Advances in Integrated and Continuous Measurements for Particle Mass and Chemical Composition
By Chow, Judith C Doraiswamy, Prakash; Watson, John G; Chen, L-W Antony; Ho, Steven Sai Hang; Sodeman, David A
ABSTRACT Recent improvements in integrated and continuous PM^sub 2.5^ mass and chemical measurements from the Supersite program and related studies in the past decade are summarized. Analytical capabilities of the measurement methods, including accuracy, precision, interferences, minimum detectable levels, comparability, and data completeness are documented. Upstream denuders followed by filter packs in integrated samplers allow an estimation of sampling artifacts. Efforts are needed to: (1) address positive and negative artifacts for organic carbon (OC), and (2) develop carbon standards to better separate organic versus elemental carbon (EC) under different temperature settings and analysis atmospheres. Advances in thermal desorption followed by gas chromatography/mass spectrometry (GC/MS) provide organic speciation of approximately 130 nonpolar compounds (e.g., n-alkanes, alkenes, hopanes, steranes, and polycyclic aromatic hydrocarbons [PAHs]) using small portions of filters from existing integrated samples. Speciation of water- soluble OC (WSOC) using ion chromatography (IC)-based instruments can replace labor-intensive solvent extraction for many compounds used as source markers. Thermal gas-based continuous nitrate and sulfate measurements underestimate filter ions by 10-50% and require calibration against on-site filter-based measurements. IC-based instruments provide multiple ions and report comparable (+-10%) results to filter-based measurements. Maintaining a greater than 80% data capture rate in continuous instruments is labor intensive and requires experienced operators. Several instruments quantify black carbon (BC) by optical or photoacoustic methods, or EC by thermal methods. A few instruments provide real-time OC, EC, and organic speciation. BC and EC concentrations from continuous instruments are highly correlated but the concentrations differ by a factor of two or more. Site- and season-specific mass absorption efficiencies are needed to convert light absorption to BC. Particle mass spectrometers, although semiquantitative, provide much information on particle size and composition related to formation, growth, and characteristics over short averaging times. Efforts are made to quantify mass by collocating with other particle sizing instruments. Common parameters should be identified and consistent approaches are needed to establish comparability among measurements.
This review describes and evaluates integrated and continuous particulate matter (PM) measurements from the Supersite program and related studies from the past 7-10 yr, building on previous reviews.1-5 In addition to measurement methods for compliance networks, measurements related to aerosol properties (number concentration, density, light extinction/absorption efficiencies, refractive index, gas/particle equilibrium, particle size distribution, and composition) are given. Advances in measurements for PM mass, nonvolatilized and volatilized inorganic ions, organic carbon (OC) and elemental carbon (EC), water-soluble OC (WSOC), thermal- and solvent- extractable organic compounds, and single particle mass spectrometry (MS) are evaluated. The reported accuracy, precision, detection limits, interference, and comparability of these measurements are documented. Recent developments in size-selective inlets and their equivalence status are explained. Comparisons within and between samplers are summarized. Table 1 lists abbreviations for the integrated and continuous instruments included in the review. Table 2 summarizes the web-based supplemental information (Tables 3-17), which is available online only at http://secure.awma.org/journal/pdfs/ 2008/ 2/10.3155-1047-32126.96.36.199_supplmaterial.pdf) that was compiled for this article, but is too lengthy for print. Data from these measurement technologies are used with receptor models to identify and quantify source contributions6 (this issue). This paper complements Solomon and Sioutas7 (this issue), which examines comparability among integrated and continuous methods with recommendations on their application in existing air quality monitoring networks. It also complements Demerjian and Mohnen8 (this issue), which illustrates temporal variations of PM size and components. Lessons learned from time-resolved PM measurements are summarized by Wexler and Johnston9 (this issue).
The following questions are answered:
* What advances have occurred in sampling and analytical methods for measurement of mass, components of mass, semi-volatile species, and precursor species, including integrated, semicontinuous, continuous, and single particle methods?
* What are the analytical characteristics of the methods: uncertainty (accuracy, precision), minimum detectable limits (method detection limits [MDLs]), data capture, and reliability?
* What have we learned about interferences and the variables that affect those interferences, how are they accounted for in reported data, and how do they affect the uncertainty in the method?
Size-selective inlets used on several instruments in Table 1 achieve size separation using impactors, virtual impactors, cyclones, and selective filtration.10,11 Table 3 updates information from Chow1 and Watson and Chow12 for U.S. Environmental Protection Agency (EPA) PM^sub 2.5^ and PM^sub 10^ Federal Reference Methods (FRM) and Federal Equivalent Methods (FEM) used for regulatory compliance determinations. With the implementation of the fine particulate matter (PM^sub 2.5^) National Ambient Air Quality Standard (NAAQS), the Well Impactor Ninety-Six (WINS)13 was a designated inlet for the PM^sub 2.5^ FRM sampler. 14,15 The impactor design requires a 50% cutpoint (d50) of 2.48 [mu]m and a geometric standard deviation (GSD) of 1.18.13 The regulations require cleaning the WINS impactor every 5 days of 24-hr sampling.16 Vanderpool et al.17 showed that after 120 hr of sampling, cutpoints of 13 WINS impactors ranged from 2.32 to 2.51 [mu]m (averaged 2.41 [mu]m), with a maximum bias of -2.1% (within the allowed 5% bias) in the PM^sub 2.5^ mass.
The capacity of WINS was further evaluated against the Sharp-Cut Cyclone (SCC).18-20 After 4 wk of sampling at a parking lot, the SCC d50 was reduced by approximately 4.5% (2.46 [mu]m to 2.35 [mu]m) and the WINS d^sup p^ was reduced by 12% (2.44 [mu]m to 2.15 [mu]m).18 Maximum bias in PM^sub 2.5^ mass based on an idealized size distribution was estimated to be approximately +5% for the SCC and – 8.8% for the WINS.18 Follow-up development of the Very Sharp-Cut Cyclone (VSCC; BGI, Inc.) resulted in an approved FEM inlet, with a d50 of 2.5 [mu]m and a GSD of 1.16.21 Laboratory experiments showed that the VSCC cutpoint was not affected at high loadings (e.g., 150 [mu]g/ m^sup 3^) for approximately 90 days of 24-hr sampling, with a slight increase in GSD from 1.16 to approximately 1.2. Field evaluations at Hartford, CT, and Phoenix, AZ,21 showed that PM^sub 2.5^ samplers with VSCCs satisfied EPA criteria for equivalency.22 As shown in Table 3, eight instruments (e.g., BGI PQ 200/200A-VSCC; R&P-2000/2000A/2025-VSCC; RAAS-100/200/300-VSCC) have been designated Class II FEMs for PM^sub 2.5^ NAAQS compliance since Watson and Chow.12 Several of the continuous instruments and PM speciation samplers have been equipped with a SCC or VSCC.23
Compliance inlets for PM^sub 2.5^ and PM^sub 10^ FRMs and FEMs are insufficient for understanding many PM properties. Research inlets were developed for this purpose. Chakrabarti et al.24 designed a low pressure-drop ultrafine particle (UP) impactor (d50 of 148 +- 10 nm, GSD of ~1.38) for use with the Beta Attenuation Monitor (BAM) as a nano-BAM. Comparability (average ratio 0.92 +- 0.12; R^sup 2^ = 0.92) was found between nano-BAM and microorifice uniform deposit impactors (MOUDIs). The nano- BAM captures real- time concentration peaks that are not quantified by the Scanning Mobility Particle Sizer (SMPS), probably because of particle agglomeration. By adjusting the upstream pressure (0.05-1 standard atmospheric pressure [atm]), Middha and WexleR^sup 2^5 designed a high-volume slot-type UP virtual impactor with a minor flow that was 10-14% of the total flow at 1 atm. The d^sub 50^ of this UP inlet can be adjusted from 13 to 200 nm by increasing the operating pressure without major changes to the minor to total flow ratio. The sampling efficiencies of this slot virtual impactor need to be tested for field operation.
FILTER-BASED INTEGRATED SAMPLERS
Nine integrated sampling systems (including inlet, surface material, denuder [optional], filter medium and holder, and pump) have become commercially available since 1997 (Table 1). Several types have been applied in the U.S. urban Speciation Trends Network (STN, which is part of the Chemical Speciation Network [CSN])26,27 and non-urban Interagency Monitoring of PROtected Visual Environments (IMPROVE)28 network. Reviews by Chow,1 Watson and Chow,12,29 Solomon et al.,30,31 Ashbaugh and Eldred,32 and McMurry et al.33 detail different configurations of speciation samplers in the CSN and IMPROVE networks. These instruments use annular or honeycomb denuders with an aluminum or glass surface coated with sodium carbonate (Na^sub 2^CO^sub 3^) or magnesium oxide (MgO) to remove nitric acid (HNO^sub 3^). Characteristics and comparisons can be found in Solomon et al.30,31 Other samplers, such as the DRI- sequential filter sampler (SFS),34-40 PC-BOSS,41 HEADS,42 ARA- particle composition monitor (PCM),43 the University of Wisconsin sampler, 44,45 the CMU sampler,46-48 and the Hi-Flow sampler49 identified in Table 1 have been developed for research studies (Tables 12-16). These samplers often include upstream denuders and tandem filter packs that allow the evaluation of sampling artifacts and gas/particle equilibrium at variable flow rates.
Several instruments have been applied for particle size and chemical composition: (1) MOUDI (MSG Corp.)50-52; (2) electrostatic low-pressure impactor (ELPI), which determines mass by counting the charge-induced current from each impactor plate with an assumed spherical particle and user defined density, (1-sec average)53-55; (3) low-pressure impactor (LPI)56,57; and (4) the Davis rotating- drum uniform size-cut monitoring (DRUM).58 Comparisons between MOUDI, FRM, and non-FRM samplers are given in Table 12 for mass and Tables 13-16 for elements, ions, and carbon species.
Chemical Speciation of Filter Samples
Analytical Methods. Several analytical methods have been applied to filter samples. These include: (1) mass by gravimetry14,15,59; (2) light transmission (babs) by densitometer or reflectance60; (3) elements by X-ray fluorescence (XRF),61 proton-induced X-ray emission (PIXE),62 instrumental neutron activation analysis (INAA),63 atomic absorption spectroscopy (AAS)64 and inductivelycoupled plasma/MS (ICP/MS)65; (4) cations (water-soluble sodium [Na^sup +^], potassium [K^sup +^]], calcium [Ca2^sup +^]], and magnesium [Mg2^sup +^]]) by AAS65; (5) anions (nitrate [NO^sub 3^] ^sup -^]], sulfate [SO^sub 4^] ^sup 2-]^) by ion chromatography (IC)66; (6) multiple ions (e.g., chloride [Cl^sup -^]], NO^sub 3^] ^sup -^], SO^sub 4^] ^sup 2-^], ammonium [NH^sub 4^) ^sup +^]]) by IC, automated colorimetry (AC),67 ion-selective electrodes (ISE),68 and Fourier transform infrared spectrometry (FTIR)69; and 7) carbon (OC and EC) by thermal/optical reflectance (TOR) and transmittance (TOT) methods.70-78 Table 4 shows that these methods typically report 2-5% accuracy, 5-10% precision for mass, elements, ions, and total carbon (TC), and 10-20% precision for OC and EC. MDLs (assuming a 24-m^sup 3^ sampling volume) range from 0.004 to 0.36 [mu]g/m^sup 3^ for elements, 0.04-1 [mu]g/m^sup 3^ for mass and ions, and 0.03-0.8 [mu]g/m^sup 3^ for carbon. Although matrix interferences (e.g., AAS, ICP/ atomic emissions spectroscopy, ICP/ MS), peak overlap (e.g., IC), spectral interferences (e.g., XRF, PIXE, INAA, ICP/MS, FTIR), and lack of reference standards (e.g., EC and organic species) may result in analytical uncertainties, many discrepancies are attributable to inhomogeneous particle deposits, adsorption of gases on filter material, and variabilities in filter blanks.
OC and EC. Watson et al.76 showed that the many different analysis protocols used throughout the world can result in factors of 2 to 7 difference in EC.79 Because OC usually constitutes a large fraction (75-95%) of TC in ambient aerosols, its differences are smaller. Major differences between TOR and TOT analyses are attributed to charring of adsorbed gases or dissolved organic compounds collected within the filter during thermal analysis. 72,78 Good agreement was observed for diesel soot but not for ambient samples. Temperature differences between the sample and the sensor in the thermal/optical carbon analyzers influence the quantification of thermally- resolved carbon fractions, and a temperature calibration protocol80 needs to be implemented to minimize the differences.
In the United States, carbon measurements in CSN and the IMPROVE network will be consistent after 2007, because the same sampler (IMPROVE Module C or URG 3000N carbon sampler; URG corporation) and the IMPROVE carbon analysis protocol (i.e., IMPROVE_A,74 which includes both TOR and TOT) are implemented in both networks.81 Using identical sampling and analysis protocols to differentiate between OC and EC is essential because these components have different effects on urban air pollution and regional haze and are used to identify and quantify source contributions.6
WSOC. Cost-effective analytical techniques using existing filter remnants or distilled-deionized water (DDW) extracts (commonly applied for inorganic ions) are needed to identify and quantify nonpolar and polar species. In addition to solvent extraction followed by gas chromatography/MS (GC/MS),1,82,83 efforts have been made to better explain the composition of OC and to obtain a large database cost effectively for: (1) total WSOC and polar organic compounds by IC with electrochemical detection (ECD) and high- performance liquid chromatography (HPLC) with MS or pulsed amperometric detection (PAD); and (2) nonpolar organic compounds by thermal desorption (TD)-GC/MS.84 Figure 1 illustrates recent developments in practical carbon speciation.
Several methods have been used to analyze total WSOC on the basis of oxidation to carbon dioxide (CO2) followed by infrared (IR) detection,85-97 flame ionization detection (FID),92,98-100 and pyrolysis GC/MS.101 MDLs for WSOC are 0.01-0.23 [mu]g/m^sup 3^, with precision of +-3- 10%. Elemental compositions of hydrogen (H), carbon (C), nitrogen (N), and sulfur (S) in WSOC determined with an elemental analyzer using thermal conductivity detection (TCD; EA CHNS-O 1108; Carlo Erba)90 have a precision of approximately 2%. Dissolved organic nitrogen (DON) can be analyzed by ultraviolet (UV) and persulfate oxidation methods.102 The MDLs for organic and inorganic nitrogen are 0.001 [mu]g/m^sup 3^ and 0.071 [mu]g/m^sup 3^, respectively, with precisions from +-5 to +-30%. WSOC species have been quantified for: (1) neutral polyols and polyethers by GC/ MS, HPLC/MS, or IC/ECD87,89,95,103-106; (2) mono- and dicarboxylic acids (e.g., formate, and oxalate) by IC or GC/ MS86,94,97,98,103,107-111; (3) amino acids by HPLC/ fluorescence98,112,113; (4) polycarboxylic acids (e.g., unsaturated aliphatic and aromatic species) by GC/MS and HPLC/UV89; and (5) humic-like substances (HULIS) by HPLC/MS and HPLC/evaporative light scattering detection (ELSD).114,115 Table 4 shows that the method precisions for these species also range from +-5 to +-30%. Interferences result from incomplete extraction efficiency, incompleteness of derivatization, contamination introduced during sample preparation, and chromatographic peak overlap and resolution.
Specific efforts have been made toward better quantification of HULIS and levoglucosan (a biomass burning marker). HULIS are a group of macromolecular-size polycarboxylic acids.116,117 They are present in atmospheric (both primary and secondary),101,114,118 terrestrial (e.g., soil), marine, and biomass burning sources.119 Although there is no chromatographic technique available to resolve single compounds from HULIS,120 they may contribute up to 30% of the organic matter (OM = OC times a multiplier of 1.2-2.685,121 and 15- 60% of the total WSOC.122 HULIS are analyzed by size exclusion chromatography119,123 and ion-exchange chromatography.86 Spectrometric detection (i.e., UV, IR) and nuclear magnetic resonance (NMR)86,116,119,124,125 provide chemical structural information and quantification. Other quanti- fication methods include ELSD,115 electrospray ionization/ MS (ESI/MS), laser desorption ionization/MS (LDI/ MS),122,126-130 pyrolysis GC/MS,101 and capillary electrophoresis (CE).116,131
Engling et al.105 analyzed water extracts for levoglucosan by IC/ ECD instead of solvent extraction followed by derivatization with GC/ MS detection (e.g.,104,107). Comparison with HPLC/MS and GC/MS methods for five samples showed comparability of 97 and 89%, respectively. The IC/ECD method shortens the sample preparation and analytical time. Accuracy, precision, MDLs, and comparability with other methods of these newly developed techniques still need to be documented. Laboratory tests showed a precision of +-20% and an MDL of 0.01 ng/m^sup 3^ can be achieved for levoglucosan using IC/ECD.
Organic Speciation. In-injection port TD-GC/MS offers a fast, simple, and sensitive approach for organic speciation of PM samples.84,132 A quartz-fiber filter punch (0.3-5 cm2) is heated to approximately 275 [degrees]C and the desorbed organic compounds are passed through a GC/MS system for identification and quantification of approximately 130 nonpolar organic compounds. This method can be applied on the remnants of archived quartz-fiber filters (from the speciation sampler) after ionic and carbon analyses for single or multiple samples. It does not require an additional sampling channel or a separate sampling system with a higher flow rate.132-134 Current TD-GC/MS methods analyze nonpolar compounds (i.e., n- alkanes, iso/anteiso-alkanes, branched alkanes, cyclohexanes, hopanes, steranes, alkenes, phthalates, and polycyclic aromatic hydrocarbons (PAHs).84,135 The TD approach is cost-effective, does not require sample pretreatment, and avoids the use of organic solvents.134,136-138 It minimizes sample contamination, reduces uncertainties from extraction ef- ficiencies,139-141 and increases sensitivity.84,132,142,143
Figure 2 shows examples of chromatograms from TDGC/ MS for vegetative burning (grass, mesquite wood) and cooking (charcoal hamburger and smoked chicken) samples. Abundances of individual peaks show different patterns by different sources.144 A lower C^sub max^ number (the n-alkane that has the highest concentration among the n-alkane homologs) is found in cooking profiles (e.g., n-C^sub 23^ for charcoal hamburger and n-C^sub 25^ for smoked chicken) than in vegetative burning profiles (e.g., n = C^sub 29^ for mesquite wood burning and n = C^sub 31^ for grass burning). The carbon preference index (CPI; the ratio of oddto even-number n-alkanes)145 is approximately 2 for the vegetative burning and close to unity for cooking emissions. Figure 2 also shows that some polar compounds are not well resolved; this is apparent in the grass burning and charcoal hamburger cooking profiles. Other source markers, such as hopanes, steranes, and PAHs acquired from TD-GC/MS can also be obtained for source apportionment. A needed advancement is pattern recognition software that can extract consistent chromatographic features for different sources, even when they cannot be associated with a specific compound. Table 5 compares analytical specifications and shows that PAHs by TD-GC/MS agree within +-5% with other methods, with a maximum accuracy of +-29.4%, compared with approximately +-22% by solvent extraction. Better precision is found for TD (+-3.2%) than solvent extraction (+-23%). Ho and Yu132 showed low MDLs for n-alkanes (0.061- 0.97 ng/m^sup 3^), hopanes (0.030 – 0.14 ng/m^sup 3^), steranes (0.018 – 0.063 ng/m^sup 3^), and PAHs (0.016 – 0.48 ng/m^sup 3^). Jeon et al.137 showed that longchain organic acids by TD-GC/MS gave reasonable precisions (+-10-29%). Interferences for TD-GC/MS include loss of high-polarity analytes due to adsorption onto the surface of the injector and incomplete desorption due to strong binding with filter matrices.84,132 Interferences for solvent extraction include contamination, loss of volatile compounds, incomplete extraction, and incomplete derivatization.
The recovery and reproducibility in quantification of polar compounds such as n-carboxylic acids by TDGC/ MS are currently semiquantitative. Solvent extraction followed by derivatization with GC/MS detection quantifies polar organic compounds (e.g., organic acids, polyols, and sugars)146-148 with accuracy and precision of +- 4-8% and +-10-30%, respectively. MDLs range from 0.01- 0.03 ng/ m^sup 3^, but higher MDLs of 10 ng/m^sup 3^ for levoglucosan and 8.3 ng/m^sup 3^ for oxalic acids have been reported.149,150
Comparability between solvent extraction and TDGC/ MS is within +- 5-20% (e.g., n-alkanes [88 -105%], hopanes [99 -105%], steranes [95- 100%], and PAHs [80-115%]). Lower correlations are often found for polar organic species (e.g., R^sup 2^ = 0.73 for the organic acids).137
COMPARABILITY OF INTEGRATED FILTER SAMPLERS
Peters et al.151 summarize field comparisons (1996-1999) of eight PM^sub 2.5^ FRM samplers in eight cities (i.e., Bakers- field, Azusa, and Rubidoux, CA; Phoenix and Tucson, AZ; Denver, CO; Research Triangle Park, NC; and Birmingham, AL). The collocated precision for these FRM samplers ranged from 1.5 to 6.2%, on the basis of the average of daily coefficients of variation (CV)152 and range from -1.2 to +3.2% with respect to an FRM audit sampler used as a benchmark.
Comparability (R^sup 2^ = 0.98, slope = 1.03, precision = ~3%) was also found for collocated FRMs (i.e., RAAS 100 and R&P 2000) at Atlanta.31 Chow et al.38 reported similar findings (R^sup 2^ = 0.98, slope = 0.91 to 0.95, intercept 0.34- 0.64 [mu]g/m^sup 3^) for collocation of identical samplers at Fresno (i.e., between the two RAAS-300 or two R&P-2000 FRMs). Between-instrument variations (e.g., R&P vs. RAAS FRM) also showed reasonable comparability (slope = 0.87 to 0.91, intercepts = 1.11 to 1.18 [mu]g/m^sup 3^)38 with an average difference of +-4%.
Compared with the FRM, the dichotomous sampler (Dichot) yielded approximately 12% higher PM^sub 2.5^ at Atlanta, 31 10% lower PM^sub 2.5^ at Birmingham, 3-6% lower PM^sub 2.5^ at Phoenix and Tucson,153 and approximately 7-10% lower PM^sub 2.5^ at several locations in California.38,154 In the western United States, NO^sub 3^ ^sup -^ is a major atmospheric constituent. John et al.155 demonstrated that the dichotomous sampler’s anodized aluminum inlet effectively removes HNO^sub 3^, and Hering and Cass156 found similar HNO^sub 3^ removal in the inlets of FRM samplers. Differences in PM^sub 2.5^ mass between the FRM and dichotomous sampler may be related to differences in abundances of ambient HNO^sub 3^ and NO^sub 3^ ^sup – ^ and denuder efficiencies, gasto- particle ratios, and pressure drops through the sampler. 157,158 Even complex thermodynamic equilibrium modeling based on comprehensive precursor gas and particle measurements along with meteorological variables (e.g., temperature, relative humidity [RH]) cannot adequately estimate these evaporative losses.159-163 The most straightforward method is to directly measure both nonvolatilized and volatilized NO^sub 3^ ^sup -^ using an upstream HNO^sub 3^ denuder, a particle filter, and an HNO^sub 3^- absorbing backup filter.
PM^sub 10^ mass by MOUDI was approximately 20% lower than that of the dichotomous sampler (possibly due to loss of coarse particles in the MOUDI sampler), but PM^sub 2.5^ agreed within 3% of the dichotomous sampler in Pittsburgh. 164 PM^sub 2.5^ mass from the dichotomous sampler was approximately 17% lower than MOUDI in Los Angeles,165 attributed to NO^sub 3^ ^sup -^ volatilization. At a low flow rate (2 L/min), Williams et al.166 showed that a PM^sub 2.5^ personal environmental monitor (PEM; MSP, Inc.) used for exposure assessments167-169 measured 16-18% higher mass than a collocated FRM. Table 12 shows that various FRM measurements were within 10% of each other, whereas FRM was within +-10-20% of non-FRM PM^sub 2.5^ mass. The differences between the different samplers are due to differences in: (1) sampler design (with or without denuder); (2) inlet cutpoints and characteristics29,170; (3) sampling surfaces; (4) sampling environments31; (5) loss of semivolatile species (including NO^sub 3^ ^sup -^ and semi-volatile organic compounds [SVOCs]); (6) adsorption of gases by filter media171; and (7) filter handling and processing.172
Comparisons of eight integrated samplers at Atlanta showed that elemental concentrations were within 15- 20% of each other, except for R&P-2300 and ARA-PCM samplers.31 The variability in these two samplers was attributed to differences in inlet cutpoints and GSDs, and possibly different calibrations for the laboratory XRF analyzers.
Energy-dispersive XRF has been widely used to determine PM elemental concentrations. It requires minimal sample handling, can determine a wide variety of elements, is nondestructive, and has low detection limits.61 XRF and ICP/MS are complementary methods. ICP/ MS is useful in addition to XRF when rare-earth elements (lanthanide series) or light elements (Li, Be, and B) are sought at low levels. It can use half or whole filters, minimizing uncertainties due to inhomogeneous sample deposits. There is increasing interest in acquiring multiple elements by high-sensitivity ICP/MS for health studies.173-181 The disadvantages of ICP/MS include: (1) labor intensive operation, (2) use of strong acids, (3) potential contamination during sample extraction, (4) incomplete extraction efficiencies, and (5) spectral interferences from the Argon carrier gas and the extraction acids. MDLs are 0.4-30 ng/m^sup 3^ for XRF61 and 0.004-25 ng/m^sup 3^ for ICP-MS.181
Transition metals (e.g., chromium [Cr], iron [Fe]) in specific valence states have been identified as hazardous PM components. Their comparative toxicities need to be explored across the relevant health outcomes.182 A method of extracting Fe from PM and quantifying Fe (II) and Fe (III) has been developed that extracts Fe from Teflon filters using an acetate buffer solution, complexes the Fe (II) with Ferrozine, and quantifies absorption at 562 nm.183 Fe (III) is reduced to Fe (II) by the addition of a hydroxylamine hydrochloride solution. Fe (III) is determined by the difference between the total and initial Fe (II) concentrations.
Anions and Cations
The 1979 HNO^sub 3^ 184 and 1985 nitrogen species method comparisons156,185,186 documented the importance of gasparticle equilibrium for nitrogen species. At Atlanta, 12 integrated PM^sub 2.5^ samplers reported comparability of 30- 35% for NO^sub 3^ ^sup – ^.31 Using a Teflon-membrane front filter (nonvolatilized NO^sub 3^ ^sup -^) with a nylon backup filter (volatilized NO^sub 3^ ^sup – ^), John et al.155 found that volatilized NO^sub 3^ ^sup -^ accounted for approximately 90% of total particulate NO^sub 3^ ^sup – ^ (PNO^sub 3^ ^sup -^ = nonvolatilized NO^sub 3^ ^sup -^ = volatilized NO^sub 3^ ^sup -^) during daytime and approximately 60% during nighttime at Claremont, CA. Hering and Cass156 determined that PM^sub 2.5^ nonvolatilized NO^sub 3^ ^sup -^ from Teflonmembrane filters was 80-90% lower during daytime and 40-60% lower during nighttime than that measured by MgO-denuded nylon filter (i.e., PNO^sub 3^ ^sup -^). Volatilized NO^sub 3^ ^sup -^ loss from quartz-fiber filters accounted for less than 10% of PNO^sub 3^ ^sup -^ in winter and more than 80% during summer in central California.187 Experiments in Mexico City188,189 showed that NO^sub 3^ ^sup -^ volatilization was less than 10% during the early morning (12:00 to 6:00 a.m.) and late evening (6:00 p.m. to 12:00 a.m.) period with the highest NO^sub 3^ ^sup -^ volatilization (20- 40%) found during the afternoon period (12:00 to 6:00 p.m.) when ambient temperatures were high. Chow et al.187 also found that the nonvolatilized NO^sub 3^ ^sup -^ on quartz-fiber filters collected by FRM samplers were similar to NO^sub 3^ ^sup -^ on the denuded front quartz-fiber filters measured by the denuded DRI-SFS, con- firming that the inlet and sampling surfaces in the FRM sampler served as an effective HNO^sub 3^ denuder.156 Volatilized NO^sub 3^ ^sup -^ needs to be accounted for as part of PM^sub 2.5^ mass. Chow et al.187 estimated that volatilized ammonium nitrate (NH^sub 4^)NO^sub 3^) accounts for 32-44% of actual PM^sub 2.5^ mass (i.e., measured mass + volatilized NH^sub 4^)NO^sub 3^) during summer in Fresno. Similar mass losses of 10-20% were found for the IMPROVE network.32 At Pittsburgh, the MOUDI sampler (with Teflon filter) underestimated PNO^sub 3^ ^sup -^ concentrations by approximately 70% (~0.5 [mu]g/m^sup 3^) compared with the CMU sampler (denuded Teflon with backup nylon filter) because of NO^sub 3^ ^sup -^ volatilization under the MOUDI’s high-pressure gradients. 164 Similar underestimations of PNO^sub 3^ ^sup -^ by MOUDI (~54-80%) and FRM (~37-73%) were observed at Atlanta when compared with nine samplers with denuders and backup filters.31
In addition to differences in ambient temperature, aerosol composition, sampler configuration, and flow rate, removal efficiencies of HNO^sub 3^ using different chemical coatings have not been sufficiently verified. Possible contamination of the filter downstream by the addition of glycerol (to lower freezing point and to retain moisture) as part of the impregnating solution and the reaction of Na^sub 2^CO^sub 3^ (part of the denuder coating) with nitric oxide (NO) and nitrogen dioxide (NO^sub 2^) need to be further explored.
Comparisons among 12 filter samplers at Atlanta reported high correlations (R^sup 2^ > 0.90) for SO^sub 4^ ^sup 2-^, and 5-10% precision with the exception of the MOUDI.31 This precision is similar to the 6-7% range156,190 found during the 1987 South Coast Air Quality Study (SCAQS). Irrespective of season, MOUDI SO^sub 4^ ^sup 2-^ was consistently 13- 20% lower than SO^sub 4^ ^sup 2-^ from other speciation samplers at Atlanta31 and Los Angeles.191 Although MOUDI SO^sub 4^ ^sup 2-^ at the Pittsburgh site was similar to that of the CMU sampler during winter (Y = 0.97X), the correlation was low (R^sup 2^ = 0.48).164 Differences in inlet cutpoints and particle bounce in the MOUDI, variations in flow rates, and higher uncertainty for MOUDI stages (e.g., separate analyses per stage and sum of different stages to obtain PM^sub 2.5^ concentrations) may account for some of these discrepancies.
Few comparisons were made for filter NH^sub 4^ ^sup +^ measurements. Pooled comparisons at Atlanta showed that particulate NH2^sub 4^ ^sup +^ concentrations from different filter samplers were within 10-15% of the all-sampler average.31 Differences may exceed 15% in regions with large NH^sub 4^)NO^sub 3^, compared with ammonium sulfate [(NH^sub 4^)2SO^sub 4^] concentrations because NO^sub 3^ ^sup -^ volatilization also results in NH^sub 4^ ^sup +^ conversion to ammonia (NH^sub 3^) gas.
Carbon Fractions and Organic Speciation
Carbon measurements are subject to positive (adsorption of organic vapors on quartz-fiber filters) and negative (evaporation of semi-volatile compounds) sampling artifacts. 192-201 Solomon et al.31 reported variations of 20- 50% for OC and 20-200% for EC using 11 filter samplers and four analytical protocols (Table 16). Subramanian et al.48 and Chow et al.40 examined organic artifacts by placing a backup quartz-fiber filter behind the front Te- flon- membrane (QBT) or quartz-fiber (QBQ) filter. In the nondenuded channels, the positive artifact from QBT (24-34%, up to 4 [mu]g/ m^sup 3^ OC) was nearly twice that from QBQ (13-17%) for both Pittsburgh and Fresno samples (Table 16). With preceding organic denuders, the negative artifact was 5-10%,40,48,202 lower than the 40% reported by Eatough et al.203 Although Subramanian et al.48 suggested that equilibrium is attained in 24-hr samples at Pittsburgh, Chow et al.40 found equilibrium differences throughout the day at Fresno. Denuder efficiency was tested at Fresno202 and no difference was found between an XAD-coated denuder and a parallel plate denuder using carbon-impregnated charcoal filters (CIF) with frequent denuder changes. Huebert and Charlson204 observe that changes in temperature and pressure using tandem filter packs may hinder a quantitative analysis of the artifacts. For a single filter configuration, Huebert and Charlson204 estimated 30-50% and 50% for positive and negative artifacts, respectively.
COMPARABILITY OF INTEGRATED FILTER AND CONTINUOUS/SEMICONTINUOUS SAMPLERS
In situ continuous mass and speciation monitors allow temporal profiles to be observed, thereby enabling a better understanding of PM^sub 2.5^ formation, sources, transformation, transport, and sinks.7,205 Brief descriptions of these instruments are given in Tables 6-11 in the supplemental material online.
Continuous mass measurements have been proven to be sensitive, reliable, and easy to operate. MDLs are 0.06 [mu]g/m^sup 3^ for tapered element oscillating microbalance (TEOM) and 5 [mu]g/m^sup 3^ for BAM and the continuous ambient mass monitor (CAMM). Collocated precisions are within +-10% for TEOM and BAM and +-30% for CAMM (Table 6). Development of the SES-TEOM (TEOM with a sample equilibration system, consisting of a Nafion dryer at the inlet), D- TEOM (TEOM with electrostatic precipitator [ESP]), and FDMS (filter dynamics measurement system; TEOM with 4 [degrees]C filters) allow the estimation of SVOCs while minimizing thermal expansion of the tapered elements and effects of RH. Field tests showed that SES- TEOM still loses some semi-volatile species.206-212 Standard TEOMs underestimated PM^sub 2.5^ FRM mass by 20- 30% at Atlanta,213 Fresno,38 and Houston,214 regardless of whether the oscillating element was maintained at 50 [degrees]C38,210,211,214,215 or 30 [degrees]C.211,213,215 At Fresno, the 50 [degrees]C-TEOM reported approximately 20% lower mass than the 30 [degrees]C-TEOM, whereas the 30 [degrees]C-TEOM was approximately 50% lower than the FDMS.215 Compared with the FRM mass during summer, the FDMS reported approximately 30% higher PM^sub 2.5^ at Rubidoux, CA,211 and approximately 25 and 9% higher at urban and rural sites, respectively, in New York.210 The SES-TEOM was comparable (within 2%) to the FRM mass at Atlanta216 and Houston.214 However, at Pittsburgh, Rees et al.217 found that PM^sub 2.5^ volatilization resulted in approximately 50% lower mass than that from a collocated FRM for a SESTEOM under low (<-6 [degrees]C) ambient temperature with low (<20 [mu]g/m^sup 3^) PM^sub 2.5^ concentrations. The SES- TEOM loses semi-volatile material in Houston.214,216 High correlations (R^sup 2^ = 0.94-0.99) and near unity (~0.9) slopes were found for collocated D-TEOMs at Queens, NY, during July 2001218; at Rubidoux, CA, during August-September 2001218; at Claremont, CA, during February-August 2002165; and at Fresno, CA, during December 2003.215 The Claremont comparison showed that D- TEOM PM^sub 2.5^ was approximately 18% higher than that from the MOUDI (R^sup 2^ = 0.86).
Motallebi et al.154 reported good comparisons between BAM and FRM PM^sub 2.5^ in central California (slope = 0.91-0.97, intercept = 0.8-3.25 [mu]g/m^sup 3^, R^sup 2^ = 0.96-0.98). At Fresno, the PM^sub 2.5^ BAM (without an inlet heater) correlated with the FRM but overestimated the FRM PM^sub 2.5^ mass by approximately 30% from 1999 to 2003. Chow et al.38 attributed this to water absorption by hygroscopic species. Chung et al.219 reported +-5% comparability in PM^sub 2.5^ mass between a BAM (with a heated inlet to keep RH <60%) and an FRM (R^sup 2^ = 0.99) at Bakersfield, CA, during January 1999; FRM NO^sub 3^ ^sup -^ losses were less than 10% during winter.187 Using a smart heater (maintaining RH <45%) at the urban New York site during 2004, Schwab et al.210 showed that a BAM and a FDMS agreed within 2%, although measuring approximately 28% higher mass than the collocated FRM, similar to the approximately 30% higher mass reported by Chow et al.38 at Fresno. Chang et al.220 demonstrated that the BAM was affected by water absorption of hygroscopic aerosols, but some of the water evaporated because of heating by the beta source. Chang and Tsai221 further showed that the absorbed water may volatilize completely at RH less than 85%. Using inlet heaters to control RH may minimize water absorption but can also result in loss of SVOCs and NH^sub 4^NO^sub 3^.
Field tests of the CAMM showed inconsistent results. 216,222,223 Collocated precisions for the CAMM ranged from 15.9% (24-hr average) to 28.1% (1-hr average) for 14 days at Philadelphia (R^sup 2^ = 0.75- 0.88). Poor precision was attributed to an inconsistent sealing mechanism, leaks, and a variable baseline.224 Reasonable CAMM correlations (R^sup 2^ = 0.80-0.89) and agreements (within 2%) were found between the SES-TEOM and FRM at Houston.223 However, at Rubidoux no correlation was found between the CAMM and the D-TEOM or the FRM PM^sub 2.5^ during August-September 2001.222 The CAMM showed little temporal variability at Rubidoux, indicating the lower sensitivity of the CAMM to SVOCs.222
Collocation of two NGN nephelometers (heated to maintain RH <60%) at Bakersfield, CA during December 1998 showed good agreement for dry particle scattering (bscat; slope = 1.01, intercept = -1.64 [mu]g/m^sup 3^, R^sup 2^ = 0.99).219 Good precision (5%) was also reported by Andersen and Ogren225 for a three-color TSI nephelometer. The Dust- Trak overestimated FRM PM^sub 2.5^ by three- fold with the factory calibration to Arizona road dust. For dp < 0.25 [mu]m, the amount of light scattered by the DustTrak is proportional to the sixth power of the particle diameter (dp 6), causing an overestimation of mass for small particles. 219 Chow et al.39 noted that conversion factors from particle scattering to mass concentrations (mass scattering efficiencies) varied by site and by season in the range of 2-6 m2/g. Nephelometers should be collocated with filter measurements to derive site- and season-specific conversion factors. The use of a smart heater (e.g., Radiance Research M903; Ecotech M9003) resulted in better correlation between bscat and PM^sub 2.5^ mass.39 Chow et al.39 recommended deployment of relatively inexpensive and portable nephelometers to a large spatial monitoring network to capture diurnal variations and to estimate PM^sub 2.5^ between every third- and sixth-day 24-hr filter samples. Details on PM^sub 10^-2.5 mass measurements using TEOM, BAM, and SMPS-APS are found in Peters226 and Solomon and Sioutas.7 Particle Sizing
Measurements of particle number and size distributions were taken over multiyear periods at the Supersites227-233 using SMPS, optical particle counters (OPCs), and aerodynamic particle sizers (APS). The APS was modified with respect to previous versions to eliminate artifacts from larger particles (>10 [mu]m).234-236 Nano- differential mobility analyzers (nano-DMAs) have been integrated with the SMPS, allowing the detection of dp >/= 3 nm.237 Four collocated SMPSs were tested at Fresno: (1) TSI Model 3935 SMPS; (2) TSI Model 3936 nano-SMPS (TSI, Inc.); (3) Grimm Model 5.400 SMPS (Grimm Aerosol Technik GmbH & Co. KG); and 4) MSP Wide-Range Particle Spectrometer (WPS, 10 nm to 10 [mu]m, couples the SMPS and an OPC into one instrument; MSP Corp.). The two TSI instruments agreed within 5%, and the GRIMM and TSI nano-SMPS were indistinguishable from each other in the 30- to 50-nm size range. The WPS and TSI nano-SMPS agreed within 30%.238 For a laboratory comparison using diluted diesel exhaust, the GRIMM SMPS agreed within 20% of the TSI nano-SMPS, and the MSP WPS agreed within 60% of the GRIMM SMPS and TSI nano SMPS.239
A water-based condensation particle counter (CPC; Models 3785 and 3786, TSI, Inc.)240-242 has shown similar performance to conventional butanol based-CPCs243 while avoiding the use of a toxic organic solvent. Fastresponse instruments (e.g., the Engine Exhaust Particle Sizer [EEPS; 10 Hz], the Fast Mobility Particle Sizer [FMPS; 1 sec])244,245 have been used for source testing of exhaust emissions. Mohr et al.246 show that fast-response particle measurement methods are feasible for detecting low vehicle emissions with better precision and lower MDLs than gravimetric analysis of integrated samples. Thermal denuding and/or hot dilution is needed to minimize nucleation of volatile species in the sampling line. Nonmassbased continuous instruments (e.g., number- or surfacearea methods) may be needed for low-level mass determination. Dosimetry- based ambient PM metrics and standards have been proposed to relate PM data with human exposure.247
Elements and Particle Water Content
The Semicontinuous Elements in Aerosol Sampler (SEAS)248,249 has been tested for semicontinuous measurement of 7-12 elements. It acquires time-resolved samples in the field, but these need to be taken to a laboratory for analysis. The Dry Ambient Aerosol Size Spectrometer (DAASS) has been used to measure in situ particle water content. It showed that particles may be hydrated even at ambient RH below 30%, and is useful for identifying “dry” and “wet” particles.250 Some limited comparisons are shown in Table 13.
Anions and Cations
NO^sub 3^ ^sup -^. Seven different single and multiple ion continuous monitors are commercially available.251,252 MDLs for hourly sampling are in the range of approximately 0.5 [mu]g/m^sup 3^ for the thermal-gas method (i.e., heating the sample stream, followed by gas determination) and 0.1 [mu]g/m^sup 3^ for the IC- based methods (Table 8). These MDLs are 2-10 times higher than those obtained in the laboratory (Table 4).
For measurements from the Five-Cities study (Seattle, Phoenix, Houston, Indianapolis, and Chicago), Vaughn et al.253 estimated the hourly NO^sub 3^ ^sup -^ precision (expressed as CV) of the R&P- 8400N (Thermo Electron) in the range of from 1.8-6.9%, with an average of 5.2%. This precision increased to 6.3-23% (average 8.7%) for 10-min averages, including flow rate uncertainties (2%), NO^sub 3^ ^sup -^ to oxides of nitrogen (NOx) conversion efficiency (4.4%), reaction cell pressure correction (0.6%), calibration error (2.6%), and peak integration uncertainty (~4%).254
Hourly NO^sub 3^ ^sup -^ averaged over 24 hr agreed within 20- 30% of the filter NO^sub 3^ ^sup -^ measurements except for the ARAN252 (Table 14). The R&P-8400N underestimated the filter NO^sub 3^ ^sup -^ concentration by 20-45% at the Fresno,37 Atlanta, 252 Pittsburgh,47 Baltimore,254 and New York251,255 Supersites. Size- segregated continuous NO^sub 3^ ^sup -^ (ADI-N) was approximately 20% less than filter NO^sub 3^ ^sup -^ in Los Angeles. 191 Discrepancies were less pronounced during early morning hours (12:00 to 10:00 a.m.) in Bakersfield.37 Harrison et al.254 found that differences were smaller for lower temperature differences between the instrument and ambient air. Although potassium nitrate (KNO^sub 3^) is recommended by the manufacturer for calibration of the R&P- 8400N, Rattigan et al.255 observed that the conversion efficiency was lower when using a combination of KNO^sub 3^ and (NH^sub 4^)2SO^sub 4^, suggesting that the aerosol composition (matrix) influences NO^sub 3^ ^sup -^ to NOx conversion. Data adjustments (for conversion efficiencies, flow rate, vacuum drift of reaction cell pressure, instrument blank, calibration drift, etc.) were insufficient to establish mass closure.47,254 Vaughn et al.253 found that the extent of underestimation increased at higher NO^sub 3^ ^sup -^ concentrations (>3 [mu]g/m^sup 3^). Harrison et al.254 and Hogrefe et al.251 found that the use of ridged flash strips (as opposed to flat strips) in the R&P-8400N at Baltimore and New York, respectively, lowered the dissociation losses.
Comparisons in New York showed that the particleinto- liquid sampler-IC (PILS-IC) measured approximately 37% more NO^sub 3^ ^sup – ^ than the R&P-8400N, and was within 10% of the Na2CO3-denuded nylon filter-based NO^sub 3^ ^sup -^ from a Partisol 2300 speciation sampler.251 Grover et al.215 reported reasonable comparisons (~5% bias, slope = 0.71, intercept = 3.2 [mu]g/m^sup 3^, R^sup 2^ = 0.91) between Dionex-IC and PC-BOSS measurements at Fresno. The Dionex-IC NO^sub 3^ ^sup -^ was higher than the PC-BOSS at lower concentration ranges.
Field comparisons showed that data recovery for the R&P-8400N was above 80%.47,251,254-256 The thermal-gas methods require collocated filter measurement for calibration and adjustment.47,254 Overall, continuous instruments were within 10% (for IC-based) to 45% (thermalgas methods) of filter NO^sub 3^ ^sup -^ measurements and 30- 40% between continuous instruments.
SO^sub 4^ ^sup 2-^. During the Five-Cities Study,253 the precision of the R&P-8400S varied from 2.3-17.2% (average 8.3%). Rattigan et al.255 estimated uncertainties of 25% for 10- min average measurements, ranging from greater than 30% at low (<2 [mu]g/ m^sup 3^) to less than 15% at high (>9 [mu]g/m^sup 3^) concentrations.
Similar to NO^sub 3^ ^sup -^, SO^sub 4^ ^sup 2-^ from the flash volatilization method reported approximately 30% lower concentrations than collocated filter concentrations at Pittsburgh.47 At Fresno, Grover et al.215 reported a low correlation (R = 0.68) but an adequate regression (slope = 0.95, intercept = 0.3 [mu]g/m^sup 3^) for the R&P-8400S and Dionex-IC during the low (<2 [mu]g/m^sup 3^) concentration winter period. Concurrent Dionex-IC and PC BOSS measurements agreed within +-6% (R^sup 2^ = 0.98).215
At New York, the continuous ambient sulfate monitor (CASM; prototype of the TE-5020) underestimated filter SO^sub 4^ ^sup 2-^ by approximately 15% (24-hr average) to approximately 30% (6-hr average).251,257 The continuous instruments performed better (2-6% precision, R^sup 2^ = 0.87-0.94) at the urban site than at the rural site, though the extent of underestimation increased for SO^sub 4^ ^sup 2-^ above 5 [mu]g/m^sup 3^. Although the R&P-8400S, aerosol mass spectrometer (AMS), and CASM were all 10-30% lower than filter measurements at urban and rural sites, the PILS-IC was lower (17- 28%) at the urban site, and higher (7-11%) at the rural site.251 The underestimation of continuous SO^sub 4^ ^sup 2-^ was attributed to inlet and transport losses and incomplete collection/conversion of different forms of SO^sub 4^ ^sup 2-^ to a detectable form.47,257
The TE-5020 also reported 20% lower SO^sub 4^ ^sup 2-^ than collocated filter measurements in New York, but it was comparable to filter SO^sub 4^ ^sup 2-^ at St. Louis. Schwab et al.258 postulated that: (1) a systematic error occurred in the instrument, or (2) SO^sub 4^ ^sup 2-^ in St. Louis was more easily converted to SO2 than in New York. Laboratory evaluations confirmed that the conversion efficiencies for different salts were: 0.78 +- 0.12 for (NH^sub 4^))2SO^sub 4^, 0.63 +- 0.05 for potassium sulfate (K2SO^sub 4^), 0.20 +- 0.04 for sodium sulfate (Na2SO^sub 4^), and 0.04 +- 0.02 for calcium sulfate (CaSO^sub 4^).258 This suggests that the TE- 5020 instrument may underestimate SO^sub 4^ ^sup 2-^ concentrations in regions influenced by sea salt or mineral dust and where some SO^sub 4^ ^sup 2-^ may be in the coarse mode. Further laboratory tests showed conversion efficiencies ranging from 81 to 85%. Interferences from organic aerosols (maximum of 0.8% interference from succinic acid) and NO^sub 3^ ^sup -^ (maximum of 1.1% from KNO^sub 3^) aerosols were minimal.
Similar conversion efficiencies were found for R&P- 8400S SO^sub 4^ ^sup 2-^. Rattigan et al.255 noted lower conversion efficiencies when less than a four-fold excess of oxalic acid over (NH^sub 4^))2SO^sub 4^ was used during calibration. The R&P- 8400S reported data recovery rates of 84-95% and was subject to frequent (every fourth to sixth day) flash strip failures and/or vacuum pump malfunction.47,251,255-257 It required weekly or biweekly maintenance by trained personnel, making it more labor-intensive than the TE-5020 (data capture of >88%).258
NH^sub 4^) ^sup +^. At Atlanta, NH^sub 4^) ^sup +^ by PILS-IC and the ECN were approximately 5% lower than the all-sampler average, including 10 integrated filter and three continuous instruments. 259 Recoveries similar to those for NO^sub 3^ ^sup -^ and SO^sub 4^ ^sup 2-^ (65-70%) were found by the PILS-IC at New York. The instrument experienced tubing failure (e.g., wider tubing resulted in large aqueous flow, which increased dilution), losses of NH^sub 4^) ^sup +^ in the Teflon tubing, DDW contamination, presence of air in the system, and corrosion of parts.251 Carbon
OC and EC. Good collocated precision (R^sup 2^ = 0.97-0.98) was found for the Sunset OCEC instrument at Los Angeles (December 2004- May 2005). The discrepancies in regression slope (0.82 +- 0.02) and intercept (0.2 +- 0.04 [mu]g/m^sup 3^) were attributed to a 17-sec delay for the OC/EC split in one of the instruments.260 Bae et al.44 reported Sunset precisions of approximately 10% for OC and approximately 14% for EC in St. Louis. Lim et al.261 reported collocated precision (pooled CV) of 13% for OC (ADI-C, R&P-5400, RU- OGI), 26% for EC (particle soot absorption photometer [PSAP], AE-16 aethalometer, R&P-5400 and RU-OGI) and 7% for TC (R&P-5400 and RU- OGI) in Atlanta.
In addition to different OC/EC split times, different temperature programs were used at different sites, resulting in wide variations (Table 16) in EC comparisons. The R&P-5400 showed good agreement (average ratio of 1.02, slope = 0.96, R^sup 2^ = 0.83) with the RU- OGI carbon analyzer for TC, although differing in the OC/EC split at Atlanta. 261 R&P-5400 and filter sample carbon were not comparable at Fresno, where R&P TC was 40-60% higher than filter TC.202,262 At New York, the R&P-5400 measured approximately 23% lower TC (R^sup 2^ = 0.83) than filter TC.263 Without adjustments for OC artifacts, variations in denuders and measurement protocols, the differences between continuous and filter measurements were 28% for OC, 66% for EC, and 25% for TC. Continuous BC (by optical methods) or EC (by thermal methods) was 1.2-2.5 times higher than filter EC at Atlanta and Baltimore.261,264
Agreement (~7% precision, R^sup 2^ = 0.9 for OC, 0.6 for EC) between Sunset OCEC and denuded filter samples was observed at St. Louis using the same temperature protocol and analysis method (TOT).45 At Baltimore, different temperature protocols were used for Sunset OCEC (870 [degrees]C) and filter analysis (920 [degrees]C) for OC by TOT. Denuded Sunset OCEC reported 22% lower OC and 11.5% lower EC compared with the nondenuded filter samples.264 In addition to the difference in temperature settings, the absence of an upstream denuder in the speciation sampler may have contributed to higher OC on the filter. At New York, the Sunset OCEC (with denuder) was approximately 25% lower for OC, 21% higher for EC, and approximately 14% lower for TC than the nondeunded filter samples.263
As summarized by Park et al.,265 filter-based optical instruments (i.e., BC) may be subject to multiple scattering effects by sampled particles and the filter matrix, resulting in an absorption enhancement. Empirical corrections have been proposed for such effects.266-268 Park et al.265 noted that BC by optical and photoacoustic methods and EC by thermal methods are highly correlated but have different absolute values. Particle- and EC- absorption efficiencies are both wavelength dependent. The assumption that these efficiencies scale with inverse wavelength (the Angstrom Power Law [lambda^sup -alpha^] assuming alpha = 1) was shown to be untrue at Fresno265 and other locations. 269 By comparison with IMPROVE_TOR EC, sitespecific absorption efficiencies were 5.5 m^sup 2^/g for the multiangle absorption photometer (MAAP), 10 m^sup 2^/g for the aethalometer (880 nm), and 2.3 m^sup 2^/g for the photoacoustic instrument (DRI-PA), which differ from the 6.5, 16.6, and 5 m^sup 2^/g default efficiencies, respectively.265 Although the aethalometers and the PSAP correlated well (R^sup 2^ = 0.97) in Atlanta, BC differed by approximately 50%, con- firming the need to adjust the efficiencies used to convert filter transmittance to BC.261 Although aethalometer (AE- 16) BC was higher (~12%) relative to the TOT-based RUOGI analyzer at Atlanta,261 BC was lower by approximately 20-25% than IMPROVE_TOR-based filter EC concentrations at Fresno.202 Similar differences were found by Venkatachari et al.263 in New York, where aethalometer (AE-20) BC was approximately 14% lower than Sunset EC and STN_TOT-based filter EC. Large variations (over a factor of two) were found among continuous BC/EC measurements and between continuous and filter EC measurements. Park et al.265 observed that MAAP BC was within 18- 40% of IMPROVE_TOR and Sunset EC at Fresno. Although the photoacoustic instrument showed good correlation (R^sup 2^ > 0.80), it differed by more than 40% compared with the aethalometer, MAAP, and IMPROVE_TOR EC. These instruments require calibration as a function of sampling location and time.263 In addition, the BC absorption efficiency is not constant over the range of measured babs.261,265 The comparability also depends on the EC thermal and optical monitoring (TOR or TOT) protocols. Sunset optical BC was 80% of Sunset thermal EC at Fresno265 and approximately 58% of the Sunset thermal EC and approximately 62% of the aethalometer BC at New York.263
WSOC. Sullivan et al.270 combined the PILS with a total organic analyzer (TOA; i.e., PILS-WSOC) to measure total WSOC at St Louis with uncertainties of +-5-10%. These in situ measurements were compared with collocated filters extracted in DDW with a correlation of R^sup 2^ = 0.71. Higher WSOC with PILS-WSOC may be associated with sampling artifacts, incomplete extraction of organic compounds from filter, and differences in concentrations affecting the solubility of OC compounds.
Organic Species. Coupling TD with either a particle beam (PB)/ MS, chemical ionization (CI)/MS, or GC/MS has been used for in situ continuous organic speciation.271-277 Chamber-produced secondary organic aerosol (SOA)271-276 was analyzed for composition by TD-PB/ MS. Smith et al.278 used TD-CI/MS to characterize particle molecular compositions with dp of 6-20 nm. Inorganic substances such as (NH^sub 4^))2SO^sub 4^ and organic compounds such as oxalate and methanesulfonate were quantified. Using TD-GC/MS-FID, Williams et al.277 reported 100 nonpolar and polar compounds, but the MDLs for polar levoglucosan (12.1 ng/m^sup 3^) and decanoic acid (1.67 ng/ m^sup 3^) were 2-3 orders of magnitude higher than that for nonpolar chrysene (0.10 ng/m^sup 3^). Low regression slopes in the calibration of polar species raise uncertainties about their quantification. MDLs for nonpolar species, such as n-alkanes and PAHs, were in the range of 0.11-0.25 ng/m^sup 3^ and 0.10-0.14 ng/ m^sup 3^, respectively. Precisions for most measurable compounds ranged from +-0.05 to 11.5%. Collocated comparisons with integrated filter samplers followed by either solvent extraction- or TD-GC/MS is needed to better determine comparability.
Particle Mass Spectrometers
Particle mass spectrometers were developed in the mid- 1990s to provide real-time, size-fractionated PM chemical composition. The four most commonly used types are: (1) particle analysis by laser MS (PALMS; National Oceanic and Atmospheric Administration [NOAA])279; (2) rapid single particle mass spectrometer (RSMS; University of Delaware)280,281; (3) aerosol time-of-flight MS (ATOFMS; TSI, Inc.)282; and (4) AMS (Aerodyne).283 These instruments decompose individual particles and ionize their components followed by time- of-flight (TOF) or quadrapole MS detection. PALMS, RSMS, and ATOFMS use high-powered lasers to ablate the particles. The PALMS and RSMS use an ArF laser (193 nm) whereas the ATOFMS uses a Nd:YAG laser (266 nm). AMS fragments particles after impaction onto a heated filament and ionizes them by electron impact. Because AMS analyzes all particles that are impacted and ionized, it is an ensemble of particles, not a “single” particle method. The temperature of the impaction filament in the AMS typically ranges from 550 [degrees]C to 700 [degrees]C.251,283,284 Middlebrook et al.285 compared these four spectrometers collocated at the Atlanta Supersite. Three single particle instruments yielded similar particle types and abundance levels. The main particle types were OC with SO^sub 4^ ^sup 2-^, sodium with potassium and SO^sub 4^ ^sup 2-^, soot/hydrocarbon, and a mineral particle type. The mole ratio of continuous ion (e.g., NO^sub 3^-, SO^sub 4^ ^sup 2-^) instruments was compared with the AMS (mole ratio) and the RSMS-II (ion ratio) with good correlations (R^sup 2^ = 0.92 for AMS and R^sup 2^ = 0.71 for RSMS-II). A more detailed history of the development of the different particle mass spectrometers and their applications can be found in Suess and Prather,286 Noble and Prather,287 Sullivan and Prather,129 and Nash et al.288
Early reports on these spectrometers focused on qualitative patterns, such as detection of negative and positive ion mass spectra simultaneously,289 sulfur speciation,290 correlation between size and chemical composition,291 detection of airborne bacteria,292 detection of fireworks, 293 and observations of heterogeneous chemistry. 294 Mass spectrometers were operated at Fresno (ATOFMS), 295,296 Houston (RSMS-II),297 Pittsburgh (AMS and the RSMS-III),298- 300 Baltimore (RSMS-III),301,302 and New York (AMS) Supersites.251,303,304 Most findings were semiquantitative, with the exception that AMS quantified SO^sub 4^ ^sup 2-^, NO^sub 3^ ^sup – ^, and OM.298,305 These experiments acquired up to one million mass spectra of individual particles (~1-wk to 1-month sampling) and are labor intensive, requiring highly skilled personnel for operation and data interpretation. Wenzel et al.305 collocated an ATOFMS with sizing instruments (e.g., laser particle counter or SMPS) to estimate number concentrations of different particle types in Atlanta, as well as from light-duty gasoline306 and heavy duty diesel307 vehicles. The ATOFMS was collocated with MOUDI measurements to estimate mass.296,308-310 However, neither of these scaling techniques allows for comparison to bulk chemical species (other than total mass), because the fraction of each chemical species in each particle type is unknown. Spencer and Prather311 constructed a linear calibration curve for the determination of OC/ EC ratios, and a comparison with Sunset OCEC yielded a linear correlation (R^sup 2^ = 0.69). With additional research, a correlation between the mass spectral signatures and bulk speciated concentrations might be achieved. DATA ANALYSIS TECHNIQUES AND DEFINITIONS
Different definitions and methodologies have been used to evaluate accuracy, precision, validity, and comparability at the different Supersites, so the data summarized here must be considered approximate. EPA312 specifies the following criteria for equivalence of PM^sub 2.5^ mass:
Two methods are “equivalent” if the linear regression slope is 1 +- 0.05, intercept is 0 +- 1 [mu]g/m^sup 3^, correlation coefficient (r) is >/=0.97 (or R^sup 2^ >/= 0.94) and the collocated precision is 2 [mu]g/m^sup 3^ or 5%, (whichever is larger).
This is a stringent requirement achieved in a small number of tests, even for collocated FRMs. Methods are comparable when slope equals unity and the intercept equals zero within three standard errors, r is >0.90 (or R^sup 2^ > 0.81), and average ratios equal unity within one standard deviation; predictable if r > 0.90, but slope and intercept fail the requirements for comparability; and not related if r < 0.90.202,313
Most studies use correlation and linear regression of collocated measurements. Watson and Chow202 and Park et al.265 evaluated equivalence, comparability, and predictability on the basis of the above criteria to maintain consistency with EPA.312 With the exception of Peters et al.151 (using the EPA criteria to estimate equivalence), most researchers evaluate comparability on the basis of slopes and correlation coefficients derived from ordinary unweighted least squares (OLS) regression. Selection of variables as the x-axis or benchmark is also arbitrary. Conclusions are based on OLS with a forced zero-intercept, 211,215,251 OLS with intercept,202,210 and/or effective variance (EV) weighted least squared regression.202 These comparison metrics lead to different conclusions on comparability. For example, Grover et al.215 suggested that the Dionex-IC NO^sub 3^ ^sup -^ agreed with PC-BOSS with a zero-intercept slope of 0.98 and R^sup 2^ = 0.74, whereas OLS reported a slope of 0.71, an intercept of 3.2 [mu]g/m^sup 3^, and R^sup 2^ = 0.91, suggesting that the Dionex-IC overestimated NO^sub 3^ ^sup -^ at lower concentrations. Thus, multiple measures should be used to evaluate comparability.202 Some researchers de- fined differences in terms of absolute or relative bias whereas others presented inferences on the basis of parametric statistical tests. Collocated precisions are calculated differently.165,202,258 The regulatory definition for precision of mass measurements (see ref. 314, Appendix A, Section 5.5.2) is rarely used; and there are no evaluation criteria for chemical species. There is a need to establish a common set of conditions and parameters to estimate collocated precision, comparability, and predictability.
SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS
Advances in PM measurements are apparent since the reviews by Chow1 and McMurry.2 The VSCC (a PM^sub 2.5^ FEM inlet) eliminates particle overloading/penetration and minimizes frequent cleaning. Limited tests have been performed on a nano-BAM particle inlet (d^sub 50^ = 148 +- 10 nm) and on a high-volume slot virtual impactor (d50 of 13-200 nm) for UP measurements. Different inlets equipped with continuous mass and chemical composition measurements were compared with FRM and speciation samplers, but causes of discrepancies are still poorly understood. Examples of very good or very poor comparabilities can be cited.
Most integrated filter sampling systems have been proven robust and can be applied in urban and remote environments. Laboratory analytical precisions are 5-10% for mass, elements, ions, and total carbon. Lack of standardization results in poor reproducibility (10- 30%) for OC, EC, and organic species (e.g., n-alkanes, hopanes, steranes, PAHs, and WSOC). Many of these variations result from analytical uncertainties such as peak overlap, matrix interferences, spectral interferences, and lack of reference standards for EC and most organic species. Intersampler comparisons are 20% for mass, 30- 35% for NO^sub 3^ ^sup -^ and NH^sub 4^) ^sup +^, 10% for SO^sub 4^ ^sup 2-^, 20-50% for OC, and 20-200% for EC.
Although good collocated precisions (~10%) have been reported for PM^sub 2.5^ FRMs, they underestimate PM^sub 2.5^ mass by 10-40%, especially at locations with abundant NH^sub 4^)NO^sub 3^ and SVOCs. FRMs report 3-12% lower PM^sub 2.5^ mass than collocated dichotomous and MOUDI samplers. Comparisons of non-FRM samplers reported precisions of 10-20%.
With HNO^sub 3^ denuders, NO^sub 3^ ^sup -^ volatilization accounts for 80-90% of total PNO^sub 3^ ^sup -^ during daytime in summer. 155,156,187 The extent of volatilization is reduced to 40- 60% during summer nights156 and less than 10% during winter.187 With diurnal sampling, Chow et al.188 reported less than 10% NO^sub 3^ ^sup -^ volatilization during the early morning and late evening period whereas low ambient temperature favors the presence of particle phase NH^sub 4^)NO^sub 3^. Evaporative loss of NH^sub 4^)NO^sub 3^ negatively biases actual PM^sub 2.5^ mass (i.e., measured mass + volatilized NH^sub 4^)NO^sub 3^) measurements, especially in the western United States where NH^sub 4^)NO^sub 3^ is abundant.32,187 These losses cannot be estimated, even with complex thermodynamic equilibrium modeling. To evaluate negative biases, PM^sub 2.5^ measurements should be taken with preceding HNO^sub 3^ denuders followed by HNO^sub 3^ absorbent back filters during nonwinter periods.
Using tandem filter packs without preceding organic denuders, positive artifacts (24-34%) from QBT was nearly double that of QBQ (13-17%) at Pittsburgh and Fresno.40,48 With a preceding denuder, the negative artifact was 5-10%. There is no consensus regarding the extent of organic sampling artifacts