Indoor air quality at a French university: a participatory CO2 measurement campaign highlights the wide gap between reality and the law

2.3.1 CO₂ measurements using the “participatory campaign”

Four Aranet4 sensors were made available to participants. These sensors complied with the recommendations of the French Environment Code [18] for CO₂ measurements (measurement by non-dispersive infrared absorption spectrometry, measurement range 0 to 5,000 ppm; measurement uncertainty ± [30 ppm + 3% of reading]). They measure CO₂ concentrations in ppm. They had been calibrated in accordance with the manufacturer’s recommended procedure. This procedure consists of installing the sensor in a place where the CO₂ concentration is close to 420 ppm. In order to be as close as possible to this value, the calibration was performed outdoors in an environment with adequate wind exposure, sparsely built-up, far from industrial activities and car traffic, with the sensor positioned at a distance from plants. The procedure is initiated from the application installed on a mobile phone and requires no intervention other than starting the calibration and waiting for it to complete (which takes about 15 minutes). Regular calibration checks were carried out, verifying that at certain times of the day they measured outdoor concentrations between 420 and 470 ppm. No recalibration was necessary. These lightweight, portable sensors record one measurement per minute, and digitally transfer the data in .csv format to a smartphone via an app that can be downloaded from any app store. Spot measurements were taken outside teaching buildings by some participants.

2.3.2 Access to data collected by fixed UGA sensors

CO₂ measurements from the “participatory campaign” were enriched by data captured by 117 Carbon Nexelec sensors deployed by the UGA. These sensors take a reading every 10 minutes and transmit their data over a private wireless long-range wide area network. The UGA makes all these measurements available to its staff, via its intranet, on a Grafana® platform.

2.5.1 Processing data measured during room occupancy

For each measurement session, an Excel interpretation file was created, containing both the measurement conditions and the measurements themselves. A graph of the CO₂ curve versus time was generated, also showing the HCSP target values (800 and 1,500 ppm), the legal value (1,300 ppm), and the thresholds of 2,500 and 4,000 ppm. The following parameters were systematically calculated for each session, when the necessary information was available: gauge J, median N of the number of occupants, occupancy rate R_Occ = N/J, volume per occupant when R_Occ = 100% V_J = V/J, median of CO₂, Q₃ value of the third quartile of CO₂, portion of the duration of the session spent with CO₂ > 1,500 ppm and with CO₂ > 2,500 ppm, clean air flow per occupant D_N when N occupants are present in the room (so D_J is the clean air flow per occupant when R_Occ = 100%), total clean air flow D in the room.

In order to estimate D_N, the classical model of CO₂ concentration in a well-mixed space was utilized [19]. This model states that, in a room with N occupants, where c(t) is the concentration of CO₂ at time t, D_exp is the average CO₂ generation rate by human beings, c_ext is the concentration of CO₂ outside the building, and is the air renewal rate. As a consequence, when c(t) reaches a steady state c_limit hence This model requires an estimation of D_exp, the average CO₂ generation rate by human beings. This rate has been determined to be 20.2 L/h [20]. We propose to estimate c_limit by Q₃. The reasons for these choices of the estimations of D_exp and of Q₃ are discussed in section 4.1.2. As a consequence, if c_ext is expressed in ppm, D_N (m³/h) was estimated by As a consequence,

For university rooms without mechanical ventilation, French law imposes V_J ≥ 15 m³/occupant. Accordingly, the legal gauge can be defined as J_Legal = min [J ; J V_J / 15]. Although there is no legal maximal value of CO₂ in university rooms in France, the HCSP Gauge J_HCSP can be defined as the maximal number of occupants not to overpass the 1,500 ppm threshold recommended by HCSP. The previously mentioned model of CO₂ concentration in a well-mixed space implies that, if the CO₂ limit concentration is 1,500 ppm, the minimum flow of clean air per occupant D_HCSP satisfies D_HCSP = 20,200/(1,500-c_ext), where c_ext is the outdoor concentration. From this it can be deduced that J_HCSP = min [J ; J D_J / D_HCSP].

Since the true value of the outdoor CO₂ concentration was rarely available, we opted to overestimate D by using the maximum value of CO₂ measured outdoors (500 ppm), and to underestimate D_HCSP by using the minimum possible value (417 ppm, world average CO₂ concentration [21]). These choices all maximize the estimate of the J_HCSP gauge. The reasons for this choice will be discussed in section 4.1.2.

For rooms with mechanical ventilation, French law imposes D_J ≥ 25 m³/occupant in universities. The legal gauge can then be defined as J_Legal = min [J ; J D_J / 25].

2.5.2 Processing data measured after room occupancy, with openings closed

The participant had to leave the CO₂ sensor in the room for at least 1 hour, after the room had been used and CO₂ had risen above the concentration outside the building. Equation (1) can be written as c(t) = ae^−kt + c_ext. When c_ext had been measured, this value was used. Otherwise, c_ext was assumed to be 500 ppm (this leads to a limited overestimation of k, that will be discussed in section 4.1.2). To estimate parameters a and k, Excel’s SLOPE and INTERCEPT functions were used, on data {t_i; ln(c(_t) − c_ext)}. Quality of fit was estimated by the correlation coefficient between measurements and model predictions, obtained by Excel’s CORREL function. Time being expressed in minutes, when the volume V of the room and its gauge J were known, the clean air flow per occupant at 100% occupancy was obtained by the formula D_J=60 kV/J. The legal gauge J_Legal is here also estimated by J_Legal = min [J ; J D_J /25].

3.4.1 Study of data collected during room use

CO₂ concentrations were studied during 57 classroom sessions in 41 mechanically ventilated rooms in 15 UGA buildings. The main parameters describing characteristics and room occupancy, CO₂ distribution and mechanical ventilation performance in these rooms are synthesized in Table 3.

In 47 out of the 57 classroom sessions (82%), the clean air flow at 100% occupancy was estimated below the legal threshold of 25 m³/h/occupant at 100% occupancy. In 32 out of the 41 rooms (78%), at least one session led to an estimation of a clean air flow at 100% occupancy below 25 m³/h/occupant. Figure 3 shows a typical measurement in a 300-seat, 1,360 m³ amphitheatre, occupied to 32% of gauge.

• Download figure
• Open in new tab

Figure 3.

Typical CO₂ profile in room Ta, a room of 1,360 m³ occupied by 95 people for a gauge of 300 (32% occupancy). 3^rd quartile of CO2 = 1071 ppm. Clean air flow during session = 35 m³/h/occupant. Clean air flow at 100% occupancy = 11 m³/h/occupant. Percentage of gauge reduction required to meet legal airflow of 25 m³ / hour / occupant at 100% occupancy = 55% (78% to reach the 50 m³/h/occupant target of HCSP).

3.4.2 Study of CO₂ decay data after room use

The decay of the CO₂ concentration was studied subsequent to the cessation of room use and the closure of all apertures in nine rooms (including one, the Ne amphitheatre, before and after repair of the air handling unit). The results obtained are presented in Table 4. Figures 4 and 5 show the decay curves for amphitheatre Ne, before and after repair of the air handling unit.

• Download figure
• Open in new tab

Figure 4.

Model of CO₂ removal in room Ne, a room of 1,040 m³ with a gauge of 240 (room emptied after use, all doors and windows closed) during one night (800 minutes). r = 0.994. Renewal rate = 0.04 volume/h. Clean air flow per occupant at 100% occupancy = 0.2 m³/h/occupant (< 1% of the regulatory floor).

• Download figure
• Open in new tab

Figure 5.

Model of CO₂ removal in room Ne, a room of 1,040 m³with a gauge of 240 (room emptied after use, all doors and windows closed) during 45 minutes. r = 0.996. Renewal rate = 4.6 volume/h. Clean air flow per occupant at 100% occupancy = 20 m³/h/occupant (80% of the lower legal limit). Note: This measurement was made after the air handling unit was repaired.

4.1.1 Involvement of the lecturers

The involvement of the lecturers made it possible to collect precise data on the number of occupants and the condition of the openings (all the measurement files sent were accompanied by the required details on this point). The lecturers taking part in the study reported no difficulties in collecting this information, which is a major advantage over processing data from fixed sensors installed in the rooms, with no information on the number of occupants or the state of the openings. Using the mobile CO₂ sensors made available to the teachers proved very easy, with all of them having no difficulty installing the necessary application on their mobile phones. Transferring the measurements and information about the measurement conditions to the cloud made available to them posed no particular problem either. The study also raised participants’ awareness of the need for better ventilation in premises with insufficient natural ventilation, and of the reality of the problem of insufficient mechanical ventilation in many rooms.

4.1.2 Uncertainty of the estimation of the indicators of air quality from our measurements

As described in sections 2.5.1 and 2.5.2, the clean air flow is estimated on the basis of CO₂ measurements, using the classical model of CO₂ concentration in a well-mixed space [19]. CO₂ measurements are known with an uncertainty of ± (30 ppm + 3% of reading), which corresponds to ± 5% ppm for a concentration of 1,500 ppm (and between 3 and 5% for higher concentrations).

For estimations of air flow based on CO₂ measurements while the room is occupied, the CO₂ growth model operates under the assumption of homogeneous mixing of the air within the room. However, it is well known that CO₂ concentrations can vary from one point of the room to another, depending on natural convective air movements [22], especially when an air current is established after opening windows or doors, or on the proximity of room occupants [23]. These differences can reach 10% of measurements [24]. Therefore, the optimal scenario would involve the strategic placement of multiple CO2 sensors throughout the room, whereas only a single sensor was utilized per session of measurements. Given the utilization of a single sensor in each session, the total uncertainty in the CO2 measurements was considered to be ±15% (5% due to the sensor uncertainty, 10% due to the inhomogeneity).

In order to estimate the limit concentration of CO₂, that was reached at steady state, the third quartile Q₃ of CO₂ measurements obtained during a session of room use was utilized in Equation (4) – rather than the maximum – for two reasons. Firstly, this approach excluded potential outliers. Secondly, this deliberate overestimation of the clean air flow ensured that rooms considered incompatible with the law truly were. This approach was also adopted to determine the outdoor concentration of CO₂, which was set at 500 ppm, representing the highest concentration measured. The utilization of Q₃ as opposed to the maximum of CO₂ offers a distinct advantage, namely the fact that the associated uncertainty is considerably less pronounced in comparison to the uncertainty inherent in discrete CO₂ measurements. Nonetheless, the estimation of the uncertainty on Q₃ will be maintained at 15%. Given that Equation (4) utilized to estimate the clean air flow we use model is D_N = 20,200/(Q₃-500), an uncertainty of ± 15% in Q₃ yields an uncertainty of D_N inferior to 22.5% when Q₃ ≥ 1,500 ppm.

Another source of uncertainty comes from estimating the average CO₂ generation rate by the human beings present in the room. The value of 5.6 mL/s (or 20.2 L/h) was utilized, as proposed by the Air Infiltration and Ventilation Centre [20]. This value is compatible with the inhalation rates used by [25] for different activity levels (0.49 m³/h for resting, 0.54 m³/h for standing), which, taking into account an average CO₂ concentration of 4% [26] correspond to 19.6 L/h and 21.6 L/h, respectively. However, other publications suggest lower CO₂ generation rates. For example, [27] suggests an average value of 16.2 L/h for a classroom of 64 students and one teacher, and [28] considers that an activity of “sitting reading, writing, typing” corresponds to a metabolism of 1.3 times the basal metabolism, corresponding to an average value of 18.0 L/h. Once more, we elected to overestimate the clean air flow, which consequently guided our decision to utilize the estimate proposed by [20].

The variability of estimates of average human CO₂ production rates underscores the importance of preferring the CO₂ decay method. In addition, during decay measurements, the risk of mixture inhomogeneity is reduced because all openings are closed and the room is empty, so the uncertainty is also reduced with this method. Finally, the fact that the model is fitted to all measurements also reduces the uncertainty. This explains why the confidence interval of the decay model is small, as can be seen in Figure 8. For all these reasons, this method should be used whenever possible.

4.1.3 Reliability of the estimation of the indicators used to characterize violations of legal requirements for air quality

Analysis of the estimated clean air flows shows that an underestimation of 22.5% of the clean air flow would lead to classifying 24 instead of 32 (59% instead of 78%) of the 41 rooms with mechanical ventilation as uncompliant with the legal minimum of 25 m³/h/occupant. Such an error would have no practical consequences. In fact, universities do not have the resources to immediately address all air handling unit deficiencies in more than half of their mechanically ventilated rooms, so an error in the classification of 20% of the rooms will not have major consequences: university administrators will have to work first on improving their worst rooms in terms of air quality.

4.1.4 Comparison of flow estimates during room occupancy and CO₂ decay

In rooms with mechanical ventilation, where the clean air flow per occupant could be estimated both by estimating the limit concentration of CO₂ at steady state and by modelling the decay of CO₂ after evacuation of the room and closure of the openings, the estimate obtained by the first method was systematically higher than that obtained by the second. The first reason for this difference is that steady state is often not reached before the end of the session. A second reason is the deliberate choice of the 3^rd quartile of CO₂ concentrations as the limit value, which was justified in the previous section. Finally, in certain circumstances, opening of doors and windows provides more clean air than mechanical ventilation. It is therefore not surprising that the estimate obtained after evacuating the room and closing the openings is systematically lower than that obtained in real conditions, where the openings can be opened. This is illustrated by the example of room Ne, where the clean air flow was estimated during 6 classroom sessions. The 6 estimates of clean air flow (in m³/h/occupant at 100% occupancy) for this room are 3.5, 4.0, 4.3, 4.6, 5.4, 8.1. The relative homogeneity of the first 5 estimates should be noted, with the estimate of 8.1 m³/h/occupant appearing to be an outlier. These measurements were obtained when the mechanical ventilation in this 240-seat auditorium was not operational. This auditorium has the unusual feature of having 3 doors, which were open (two main doors at the bottom of the amphitheatre, opening onto a patio where a large number of students may walk, and where the CO₂ concentration can therefore vary significantly, and a third at the very top, opening onto a corridor). A slight draught was felt by the lecturer at the bottom of the amphitheatre. The estimated clean air flow per occupant for this same room using the decay method (Figure 4) is 0.2 m³/h/per occupant at 100% occupancy. It can be seen that opening the doors provides a significant flow of clean air, although it is far from sufficient.

Estimating the flow of clean air at 100% occupancy using the CO₂ decay method after evacuating the room and closing the doors proved particularly simple to implement. The first tests of this method were carried out on the Ne room, when its mechanical ventilation was down. As a result, the sensor had to be left in place overnight to observe a sufficient drop in CO₂, which posed practical constraints when it came to retrieving it the following morning. But when the mechanical ventilation is working more or less correctly, measurements over an hour or so after the doors have been closed are sufficient. However, it is difficult to ask a lecturer to stay for an hour after the end of the lesson. Extending this method to all classrooms would therefore require institutional mobilisation, with the involvement of the building maintenance staff.

For rooms with mechanical ventilation, this method of modelling CO₂ decay seems the simplest to implement, for rapidly characterising all the rooms with mechanical ventilation in a university. By simply taking this measurement after the end of the last lesson of each day, it is possible with one sensor to characterise 20 rooms in one month, i.e. around 20 working days. For an investment of around €2,000, or 10 sensors, 200 rooms could be characterised each month, at the cost of a moderate human investment (a maximum of ten minutes to transmit and process each series of measurements characterising a room). However, this requires institutional mobilisation and political will.

4.2.1 Risk situations in rooms without mechanical ventilation

Table 1 shows that none of the studied rooms meets the legal limit of at least 15 m³/occupant. Of the 14 rooms without mechanical ventilation in our study, where the volume was known, to meet this minimum:

- 2 rooms (14%) should reduce their gauge by less than 50%,

- 6 rooms (43%) should reduce their gauge by between 50 and 70%,

- 6 rooms (43%) should reduce their gauge by more than 70%.

As early as December 2023, the participatory campaign demonstrated that failure to comply with the minimum volume per occupant in rooms without mechanical ventilation led very quickly (typically within 15 minutes), when the openings (doors and windows) were closed, to the threshold of 1,300 ppm being exceeded, with regular measurements of over 4,000 ppm, and even 5,000 ppm, as Table SM.1, Table 2, and Figures 1, 2 and 6 amply demonstrate. Figure 7 illustrates the frequency with which 2,500 ppm concentrations are exceeded in a 154-seat amphitheatre (2.8 m³/occupant) without mechanical ventilation.

Of course, the method used in no way allows us to claim that the 15 rooms without mechanical ventilation studied by the participatory measurement campaign are representative of the situation in all the university’s rooms. They were not chosen at random; these were the rooms where the campaign volunteers taught. However, the observations of the participatory campaign are compatible with the analysis of the data from the fixed CO₂ sensors. It is very worrying to note that, as table 2 shows, in 52% of cases, the first quartile of CO₂ concentrations measured in these rooms exceeded the threshold of 1,300 ppm (and 97% of this first quartile exceeded the threshold of 800 ppm recommended by the HCSP). These rooms, which have no mechanical ventilation, are usually of relatively limited gauge (typically designed for 30 to 40 occupants, and are usually classrooms for tutorials or practical work). However, there is at least one case where a 154-seat lecture theatre has no mechanical ventilation. This is room Xa, where the volume per occupant is 2.8 m³/occupant, or 19% of the legal minimum threshold of 15 m³/occupant (see Figure SM.1 for photos and plans of this room). Under these conditions, it is not surprising that plateaus of over 5,000 ppm for more than an hour are regularly recorded there, even with an occupancy rate of less than 50%. Such levels have been observed in particular during examination periods. This is particularly problematic in view of publications showing a decline in psychomotor and cognitive abilities at concentrations above 1,000 ppm, and especially above 2,500 ppm [29] [10], as well as an increase in the rate of headaches [10] [12].

Our campaign therefore suggests that any room without mechanical ventilation that does not meet the threshold of 15 m³/occupant should be considered hazardous to the health of occupants, as long as the openings cannot be permanently opened when the room is in use.

4.2.2 Risk situations in rooms with mechanical ventilation

Here again, our methodology cannot guarantee that all 41 rooms in 15 of the university’s buildings are representative. However, it is noteworthy that if the legal minimum of clean air flow of 25 m³/h/occupant is considered the objective, a mere nine of the 41 rooms examined (22%) would not necessitate a reduction in gauge. Conversely, if the objective is to adhere to the most recent HCSP recommendation (a minimum clean air flow of 50 m³/h/occupant, corresponding to a maximum CO₂ concentration of approximately 800 ppm), it is observed that merely two of the 41 rooms meet this criterion (with doors of these two rooms open to the outside during the session). Consequently, in order to adhere to the recommendations outlined by the HCSP, it is imperative that at least 39 of the 41 rooms under consideration undergo a reduction in gauge, with 35 of these rooms required to undergo a reduction greater than 70%. It should also be remembered that the estimates of clean air flow during classroom sessions are deliberately “optimistic” for the reasons discussed in section 4.1.2, and that measuring the flow of clean air per occupant using the CO₂ decay method would certainly give even more drastic estimates of the reduction in gauge.

4.2.3 The university’s ability to respond to air quality alerts

The university has two levers at its disposal to reduce the risk to which its staff and students are exposed as a result of poor indoor air quality. The first of these levers is technical, and can be activated when the problem arises from a malfunction in the air handling units. This was the case for the Ne amphitheatre. When the relevant university departments intervened, it was found that the air handling unit in this amphitheatre had stopped working. Simply restarting it brought about a considerable improvement in the flow of clean air per occupant, which rose overnight from 0.2 m³/h/occupant to 20 m³/h/occupant (80% of the legal minimum of 25 m³/h/occupant, and 40% of the HCSP recommendation of 50 m³/h/occupant).

The second lever is political and organisational. The university needs to become aware of the seriousness of the public health problem posed by the quality of its indoor air. In reality, despite explicit advice from the country’s highest public health authority, the HCSP, the message is struggling to get through. Such awareness would provide the means to detect malfunctions in the air handling units and correct them where possible, as well as defining a policy for adapting the use of rooms, which could be used to improve indoor air quality.

The fixed sensors deployed by the university have the advantage of constantly monitoring the CO₂ concentration in the rooms equipped with them. This is very useful for detecting malfunctions in the air handling units. However, during the course of our study, no one was responsible for sounding the alarm if the thresholds were exceeded. Thus, it was within the framework of the study described here that analysis of the CO₂ signal recorded continuously in the Vc amphitheater (see Figure SM.2 for photos and plans of this room) made it possible to detect the shutdown of the air handling unit in this building, the consequences of which are summarized in Figure 8. This shutdown during an examination led to considerable exceedances of the legal thresholds, with probably a peak of CO₂ concentration around 10,000 ppm. Such levels are utterly unacceptable, particularly during an examination period when students are mobilizing all their cognitive capacities. It has previously been noted that these capacities are diminished above 1,000 ppm, and more specifically above 2,500 ppm. At 10,000 ppm, the risk of contamination by airborne infectious agents is considerably increased (21% of the air breathed in comes from the occupants’ lungs). But direct toxicity must also be considered. This is characterized by the Occupational Exposure Limit (OEL), which must not be exceeded for 8 hours (OEL 8h) or for 15 minutes (Short Term OEL). The 8h OEL for CO₂ is 5,000 ppm [30]. Short Term OEL values vary from country to country (10,000 ppm in Germany, 30,000 ppm in the USA, France has not defined one). The concentrations shown in Figure 8 are therefore particularly problematic.

It should also be possible to put in place a suitable organisation to prevent the recurring risks of CO₂ limits being exceeded. Once again, the example of the Vc amphitheatre is typical. In building V, the air handling unit stopped working due to a malfunction in the dialogue between this unit and the fire detection system. This dialogue regularly leads to the air handling unit shutting down, which can only be restarted by the intervention of a member of staff. Despite repeated warnings, it has not been possible to arrange for the CO₂ concentration in the two lecture theatres in Building V with fixed CO₂ sensors to be read at least daily. This is how the incident summarised in Figure 8 occurred, even though reading the signals from these sensors several days before the examination summarised in Figure 8 would have made it possible to detect the shutdown of the air handling unit and therefore to restart it before the examination that led to the 10,000 ppm peak. Several weeks before the incident summarised in Figure 8, the authorities had been asked to put in place a procedure to ensure that the CO₂ sensors in this building were checked twice a day and that action was taken immediately to restart the air handling unit in the event of an unexpected shutdown. The alert had even been given 48 hours before the examination summarised in Figure 8. No measures were taken to protect the occupants. Worse still, although a specific alert had been filed following this incident on the official register, a new similar incident occurred a few weeks later, with the 5,000 ppm limit value again being exceeded during an examination.

The university also needs to organize itself to react to a known systemic risk. Signals showing the reality and seriousness of the problem in a large proportion of the rooms at UGA had been available at the highest levels of the university since July 2022. The poor quality of the indoor air was discussed on numerous occasions by the university’s specialist risk management bodies, without appropriate corrective or preventive action being systematically taken in line with the HCSP recommendations for CO₂ levels in excess of 1,500 ppm. Figures similar to those in Figure 6 have been discussed several times at the highest level, without the university’s website mentioning the existence of the problem, and without appropriate warning measures being systematically taken (for example, during months, most of the rooms concerned did not display a poster warning of the existence of a danger, encouraging users to open the openings as much as possible, asking people entering the room to use a surgical mask at the slightest sign of respiratory infection, and suggesting that immuno-compromised people protect themselves). No decision was taken to reduce the gauge of the rooms, even in the face of major inadequacies. Amphitheatre Xa, for example, continued to be used to accommodate students in conditions that regularly led to considerable breaches of the legal thresholds (up to more than 5,000 ppm, see Figure 7), even though the failure to comply with the minimum threshold of 15 m³/occupant for this room without mechanical ventilation should have led to the 154-seat gauge being reduced to 29.

Finally, when a malfunction is reported in a particular room, it is important to organise a rapid response. The university has a specific register in which its staff can officially report any risk to their health or that of their colleagues and students. In our experience, this register did not allow for rapid reactions when exceedances of the legal CO₂ thresholds were reported. For example, an initial informal alert following a CO₂ measurement in lecture theatre Ne was issued in January 2022. As this alert did not lead to any action being taken, a second alert was filed on the ad hoc register in January 2023, which was also not acted upon. A third alert was filed in January 2024. This alert was initially closed by the administration, which considered that the problem was already known. The staff member concerned had to challenge this decision to close the alert before the university’s health and safety committee. A specialist was finally sent to check the operation of the air handling unit, found that it had been shut down, and restarted it. This highlights how difficult it is for a university to listen to the alarm signals it receives.