamic recalibration of the devices installed at various settings over the internet. With this auto-calibration, sensors would maintain the highest accuracy possible and users would not have to perform manual calibration themselves.

AMS can be researched for reduction of power consumption, while still providing the comprehensive set of all 22 data streams, thus maximizing its portability.

AMS can be used within households to compare indoor air quality levels between neighbors, or on a grander scale within communities. Large-scale distribution of AMS can lead to an expansion of preventative health approaches from an individual level to a community level and can provide valuable insights into public health.

Other possible enhancements include making the high-power model portable, and adding physical indicators, such as sound and light on AMS hardware, which can help indicate the hazardous levels of pollutants in a more noticeable way.

Cybernetic and navigational health approaches enable individuals to be in control of their health throughout their lives so that they have the necessary information to always maintain an ideal state of health [85,86]. Continued research in this field will focus on expanding into other health domains, improving quality metrics, and developing methods to combine atmosome data with other data streams to provide uniquely tailored lifestyle recommendations.

Last, AMS could implement more sophisticated software algorithms, such as an artificial neural network, as demonstrated in [27], and machine learning, as demonstrated in [30], and adaptive thresholding. In particular, the simplest algorithm to implement would be to account for environmental variation (eg, temperature and humidity) using an adaptive threshold to improve sensor accuracy. Specifically, and referring to [54], an adaptive threshold or the computed gas concentration could be computed as a function of temperature and humidity. This would significantly increase the sensitivity of the MQ sensors. Other algorithms likely exist and could be explored with the intention of maintaining a low-cost and small sensor solution as opposed to transitioning to an alternative sensor.

In this work, the concept of an atmospheric exposome (atmosome) was presented and a low-cost approach to leverage multimodal sensors and cloud data storage in building a personal atmosome was proposed. This work shows that AMS offers both the concept and system to quantify and measure the atmosphere with continuous real-time data streams, which provide users and researchers access to pertinent air quality data through a cloud computing architecture. These data can be used by both members of the public and researchers. For example, AMS could be used to alert an elderly user with dementia about a gas leakage.

Further, AMS can find the effect of other atmospheric streams such as temperature and humidity on the user’s observed health and behavioral outcomes. This system monitors and analyses VOCs, which could help ensure that pregnant women breathe safe air, safeguarding not only their own health, but also that of their fetus(es). Particulate matter detection and warnings can help users to act on remediating the environment to avoid pulmonary complications in all age groups. Overall, the findings indicate that the multidimensional model of AMS is a step closer in considering a variety of pollutants and atmospheric characteristics that guide users toward a healthy lifestyle.

While the system can predict respiratory triggers, such as asthma attacks in vulnerable people, it is yet to be tethered to other health conditions (eg, proneness of a patient with diabetes to blood glucose variations due to air quality). AMS could certainly provide insight into such correlations and dramatically improve personal health.