Simulating Particle Size Distributions over California and Impact on Lung Deposition Fraction

Abstract

Reliable simulations of particle mass size distributions by regional photochemical air quality models are needed in regulatory applications because the U.S. EPA's National Ambient Air Quality Standards specify limits on the mass concentration of particles in a specific size range (i.e., aerodynamic diameter <2.5 μm). Considering the associations between adverse health effects and exposure to ultrafine particles, air quality models may need to accurately simulate particle number size distributions in addition to mass size distributions in future applications. In this study, predictions of particle number and mass size distributions by the Community Multiscale Air Quality model with the standard and an updated emission size distribution are evaluated using wintertime observations in California. Differences in modeled lung deposition fraction for simulated and observed particle number size distributions are also evaluated. Simulated mass size distributions are generally broader and shifted to larger diameters than observations, and observed differences in inorganic and carbon (elemental and organic) distributions are not captured by the model. These model limitations can be reasonably accounted for in regulatory modeling applications. Simulated number size distributions are considerably less accurate than mass size distributions and are difficult to represent in air quality models due to large sub-grid-scale concentration gradients. However, modeled number size distributions are responsive to updates of the emission size distribution, and reasonable simulation of background number size distributions might be possible with an improved treatment of emission size distributions. Modeled lung deposition fractions for simulated number size distributions peak in the same lung region as those for number size distributions observed in the background. However, differences in modeled and observed total number concentrations generally suggest large differences in the total number of deposited particles. Future model development on simulating particle mass size distributions should focus on improving predictions of the mass fraction of particles <2.5 μm. Model development for particle number size distributions should focus on reducing differences in modeled lung deposition for modeled and observed distributions.

INTRODUCTION

The U.S. EPA's National Ambient Air Quality Standards (NAAQS) specify limits on observed mass concentrations of particulate matter with aerodynamic diameter <2.5 μm (PM_2.5; U.S. EPA 1997, 2006). The specification of diameter in this regulation reflects an understanding that fine particles (diameter <2.5 μm) are transported deep into the lung and have been correlated with adverse health effects (U.S. EPA, 2004). Some studies report that ultrafine particles (diameter <0.1 μm) of certain composition are more toxic than larger fine particles of the same composition (Karlsson et al. 2009; Oberdörster 2001). Atmospheric ultrafine particles dominate the particle number size distribution but contribute little to the particle mass or volume size distributions. Woo et al. (2001) found no correlation between fine-particle number and volume concentration in Atlanta, GA and suggested that strategies for attaining the PM_2.5 NAAQS may not protect human health if ultrafine particles are responsible for adverse health effects. Considering that adverse health effects may be associated with both ultrafine and fine particles, the ability of photochemical air quality models to simulate both particle number and mass size distributions must be evaluated and likely enhanced.

Reliable predictions of particle size distributions by air quality models are needed for reasons in addition to their relevance to health-based standards. Since light extinction by particles is size dependent, modeling regional haze and visibility impairment requires accurate estimates of particle size distributions (Ying et al. 2004). Particle settling velocity, and therefore dry deposition, is also size dependent, and so accurate size-distribution predictions are needed for determining the impact of particle-phase nutrients on sensitive ecosystems (Evans et al. 2004; Pryor and Sorensen 2000). Furthermore, as air quality models become increasingly coupled with meteorological models (Pleim et al. 2008), and online photolysis calculations become the norm (Foley et al. 2010), accurate particle size-distribution predictions will be needed for calculations of cloud condensation nuclei (CCN), photolysis rates, and radiative feedbacks of particles on meteorology.

Despite the need for reliable simulations of particle size distributions, few studies have evaluated size-distribution predictions of the U.S. EPA's Community Multiscale Air Quality (CMAQ) model, which is commonly used in developing State Implementation Plans for attaining the NAAQS. Zhang et al. (2006) compared volume size-distribution predictions of CMAQ with a distribution calculated from particle observations (diameter <2 μm) in Atlanta assuming a lognormal function for the observations. Elleman and Covert (2009a) evaluated predictions of total mass concentration in two <1-μm diameter ranges for CMAQ simulations of the Pacific Northwest. Kelly et al. (2010) evaluated mass size-distribution predictions for inorganic particle components against cascade impactor measurements for a wide range of aerodynamic diameters (D_aero; ∼0.056–18 μm) at coastal sites in Tampa, FL. These studies indicate that geometric mean diameters (GMDs) and geometric standard deviations (GSDs) for particle mass size distributions are sometimes over-predicted by CMAQ. However, the studies were limited to summer conditions and did not evaluate size-distribution predictions with speciated measurements of elemental carbon (EC) or organic matter (OM).

Evaluations of particle number size-distribution predictions by CMAQ have also been scarce until very recently. Park et al. (2006) evaluated CMAQ number distributions for a 20-month period from January 1999 to August 2000 in Atlanta. They reported an under-prediction of particle number concentration in the diameter range 0.003–2 μm. Zhang et al. (2006) evaluated CMAQ number distributions at the same site for 10 days in July 1999 and reported an over-prediction of particle number concentration in the diameter range 0.1–2 μm. Fan et al. (2006) reported under-prediction of number concentration by CMAQ over Houston, TX, although predictions improved after accounting for enhanced particle formation due to organic acids (Zhang et al. 2004). Elleman and Covert (2009a, b, 2010) evaluated number size-distribution predictions for the Pacific Northwest and investigated the ability of ternary nucleation parameterizations (NH₃−H₂SO₄−H₂O) and updated emissions distributions to rectify large under-predictions of number concentration. Zhang et al. (2010) and Chang et al. (2009) also recently evaluated the impact of numerous treatments of new particle formation on CMAQ predictions of particle number concentration in the U.S.

A challenge to evaluating model predictions of particle size distributions is the lack of routine observations. However, as part of the California Regional PM₁₀/PM_2.5 Air Quality Study (CRPAQS; Watson et al. 1998), numerous measurements of size-segregated PM composition (i.e., size-composition distributions) were made in and at the boundaries of the San Joaquin Valley (SJV) of California between December 2000 and February 2001 (Chow et al. 2008; Herner et al. 2005). Herner et al. (2005) noted that this study appears to represent the largest measurement of size-composition distributions using cascade impactors outside of Los Angeles. Size-resolved observations of particle number concentration are also available from this study (Watson et al. 2002; Herner et al. 2006). Yet, to date, CMAQ simulations of particle size distributions have not been evaluated against these observations.

The diameter dependence of associations of PM with health effects could be due in part to the diameter dependence of particle deposition in the lung. For instance, particles that deposit in the pulmonary region of the lung, which contains alveolar sacs, have easier access to the blood stream and may have longer lung residence times than particles that deposit in the extra-thoracic or conducting regions (Asgharian et al. 1995; Schlesinger 1990). Oberdörster et al. (2005) indicate that 20-nm particles have the highest deposition efficiency in the pulmonary region. Since air quality models are often applied in the context of health-based regulations, a demonstration of similar lung deposition patterns for simulated and observed particle size distributions would provide additional support for using the models in regulatory applications. Models that simulate size-resolved particle deposition in the lung (Alföldy et al. 2009; Asgharian et al. 2001; Hofmann 2009) could be used for this purpose. However, to our knowledge, such models have never been applied in the context of evaluating air quality model performance.

TABLE 1 Summary of Aitken and accumulation mode parameters used for mapping particle emissions to lognormal distributions. Coarse mode parameters not changed in this study

Mode	Scenario^a		% Mass^b	D^c g (nm)	D^d gv (nm)	GSD^e
Aitken	CMAQv4.6	SO^2− 4, NO^− 3, other	0	13	30	1.7
		OC & EC	0.1	13	30	1.7
	Improved Emission		10	25	60	1.7
Accumulation	CMAQv4.6	SO^2− 4, NO^− 3, other	100	70	300	2.0
		OC & EC	99.9	70	300	2.0
	Improved Emission		90	120	280	1.7
^aCMAQv4.6 is standard model; Improved Emission from Elleman and Covert (2010).
^bPercentage of emitted mass apportioned to mode.
^cGeometric mean diameter of particle number-distribution mode.
^dGeometric mean diameter of particle volume-distribution mode.
^eGeometric standard deviation of mode.

The objectives of this study are as follows: (1) to evaluate particle mass and number size-distribution predictions of CMAQ with the standard-emission distribution and the “Improved Emission” distribution of Elleman and Covert (2010) using observations from CRPAQS, (2) to investigate whether differences in modeled and observed particle number size distributions are significant in the context of particle deposition in the lung, (3) to consider the implications of sub-grid scale concentration gradients for simulating particle number size distributions with regional air quality models, and (4) to identify areas for future model development.

MODELING

Air Quality Modeling

The meteorological fields used by the air quality model were generated with the 5th generation Penn State/National Center for Atmospheric Research Mesoscale Model (MM5 version 3.7.4; Grell et al. 1994). CMAQ-ready meteorological fields were created by processing MM5 output with the Meteorology–Chemistry Interface Program version 3.3. The meteorological model was configured with 30 vertical layers extending up to 100 mb, the Mellor-Yamada (Eta) boundary layer scheme (Janjiæ 1990, 1994), the Dudhia simple ice (Dudhia 1989) and cloud radiation schemes (Dudhia 1993), and the five-layer soil model (Dudhia 1996). Gridded emission files were based on the 2000 emission inventories prepared by the California Air Resources Board (http://www.arb.ca.gov/ei/emsmain/emsmain.htm). Biogenic and on-road mobile source emissions were calculated using day-specific air-temperature fields based on objective analysis of observations. Weekday and weekend differences in area- and point-source emissions were represented.

Air quality simulations were conducted with CMAQ version 4.6 on a domain that includes all of California and Nevada as well as part of the Pacific Ocean. The simulation domain is illustrated in Figure S1 of the article supplement and covers the area from (32.36°N, 125.76°W) to (42.18°N, 113.16°W) with 4 km × 4 km horizontal grid cells (269 × 269 grid points). Vertically, the model extends from the surface up to 100 mb using 15 layers, the lowest nominally 30-m deep. The simulations cover the time period December 8, 2000 to January 31, 2001 and were initialized with output from a simulation for December 1–7, 2000. Chemical boundary conditions, based on output from the MOZART-4 global chemical transport model (Emmons et al. 2009; Tang et al. 2009), varied spatially and monthly. The SAPRC99 chemical mechanism (Carter 2000) was used for simulating gas-phase chemistry with the Euler Backward Iterative solver, and the aero4 module was used for simulating aerosol processes.

CMAQ represents the atmospheric particle distribution as the superposition of three log-normal distributions, referred to as modes. A mode is defined by its GMD, GSD, and magnitude, where the magnitude is number concentration in the case of the number distribution and mass in the case of the mass distribution. Particles are assumed to be spherical, and processes including binary homogeneous nucleation (H₂SO₄−H₂O; Kulmala et al. 1998), coagulation, and condensation-evaporation are simulated. Additionally, a mode-merging algorithm transfers particle number, surface area, and mass concentration from the Aitken mode to the accumulation mode when the Aitken mode growth rate and number concentration exceed those of the accumulation mode (Binkowski and Roselle 2003).

Particle number concentration emission rates are determined from mass emissions rates in CMAQ using an assumed emission size distribution according to Equations (7a– c) of Binkowski and Roselle (2003). The emission size distribution specifies the GMD and GSD for the particle modes, and non-sea-salt components are apportioned to the modes in fixed, component-specific amounts. For instance, all primary EC emissions have the same emission size distribution independent of source and the same modal GMDs and GSDs as the other primary non-sea-salt components (largely OM in this study). Sea-salt emissions are modeled interactively in CMAQ (Sarwar and Bhave 2007; Kelly et al. 2010) and are of minor significance in this study. Particle number size distributions are influenced by numerous processes in CMAQ including emission and deposition of particles, nucleation, coagulation, and boundary conditions. See Binkowski and Roselle (2003) for further details on CMAQ's aerosol module.

This study focuses largely on two air quality simulations: one based on the standard version of CMAQv4.6, and one based on the Improved Emission scenario of Elleman and Covert (2010). These simulations differed only in the parameters used for apportioning particle-mass emissions into the two fine-diameter modes and the modal GMDs and GSDs as summarized in Table 1. The Improved Emission case was recently developed by Elleman and Covert (2010) from a literature review of modern measurements corresponding to 4–15 km spatial scales in mid-latitude regions influenced by urban emissions. Although the emission size-distribution parameters for this case are not specific to the individual sources in urban environments, they represent an improvement over CMAQ's standard parameters, which are based on outdated measurements and inappropriate spatial scales (Elleman and Covert 2010). The biggest difference between the standard and Improved Emission size distributions is a more prominent Aitken mode with higher particle number concentrations for the updated distribution.

FIG. 1 Particle size-distribution observation sites and terrain in California.

Lung Deposition Modeling

The Multiple-Path Particle Dosimetry (MPPD) model (MPPDv2.1; http://www.ara.com/products/mppd.htm; Anjilvel and Asgharian 1995; Asgharian et al. 2001) was used to predict the deposition of observed and simulated particle number size distributions in the lung. The MPPD model calculates deposition in lung airways using analytic formulas for deposition efficiency by mechanisms of impaction due to inertial forces, diffusion due to thermodynamic forces, and sedimentation due to gravitational forces. For the ultrafine particles considered here, diffusion is the dominant deposition mechanism, and deposition occurs when the residence time of a particle in an airway is longer than the time required for a particle to diffuse to the airway surface. An empirical formula is used to estimate deposition in the extra-thoracic region because of the complexity of this geometry. Flow rate in each airway is proportional to the airway's distal volume, and flow behavior is not influenced by the presence of particles.

The five-lobe symmetric but structurally different lung geometry developed by Yeh and Schum (1980) from measurements of Raabe et al. (1976) was used for all lung deposition calculations. This lung geometry approximates an adult, healthy, and nonsmoking male person. The five-lobe symmetric geometry is suitable for predicting average particle deposition per airway generation, although more advanced modeling approaches are possible with stochastic lung geometries (Asgharian et al. 2001; Hofmann et al. 2002). Values for simulation parameters (e.g., breathing frequency, tidal volume) were chosen to match those of Asgharian et al. (2001), except that oral rather than endo-tracheal breathing is simulated here. Also, particle density was set to 1.3 g cm^–3 to approximate wood smoke particles (Herner et al. 2006), although increasing the density to 1.8 g cm^–3 had little impact on results.

Deposition for each size distribution was calculated as the number-weighted average of deposition fractions calculated for each diameter of the size distribution measurements (see below) or similarly discretized CMAQ size distributions. Deposition fraction is the number of particles deposited divided by the total number of particles inhaled. Average deposition fraction for the distributions was determined as a function of lung airway generation (i.e., generation 1 is the trachea, generation 2 is the first bifurcation, and so on, down the lung tree). Note that this study investigates whether differences in observed and simulated particle number size distributions are significant in terms of particle deposition in the lung but does not consider the health effects associated with deposited particles.

OBSERVATIONS

Simulated particle mass size distributions were evaluated using observations of size-composition distributions reported elsewhere (Chow et al. 2008; Herner et al. 2005) for eight sites in California (Figure 1): Bodega Bay, Sacramento, Davis, Modesto, Fresno, Angiola, Bakersfield, and Sequoia. These observations, based on Micro-Orifice Uniform Deposit Impactor (MOUDI) samples, included the following particle components: sulfate (SO²⁻ ₄), ammonium (NH⁺ ₄), nitrate (NO⁻ ₃), sodium (Na⁺), chloride (Cl⁻), organic carbon (OC), and elemental carbon (EC). OC measurements were multiplied by 1.4 to obtain estimates of organic mass comparable to modeled OM (Chow et al. 2006; Turpin and Lim 2001). Observations from two Intensive Operating Periods (IOPs) within a three-week stagnation event (Herner et al. 2005) are considered here for evaluating modeled mass size distributions: IOP1 during December 15–18, 2000 and IOP3 during January 4–7, 2001.

Herner et al. (2005) provide details on the size-composition distribution measurements at Bodega Bay, Sacramento, Davis, Modesto, Bakersfield, and Sequoia. At these sites, MOUDIs were equipped with a cyclone to remove particles with D_aero> 1.8 μm and six fractionated impactor stages with nominal d ₅₀s (D_aero for which 50% of particles are retained by the stage) of 0.056, 0.1, 0.18, 0.32, 0.56, and 1.0 μm. OC and EC were analyzed by thermal-optical transmittance following the NIOSH 5040 protocol (Birch and Cary 1996). Chow et al. (2008) provide details on the size-composition distribution measurements at Angiola and Fresno. At these sites, MOUDIs were equipped with eight fractionated stages with nominal d ₅₀s of 0.1, 0.18, 0.32, 0.56, 1.0, 1.8, 2.5, and 5.6 μm for IOP1, while the 1.8-μm stage was replaced by a 0.056-μm stage before IOP3. At Angiola and Fresno, OC and EC were analyzed by thermal-optical reflectance following the IMPROVE protocol (Chow et al. 2001).

Simulated particle number size distributions were evaluated using Scanning Mobility Particle Sizer (SMPS) measurements reported elsewhere (Herner et al. 2005; Zhu et al. 2004; Watson et al. 2002) for four sites in California: Modesto, Fresno, Bakersfield, and LA. For the Fresno site, continuous SMPS observations were available for the entire simulation period with 52 channels in the 0.009–0.392 μm electric mobility diameter range. Since the average temperature in the SMPS cabinet was 17°C higher than the ambient temperature, Watson et al. (2002) assumed the observations represented dry particles. For the Modesto site, SMPS observations were available during IOP3 with 108 channels in the 0.016–0.764 μm electric mobility diameter range (Herner et al. 2005). Similar unpublished SMPS measurements were obtained for Bakersfield during IOP3; however, these observations cover only January 4–5, 2001 due to an instrument malfunction beginning on January 6. SMPS observations at several distances from Highway 710 in LA were available for January 21–25, 2002 with 108 channels in the 0.006–0.225 μm electric mobility diameter range (Zhu et al. 2004). Although these observations do not overlap the simulation period, they are considered in the discussion of sub-grid-scale concentration gradients.

FIG. 2 Average observed and modeled mass size distributions for CRPAQS Intensive Operating Period 1 (IOP1). See upper-left FIG. for legend (Obs: Observations; Imp. Emiss.: Improved Emission scenario of Elleman and Covert, 2010). Dashed vertical line indicates aerodynamic diameter of 2.5 μm.

FIG. 3 Same as Figure 2, but for IOP3.

TABLE 2 Observed and modeled PM_2.5 (Angiola and Fresno) or PM_1.8 (other sites) associated with Figs. 2 and 3 (μg m⁻³)

		SO^2− 4		NH^+ 4		NO^− 3		Na^+		Cl^−		OM		EC
Site	Category	IOP1	IOP3	IOP1	IOP3	IOP1	IOP3	IOP1	IOP3	IOP1	IOP3	IOP1	IOP3	IOP1	IOP3
Bodega Bay	Observation	0.79	1.50	0.18	3.64	0.34	7.60	1.77	1.10	3.07	1.80	1.53	4.89	0.23	0.71
	CMAQv4.6	0.53	0.74	0.53	2.44	1.17	7.48	0.02	0.04	0.01	0.04	2.09	3.30	0.44	0.93
	Imp. Emiss.	0.56	0.77	0.56	2.56	1.25	7.88	0.02	0.04	0.01	0.05	2.19	3.49	0.46	0.98
Sacramento/Davis^a	Observation	0.65	0.64	1.19	5.85	2.91	12.93	0.16	0.03	0.15	0.10	5.83	10.05	1.06	1.75
	CMAQv4.6	0.88	0.97	2.31	3.61	6.85	11.20	0.01	0.01	0.01	0.01	5.16	3.11	1.76	1.29
	Imp. Emiss.	0.94	1.04	2.48	3.90	7.36	12.10	0.01	0.01	0.01	0.01	5.56	3.50	1.90	1.45
Modesto	Observation	1.67	2.73	2.87	16.41	5.73	34.15	0.19	0.12	0.32	0.55	9.34	25.34	1.59	3.74
	CMAQv4.6	1.78	3.83	2.96	5.90	7.90	15.38	0.00	0.01	0.01	0.01	4.08	4.95	1.60	2.00
	Imp. Emiss.	1.90	3.92	3.19	6.14	8.57	16.10	0.01	0.01	0.01	0.01	4.47	5.45	1.75	2.20
Fresno	Observation	1.23	2.26	1.69	11.09	3.90	31.85	0.14	0.10	0.18	0.17	23.91	45.79	2.86	4.15
	CMAQv4.6	3.46	4.60	5.36	6.62	14.00	16.89	0.00	0.02	0.01	0.03	6.65	6.44	2.82	2.60
	Imp. Emiss.	3.63	4.74	5.69	6.86	14.91	17.55	0.01	0.03	0.01	0.03	7.09	6.76	3.01	2.73
Angiola	Observation	1.26	3.36	1.83	16.89	4.03	50.02	0.21	0.10	0.41	0.09	13.29	27.65	1.16	2.21
	CMAQv4.6	0.62	0.98	2.98	6.24	9.48	20.23	0.01	0.02	0.02	0.03	1.68	1.92	0.63	0.77
	Imp. Emiss.	0.66	1.00	3.18	6.34	10.10	20.55	0.02	0.02	0.02	0.03	1.84	2.03	0.69	0.81
Bakersfield	Observation	2.12	3.97	4.23	25.92	11.87	53.57	0.08	0.24	0.58	0.32	9.67	24.33	1.59	3.64
	CMAQv4.6	3.22	4.47	4.17	6.84	10.21	17.80	0.01	0.01	0.01	0.02	5.02	4.98	3.30	2.94
	Imp. Emiss.	3.41	4.63	4.45	7.22	10.94	18.90	0.02	0.02	0.02	0.02	5.41	5.34	3.54	3.14
Sequoia	Observation	0.68	0.60	0.89	1.02	1.38	2.33	0.08	0.39	0.09	0.12	2.47	2.35	0.23	0.28
	CMAQv4.6	0.39	0.43	0.41	0.21	0.91	0.18	0.00	0.01	0.00	0.01	0.58	0.44	0.17	0.14
	Imp. Emiss.	0.40	0.42	0.42	0.21	0.96	0.19	0.01	0.01	0.01	0.01	0.62	0.43	0.19	0.14
^aIOP1 values are for Sacramento; IOP3 values are for Davis.

Particle size distributions for CMAQ simulations were constructed for comparison with observations using model output for modal wet diameters for all cases except for number distributions at Fresno, where modal dry diameters were used. Simulated Stokes diameters were converted to D_aero for comparisons with MOUDI measurements. For comparison with SMPS data, no adjustment was made to simulated diameters due to the lack of information on particle dynamic shape factors needed for calculating mobility equivalent diameters (Kasper 1982). Grid-cell average predications are compared with point measurements matched in space and time in this study. Descriptions of the observation sites considered here are available in the following articles: Watson et al. (2002), Zhu et al. (2004), Herner et al. (2005), and Chow et al. (2006).

RESULTS

Particle Mass Size Distributions

Average observed and modeled speciated particle mass size distributions for IOP1 and IOP3 are shown in Figures 2 and 3, respectively. The speciated PM_2.5 (Fresno and Angiola) or PM_1.8 (other sites) concentrations associated with these figures are provided in Table 2. In general, both models correctly predict that the concentrations of SO²⁻ ₄, NH⁺ ₄, and NO⁻ ₃were greater at sites in the southern portion of California's Central Valley (i.e., SJV) than at the non-SJV sites (i.e., Bodega Bay, Sacramento, Davis, and Sequoia). However, contrary to observations, the models predicted lower PM_2.5 concentration at the rural Angiola site than PM_1.8 at the urban Sacramento site for SO²⁻ ₄ during IOP1 (Table 2). The models correctly predicted higher OM and EC concentrations for the Central Valley sites than for the coastal Bodega Bay site and elevated (535 m above msl) Sequoia site in general. However, contrary to observations, the models predicted lower PM_2.5 concentration at Angiola than PM_1.8 at Bodega Bay for OM. This behavior is consistent with Ying et al.'s (2008) hypothesis that the emissions inventory around the Angiola site is incomplete. OM in the SJV was largely due to primary emissions during the wintertime IOPs, although secondary OM is believed to be non-negligible for these periods (Chen et al. 2009).

NH⁺ ₄and NO⁻ ₃ were major PM components in the SJV during the simulation period (Chow et al., 2006). For these components, the models tend to over-predict the relatively moderate concentrations observed during IOP1 and under-predict the extremely high concentrations observed during IOP3. This behavior could be due in part to limitations of using prognostic meteorological fields in simulating air quality for stagnant periods on domains with complex topography (Hu et al., 2009; Jackson et al., 2006). However, over-prediction of PM during IOP1 and under-prediction during IOP3 has also been reported for air quality simulations based on objective-analyses meteorological fields (Ying et al. 2008) despite evidence that objective-analyses fields may result in better air quality model performance (Jackson et al. 2006; Hu et al. 2009).

Kelly et al. (2010) reported that concentration peaks in modeled size distributions for inorganic particle components occurred in the adjacent larger diameter bin to the observed peaks when CMAQ's lognormal distributions were mapped to the discrete bins of impactor measurements in Tampa. For the California simulations in the present study, CMAQ correctly predicts the peak-concentration diameter bin in some cases. For instance, the correct bin appears to be predicted for SO²⁻ ₄, NH⁺ ₄, and NO⁻ ₃ distributions at Sacramento and Bakersfield during IOP1 and for the NO⁻ ₃ distribution at Fresno and Bodega Bay during IOP3 (see supplementary Figures S2 and S3 for versions of Figures 2 and 3 with CMAQ distributions mapped to impactor stages). However, over-predictions of peak-concentration diameter similar to those reported by Kelly et al. (2010) are clearly evident for SO²⁻ ₄, NH⁺ ₄, and NO⁻ ₃ distributions at Modesto and Angiola during IOP1 (Figure 2) and Angiola and Bakersfield during IOP3 (Figure 3). For these cases, observed peaks were in the 0.56–1 μm diameter bin, whereas modeled peaks were in the 1–1.8 μm or 1–2.5 μm diameter bin. Differences in emission size distributions (i.e., standard or Improved Emission) had little impact on predictions of peak-concentration diameter.

The concentration peaks for OM and EC distributions were observed at smaller diameters than were the peaks for the inorganic components (Figures 2 and 3). For all sites and periods, CMAQ over-predicted the peak-concentration diameter for OM and EC size distributions, sometimes by several bins (e.g., OM at Fresno during IOP1). Differences in peak-concentration diameter for organic and inorganic PM could be due to external mixing (i.e., different particles of a given size have different composition) of PM with different size distributions. However, PM is internally mixed during MOUDI sampling (i.e., all particles of a given size are mixed together on an impactor stage), and so particle mixing-state cannot be determined from MOUDI observations. In any case, CMAQ's aerosol formulation, which is based on two internally mixed sub-micrometer modes, does not appear flexible enough to capture the large observed differences between carbon (OM and EC) and inorganic size distributions. Note that inorganic aerosol equilibrium calculations are performed on the combined sub-micrometer modes in CMAQ, and components are then distributed to each mode in proportion to its condensation moment. Therefore a scenario could not occur in which the Aitken and accumulation modes separately simulated the carbon (OM and EC) and inorganic distributions to reproduce observations in Figures 2 and 3. At some sites (e.g., Fresno and Modesto), the Improved Emission simulation produces a tail of relatively high OM and EC concentrations for <0.2 μm diameters. This behavior is a minor improvement over results for the standard-emission simulation.

Previous studies (Zhang et al. 2006; 2010) have reported that size-distributions simulated by CMAQ are sometimes too wide compared with observations (i.e., particle-mode GSDs are over-predicted). Here, the widths of simulated distributions appear to agree with observations in some cases (e.g., OM at Fresno during IOP1), although reliable quantitative comparisons are not possible due to structural differences in the impactor and log-normal distributions. However, despite occasional agreement, the simulated distributions are generally much broader than the observed distributions at the California sites (e.g., SO²⁻ ₄, NH⁺ ₄, and NO⁻ ₃ distributions at Angiola and Bakersfield). Also, the observations indicate narrower distributions with sharper peaks for SO²⁻ ₄, NH⁺ ₄, and NO⁻ ₃ distributions at Angiola and Bakersfield than at the other sites, but these spatial differences are not captured by the models (Figures 2 and 3). Average accumulation-mode GSDs for the Improved Emission simulation (1.95–2.19, IOP1; 2.02–2.25, IOP3) are less than those of the standard-emission simulation (2.16–2.33, IOP1; 2.23–2.34, IOP3) due to the lower GSD for the updated emission distribution (1.7 vs. 2, Table 1). However, differences in distribution widths for the two simulations are much smaller than the differences between observed and simulated widths.

For regulatory modeling applications, particle mass size distributions should be simulated reliably enough to capture the fraction of the distributions with D_aero< 2.5 μm (i.e., the PM_2.5 fraction). The over-predictions of peak-concentration diameter and distribution width just mentioned could lead to under-predictions in the PM_2.5 fraction by increasing the mass fraction in the >2.5 μm diameter range. For the Angiola and Fresno sites, the observed and simulated PM_2.5 fraction is given in Table S1 of the article supplement for the distributions in Figures 2 and 3. Although the models provide reasonable predictions of the PM_2.5 fraction for NO⁻ ₃ at Fresno during IOP3, the PM_2.5 fraction is under-predicted in general. The under-prediction is relatively large for NH⁺ ₄ and NO⁻ ₃ at Angiola, where observed fractions are 0.95–0.98 and simulated fractions are 0.69–0.79. Under-predictions of the PM_2.5 fraction are also considerable for OM and EC at Fresno, where observed values are 0.98–1 and simulated values are 0.84–0.89. Estimates of the PM_2.5 fraction for the Improved Emission simulation are slightly better than those for the standard-emission simulation (Table S1). Although errors in speciated PM_2.5 fractions should be considered when calculating total PM_2.5, under-predictions in PM_2.5 fraction are generally small compared to differences in modeled and observed concentrations in Table 2, and the simulations of particle mass size distributions may be considered adequate in this context.

Coastal sites typically experience high concentrations of Na⁺ and Cl⁻ in the coarse-particle diameter range (Nolte et al. 2008). Although the MOUDI observations at the coastal Bodega Bay site only covered fine-diameter particles (D_aero< 1.8 μm), results in Figure 3 of Herner et al. (2006) suggest that coarse Na⁺ and Cl⁻ were substantially under-predicted here. This under-prediction is primarily due to the lack of enhanced sea-salt emissions from the coastal surf zone in CMAQv4.6. A surf-zone emission parameterization has recently been incorporated into CMAQv4.7 to address this model limitation (Kelly et al. 2010). Other recent updates to CMAQ in version 4.7 (Foley et al. 2010) would be insignificant for our study conditions.

Particle Number Size Distributions

In Figure 4, the time series of observed and modeled number concentration for particles with diameter >9 nm at Fresno is shown for December 8, 2000 to January 31, 2001. Observations were averaged to hourly time resolution for this comparison. The mean particle number concentration for the Improved Emission simulation (17 × 10³ cm⁻³) was about 20% lower than the mean observed concentration (21 × 10³ cm⁻³), while the value for the standard-emission simulation (4.5 × 10³ cm⁻³) was nearly a factor of 5 too low. The concentration range for the Improved Emission simulation (1.5 × 10³ – 38 × 10³ cm⁻³) also better matched the observed range (0.5 × 10³ – 83 × 10³ cm⁻³) than did the range of the standard-emission simulation (0.1 × 10³ – 11 × 10³ cm⁻³). Therefore the updated emission distribution greatly improved particle number concentration predictions for this site and time period.

FIG. 4 Time series of observed and modeled number concentration (number cm^–3) for particles with diameter > 9 nm at Fresno for 12/8/00–1/31/01.

FIG. 5 Average diurnal number-concentration (number cm⁻³) profile at Fresno associated with time series in Figure 4.

FIG. 6 Average observed and modeled number size distributions at Fresno for 12/8/00–1/31/01 for (A) 7:00–8:00 PST, (B) 20:00–21:00 PST, and (C) all hours.

The average diurnal particle number concentration profile associated with Figure 4 is shown in Figure 5. The peak in the observed concentration profile during 5–9 PST was primarily caused by morning traffic emissions, while the peak during 16–21 PST was caused by residential wood combustion and traffic emissions in combination with a nighttime temperature inversion (Watson et al. 2002). The Improved Emission curve displays diurnal variations similar to but weaker than the observations, while the standard-emission simulation produced little diurnal variation (Figure 5). The lack of diurnal variation in simulated number concentration could be due to under-prediction of the range of planetary boundary layer (PBL) heights in addition to grid-cell averaging issues, too little evening wood-combustion emissions, and limitations in PM process modeling.

TABLE 3 Average peak-concentration diameter (nm) for particle number size distributions. Both wet and dry diameters are given for modeled distributions

	Bakersfield (All Hours)	Modesto (All Hours)	Fresno (All Hours)	Fresno (7–8 PST)	Fresno (20–21 PST)
Observation	79	79	78	35	78
CMAQv4.6 (wet)	118	131	120	143	114
CMAQv4.6 (dry)	76	80	76	76	75
Imp. Emiss. (wet)	34	43	40	42	40
Imp. Emiss. (dry)	25	31	30	28	30

Average modeled and observed particle number size distributions at the Fresno site are shown in Figure 6A and 6B for morning (7–8 PST) and evening (20–21 PST) hours during the simulation period. Modeled dry diameters are used for this comparison in an attempt to account for the effects of sample heating in the instrument cabinet (Watson et al. 2002). The observed size distribution is relatively broad in the morning and peaks at a diameter of 35 nm (Figure 6a). The peak-concentration diameter for the Improved Emission simulation (28 nm) is similar to that of the observations, while the peak-concentration diameter for the standard-emission simulation (76 nm) is over-predicted by a factor of two. The particle number size distribution for the Improved Emission simulation overlaps the observed distribution for a range of diameters (∼13–30 nm) in the morning, although the simulated distribution is much narrower than the observed. The size distribution for the standard-emission simulation is a poor estimate of the observations due to the drastic under-prediction of particle number concentration.

FIG. 7 Average observed and modeled number size distributions at Modesto (1/4/01–1/7/01) and Bakersfield (1/4/01–1/5/01).

For the evening hour (Figure 6b), the peak-concentration diameter for the observed particle number size distribution (78 nm) is twice that of the morning hour and reflects the prevalence of wood-combustion emissions in the evening. Contrary to observations, the simulated size distributions do not vary much from morning to evening because the emission size distributions used by the models are independent of the emission source (i.e., traffic or wood combustion). The better agreement of the peak-concentration diameter for the Improved Emission distribution with the morning observations likely reflects the dominance of traffic emissions and lack of wood-combustion emissions in the dataset used by Elleman and Covert (2010). Doubling the Aitken-mode diameter of the Improved Emission distribution in a sensitivity simulation for December 2000 increased the simulated peak diameter by a factor of 2.2 over all hours but decreased number concentration by a factor of 3.5. Therefore simulated distributions are strongly impacted by the emission size distribution for our conditions, but more work is required to adequately represent wood-combustion emissions in the model.

The shape of the average observed size distribution for all hours (Figure 6c) resembles the average distribution for the evening hour (Figure 6b) more than that for the morning hour (Figure 6a) and indicates the importance of wood-combustion emissions to the particle number size distributions at Fresno. The shape of the distribution for the Improved Emission simulation is similar to observations for this case, but the simulated distribution is shifted to relatively small diameters (Figure 6c). The greater peak-concentration diameter for the standard-emission simulation than the Improved Emission simulation is due to the much greater apportioning of mass emissions to the accumulation mode for that case (Table 1).

Average modeled and observed particle number size distributions at the Modesto and Bakersfield sites during IOP3 are shown in Figure 7. For both sites, the shape of the distribution for the Improved Emission simulation adequately reflects the observed shape. However, the simulated distribution is shifted to relatively small diameters and number concentration is over-predicted. Conversely, the standard-emission simulation under-predicts number concentration and over-predicts peak-concentration diameter at Modesto and Bakersfield. These results suggest that modeled particle number size distributions are highly sensitive to emission size distributions and that an improved treatment of emission distributions may be required to adequately simulate number distributions with air quality models. Note that the assumption of internally mixed particles discussed above for mass size distributions could also impact model performance for number size distributions.

FIG. 8 Average modeled deposition fraction in the 5-lobe symmetric lung geometry as a function of airway generation number for average particle number size distributions in Figures 6c and 7. Average particle number concentration (number cm^–3) is given in parenthesis in the legend for the following diameter ranges (nm): 5–1000 for model, 9–392 for observations at Fresno, and 16–764 for observations at Modesto and Bakersfield. “H” indicates extrathoracic region; generation 1 is trachea.

Dry particle diameters were used in the number size distribution evaluation for the Fresno site because of evidence of particle drying for those measurements (Watson et al. 2002). Particles may have also been heated to some degree in the measurements at Modesto and Bakersfield. However, estimating the influence of sampling on particle diameter is challenging for reasons including the variability in atmospheric conditions, uncertainty in insulating capacity of the conductive tubing, and hysteresis effects on particle phase state. In Table 3, both wet and dry peak-concentration diameter predictions are listed with observed peaks. For the Improved Emission simulation, both wet and dry peak diameters are significantly lower than the observed peaks for all cases except Fresno in morning. For the standard-emission simulation, wet peak diameters are consistently greater than the observations while dry ones generally agree.

Particle Deposition in the Lung

In evaluating predictions of particle mass size distributions above, model performance was considered in the context of the PM_2.5 NAAQS by evaluating the model's ability to simulate the PM_2.5 fraction. Good model performance for the PM_2.5 fraction provides additional support for using CMAQ in regulatory applications, although effects of inaccurate PM_2.5-fraction predictions are reduced in regulatory modeling through use of relative response factors (U.S. EPA 2007). In evaluating predictions of particle number size distributions, however, a U.S. air quality standard is not available to provide context. In this section, differences in modeled particle deposition fraction in the lung for observed and simulated particle number size distributions are examined. Similarity in lung deposition fraction for modeled and observed distributions would be an indicator of good model performance for particle number size distribution simulations.

In Figure 8, the number-weighted average particle deposition fraction is shown as a function of airway generation for the modeled and observed distributions at Fresno, Modesto, and Bakersfield. The total number concentration of particles for each distribution is given in the figure legend. A peak occurs in the figures because deposition for ultrafine particles (i.e., diffusion dominated) initially increases with lung generation as airway flow rates and volume-to-surface area ratios decrease. Eventually, particles with sufficient diffusivity to deposit are removed from the air flow and deposition fraction decreases in subsequent airway generations. Distributions with smaller peak-concentration diameters have more particles with high diffusivity and tend to have peak deposition fractions in earlier airway generations. At all sites, the lung generation of peak deposition fraction for the standard-emission simulation matches that of the observations (i.e., generation 20). For the Improved Emission simulation, the lung generation of peak deposition fraction agrees with observations at Modesto, but occurs in generation 19 at Fresno and Bakersfield. The pulmonary region of the modeled lung geometry begins at generation 16–18 (depending on lobe), and so both simulated distributions correctly yield peak deposition fractions in the pulmonary region of the lung.

Lung deposition fractions for number size distributions of the Improved Emission simulation were greater than those for the observed distributions at all sites, while deposition fractions for distributions of the standard-emission simulation were often less (Figure 8). The peak deposition fraction associated with observations ranged from 0.041 (Bakersfield) to 0.048 (Fresno), while the range was from 0.05 (Bakersfield) to 0.054 (Fresno) for the Improved Emission simulation and from 0.036 (Modesto) to 0.044 (Fresno) for the standard-emission simulation. The greater lung deposition fractions associated with the Improved Emission simulation are caused by the under-prediction of particle diameter in that case, which enhanced particle diffusivity and lung deposition. Conversely, the lower lung deposition fractions associated with the standard-emission simulation than the observations for the Modesto site are caused by the over-prediction of particle diameter in that case. However, the decent agreement in deposition fraction for the Fresno and Bakersfield sites indicates that differences in the distributions for the standard-emission simulation and observations could be considered minor by a lung deposition fraction metric. The impact of errors in modeled size distributions on lung deposition fraction varies with particle diameter range. Since lung deposition as a function of diameter reaches a minimum between 0.1 and 1 μm (Asgharian et al. 2001), model errors near this size range have relatively small impact on lung deposition fraction.

In Figure 8, lung deposition is presented as a fraction, and so similarity in the curves does not imply similar particle lung doses for the observed and simulated distributions. Since the total number concentration of particles differs among the distributions, the number of deposited particles for a given tidal volume would be different for the same particle deposition fraction. For instance, the under-prediction of particle number concentration at Fresno for the standard-emission simulation would lead to a much lower number of deposited particles for the same deposition fraction compared with observations. However, at the Bakersfield site, both the number concentrations and lung deposition fractions associated with the standard-emission simulation are in reasonable agreement with values for the observations.

MODELING LIMITATIONS

Although 4-km horizontal grid resolution is considered very high in the field of regional air quality modeling, near-roadway number-concentration gradients cannot be resolved using this resolution. In Figure 9, particle number size distributions observed by Zhu et al. (2004) at distances from Highway 710 in LA in January 2002 are shown with simulated size distributions for January 2001. Since there were no significant changes in engine technologies or traffic patterns between 2001 and 2002, this comparison should be valid. Particle number concentrations decrease sharply with distance from the highway due to rapid dilution with background air. For instance, the observed particle number concentration decreases by a factor of three from 17 m to 90 m from the roadway. Approaches for characterizing sub-grid-scale variability have been applied to CMAQ simulations of benzene and formaldehyde (Isakov et al 2007). For particle number distributions, near-roadway evolution has only been studied using box-model or small-scale three-dimensional models (Jacobson and Seinfeld 2004; Zhang and Wexler 2004; Albriet et al. 2009).

FIG. 9 Observed (Zhu et al. 2004) and modeled particle number size distributions near Highway 710, LA in January. Observations are from 2002; model results are from 2001. Numbered labels indicate distance from roadway in meters, Bkgrd corresponds to background location, and Imp. Emiss. is Improved Emission scenario of Elleman and Covert (2010).

In Figure 10, particle deposition fraction as a function of lung airway generation is shown for the size distributions in Figure 9. The deposition fraction for the distribution observed at 17 m from the roadway peaks at an earlier lung generation and has higher extra-thoracic deposition than the observed background distribution due to the higher concentration of small particles (with high diffusivities) near the roadway. The lung deposition fractions associated with the Improved Emission simulation are in reasonable agreement with those for the observed background distribution, although the number concentration of particles is over-predicted by about a factor of two. Results in Figures 9 and 10 suggest that CMAQ is sometimes capable of providing reasonable estimates of background number size distributions, but that fine-scale models are required to simulate the evolution of number size distributions near the roadway. Near-roadway models could also be used in generating emission size distributions at spatial scales relevant to regional models. Note that gradients in mass concentration of black carbon have also been observed near roadways (Ning et al. 2010). However, near-roadway concentration gradients for total mass concentration are likely to be smaller than for number concentration because key contributors of particle mass (e.g., NH⁺ ₄ and NO⁻ ₃) are regional pollutants and should not vary sharply near roadways.

FIG. 10 Average modeled deposition fraction in the 5-lobe symmetric lung geometry as a function of airway generation number for particle number size distributions in FIG. 9. Average particle number concentration (number cm^–3) is given in parenthesis in the legend for the following diameter ranges (nm): 5–1000 for model, 6–225 for observations. “H” indicates extrathoracic region; generation 1 is trachea.

In addition to sub-grid-scale concentration gradients, particle nucleation and nuclei processing present modeling challenges. Several recent studies have examined various treatments of new particle formation in CMAQ (Fan et al. 2006; Chang et al. 2009; Elleman and Covert 2009b; Zhang et al. 2010); however, the best approach to use for a given simulation is still unclear. For our wintertime study conditions and SJV sites, the impact of freshly-nucleated particles on particle number size distributions may be small. Observations of particle number concentrations (⩾3 nm) at Fresno from August 2002 to July 2003 indicate that concentrations associated with solar irradiance peaks never exceeded 2.44 × 10^–4 cm^–3 and an increase in number concentration with decreasing size below 10 nm very rarely occurs (Watson et al. 2006). A lack of freshly nucleated particles in Fresno could be explained in part by the very low sulfur concentrations in SJV and high particle surface area.

In standard release versions, CMAQ treats particle nucleation according to Kulmala et al. (H₂SO₄-H₂O; 1998) and assumes that all nuclei survive to reach the Aitken mode. Sensitivity simulations were conducted for cases without nucleation and with the parameterization of Vehkamäki et al. (2002), which corrects and extends the Kulmala et al. (1998) approach. Modeled number size distributions were insensitive to these changes. Also, despite significant ammonia concentrations in SJV, results of Zhang et al. (2010) indicate that particle number concentrations in California based on CMAQ simulations using a recent ternary nucleation parameterization (NH₃−H₂SO₄−H₂O; Merikanto et al. 2007, 2009) are similar to those with the binary schemes just mentioned. For conditions where nucleation has a significant impact on particle number concentrations, the growth of fresh nuclei should be treated either by analytical estimates (Kerminen and Kulmala 2002; Lehtinen et al. 2007) or by explicitly simulating a nucleation mode (Sartelet et al. 2006; Kazil et al. 2010).

CLOSING REMARKS

In this study, predictions of particle mass and number size distributions by the CMAQ model were considered for two simulations: one based on standard CMAQv4.6, and one based on CMAQv4.6 using the Improved Emission size distribution of Elleman and Covert (2010). The updated emission distribution slightly improved predictions of speciated particle mass size distributions and PM_2.5 fractions. However, differences in speciated mass distributions for the two simulations were small compared to differences between predictions and observations. Observed OM and EC peak-concentration diameters were generally smaller than those of inorganic components by 1–2 diameter bins, while the models predicted similar peak-concentration diameters for all components. For some conditions (e.g., Fresno during IOP3), a significant fraction of NH⁺ ₄ and NO⁻ ₃ mass (∼15%) was observed in the coarse-diameter range, while nearly all OM and EC mass was observed in the fine-diameter range. In such cases, simulating differences between inorganic and carbon (elemental and organic) size distributions could be necessary to adequately predict total PM_2.5 concentration. Future model development should focus on this issue as well as on resolving the over-predictions of particle-mode GSDs and GMDs reported here and elsewhere. Despite the need for model development, the limitations in particle mass size distribution predictions can be reasonably accounted for in regulatory applications by using relative response factors and by considering the total mass in the two fine-particle modes in addition to modeled PM_2.5.

Particle number size-distribution predictions differed dramatically for the two simulations and indicate that number-distribution predictions are highly sensitive to emission size distributions. The updated emission distribution greatly improved predictions of particle number concentration at the Fresno site; however, peak-concentration diameter was under-predicted during hours where wood-combustion emissions were dominant. Modeled lung deposition fractions for simulated number size distributions peak in the same lung region as those for number size distributions observed in the background. However, differences in modeled and observed total number concentrations suggest large differences in the total number of deposited particles in the lung for the distributions. Modeled lung deposition fractions associated with number distributions from the Improved Emission simulation were higher than those of observations, while lung deposition fractions for the standard-emission simulation tended to be lower. Further improvements in emission size distributions are likely needed to improve simulations of particle number size distributions with CMAQ. However, the high degree of sensitivity of simulated number size distributions to emission size distributions could make adequately representing emission distributions a challenge. Once emission size distributions are reasonably represented in CMAQ, the reliability of nucleation and other parameterizations that affect number-distribution predictions should be examined in further detail.

Particle number concentrations decrease sharply with distance from the roadway, but such variations cannot be captured with grid resolutions typical of regional air quality models. Small-scale models are better suited for simulating the evolution of particle number distributions near the roadway and should be used to generate emission distributions at spatial scales relevant for regional air quality models. Comparisons of modeled number distributions with observations at distances from Highway 710 in LA suggest that CMAQ is capable of providing reasonable predictions of background number size distributions in some cases. Simulated number size distributions may also yield lung deposition fractions similar to those associated with observed background distributions. Model development should focus on improving particle number size distribution predictions such that the total number of deposited particles and lung deposition fractions for simulated and observed number distributions are in reasonable agreement over regional domains. The sensitivity of visibility and CCN calculations to differences in simulated and observed particle number distributions should also be explored in the context of model performance evaluation for particle size distributions.

Acknowledgments

We kindly thank Bahman Asgharian and Owen Price of ARA, Inc. for providing the MPPD model. Also, we thank Michael Kleeman (UC–Davis) and Jorn Herner (CARB) for MOUDI and SMPS data at Bodega Bay, Sacramento, Modesto, Davis, Bakersfield, and Sequoia. We thank Dana Trimble and John Watson of DRI for SMPS data at Fresno and Yifang Zhu of Texas A&M University–Kingsville for near-roadway data in LA. We also thank Daniel Chau of CARB for providing the meteorology fields used in our simulations.

This article has been reviewed by the staff of the California Air Resources Board and has been approved for publication. Approval does not signify that the contents necessarily reflect the views and policies of the California Air Resources Board, nor does mention of trade names or commercial products constitute endorsement or recommendation for use.

[Supplementary materials are available for this article. Go to the publisher's online edition of Aerosol Science and Technology to view the free supplementary files.]