Introduction

The US Environmental Protection Agency (EPA) employs a variety of computer-based models to estimate population exposure to air pollution (Ott et al., 1988; US EPA, 1991; Johnson, 1995). These models typically estimate exposures by simulating the movement of specific population groups through defined microenvironments. The accuracy of the resulting exposure estimates is highly dependent on the validity of the algorithms used to estimate pollutant concentrations in each microenvironment. Many screening-level exposure models assume that pollutant concentrations within enclosed microenvironments can be estimated as a linear function of the outdoor concentration, for example,

in which the additive term (A) accounts for the influence of indoor sources and the multiplicative term (B) is the average ratio of indoor to outdoor concentration in the absence of indoor sources. In more sophisticated exposure models, a mass balance model is used to calculate the pollutant concentration within an enclosed microenvironment as a function of outside concentration, air exchange rate, decay rate, and deposition rate, as appropriate. For example, the pNEM series of exposure models developed by EPA's Office of Air Quality Planning and Standards (OAQPS) uses the following mass balance model to estimate pollutant concentrations in enclosed microenvironments:

in which Cin is the indoor concentration (units: mass/volume); FB is the fraction of outdoor concentration intercepted by the enclosure (dimensionless fraction); Fd is the pollutant deposition or decay coefficient (1/time); ν is the air-exchange rate (AER) (1/time); Cout is the outdoor concentration (mass/volume); S is the indoor emission rate (mass/time); cV is the effective indoor volume where c is a dimensionless fraction (volume); m is the mixing factor (dimensionless fraction); q is the flow rate through air-cleaning device (volume/time); and F is the efficiency of the air-cleaning device (dimensionless fraction). Under steady-state conditions, Eq. (2) can be expressed as a linear expression equivalent to Eq. (1).

In residences, AER is likely to be significantly affected by wind speed and direction, indoor–outdoor temperature gradients, open windows and doors, and the use of heaters, air conditioners, and fans. With the exception of wind speed and direction, each of these factors can be varied in a test house according to a predetermined script as pollutant concentrations and AERs are measured. A pilot study was conducted in Columbus, OH, from January 30 through February 1, 2002, to determine whether this approach could be used to identify the key factors affecting AER and indoor concentrations of six traffic-related pollutants (benzene, formaldehyde (HCHO), polycyclic aromatic hydrocarbons (PAH), carbon monoxide (CO), nitric oxide (NO), and nitrogen oxides (NOx). This article describes the procedures used in this study, presents results of a statistical analysis of the data, and provides recommendations for follow-up studies.

Methods

To identify a representative test home, researchers surveyed traffic patterns in the Columbus, OH, metropolitan area, and contacted the occupants of desirable homes in areas of known high traffic volume. A suitable test home was located on Dublin-Granville Road east of the intersection of US 71 and Ohio State Route 161. Researchers prepared an accurate floor plan of the home, carefully noting the locations of internal and external doors, windows, ceiling fans, furnace, and air conditioner. This information was used to develop a script that specified hourly changes to the window and door positions (open or closed), the furnace and air-conditioning settings (on or off), and the ceiling fan speed (on or off). The script was implemented over the 3-day test period during which technicians collected hourly-average data for AER, indoor, and outdoor air concentrations for the six pollutants, and selected meteorological variables. AER was determined by releasing sulfur hexafluoride (SF6) within the home and measuring the rate of decrease in concentration as a function of time.

Location and Characteristics of the Test Residence

Figure 1 is a photograph of the selected residence. The home was located on a frontage street about 200 feet northeast of the intersection of Ambleside Drive and Dublin-Granville Road, which is a divided six-lane road (Figure 2). Actual 5-min traffic counts at the intersection of Ambleside Drive and Dublin-Granville Road were taken 23 times during the 3-day test. These counts, representing traffic conditions between 0500 and 0100, yielded an average traffic volume of 2076 vehicles/h.

Figure 1
figure 1

View of selected home, looking north from Dublin-Granville Road.

Full size image
Figure 2
figure 2

Location of test home (shaded section within the circle) relative to intersection. North is at the top of the figure.

Full size image

The home was occupied by one female resident who chose to stay in the home during the performance of this test. This resident rarely left the home, and was visited only by a relative who stayed in a basement bedroom for one night during the test period.

The house (built in 1962) was a wood-frame, split-level structure with aluminum siding on the main floor and brick on the basement. The upper floor and basement floor each had an area of 740 square feet (165 square meters). (The basement floor was sealed off during the study; all specifications provided hereafter pertain to the main floor only.) A fully vented, gas-fired furnace heater was located in the northeast corner of the basement, and an electric air conditioner was installed in a permanently closed rear window in the dining room (designated DR in Figure 3). The return air vent for the heater was located in the wall of the dining room nearest the entrance to the living room (LR). Three rooms (sampling room (SR), kitchen (K), and dining room) had ceiling fans. All windows were single glazed. The upper (main) floor had two exterior doors at the front and rear of the house (OD1 and OD2, respectively). The sampling room and rear bedroom could be sealed off from the rest of the main floor by doors (ID1 and ID2, respectively).

Figure 3
figure 3

Plan for the upper floor of test home.

Full size image

The home was chosen in part because of the absence of indoor sources of the target pollutants. As noted above, the home had a fully vented gas furnace, and no other heat or combustion sources were present. The kitchen range and oven were electric, and were minimally used during the test period. Neither the resident nor the visiting relative were smokers, and there was no evidence of smoking being permitted in the home. Small combustion sources (e.g., candles, incense, matches, oil lamps) were similarly absent. The home had no new construction and was fully, but not newly, furnished. Potential indoor sources for the target pollutants appeared to be limited to the common building materials, textiles, and furnishings typical of residential environments (e.g., Otson and Fellin, 1992; Kelly et al., 1999).

The garage was attached to the basement of the home by a closed doorway that the resident indicated was never opened. The resident parked her own car in the driveway outside, and not in the garage. The garage did contain a nonfunctioning automobile that the resident indicated contained no fuel. Inspection of the garage showed no apparent indications of stored fuel or other potential sources of benzene. Both the interior door into the garage and the exterior garage door remained closed at all times during the test.

Script Development

Researchers developed a set of 18 scenarios in which each scenario defined a unique set of air-flow conditions for the house. Table 1 lists the 18 scenarios and identifies each with a brief title and number. (Note that the scenarios are numbered 31–48 only to distinguish the final set of scenarios from earlier sets under evaluation.) Each scenario specifies the status of 14 elements that could potentially affect air flow: two exterior doors (open/closed), two interior doors (open/closed), seven windows (open/closed), the heater (off/on with temperature setting), the air conditioner (off/on with temperature setting), and the ceiling fans (off/on). The column labeled “Heating” in Table 1 indicates furnace operation (on or off); the thermostat temperature setting is listed in parenthesis. The windows of the home were labeled W1–W7, starting with the window in the southwest corner of the sampling room and moving clockwise around the home. The floor plan of the main floor of the home (Figure 3) indicates the positions of these features and their respective designations.

Table 1 Scenarios represented in test script.
Full size table

The 18 scenarios were used to construct the script presented in Table 2. The script is a sequence of 67 1-h periods in which each period is assigned one of the scenarios listed in Table 1. The script was implemented over the 67-h period from 5 am, January 30, 2002 to 12 midnight, February 1, 2002.

Table 2 Test script order during field test (bold=likely periods of heavy traffic, underlining=periods of canister sampling).
Full size table

Sampling System

Indoor and outdoor air samples were drawn through a 1/4 in stainless-steel tube (for PAH) and 1/4 in Teflon sampling lines (for NO, NOx, CO, and HCHO) connected to continuous monitors located in the basement below the sampling room. The indoor air samples were collected at a height of 4 feet in the center of the sampling room in the southwest corner of the upper floor. The outdoor samples were collected at a height of approximately 3.5 feet in front of the home, centered on window W1. Technicians placed the continuous monitoring equipment in the basement, so that (1) heat from the equipment would not affect the test and (2) the noise generated by the equipment would be less intrusive to the occupant. The outdoor and indoor measurements were performed using continuous monitors with the sampled air switched between indoors and outdoors at 5-min intervals using an electric solenoid valve and an automatic 5-min timer.

Indoor and outdoor air samples for benzene analysis were collected by simultaneously opening evacuated 6-L Summa canisters at the same locations as the indoor and outdoor sampling line inlets. Each canister was fitted with a restricting orifice that permitted an integrated air sample to be taken over the 1-h measurement period.

Chemical Measurements

Concentrations of CO, HCHO, PAH, NOx, and NO were measured on a continuous basis for the entire test, while 1-hour integrated benzene samples were collected during selected periods. SF6 concentration was measured at 5-min intervals for determinations of AER. In addition, meteorology data were recorded using a Met One portable meteorological station capable of measuring and storing ambient temperature, relative humidity, wind direction, and wind speed. This station was set up in the backyard of the test home. Indoor air temperature was monitored in the sampling room using a portable thermocouple sensor.

A ThermoEnvironmental 48C ambient CO monitor measured CO over a 10 ppm range, with a detection limit of about 0.1 ppm. Particle-bound PAHs were measured using an EcoChem PAS 2000 continuous photoionization monitor with a 90% response time of 10 s. The standard conversion factor was used to give PAH output in ng/m3. A ThermoEnvironmental 42S chemiluminescent monitor was used to measure NOx and NO over a 200 ppb range. This monitor has a detection limit of 0.1 ppb and a response time of 10 s. HCHO was measured using a fluorescence monitor developed by Battelle (Kelly and Fortune, 1994; Pedditzi et al., 1999) that has a detection limit of 0.1 ppb and a response time of 90 s.

The concentration of SF6 was monitored using a Shimadzu GC-Mini2 gas chromatograph (GC) with a Ni-63 electron capture detector. SF6 was injected into the home using a 50 cc syringe at intervals of several hours, and the GC measured the concentration of the SF6 at the specified 5-minute cycle. The slope of a plot of ln(SF6 concentration) vs. time was used to determine the AER for each of the 1-hour scripted conditions.

Indoor and outdoor hourly samples for benzene were collected simultaneously from 0500 to 1000 and from 1600 to 0100 on each test day (to midnight on February 1), producing a total of 82 canister samples (41 indoor, 41 outdoor) over the 3-day test. The canister samples were analyzed according to EPA Method TO-14 at Battelle's laboratory. Pressure checks of each canister after sampling (but before analysis) revealed that one of the sampling orifices became plugged during the test, allowing very little sample to be collected. In all, 13 canisters were found to be less than half full after sampling (i.e. vacuum readings greater than 16 in Hg). The data from those 13 canisters were excluded from the statistical analysis, leaving 69 total canister samples and 28 hourly periods with simultaneous indoor and outdoor benzene concentrations.

Data Collection

A Campbell Scientific CR23X Micrologger attached to a PC was used to collect data from the ambient continuous monitors and the indoor thermocouple. Data were collected at 1-min intervals and later averaged over the 1-h scripted periods. Meteorological data were recorded every 5 min using the Met One data-storage system and later downloaded into spreadsheet format for comparable averaging. The 5-min wind direction data were used to calculate a 1-h wind direction standard deviation to provide an indication of wind direction stability over the hour.

Results

Descriptive Statistics

A database was compiled listing all valid hourly-average data for the 67 test periods comprising the study. Nine PAH concentrations and one HCHO concentration were below the detection limits estimated for these pollutants (0.1 ng/m3 and 0.1 ppb, respectively). These concentrations were represented in the database by values equal to 50 percent of the specified detection limit. A negative AER value reported for the hour starting at 1100 on January 30 was omitted from the database. Table 3 lists the hourly values included in the database for each meteorological parameter and for AER. Table 4 lists the hourly-average indoor and outdoor pollutant concentrations included in the database. Table 5 presents descriptive statistics for the hourly values included in Tables 3 and 4.

Table 3 Hourly summary of meteorological data and measured air-exchange rates.
Full size table
Table 4 Hourly concentration summary of indoor and outdoor pollutants.
Full size table
Table 5 Descriptive statistics based on 1-h data listed in Tables 3 and 4.
Full size table

Meteorological Conditions

Hourly average wind speed varied between 1.1 and 9.1 mph during the study, averaging 3.5 mph. Wind direction was generally south-westerly from the direction of Dublin-Granville Road during most of the test period (January 30–February 1, 2002). However, weather conditions ranged widely during the study with easterly winds occurring frequently during the first day of sampling. Periods of heavy rain and strong, gusty winds occurred during a few hours of the test. Outdoor temperatures ranged between 28.9 and 63.6°F during the test, averaging 44.6°F. Indoor temperatures averaged 67.0°F with 50 percent of the values falling between 66.0 and 69.9°F.

The far-right column in Table 3 lists the AER determined for each hour of the test. The AERs for all test scenarios ranged between 0.36 and 15.8 h−1. AERs for the most common test scenario (Scenario 34 in Table 1) ranged from 0.36 to 1.51, with an arithmetic mean of 0.74, an arithmetic standard deviation of 0.24 h−1, and a median of 0.72 h−1.

Chemical Measurements

Table 4 lists the hourly-average values of indoor and outdoor pollutant concentrations for each hour for each 1-h test period. As canister samples were not collected between the hours of 1000 and 1600 and between 0100 and 0500, no benzene concentration data are reported for these intervals. In addition, 13 of the benzene samples were found to be unusable because of a problem with one of the orifices used to provide a 1-h integrated sample; these samples were excluded from Table 4 and from the statistical analysis. The remaining 69 benzene measurements ranged from 0.31 to 1.15 ppb.

Indoor CO concentrations ranged from 115 to 1013 ppb. Outdoor concentrations spanned a comparable range (122–1166 ppb). PAH concentrations varied from 0.05 to 29.3 ng/m3 indoors and from 0.05 to 26.4 ng/m3 outdoors. (Note that PAH values below the detection limit of 0.10 ng/m3 were represented in the database by values of 0.05 ng/m3.) HCHO outdoors averaged 1.17 ppb with a minimum near zero and a maximum of 2.3 ppb. Indoor HCHO ranged up to 13.2 ppb with a mean of 8.23 ppb. The HCHO monitor experienced problems in reagent delivery during the last 5 h of the test, and no measurements were made.

Indoor NOx and NO concentrations varied from 15 to 123 ppb and from 3.6 to 75.6 ppb, respectively. Outdoor NOx and NO concentrations ranged from 20 to 112 ppb and from 3 to 89.4 ppb, respectively.

Variation of AER with Ventilation Conditions

The hourly-average AER measured in the test house varied from 0.36 to 15.8 h−1 over the course of the study. Table 6 presents descriptive statistics for the AERs measured during various ventilation categories, defined by the number and location of open exterior windows and doors (hereafter collectively referred to as apertures).

Table 6 Descriptive statistics for hourly-average air-exchange rates according to conditions of exterior windows and doors.
Full size table

The 39 AER values measured during the study when all exterior doors and windows were closed varied from 0.36 to 2.29 h−1 with a geometric mean (GM) of 0.77 h−1 and a geometric standard deviation (GSD) of 1.435. The 27 AER values measured when at least one exterior door or window was opened varied from 0.50 to 15.8 h−1 with a GM of 1.98 h−1 and a GSD of 1.902. Consistent with expectations, the median AER increases as the number of apertures increases: 0.76 h−1 for no apertures, 1.51 h−1 for one aperture, 2.30 h−1 for two apertures, and 2.75 h−1 for three or more apertures. The median air exchange for all test periods is 0.98 h−1.

Researchers were particularly interested in the effect of opening one or both windows in the sampling room (windows W1 and W2 in Figure 3). The median AER for the 16 test periods during which one or both sample room windows were open is 1.98 h−1, falling midway between the medians for one aperture and two apertures.

Variation of Indoor–Outdoor Relationships by Pollutant and AER

To determine the effect of AER on the relationship between indoor and outdoor pollutant concentrations, researchers performed a series of linear regression analyses in which the dependent variable was hourly-average indoor pollutant concentration and the independent variable was the simultaneous hourly-average outdoor concentration for the same pollutant. Separate sets of analyses were performed using data collected during test periods in which (1) the AER varied from 0.36 to 0.99 h−1, (2) the AER varied from 1.0 to 15.8 h−1, and (3) the AER varied between 0.36 and 15.8 h−1 (i.e., all periods with usable data).

Table 7 presents the results of these analyses. Each entry in the table lists the pollutant, the number of paired indoor and outdoor values analyzed, the intercept and slope of the fitted regression line, the coefficient of determination (R2), and a “P-value” indicating the probability that the R2-value is not different from zero. In cases where the intercept value was found not to be statistically significant, the regression analysis was repeated with the regression line forced through the origin (i.e., intercept=zero). Results for these “zero intercept” regression analyses are presented in italics.

Table 7 Results of linear regression analyses in which indoor pollutant concentration is the dependent variable and outdoor pollutant concentration is the independent variable.
Full size table

The following patterns can be observed in Table 7.

  • The intercept associated with the low AER classification (0.36–0.99 h−1) is larger than the intercept associated with the high classification (1.0–15.8 h−1) for each of the six pollutants.

  • R2-values are relatively large (ranging from 0.5597 to 0.8879) for the regression analyses associated with CO, PAH, NO, and NOx. R2-values for these pollutants tended to be higher under high AER conditions than under low AER conditions.

  • The regression slopes for low AERs are lower than the corresponding slopes for high AERs for CO, PAH, NO, and NOx.

  • Intercepts for CO, PAH, and NOx are not statistically different from zero under the high AER conditions. When the regression line is forced through the origin, the resulting slopes are near unity, ranging from 0.913 (PAH) to 1.005 (CO).

  • The three R2-values associated with benzene are relatively small (ranging from 0.2005 to 0.4449) with two of the three P-values exceeding 0.05. The slopes range from 0.302 (low AER) to 0.392 (high AER). The intercepts range from 0.476 ppb (low AER) to 0.590 ppb (high AER) — values comparable to the mean indoor benzene concentration of 0.774 ppb.

  • Each of the three R2-values for HCHO is less than 0.02. All three P-values exceed 0.32, indicating that the relationship between indoor and outdoor HCHO concentrations is not statistically significant.

The lack of correlation between indoor and outdoor concentrations of formaldehyde suggests that one or more unidentified sources of HCHO were present inside the house. Indoor sources may have contributed also to the relatively low correlations observed between indoor and outdoor concentrations of benzene and to the relatively high intercept values, although the relatively small sample size (low number of usable values) for benzene complicates this interpretation.

Indoor and outdoor concentrations were highly correlated for CO, PAH, NO, and NOx. Consistent with the mass balance model, the indoor–outdoor correlation values were larger for these pollutants under high AER conditions than under low AER conditions.

Stepwise Linear Regression Analyses

Researchers performed a series of stepwise linear regression (SLR) analyses with the goal of identifying variables that could be used to estimate AER and/or indoor pollutant concentration. Table 8 lists and defines 62 variables selected as candidate predictor variables for these analyses. The variables indicate operation of heater, air conditioner, and fans (e.g., AC_ON); temperature settings (HEAT_SET); number and identity of open windows and doors (e.g., APERNUM, W1_OPEN); AER (e.g., AER, INVAER); indoor pollutant concentrations for preceding hour (e.g, BENZ_INL); outdoor pollutant concentrations (e.g., BENZ_OUT); indoor and outdoor temperatures (e.g., TEMP_IN, TEMPDIFF); and wind speed (e.g., WINDSPEED, WINDSP_N4). Additional variables are based on the products and quotients of outdoor pollutant concentration and AER (e.g., BENZ_O1, BENZ_O2). Researchers used a two-stage statistical approach to identify the parameters that exhibited the highest predictive power with respect to a particular dependent variable (e.g., indoor CO concentration) under a specified set of ventilation conditions (e.g., all windows and doors closed).

Table 8 Candidate predictor variables for stepwise linear regression analyses.
Full size table

In the first stage of each analysis, researchers defined an initial pool of candidate predictor parameters to be considered for each combination of dependent variable and ventilation category. The pool consisted of one or more groups of variables. Table 8 lists the variables included in each group. Researchers then performed SLR on small subsets of parameters (10 or less) selected from the pool. The subsets were constructed such that each of the parameters under investigation was included in one or more subsets. Preference was given to parameters with complete or nearly complete data, as SLR can only be performed on cases that provide a value for each parameter under evaluation.

Analysts avoided assigning parameters that were likely to exhibit collinearity (e.g., indoor temperature and heater setting) to the same subset. In addition, analysts checked appropriate diagnostics provided in the SLR outputs (e.g., the variance inflation factor) and restructured subsets that exhibited significant collinearity. In performing the SLR analyses, analysts specified P-to-enter and P-to-exit values of 0.05 for adding and removing variables, consistent with recommendations by Draper and Smith (1981). The parameter(s) selected into the regression equation by each SLR analysis (if any) were compiled into a master list of candidate parameters for the specified combination of dependent variable and ventilation conditions.

In the second stage of each analysis, SLR was performed on the candidate parameters included in the master list for each combination of dependent variable and ventilation category. This SLR analysis produced a regression equation that included one or more of the candidate parameters. As before, the sample sizes available for each analysis varied according to the data completeness of the parameters included in the analysis. Tables 9, 10, 11, 12, 13, 14 and 15 present the results of these SLR analyses.

Table 9 Results of SLR analyses in which air-exchange rate is dependent variable.
Full size table
Table 10 Results of SLR analyses in which indoor benzene concentration is the dependent variable.
Full size table
Table 11 Results of SLR analyses in which indoor CO concentration is the dependent variable.
Full size table
Table 12 Results of SLR analyses in which indoor HCHO concentration is the dependent variable.
Full size table
Table 13 Results of SLR analyses in which indoor PAH concentration is the dependent variable.
Full size table
Table 14 Results of SLR analyses in which indoor NO concentration is the dependent variable.
Full size table
Table 15 Results of SLR analyses in which indoor NOx concentration is the dependent variable.
Full size table

Table 9 presents the results of four SLR analyses in which AER was the dependent variable. The first entry in the table shows the results of an SLR analysis applied to all valid AER values, including an outlier value of 15.8 h−1. The two-stage approach described above was applied to the candidate predictor variables included in Groups A and C. The two parameters ultimately selected into the regression equation (W3_OPEN and TEMP_IN) are listed according to the order in which they were selected. The table lists the regression coefficient of each parameter and indicates the cumulative R2-value. Definitions of the parameters can be found in Table 8. The resulting regression equation can be expressed as

in which e is the residual term.

The sign of each regression coefficient provides a directional indicator of the parameter's effect on AER, with the caveat that the value of regression coefficient may vary according to the other parameters included in the regression equation. The positive regression coefficient for W3_OPEN indicates that AER tends to increase when window no. 3 is open (i.e., W3_OPEN=1). The negative coefficient for TEMP_IN suggests that AER tends to decrease as indoor temperature increases. The regression model based on these two predictor variables explained approximately 67 percent of the variability in the 1-h AER values (R2=0.6724). The sample size (n=66) used in the SLR analysis was one less than the total number of 1-hour test periods (n=67) because of the exclusion of a negative AER discussed previously.

It should be noted that window no. 3 (the bathroom window) was only opened during scenario no. 31. During this scenario, technicians opened all seven windows in the test house. As the largest AER value (15.8 h−1) was measured during these conditions, the regression results discussed above show W3_OPEN to be a relatively strong predictor of AER. Researchers judged the result to be a statistical artifact of the script design, and repeated the analysis using the same pool of variables (Groups A and C) but excluding the outlier AER value (15.8 h−1). In this case, the resulting regression equation indicates that AER tended to increase with the number of apertures (open exterior windows and doors) and decrease when interior door no. 1 was open (i.e., ID1_OPEN=1). The R2-value is smaller (0.5438).

The third entry in Table 9 presents the results for an SLR analysis limited to the 39 AER values measured when all exterior windows and doors were closed (i.e., when APER0=1). The pool of candidate variables (Group A) omits all variables concerned with the position of exterior doors and windows, as these variables do not vary with AER under the stated conditions. The results suggest that AER increased as indoor temperature (TEMP_IN) decreased, as wind speed perpendicular to the eastern side of the house decreased (WIND_E4), and as the difference between indoor and outdoor temperature (TEMPDIFF) increased. These three variables explain approximately 44 percent of the variation in AER observed under the stated conditions.

The final entry in Table 9 provides results for an SLR analysis based on the 26 AER values measured when one or more exterior windows or doors were open, omitting the 15.8 h−1 value. The candidate predictor pool included variables in Groups A and C. The SLR analysis identified only one significant predictor variable (W2_OPEN) — window no. 2 located in the sampling room. The R2-value is 0.3025.

Table 10 presents four sets of SLR results for analyses in which indoor benzene concentration was the dependent variable. The first entry provides results for all ventilation conditions using Groups A–D as the pool of candidate predictor variables. The regression equation indicates that indoor benzene concentration tended to decrease with AER and increase with outdoor benzene concentration (BENZ_OUT). The analysis was repeated with variables related to AER (Group B) excluded from the pool of candidate variables. In this case, indoor benzene was found to increase with outdoor benzene concentration and decrease with the number of apertures (APERNUM). The third entry in the table shows results for an analysis in which all exterior windows and doors were closed (APER0=1). Only one variable (BENZ_OUT) was selected into the regression equation. The fourth entry provides SLR results for ventilation conditions in which one or more exterior doors or windows were open. Indoor benzene tended to increase with increases in BENZ_O2 (defined as outdoor benzene concentration divided by AER) and decreased when AERLT1 equaled 1 (i.e., when the AER was less than 1). The R2-values for the four regression equations listed in Table 10 vary from 0.4329 to 0.5609.

Table 11 presents SLR results for CO. The regression equation for all ventilation conditions contains three variables that indicate indoor CO tended to increase with increasing outdoor CO concentration (CO_OUT), increasing indoor CO concentration during the preceding hour (CO_INL), and increasing wind speed perpendicular to the east side of the house (WINDSP_E4). The R2-value is relatively large (0.8756). The results for the closed-house conditions suggest that indoor CO tended to increase when indoor CO during the preceding hour increased, outdoor CO increased, and the AC was being operated (AC_ON=1). The R2-value is quite high (0.9538), indicating that the regression equation explains almost all of the variation observed in the indoor CO concentration. The final entry in Table 11 presents results for conditions in which one or more doors or windows were open. Indoor CO was found to increase with increasing outdoor CO and increasing wind speed from the east (WINDSP_E). R2 equals 0.8756.

Results for HCHO are presented in Table 12. The regression equation for all ventilation conditions contains five variables and has an R2-value of 0.7941. The results suggest that HCHO increased with increasing indoor HCHO during the preceding hour, conditions in which all exterior windows and doors are closed, increasing outdoor temperature, decreasing wind speed, and increasing indoor temperature. Under closed-house conditions, the regression equation indicates that indoor HCHO increased with increasing outdoor temperature, decreasing wind speed, and increasing indoor HCHO concentration during the preceding hour. The R2-value is slightly lower (0.6977). With one or more doors or windows open, indoor HCHO tended to increase with increasing indoor and outdoor temperature (R2=0.8169).

Table 13 provides SLR results for PAH. The R2-value is relatively large (0.8350) for the “all ventilation conditions” classification in which indoor PAH tended to increase with increasing outdoor PAH concentration (PAH_OUT), increasing AER, and conditions in which the number of open doors and windows equaled two (represented in the data set by APER2=1). When all exterior windows and doors were closed, indoor PAH tended to increase with increasing PAH_OUT and with increasing indoor PAH concentration during the preceding hour (PAH_INL). This regression equation has an R2-value of 0.9011. The SLR results for conditions in which one or more doors or windows were open show indoor PAH increased when outdoor PAH increased and when window no. 2 was open (R2=0.9372).

SLR results for NO are presented in Table 14. The same three variables (NO_INL, NO_OUT, and RELHUM_O) are included in the regression equations obtained for all the ventilation conditions and for the closed-house conditions. In both cases, the results show that indoor NO tended to increase with increasing indoor NO during the preceding hour, increasing outdoor NO, and increasing outdoor relative humidity (RELHUM_O). The R2-value associated with the first analysis (0.9041 is almost identical to that of the second analysis (0.9093)). The SLR results for conditions in which one or more doors or windows were open indicate that indoor NO increased when outdoor NO increased, when indoor NO during the preceding hour increased, and when the number of apertures decreased. The R2-value is very high (0.9477).

The results for all ventilation conditions in Table 15 show that indoor NOx tended to increase with increasing outdoor NOx concentration and with increasing indoor NOx concentration during the preceding hour. The same relationship was found for conditions in which all exterior doors and windows were closed and for conditions in which one or more doors or windows were open. The R2-values for the three analyses vary from 0.7759 to 0.9572.

Correlation of Outdoor Pollutant Concentrations with Outdoor CO Concentration

In designing this study, researchers assumed that nearby roadways would serve as the predominant source of outdoor pollutant levels for benzene, CO, PAH, NO, and NOx — pollutants primarily associated with emissions from motor vehicles. CO was selected as a “tracer” pollutant to indicate the presence of vehicle exhaust. Researchers performed a series of linear regression analyses to determine whether outdoor CO concentration was a good predictor of the outdoor concentrations of the other five pollutants measured during the study. Table 16 presents the results of these analyses. The R2-values in descending order are 0.6958 (NOx), 0.6497 (NO), 0.5341 (PAH), 0.4683 (benzene), and 0.0002 (HCHO). The correlation between outdoor CO and outdoor HCHO was not statistically significant.

Table 16 Results of linear regression analyses in which outdoor concentration for indicated pollutant is the dependent variable and outdoor CO concentration is the independent variable.
Full size table

Discussion and recommendations

Air-Exchange Rate

The 66 hourly-average AERs measured in this study varied from 0.36 to 15.8 h−1 with a GM of 1.129 h−1 and a GSD of 1.974. The 39 AER values measured when all exterior doors and windows were closed varied from 0.36 to 2.29 h−1 with a GM of 0.77 h−1 and a GSD of 1.435. The 27 AER values measured when at least one exterior door or window was opened varied from 0.50 to 15.8 h−1 with a GM of 1.98 h−1 and a GSD of 1.902. Consistent with expectations, the results in Table 6 indicate that AER tended to increase with the number of open exterior windows and doors. Results of the SLR analyses in Table 9 support this finding and suggest that AER was also affected by temperature and wind speed, particularly when exterior windows and doors were closed.

An article by Murray and Burmaster (1995) described their analysis of residential AER data compiled by the Brookhaven National Laboratory (BNL). The BNL data included AERs for 2844 residences in the United States, classified according to four geographic regions and the four seasons. The BNL data for Ohio and seven other states (Colorado, Connecticut, Illinois, New Jersey, southern New York, Pennsylvania, and Washington) are included in Region 2. Season 1 includes December, January, and February. The BNL data for Region 2 — Season 1 (i.e., the data collected under conditions most comparable to those of the Columbus study described here) have a GM of 0.45 h−1 and a GSD of 1.960.

The BNL data were not coded as to the status of windows and doors in each surveyed residence; however, it is likely that nearly all of the BNL data represent closed residences. Consequently, the BNL statistics are best compared to the results in Table 6 for scenarios in which all windows and doors were closed (GM=0.766, GSD=1.435).

Comparable AER data for residences with windows open are relatively scarce. The American Petroleum Institute (API) conducted a study of a typical suburban house in Research Triangle Park (RTP), North Carolina, over a 24-h time period during April 1996 (Johnson et al., 1998). Using an approach very similar to that employed in the present study, researchers altered the ventilation characteristics of the house each hour according to a prepared script and measured the resulting hourly-average AER. The resulting 24 AER values ranged from 0.47 to 2.92 h−1 with a GM of 0.81 h−1 and a GSD of 1.66. The nine AER values measured when one or more windows were open varied from 0.78 to 2.92 h−1 with a GM of 1.34 h−1 and a GSD of 1.55 (Appendix H in Johnson, Mihlan, LaPointe, Fletcher, et al., 2000). As noted above, the 27 AER values measured under comparable “open” conditions during the present study varied from 0.50 to 15.8 h−1 with a GM of 1.98 h−1 and a GSD of 1.902.

Johnson et al. (1998) performed a series of SLR analyses to identify factors that significantly affected AER in the RTP test house. The authors concluded that AER increased significantly when either the window or door of the sampling room was open. AER also increased with increasing wind speed from the north and east. None of the tested temperature variables was found to be a significant predictor of AER in the RTP house — a result that differs from the findings of the present study.

In an earlier study reported by Wallace and Ott (1996), researchers measured AERs in a detached house in Redwood City, CA. A continuous monitor (Bruel & Kjaer Model 1302) was used to track the decay of SF6 under a variety of conditions over a period of 16 months. The majority of measurements (88 of 101) were made with all external doors and windows closed. Nine measurements were made under maximum air exchange conditions in which all windows and one or two doors were open. The most useful data were obtained from 27 measurements in which one or two windows were opened to varying widths in a single room (the den). The measured AERs varied from 0.35 to 5.6 h−1, with half the rates between 0.59 and 2.75 h−1. Linear regression analyses indicated that the AER increased by about 0.12 h−1 for every inch that the window was opened. The results of this study appear to be generally consistent with those of the Johnson et al. (1998) study cited above. Wallace and Ott (1996) also provide a useful survey of other studies that have measured AERs. In most cases, the available data represent buildings with closed windows or buildings for which the window status is unknown.

Howard-Reed et al. (2002) supplement Wallace and Ott (1996) by comparing findings from the California home to a similarly instrumented East Coast home in Reston, VA. They conclude that maximal AER typically occurs during mild weather conditions with low winds and small indoor/outdoor temperature gradients and not the reverse as is often assumed. This pattern occurs because occupants are more likely to open home windows and doors under such mild conditions. A single such opening can result in an AER higher than those measured even under extreme weather conditions in a completely closed home. Multiple window openings led to measured AER values of 1.7–2.8 h−1 in homes with a AER of about 0.5 h−1 when closed. The authors conclude that window-opening behaviors must be determined for estimating occupant exposures to outdoor pollutants given that open doors and windows substantially affect home AER and indoor pollutant concentrations.

Wallace et al. (2002) present a year-long time-series of AER measurements made in the Reston, VA home discussed above. The AER, which averaged 0.65 h−1 for this home over the year, increased to 2 h−1 for extended periods and to 3 h−1 for short periods when external windows were opened. Use of an attic fan increased AER by amounts up to 1 h−1, although furnace fan use had little effect. Indoor/outdoor temperature gradients up to 30°C could increase AER by an additional 0.6 h−1, but wind speed had relatively little effect.

Indoor Pollutant Concentrations

Results of the linear regression analyses summarized in Table 7 and the SLR analyses summarized in Tables 10, 11, 12, 13, 14 and 15 indicate that outdoor concentration is a strong predictor of indoor concentration for CO, PAH, NO, and NOx. The results in Table 7 for these pollutants show regression slopes near unity and R2-values ranging from 0.8012 to 0.8806 when AER equals or exceeds 1 h−1, suggesting that indoor concentrations closely tracked outdoor concentrations when AER was relatively high.

The relationship between indoor and outdoor concentrations is noticeably weaker for benzene. Table 7 lists relatively small values for the regression slope (0.392) and R2 (0.2005) under conditions where AER equals or exceeds 1. The SLR results in Table 10 show that indoor benzene concentration tended to increase as AER and number of apertures decreased, suggesting that an unknown indoor source of benzene may have existed in the house. Although the door between the garage and basement was sealed during the study, there is a possibility that residual benzene from past gasoline storage entered the living area from the garage.

The median indoor HCHO concentration (7.9 ppb) was about seven times the median outdoor concentration (1.1 ppb), and the correlation between indoor and outdoor concentrations was not statistically significant. However, the SLR results in Table 12 show HCHO_INL (HCHO concentration during the preceding hour) to be a relatively strong predictor of indoor HCHO concentration when all ventilation conditions are considered. Indoor HCHO concentration was also found to increase under conditions in which all external windows and doors were closed (APER0=1) or when indoor or outdoor temperature increased. Overall, the results suggest that the test house contained a significant indoor source of HCHO that tended to overwhelm any contribution from outdoor HCHO levels.

The indoor concentration during the preceding hour was a significant predictor of indoor concentration for CO (according to two of the three SLR analyses summarized in Table 11), for HCHO (two analyses), PAH (exterior windows and doors closed), NO (all three analyses), and NOx (all three analyses). This variable was not found to be significant for any of the benzene SLR analyses summarized in Table 10, possibly due to the relatively small sample sizes (n≤28) available for the four SLR analyses.

In general, the results suggest that (1) outdoor pollutant concentration and (2) indoor pollutant concentration during the preceding hour are the “predictors of choice” for modeling indoor pollutant concentration in the test house under the conditions of this study. Depending on the pollutant and ventilation conditions, one or more of the following variables produced a small, but significant increase in the R2-value of the regression equation: AER, number and location of apertures, wind speed, air conditioning operation, indoor temperature, outdoor temperature, and relative humidity.

A report by Spicer et al. (1993) describes an earlier Columbus, OH study by Battelle in which researchers measured hourly indoor and outdoor pollutant concentrations at a test house (Site 2) for 240 consecutive hours during June and July of 1989. The house was about 30 years old and occupied by a nonsmoking family of five who maintained a log of their activities, including opening and closing windows. Relevant to the present study, the report provides the results of fitting a statistical model to the indoor and outdoor data that accounted for the effect of window position. The model assumed that indoor concentration is a linear function of outdoor concentration with the intercept value and slope of the regression line taking on different values depending on whether or not the windows in the house were open or closed. The fitted model can be expressed as

in which Cindoor is the indoor pollutant concentration; Coutdoor — windows closed is the outdoor pollutant concentration with windows closed; Coutdoor — windows open is the outdoor pollutant concentration with windows open; the parameters αc and βc, are constants for windows closed; and αo and βo are constants for windows open. Table 17 presents parameter estimates for five pollutants, two of which were also measured in the current study.

Table 17 Results of regression analyses performed by Spicer et al. (1993) in which dependent variable was indoor concentration and independent variables were outdoor concentrations with windows open and closed.
Full size table

The results for benzene data indicate that 77 percent of the variation in indoor benzene concentrations could be explained by variations in outdoor benzene concentration and window position (open or closed). The regression coefficient for outdoor benzene concentration was larger for windows open than for windows closed. Linear regression analysis of HCHO data indicated that 55 percent of indoor concentrations could be explained by variations in outdoor HCHO concentration and window position. Consistent with the benzene results, the regression coefficient for outdoor HCHO concentration was larger for windows open than for windows closed. The regression results for both the pollutants indicate the likely presence of indoor sources.

The R2-values obtained in the 1989 study are higher than those obtained in the current study under similar conditions. The indoor/outdoor relationship for benzene is consistent with the results of the current study in that the regression slope is larger when windows are open than when windows are closed.

Many other studies have measured indoor and outdoor concentrations of one or more of the pollutants included in the current study (e.g., Thomas et al., 1993; Brown and Crump, 1995; Baek et al., 1997; Crump et al., 1997; Clayton et al., 1999; Edwards et al., 2001; Ilgen et al., 2001; Lee et al., 2002). EPA's OAQPS is currently surveying these and other studies with the goal of developing distributions for pollutant-specific “penetration factors” that can be used in enhanced versions of the Hazardous Air Pollution Exposure Model (HAPEM). (Penetration factor — defined as the ratio of indoor to outdoor pollutant concentration in the absence of indoor sources — is equivalent to the “B” term in Eq. (1).) A preliminary summary of the pertinent literature can be found in a report prepared for OAQPS by ICF Consulting and TRJ Environmental (2000) in support of the National Air Toxics Assessment (NATA). The majority of surveyed studies employed averaging times of 8, 12, or 24 h when measuring pollutant concentrations associated with residences. Although such studies may provide data for estimating long-term or “steady-state” relationships between indoor and outdoor concentrations, they typically do not lend themselves to analyses of the factors affecting short-term changes in indoor pollutant concentrations.

Evaluation of the Methodology

The study described in this article served as a test of a proposed “scripted house” methodology for characterizing the effects of various ventilation conditions on AER and on the relationships between indoor and outdoor pollutant concentrations. Overall, the results of the study support the use of this approach in future studies. Field personnel experienced no problems in implementing the script developed for this project. Although the data set obtained from the study was relatively small, the variation in conditions was sufficient to permit analysts to identify the important factors affecting AER and indoor pollutant concentration.

An apparent negative AER observed for the 1100–1200 test period on January 30th was determined to be an artifact of closing the sample room door before fully mixing SF6 throughout the house. In addition, 13 of the benzene samples were found to be unusable because of a problem with one of orifices used to provide a 1-h integrated sample. These measurement problems can be avoided in future studies.

Implications for Exposure Modelers

Consistent with expectations, the SLR results presented in Tables 10, 11, 12, 13, 14 and 15 show outdoor pollutant concentration to be a strong predictor of indoor concentration, particularly in cases where no indoor sources are present, and that low AER (closed windows and doors) in such cases reduces indoor pollutant levels substantially. The results also show that the preceding indoor pollutant concentration can be an equally important predictor. Sophisticated exposure models such as pNEM (Johnson et al., 2000) employ mass balance models to estimate hourly-average indoor concentrations. These models explicitly account for the influence of both outdoor pollutant concentration and the preceding indoor concentration on short-term variations in indoor pollutant concentration. They also incorporate empirical patterns of window opening behaviors (varying AER) as a function of season, temperature, and time of day.

Hourly-average exposure models that employ Eq. (1) or a simple indoor/outdoor ratio to estimate indoor concentrations will, by definition, disregard the short-term influence of the preceding indoor concentration. Such models may produce acceptable estimates for long-term indoor concentrations, but are unlikely to correctly model short-term variations in indoor levels.