Evaluation and quality control of personal nephelometers in indoor, outdoor
and personal environments
CHANG-FU WU,aRALPH J. DELFINO,bJOSHUA N. FLORO,bBEHZAD S. SAMIMI,c PENELOPE J.E. QUINTANA,cMICHAEL T. KLEINMAN,dAND L.-J. SALLY LIUa aDepartment of Environmental and Occupational Health Sciences, University of Washington, Seattle, WA 98195, USA bEpidemiology Division, Department of Medicine, University of California at Irvine, Irvine, CA 92697, USA
cDivision of Occupational and Environmental Health, San Diego State University, San Diego, CA 92182, USA dDepartment of Community and Environmental Medicine, University of California at Irvine, Irvine, CA 92697, USA
Personal nephelometers provide useful real-time measurements of airborne particulate matter (PM). Recent studies have applied this tool to assess personal exposures and related health effects. However, a thorough quality control (QC) procedure for data collected from such a device in a large-scale exposure assessment study is lacking. We have evaluated the performance of a personal nephelometer (personal DataRAM or pDR) in the field. We present here a series of post hoc QC procedures for improving thequality of thepDR data. Thecorrelations and theratios between thepDRs and the collocated gravimetric measurements were used as indices of the pDR data quality. The pDR was operated in four modes: passive (no pump), active (with personal sampling pumps), active with a heated inlet, and a humidistat. The pDRs were worn by 21 asthmatic children, placed at their residences indoors and outdoors, as well as at a central site. All fixed-site pDRs were collocated with Harvard Impactors for PM2.5(HI2.5). By examining the differences
between the time-weighted average concentrations calculated from the real-time pDRs’ readings and recorded internally by the pDRs, we identified 9.1% of the pDRs’ measurements suffered from negative drifts. By comparing the pDRs’ daily base level with the HI2.5measurements, we identified 5.7% of
the pDRs’ measurements suffered from positive drifts. High relative humidity (RH) affected outdoor pDR measurements, even when a heater was used. Results from a series of chamber experiments suggest that the heated air stream cooled significantly after leaving the heater and entering the pDR light-scattering chamber. An RH correction equation was applied to the pDR measurements to remove the RH effect. The final R2values between the
fixed-sitepDRs and thecollocated HI2.5measurements ranged between 0.53 and 0.72. We concluded that with a carefully developed QC procedure, personal
nephelometers can provide high-quality data for assessing PM exposures on subjects and at fixed locations. We also recommend that outdoor pDRs be operated in the active mode without a heater and that the RH effect be corrected with an RH correction equation.
Journal of Exposure Analysis and Environmental Epidemiology (2005) 15, 99–110. doi:10.1038/sj.jea.7500351 Published online 24 March 2004
Keywords: fine particulate matter, personal exposure, light scatter, continuous measurement.
Introduction
To monitor short-term variation and identify sources of personal exposure to particulate matter (PM), a real-time instrument is needed. The personal DataRAM (pDR-1000, Thermo Andersen, Smyrna, GA, USA) is one of the commercially available personal nephelometers, and has been used widely in many recent and ongoing exposure and health effect studies (Muraleedharan and Radojevic, 2000; Quintana et al., 2000; Williams et al., 2000; Quintana et al., 2001; Rea et al., 2001; Liu et al., 2002; Wallace et al., 2003). Although previous studies have validated the pDR as a useful
personal PM monitor, several drawbacks remain. The response of the light-scattering devices depends on the size, density, shape, and optical properties of the sampled aerosols. Such devices need to be calibrated against gravimetric devices in the field, rather than relying on the calibration curve developed by the manufacturer with standard test dust. In addition, light-scattering is affected by relative humidity (RH) (McMurry and Stolzenburg, 1989; Thomas and Gebhart, 1994; Brauer, 1995; Day et al., 2000; Quintana et al., 2000; Day and Malm, 2001). Previous papers showed that the RH effect may be corrected by either physical (Sloane, 1984; Sloane and Wolff, 1985; Lowenthal et al., 1995; Sioutas et al., 2000) or statistical models (Richards et al., 1999), or prevented by using a heater (Liu et al., 2002) or a diffusion drier (Sioutas et al., 2000). Another issue pertaining to pDR is the baseline or zero drift, which appears to result from particle contamination of the light source chamber. Although Liu et al. (2002) reported minimal zero drift (o2 mg/m3), Howard-Reed et al. (2000) Received 14 July 2003; accepted 9 January 2004; published online 24
March 2004
1. Address all correspondence to: Chang-Fu Wu, Box 354695, Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, WA 98195, USA. Tel.: þ 1-206-616-9453. Fax: þ 1-206-543-8123. E-mail: cfwu@u.washington.edu
reported a daily baseline drift from nine pDRs worn by elderly subjects for two 24-h periods ranging from 3 to 6 mg/m3. Quintana et al. (2001) reported a baseline drift of approximately 40 mg/m3 for one pDR carried for seven consecutive days.
In this study, the pDR was used to monitor PM exposures among young children and at their residences in Alpine, CA, USA. Although we had tested the pDR extensively (Quintana et al., 2000, 2001) before deployment in the field, we found unexpected outliers and baseline drifts in the final field measurements. In light of these pDR measurement errors, we developed a series of physically justifiable strategies to identify and remedy these problems, and evaluated the pDR performance after each quality control (QC) strategy.
Methods
Field Study Design and Instruments
This study was conducted in fall 1999, consisting of two sampling sessions, and spring 2000, consisting of five sampling sessions. Each sampling session lasted for 14 consecutive days. During each session, PM concentrations (recorded at 1-min intervals) were measured using the pDRs on up to four subjects, inside and outside at their residences, and at a central site. An RH and temperature sensor (HOBOsdata logger, Onset Computer Corp., Pocasset, MA, USA) was attached on top of each pDR, also recording at 1-min intervals. At all fixed-locations, the pDRs were collocated with 24-h integrated Harvard Impactors for PM2.5 (HI2.5) (Air Diagnostics and Engineering, Inc., Naples, ME, USA) using 37 mm Teflon filters (2.0 mm PTFE membrane, Gelman Laboratories, Ann Arbor, MI, USA) operated at a flow rate of 10 l/min. The HI2.5 data were used as the reference for the pDR measurements. Both Babich et al. (2000) and Liu et al. (2003) reported a good agreement between the HI2.5and the federal reference method for PM2.5(R240.97). Although the pDRs were not preceded by a PM2.5 inlet, Quintana et al. (2000) and Liu et al. (2002) reported high correlation between thepDR and HI PM2.5(R2¼ 0.66B0.84).
In all, 21 asthmatic children wore the pDRs in a fanny pack at waist height during waking hours, and placed the instrument on a table near their breathing zone while asleep. Theindoor pDRs werelocated wheretheparticipants spent most of their time, such as the living room, bedroom, or other main activity areas. Thehomeoutdoor pDRs were located under an open rain shield in the yard at least 1 m away from thewalls. Themonitor inlets wereapproximately 0.9 m abovetheground at homesites.
The pDRs were operated in four modes (Quintana et al., 2000): (1) passive (PAS): This is thedefault configuration of thepDR; (2) active and unheated (AUH): A personal sampling pump (Model 224, SKC Inc., Eighty Four, PA, USA) drew 2 l/min of air through the pDR’s sampling
chamber to avoid potential undersampling in stagnant air; (3) active and heated (AH): A he ate r (DR-TCH, The rmo Andersen, Smyrna, GA, USA) was attached to the pDR inlet and it raised the ambient temperature by 301F to avoid RH artefacts. This temperature increment reduced the RH to below 50%, as estimated from a standard psychro-metric chart (MIE, 1994); and (4) AH with a humidistat that switches on and off the electricity based on a pre-set RH level (AHH): This modeis thesameas theAH modeexcept that the heater was set to be activated only when the ambient RH was over 85%. Although several studies (e.g., Thomas and Gebhart, 1994; Sioutas et al., 2000) observed the RH effect starting at 60%, our previous field tests (Quintana et al., 2000) showed that the RH effect on the pDR in the AUH mode was moderate between 60% and 85% and prominent above85%. This, in conjunction with theconcern of losing particle-bound semivolatilecompounds dueto heating the incoming air all the time (Bergin et al., 1997), prompted us to set the humidistat at 85%. For personal and home indoor monitoring, the pDRs were operated in the PAS mode. For homeoutdoor and central sitemonitoring, theAUH and AH modes were used in the two fall sessions, whereas the AHH modewas used in all fivespring sessions. An additional pDR in thePAS modewas located at thecentral siteduring both thefall and spring sessions.
The sampling procedures in the fall sessions were described in detail in Quintana et al. (2000). Briefly, the pDRs and HI2.5 were operated over 24-h periods from early evening to early evening. The pDR was zeroed at the beginning of each runday with the zeroing bag (Z-Pouch) attached to a small HEPA filter as supplied by the manufacturer. Data were downloaded at the end of each runday. All filters were weighed on a microbalance (Model 30, Cahn Instruments, Madison, WI, USA) and corrected for field blanks (one indoor blank and one outdoor blank per runday). A similar sampling procedure was used in the spring sessions except for the zeroing frequency. In considering that the manufacturer’s zeroing set may not provide adequate zero air, the pDR was zeroed only at the beginning of each session with zero air through a HEPA filter connected to a pump (air flow rate: 4 l/min). Subjects were trained to use an electronic diary programmed on Palms personal digital assistants (Palm, Inc., Milpitas, CA, USA) to report their times, locations, and activities (time-activity diary, or TAD) with a 15-min resolution.
The 1-min pDR measurements (pDR1 min) were averaged over 15-min intervals (pDR15 min) to match theTAD information. Thesamplesizes for thepDR15 min on the subjects, at homeindoors, at homeoutdoors, and at the central site were 25,930, 26,420, 26,505, and 9067, respectively. They were also averaged over 24 hours (pDR24 h) with on and off times corresponding to the collocated HI2.5 sampling period. The SAS statistical program (Release 8, SAS Institute, Inc., Cary, NC, USA) was used for data analysis.
Chamber Experiments
From the QC procedures (as described in the next section), it was identified that there might be some operational issues associated with the heaters and humidistats. To test the performance of the heaters, a series of validation experiments was conducted inside a 46 cm (W) 76 cm (L) 51 cm (H) chamber, with a commercially available humidifier (Model # DH-903, Duracraft Corp., Whitinsville, MA, USA) adjust-ing RH between 58% and 95%. Two sampladjust-ing trains were set up. Onesampling train included a pump drawing theair through theheater continuously at 2 l/min, with a NIST traceable thermohygrometer (Model # 03313-66, Cole-Parmer, Vernon Hills, IL, USA) measuring the temperature and RH at theoutlet of theheater. Thesecond sampling train included a pump drawing theair through theheater and the pDR continuously at 2 l/min. The thermohygrometer measured thetemperatureand RH at theoutlet of thepDR. Two pDRs were used in the second sampling train. A leak test indicated that the two pDRs had an air leak of 0.6% and 20.4%, respectively, at a 2 l/min flow rate. During the experiments, a HOBOsdata logger with built-in tempera-ture and RH sensors recorded the temperatempera-ture and RH at the heater’s air inlet. Furthermore, three humidistats were tested in the same chamber at RH between 90% and 95% with the dials set at 85%. Light bulbs were connected to the electrical outlets on thehumidistats to determineif thelinevoltage were activated.
QC Procedures
We developed a series of post hoc procedures to identify pDR artefacts and drifts and strategies to resolve these measure-ment problems.
Identification of Outliers
A simple time-series plot and a univariate analysis of the pDR1 min data helped identify potential outliers, especially unexplainableand unjustifiableextremevalues. In onecase, pDR1 minincreased from 39 to 7390 mg/m3in 1 min and then decreased to 133 mg/m3 thenext minute. As theTAD data showed no potential PM generating activities, such a phenomenon was very likely due to pDR artefacts. In light of these extreme values, we developed two screening criteria to identify them. The first criterion identifies physically impossiblevalues, that is, thepDR1 min remained over 4000 mg/m3 (499th percentile) for more than 2 min. The second criterion identifies unlikely values based on natural exponential decays, that is, the pDR1 min was gre ate r than 2000 mg/m3and at least 25 times (499th percentile) greater than the previous or subsequent pDR1 min. All ide ntifie d outliers were confirmed by checking the TAD data to ensure that no PM-generating events or activities were reported. Sincetheseidentified outliers may still contain partial
information about the actual PM levels, they were sub-stituted with smoothed filter values derived from a low-pass linear filter (Shumway, 1988) based on a 31-min center-weighted moving average, in which the weighting factor for the outliers was 23.7%. Although only less than 0.01% of thepDR1 min outliers were detected, they affected the pDR15 min values significantly (Po0.01 for paired t-test between the pDR15 mincalculated with or without pDR1 min outliers). In the example given earlier, the pDR15 minfor that timeperiod was 572.6 mg/m3when the outlier (i.e., 7390 mg/ m3) was included and was 85.6 and 198.1 mg/m3 when the outlier was deleted or replaced with the smoothed value (i.e., 1687.5 mg/m3), respectively.
Even after smoothing the pDR1 min outliers, potential outliers were still apparent in the pDR15 mintime-series data. We used two similar but more stringent criteria applied to the smoothed 15-min averages to identify extreme pDR15 min values that were not supported by any reported personal or indoor activities: (1) the pDR15 minwas greater than 2000 mg/ m3 and (2) thepDR
15 minwas at le ast 10 time s gre ate r than the previous or subsequent pDR15 min. Both criteria represent values greater than 99th percentile of the distribution. Only seven pDR15 minoutliers (0.03%) were found and removed from thedataset.
Identify Negative Baseline Drift
The pDR comes with a built-in function that reports the time-weighted average concentration (TWACpDR) of PM during the monitor operation period. Such averages can also be calculated manually once the measurements have been downloaded (TWACman). Although TWACpDR should be identical to TWACman, this is only thecasewhen thereis no baseline drift below 0 mg/m3. Such a negative drift may be caused by zeroing errors (e.g., contamination of the zero air) or by changeof theenvironmental conditions. ThepDR’s internal memory that determines the TWACpDR registers negative signals, whereas the memory that stores the time-series measurements only records zero or positive values. For example, if the baseline drift is5 mg/m3when the true PM concentration is 3 mg/m3, the internal memory would register 2 mg/m3 whiletheexternal memory logs 0 mg/m3. As long as thevalueof thenegativedrift is greater than theambient concentration, the 24-h TWACmanwould begreater than the 24-h TWACpDR. In all, 31 observations (3.5%) were identified where the 24-h TWACpDRwas 2 mg/m3 or more lower than the calculated 24-h TWACman value(Table1). Incidentally, these 24-h averages were from pDR1 min time series registering more than 20% (or Z5 h) of zero values. There were 11 (1.2%) additional zero 24-h TWACpDRwith a high frequency of zero pDR1 min (458%) during the rundays (Table 1). Such a high frequency of zero values measured by the pDRs in the field often implies a negative drift problem. However, the frequency of zero pDR1 min should not beused as thesoleevidenceof negativedrifts since
days of low ambient PM levels exist. In our data set, there were four observations (mean HI2.5¼ 3.8 mg/m3) with more than 20% zero pDR1 min during therundays, whilethe
TWACpDRwas neither zero nor lower than the TWACman. Currently, it is not possibleto estimatethemagnitudeof such a negative bias and when it occurred, thus time series with evidence of negative drift were removed from analysis.
Theidentification of thenegativedrift was further evaluated by comparing the pDR24 h with thecollocated HI2.5measurements. The pDR responds most efficiently to particles with aerodynamic diameter of 0.6 mm, with the sensitivity falling off to about 16% at diameters larger than 5 mm and to about 10% at diameters below 0.15 mm. Since the pDR also responds to particles with diameters larger than 2.5 mm, its measurements are expected to be greater than thoseof HI2.5. Furthermore, assuming that aerosol density is the major different characteristic between the ambient and calibration aerosols, the density of the aerosols sampled by thepDRs can beestimated as 2.6*(HI2.5/pDR), where 2.6 g/cm3is thedensity of thecalibration dusts used by the pDR manufacturer (Liu et al., 2002). Previous studies (Ames et al., 2000) estimated a density of 1.4 g/cm3 for organic carbon (OC) and 2.0 g/cm3for elemental carbon (EC), and thedensity of ammonium nitrateand ammonium sulfateis 1.72 and 1.77 g/cm3, respectively (O’Neil, 2001). As PM2.5 in southern California (Downey, CA, USA) consisted of 6% EC, 46% OC, 24% ammonium nitrateand 16% ammonium sulfate (Sioutas et al., 2000), a reasonable estimate for PM2.5 density weighted by composition is 1.5 g/cm3 and possible values could range between 1.4 (100% OC) and 1.8 g/cm3 (100% ammonium sulfate). Consequently, the average ratio of pDR/HI2.5should beapproximately 1.7 or rangebetween 1.4 and 1.9. This estimated rangealso should beapplicableto most cities in theUS sincesulfate, nitrateand organic carbon arethemajor outdoor PM2.5components as measured by the IMPROVE network (EPA, 1999) and the PM Speciation Trend Network (EPA, 2003).
Although light-scattering responses depend on many factors other than particle density, previous studies have shown that the pDR is on average 1.5 times (which falls within our estimated range of 1.4–1.9) higher than the collocated gravimetric measurements across homes and cites (Howard-Reed et al., 2000; Liu et al., 2002; Wallace et al., 2003), or
pDR24 h¼ X96
i¼1
pDR15 min; i 1:5PM2:5 ð1Þ Having a pDR24 hvaluesmaller than theHI2.5indicates that either the density of the measured aerosols was greater than 2.6 g/cm3, which is unlikely, or the pDR underestimated the concentration of theambient PM dueto a negativedrift. In all, 24 (77%) fixed-site pDR observations that meet the previous criteria, that is, (TWACpDR–TWACmanr2) or (TWACpDR¼ 0) arealso 2 mg/m3 lower than the collocated HI2.5measurements (Table 1), providing confirmation of the negative drift identification criteria. In addition, 25 (4.1%) fixed-site pDR24 h value s we re found to be 2 mg/m3 lower
Table 1. Summary tablefor the24-h averageobservations of (A) (TWACpDR– TWACman)o2 mg/m3, and (B) TWACpDR¼ 0 but (TWACpDR– TWACman)42 mg/m3.
Location TWACpDRa (mg/m3) TWACmanb (mg/m3) DIFFc (mg/m3) HI2.5 (mg/m3) Zero Freq(%)d (A) Personal 0.0 3.8 3.8 F 1085 (77) 0.0 3.6 3.6 F 1039 (71) 2.0 6.7 4.7 F 940 (69) 2.0 6.3 4.3 F 862 (60) 16.0 19.5 3.5 F 817 (54) 2.0 6.1 4.1 F 735 (52) 6.0 15.0 9.0 F 727 (54) 24.0 26.5 2.5 F 570 (41) 2.0 6.7 4.7 F 437 (32) 18.0 20.6 2.6 F 427 (30) 8.0 10.9 2.9 F 393 (31) Indoor 0.0 3.6 3.6 8.4 1056 (72) 0.0 2.8 2.8 7.4 889 (62) 1.0 3.6 2.6 5.4 841 (58) 0.0 3.7 3.7 11.1 808 (59) 2.0 5.2 3.2 9.2 803 (56) 0.0 4.0 4.0 10.3 738 (49) 1.0 6.6 5.6 11.5 724 (53) 0.0 6.3 6.3 9.4 703 (49) 1.0 3.8 2.8 4.9 690 (49) 1.0 5.5 4.5 10.0 610 (43) 0.0 5.1 5.1 10.9 588 (41) 2.0 5.7 3.7 12.1 437 (31) 12.0 14.6 2.6 13.0 430 (30) 5.0 7.6 2.6 13.0 296 (21) Outdoor 0.0 4.5 4.5 14.1 1000 (71) 0.0 4.6 4.6 10.3 840 (48) 7.0 18.3 11.3 15.6 775 (55) 1.0 3.7 2.7 1.7 666 (42) 15.0 21.5 6.5 20.1 468 (33) 24.0 31.1 7.1 24.1 415 (35) (B) Indoor 0.0 1.3 1.3 5.6 1040 (75) 0.0 1.8 1.8 7.5 942 (65) Outdoor 0.0 0.2 0.2 4.8 1310 (97) 0.0 0.2 0.2 4.8 1310 (97) 0.0 0.6 0.6 3.5 1000 (67) 0.0 0.6 0.6 3.5 1000 (67) 0.0 1.2 1.2 6.2 875 (58) 0.0 0.3 0.3 11.9 1324 (89) 0.0 1.2 1.2 4.2 1198 (88) 0.0 0.6 0.6 10.1 982 (69) 0.0 1.0 1.0 9.3 680 (63) a
Time-weighted average concentrations reported from the pDR’s built-in function.
bTime-weighted average concentrations calculated from the pDR 1 min. c
TWACpDR–TWACman.
dFrequency (and percentage) of zero pDR
1 minduring onerunday.
than thecollocated HI2.5 values when both measurements werewell abovethelimit of detection of thepDR, that is, 2 mg/m3(Wallace et al., 2003). These 25 observations passed the(TWACpDR–TWACmanr2) criteria probably because thevalueof thenegativedrift was smaller than theambient concentration (HI2.5¼ 6.6–23.5 mg/m3).
Positive Baseline Drift
Positivebaselinedrifts occur when thepDR light-scattering chamber is contaminated. In this case, the pDR over-estimates PM2.5 levels, resulting in an unreasonably low estimated particle density. We identified the potential positive drifts by comparing thepDR’s daily ‘‘baselevel’’ with the collocated 24-h HI2.5measurements. The pDR’s daily base level was calculated based on a 75-min interval, defined as the minimum of 5-point centered moving average of pDR15 min during each runday. The lowest pDR15 minduring any 24-h monitoring period, or thepDR’s daily baselevel should be lower than 1.5 times the collocated 24-h PM2.5valuefor Eq. (1) to be valid. Taking into account the measurement uncertainty and allowing for fluctuations in the density of ambient aerosols, we defined a positive drift when the pDR’s daily baselevel was 42.6*HI2.5. Theobservations that do not pass this criterion will havetheestimated particledensity at an unreasonably low range (o1 g/cm3). An example of positivedrifts is shown in Figure1, wherethepDR24 h is 170.7 mg/m3, the pDR’s daily base level was estimated as 70.6 mg/m3, and the HI2.5 was 19.6 mg/m3. Curre ntly, the absolutevalueof thepositivedrift could not beestimated because collocated real-time gravimetric measurements were not availablein this study. Weidentified a positivedrift for 35 (5.7%) fixed-site observations and removed them from analysis. For these observations, the average ratio of pDR’s daily baselevel to HI2.5was 3.49, as compared to 0.88 for therest of theobservations.
For pe rsonal pDRs, which we re not collocate d with any gravimetric measurements, the positive drift was identified
using the ‘‘collocated’’ home indoor pDRs when the subjects were asleep. Personal pDR measurements from six subjects were found to be consistently higher than the collocated indoor pDRs. Figure 2 shows an example of an elevated personal pDR as compared with the collocated indoor pDR. Themagnitudeof thepositivedrift or thecorrection factor for a personal pDR can be estimated using the average difference between the indoor and personal pDRs. Then the personal pDR’s readings were subtracted by the correction factor to give the calibrated personal pDR measurements. As not all subjects havetheindoor pDRs placed in thebedrooms (e.g., in the living rooms), occasional excursions occurred on one pDR but not on the other (e.g., when subjects stirred at night). Measurements during such excursions should not be included in the determination of the correction factor. Excursions were defined when (1) there is a change in one pDR but not in theother pDR. That is, thedifference between the pDR15 min, t at time t and pDR15 min, t1 or pDR15 min, tþ 1 is greater than 5 mg/m3 for oneof thetwo collocated pDRs, but thedifferenceamong thepDR15 min, t, pDR15 min, t1, pDR15 min, tþ 1 for thesecond pDR stayed within 2 mg/m3; and (2) thereareexcessivefluctuations in personal or indoor pDRs. That is, the difference between pDR15 min, t and pDR15 min, t1 or pDR15 min, tþ 1 is greater than 20 mg/m3 (4the 99th percentile). In Figure 2, the personal pDR15 min between 0300 and 0400 hours (filled symbols) were identified as excursions according to the first criterion. For the six pDRs, 7.4% (range: 3.7–11.9%) of the pDR15 min were identified as excursion data points. The correction factor ranged between 13.5 and 129.6 mg/m3 (mean7SD ¼ 76.2733.0 mg/m3).
RH Effects and Heater Performance
Pearson’s correlation coefficients between the RH and fixed-sitepDR15 min in thePAS (indoor), AUH (outdoor), AH (outdoor) and AHH (outdoor) modes are 0.20, 0.26, 0.32, 0.67 and 0.27, respectively. Controlling for the actual
Figure 1. An exampleof thepDRs that experienced thepositivebaselinedrifts. Thesolid lineis thepDR15 mindata. The pDR daily base le ve l (dashed line) was estimated as the minimum 75-min centered moving average of pDR15 minduring the 24-h period. The dotted line is the HI2.5 measurement.
ambient PM concentrations, significant positive correlations between the pDR/HI2.5 ratio and RH were found for outdoor pDRs in theAUH (Figure3b), AH (Figure3c) and AHH (Figure3d) modes (Po0.05). This indicates that the heating system did not effectively reduce the RH to the expected level. In addition, the mean ratio of the collocated 1-h pDR in thePAS modeand 1-h pDR in theAH modeat thecentral siteat RH 460% was 1.1170.17 (r ¼ 0.89). This suggests that the RH effects persisted for pDRs in the AH mode and provides further evidences that the heater did not function properly. For indoor pDRs, the pDR/HI2.5ratio and RH was not correlated (Figure 3a) because 97% of the timetheindoor RH was below 60% (mean¼ 36.6%, SD¼ 12.8%).
The RH effect could be corrected using physical models when theparticlesizeand composition areknown (Sloane, 1984; Sloaneand Wolff, 1985; Lowenthal et al., 1995; Sioutas et al., 2000). Richards et al. (1999) also constructed a simple regression model to correct for the RH effects without knowing theaerosol sizeand composition:
ln PM2:5 Bsp PM2:5 true
¼ a lnð1 RHÞ b ð2Þ
where PM2.5_Bsp is PM2.5 measured by a light-scattering device; PM2.5_true is the RH corrected (true) PM2.5 concentration; and a and b are empirically determined parameters. They used the measurements from a heated nephelometer as the PM2.5_true and the measurements from an unheated nephelometer as PM2.5_Bsp. Two se ts of parameters were calculated {(a,b)¼ (0.68, 0.35) and (0.31, 0.01)} based on the RH readings reported from an RH sensor located at themonitoring siteand on the nephelometer, respectively. Figures 3e–h show the correlations between the pDR/HI2.5 ratio and RH after correcting for the RH effect using the first set of parameters for pDR15 min measured between 60% and 95% RH. Correction for pDR measurements taken at RH
below 60% is not necessary as such RH level has little effect on the pDRs (Thomas and Gebhart, 1994; Quintana et al., 2000). At high RH (495%) thecorrection curveis too sensitiveto RH such that an accuratecorrection is not possible(Richards et al., 1999). Thus pDR15 min data taken at RH495% were removed. After the RH correction, no correlation was observed between the pDR/ HI2.5 ratio and RH for outdoor pDRs in theAUH (Figure3f) and AHH (Figure3h) modes. Although the correlation remained significant for the AH mode, the R2 decreased from 0.64 (Figure 3c) to 0.25 (Figure 3g). Note that using the second set of parameters by Richards et al. (1999), the corrected pDRs also showed decreased correla-tions between the pDR/HI2.5ratio and RH (R2¼ 0.04, 0.41, 0.04 for outdoor pDRs in theAUH, AH, and AHH modes, respectively).
Results and discussion
Table 2 summarizes the comparison results between the collocated fixed site pDR24 hand HI2.5after each data QC procedure. N indicates the number of remaining samples. We used the coefficient of determination, R2, be twe e n the pDR24 h and HI2.5 and median of the pDR24 h/HI2.5 ratio to determine the adequacy of each QC procedure. The R2 generally increased after each QC procedure. Figure 4 shows the correlations between the pDR24 h and HI2.5after these QC procedures. Both Spearman’s and Pearson’s correlation coefficients were calculated for pDR24 h and HI2.5. The differences between the two correlation coefficients are all smaller than 0.03. The final median pDR/HI2.5ratios for all pDR types were between 1.2 and 1.9 (Table 2). For the indoor pDRs in thePAS mode, theR2 and median pDR/ HI2.5ratio did not change much after the RH correction. This is mainly becausetheindoor RH was usually below 60%. For theAUH mode, theRH correction increased the
Figure 2. An exampleof thepersonal pDR that experienced thepositivebaselinedrifts. ThepDR15 minbetween 0300 and 0400 hours were the identified excursions points (filled symbols) and were not included for calculating the correction factor of the personal pDRs.
R2 from 0.4 to 0.7, and themedian pDR/HI2.5 ratio decreased from 2.3 to 1.6.
For thepDRs in theAH mode, themedian pDR/HI2.5 ratio was 1.2 (Table2, Column V.1), suggesting that the PM2.5 levels were underestimated even after all the QC procedures. This could bedueto theremoval of semivolatile compounds by overheating from the heaters (Bergin et al.,
1997). At RHo60%, collocated pDRs running in the PAS and AH modes at the central site were expected to have comparable readings. However, the mean ratio of the 1-h pDR measurements in the PAS mode to the AH mode was 1.34 (SD¼ 0.45) at RHo60%. Figure5 shows that high ratios generally occurred at high ambient temperature (4801F) and low RH (o15%). Theratio (mean ¼ 1.55,
Figure 3. Scatter plots of the pDR/HI ratio vs. RH before (a–d) and after (e–f) the RH corrections using Eq. (1) with the first set of parameters developed by Richards et al. (1999) (PAS: passive; AUH: active unheated; AH: active with a heated inlet; AHH: active with a heated inlet and a humidistat).
SD¼ 0.65) was significantly higher at ambient temperature 4801F and RHo15% than that (mean ¼ 1.24, SD ¼ 0.26) at the other range of temperature and RH. Removing these points resulted in a reasonable pDR/HI2.5 ratio of 1.5 (Table2, Column V.1). However, themass loss dueto semivolatilecomponents of theaerosol, that is, the‘‘over-heating effect’’, is a function of temperature, RH, and particlecomposition. Thus, using theabovecriterion alone may result in unnecessary removal of good data points, as evidenced by the lower R2 in Table2 (Column V.1, compared to the R2 before correcting for the overheating effect). The correlation between the pDR/HI2.5ratio and RH in Figure3g becameinsignificant (R2¼ 0.02, P ¼ 0.58) after correcting for the overheating effect.
In correcting for the RH effect (Eq. (2)), we applied the ambient RH measured by the HOBOsattache d to the top of thepDR, assuming that theheating system on the pDR was not effective. Our chamber experiments showed an effective reduction of 32% in RH between the inlet and outlet of the heater. However, the RH reduction between the heater’s inlet and the pDR’s outlet is only 10% and 8% for the pDR with 1% and 20% air leakage, respectively. Evidently, air cooled substantially after entering the pDR, effectively raising the RH inside the pDR. This resulted partially in the observed RH effect on pDRs in the AH mode (Figure3c). Theair leakageof theactivepDRs also
accounted for the RH effect. Therefore, the ambient RH is a good surrogatefor theRH insidethepDR. Notethat the actual RH insidethepDR could not bemeasuredF our modeling attempt based on an exponential decay curve from our limited experimental data failed to produce a better R2 than thoseusing theambient RH. Compared to theair cooling phenomenon insidethepDRs, thedegreeof RH effect dueto theair leakagemay beless substantial, judging from the small difference in the pDRs’ RH reduction (8% vs. 10%) between the 1% and 20% of air leakage in our chamber experiments.
For pDRs in theAHH mode, theRH correction increased the R2 between pDR24 h and HI2.5 by 0.1 and decreased the median pDR/HI2.5 ratio from 2.3 to 1.7 (Table2, Column V.1). Although thehumidistat was set to activate the heater at ambient RH485%, our chamber experiments indicated that only one out of three humidistats
activated thelinevoltagewhen RH 4 90%.
Thesehumidistats wereprobably not suitableor designed for applications in high RH conditions. In addition, the R2 and median pDR/HI
2.5 ratio in the AHH mode were moresimilar to thosein theAUH modethan thosein theAH mode(Table2). It is thus likely that most humidistats did not activate the heaters at high RH. Therefore, the ambient RH was a good surrogatefor theRH insidethepDR when applying the RH correction equation (Eq. (2)). Note that
Table 2. Comparisons between the pDR24 hand HI2.5after each data quality control and correction procedure. CumulativeQC proceduresc
pDR modea VariableI II III IV V.1 V.2
PAS (indoor) N 260 244 234 233 226 226 Median of pDR24 h/HI2.5 1.6 1.6 1.7 1.7 1.7 1.7 R2b 0.50 0.50 0.53 0.55 0.54 0.54 AUH (outdoor) N 48 48 45 42 36 36 Median of pDR24 h/HI2.5 2.3 2.3 2.4 2.3 1.6 1.9 R2b 0.35 0.35 0.40 0.41 0.72 0.73 AH (outdoor) N 60 55 49 48 47 (20)d 47 (20)d Median of pDR24 h/HI2.5 1.1 1.1 1.2 1.2 1.2 (1.5) 1.2 (1.5) R2b 0.50 0.48 0.53 0.60 0.70 (0.53) 0.68 (0.48) AHH (outdoor) N 246 236 230 200 174 174 Median of pDR24 h/HI2.5 2.3 2.4 2.4 2.3 1.7 1.8 R2b 0.31 0.33 0.36 0.53 0.62 0.62
aPAS: passive; AUH: active unheated; AH: active with a heated inlet; AHH: active with a heated inlet and a humidistat. bR2values between pDR
24 hand HI2.5. cI: Correction for outliers.
II: I & correction for negative baseline drifts (compared to TWACpDR).
III: II & correction for negative baseline drifts (compared to HI2.5).
IV: III & correction for positive baseline drifts.
V.1: IV & RH corre ction with the 1st se t of parame te rs in Richards et al. (1999). V.2: IV & RH correction with the 2nd set of parameters in Richards et al. (1999).
dNumbers in parentheses are statistics after removing pDR
15 minwith temperature 4801F and RHo15%.
sincetheheaters wereoff most of thetime, pDRs in the AHH mode were not expected to be influenced by the overheating effect.
Although collocated pDRs in the PAS and AHH modes at the central site were expected to have comparable readings at RHo60%, 10% of theobservations had PAS/AHH ratios higher than 2. A weak but significantly positive correlation (Pearson’s correlation coefficient¼ 0.13, P ¼ 0.002, n ¼ 588) was found between the ratio and wind speed, suggesting that thepassivepDR may over sampleunder high wind conditions. Nevertheless, since such a wind effect is infrequent, our reported overheating effect for the pDRs in theAH modeat low RH still holds.
TheactivepDR is a sealed system. Theair leakagefound in our study could beattributed to looseconnectors or damage of the sealant near the optical chamber assembly. Loose connectors can be easily fixed by the field technicians, but problems involving internal components need to be repaired by the manufacturer. It is essential to check the air leak on a regular basis to minimize the RH effects from the air leakage.
Thechoiceof thescreening criteria for outlier identification should depend on the distribution of the pDR1 mindata and the characteristics of the study subjects and homes. In this present study of nonsmokers living in nonsmoking resi-dences, 4000 mg/m3was used because it was above the 99th percentile of the pDR1 min measurements. This screening criterion for outliers was further supported by results of Howard-Reed et al. (2000) and Quintana et al. (2001), who reported that the maximum pDR1 min measurements were approximately 2900 mg/m3. This screening criterion should be shifted upward for study populations or regions with prominent PM sources. Wallace et al. (2003) reported in their Inner City Asthma Study that the 1-h pDR measure-ments were over 4000 mg/m3 in the smoking residences. Balakrishnan et al. (2002) also reported that the personal exposures (average sampling duration¼ 2.7 h) to respirable
Figure 4. Scatter plots of the pDR24 h vs. HI2.5 after the RH corrections using Eq. (1) with the first set of parameters developed by Richards et al. (1999) (PAS: passive; AUH: active unheated; AH: active with a heated inlet; AHH: active with a heated inlet and a humidistat).
Figure 5. Ratio of 1-h pDRs (in thePAS mode) to 1-h pDRs (in the AH mode) as a function of RH at thecentral siteat RHo60%.
PM for cooks using wood fuels during cooking times in rural India could behigher than 4000 mg/m3.
Both sets of parameters reported by Richards et al. (1999) were applied in Eq. (2). The first set reduced the RH effect more successfully than the second set. Although these two sets of parameters were calculated from data collected in Bakersfield, CA, reasonable pDR/HI2.5ratios and good R2 values between pDR24 h and HI2.5after the RH corrections (Table 2) indicated that they were applicable in this study. Previous studies have also reported that the chemical composition of PM2.5 did not affect significantly the light-scattering response of the DataRAM (Sioutas et al., 2000).
Light-scattering responses depend on particle density, refractive index, and particles size (Sioutas et al., 2000). Sincetherefractiveindex is similar across major outdoor PM2.5 species (1.47–1.55) (Sioutas et al., 2000) and the calibration dust (1.5), theaccuracy of theparticledensity estimation equation (i.e., 2.6*HI2.5/pDR) is affected by the particlesizedifferencebetween thecalibration dust (mass media diameter: 2–3 mm) and the measured PM.
In our QC procedures, we identified the baseline drifts before correcting for the RH effects as drifts occurred throughout thesampling period whiletheRH artefacts most likely occurred during the nighttime. Performing RH correction before the baseline drift identification decreased thefinal R2values for pDR24 hand HI2.5from 0.72 to 0.63 (AUH mode) and 0.62 to 0.57 (AHH mode), while the R2 values for the PAS and AH modes were approximately the same.
To identify the negative drifts, we first compared the TWACpDRto TWACman. This approach may not provide sufficient sensitivity to identify all the observations with negativedrifts, especially thosewith small drifts. Theremay be cases when the negative drift produced negative numbers, but there were not enough of them to create a 2 mg/m3 difference during a 24-h period. To examine the extent of this issue, we simulated 520 24-h time series at 1-min resolution with a geometric mean ranging between 5 and 30 mg/m3with a 1 mg/m3 interval, and a geometric standard deviation ranging between 1.1 and 3.0 with a 0.1 interval. The spline-smoothed lines of the TWACman–TWACpDR values and percent of zero pDR1 min of the simulated time series with different drift values were presented in Figures 6a and b, respectively. The percent of zero pDR1 min required to producea 2 mg/m3 difference in TWACman–TWACpDR depends on both the true PM2.5and thedrift values. The simulation results indicated that the (TWACpDR– TWACman42) criterion underestimated the percentage of pDR’s having negative drifts at high ambient PM2.5 levels and at small negative drift values (o8 mg/m3). At truePM2.5 of 10 mg/m3, this criterion identifies observations with negative drift values greater than 8 mg/m3(Figure6a), where thereareapproximately 52% of zero pDR1 min(Figure6b). When the true PM2.5is 20 mg/m3, this criterion only identifies
most observations with negative drift values greater than 12 mg/m3(Figure 6a), where there are approximately 38% of zero pDR1 min(Figure6b).
Thelargepositivedrifts found for thepersonal pDRs were attributed to elevated personal activities and infrequent zeroing. Based on the activity levels recorded on the TAD (i.e. levels 1–5 for activity levels ranging from sleep, rest, light, moderate to strenuous), four of the six subjects carrying pDRs with positivedrifts weremoreactive(mean level¼ 2.0–2.2) than the other 17 subjects (mean level¼ 1.3–1.9). This also explains why thepositivedrifts observed in our study were higher than those reported earlier (Howard-Reed et al., 2000; Liu et al., 2002). Howard-Reed et al. (2000) reported that the highest baseline drift was 6 mg/ m3, while Liu et al. (2002) observed minimal zero drift (o2 mg/m3). This is because they had mostly elderly subjects who wereless activethan thechild subjects and daily zeroing was deployed in both studies. Furthermore, all personal pDRs with positivedrifts wereoperated in thespring sessions, during which the pDRs were zeroed only once every 14 days as compared to the daily zeroing in the fall sessions. This is also evident from the outdoor pDR measurements. In the fall sessions, the percentage of measurements with positive drifts was 6% and 1.7% for the pDRs in the AUH and AH modes, respectively. In the spring sessions, therateof positivedrifts was 12% for the pDRs in the AHH modes. This suggests that daily zeroing may alleviatepositivedrifts. Thefact that theoutdoor pDRs in thefall sessions werestill affected by thepositivedrifts despite the daily zeroing procedures indicated that using the manufacturer-provided bag might not eliminate the drift problems totally. An improved zeroing protocol, such as using clean air with active sampling system, should be considered. If the pDRs are zeroed correctly at the beginning of each sampling day, the occurrence and magnitude of the baseline drift could be estimated by checking the results of the postexposure zeroing, and by comparing the ambient PM levels recorded by the pDRs after the 24-h sampling period to the readings from a newly zeroed pDR. In addition to the daily zeroing procedures, annual or semi-annual cleaning and recalibration of thepDRs by themanufacturecould also help to removethebuild-up (e.g. greasefrom cooking sources) inside the sensing chamber, thus alleviating serious zeroing problems.
Conclusion
The personal nephelometer is a valuable tool for accessing real-time PM exposures. Many operational issues of the pDR ne e d to be ide ntifie d and handle d prope rly be fore data analysis. In this paper, we presented a series of QC procedures to identify pDR measurement artefacts. By comparing theTWACman to TWACpDR and pDR24 h to
HI2.5, negative drifts were identified. By comparing the pDR’s daily baselevel to HI2.5measurements, positive drifts were identified. We successfully removed most RH effects by applying a statistical RH correction equation. The final results were satisfactory. Our R2for HI2.5and pDR24 hare similar to the R2 of 0.66 (personal filter samples vs. pDR) reported by Howard-Reed et al. (2000) but lower than the R2 of 0.77–0.84 found by Liu et al. (2002). Our pDR/HI2.5 ratios aresimilar to theratios of 1.5–1.6 reported in other studies (Howard-Reed et al., 2000; Liu et al., 2002; Wallace et al., 2003). For identifying drifts of pDRs, we suggest to have collocated gravimetric measurements or consider using active pDRs with an integrated cyclone and filter (1200, MIE, Bedford, MA, USA). Theperformanceof pDR-1200 can befound in Lanki et al. (2002). Although we
attempted to minimize the potential zeroing errors in the field by zeroing the pDRs only at the beginning of each session with ‘‘truezero air’’ in thespring, our study results indicate that the benefit of daily zeroing of the pDR outweighs the zeroing errors.
ThepDR in thepassivemodeis adequatefor indoor PM monitoring, given thereasonablepDR/HI2.5 ratio and R2 between pDR24 hand HI2.5. ThepDR in thepassivemodeis adequate for personal PM monitoring with our QC procedures and calibration against a fixed-site pDR or other gravimetric samplers. When applied outdoors, pDRs can be cost-effectively operated in the AUH mode. In the AH mode, our heater had difficulty effectively eliminating sufficient RH at high RH levels and overheated the air under hot and dry conditions. The RH effect for pDRs in the
Figure 6. The spline-smoothed lines of (a) the TWACman–TWACpDRvalues and (b) percent of zero pDR1 minof the simulated time series with different drift values.
AUH mode can be corrected with the RH correction equation. To entirely avoid the RH and overheating effects, an upstream diffusion drier (Sioutas et al., 2000) may be a good alternative, although such a device typically cannot operate for 24 h without replacement of the absorbing material (Quintana et al., 2000; Lanki et al., 2002).
Acknowledgements
This study was funded by the NIH/NIEHS Grant ES-06214 and EPA Northwest Research Center for Particulate Air Pollution and Health Grant R827355. We also thank Dr. Timothy Larson for his insightful comments.
References
Ames R.B., Hand J.L., Kreidenweis S.M., Day D.E., and Malm W.C. Optical measurements of aerosol size distributions in Great Smoky Mountains National Park: dry aerosol characterization. J Air Waste Manage Assoc 2000: 50(5): 665–676.
Babich P., Davey M., Allen G., and Koutrakis P. Method comparisons for particulate nitrate, elemental carbon, and PM2.5 mass in seven US cities. J Air Waste Manage Assoc 2000: 50(7): 1095–1105.
Balakrishnan K., Parikh J., Sankar S., Padmavathi R., Srividya K., Venugopal V., and Prasad S., et al. Daily average exposures to respirable particulate matter from combustion of biomass fuels in rural households of southern India. Environ Health Perspect 2002: 110(11): 1069–1075.
Bergin M.H., Ogren J.A., Schwartz S.E., and McInnes L.M. Evaporation of ammonium nitrate aerosol in a heated nephelometer: Implications for field measurements. Environ Sci Technol 1997: 31(10): 2878–2883.
Brauer M. Assessment of indoor aerosols with an integrating nephelometer. J Expos Anal Environ Epidemiol 1995: 5(1): 45–56.
Day D.E., and Malm W.C. Aerosol light scattering measurements as a function of relative humidity: a comparison between measurements made at three different sites. Atmos Environ 2001: 35(30): 5169–5176.
Day D.E., Malm W.C., and Kreidenweis S.M. Aerosol light scattering measurements as a function of relative humidity. J Air Waste Manage Assoc 2000: 50(5): 710–716.
EPA US. National Air Quality and Emissions Trends Report. Office of Air Quality Planning and Standards,, 1999.
EPA US. Fourth External Review Draft of Air Quality Criteria for Particulate Matter. Office of Research and Development,, 2003.
Howard-Reed C., Rea A.W., Zufall M.J., Burke J.M., Williams R.W., Suggs J.C., and Sheldon L.S., et al. Use of a continuous nephelometer to measure personal exposure to particles during the US Environmental Protection Agency Baltimore and Fresno panel studies. J Air Waste Manage Assoc 2000: 50(7): 1125–1132.
Lanki T., Alm S., Ruuskanen J., Janssen N.A.H., Jantunen M., and Pekkanen J. Photometrically measured continuous personal PM2.5 exposure: levels and
correlation to a gravimetric method. J Expos Anal Environ Epidemiol 2002: 12(3): 172–178.
Liu L.J.S., Box M., Kalman D., Kaufman J., Koenig J., Larson T., and Lumley T., et al. Exposure assessment of particulate matter for susceptible populations in Seattle. Environ Health Perspect 2003: 111(7): 909–918.
Liu L.J.S., Slaughter J.C., and Larson T.V. Comparison of light scattering devices and impactors for particulate measurements in indoor, outdoor, and personal environments. Environ Sci Technol 2002: 36(13): 2977–2986.
Lowenthal D.H., Rodgers C.F., Saxena P., Watson J.G., and Chow J.C. Sensitivity of estimated light extinction coefficients to model assumptions and measurement errors. Atmos Environ 1995: 29(7): 751–766.
McMurry P.H., and Stolzenburg M.R. On the sensitivity of particle-size to relative-humidity for Los-Angeles aerosols. Atmos Environ 1989: 23(2): 497–507.
MIE. Application Note10-B. Monitoring with DataRAM in wet/high humidity environments. MIE, Inc., 1994.
Muraleedharan T.R., and Radojevic M. Personal particle exposure monitoring using nephelometry during haze in Brunei. Atmos Environ 2000: 34(17): 2733–2738.
O’Neil M.J., (ed.). The Merck index: An Encyclopedia of Chemicals, Drugs, and Biologicals. Merck & Co., Inc., Rahway, 2001.
Quintana P.J.E., Samimi B.S., Kleinman M.T., Liu L.J., Soto K., Warner G.Y., and Bufalino C., et al. Evaluation of a real-timepassivepersonal particle monitor in fixed site residential indoor and ambient measurements. J Expos Anal Environ Epidemiol 2000: 10(5): 437–445.
Quintana P.J.E., Valenzia J.R., Delfino R.J., and Liu L.J.S. Monitoring of 1-min personal particulate matter exposures in relation to voice-recorded time-activity data. Environ Res 2001: 87(3): 199–213.
Rea A.W., Zufall M.J., Williams R.W., Sheldon L., and Howard-Reed C. The influence of human activity patterns on personal PM exposure: A comparative analysis of filter-based and continuous particle measurements. J Air Waste Manage Assoc 2001: 51(9): 1271–1279.
Richards L.W., Alcorn S.H., McDadeC., CoutureT., Lowenthal D., Chow J.C., and Watson J.G. Optical properties of the San Joaquin Valley aerosol collected during the 1995 integrated monitoring study. Atmos Environ 1999: 33(29): 4787–4795.
Shumway R.H. Applied Statistical Time Series Analysis. Prentice-Hall, Englewood cliffs, NJ, 1988.
Sioutas C., Kim S., Chang M.C., Terrell L.L., and Gong H. Field evaluation of a modified DataRAM MIE scattering monitor for real-time PM2.5 mass concentration measurements. Atmos Environ 2000: 34(28): 4829–4838. Sloane C.S. Optical-properties of aerosols of mixed composition. Atmos Environ
1984: 18(4): 871–878.
Sloane C.S., and Wolff G.T. Prediction of ambient light-scattering using a physical model responsive to relative-humidityF validation with measurements from Detroit. Atmos Environ 1985: 19(4): 669–680.
Thomas A., and Gebhart J. Correlations between gravimetry and light-scattering photometry for atmospheric aerosols. Atmos Environ 1994: 28(5): 935–938. WallaceL.A., Herman M., O’Connor G.T., Lucas N., Morton L., Meyer K., and
Jane K., et al. Particle concentrations in inner-city homes of children with asthma: Theeffect of smoking, cooking, and outdoor pollution. Environ Health Perspect 2003: 111(9): 1265–1272.
Williams R., Suggs J., Zweidinger R., Evans G., Creason J., Kwok R., and Rodes C., et al. The 1998 Baltimore particulate matter epidemiology-exposure study: Part 1. comparison of ambient, residential outdoor, indoor and apartment particulate matter monitoring. J Expos Anal Environ Epidemiol 2000: 10(6): 518–532.
