A Shelter to Protect a Passive Sampler for Coarse Particulate Matter, PM_{10 − 2.5}

Abstract

This work designed and tested a shelter to protect a passive sampler for measuring coarse particulate matter, PM _{10 − 2.5} . The shelter protects the sampler from precipitation and reduces the effects of wind on the deposition of particles to its collection surface. Six shelters were tested in a wind tunnel at three wind speeds: 2, 8, and 24 km hr ⁻¹ . Shelter performance was expressed as the ratio of PM _{10 − 2.5} measured with the passive samplers to that measured with a filter-based dichotomous sampler. For most shelters, the PM _{10 − 2.5} ratio averaged across wind speeds was well above one (2.4 to 8.5) and was generally dependent on wind speed. However, the PM _{10 − 2.5} ratio for one shelter, the Flat Plates shelter, was 1.04 with substantially less effect on particle deposition from wind speed. Eight week-long field tests were conducted to compare PM _{10 − 2.5} measured with a passive sampler installed in a Flat Plates shelter to that measured with a collocated filter-based dichotomous sampler. In these tests, the mean PM _{10 − 2.5} ratio was 1.29. The linear relationship between PM _{10 − 2.5} measured passively to that measured with the filter-based sampler had a Pearson correlation coefficient of 0.97 and was not significantly affected by the addition of weekly mean wind speed (p = 0.35). Although temperature was significant in this regression model (p = 0.02), it only improved the relationship marginally. The passive sampler in a Flat Plates shelter offers an inexpensive means to assess ambient PM _{10 − 2.5} without on-site measurement of wind speed.

NOMENCLATURE

A	=	projected area of particle
AT	=	total area of the sample imaged
C	=	mass concentration of single particle
Cp	=	particle circularity
D	=	diffusivity of a particle
da	=	aerodynamic diameter
dpa	=	projected-area diameter
Dx	=	horizontal distance
Dy	=	vertical distance
E	=	sampling efficiency
F	=	mass flux of the particle to the deposition surface
g	=	gravitational constant
k	=	von Karman's constant
m	=	mass of single particle
P	=	perimeter of particle
PM10 − 2.5	=	coarse particulate matter
PM10	=	particle matter smaller than 10 μm
SV	=	volumetric shape factor
u*	=	friction velocity
Vambient	=	settling velocity of a particle outside of the mesh cap
Vdep	=	deposition velocity of a particle to the deposition surface inside the mesh cap
Vt	=	settling velocity of a particle in still air = τ g
Vx	=	horizontal wind velocity
Vy	=	settling velocity of particle
z	=	height above ground that wind speed is measured
z o	=	surface roughness height
γmesh	=	mesh factor
ρ p	=	particle density
τ	=	particle relaxation time
ν	=	kinematic viscosity

INTRODUCTION

Adverse respiratory (Chen et al. 2004; Fung et al. 2006; Lin et al. 2002) and cardiovascular health effects (Burnett et al. 1997; Lipsett et al. 2006; Villeneuve et al. 2003) have been associated with exposure to ambient coarse particulate matter, PM_{10 − 2.5}. Typically, these studies rely on data collected with filter-based samplers at sparsely distributed central monitoring stations to represent exposures for people residing throughout large geographical areas. This approach leads to progressively greater exposure misclassification as the magnitude of spatial variability becomes greater (Brunekreef and Forsberg 2005; Hoek et al. 2001; Jerrett et al. 2005; Wilson et al. 2005). Such misclassification is particularly likely for coarse particles, which are often substantially more heterogeneously distributed than fine particles (Hoek et al. 2001; Jerrett et al. 2005; Monn 2001; Wilson et al. 2005; Wilson and Suh 1997).

Exposure assessment of PM_{10 − 2.5} may be improved by increasing the number of monitoring sites. However, this approach is resource intensive because of the costs and manpower associated with installation and operation of filter-based particulate matter samplers. Indirect methods such as geospatial statistics and dispersion modeling may be incorporated with central monitoring to augment knowledge about the variability of contaminants within an air shed (Jerrett et al. 2005; Wilson and Zawar-Reza 2006). These techniques, however, require extensive knowledge of sources, meteorology, and geography and are generally ineffective in estimating short-term concentrations for contaminants that exhibit great spatial variability (Nieuwenhuijsen et al. 2006).

Passive sampling has been used as an alternative to central monitoring to assess the spatial variability of ambient gases. Passive samplers are substantially less costly to construct and deploy than typical central monitoring equipment because they do not require a pump, flow control equipment, or a power source (Krupa and Legge 2000; Nothstein et al. 2000). Briggs et al. (2000) used passive samplers to measure nitrogen dioxide at 80 sites in four metropolitan areas to develop an air quality model. Norris and Larson (1999) used passive nitrogen dioxide samplers to validate measurements made on a mobile monitoring platform. Ballesta et al. (2006) utilized 1,600 passive samplers to measure benzene levels outdoors, indoors, and on people in six cities across Europe.

Wagner and Leith (2001a; 2001b) introduced a passive sampler to measure particles in occupational settings. In their sampler, particles deposit by diffusion, gravity settling, and turbulent inertial forces onto a collection substrate. They introduced a deposition velocity model to translate particle surface loading on the substrate, determined through microscopy, to airborne mass concentration. They found good agreement between the mass concentration by size measured with the passive sampler and that derived from an impactor in occupational settings where air velocities were low. However, ambient use of this sampler requires knowledge of wind speed and site characteristics (surface roughness) to apply the turbulent inertial deposition model presented by Wagner and Leith (2001a). The need for wind speed data substantially reduces the sampler's ease of use and increases operational costs, and surface roughness is difficult to assess for an ambient location. Moreover, this sampler must be protected from precipitation with some shelter, which in turn may affect the deposition characteristics of the sampler.

Thus, the purpose of this study was to develop a shelter to protect the Wagner–Leith passive sampler from precipitation in a manner that reduces the need to account for wind effects when measuring PM_{10 − 2.5}. We first outline the deposition velocity model of Wagner and Leith. Then, we present wind tunnel tests that were conducted to select a shelter design. Last, we present field tests that were carried out to assess the ability of the sheltered passive sampler to measure airborne PM_{10 − 2.5}.

THEORETICAL CONSIDERATIONS: THE WAGNER AND LEITH DEPOSITION VELOCITY MODEL

Wagner and Leith (2001a) presented a deposition velocity model to estimate airborne mass concentration from microscopic images of particles collected on a passive sampler. In this model, the contribution of a single particle (i) to mass concentration, C, is calculated as:

where F is the mass flux of the particle to the deposition surface, V_dep is the deposition velocity of the particle, m_i is the mass of the particle, A_T is the total area of the sample that was imaged (A_T = number of images times the area of one image) and t is the sample time. Deposition velocity, V_dep, is calculated as:

where V_ambient is the theoretical deposition velocity and γ_mesh is the mesh factor. The mesh factor is an empirical factor that accounts for the fact that the sampler's mesh cover reduces the deposition of particles to the substrate. For coarse particles, it may be calculated as:

where d_a is the particle's aerodynamic diameter, τ is its relaxation time, g is the gravitational constant, and ν is the kinematic viscosity of air.

Wagner and Leith accounted for turbulent inertial deposition with the concept of friction velocity, u*, in the term V_ambient as follows:

where

where V_t = τg and D is the diffusivity of the particle. They suggested that the following equation may be used to estimate u*:

where k is von Karman's constant which can be assumed to be 0.4, u is the wind speed, z is the height above ground at which the wind speed is measured at, and z _o is the surface roughness length. Note that k was inadvertently omitted from the formula in the original paper as confirmed by Wagner (personal communication, February 23, 2007).

According to Wagner and Leith (2001a), turbulent inertial deposition may be neglected when u* is less than 0.4 m s⁻¹. Such a situation occurs frequently in occupational settings (Baldwin and Maynard, 1998) and greatly simplifies application of this model. Without turbulent inertial deposition, the settling velocity of a particle, V_ambient, from Equations (4), (5) (6) reduces to:

There are two concerns with applying the Wagner–Leith deposition velocity model to estimate ambient PM_{10 − 2.5} from samples collected passively. First, the model was only tested under conditions where turbulent inertial deposition was negligible. Friction velocity, u*, was shown less than 0.4 m s⁻¹ for all wind tunnel tests conducted by Wagner and Leith (2001b). Moreover, field work has generally been limited to indoor, occupational settings, where turbulent inertial deposition was negligible (Wagner and Leith 2001c; Wagner and Macher 2003).

Second, the selection of an appropriate surface roughness, z _o , presents considerable difficulties. Surface roughness is typically described as the height at which the wind speed becomes zero due to friction from the surface. Equation (6) applies to heights greater than “a few meters” for even ideal homogenous terrain or more practically from 5 m to 14 m for a sub-urban environment (Wieringa 1993). Thus, Equation (6) is probably not applicable for a passive sampler located 2 m from the ground in a populated area. In this type of area, surface roughness likely depends on site specific details, such as the position of trees or houses, and thus may be heterogeneous in many environments.

METHODS AND MATERIALS

Passive Sampler Operation

The passive samplers used in this work were identical to those described by Wagner and Leith (2001a) and were operated and analyzed with the modifications described previously (Ott et al. accepted). Briefly, digital images of the particles on the sample media were captured at 100× magnification (10× objective lens) with a light microscope (Leica DMLSP, Leica Microsystems, Wetzlar, Germany) equipped with a digital camera (Leica DFC 280, Leica Microsystems, Wetzlar, Germany). A stage micrometer was used to determine that the pixel resolution of digital images was 1.91 pixels per micron. The size of each image was 1064 × 1280 pixels. Particles were counted and sized automatically with ImageJ (NIH, Bethesda, MD). The software output included circularity for each particle detected in the images.

This imaging process was validated by sizing borosilicate glass microspheres (Catalog # 9005, Duke Scientific, Fremont, CA) with a 5-μ m physical diameter certified to NIST traceable standards. Adjustments were made to the microscope and camera setup to ensure that the diameter of particles measured with this automated counting and sizing procedure were within 5% of the certified diameter before samples were analyzed. Image edge effects were neglected because the area of the image was very large compared to the size of the particles of interest.

Computation of Mass Concentration

Particle mass concentration was computed following Ott and Peters (2008). Friction velocity, u*, was not included when applying the deposition velocity model to facilitate comparison of different shelters, thus, Equation (7) was the deposition velocity model. Particle aerodynamic diameter was assumed equal to the projected area diameter obtained from microscopy to eliminate the need to estimate the aerodynamic shape factor, S_D. The mass of a single particle, m, was then calculated as:

where ρ_p is the density of the particle, d_pa is the projected diameter of the particle determined from microscopy, and S_V is the volumetric shape factor.

A unique volumetric shape factor, S_V, was determined from particle circularity, C_p, output by ImageJ as:

where P is the perimeter and A is the projected area of the particle. A circularity of unity indicates that the particle is a perfect circle, while circularity progressively decreases from unity the more irregularly shaped a particle appears. The contribution of a single particle to mass concentration was then calculated with Equation (1) and PM_{10 − 2.5} was calculated as:

where E is the PM₁₀ curve as defined by Hinds (1999):

and d_a,i is in μ m.

Shelter Design

As shown in Figure 1, six shelters were tested in this work. The design of five of the shelters (Figure 1a, Figure 1b, Figure 1c, Figure 1d, Figure 1e) were based on preliminary wind tunnel tests. The Flat Plates shelter (Figure 1a) consisted of two parallel plates with the top plate (diameter = 20.32 cm) 1.6 cm above a bottom plate (diameter = 12.7 cm). The passive sampler was placed into a rubber grommet set in the middle of the bottom plate. Based on preliminary tests with a garden hose to simulate rain, a shallow groove was cut around the perimeter of the underside of the top plate, 1 cm from the edge, to break surface tension of water drops hanging on the underside of the top plate. These tests also led to the incorporation of rubber washers to seal the screws that hold the top plate in place. This shelter restricted atmospheric turbulence to a height less than the distance between plates and minimized generation of turbulence as air flowed into it.

FIG. 1 Shelters tested in the wind tunnel.

The top plate in the Flat Plates shelter was sized to allow coarse particles to enter the shelter but to minimize the possibility that rain would hit the bottom plate. We considered a rain drop with a settling velocity, Vy, being carried in a horizontal wind with a speed, Vx. The rain drop would just hit the bottom plate if Dx/Vx = Dy/Vy, where Dx is the horizontal distance between the edge of the top plate and the edge of the bottom plate (Dx = 3.8 cm) and Dy is the separation distance between the plates (Dy = 1.6 cm). The smallest raindrop is about 1000 μ m in diameter and has a terminal settling velocity of 4 m s⁻¹ (Ahrens 2007). A wind speed, Vx, of 34.3 km hr⁻¹ would be required for a droplet of this size to just reach the bottom plate.

The Louvered Plates shelter (Figure 1b) was similar to the Flat Plates shelter in that it allowed air to pass through relatively unobstructed. It consisted of two plates with edges bent downward to serve as a drip edge for rain. This design was motivated by the approach used to prevent rain intrusion into the PM₁₀ inlet for the dichotomous sampler (Tolocka et al. 2001).

The Fence, Drum, and Funnel Lid shelters were designed with the intent of blocking the wind to provide a calm sampling environment for the passive sampler. In the Fence design (Figure 1c), a top plate was spaced apart from a bottom plate with 24 posts that were intended to obstruct and slow the air flow. The Drum shelter (Figure 1d) consisted of a 20.32-cm-diameter pipe as the main housing with flanges on the top and bottom. Two pie pans were attached 2.5 cm above the openings in both ends to force air into the main housing. A hole was cut into the top pan to allow particles to settle into the shelter during low wind speed events, and a plate was placed in the center of the pipe to block any direct impaction from the altered air flow from above. In the Funnel Lid shelter (Figure 1e), a funnel was welded to the top plate which was then attached over an enclosed pipe. This shelter was designed to reduce turbulence in its large internal cavity.

A sixth shelter, the Gill Screen (Figure 1f), has been used in conjunction with passive sampling by other researchers (Leith et al. 2007). The Gill Screen is a commercially available (Model 41003, RM Young and Co., Traverse City, MI) component that is used to shield thermometers from solar radiation. For this project, it was modified by cutting holes through all but the top two plates to accommodate a supporting stand for the passive sampler.

Wind Tunnel Tests

As shown in Figure 2, laboratory tests were conducted in a straight, 20-m-long wind tunnel with a square cross-sectional test section (61 cm × 61 cm). Test dust (primarily road dust, ρ _p = 2.65 g cm⁻³, ISO 12103-1, A3 Medium, Powder Technology, Inc., Burnsville, MN) was injected into the tunnel 13 m upstream of the test section with a Venturi nozzle (Model JD-90M, Vaccon, Medfield, MA). A panel that blocked the center two-thirds of the tunnel was positioned 1 m downstream of the dust injection port to force mixing. A perforated steel sheet (0.95 cm diameter hole size, staggered spacing with a distance of 1.43 cm center-to-center, catalog no. 9255T911, McMaster-Carr, Atlanta, GA) 4 meters further downstream provided a uniform distribution of air flow.

FIG. 2 Experimental setup for laboratory tests.

A passive sampler was placed in each of the six shelters, and the shelters were installed in one-half of the test section. Four of the shelters (Gill Screen, Flat Plates, Louvered Plates, and Fence) were small enough to fit into the test section of the tunnel at once so those runs were completed together. The Drum and Funnel Lid shelters were tested individually because of their larger size. The inlet of a dichotomous sampler (Model SA241CUM, Graseby-Andersen, Fulton, GA) was centered in the other half of the test section. 37-mm PTFE filters (Part # R2PJ037, Gelman Sciences, Ann Arbor, MI) were used to collect coarse and fine particulate in the dichotomous sampler. Using a volumetric flow meter (TriCal, BGI, Inc., Waltham, MA) the coarse and fine air flows of the dichotomous sampler were set before and verified after each run to be within 5% of their design values (coarse flow = 1.67 Lpm and fine flow = 15.0 Lpm). These filters were analyzed gravimetrically with a micro-balance (Model MT5, Mettler-Toledo, Columbus, OH). The coarse mass concentration was adjusted for fine particle contamination in the calculation (Poor et al. 2002). All filters were equilibrated in the weigh room for 24 hours prior to pre-run and post-run weighing.

Each shelter was tested three times at each of three wind velocities required for PM₁₀ inlet testing by the US EPA (2, 8, and 24 km hr⁻¹; 40 CFR Part 53.42), except for the Drum and Funnel Lid shelters that were only tested at 24 km hr⁻¹. Wind velocity was measured with a thermal anemometer (Velocicalc 8347, TSI, Inc., Shoreview, MN) to be uniform in the test section (5 m downstream of the laminar-flow panel) at 9 points that were equally spaced across the face of the test section for each wind speed. The overall CV for wind speed measurements averaged 7.1% (3.4% at 24 km hr⁻¹, 4.3% at 8 km hr⁻¹, and 13.5% at 2 km hr⁻¹). Spatial variability within the tunnel was measured at 9 points with uncovered passive samplers and found to have CV of 20.1%. For each run, the air flow velocity in the center of the test section was set with the anemometer. Run times were varied to account for different mass concentration in the test section for different wind speeds: 15 min for 2 km hr⁻¹; 40 min for 8 km hr⁻¹; and 120 min for 24 km hr⁻¹. Mean mass concentration was 6 mg m⁻³ at 2 km hr⁻¹, 2 mg m⁻³ at 8 km hr⁻¹, and 1 mg m⁻³ at 24 km hr⁻¹.

Three blank passive samplers and a blank filter were transported and analyzed for each test. Passive sampler blanks were treated in the same manner as samples except that their cover was immediately replaced upon opening. Airborne mass concentration from blank passive samplers was subtracted from the corresponding samples. This subtraction accounted for 0.3% of the total mass on average.

Field Tests

The Flat Plates shelter was selected for further evaluation in field tests. Eight seven-day field trials were conducted at three sites in Iowa City, IA. The first site was adjacent to the Institute for Rural & Environmental Health (IREH) Building on the University of Iowa Oakdale Campus, the second was in an open field just east of the outer-most buildings at the Iowa City Airport, and the third was at a residence in a well-established neighborhood in northeast Iowa City. For the IREH site, potential sources of PM_{10 − 2.5} included commercial construction projects, Interstate 80, a high-traffic commercial complex, and several factories within 2 km. Two large stone quarries were located within 4.5 km of the IREH site. The site at the airport was potentially impacted by a small quarry operation and industrial facilities within 0.5 km. The residential site was surrounded by homes within one km with other potential sources including a major construction project 1.5 km away and Interstate 80 and industrial facilities more than 2.5 km away. The residence was 4.4 km from the airport site and 4.8 km from the airport runway, where the official aviation weather sensors were located. IREH was 9.4 km from the airport and 10.2 km from the residence site.

In each trial, three passive samplers were each installed in a Flat Plates shelter and collocated with a dichotomous sampler. On-site meteorological equipment (Weather Transmitter WXT510, Vaisala, Helsinki, Finland or Oregon Scientific WMR968, Portland, OR) recorded temperature, relative humidity, and wind speed within 2 m of and at the same height as the passive samplers. Each passive sampler shelter was placed on top of its own support post at 1.8 m above the ground. All samplers were placed in a range of 3 m to 6 m from each other at a given site. The dichotomous sampler was operated as described in the laboratory tests. Blank passive samplers and blank dichotomous sampler filters were taken into the field for each of the trials. Samples were analyzed as before except density was assumed to be 2.0 g cm⁻³ as done previously for ambient PM (Wagner and Macher 2003). The samplers were routinely observed during the sample periods to note the effectiveness of the shelter to prevent precipitation intrusion.

Data Analysis

For laboratory tests, the primary indicator of shelter effectiveness was the PM_{10 − 2.5} ratio, the ratio of PM_{10 − 2.5} measured with the passive sampler to that measured with the dichotomous sampler. For each shelter, the coefficient of variation, CV, of PM_{10 − 2.5} measured with the passive sampler was calculated for each wind speed and then averaged across all wind speeds to obtain overall CV. The overall CV for the dichotomous sampler was calculated in a similar fashion. CV for the passive sampler in the field tests was calculated with the data derived from the three collocated samplers from a single test period. Calculations were carried out in an Excel spreadsheet (Microsoft Corp., Redmond, WA).

For field tests, a linear model was fit to the data using PROC GLM in SAS (SAS Institute Inc., Cary, NC) with PM_{10 − 2.5} measured with the passive sampler as the dependent variable and that measured with the dichotomous sampler as the independent variable. The mean wind speed, mean temperature, mean relative humidity, and mean daily precipitation of the sample period were included as possible predictors in the model. Starting with all variables in the model, backwards elimination was performed until only variables with a p-value less than 0.05 remained. The number of data points prevented interactions from being entered into the model.

Meteorological data was downloaded for the Iowa City Airport from the National Climactic Data Center (NCDC) to compare the data collected at each site during the sample period. This analysis served to determine the uncertainty associated with using readily available airport wind data instead of onsite measurements.

RESULTS

Wind Tunnel Tests

Table 1 summarizes the wind tunnel tests averaged over all wind speeds and for the highest wind speed. A PM_{10 − 2.5} ratio of 1.0 would indicate that PM_{10 − 2.5} measured with the passive sampler was equal to that measured with the dichotomous sampler. The CV is an indicator of sampler precision; lower CVs reflect better precision. The PM_{10 − 2.5} ratio averaged over all wind speeds was nearest to one for the Flat Plates shelter (1.04). The CV was lowest for this design (13.3%) and even lower than that for the dichotomous sampler (15.4%). The PM_{10 − 2.5} ratios for other shelters were substantially greater than unity and ranged from 2.4 to 21.8. CVs for these shelters ranged from 16.1 to 49.9%.

TABLE 1 Summary of wind tunnel tests averaged over all wind speeds and for the highest wind speed (24 km hr⁻¹)

	PM10 − 2.5 Ratio	PM10 − 2.5 Ratio
Sampler	(all wind speeds)	(24 km hr^−1)	CV%
Dichot	–	–	15.4
Flat Plates	1.04	1.4	13.3
Louvered Plates	2.36	1.6	40.7
Fence	4.86	8.7	25.2
Drum	**	10.5	16.1
Funnel Lid	**	21.8	49.9
Gill Screen	8.48	13.1	40.5
**only done at high speed.

Figure 3 presents the PM_{10 − 2.5} ratio for 4 of the shelters. A dotted line indicates an ideal ratio of 1.0. At the low wind speed, the PM_{10 − 2.5} ratio was greater than 2 for all shelters other than the Flat Plates shelter. The ratios progressively increased with wind speed to 13.1 for the Gill Screen shelter and to 8.7 for the Fence shelter. Although the ratio for the Louvered Plates shelter did not increase with wind speed, it was consistently greater than unity. In contrast, the PM_{10 − 2.5} ratio for the Flat Plates shelter (Figure 3b) was substantially closer to unity than that of the other designs and ranged from 0.8 at 2 km hr⁻¹ to 1.4 at 24 km hr⁻¹.

FIG. 3 PM_{10 − 2.5} ratio for four shelters by wind speed: A. Gill Screen, Fence, and Louvered Plates shelters; B. Flat Plates shelter.

Figure 4 shows particle count distributions, normalized by the number of images for each analysis, as measured with the passive sampler housed in four of the shelters. For particles smaller than 2.8 μ m, the counts per image were similar in magnitude for all shelters. In contrast for particles larger than this size, substantially lower counts per image (100× lower in some cases) were observed for the sample collected in the Flat Plates shelter compared to that collected in other shelters. At the low wind speed (Figure 4a), the counts were distributed similarly for samples collected in the Gill Screen, the Fence, and Louvered Plates shelters. These distributions diverged for higher wind speeds (Figures 4b, c).

FIG. 4 Particle size distributions measured on the substrates for the four shelters tested at three wind speeds: A. 24 km hr⁻¹; B. 8 km hr⁻¹; C. 2 km hr⁻¹. Note that Y-axis scales vary.

Field Tests

Table 2 summarizes the eight field tests using the Flat Plate design. The PM_{10 − 2.5} ratio ranged from 1.11 to 1.49 with a mean of 1.29 ± 0.16. The CV of the eight tests ranged from 3.4 to 20.8% with an overall average of 9.5%. Mean weekly wind speed ranged from 2.66 to 9.29 km hr⁻¹ with an overall mean of 5.4 ± 2.3 km hr⁻¹. Mean weekly relative humidity ranged from 64.8 to 84.5%, and mean weekly temperature ranged from −1.1 to 7.3°C.

TABLE 2 Summary of field tests

	Test	Dichot PM10 − 2.5	Passive/Dichot	Passive	Mean Wind	Mean Precip	Mean	Mean
Site	Week #	(μ g m^−3)	Ratio	CV %	Speed (km hr^−1)	(cm day^−1)	RH%	Temp (°C)
Airport	2	21.2	1.39	3.8	5.05	0.01	70.3	6.9
Airport	3	13.9	1.27	7.3	7.81	0.43	72.8	6.0
Airport	4	12.8	1.11	19.9	9.29	0.01	71.5	−1.1
Airport	5	13.4	1.25	7.2	6.00	0.00	70.1	1.7
Airport	6	3.7	1.18	9.4	5.90	0.65	84.5	0.8
IREH	1	20.9	1.43	4.1	3.22	0.01	64.8	4.6
IREH	2	12.5	1.56	3.9	3.70	0.32	75.8	7.3
Residence	4	13.4	1.11	20.8	2.66	0.01	71.5	−1.1
Mean	—	14.0	1.29	9.5	5.45	0.18	72.7	3.1
St. Dev.	—	5.48	0.16	7.0	2.29	0.25	5.70	3.49

As shown in Table 3, only mean weekly temperature of the meteorological parameters was significant (p = 0.024) in the linear model of PM_{10 − 2.5} measured with the passive samplers regressed on that measured with the dichotomous sampler. The r² value was increased from 0.943 without temperature to 0.987 with temperature. The coefficient for PM_{10 − 2.5} measured with the dichotomous sampler would be equal to unity if the PM_{10 − 2.5} derived from the passive sampler agreed perfectly with that from the dichotomous sampler. This coefficient was greater than unity in both models, although the addition of temperature significantly reduced it from 1.47 ± 0.15 to 1.31 ± 0.10. The intercept in the model would be zero if the passive sampler performed as a perfect indicator for the dichotomous sampler. In the model without meteorological variables, the intercept was −2.2 and was not statistically different from zero (p = 0.35). With temperature included in the model, the intercept was −11.1 and was statistically different from zero (p = 0.015). Wind speed was found to be insignificant when added to the original model (p = 0.35).

TABLE 3 Summary of PM_{10 − 2.5} measured with passive sampler regressed on that measured with dichotomous sampler: with and without meteorological data

Model without Meteorological Data
Parameter	Estimate	Std. Error	p-value
Intercept	−2.22	2.20	0.352
Dichot	1.47	0.15	< 0.0001
r^2= 0.943
Final Model with Meteorological Data
Intercept	−11.08	3.03	0.015
Dichot	1.31	0.10	< 0.0001
Temperature	0.30	0.09	0.022
r^2 = 0.987
Full Model with Meteorological Data			(Type III)
Intercept	−8.86	28.09	0.782
Dichot	1.24	0.46	0.116
Wind speed	−0.28	0.30	0.444
Temperature	0.31	0.22	0.297
R.H.	−0.0007	0.35	0.999
Precipitation	−0.99	7.07	0.902
r^2 = 0.988

Figure 5 shows that hourly wind speeds measured at passive-sampler height at two of the field sites (Airport on site and Residence) were substantially lower than that measured at the airport runway (on tower, height = 10 m). Regression analysis (Table 4) shows that hourly wind speeds recorded on the airport tower were correlated moderately with those measured at the airport site (r² = 0.51) and to a lesser extent with those measured at the residence (r² = 0.42). The slope of the linear regression indicates that hourly wind speeds recorded on the airport tower were substantially greater than those recorded at the airport site (slope = 0.41) and at the residence (slope = 0.15). The regression of the hourly wind speed measured at the residence on that measured at the airport site shows that these values are only moderately correlated (r² = 0.43). The mean and standard deviation of the weekly wind speed averages for the two years previous to this study at the NCDC airport site was 9.2 ± 4.0 km hr⁻¹. The result for the weekly averages at this same site for the weeks of the study was 8.9 ± 3.0 km hr⁻¹.

FIG. 5 Box and whisker plot of wind speed measured at different sites. Minimum, maximum, median, lower quartile, and upper quartile are shown. A plus symbol represents outliers with values greater than 1.5 times the median value.

TABLE 4 Pairwise comparison of hourly wind speed at different stations

	Airport	Airport	Airport
X-Station	(on tower)	(on tower)	(on site)
Y-Station	Airport (at site)	Residence	Residence
N	1379	738	528
Slope ± St. Err.	0.41 ± 0.01**	0.15 ± 0.01**	0.33 ± 0.01**
Intercept ± St.	1.98 ± 0.16**	−0.286 ± 0.10**	−0.76 ± 0.11**
Err., km hr^−1
r^2	0.51	0.42	0.53

There were significant rain events during Week 2 and Week 3 and light snow and ice during Week 4. Heavy rain and periods of fog and freezing fog occurred during Week 8. No major infiltration of precipitation was noted in any of the shelters during these precipitation events. Frost was commonly found on the top of the topmost flat plate in early morning but not on the bottom plate nor on the passive sampler. Frost formed on one passive sampler once after conditions of high humidity at night but quickly evaporated during the day. During drizzle, some droplets were seen on the top of the bottom plate but not on the mesh of the passive sampler. There was no apparent impact of frost or drizzle on the appearance of the particles in microscopic analysis conducted on this sample.

Further tests with water sprayed from a garden hose to simulate rain confirmed the ability of the Flat Plates shelter to protect the passive sampler from very heavy rain at a steep angle of inclination (30 degrees from vertical). These tests also identified the importance of mounting the shelter so that the bottom plate fully covers its support to eliminate splash of drops into the shelter.

DISCUSSION

This work demonstrates that the Flat Plates shelter protects the Wagner-Leith passive sampler from precipitation and from the effects of wind. This shelter is simple to construct and deploy. In wind tunnel tests, PM_{10 − 2.5} measured with a passive sampler installed in this shelter was near that measured with a dichotomous sampler (ratio = 1.04 averaged over all wind speeds) with precision comparable to that of the dichotomous sampler. The field comparisons demonstrated that PM_{10 − 2.5} measured with passive samplers in the Flat Plates shelter was correlated highly with that measured with the dichotomous sampler. They also confirmed that the wind speed does not significantly affect this relationship (p-value = 0.44).

The reduced effect of wind on particle deposition offered by the Flat Plates shelter resolves several major issues with application of the Wagner-Leith passive sampler to ambient environments. It eliminates the need for on-site collection of wind speed data. It also greatly simplifies application of the deposition velocity model by allowing V_ambient in Equation (2) to be modeled as terminal settling velocity (τ g), thereby rendering Equations (4) through (6) unnecessary. Consequently, there is no need to assume a value for surface roughness length, z _o required to estimate friction velocity, u*, in the turbulent deposition component of the model.

Although mean weekly temperature was significant in the linear model used to explain PM_{10 − 2.5} measured with passive samplers with that measured with the dichotomous sampler, it only improved the relationship marginally. Temperature does affect air density and air viscosity that in turn may alter the settling velocity of a particle. However, the small temperature differences encountered in this field study were insufficient to significantly influence deposition rates to the passive sampler. Further work is needed to investigate if temperature should be included in the deposition velocity model.

Several sources of uncertainty may explain why PM_{10 − 2.5} ratios were greater than unity in field tests. The values assumed for dynamic shape factor, volumetric shape factor, or particle density may have been inaccurate for use with the particles encountered in this study. These values are not definitively known for ambient particles and may vary with site and season (Noll et al. 1988; Pratesi et al. 2007). Sources of uncertainty in analysis include focus of the microscope, which is somewhat dependent on human judgment and influences particle sizing. For example, the position of the passive sampler relative to upwind obstacles may influence particle deposition. In our study, the IREH site had a solid fence and a two-story building within 4 m of the passive samplers and had PM_{10 − 2.5} ratios greater than at any other site (Table 2).

Compared to the Flat Plates shelter, other shelters caused a dramatic increase in deposition of particles to the surface of the passive sampler. Although the PM_{10 − 2.5} ratio for the Louvered Plates shelter was relatively insensitive to wind speed (Figure 3), it was substantially greater than one for all wind speeds. The louvers may have introduced turbulence to the air flow as it passed into the shelter and over the passive sampler. Such turbulence would explain the greater counts per image across all particle sizes for the Louvered Plates shelter compared to that of the Flat Plates shelter (Figure 4).

PM_{10 − 2.5} measured with passive samplers in the Gill Screen and Fence shelters was influenced greatly by wind speed (Figure 3a). The dramatic increase in PM_{10 − 2.5} ratio for the Gill Screen and the Fence shelter with wind speed may be attributed to an increase in the deposition of particles larger than 5 μ m (Figure 4). This increased deposition may partially result from turbulence introduced as the air flow interacts with the shelter features. A secondary impaction mechanism resulting in directional change in the air flow above the passive sampler may play a role in the deposition of large particles in these shelters. Also, turbulence may impart a net loss in horizontal momentum that allows particles to drop out of the forward-moving flow. Similar mechanisms may explain why PM_{10 − 2.5} ratios were substantially greater than unity for the Drum and Funnel Lid shelters at high wind speed. Consequently, wind speed must be taken into account when PM_{10 − 2.5} is measured with passive samplers in shelters other than the Flat Plates shelter.

The analysis of wind speed data revealed important spatial differences relevant to the discussion of passive sampling with wind-speed dependent shelters. Mean hourly wind speeds measured at the runway (10 m height) were substantially greater and only moderately correlated with those obtained from equipment collocated with and at the same height as the passive samplers. Since correlation is not high, incorporation of NCDC data would lead to error in the calculation of turbulent inertial deposition with the model of Wagner and Leith. This source of error would suggest that wind speed data should be collected with meteorological equipment at the same site as the passive sampler if wind-speed dependent shelters are used. In addition, substantial error may be introduced if arithmetic mean wind speeds are used to correct passive sampler data for shelters dependent on wind speed because the deposition rates of larger particles may not be linearly related to wind speed. Short durations of high winds may cause substantial deposition of large particles (Figure 4) to the passive sampler leading to an overestimate of PM_{10 − 2.5}.

In conclusion, the Flat Plates shelter protects the Wagner-Leith passive sampler from precipitation and eliminates the need to account for wind speed effects when measuring PM_{10 − 2.5}. Wind tunnel tests revealed that particle deposition to a passive sampler installed in this shelter was much less effected by wind speed than in other shelters. Field tests showed that PM_{10 − 2.5} measured with a passive sampler in the Flat Plates shelter correlated highly with that measured with a dichotomous sampler (r² = 0.987) and that wind speed did not improve the correlation. This arrangement enables passive measurement of PM_{10 − 2.5} without on-site wind-speed data and without the assumptions inherent to the deposition velocity model of Wagner and Leith to account for turbulent inertial deposition. Passive samplers in Flat Plates shelters may be used to assess the variability of ambient PM_{10 − 2.5} in epidemiological studies at fine spatial resolution with little demand on resources compared to that required to operate filter-based samplers.

Acknowledgments

The authors greatly appreciate the financial support from the Center for Health Effects of Environmental Contamination (University of Iowa) and the US Environmental Protection Agency (PR-RT-06-01007/U2C694).

Notes

**value is significantly different from 0 at α = 0.05 significance level.