2.1. Experimental Setup

The T, RH, pressure (P), and concentrations of CO₂, O_x (=O₃+NO₂), NO_2, and NO were monitored using two AQB systems in a classroom (25°0 ^′53 ^″ N, 121°32 ^′20 ^″ E) at the National Taiwan University from September to October 2020 and June 2022, corresponding to the academic semester and summer vacation, respectively. The classroom has a volume of 289 m³ and is open to the outdoors through regularly open windows and gate doors during daytime on working days (typically from 8:00 to 17:00 every Monday to Friday). The walls are painted, and the floor and ceiling are covered with wood. The surface area of the walls is approximately 113 and ~214 m² for the floor and ceiling, while the surface area of the chairs (covered with fiber) and desks (made with wood) is about 154 m². The room has an energy recovery ventilator (including PM filtration) with five inputs and five outputs distributed on the top of the room to exchange indoor and outdoor air. Additionally, there are two doors on one side (Figure 1), which exchange air between the room and the hallway. There are windows in the classroom, and all windows were kept closed during the experiment periods. The exchange flow of the ventilator is 0.097 m³ s⁻¹ (corresponding to a rate constant of 1.20 h⁻¹) and is programmed to operate from 6:00 to 23:00 every Monday to Friday. The doors are closed at nighttime, and their open/closed time and air exchange efficiency can be determined based on the observed CO₂ profile simulated using a box model. The number of occupants in the classroom varies between 10 and 50 people, depending on the courses. This estimation is based on official reports of students enrolled in classes and cross-checked with attendance records maintained by teaching assistants. The surface area of each occupant is assumed to be 1.7 m² per person [46]. The room air conditioner is controlled manually and has internal circulation only. A visual representation of the classroom and an overview of the environmental setup are provided in Figure S1.

Three home-built AQB systems were applied to monitor the selected gas concentrations, including CO₂, NO, NO₂, and O_x. One AQB was placed on a desk at the front of the classroom and another at the back to assess indoor air quality. As shown in Figure S2, gas concentration measurements varied minimally between these two locations. Given the generally uniform seating arrangement of occupants and the consistent distribution of gases, indoor gas concentrations were assumed to be well mixed. Due to this uniformity and the stability of the instrument performance, data from the AQB at the front of the classroom was selected for subsequent analysis. The third AQB was placed in the corridor near the front door to measure the building’s concentration. The AQB comprises a Seeeduino (SHT31) for T and RH, a nondispersion infrared (NDIR) sensor (T6713-5K, Amphenol Advanced Sensors) for CO₂, and electrochemical sensors (Alphasense Inc., United Kingdom) for other gas species. It has dimensions of 25 × 16 × 8 cm and weighs 730 g. The sampling flow rate is controlled at ~5.6 L min⁻¹ by a fan, corresponding to a residence time of 34 s in the box. The CO₂ sensor has an accuracy of ±30 ppmv or ±3% of the reading value, while other gases are detected at parts per billion by volume levels with a time resolution of seconds. Ozone concentration is calculated from the difference between O_x and NO₂ concentrations. Outlier data points exceeding two standard deviations from the mean are excluded. Minute-averaged data are applied for further analysis.

AQB has the advantage of being lower cost, portable, and operating quietly, making it comparable to the measurements obtained at Environmental Protection Administration (EPA) stations. Typical measurements operated by the EPA have issues with the by-product generation in the exhaust air and the noise produced by pumps. Taking the example of EPA’s NO_x measurement (ML9841, Ecotech), it relies on a NO₂-to-NO converter for NO_x detection and needs a connected pump with a charcoal scrubber to remove excess ozone and discharge the exhaust gas outdoors. In contrast, AQB, equipped with electric gas sensors, does not generate by-products or pumping noise, making it more suitable for indoor environment monitoring. For the calibration issue, the calibration coefficients of the AQB were established by colocation with the Songshan EPA station, serving as the reference. The instruments at EPA stations exhibit higher precision and undergo calibration with standard gases every midnight. Moreover, the Songshan EPA station, the nearest major station with CO₂ measurement, provides environmental conditions closely aligned with the experimental conditions. A more detailed description of AQB data calibration is provided in the supporting information (Section S1 and Tables S1 and S2). Overall, the correlation coefficients for monitored environmental data between EPA instruments and AQB range from 0.47 to 0.81 for the studied chemical species. Low-cost electrochemical sensors might have a cross-sensitivity issue, while there were no significant interferences of the target pollutant measurement in this study [47–49].

As the electrochemical sensors are sensitive to T variations, sudden changes in T can cause an unrealistic signal variation for NO_x. To address this issue, a multiple linear regression that takes T into account was employed to calibrate NO_x concentration measurements. However, the issue persisted due to rapid T changes leading to baseline shifts. During class sessions, the initial operation of the air conditioner can cause a T drop of ~2°C and a 10% decrease in RH, which might lead to an additional baseline shift. Because of varying ventilation conditions within a whole day, resulting in distinct steady-state ozone levels, two different types of baseline correction were employed. During the periods with ventilation turned off and no occupants, the ozone concentration should be consumed entirely by the deposition onto room surfaces, resulting in approximately zero ozone concentration. Therefore, the baseline is corrected by offsetting the ozone concentration value by a range of −1 to +8 ppbv before the ventilator is turned on. During the periods with ventilation turned on, ozone had a nonzero concentration at a steady state due to the ventilation effect transporting ozone from outdoors. Hence, the concentration of ozone in each period requires baseline correction by 1.5–4.0 ppbv based on the steady-state concentration calculated from the model. Concerning the low concentration of ozone and NO_x in the indoor environment, the observed patterns of ozone and NO_x are verified using a personal ozone monitor (POM, 2B Technologies) and a 405 nm NO₂/NO/NO_x monitor (NO_x-405 nm, 2B Technologies) collocated with the AQB. The colocation comparisons took place on August 3 and 4, 2022, and March 8 and 9, 2023, under the same ventilation conditions as the experiments conducted from September to October 2020 and June 2022. Moreover, the presence of indoor VOCs was confirmed through VOC sampling conducted on June 27 and July 1, 2022. The detailed VOC sampling is described in the supporting information.

As there are limited CO₂ monitoring sites in Taiwan, the outdoor CO₂ concentration was based on the results of the Songshan EPA station (25°3 ^′0 ^″ N, 121°34 ^′43 ^″ E), located 5.5 km away from the classroom, while other gas concentrations were based on the Guting EPA station (25°1 ^′14 ^″ N, 121°31 ^′46 ^″ E), 1.2 km away from the classroom. The low reactivity of CO₂ leads to no significant difference in inlet concentration from the ventilation or doors. Additionally, the accumulation of CO₂ in the corridor is considered negligible due to the high air exchange rate with the outdoors. Further details are provided in the supporting information. However, ozone might be consumed due to reactions occurring inside the air parcel or by reacting with the surfaces of ventilator ducts and the interior wall of the building as outdoor air flows through ducts and enters the building. The resulting changes in concentration can be expressed as residual fractions α and β. α was derived using ozone modeling in a condition with the ventilator on but without occupants, while β was evaluated based on the linear relation between observed ozone concentration inside the building and those measured at the Guting EPA station.

2.2. Model Simulation

3.1. Diurnal Patterns of Indoor Air Parameters

Figures 2 and 3 show the temporal profiles of monitored parameters for indoor air without and with occupants, respectively, as measured by the AQB on the desk in front of the classroom. Similar diurnal trends were observed across course sessions for a given occupancy condition (with or without). For simplification, 4 days of observation data are used here to provide a comprehensive overview of the trends throughout the entire period. When the classroom was unoccupied (Figure 2), the T remained within a 1°C difference on the same day. The overall CO₂ concentration remained at ~420 ppmv during this period. When the air ventilator was on (06:00–23:00 on weekdays), indoor CO₂ had a similar variation trend as the outdoor CO₂. In contrast, when the air ventilation was off (23:00–06:00 the next day and weekends), indoor CO₂ remained relatively constant, even though outdoor CO₂ changed. This observation indicated no apparent source and sink of CO₂ inside the classroom except for the ventilation effect. Moreover, during no ventilation periods, the concentrations of ozone and NO₂ decreased, but NO increased. The observed decrease of ozone and NO₂ might be due to reactions with the room surface, while the increase of NO might result from the chemical reactions of NO₂ with the room surface under negligible ventilation conditions. This varying pattern shows that ~24% of NO₂ was converted to NO as the ventilation was turned off for 7 h. Spicer et al. reported that NO₂ could be removed in the indoor environment, and ~33% of NO₂ was converted to NO on the surface of wallboard paper after 66 h at inlet NO₂ concentrations of 1.2–1.5 ppmv [53]. A similar increasing trend was also observed during the weekend (Saturday in Figure 2), as there were no occupants, and the air ventilator remained off.

As the T variation was not significant in the isolated classroom (i.e., with ventilation off and no occupants present), the retrieved composition trend would reflect the environmental conditions. However, the absolute values might not be accurate due to the lack of daily baseline calibration. The temporal patterns of ozone, NO, and NO₂ monitored by colocating AQB with calibrated POM or NO_x-405 nm verify similar trends and amplitudes with ozone decreasing, NO₂ decreasing, and NO increasing, as shown in Figure S3. Both AQB and POM exhibit consistent ozone profiles, showing a reduction of ~5 ppbv after the room was isolated for 1 h. During periods with no production or transport but only consumption processes affecting ozone, the baseline of ozone can be determined as the concentration reaches a minimum and remains unchanged. The profiles with baseline correction for both AQB and POM are consistent, with an RMSE of 1.62 ppbv. Based on this baseline correction approach, further analysis of ozone-related simulation is conducted. Regarding the NO and NO₂ profiles, the consistently aligned trends between NO_x-405 nm and AQB support the observed NO and NO₂ trends using AQB, with RMSE at 1.92 and 3.65 ppbv, respectively. Additionally, NO_x-405 nm indicated that ~32% NO₂ was converted to NO after 7 h of no ventilation. The increased NO concentration during dark periods (when ozone was depleted) might be attributed to the biological denitrification processes [54]. Alternatively, it might arise from the heterogeneous hydrolysis of NO₂, leading to the generation of HONO (or NO) on surfaces capable of adsorbing water and forming a surface film [55–57]. The observed NO and NO₂ variations provide insights into the possible interconversion of NO_x indoors.

When the ventilator was turned on with no occupants present, the indoor ozone and NO₂ concentrations increased due to the inflow of outdoor air having higher concentrations. Conversely, the indoor concentration of NO decreased because the outdoor concentration of NO was relatively lower. The variation pattern of ozone was mainly influenced by the outdoor ozone photolysis reaction, with the highest levels observed at midday and the lowest levels at dawn and dusk. Meanwhile, the profiles of NO₂ and NO were primarily dominated by pollution episodes emitted from vehicles near the campus.

During weekdays with class sessions, as shown in Figure 3, significant variations in CO₂ concentration were observed when there were occupants. The increase in CO₂ concentration was attributed to occupants’ exhalation, while the decrease was due to reduced occupancy and higher air exchange rates. The concentration of ozone and NO₂ decreased more rapidly than periods with no ventilation, likely due to the additional reactions of ozone and NO₂ with human surfaces and the inhalation loss. Furthermore, there was a more accelerated increase in NO concentration compared to periods with ventilation off, suggesting that the observed NO might be from human exhalation [53], which is detectable in AQB. The trends of ozone, NO, and NO₂ were significantly influenced by surface reactions during transport and interactions with objects such as walls, furniture, and people, even though the ventilation system introduced outdoor air with higher ozone and NO₂ concentrations but lower NO levels than indoors.

3.2. CO₂ Simulation

The indoor CO₂ concentration can be simulated using Equation (1), along with records of class session duration and the estimated student attendance. This simulation can then be compared with observed data to extract additional ventilation details, including door openings and fluctuations in student numbers. Figure 4 shows simulation results for September 22, including two courses scheduled from 8:10 to 10:00 and from 10:20 to 12:10. During these sessions, a minimum air exchange rate of 1.20 h⁻¹ was maintained (the ventilator was on), with a higher air exchange rate observed when doors were opened. In the first course, with approximately 47 occupants, an air exchange rate of 1.20 h⁻¹ was maintained, leading to CO₂ levels nearing 3000 ppmv by the end of the class. Conversely, in the second course, with ~42 occupants and a short recess during 11:05–11:30, a lower CO₂ level of ~2200 ppmv was observed at the end of the class due to fewer people and increased ventilation (one door opened). Recesses with fewer occupants and higher ventilation decrease CO₂ concentration compared to class sessions. The high coefficient of determination (R²) of 0.996 indicated the simulation could identify possible variations in the number of occupants, such as students entering the classroom before the start of the first course and remaining in the room between the first and second courses, as shown in Figure 4. The ventilation efficiency of each door was estimated to be approximately 1.05 h⁻¹, indicating that the ventilator efficiency is higher than with one door opened but lower than with two doors opened. Both turning on the ventilator and keeping two doors open in this classroom can help mitigate the CO₂ impact on human health and ideally keep CO₂ levels below 1000 ppmv, the standard of the Taiwan Indoor Air Quality Management Act implemented by the Taiwan EPA. Furthermore, by quantitatively estimating physical processes, including ventilation efficiency and occupancy variation, the chemical processes of ozone can be further elucidated.

3.3. O₃ Profiles and VOC Formation

Figure 5 shows the extracted normalized data for each weekday nighttime from 23:00 to 06:00 the next day relative to the concentration at 23:00 when the ventilation was off. The persistent decay trend indicates the uptake of ozone by room surfaces (including furniture and walls) in nonoccupancy and unventilated conditions. With a first-order reaction assumption, the simulation results are highly similar to the observations (average R² = 0.989), and the deposition velocity of ozone removed by room surfaces was determined at 0.019 ± 0.005 cm s⁻¹, consistent with previous studies ranging from 0.019 to 0.036 cm s⁻¹ (Table 1) [38–40]. With the derived room uptake rate of ozone on room surfaces, the simulation results in the absence of occupants are shown in Figure 6 (average R² = 0.963). The deposition velocity of ozone removed by human surfaces was estimated as 0.45 ± 0.15 cm s⁻¹, similar to literature ranges (0.20–0.50 cm s⁻¹) and summarized in Table 1 [27, 38, 41]. Human surfaces exhibit a significantly higher deposition rate for ozone (~24 times), suggesting an equivalence when the human surface area is about a quarter of the room surface area (~12 people in this classroom). The residence time of ozone in the classroom is 0.88 h without occupants and reduced to 0.26 h with 40 students present, mainly consumed by human surfaces. The inhalation consumption of ozone has a residence time much longer, ~9.1 h, and is not significant compared to other surface reactions. Hence, in the typical condition of this classroom with 40 occupants, room surfaces account for 22.2% of ozone removal, while human surfaces account for 75.6% and human inhalation accounts for 2.2%. These contributions are influenced by surface materials and the number of occupants. For instance, Xiong et al. reported the contribution to be 42%, 56%, and 2%, corresponding to room surfaces, human surfaces, and human inhalation for a classroom covered with latex paint and a hard tile floor with 57 occupants (air exchange rate ~5 h⁻¹) [38]. Furthermore, although the overall ozone removal rates are similar between the two studies, the significantly lower air exchange rate in this study resulted in a notable decrease in indoor ozone concentration. The reached steady-state concentration varied with outdoor concentrations, as shown in Figure 6. This finding suggests that outdoor ozone primarily affects indoor ozone levels, though the ventilation rate and occupant density may also play a role in different indoor environments. The observed negligible contribution of human inhalation removal is consistent with the previous analysis (2%–4%) [38, 41], supporting the model simulation’s exclusion of this removal effect. The impact of internal air circulation was not quantified in this study, potentially leading to an underestimation of room surface effects and an overestimation of human surface effects on ozone removal.

Ozone reacts with squalene in human skin lipids, producing organic compounds such as Ga, 6-MHO, and 4-OPA [27, 31, 52, 58]. The simulation results of three major products (Ga, 6-MHO, and 4-OPA) under conditions of Figure 6 are shown in Figure S4, with concentrations in a range of 0.7–1.2, 0.2–0.4, and 0.4–0.7 ppbv, respectively, varying with ozone concentrations and the number of occupants. These simulated concentrations likely represent the upper limits of produced VOCs due to the simplified simulation without specific sinks, such as adsorption of VOCs on room surfaces, gas-phase reactions with ozone or other oxidants, and the possibility of ozone reacting with other gas species. These simulated concentrations are consistent with the values observed in previous studies, such as those conducted in a university classroom by Xiong et al. [38] and an occupied residence by Liu et al. [37]. Additional sampling using sorbent tubes with gas chromatography analysis had a mean concentration of 0.037–0.144 ppbv for 6-MHO and 0.031–0.050 ppbv for 4-OPA (Table S2), conducted under three ventilation and occupancy conditions: (1) no ventilation and no occupants, (2) with ventilation but no occupants, and (3) with ventilation and 10 occupants. In Case 1, lower levels of 6-MHO and 4-OPA were observed, indicative of the background or residual concentrations. Case 2 has higher 6-MHO and 4-OPA than Case 1, suggesting the ozone-induced production of VOCs. Case 3 shows similar or lower levels than Case 2. This could be attributed to either the relatively low levels of outdoor ozone and the applied sampling time insufficient to observe VOCs accumulating with only 10 occupants or the nonhomogeneity of indoor VOCs resulting in variance during collection. Even though the sampling lacked temporal profiles, it verifies the presence of these VOCs indoors resulting from the coreactions of ozone and organic compounds.

Figure 7 shows the sensitivity test results for CO₂, ozone, and VOCs under varied ventilation rates and ozone removal efficiencies of the ventilator (i.e., α parameter) with a fixed occupancy of 40 people after 2 h. Increasing the air exchange rate can effectively reduce indoor CO₂ concentration and alleviate human burnout. However, elevated ventilation rates often lead to increased indoor ozone concentrations and VOC production, yet these produced VOCs can be efficiently ventilated outdoors. Comparing VOC levels to the most unhealthy condition (no ventilation and α = 0.37), high ventilation removes 78%–86% of VOCs, while high ozone removal efficiency (i.e., high α) can reduce 86%–88% of VOCs, indicating effective VOC mitigation (Table S3). From an energy-efficiency perspective, simply opening doors can significantly reduce VOC levels without extra energy consumption. Conversely, integrating an ozone scrubber yields lower VOC production but requires regular equipment maintenance. Various room parameters, including room size, ventilation, occupants, and outdoor conditions, influence the tailoring of ventilation strategies.