Development of a probabilistic multi-zone multi-source computational model and demonstration of its applications in predicting PM concentrations indoors

Science of the Total Environment 490 (2014) 798–806

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

Development of a probabilistic multi-zone multi-source computational model and demonstration of its applications in predicting PM concentrations indoors J.A. McGrath a,⁎, M.A. Byrne a, M.R. Ashmore b, A.C. Terry b, C. Dimitroulopoulou c a b c

School of Physics & Centre for Climate and Air Pollution Studies, Ryan Institute, National University of Ireland Galway, Galway, Ireland Environment Department, University of York, York, UK Centre for Radiation, Chemical and Environmental Hazards, Public Health England, Chilton, Oxon, UK

H I G H L I G H T S • • • • •

A probabilistic model predicting PM concentrations in residential environments The effect that source location and household layout has on PM concentrations A modified approach for calculating PM10 deposition rate Analysing peak and mean PM concentrations from six independent emission events A sample 24-hour activity profile demonstrates multi-source emissions

a r t i c l e

i n f o

Article history: Received 1 October 2013 Received in revised form 9 May 2014 Accepted 19 May 2014 Available online xxxx Editor: Lidia Morawska Keywords: Modelling Indoor air quality PM10 PM2.5 Emissions Validations

a b s t r a c t This paper highlights the development and application of the probabilistic model (IAPPEM), which predicts PM10 and PM2.5 concentrations in the indoor environments. A number of features are detailed and justified through simulated comparison, which are shown to be necessary when modelling indoor PM concentrations. A one minute resolution predicts up to 20% higher peak concentrations compared with a 15 min resolution. A modified PM10 deposition method, devised to independently analyse the PM2.5 fraction of PM10, predicts up to 56% higher mean concentrations. The application of the model is demonstrated by a number of simulations. The total PM contribution, from different indoor emission sources, was analysed in terms of both emission strength and duration. In addition, PM concentrations were examined by varying the location of the emission source. A 24-hour sample profile is simulated based on sample data, designed to demonstrate the combined functionality of the model, predicting PM10 and PM2.5 peak concentrations up to 1107 ± 175 and 596 ± 102 μg m−3 respectively, whilst predicting PM10 and PM2.5 mean concentrations up to 259 ± 21 and 166 ± 11 μg m−3 respectively. © 2014 Elsevier B.V. All rights reserved.

1. Introduction The adverse effects on human health of Particulate Matter (PM) are well documented (e.g. COMEAP (2006, 2009, 2010)). According to the recently finished WHO project REVIHAAP (WHO, 2013), there is a large body of evidence of the effects of long term PM exposure on both all causes and cardiovascular mortality. The PM related morbidity outcomes include are associated with bronchitis symptoms in children, chronic bronchitis in adults, asthma attacks in all ages, cardiovascular, cerebrovascular (possibly) and respiratory hospital admissions, in all ages, emergency care visits due to asthma and cardiovascular disease ⁎ Corresponding author. E-mail address: [email protected] (J.A. McGrath).

http://dx.doi.org/10.1016/j.scitotenv.2014.05.081 0048-9697/© 2014 Elsevier B.V. All rights reserved.

in all ages, and finally restricted activity days for adults. The associations between long-term PM2.5 exposure and the incidence of coronary events remain for concentrations below the current European limits, supporting lowering of these limits in order to protect public health (Cesaroni et al., 2014). Short-term exposure to PM2.5 over a few hours to weeks can trigger cardiovascular morbidity and mortality events (Brook et al., 2010). Based on toxicological and clinical studies, there is significant evidence that peak exposures of short duration (ranging from less than an hour to a few hours) to combustion-derived particles leads to immediate physiological changes (WHO, 2013). Indoor PM concentrations are affected by the infiltration of outdoor particles, meteorological parameters and seasonal effects, indoor activities of occupants, emissions from indoor PM sources, removal of

J.A. McGrath et al. / Science of the Total Environment 490 (2014) 798–806

799

particulates by deposition, the dilution of indoor PM through external and internal air exchange and internal house layout (Ferro et al., 2009; Singer et al., 2002; Ott, 1999). Ebelt et al. (2005) showed that the health effects associated with indoor-generated particles differ from those arising from particles of outdoor origin. Research has highlighted that indoor activities contribute to PM concentrations through combustion events such as smoking, frying, solid fuel fire and use of candles and incense (Ott and Siegmann, 2006; Ozkaynak et al., 1996; He et al., 2004) and resuspension activities; such as walking, dusting and vacuuming (Ferro et al., 2004). The indoor environment deserves particular attention since people spend approximately 89% of their time there (Klepeis et al., 2001). It is often expensive or impractical to obtain direct indoor measurements (or personal exposure measurements) for large population groups in epidemiological studies and computational models are a recognised substitute. Computational modelling has the benefits of cheaply and easily evaluating population exposure studies, along with potential PM reductions, through changes in building design related strategies or behavioural strategies. CONTAM (NIST, Gaitherburg, MD, USA) is a multi-zone, airflow and transport pollutant model considering airflow paths, ventilation system and emission sources (NIST, 2011). To date, CONTAM has been used in over 54 published applications (NIST, 2010) making it the most widely used indoor air pollutant model. Fabian et al. (2012) used CONTAM to predict NO2 and PM2.5 concentrations in low income family homes in Boston, for use in health based intervention study, highlighted the challenges imposed on simulations due to the large variations in emission strengths. CONTAM, being a deterministic model, fails to consider variations in modelling parameters. Furthermore, as an airflow model, it requires very detailed parameterisation of building elements in order to predict airflows, which impose even greater uncertainty. All of the above limit its use in complex indoor environments as every input parameter in a modelling has a level of uncertainty associated with it. Probabilistic models, using probability density functions, simulate a range of possible values for each input parameter, overcoming some of the uncertainties in experimentally obtained data, but can also encompass uncertainties in the selection of appropriate modelling parameters between studies. Burke et al. (2001), using a probabilistic model, predicted daily averages for PM2.5 exposure for the population living in Philadelphia. The model did include cooking and smoking emission sources in the residential environment but it only calculated a 12 hour average for each microenvironment. Dimitroulopoulou et al. (2006) developed a probabilistic model to predict PM10 and PM2.5 concentrations every 15 min in a three-room layout of UK homes, under smoking or cooking emission based scenarios. Whilst current models have demonstrated the capability to predict indoor PM10 and PM2.5 concentrations, the need still exists for a comprehensive probabilistic model capable of simulating more realistic representations of a home environment. The model should encompass a full range of possible emission sources situated in a range of different room locations, and simulated on a sufficiently short time scale so as to capture details on peak and mean concentrations and accurately determine the time duration for emission concentrations to fully decay.

discusses the microenvironmental aspect of the model. At each time interval, the input parameters are generated using probability density functions.

2. Model description

2.3. Preliminary validation

2.1. Background of the model

Model validations are reported in McGrath et al. (2014) where experimental vs. predicted PM concentrations were examined in the context of varying interzonal airflow, using a one-minute time resolution in a six-room apartment. Linear regression values of experimental vs. modelled PM concentrations ranged from 0.97 to 1.07 without variations in interzonal airflow. This study also concluded that the modelling approach used, accurately predicted PM concentrations for interzonal airflow variations for durations of 10 min or greater. In addition to the work reported in McGrath et al. (2014), a number of model validations were carried out in the context of indoor PM emission scenarios. The same building (room dimensions and airflow rates)

The current model, titled Indoor Air Pollutant Passive Exposure Model (IAPPEM), is based on the INDAIR model described by Dimitroulopoulou et al. (2006), which is an advanced probabilistic modelling tool that can evaluate the contribution of indoor and outdoor sources to the air pollution concentration in the indoor environment. IAPPEM calculates the change in indoor pollutant concentrations by solving the differential Eq. (1); considering the infiltration of outdoor air pollution, the generation of air pollution indoors and its transport between rooms, and the indoor deposition of air pollution. This paper


 n X dCk ðλ0k Þ A Q λik ¼ ð f k  C0 −Ck Þ−vg k Ck þ k þ ðCi −Ck Þ: dt Vk Vk V Vk i¼1 k

ð1Þ

Eq. (1) is solved for each k, where k represents each individual room. Subscripts of 0, 1 and 2 are used to represent outside, room 1 and room 2 for different parameters. Ck represents the concentrations of the pollutant in that room (μg m−3), where C0 represents the outdoor concentration (μg m−3). fk represents the building filtration factors between the outdoor and that room. vg is the deposition velocity of the pollutant (m h−1). λ(ik) is the interzonal airflow between internal rooms, e.g. λ(12) represents the transport of pollutants from room 1 into room 2, and λ(0k) the external airflow between the outside and room k (m3 h−1). Ak is the surface area of room k (m2). Vk is the volume of room k (m3). Qk is the indoor emission rate of the pollutant in room k (μg h−1). 2.2. Adaptations To date, in the domestic building context, the INDAIR model has been applied to the simulation of PM concentrations arising from (i) passive smoking in a two-room dwelling (Dimitroulopoulou et al., 2001) and (ii) to the generation of NO2, CO, PM10 and PM2.5 concentrations in a three-room dwelling, with the pollutants arising from separately-occurring smoking and cooking events (Dimitroulopoulou et al., 2006). In developing the IAPPEM model, where the aim is to demonstrate capability in simulating PM concentrations arising from a number of simultaneously-operating sources in a number of different rooms of a multi-room building, several adaptations were made to the INDAIR approach, and these are: a) Modifying the approach for calculating the PM deposition rate. Indoor air pollution models to date, calculate PM10 decay rates independently of PM2.5. However, for combustion sources, PM2.5 can constitute up to 90% of PM10 (as discussed in Section 2.5.3) which implies that the PM10 deposition velocity needs to encompass the entire particle size fraction. The approach taken in this paper is to separate the contribution of PM2.5 from PM10, calculate the deposition rate for both size fractions independently, and then recombine for the calculation of the deposited PM10 concentration. This approach is expressed in Eq. (2), where γ represents the deposition rate.     PM10deposited ¼ ðPM10 −PM2:5 Þ  γPM10−2:5 þ PM2:5  γPM2:5 :

Þ

ð2Þ

b) Incorporation of up to 12 simultaneously-operating emission sources, in up to 15 different rooms in a dwelling.

800

J.A. McGrath et al. / Science of the Total Environment 490 (2014) 798–806

Table 1 This table shows a comparison between experimental vs. simulated concentration for a range of different emission sources. The experimental uncertainty on the measured data represents ± 1 μg m−3. Emission scenario

PM2.5 concentrations Measured

Simulated

(μg m−3)

(μg m−3)

Peak concentrations A single cigarette A frying event Incense stick

181 418 593

167 ± 14 398 ± 62 633 ± 70

Mean concentrations No emission source (two-hour mean) Smoking six cigarettes (four-hour mean) A frying event (two-hour mean) Incense Stick (six-hour mean)

7.3 296 289 326

6.5 294 276 331

± ± ± ±

1.8 14 43 41

and experimental methodology were used as detailed in McGrath et al. (2014); however emission rates and deposition rates are all based on values from the current manuscript. Experimental data are shown in Table 1, and were collected to support the simulations described in the current manuscript. 2.4. Simulations Each scenario is simulated with a one minute time resolution over a 24-hour period, starting at 08:00 am in an 11-room dwelling (the layout is discussed in Section 2.5.2). The outdoor PM concentrations, room dimensions (volume and surface area) and the external airflow were held constant throughout all the simulations. Emission rates were varied depending on the emission source, and interzonal airflow varied depending on whether the doors were open or closed. An emission event refers to the period during which an emission source is present and corresponds to the duration as specified in Table 3. The separate simulations that were carried out are detailed below: Simulation (i) provides justification for the model's modifications with results summarised in Sections 3.1 and 3.2; whilst simulations (ii–iv) focus on the model's overall application with results summarised in Sections 3.3 to 3.6. i) To demonstrate the effect of modifying the deposition rate calculation in the model, and also the effect of the improved time resolution, a smoking event at 12:00 was simulated in the kitchen with the internal doors closed. ii) 12 independent simulations, with six different emission sources in each, and with the doors open and closed, were carried out. In each simulation, the emission source started at 12:00 and was assumed to be in the kitchen. Analysis of the PM concentrations in each case yielded information on the overall PM contribution from each source and permitted a calculation of the time required for an emission event to result in an exceedance of an internationally recognised exposure limit value. iii) To demonstrate the effect that source location and household layout has on PM concentrations, and to highlight the importance of a realistic representation of the home environment, six independent emission events were simulated with each emission source in a different location. Emission periods commenced at 12:00 for each simulation. iv) A sample 24-hour activity profile was simulated, to demonstrate the combination of all the adaptations to the model.

2.5. Model parameterisation To simulate the scenarios mentioned above, it is necessary to assign values for the input parameters in the model. It should be noted that

whilst Irish data was selected solely for demonstration purposes, this does not reflect any geographical limitations in the model. 2.5.1. Outdoor PM concentration Ambient outdoor PM10 and PM2.5 concentrations were estimated from data supplied by the Irish EPA (2013). The data corresponded to hourly concentrations at an urban background monitoring station, 3 km from Dublin city centre. Mean and standard deviations were extrapolated from these data to generate a 24-hour profile. 24-hour mean PM10 and PM2.5 concentrations were 17.14 ± 0.33 μg m−3 and 9.77 ± 0.24 μg m−3 respectively. 2.5.2. Room dimensions Data on room dimensions were collected from a sample of 50 houses from an online estate agent (Daft, 2013), in accordance with property classification on the Irish Census (Central Statistics Office, 2013). Room dimensions are listed in Table 2. The house layout is as follows: the hallway was connected to each bedroom, the bathroom, the kitchen and the living room. The kitchen alone was connected to a utility room and dining room, and Bedroom 1 alone was connected to an en-suite bathroom. Fig. 1 provides an illustration of a representative household layout that was used for these simulations. Previous simulations by McGrath et al. (2014) examined a six-room household. 2.5.3. Indoor sources of PM Table 3 summarises PM10 and PM2.5 indoor emission rates, together with the corresponding emission durations; emission values and durations have been compiled from literature and if multiple references were found, mean and standard deviations were calculated. When insufficient information was found in literature for PM10 emission rates, they were extrapolated based on a relationship between PM10 and PM2.5. These relationships are also summarised in Table 3. 2.5.4. Deposition velocities & penetration factors In test chamber studies, Byrne et al. (1995) found that approximately 60–65% of particles between 0.7 and 2.5 μm deposited on the floor and that 35–40% deposited on the walls, whilst 75–80% of the larger size particles (uptown 5.4 μm) deposited on the floor with the remainder depositing on the walls. Byrne et al. (1995) found that particles with a diameter of 0.7 and 2.5 μm had deposition velocities of 4.1 × 10−5 and 6.2 × 10−5 m s−1 respectively. Based on these values, a PM2.5 deposition velocity of 5.15 × 10−5 m s−1 was selected. This was shown to be in agreement with simulations by McGrath et al. (2014). In the study by Byrne et al. (1995) the largest particles present were 5.4 μm and for that reason, a PM10 deposition velocity value of 3.9 × 10− 4 m s− 1 was selected from the PTEAM study (Ozkaynak et al., 1996). Table 2 Summary of the room dimensions as an input parameter for the model. A range of possible values are generated using probability density functions. Location

Wall surface Mean (St dev) (m2)

Floor surface Mean (St dev) (m2)

Volume Mean (St dev) (m3)

Utility room Kitchen Dining room Hallway Living room Bathroom Bedroom 1 Bedroom 2 Bedroom 3 Bedroom 4 En suite bathroom

22.8 (4.7) 46.8 (11.5) 40.5 (5.6) 35.2 (9.8) 44.6 (8.7) 26.2 (5.6) 37.4 (6.4) 35.6 (6.1) 38.8 (5.6) 34.4 (6.5) 18.4 (2.1)

5.1 (1.8) 21.9 (11.4) 16.4 (4.6) 9.7 (4.8) 19.9 (7.5) 6.5 (2.4) 14.0 (5.0) 12.7 (4.7) 12.1 (3.8) 11.9 (4.8) 2.9 (0.6)

12.7 (4.4) 54.7 (28.6) 40.9 (11.5) 24.1 (11.9) 49.7 (18.8) 16.3 (6.0) 35.1 (12.5) 31.8 (11.7) 30.3 (9.6) 29.7 (12) 7.3 (1.6)

J.A. McGrath et al. / Science of the Total Environment 490 (2014) 798–806

801

Fig. 1. A diagram illustrating a representative household layout.

Chen and Zhao (2011) reviewed the current literature on infiltration factors; based on this review, a value of 0.7 ± 0.3 was selected for the building fabric infiltration factor for these simulations. 2.5.5. Airflow rates All the rooms are naturally ventilated with external windows and external doors assumed to remain closed for the entire duration. Crump et al. (2005) reported external air exchange rates ranging from 0.44 ± 0.11 air changes per hour, which were converted into volumetric air flow rates (m 3 h− 1) for each room. Internal doors were simulated as either all closed or all open. In the absence of a detailed Irish study, the U.S. volumetric air flow rates were used as a substitute, with airflow rates for the closed door condition ranging from 0.4 to 5.1 m3 h− 1 and with airflow rates for the open door condition ranging from 60 to 245 m3 h− 1 (Ott et al., 2003; Miller and Nazaroff, 2001; Ferro et al., 2009). 3. Results 3.1. Time resolution The importance of choosing an appropriate model time resolution is illustrated by analysing the comparison between a one minute resolution and a longer time resolution. A 15 minute time resolution was selected for demonstrative purposes. Each resolution independently simulating the kitchen smoking event described in Section 2.4 was independently simulated. To directly compare a one minute and 15 minute resolution, both smoking durations are set to 15 min i.e. the value of nine minutes, as specified in Table 3, was not used. For the one minute resolution test case, the peak PM10 and PM2.5 concentrations were 625 ± 245 μg m−3 and 428 ± 175 μg m−3 respectively, whereas for the 15 minute resolution, peak PM10 and PM2.5 concentrations were 555 ± 225 μg m−3 and 381 ± 170 μg m−3 respectively. This difference arises as the 15-minute

resolution case overestimates PM decay, whereas the one minute resolution case calculates PM decay at individual minutes and gives a better estimate. If the nine minute smoking duration, as specified in Table 3, is used, the effect of carrying out a simulation of 15 minute duration is that peak PM10 and PM2.5 concentrations of 338 ± 141 μg m−3 and 239 ± 108 are calculated, as compared with peak PM10 and PM2.5 concentrations of 405 ± 177 μg m− 3 and 275 ± 124 μg m− 3, associated with a one minute time resolution. This is because an additional 6 min of PM decay are incorporated in the 15 minute resolution case, as it averages over a period that is longer than the actual emission period. 3.2. Deposition The above simulation also highlights the effect of incorporating a modified deposition rate equation in the model, as was described in Section 2.2. Fig. 2 shows PM2.5 and PM10 concentrations, estimated by the modified method, together with PM10 concentrations estimated by the original method. Fig. 2 shows how the original method results in PM10 concentrations decaying faster than PM2.5, as PM10 is dominated by smaller particles for combustion sources (PM2.5 represents 66% of the PM10 for smoking emissions (Section 2.5.3)). By the new PM10 deposition method, calculated PM10 concentrations remain higher than PM2.5 concentrations, which make greater physical sense. For this simulation, the 24-hour mean PM10 concentration calculated by the method that incorporates the original decay equation was 18 ± 3 μg m−3 compared with 28 ± 2 μg m−3 calculated by the revised method; this represents a difference of 56%. 3.3. The relative importance of different sources To demonstrate the relative contributions of different sources to overall PM contribution, a series of different simulations were run in

Table 3 PM10 and PM2.5 emission rates, with the corresponding emission duration for each source. Comments are provided when extrapolating a PM10 emission rate based on PM2.5 data. PM2.5 emission strength (mg h−1)

Source and duration in minutes

PM10 emission strength (mg h−1)

Smoking (9)

145.5 ± 42.7

96.0 ± 28.2

Frying (8)

189.3 ± 163.5

160.8 ± 130.8

Incense stick (40)

108.9 ± 118.0

98.0 ± 106.9

39.6 ± 37.8 5.3 ± 2.8

12.0 ± 12.6 5.3 ± 2.8

Solid fuel fire (210) Candles (15)

References

Comments

Brauer et al. (2000); He et al. (2004); Ozkaynak et al. (1996); Ott (1999); Ott et al. (2003); Jiang et al. (2011) He et al. (2004)

Ozkaynak et al. (1996) found that PM2.5 comprised 66% of PM10. Lee et al. (2001) estimate that PM2.5 encompasses 80% of PM10 for cooking events.

See and Balasubramanian (2011); Lee and Wang (2004); Jiang et al. (2011); Jetter et al. (2002) Guo et al. (2008) Fine et al. (1999); Pagels et al. (2009)

PM2.5 emissions were assumed to be the same for PM10 Zai et al. (2006).

802

J.A. McGrath et al. / Science of the Total Environment 490 (2014) 798–806

400

Concentrations (µg m-3)

PM10 New Decay Method PM10 Old Decay Method PM2.5

300

200

100

12:00

13:00

14:00

15:00

16:00

17:00

18:00

Time (Hours) Fig. 2. Differences in mean PM10 concentrations estimated using the modified PM10 decay method, the original PM10 decay method and the corresponding mean PM2.5 concentration during a smoking event in the kitchen when doors are closed.

the kitchen, each one incorporating one of the following combustion events: smoking, frying, solid fuel fire (wood/coal/peat), incense and candles; the duration of each event, and the appropriate emission rate, is detailed in Table 3. In each case, separate simulations were run with the internal kitchen door closed and open, and “no source” simulations were also run. Fig. 3 shows a box plot of the predicted peak PM10 concentrations, demonstrating the advantages of a probabilistic model in simulating the range of possible concentrations. Table 4 summarises PM10 and PM2.5 concentrations contributed by different sources. It can be seen that PM10 and PM2.5 concentrations are both higher when the kitchen door is closed, compared with the open door case, and this is most evident in the 24-hour mean concentration data. Although frying has a marginally shorter emission duration than smoking, frying's higher emission rate generates higher PM concentrations. The incense stick and the solid fuel fire both have lower emission rates than either smoking or frying, but the longer emission durations impact substantially on PM concentrations. The highest peak PM10 and PM2.5 concentrations are associated with the burning of the incense stick are at least twice those associated with all other sources. Interestingly, higher PM10 peak concentrations are observed for solid fuel fire usage than for frying, although frying has the higher PM2.5 peak concentrations. Although the PM10 incense peak concentration is twice that of the solid fuel fire peak, there is no substantial difference between the mean PM10 concentrations for either source.

Fig. 3. A box plot showing the range of predicted peak PM10 concentrations in the kitchen when the doors are closed for the five indoor emission sources. Each emission source was simulated separately.

3.4. Exceedance of 24 hour guideline values Table 4 presents the calculated time for which each single PM emission event, as described in Table 3, results in an exceedance of the WHO (2005), 24-hour mean PM10 and PM2.5 concentration guidelines of 50 μg m−3 and 25 μg m−3 respectively. Although these guidelines represent 24-hour mean concentrations and not threshold limits, they give indications of the extent of PM concentration decay in each case. In all cases, PM2.5 concentrations exceed the guidelines for a longer time than PM10, as a consequence of the lower PM2.5 limit and the lower PM2.5 deposition velocity. With doors closed between rooms, the limits are exceeded for longer times than when doors are open. Whilst incense and the solid fuel fire have similar PM 10 mean concentrations, PM 2.5 mean concentrations for the solid fuel fire are approximately half that of the incense, but there is no noticeable difference between the time for which each exceed the WHO guidelines. This is due to the longer duration of the solid fuel burning event, which is of a timescale such that the incense-originating PM can decay to a similar concentration to those occurring that at the end of the solid fuel fire emission period.

3.5. The effect of source location on PM concentration in various rooms In this section, simulated PM concentrations in various rooms of the dwelling, according to whether they contain an emission source, or are subject to the influence of a source in another room, are examined. Table 5 shows peak PM10 and PM2.5 concentrations in Bedroom 1, associated with a series of separate emission events, of durations detailed in Table 3, in the bedroom itself, and in four other rooms; in each case, the emission event is set to commence at 12:00. When these data are compared with those shown in Table 4, it can be seen that all PM concentrations in the bedroom are lower, except in the case where candles are burnt in the bedroom itself. When the doors remain closed; frying, incense and solid fuel fire burning result in only a marginal increase, relative to background (no source scenario) in peak concentrations in the bedroom, as the room containing the sources is some distance away. However, the smoking event gives rise to higher concentrations in the bedroom, due to the source being in the adjoining room. The smoking scenario results in elevated PM concentrations, with peak PM10 and PM2.5 smoking concentrations being only 64% and 46% lower than those associated with burning an incense stick; when both

J.A. McGrath et al. / Science of the Total Environment 490 (2014) 798–806

803

Table 4 Predicted PM10 and PM2.5 peak and mean concentrations in the kitchen during six different emission scenarios, each with the doors open and then with the doors closed. The duration for which PM10 and PM2.5 concentrations exceeded the 24-hour guidelines are also included. Source

Peak concentrations PM2.5 μg m−3

PM10 μg m−3

24-hour mean concentrations

Time exceeding WHO guidelines

PM10 μg m−3

PM2.5 μg m−3

PM10 (hh:mm)

PM2.5 (hh:mm)

Doors closed No emission Smoking Frying Incense stick Solid fuel fire Candles

8 405 526 1474 670 31

± ± ± ± ± ±

1 177 248 497 88 10

6 275 410 1352 295 29

± ± ± ± ± ±

1 124 196 466 35 10

7 28 36 114 115 9

± ± ± ± ± ±

1 2 4 10 12 1

5 22 30 106 54 7

± ± ± ± ± ±

0 2 4 11 4 0

–:– 02:52 03:26 06:19 07:03 00:00

–:– 04:02 04:47 07:58 08:02 00:15

Doors opened No emission Smoking Frying Incense stick Solid fuel fire Candles

7 245 330 592 228 19

± ± ± ± ± ±

0 65 101 92 18 3

6 164 252 539 96 18

± ± ± ± ± ±

0 44 78 85 6 3

7 14 16 40 41 8

± ± ± ± ± ±

0 1 1 3 2 0

5 10 13 36 20 6

± ± ± ± ± ±

0 1 1 3 1 0

–:– 00:59 01:19 03:44 04:49 00:00

–:– 01:39 02:17 05:22 05:36 00:00

sources are in the kitchen PM10 and PM2.5 percentage differences were 75% and 80% lower. Peak concentrations that are similar to those arising from the solid fuel fire and higher concentrations than those associated with frying are observed. Although smoking still results in the second lowest mean concentrations, there is an only a marginal difference between concentrations. Interestingly, the solid fuel burning scenario, with the fire now situated in the living room, results in elevated concentrations in the bedroom, although still connected by the same number of rooms to the room containing the source, relative to the previous case where the fire was in the kitchen. The two additional rooms connected to the kitchen in combination with the living room's smaller volume, result in higher concentrations in the living room, allowing greater particle infiltration into the hall and subsequently into the bedroom. Whilst the incense stick is in the dining room (one room further away from the bedroom than is the kitchen) decreased concentrations are observed in the bedroom. With both the solid fuel fire and the incense stick in the kitchen, the incense stick resulted in 64% and 180% higher PM10 and PM2.5 concentrations respectively in the bedroom than the fire, but by adjusting only the location of both sources, increases in peak PM10 and PM2.5 concentrations reduce to only 39% and 101% respectively.

Table 5 Peak and mean PM2.5 concentrations in Bedroom 1 as different emission sources are independently simulated in different locations throughout the dwelling. Peak concentrations

Peak concentrations

Doors open

Doors closed

PM10 μg m−3

PM2.5 μg m−3

PM10 μg m−3

PM2.5 μg m−3

Emission source and location No source Smoking in hallway Frying in kitchen Incense stick in dining room Fire in living room Candles in Bedroom 1

7 79 43 124 89 30

± ± ± ± ± ±

0 16 7 16 8 4

7 54 35 116 57 28

± ± ± ± ± ±

0 11 6 16 6 5

7 29 8 8 11 43

± ± ± ± ± ±

0 6 1 1 1 11

7 22 6 6 8 41

± ± ± ± ± ±

0 11 1 16 6 5

Emission source and location No source Smoking in kitchen Frying in kitchen Incense stick in kitchen Fire in kitchen Candles in kitchen

6 34 43 136 83 8

± ± ± ± ± ±

0 4 7 18 6 0

6 25 35 126 45 7

± ± ± ± ± ±

0 3 6 19 3 0

6 6 8 10 7 6

± ± ± ± ± ±

0 1 1 1 1 1

6 6 6 10 7 6

± ± ± ± ± ±

0 1 1 1 1 1

3.6. A simulation of multiple emission events A sample 24-hour activity profile, incorporating simultaneously occurring emission sources, is presented in Table 6, and Table 7 shows the corresponding estimated PM10 and PM2.5 peak and mean concentrations, in the various rooms of the dwelling. Fig. 4 shows PM10 concentration in the living whilst the doors are open, this figure helps demonstrates the values predicted by a probabilistic model. The peak living-room concentration occurs at 19:39, prior to the end of the solid fuel fire emission, the period demonstrating the combined effect of two emission sources. During 19:39 to 21:30, the PM decay rate in the living room exceeds the emissions from the solid fuel fire; this is evident in Fig. 4 by the slow decrease in concentrations, once the solid fuel fire emissions cease at 21:30 this results in a greater decay of PM concentrations. Whilst peak kitchen concentrations remain broadly similar to those shown in Tables 4 (single frying scenario), marginally increased concentrations, due to infiltration from the living room, are discernible. Although peak kitchen concentrations occur after the second frying event, concentrations differ by less than 5%, compared with the first frying event. The highest mean PM concentrations are observed in the living room, as the solid fuel fire is the largest PM contributor modelled in this scenario (Section 3.3). As previously discussed (Section 3.5), the closing of doors between adjoining rooms can be an effective method of reducing dispersion of PM concentrations; this is evident in the current simulation set, where only marginally increased PM concentrations are observed in Bedroom 1. In the open door scenario, the hallway experiences higher mean PM concentrations than the kitchen, although the hallway contains no emission sources; this is more evident for PM10 than for PM2.5, due to the solid fuel fire's larger PM10 emission rate.

Table 6 A 24-hour profile showing a number of different emission sources present at different times and locations throughout the dwelling. Time

Location

Source

10:30–10:39 12:00–12:09 13:30–13:38 17:00–17:09 18:00–21:30 18:00–18:08 19:30–19:39 22:00–22:09

Living room Living room Kitchen Living room Living room Kitchen Living room Living room

Smoking Smoking Frying Smoking Solid fuel fire Frying Smoking Smoking

804

J.A. McGrath et al. / Science of the Total Environment 490 (2014) 798–806

Table 7 The PM10 and PM2.5 peak and mean concentrations in different rooms after simulating the 24-hour profile scenario with doors opened and doors closed. Additional information is provided on the times of peak concentrations and duration in each room for which the WHO 24-hour guideline values are exceeded. Peak times refers to when peak PM10 concentrations occurs, in most cases peak PM2.5 times occurred at the same time as PM10, except for a = 19:19 and b = 22:43. Source

Peak time

Peak concentrations PM10 μg m−3

24-hour mean concentrations

Time exceeding WHO guidelines

PM2.5 μg m−3

PM10 μg m−3

PM2.5 μg m−3

PM10 (hh:mm)

PM2.5 (hh:mm)

Doors closed Kitchen Dining room Hallway Living room Bedroom 1

18:08 19:11a 20:35 19:39 21:48b

535 27 81 1107 14

± ± ± ± ±

214 4 10 175 1

420 23 54 596 10

± ± ± ± ±

166 4 6 102 1

65 13 30 259 10

± ± ± ± ±

7 1 2 21 1

55 11 23 166 7

± ± ± ± ±

6 1 2 11 0

7:12 0:00 4:57 17:00 0:00

9:45 00:00 8:53 18:37 0:00

Doors open Kitchen Dining room Hallway Living room Bedroom 1

18:08 18:20 19:47 19:39 20:08

358 160 202 641 162

± ± ± ± ±

99 29 19 83 13

281 129 122 361 102

± ± ± ± ±

78 24 12 52 8

51 44 57 102 49

± ± ± ± ±

2 2 2 4 2

39 35 40 62 36

± ± ± ± ±

2 2 2 2 2

7:56 7:49 9:45 10:28 8:38

11:24 10:56 13:13 13:37 12:48

In the open door scenario, the higher concentrations in the hallway also affect the kitchen's PM decay. One hour after the second frying event, the kitchen's PM10 concentration decayed to 32% of the peak concentrations, compared to 18% in the single frying scenario (Table 4) resulting in overall higher mean concentrations in the kitchen; this effect is not as obvious in the closed door scenario, as the kitchen is located two closed doors away from the hallway. 4. Sensitivity analysis

PM10 Concentrations (µg m-3)

Table 8 highlights the results of the sensitivity analysis, simulating a 20% independent decrease in each input parameter. Peak and mean concentrations in the kitchen were analysed whilst the doors were open, and concentration differences between simulations with original parameters values and simulations with the 20% decreased parameter values were compared. The results are reported to two decimal places; otherwise the deposition velocity, airflow and room dimensions would appear to have no effect on PM2.5 concentrations in the no source scenario due to the lower concentrations. In the absence of indoor emission sources, outdoor PM concentrations have the greatest influence on indoor PM concentrations; without large percentage difference being observed between peak and mean concentrations. In the presence of indoor emission sources, peak concentrations are dominated by emission rates and room dimensions; and their influence on mean concentrations increases as indoor emissions increase. The influence of outdoor concentrations decreases as 700 600 500 400 300 200 100

10:00

12:00

14:00

16:00

18:00

20:00

22:00

Time (Hours) Fig. 4. PM10 concentrations in the living room. The time axis has been scaled to focus on the emission period. The y bars represent one standard deviation at each time step, highlighting the probabilistic nature of the model. Each of the five peak concentrations refers to the end of a smoking event.

indoor emission rates increase; however, the influence of ventilation rates still remains an important factor, varying from a positive to a negative influence as indoor emission rates increase. In the absence of indoor emission sources, indoor concentrations affected by the outdoor air decease as ventilation rates are reduced. However, in the presence of indoor emission sources, a reduction in ventilation rates increases indoor concentrations as indoor emission rates increase. In the absence of indoor emission sources, interzonal airflows do not appear to have a major influence on PM concentrations that are already uniform throughout the residence. In the smoking scenario (Section 2.4(i)), lower interzonal airflows increase PM concentrations in the kitchen, however in the 24 hour profile, the interzonal airflows appear to be the least important factor; this is due to the similar concentrations in the hall arising from emission sources in the living room. 5. Discussion In this paper, the main features of a multi-zone multi-source compartmental model for estimating indoor PM concentrations have been demonstrated. Accompanying validation experiments compared predicted vs. experimental PM concentrations; in all cases, predicted concentrations lie within one standard deviation of the experimental concentrations, further demonstrating the advantages of probabilistic modelling. The importance of modelling at an appropriate time resolution was illustrated in Section 3.1, where it was seen that estimating PM concentrations from a smoking source at intervals greater than one minute led to an over-estimation of PM decay, and consequently an underestimation of the peak concentration, and this effect worsened for greater modelled time intervals; there was a 20% difference in estimated peak PM concentration for a smoking event modelled over a 15 minute interval, compared with a one minute interval. The adapted PM10 deposition method demonstrates the benefit of separating PM2.5 from PM10, for the purposes of calculating PM10 deposition. Further separating PM10 or PM2.5 into smaller size fraction bins (e.g. PM1) would allow a greater level of accuracy in simulating PM deposition. However, at present, insufficient information is available in the literature on PM1 emission rates and deposition velocities. In Section 3.3, overall PM concentrations from different indoor emission sources were compared, and it was observed, when comparing the PM concentrations associated with the solid fuel fire and the incense stick, that the source giving rise to the largest peak PM concentration does not necessarily give rise to the greatest mean PM concentration. Further, it was observed that the source with the largest PM10 contribution does not necessarily have the largest PM2.5 contribution. In addition, the contribution of various emission sources was seen to be

J.A. McGrath et al. / Science of the Total Environment 490 (2014) 798–806

805

Table 8 The actual and percentage differences in modelled PM10 and PM2.5 concentrations following a 20% independent decrease in each input parameter. Peak and mean concentrations represent values in the kitchen whilst the doors are open, under three different emission scenario. Scenario

Peak PM10 μg m−3

Mean PM10 μg m−3

Peak PM2.5 μg m−3

Mean PM2.5 μg m−3

No source Deposition velocity Interzonal airflow Outdoor concentrations External airflow Room dimensions

0.63 (6%) 0.01 (0%) 2.07 (20%) −0.82 (−8%) 0.80 (8%)

0.59 (6%) 0.01 (0%) −1.87 (−19%) −0.76 (−8%) 0.73 (7%)

0.24 (3%) 0.01 (0%) −1.54 (20%) −0.45 (6%) 0.41 (5%)

0.24 (3%) 0.01 (0.14%) −1.36 (−19%) −0.30 (−5%) −0.38 (5%)

Smoking Deposition velocity Interzonal airflow Outdoor concentrations External airflow Emission strength Room dimensions

−1.74 (−1%) 18.39 (7%) −5.39 (−2%) −3.70 (−1%) −49.57 (−20%) 36.59 (15%)

0.87 (5%) 0.46 (3%) −1.85 (−11%) −0.06 (−0%) −1.34 (−8%) −1.06 (6%)

−1.57 (−1%) 10.76 (6%) −4.12 (−2%) −2.80 (−2%) −34.28 (−20%) 24.64 (15%)

0.35 (3%) 0.28 (2%) −1.37 (−11%) 0.20 (2%) −1.05 (−8%) 0.57 (5%)

24-hour profile Deposition velocity Interzonal airflow Outdoor concentrations External airflow Emission strength Room dimensions

6.49 (2%) 11.99 (3%) −3.30 (−1%) 1.86 (1%) −71.41 (−20%) 56.71 (16%)

3.14 (6%) −0.83 (−2%) −1.89 (−4%) 4.63 (9%) −8.16 (−16%) 5.10 (10%)

2.42 (1%) 11.00 (4%) 0.07 (0%) 4.68 (2%) −57.79 (−20%) 44.08 (15%)

1.03 (3%) −0.30 (−1%) −1.33 (−3%) 4.07 (11%) −6.18 (−16%) 2.84 (7%)

strongly influenced by the internal door configuration; for the series of emission sources modelled, the solid fuel fire was seen to result in the second highest PM10 peak concentration when it was in a closed room, but when the doors were open, it resulted in the second lowest peak concentration. The order of the mean PM concentrations was not however affected by door opening. When analysing the decay of particles (Section 3.4) it was found that PM10 and PM2.5 concentrations remained elevated by 60% and 54% one hour after emissions ceased. This confirms earlier findings (Ott et al., 2003) which showed that Environmental Tobacco Smoke remained elevated by 40–70% of background concentrations for an hour after one cigarette has been smoked. In Section 3.5 it was demonstrated that source location influences PM concentrations. Whilst doors remain closed, adjoining rooms experienced increased PM concentrations, although rooms separated by two or more closed doors experienced no noticeable effects. This finding demonstrates that the closing of internal doors can be an effective PM exposure control strategy in a multi-zone environment. Whilst the doors remained open, it was seen that source location also influenced PM concentrations, the additional rooms allowing for greater PM dilution and deposition. A similar finding is reported in an experimental study by Ferro et al. (2009), who observed the influence of both internal room layout and internal door configuration on measured PM concentration. In Section 3.3, the significance of a probabilistic model for predicting variations in indoor PM concentrations due to indoor emission sources is demonstrated. Based on these results, experimental studies can assess the range of PM concentrations from individual or combined emission events; explaining some of the variability in indoor PM concentrations reported between experimental studies. In Section 3.4, an important model application – that of evaluating air pollution reduction strategies – is demonstrated, by allowing the comparison of pollutant concentrations before and after the implementation of a reduction scenario, such as a combustion fuel ban. These reductions can be benchmarked in relation to the recognised international air quality guideline values. In Section 3.5, it is shown how the model can be used to estimate PM decay due to inter-zonal transfer; providing complementary data for experimental studies in multizone buildings, where pollutant concentrations are derived from a single monitoring site. In Section 3.6, the full application of the probabilistic model is demonstrated by a presentation of a 24-hour activity pattern of emission

source usage in different rooms of a dwelling. Whilst the sample profile was based upon the behaviour of an Irish population group, it serves only as a demonstration case for a model that has wide geographical applicability. As stated earlier, sample data for PM outdoor concentrations and airflow rates were used for model parameterisation in this study, so that demonstration simulations could be carried out. It is nonetheless useful to compare the calculated PM concentrations with literature values associated with comparable dwellings and emission scenarios, to provide an indication of the representativeness of the current simulations. Semple et al. (2012) reported PM2.5 concentrations from sources such as solid fuel fires, gas cooking and environmental tobacco smoke in 100 Irish and Scottish living rooms, which had a mean volume of 57 m3. 24-hour mean concentrations for wood burning were found to be 7.7 μg m− 3 (ranging from 2 to 23 μg m− 3) and for peat burning to be 15.6 μg m− 3 (ranging from 2 to 24 μg m− 3). These are lower than the concentrations reported in the current work, and there may be a number of reasons for this, including a larger mean volume and a lower outdoor PM concentration in the work of Semple et al. (2012) (a mean outdoor concentration of 8.2 μg m−3 was reported). Additionally, the solid fuel fire's emission rate could vary over time, the study by Guo et al. (2008) only reported mean emission rates. Coggins et al. (2013) observed 13 PM2.5 smoking peaks in a sample household; 11 of the 13 peaks fell in the PM concentration range of 100–250 μg m−3, with seven in the range of 100–150 μg m−3, similar to the concentration of 164 ± 44 μg m−3 estimated for the “door open” scenario in the current work (Table 4).

6. Conclusion Comparison with the studies reported in the literature demonstrated that IAPPEM, an advanced probabilistic model, is capable of simulating a realistic representation of the residential environment, with a one minute time resolution and numerous indoor emission sources. However, a validation against monitoring data is needed to confirm the above conclusion. This model has afforded an analysis of the influence of internal layout, emission source location, and internal door configuration on indoor PM concentrations, and would allow an investigation of how such parameters may be manipulated to optimise PM decay and dilution and hence minimise inhalation exposure.

806

J.A. McGrath et al. / Science of the Total Environment 490 (2014) 798–806

One application of IAPPEM is it can be used to determine the minimum feasible external airflow rates in dwellings, which when minimised to reduce heat loss and increase energy efficiency, does not negatively impact on human health. One potential strategy could examine increasing external airflow in rooms that experience higher air pollutant concentrations and decreasing external airflow in rooms that experience lower air pollutant concentrations. This paper has only demonstrated the model's adaptations in relation to the home microenvironment. However, IAPPAM includes these adaptations in the simulation of a full range of additional microenvironments that include offices, transport and recreational areas, with the aim of simulating an individual's exposure over a 24 hour period as they travel throughout a range of microenvironments. Conflict of interest I certify that there is no conflict of interest regarding the material discussed in the manuscript. Acknowledgements This work was funded by Irish Environmental Protection Agency (EPA) under the STRIVE Programme (2008-EH-MS-4). References Brauer M, Hirtle R, Lang B, Ott W. Assessment of indoor fine aerosol contributions from environmental tobacco smoke and cooking with a portable nephelometer. J Expo Anal Environ Epidemiol 2000;10(2):136–44. Brook RD, Rajagopalan S, Pope CA, Brook JR, Bhatnagar A, Diez-Roux AV, et al. Particulate matter air pollution and cardiovascular disease an update to the scientific statement from the american heart association. Circulation 2010;121(21):2331–78. Burke JM, Zufall MJ, Ozkaynak H. A population exposure model for particulate matter: case study results for PM2.5 in Philadelphia, PA. J Expo Anal Environ Epidemiol 2001;11(6):470–89. Byrne MA, Goddard AJH, Lange C, Roed J. Stable tracer aerosol deposition measurements in a test chamber. J Aerosol Sci Jun. 1995;26(4):645–53. Central Statistics Office. Census 2011 profile 5 households and families – living arrangements in Ireland – CSO — Central Statistics Office. February, Last Accessed: 09 March 2013 URL http://www.cso.ie/en/census/census2011reports/ census2011profile5householdsandfamilies-livingarrangementsinireland/, 2013. Cesaroni G, Forastiere F, Stafoggia M, Andersen ZJ, Badaloni C, Beelen R, et al. Long term exposure to ambient air pollution and incidence of acute coronary events: prospective cohort study and meta-analysis in 11 European cohorts from the escape project. BMJ Br. Med. J. 2014;348. Chen C, Zhao B. Review of relationship between indoor and outdoor particles: I/o ratio, infiltration factor and penetration factor. Atmos Environ 2011;45(2):275–88. Coggins MA, Semple S, Hurley F, Shafrir A, Galea KS, Cowie H, et al. Indoor Air Pollution and Health (IAPAH). EPA; 2013 [Last Accessed: 2013-06-16. URL http://www.epa. ie/pubs/reports/research/health/Indoor%20Air%20Pollution%20and%20Health.pdf]. COMEAP. Cardiovascular Disease and Air Pollution. Tech. rep., London. London: Department of Health Committee on the Medical Effects of Air Pollutants; 2006. [last accessed: 24 February 2014. URL http://www.comeap.org.uk/documents/reports]. COMEAP. Long-term Exposure to Air Pollution: Effect on Mortality. Tech. rep. Committee on the Medical Effects of Air Pollutants 978-0-85951-640-2; 2009. [last accessed: 24 February 2014. URL http://www.comeap.org.uk/documents/reports]. COMEAP. Review of Evidence on Health Aspects of Air Pollution Ð REVIHAAP Project: Final technical report. Tech. rep., London. Department of Health Committee on the Medical Effects of Air Pollutants; 2010. [last accessed: 24 February 2014. URL http://www.comeap.org.uk/documents/reports]. Crump D, Dimitroulopoulou S, Squire R, Ross D, Pierce B, White M, et al. Ventilation and indoor air quality in new homes. Pollut Atmosph 2005;71–76. [SPECIAL ISSUE]. Daft. Daft.ie — property for sale and houses for sale or rent in Ireland. February, Last Accessed: 12 May 2013. URL www.daft.ie, 2013. Dimitroulopoulou C, Ashmore MR, Byrne MA. Modelling the contribution of passive smoking to exposure to PM10 in UK homes. Indoor Built Environ 2001;10(3–4):209–13. Dimitroulopoulou C, Ashmore MR, Hill MTR, Byrne MA, Kinnersley R. INDAIR: A probabilistic model of indoor air pollution in UK homes. Atmos Environ Oct. 2006;40(33): 6362–79.

Ebelt ST, Wilson WE, Brauer M. Exposure to ambient and nonambient components of particulate matter — a comparison of health effects. Epidemiology May 2005;16(3): 396–405. EPA. EPA Ireland archive of PM2.5 monitoring data. February, Last Accessed: 01 June 2013. URL http://erc.epa.ie/safer/resource?id=0dc73e08-7e15-102b-aa08-55a7497570d3, 2013. Fabian P, Adamkiewicz G, Levy JI. Simulating indoor concentrations of NO2 and PM2.5 in multifamily housing for use in health-based intervention modeling. Indoor Air 2012;22(1):12–23. Ferro AR, Kopperud RJ, Hildemann LM. Source strengths for indoor human activities that resuspend particulate matter. Environ Sci Technol Mar. 2004;38(6):1759–64. Ferro AR, Klepeis NE, Ott WR, Nazaroff WW, Hildemann LM, Switzer P. Effect of interior door position on room-to-room differences in residential pollutant concentrations after short-term releases. Atmos Environ Jan. 2009;43(3):706–14. Fine PM, Cass GR, Simoneit BRT. Characterization of fine particle emissions from burning church candles. Environ Sci Technol Jul. 1999;33(14):2352–62. Guo L, Lewis JO, McLaughlin JP. Emissions from Irish domestic fireplaces and their impact on indoor air quality when used as supplementary heating source. Glob Nest J Jul. 2008;10(2):209–16. He CR, Morawska LD, Hitchins J, Gilbert D. Contribution from indoor sources to particle number and mass concentrations in residential houses. Atmos Environ Jul. 2004; 38(21):3405–15. Jetter JJ, Guo ZS, McBrian JA, Flynn MR. Characterization of emissions from burning incense. Sci Total Environ Aug. 2002;295(1–3):51–67. Jiang R-T, Acevedo-Bolton V, Cheng K-C, Klepeis NE, Ott WR, Hildemann LM. Determination of response of real-time SidePak AM510 monitor to secondhand smoke, other common indoor aerosols, and outdoor aerosol. J Environ Monit 2011;13(6): 1695–702. Klepeis NE, Nelson WC, Ott WR, Robinson JP, Tsang AM, Switzer P, et al. The National Human Activity Pattern Survey (NHAPS): a resource for assessing exposure to environmental pollutants. J Expo Anal Environ Epidemiol 2001;11(3):231–52. Lee SC, Wang B. Characteristics of emissions of air pollutants from burning of incense in a large environmental chamber. Atmos Environ Mar. 2004;38(7):941–51. Lee SC, Li WM, Chan LY. Indoor air quality at restaurants with different styles of cooking in metropolitan Hong Kong. Sci Total Environ Nov. 2001;279(1–3):181–93. McGrath J, Byrne M, Ashmore M, Terry A, Dimitroulopoulou C. A simulation study of the changes in PM2.5 concentrations due to interzonal airflow variations caused by internal door opening patterns. Atmos Environ 2014;87:183–8. Miller SL, Nazaroff WW. Environmental tobacco smoke particles in multizone indoor environments. Atmos Environ 2001;35(12):2053–67. NIST. Indoor air quality group. February, Last Accessed: 22 September 2013. URL http:// nist.gov/el/building_environment/airquality/index.cfm, 2010. NIST. Publications. February, Last Accessed: 22 September 2013. URL http://nist.gov/el/ building_environment/airquality/index.cfm, 2011. Ott WR. Mathematical models for predicting indoor air quality from smoking activity. Environ Health Perspect May 1999;107(2):375–81. Ott WR, Siegmann HC. Using multiple continuous fine particle monitors to characterize tobacco, incense, candle, cooking, wood burning, and vehicular sources in indoor, outdoor, and in-transit settings. Atmos Environ Feb. 2006;40(5): 821–43. Ott WR, Klepeis NE, Switzer P. Analytical solutions to compartmental indoor air quality models with application to environmental tobacco smoke concentrations measured in a house. J Air Waste Manage Assoc Aug. 2003;53(8):918–36. Ozkaynak H, Xue J, Spengler J, Wallace L, Pellizzari E, Jenkins P. Personal exposure to airborne particles and metals: results from the particle team study in Riverside, California. J Expo Anal Environ Epidemiol 1996;6(1):57–78. Pagels J, Wierzbicka A, Nilsson E, Isaxon C, Dahl A, Gudmundsson A, et al. Chemical composition and mass emission factors of candle smoke particles. J Aerosol Sci 2009; 40(3):193–208. See SW, Balasubramanian R. Characterization of fine particle emissions from incense burning. Build Environ May 2011;46(5):1074–80. Semple S, Garden C, Coggins M, Galea KS, Whelan P, Cowie H, et al. Contribution of solid fuel, gas combustion, or tobacco smoke to indoor air pollutant concentrations in Irish and Scottish homes. Indoor Air 2012;22(3):212–23. Singer BC, Hodgson AT, Guevarra KS, Hawley EL, Nazaroff WW. Gas-phase organics in environmental tobacco smoke. 1. Effects of smoking rate, ventilation, and furnishing level on emission factors. Environ Sci Technol Mar. 2002;36(5): 846–53. WHO. Air Quality Guidelines for Particulate Matter, Ozone, Nitrogen Dioxide and Sulfur Dioxide. Global update 2005. Tech. rep. World Health Organization; 2005. WHO. Review of Evidence on Health Aspects of Air Pollution Ð REVIHAAP Project: Final Technical Report. Tech. rep. World Health Organization; 2013. [last accessed: August 2013. URL (http://www.euro.who.int/en/what-we-do/health-topics/environmentand-health/air-quality/publications/2013/review-of-evidence-on-health-aspects-ofair-pollution-revihaap-project-final-technical-report]. Zai S, Zhen H, Jia-song W. Studies on the size distribution, number and mass emission factors of candle particles characterized by modes of burning. J Aerosol Sci 2006;37(11): 1484–96.