DRAFT MODELING PROTOCOL FOR THE

CALIFORNIA REGIONAL PM10/PM2.5

AIR QUALITY STUDY:

PRELIMINARY APPROACHES AND DATA REQUIREMENTS

Prepared By:

Karen L. Magliano, California Air Resources Board

Philip M. Roth, ENVAIR

John G. Watson, Desert Research Institute

Charles L. Blanchard, ENVAIR

Steven Reynolds, ENVAIR

Saffet Tanrikulu, California Air Resources Board

Glen R. Cass, California Institute of Technology

Version 2.0

June 1998

1. OVERVIEW OF PREFERRED MODELING APPROACHES...... 0-4

2. GRID-BASED MODELING...... 1-6

2.1 Approach...... 1-6

2.1.1 Treatment of Aerosol Thermodynamics...... 1-7

2.1.2 Treatment of the Aerosol Size Distribution...... 1-8

2.1.3 Treatment of Secondary Organic Aerosol Production...... 1-10

2.1.4 Treatment of Dry Deposition of Particles...... 1-11

2.2 Limitations...... 1-11

2.3 Outputs...... 1-12

2.4 Data Requirements...... 1-13

2.4.1 Emissions Data Needs...... 1-13

2.4.2 Meteorological Data Needs...... 1-13

2.4.3 Air Quality Data Needs...... 1-14

2.4.4 Model Performance Evaluation Needs...... 1-14

2.5 Research Needs...... 18

3. RECEPTOR MODELING...... 20

3.1 Approach...... 20

3.2 Limitations...... 21

3.3 Outputs...... 24

3.4 Data Requirements...... 25

3.5 Research Needs...... 25

4. EQUILIBRIUM CHEMISTRY AND RECEPTOR CALCULATIONS DERIVED FROM EQUILIBRIUM MODELS 27

4.1 Approach...... 27

4.1.1 Thermodynamic Equilibrium Models...... 28

4.1.2 Use of Thermodynamic Equilibrium Models to Predict Changes in Aerosol Nitrate...... 28

4.1.3 Use of Measurements to Identify Ammonia Limitation...... 28

4.2 Limitations...... 29

4.3 Outputs...... 30

4.4 Data Requirements...... 30

4.5 Research Needs...... 31

5. HYBRID ROLLBACK...... 32

5.1 Approach...... 32

5.2 Limitations...... 33

5.3 Outputs...... 34

5.4 Data Requirements...... 34

5.5 Research Needs...... 34

6. METEOROLOGICAL MODELING...... 35

7. RECOMMENDED APPROACHES FOR EPISODIC AND ANNUAL AVERAGE MODELING 36

7.1 Annual Average Source Attribution...... 36

7.1.1 Daily Grid-Based Aerosol Modeling...... 36

7.1.2 Representative Regimes Modeling...... 37

7.1.3 Receptor Modeling...... 38

7.2 Wintertime Episodic Source Attribution...... 39

7.2.1 Emissions Modeling...... 39

7.2.2 Meteorological Modeling...... 39

7.2.3 Chemical and Physical Models...... 40

7.2.4 Air Quality Modeling...... 41

7.2.5 Receptor Modeling...... 41

7.3 Fall Episodic Source Attribution...... 41

7.3.1 Emissions Modeling...... 42

7.3.2 Meteorological Modeling...... 42

7.3.3 Chemical and Physical Models...... 42

7.3.4 Air Quality Modeling...... 42

7.3.5 Receptor Modeling...... 42

8. REFERENCES...... 43

1. Overview of Preferred Modeling Approaches

Air quality modeling is an essential tool for understanding source-receptor relationships and for estimating the effects of future emissions scenarios. These mathematical simulations of atmospheric movements and chemistry are also used as part of data analysis activities. There is no universally “best” or “valid” model. Model complexity does not necessarily imply model sophistication. Every model application should: 1) determine model applicability by examining the availability of appropriate data inputs and inclusion of physical and chemical phenomena; 2) identify the potential emissions sources and their emissions characteristics; 3) evaluate model outputs and performance measures; 4) identify and evaluate deviations from model assumptions; 5) identify and remediate input data deficiencies; 6) verify consistency and stability of source contribution estimates; and 7) evaluate results with respect to other data analysis and source assessment methods.

Too often model results are evaluated only with respect to how well they duplicate measured concentrations of the relevant pollutant at a few receptors. A better approach is to stress models to an extent that they fail, and in such a way that the nature of those failures can be diagnosed. Applying both source and receptor models to the same pollution problem is one way in which that stress can be applied.

The San Joaquin Valley must develop control plans that address both the annual average and the 24-hour standards for PM10 and PM2.5. Exceedances of each of these standards have contributions from both primary and secondary material and they occur over a number of different seasons and meteorological scenarios (Watson, 1997). Therefore air quality models that can be reliably used to estimate both annual average and 24-hour PM concentrations over a broad spectrum of conditions are needed.

Unfortunately, air quality models and databases for PM have received less attention than for other pollutants, notably ozone. Several reviews of available modeling methods and initial model applications for PM have been undertaken over the last few years (Lurmann, 1996, Seigneur, 1997, Kaduwela, 1998). The results of these evaluations have been used in formulating recommendations for simulating dynamic processes in the atmosphere and for the comprehensive field program that is to be conducted as part of the California Regional PM10/PM2.5 Air Quality Study in 1999 and 2000.

No single model will be adequate to address all aspects of PM. Instead, a suite of modeling methods are proposed, each with specific strengths and ranges of applicability. Each will be used to meet a stated objective. Taken as a group, they are intended to satisfy a broad range of needs. Moreover, each of these air quality modeling approaches depend to one extent or another on “foundation models” to provide information on emissions and meteorology. Emissions and meteorological models are an integral component of the suite of modeling approaches which will comprise a comprehensive modeling system for PM.

The proposed air quality modeling methods are:

MethodPeriod of applicationComponent

Grid-based modelingAnnual and episodic Primary and Secondary

Receptor modelingAnnual and episodicPrimary

Equilibrium modelingAnnual and episodicSecondary

Hybrid RollbackAnnualand appropriate Primary, perhaps

shorter-term averagessecondary for annual

The purpose of this document is to present the proposed methods, outline the data needed to support each method, and identify further research needs. While this document discusses the primary methods to be applied, it does not present a detailed discussion of methods and the specifics of model application. Additional details on model formulation can be found in recent PM modeling assessment documents (Seigneur 1997, Roth 1998) and the respective model users guides (Kumar, 1995; Kumar 1996; Dabdub 1998; Watson, 1990). The specifics of model application will be included when the final modeling protocol is prepared in late 1998 or early 1999.

2. GRID-BASED MODELING

2.1 Approach

Grid-based modeling techniques simulate the three-dimensional physical and chemical processes that govern PM formation. A number of models have been developed, many of which are extensions of photochemical models with additional modules to address aerosol species. These models include the California Institute of Technology (CIT) model, UAM-AERO, and SAQM-AERO. The CIT model has been applied extensively in the South Coast. UAM-AERO has been applied to several historical PM10 episodes in the San Joaquin Valley (Lurmann, 1996), and to the database collected during the 1995 Integrated Monitoring Study (IMS95) (Kaduwela, 1998). The SAQM-AERO model has been applied to several South Coast episodes and will be applied to several IMS95 episodes as well (Dabdub 1998). Because the core SAQM model was developed specifically for the San Joaquin Valley, its use is recommended for future aerosol modeling applications as part of the CRPAQS study.

SAQM-AERO represents an extension of the SARMAP Air Quality Model (SAQM) that includes provisions for simulating important physical and chemical processes that lead to aerosol formation. To develop SAQM-AERO, the aerosol module developed for the UAM-AERO model (Lurmann et al., 1997; Kumar et al., 1995) was incorporated into the SAQM model. The model is computationally efficient (at least when it is run with a single aerosol size section) and incorporates a level of physical and chemical detail that is appropriate for grid models. The description of SAQM-AERO reported herein is taken from the draft report by Dabdub et al. (1998).

Since the basic features of SAQM are described elsewhere, this description focuses primarily on those aspects of SAQM-AERO concerning the treatment of aerosols. The major features of the aerosol module are:

  • Simulation of the aerosol concentrations of all the major primary and secondary components of atmospheric particulate matter (PM), including sulfate, nitrate, ammonium, chloride, sodium, elemental carbon, organic carbon, water, and other crustal material.
  • A sectional approach for characterization of the continuous aerosol size distribution, typically extenting from 0.01 to 10 um for aerosols and from 0.01 to 30 um when fog droplets are present, with user-specified size bins. The model can also be applied with a single aerosol size bin.
  • The internally mixed aerosol assumption, where all particles in a specific size range are assumed to have the same chemical composition.
  • An algorithm to simulate the mass transfer occurring between the gaseous and aerosol species during condensation and evaporation. The effects of nucleation and coagulation are ignored in the algorithm.
  • An algorithm to simulate the distribution of aerosol species concentrations based on the thermodynamics of the sulfate/nitrate/chloride/ammonium/sodium/water chemical system encoded in the SEQUILIB aerosol module (Pilinis and Seinfeld, 1987; Pandis, 1996).
  • Production of condensible organic species from oxidation of gaseous organic compounds based on the organic aerosol yields reported by Pandis et al. (1992).
  • An algorithm to approximate effects of fogwater condensation and evaporation on the growth and shrinkage of the aerosol/fog droplet size distribution.
  • An algorithm to simulate the particle deposition and gravitational settling for particles of various sizes.
  • Incorporation of ammonia (NH3) and hydrochloric acid (HCl) as gas-phase species in the model. These species only interact with the aerosol phase because the gas-phase reactions of ammonia and HCl are of negligible importance relative to their interactions with the aerosol phase.

2.1.1 Treatment of Aerosol Thermodynamics

The inorganic and organic aerosol species are distributed among the aerosol and gas phases by assuming that thermodynamic equilibrium is established over time scales smaller than the 5- to 15-minute operator-splitting time step used in the model. Testing was performed to determine the practicality of including detailed mass transfer calculations in the aerosol module, and the results suggested the computational burden for simulating the detailed mass transfer calculation was large and impractical (Wexler et al., 1994). Thus, the gas-aerosol equilibrium assumption is employed in the model, despite the potential error introduced in certain cases.

The inorganic multicomponent atmospheric aerosol equilibrium model, SEQUILIB, of Pilinis and Seinfeld (1987) with recent updates (Pandis, 1996) is used for the calculation of the total quantities of ammonium, chloride, nitrate, and water contained in atmospheric particles. The model predicts the gas-phase concentrations of NH3, HCl, HNO3, and the aerosol-phase concentrations of H2O, NH4+, SO4=, NO3-, Na+, Cl-, HSO4-, H2SO4, Na2SO4, NaHSO4, NaCl, NaNO3, NH4Cl, NH4NO3, (NH4)2SO4, NH4HSO4, and (NH4)3H(SO4)2 using a chemical mechanism involving 13 equilibrium reactions. It uses the Bromley method to obtain multicomponent activity coefficients (Bromley, 1973) and the Pitzer method to obtain the binary activity coefficients (Pitzer, 1979). Kim et al. (1993a, 1993b) suggest that the Pitzer method is more accurate than the Bromley method for multicomponent activity coefficients, and that the K-M method (Kusik and Meissner, 1978) may be more accurate than the Pitzer method for binary activity coefficients. Given the paucity of high-concentration laboratory data with which to evaluate their performance, the activity coefficient calculation methods originally coded in SEQUILIB were used. The water activity coefficients are obtained using the ZSR method (Stokes and Robinson, 1966). The equilibrium code has been relatively successful in predicting the concentrations of the various aerosol species in the South Coast Air Basin (SoCAB) (Pilinis and Seinfeld, 1987, 1988) and elsewhere (Watson et al., 1994).

Thermodynamic equilibrium is also assumed for the condensable organic vapors. When their gas-phase concentrations exceed their vapor pressure, the vapors condense to the aerosol phase in an effort to establish equilibrium. Evaporation occurs when the gas phase is subsaturated. Following Pandis et al. (1992), the aerosol module assumes a negligibly small saturation vapor pressure (0.01 ppt), which essentially places all of the condensable organic material in the aerosol phase. Due to the physical and chemical uncertainties in the secondary organic aerosol species, no attempt is made to estimate the amount of water absorbed or desorbed by the organic particles. Saxena et al. (1995) have shown that condensed organic species can alter the hygroscopic behavior of atmospheric particles, and alterations may be positive or negative depending on the location (nonurban or urban). These differential water absorption effects are not included in the model.

2.1.2 Treatment of the Aerosol Size Distribution

The model can be applied using one or more size bins. In theory, simulation of the aerosol size distribution is necessary to accurately simulate the physical and chemical evolution of the aerosol, and the aerosol removal by deposition. When the model is applied to simulate the aerosol size distribution, it is generally recommended that the model be run with at least eight sections below 10 um and one section above 10 um if fogs occur in the simulation period. Usually these sections are logarithmically spaced; however, the aerosol model algorithm can accommodate arbitrarily spaced size bins.

Using more than eight or nine sections is highly desirable, but the user should expect a proportional increase in CPU time for simulations. With the numerical methods incorporated in the model, using fewer than eight aerosol size sections can lead to excessive pseudo-diffusion of particles between size bins. For ambient PM modeling, it is important to include size sections to represent the dominant aerosol modes: the nuclei mode, the accumulation mode, and the coarse mode. The nuclei mode corresponds to particles below 0.1 um in diameter and is associated with fresh combustion emissions. The accumulation mode corresponds to particles approximately 0.1 to 2 um in diameter and is associated with particles originating from aged combustion sources, photochemical processes, and smaller fog or cloud droplets. The coarse mode corresponds to particles above 2 um and is associated with wind blown dust, mechanically generated aerosols, and larger fog and cloud droplets. Most of the surface area of ambient particles occurs in the particles with diameters below 0.2 or 0.3 um, and it is important to include sufficient size resolution to represent these particles.

There is a significant computational burden associated with simulation of the aerosol size distribution. For example, increasing the number of aerosol size sections from one to nine increases the overall model execution time by a factor of 5 to 10. This large increase in CPU time occurs not only because the number of transported species increases (10 transported species for each bin), which affects all operators except the gas-phase chemistry, but also because the number of aerosol thermodynamic calculations increases in proportion to the number of size bins. In applications of UAM-AERO to the SoCAB, the PM10 response to regional changes in VOC, NOx, SO2, NH3, and particle emissions was quite similar in one and nine section simulations (Lurmann and Kumar, 1997). Thus, for PM10 analyses, the model can be run with a single aerosol size section to explore various emission control options (i.e., screening runs). Subsequent refined simulations can be made that include size resolution for the most important emission control strategies. Incorporation of aerosol size resolution is recommended for PM2.5 and visibility analyses.

The aerosol module implemented in SAQM uses the internally mixed assumption for aerosol composition. The aerosol size composition is discretized in size sections, and all particles in each section are assumed to have the same chemical composition (Gelbard et al., 1980; Seigneur et al., 1986). The movement of these sections in the size coordinate, as a result of particle growth and shrinkage (i.e., gas-to-particle conversion, condensation, or evaporation), is initially calculated using the moving section technique (Gelbard, 1990; Kim and Seinfeld, 1990). With the moving section technique, the number of particles in each size bin is constant during the aerosol transport step, and the changes in mass due to condensation or evaporation are reflected in new mass mean diameters for the sections. In some situations where a large amount of mass is being transported between the gas and aerosol phases, multiple aerosol transport steps are taken to assure numerical stability. However, because the three-dimensional air quality model requires fixed aerosol size bins for the advection and diffusion steps, the mass in the new size distribution is reallocated to the original size bins using a mass-conserving cubic spline-fitting procedure.

To predict the size distribution of the condensable compounds, the model first calculates the gas-phase concentrations of these compounds resulting from transport and chemical reactions. For the inorganic compounds, the equilibrium concentrations of the total aerosol and vapors are determined from SEQUILIB. The amount condensed or evaporated is partitioned among the sections in accordance with a relationship involving the particle diameter, molecular diffusivity of the condensing or evaporating compound, the difference between the ambient concentration and the equilibrium particle surface concentration of the compound, the mean free path of air, and the accommodation coefficient. SEQUILIB is then used again to obtain improved estimates of the water content of each aerosol section. As a result of the condensation or evaporation, the aerosol size sections grow or shrink. Then, the mass in the new size distribution is reallocated to the original size bins using a mass-conserving cubic spline-fitting procedure. The cubic spline reallocation procedure is numerically robust; however, it introduces some pseudo dispersion into the size distributions. That is, the predicted size distributions are somewhat smoother or broader than may exist in the ambient atmosphere. During periods of rapidly increasing or decreasing moisture, the gas-aerosol transfer and resizing is performed using small time steps to ensure the size distribution evolves in a stable manner.

2.1.3 Treatment of Secondary Organic Aerosol Production

The CB-IV chemical mechanism was extended to include production of condensable organic species from higher molecular weight (C5+) gaseous VOCs. The condensable organic compound (COC) yields for the lumped organic compounds are obtained from the database of individual compound yields reported by Pandis et al. (1992) and the composition of the regional VOC emission inventories. The knowledge of the chemical composition of most condensable vapor products and the exact chemical pathways leading to their formation, including stoichiometry and rate constants, remains incomplete. Therefore, the mechanistic description of the production of low-volatility products follows the condensed gas-phase mechanisms used in regional photochemical models. The atmospheric oxidation of a hydrocarbon, HC, by an oxidant like OH, O3, or NO3 is described by a single reaction that incorporates all the individual mechanistic steps