Proxy methods for detection of inhalation exposure in simulated office environments

Summary of descriptive statistics and correlations of IAQ measurements

In order to understand spatial IAQ variations in the chamber, we examined variations of studied air pollutant concentrations in relation to monitor placement. Figure 3 shows minimum, first quartile, median, third quartile, maximum and average CO2, PM2.5, and PM10 concentrations for each monitor placement (ID 1–4) averaged across all activities and experiments. Regardless of the air pollutant type, the breathing zone concentrations were substantially higher relative to stationary concentrations. The average of breathing zone CO2 concentrations of the reference participant were approximately two times higher than the ones from stationary monitors. This finding showed a notable increase in breathing zone CO2 concentration compared to a study by Melikov et al. [39], where CO2 concentration inhaled by a breathing thermal manikin was only 16% higher than in the room exhaust. The average PM2.5 and PM10 showed 6.7× and 6.8× higher concentrations at the breathing zone than the ones at stationary monitors, respectively.

Fig. 3figure 3

The CO2, PM2.5 and PM10 concentration at different stationary monitors across all activities and experiments.

The highest average CO2 and PM10 concentration among stationary IAQ monitors were recorded at the Front edge of participant desk which was the closest stationary monitor to the reference participant. This can be a result of exhaled CO2 jet that propagates downwards during sitting activities, as well as human thermal plume that transports locally generated airborne particles to the breathing zone [40]. This was not the case of PM2.5, where the highest average concentration among stationary monitors was detected at the Exhaust 1, likely because of vigorous activities (e.g., stuffing the cabinet with paper boxes) that occurred nearby. Further, we compared the absolute mean CO2 and PM10 concentration between the Exhaust 1 and Exhaust 2 (Fig. S6), where difference and variation of mean concentration were trivial in case of CO2, while it was significant in case of PM10.

Figure 4 shows the Pearson correlation r values between stationary indoor and breathing zone CO2 and PM concentrations during sitting, standing, and combined (sitting and standing) activities. Relative to combined activities, r values for CO2 were often higher when we segregated participant activity into sitting and standing activities. The correlation r between the CO2 in the breathing zone and at the Front edge of participant desk was 45% higher during sitting activities relative to combined activities. For standing activities, the relative increase was 36% and 32% at the Exhaust 1 and Desk locations compared to combined activities. CO2 measurements at the Exhaust 1 had a moderate correlation (r = 0.526) with the breathing zone measurements during standing activities. This finding agrees in part with a study by Pei et al. [24] who reported CO2 measured at the room exhaust well correlates with the inhalation exposure to CO2 under mixing ventilation. The two highest correlations between breathing and stationary CO2 measurement were at Exhaust 1 and Desk during standing activities. This is due to the contribution of spatial air pollution gradients and the proximity between the reference participant and the sensor locations during the standing activities. During the sitting activities, a relatively weak correlation (−0.3) between CO2 at the Exhaust 1 and in the Breathing zone may be attributed to spatial non-uniformity of air pollution concentration and greater distance between Exhaust 1 and seated reference participant. Lu et al. [41] also recognized that inconsistent patterns of CO2 concentrations in breathing zone of occupants may contribute to discrepancies of correlations between room exhaust and breathing zone CO2 level.

Fig. 4figure 4

Pearson correlations of CO2, PM2.5, and PM10 measurements during sitting, standing, and combined participant activities.

The correlation r between stationary and Breathing zone PM2.5 and PM10 measurement improved marginally during sitting activities (4–7%) and did not improve during standing activities compared to combined activities (Fig. 4). Sitting activities had better correlation for PM2.5 and PM10 than standing activities by threefold. Specifically, the correlation r between Exhaust 1 and Breathing zone during sitting condition showed over 0.9 for both PM2.5 and PM10. Low correlation between stationary and breathing zone PM levels during standing activities is attributed to irregular and high-intensity activities that resulted in highly episodic particle emissions. This result confirms that human inhalation exposure can be highly dependent on human activity and its intensity [17, 42]. Further, we compared correlation r between the two exhausts with the Breathing zone measurement (Table S3). In case of PM10, r value at Exhaust 2 decreased by 41–83% compared to the one at Exhaust 1 due to the distance between the reference participant and the deployed OPCs.

Multiple linear regression models for estimating human exposureMLR models based on stationary IAQ measurements

We investigated the accuracy of human exposure estimation to CO2, PM2.5, and PM10 by using the input variables from the stationary IAQ monitors. Regression model for each studied air pollutant was proposed while considering a different number (1, 2, or 3) and combination of IAQ input variables. Table 2 shows adjusted R² values of each model under combined and separated sitting and standing activities. Segregated human activities can improve inhalation exposure estimation for all studied air pollutants. During standing activities, accuracy for estimating CO2 inhalation exposure was 77% higher compared to one under combined activities. This result agrees with the previous report (Fig. 4) of significant improvement of correlation between stationary and breathing zone CO2 measurements when participants’ activities were separated. Accuracy of PM2.5 and PM10 exposure estimation was 8% higher during sitting activities (adjusted R2 0.93 and 0.91 respectively) compared to the ones during combined activities. In case of PM, sitting activities had better estimation accuracy relative to combined activities owing to a closer distance between seated participants and the OPCs with a fewer episodic particle emission relative to standing activities. Licina et al. [42] also identified personal cloud effect with elevated PM concentration in breathing zone of seated occupants while reporting that well-mixed representation of indoor space might underestimate human exposure to coarse particles. During sitting activities, the best single input variable for PM2.5 and PM10 exposure detection was PM measurement at the Exhaust 1 (R2 of 0.91 and 0.87), which was located near the head of the reference participant.

Table 2 Adjusted R² value of MLR models for IAQ exposure estimation by using different numbers and combinations of stationary CO2 and PM measurements during combined and separated activities (sitting and standing).

The CO2 exposure estimation by using a single stationary IAQ monitor during sitting activities was not accurate (average adjusted R² = 0.25 across all single monitors, Table 2). Furthermore, the PM2.5 and PM10 exposure estimations by using a single OPC during standing activities were also not accurate (average adjusted R² of 0.24 and 0.22 across all single OPCs, Table 2). The results indicate that the single stationary IAQ monitoring location recommended by standards and guidelines [19, 20, 43] does not capture exposure well and the measurements may not be reliable particularly when complex airflow interactions exist in the space.

Using all three IAQ inputs (Front edge of participant desk + Desk + Exhaust 1) for estimating PM2.5 and PM10 exposure showed 2% and 5% higher adjusted R2 for sitting activities, and 13% and 9% higher adjusted R2 for standing activities relative to using single IAQ input. This was not the case for CO2 exposure estimation, where there was no difference between using single and multiple variables. Further, we reported regression coefficients of the models (Table S4) consisted of a single stationary IAQ measurement and participant number as input variables with the best estimation accuracy. The regression equations (Eqs. 13) are listed based on the models (Table S4) composed with one stationary IAQ measurement and participant number (partnum) as inputs. A negative correlation between participant number and CO2 inhalation exposure was observed, while a positive correlation between CO2 level at the Front edge of participant desk and CO2 inhalation exposure was detected during standing activities (Eq. 1). As indicated in Eq. (2) and Eq. (3), two inputs (partnum, partexhaust) had a positive correlation with output (inhalation exposure to PM2.5 and PM10) during sitting activities. Interestingly, inhalation exposure to PM10 was more dependent on the participant number than the stationary PM10 measurement at the ventilation exhaust, while the opposite aspect was shown for inhalation exposure to PM2.5.

$$CO_ = - 281.51part_ + 0.829CO_\\ + 1983.328$$

(1)

$$PM_ = 0.172part_ + 1.795PM_-0.007$$

(2)

$$PM_ = 2.497part_ + 1.652PM_ + 1.098$$

(3)

MLR models based on contextual measurements

We derived the MLR models by using input variables obtained from PIRs installed at three different placements; ceiling, wall, and below the participant desk. Table 3 summarizes adjusted R² values of each model with different combination of inputs under combined and separated sitting and standing activities. The estimation accuracy did not show any significant R² values throughout all proposed models, meaning that the human presence/absence data is generally not effective in detecting personal exposures. However, data obtained by all three PIRs was moderately effective (R² > 0.5) in estimating inhalation exposure to CO2 during standing activities. Our results point towards conclusion that the PIR alone is able to detect human presence in the space (see β = 0.26, Table S5), but none of the three PIRs showed a sufficient ability to estimate inhalation exposure solely.

Table 3 Adjusted R² value of MLR models for IAQ exposure estimation by using different combinations of PIRs measurements during combined, sitting, and standing activities.MLR models based on physiological measurements

We also examined MLR models composed of physiological measurements from wearable wristband (E4), which included the skin temperature (Tskin), heart rate (HR), and resultant three-axis acceleration (ACC) of the reference participant. Adjusted R² values of each model under combined, sitting and standing activities are presented in Table 4. In general, physiological measurements gave poor estimate of inhalation exposures for the investigated scenarios except the CO2 exposure in standing activities that had a moderate accuracy (R² > 0.5). A discrepancy of estimation accuracy between sitting and standing activities is aligned with the findings of two experimental studies [44, 45] that indicated a complex relationship of human physiological status and indoor CO2 concentration. Having more than one physiological parameter could improve the estimation accuracy relative to single measurement in some cases. For example, the model accuracy for detecting PM2.5 and PM10 exposure by multiple inputs showed 5 and 10% increase in sitting activities and showed 10% increase in standing activities in case of CO2 compared to the model with a single input. However, overall model accuracy by physiological inputs was still insufficient to estimate inhalation exposures. Further, we reported regression coefficients of a model that best estimated CO2 exposure (adjusted R² = 0.594) by physiological inputs, where large β coefficient was shown in order of participant number, Tskin, and HR (Table S6).

Table 4 Adjusted R² value of MLR models for IAQ exposure estimation by using different combinations of wearable wristband measurements during combined, sitting, and standing activities.MLR models based on multiple parameter measurements

We finally derived MLR models by combining stationary IAQ, physiological (E4) and contextual (PIR) parameters and compared the results with the models composed of a single parameter. We examined the models under segregated activities (sitting and standing), which was more advantageous in terms of model accuracy relative to combined activities as previously noted in “MLR models based on stationary IAQ measurements”. Adjusted R² values of each model were reported with relevant input variables listed in parentheses (Table 5). In case of sitting activities, the estimation accuracy showed twofold (101%) increase by using multiple parameters (IAQ+E4+PIR) compared to the model with a single stationary CO2 measurement. When participants were moving around, CO2 exposure estimation was better by integrating stationary CO2 measurements with wearable (Tskin, HR) and PIR (PIR_Wall) measurement, however, the improvement was small (4–6% increase).

Table 5 Adjusted R² value (relevant input variables) of MLR models with combined input parameters for IAQ exposure estimation during sitting and standing activities.

The relevant inputs for PM2.5 and PM10 estimation during standing activities were stationary PM measurements but did not include any contextual or physiological indicators. During sitting activities, however, physiological state (Tskin, HR) of the participant was included as relevant input for PM exposure detection. Particularly, the skin temperature (Tskin) was advantageous in estimating PM10 exposure while heart rate (HR) was useful in estimating both PM2.5 and PM10 exposures. By combining IAQ with wearable and PIR measurements, adjusted R² for PM2.5 and PM10 exposure estimation models slightly improved (3–6% increase in sitting activities). During standing activities, having two stationary monitors increased the estimation accuracy by 14% compared to having a single OPC monitor. This increase, however, has little relevance as the single IAQ input was sufficient to accurately estimate the exposure.

Except a notable improvement (twofold increase) of using combined parameters in CO2 exposure estimation, the increase of model accuracy by combining the parameters was trivial. The regression equations of the best models with combined input parameters are reported as Eqs. (S1S5). We also included normality test of the final regression models (Fig. S7) in order to make valid future inferences of the models. Lastly, we presented additional regression models that used single and combined parameters during combined (sitting + standing) activities (Table S7). As expected, the best model accuracy for estimating personal exposure to CO2, PM2.5 and PM10 was not apparent when participants’ activities were mixed. This finding confirms the importance of having contextual information, particularly occupant activities, for evaluating personal exposures.

Study limitations

Our study has several limitations. Firstly, our findings are limited to a handful of selections of office setups, activities, single air change rate, and single room air distribution strategy, which means our propositions may not be applicable to completely different circumstances. Our models might have been different if the exhaust vent was not positioned near the seated reference participant, as evidenced by analyzing indoor air pollution and correlation with breathing zone concentration between two different placements of exhaust (Exhaust 1 and 2). Furthermore, being limited to measuring personal exposure of one participant, we cannot generalize expiratory characteristics (e.g., the geometry of a person’s nose, lung capacity, the position of a head) to all population. Physical intrusiveness of measurements to the participants remains a weakness because it could have influenced their movements. Lastly, experimental instruments were worn by the reference participant with a real-time camera recording, which would not be possible in a real-life scenario due to intrusiveness and privacy issues [46, 47]. To tackle these limitations, one promising technology is a novel camera-based human activity detector algorithm named PifPaf [48] that gives information about total number of participants and estimates the posture of participants containing 17 joints, without violating privacy issues.

留言 (0)

沒有登入
gif