AIR QUALITY MODELING

Overview

Visibility impairment occurs when fine particulate matter (PM2.5) in the
atmosphere scatters and absorbs light, thereby creating haze. PM2.5 can
be emitted into the atmosphere directly as primary particulates, or it
can be produced in the atmosphere from photochemical reactions of
gas-phase precursors and subsequent condensation to form secondary
particulates. Examples of primary PM2.5 include crustal materials and
elemental carbon; examples of secondary PM include ammonium nitrate,
ammonium sulfates, and secondary organic aerosols (SOA). Secondary PM2.5
is generally smaller than primary PM2.5, and because the ability of
PM2.5 to scatter light depends on particle size, with light scattering
for fine particles being greater than for coarse particles, secondary
PM2.5 plays an especially important role in visibility impairment.
Moreover, the smaller secondary PM2.5 can remain suspended in the
atmosphere for longer periods and is transported long distances, thereby
contributing to regional-scale impacts of pollutant emissions on
visibility. 

The sources of PM2.5 are difficult to quantify because of the complex
nature of their formation, transport, and removal from the atmosphere.
This makes it difficult to simply use emissions data to determine which
pollutants should be controlled to most effectively improve visibility.
Photochemical air quality models offer opportunity to better understand
the sources of PM2.5 by simulating the emissions of pollutants and the
formation, transport, and deposition of PM2.5. If an air quality model
performs well for a historical episode, the model may then be useful for
identifying the sources of PM2.5 and helping to select the most
effective emissions reduction strategies for attaining visibility goals.
Although several types of air quality modeling systems are available,
the gridded, three-dimensional, Eulerian models provide the most
complete spatial representation and the most comprehensive
representation of processes affecting PM2.5, especially for situations
in which multiple pollutant sources interact to form PM2.5. For less
complex situations in which a few large point sources of emissions are
the dominant source of PM2.5, trajectory models (such as the California
Puff Model [CALPUFF]) may also be useful for simulating PM2.5.

Air Quality Models

The WRAP RMC utilized two regulatory air quality modeling systems to
conduct all regional haze modeling.  A brief discussion of each of these
models is provided below.

Community Multi-Scale Air Quality Model 

EPA initially developed the Community Multi-Scale Air Quality (CMAQ)
modeling system in the late 1990s. The model source code and supporting
data can be downloaded from the Community Modeling and Analysis System
(CMAS) Center (  HYPERLINK "http://www.cmascenter.org/" 
http://www.cmascenter.org/ ), which is funded by EPA to distribute and
provide limited support for CMAQ users. CMAQ was designed as a “one
atmosphere” modeling system to encompass modeling of multiple
pollutants and issues, including ozone, PM, visibility, and air toxics.
This is in contrast to many earlier air quality models that focused on
single-pollutant issues (e.g., ozone modeling by the Urban Airshed
Model). CMAQ is an Eulerian model—that is, it is a grid-based model in
which the frame of reference is a fixed, three-dimensional (3-D) grid
with uniformly sized horizontal grid cells and variable vertical layer
thicknesses. The number and size of grid cells and the number and
thicknesses of layers are defined by the user, based in part on the size
of the modeling domain to be used for each modeling project. The key
science processes included in CMAQ are emissions, advection and
dispersion, photochemical transformation, aerosol thermodynamics and
phase transfer, aqueous chemistry, and wet and dry deposition of trace
species. CMAQ offers a variety of choices in the numerical algorithms
for treating many of these processes, and it is designed so that new
algorithms can be included in the model. CMAQ offers a choice of three
photochemical mechanisms for solving gas-phase chemistry: the Regional
Acid Deposition Mechanism version 2 (RADM2), a fixed coefficient version
of the SAPRC90 mechanism, and the Carbon Bond IV mechanism (CB-IV). 

Comprehensive Air Quality Model with Extensions 

The Comprehensive Air Quality Model with extensions (CAMx) model was
initially developed by ENVIRON in the late 1990s as a nested-grid,
gas-phase, Eulerian photochemical grid model. ENVIRON later revised CAMx
to treat PM, visibility, and air toxics. While there are many
similarities between the CMAQ and CAMx systems, there are also some
significant differences in their treatment of advection, dispersion,
aerosol formation, and dry and wet deposition.

Model Versions

Both EPA and ENVIRON periodically update and revise their models as new
science or other improvements to the models are developed. For CMAQ, EPA
typically provides a new release about once per year. The initial 2002
MPE for WRAP used CMAQ version 4.4, which was released in October 2004.
In October 2005 EPA released CMAQ version 4.5, which includes the
following updates and improvements to the modeling system:

A new vertical advection algorithm with improved mass conservation

Changes in deposition velocities for some PM species

A new sea-salt emissions model and inclusion of sea salt in the aerosol
thermodynamics

An option to make vertical mixing parameters vary as a function of land
use type

The RMC completed the initial CMAQ MPE using CMAQ v.4.4. When version
4.5 was released in October, the modeling was revised and a comparison
of the model performance using the two versions was compared.  Note that
some of the new features in CMAQ v4.5 (e.g., sea salt in the AE4 aerosol
dynamics module, and percent urban minimum vertical diffusivity) require
the reprocessing of the MM5 data using the new version of MCIP (MCIP
v3.0). However, because such reprocessing could potentially jeopardize
the WRAP modeling schedule, WRAP elected to operate CMAQ v4.5 using the
MM5 data processed using a previous MCIP version, MCIP v2.3, and the AE3
aerosol module that does not include active sea salt chemistry.

ENVIRON releases updated versions of CAMx approximately every two years,
or as new features become available. The version used for the comparison
of CMAQ and CAMx was CAMx v4.3.  There are many similarities between
CMAQ and CAMx regarding the science algorithms and chemical mechanisms
used, including the CB-IV gas-phase and RADM aqueous-phase chemistries,
ISORROPIA aerosol thermodynamics, and PPM horizontal advection scheme.
In the past, the treatment of vertical advection was a major difference
between the two models; however, the incorporation of the new mass
conservation scheme in CMAQ v4.5 makes its vertical advection algorithm
much more similar to that of CAMx. 

Major differences between the two models that still exist are in the
basic model code, in the treatment of horizontal diffusion SOA formation
mechanisms, and in grid nesting (CAMx supports one-way and two-way
nesting, whereas CMAQ supports just one-way grid nesting). Both models
include process analysis for the gas-phase portions of the model. The
publicly released version of CAMx supports ozone and PM source
apportionment through its Ozone and PM Source Apportionment Technology
(OSAT/PSAT) probing tools, while for CMAQ there are research versions of
the model that include Tagged Species Source Apportionment (TSSA) for
some PM species (e.g., sulfate and nitrate). There are also research
versions of CMAQ and CAMx that support the Decoupled Direct Method (DDM)
sensitivity tool for PM and ozone. 

The CAMx model is computationally more efficient than CMAQ. However,
CAMx is currently supported for use on only a single central processing
unit (CPU) and can perform multiprocessing using Open Multi-Processing
(OMP) parallelization (i.e., shared memory multiprocessors). CMAQ
parallelization, on the other hand, is implemented using Message Passing
Interface (MPI) multiprocessing and therefore can be run using any
number of CPUs. Depending on the number of model simulations to be
performed and the manner in which they are set up, there can be a slight
advantage either to CAMx or to CMAQ in regard to computational
efficiency.

Model Simulations

In support of the WRAP Regional Haze air quality modeling efforts, the
RMC developed air quality modeling inputs including annual meteorology
and emissions inventories for a 2002 actual emissions base case, a
planning case to represent the 2000-04 regional haze baseline period
using averages for key emissions categories, and a 2018 base case of
projected emissions determined using factors known at the end of 2005.
All emission inventories were developed using the Sparse Matrix Operator
Kernel Emissions (SMOKE) modeling system. Each of these inventories has
undergone a number of revisions throughout the development process to
arrive at the final versions used in CMAQ and CAMx air quality modeling.
 The development of each of these emission scenarios is documented under
the emissions inventory sections of the TSS.  In addition to various
sensitivities scenarios, the WRAP performed air quality model
simulations for each of the emissions scenarios as follows: 

The 2002 base case emissions scenario, referred to as “2002 Base
Case” or “Base02”.   The purpose of the Base02 inventory is to
represent the actual conditions in calendar year 2002 with respect to
ambient air quality and the associated sources of criteria and
particulate matter air pollutants.  The Base02 emissions inventories are
used to validate the air quality model and associated databases and to
demonstrate acceptable model performance with respect to replicating
observed particulate matter air quality. 

The 2000-04 baseline period planning case emissions scenario is referred
to as “Plan02”. The purpose of the Plan02 inventory is to represent
baseline emission patterns based on average, or “typical”,
conditions.  This inventory provides a basis for comparison with the
future year 2018 projected emissions, as well as to gauge reasonable
progress with respect to future year visibility.  

The 2018 future-year base case emissions scenario, referred to as
“2018 Base Case” or “Base18”.  These emissions are used to
represent conditions in future year 2018 with respect to sources of
criteria and particulate matter air pollutants, taking into
consideration growth and controls. Modeling results based on this
emission inventory are used to define the future year ambient air
quality and visibility metrics.

Data Sources

The CMAQ model requires inputs of three-dimensional gridded wind,
temperature, humidity, cloud/precipitation, and boundary layer
parameters.   The current version of CMAQ can only utilize output fields
from the PSU/NCAR MM5 meteorological model.  MM5 is a
state-of-the-science atmosphere model that has proven useful for air
quality applications and has been used extensively in past local, state,
regional, and national modeling efforts.  MM5 has undergone extensive
peer-review, with all of its components continually undergoing
development and scrutiny by the modeling community.  In-depth
descriptions of MM5 can be found in Dudhia (1993) and Grell et al.
(1994), and at   HYPERLINK "http://www.mmm.ucar.edu/mm5/" 
http://www.mmm.ucar.edu/mm5 .  All meteorological data used for the WRAP
air quality modeling efforts are derived from MM5 model simulations. 
The development of these data is documented in (Kemball-Cook, S. et al.,
2005)

Emission inventories for all WRAP air quality simulations were developed
using the Matrix Operator Kernel Emissions (SMOKE) modeling system.  The
development of these data has been discussed and documented elsewhere
(Tonnesen, G. et al., 2006)

Initial conditions (ICs) are specified by the user for the first day of
a model simulation. For continental-scale modeling using the RPO Unified
36-km domain, the ICs can affect model results for as many as 15 days,
although the effect typically becomes very small after about 7 days. A
model spin-up period is included in each simulation to eliminate any
effects from the ICs. For the WRAP modeling, the annual simulation is
divided into four quarters, and included a 15-day spin-up period for the
quarters beginning in April, July, and October. For the quarter
beginning in January 2002, a spin-up period covering December 16-31,
2001, using meteorology and emissions data developed for CENRAP were
used..

Boundary conditions (BCs) specify the concentrations of gas and PM
species at the four lateral boundaries of the model domain. BCs
determine the amounts of gas and PM species that are transported into
the model domain when winds flow is into the domain. Boundary conditions
have a much larger effect on model simulations than do ICs. For some
areas in the WRAP region and for clean conditions, the BCs can be a
substantial contributor to visibility impairment. 

For this study BC data generated in an annual simulation of the
global-scale GEOS-Chem model that was completed by Jacob et al. ( 
HYPERLINK "http://www-as.harvard.edu/chemistry/trop/geos/" 
http://www-as.harvard.edu/chemistry/trop/geos/ ) for calendar year 2002
were applied. Additional data processing of the GEOS-Chem data was
required before using them in CMAQ and CAMx. The data first had to be
mapped to the boundaries of the WRAP domain, and the gas and PM species
had to be remapped to a set of species used in the CMAQ and CAMx models.
This work was completed by Byun and coworkers (  HYPERLINK
"http://www-as.harvard.edu/chemistry/trop/geos/meetings/2005/ppt/Expandi
ng_Model_Capabilities/GEOS-CMAQ_april_4_Byun.ppt" 
http://www-as.harvard.edu/chemistry/trop/geos/meetings/2005/ppt/Expandin
g_Model_Capabilities/GEOS-CMAQ_april_4_Byun.ppt 

The CMAQ model options and configuration used for the WRAP 36-km model
simulations are described in Tonnesen, G. et al., 2006.

Model Run Specification Sheets

In order to provide documentation for each of the CMAQ and CAMx air
quality model simulations conducted by the WRAP RMC during Calendar year
2006, a series of Model Run Specification Sheets were developed.  These
“Spec Sheets” provide a description of each simulation, the various
air quality model options and configurations used and detailed listing
and description of the meteorological data and emission inventories for
each scenario.  These Spec Sheets also provide a means for the RMC to
track the development of each of the input data sets and defined the
modeling schedule.  The purpose of each simulation, and expected
results, including their implications, are also included.  A link to
each of the individual Specification Sheets for the model simulations
can be found on the RMC web site at:    HYPERLINK
"http://pah.cert.ucr.edu/aqm/308/cmaq.shtml" 
http://pah.cert.ucr.edu/aqm/308/cmaq.shtml .

2002 Base Case Modeling

Base02 Sensitivity Simulations

The purpose of the 2002 Base Case modeling efforts was to evaluate air
quality/visibility modeling systems for a historical episode—in this
case, for calendar year 2002—to demonstrate the suitability of the
modeling systems for subsequent planning, sensitivity, and emissions
control strategy modeling. Model performance evaluation is performed by
comparing output from model simulations with ambient air quality data
for the same time period. After creating emissions and meteorology
inputs for the two air quality models, CMAQ and CAMx, the next step was
to perform the visibility modeling and the model performance
evaluations, which are described below. A detailed discussion of the
results of the CMAQ and CAMx model simulations can be found in Tonnesen,
G. et al., 2006.  Also documented in Tonnesen, G. et al., 2006 are the
results of the model performance evaluation, a model inter-comparison
and discussion of various sensitivity simulations. This information was
used as the basis for recommending the selection of CMAQ and/or CAMx to
complete the remaining modeling efforts in RMC’s support of WRAP. 

Model Performance Evaluation

The objective of a model performance evaluation (MPE) is to compare
model-simulated concentrations with observed data to determine whether
the model’s performance is sufficiently accurate to justify using the
model for simulating future conditions. There are a number of challenges
in completing an annual MPE for regional haze. The model must be
compared to ambient data from several different monitoring networks for
both PM and gaseous species, for an annual time period, and for a large
number of sites. The model must be evaluated for both the worst
visibility conditions and for very clean conditions. Finally, final
guidance on how to perform an MPE for fine-particulate models is not yet
available from EPA. Therefore, the RMC experimented with many different
approaches for showing model performance results. The plot types that
were found to be the most useful are the following:

Time-series plots comparing the measured and model-predicted species
concentrations

Scatter plots showing model predictions on the y-axis and ambient data
on the x-axis

Spatial analysis plots with ambient data overlaid on model predictions

Bar plots comparing the mean fractional bias (MFB) or mean fractional
error (MFE) performance metrics 

“Bugle plots” showing how model performance varies as a function of
the PM species concentration

Stacked-bar plots of contributions to light extinction for the average
of the best-20% visibility days or the worst-20% visibility days at each
site; the higher the light extinction, the lower the visibility

Examples of each of these MPE metrics and analysis products can be found
in Tonnesen, G. et al., 2006.  The results of the MPE are available from
the WRAP RMC website (http://pah.cert.ucr.edu/aqm/308/eval.shtml) 

2002 Planning Scenario

The 2000-04 baseline period planning case scenario is referred to as
“Plan02”. The purpose of the Plan02 scenario is to simulation the
air quality representative of baseline emission patterns based on
average, or “typical”, conditions.  This scenario provides a basis
for comparison with the future year 2018 scenario based on projected
emissions, as well as to gauge reasonable progress with respect to
future year visibility.  

Plan02 Simulations Input Data 

Input data used for the 2002 Planning model simulations consisted of the
same meteorology as for the 2002 Base Case and the Plan02 emission
inventories described under the Emissions Modeling section of the TSS.  

The setup of the CMAQ model (including science options, run scripts,
simulation periods, and ancillary data) for the Plan02 cases was
identical to that used in the Base02 modeling, as  described in the 2002
MPE report (Tonnesen et al., 2006). In summary, CMAQ v4.5 (released by
EPA in October 2005) was used on the RPO Unified 36-km domain. The
Carbon Bond Mechanism version 4 (CB4) with RADM aqueous chemistry, the
SORGAM organic aerosol algorithm, and all other science algorithms
detailed in Tonnesen et al., 2006 were used. Initial condition (IC) data
for January 1, 2002, were developed using a 15-day spin-up period
(December 16-31, 2001). Boundary condition (BC) data were generated in
an annual simulation of the global-scale GEOS-Chem model that was
completed by Jacob et al. (  HYPERLINK
"http://www-as.harvard.edu/chemistry/trop/geos/" 
http://www-as.harvard.edu/chemistry/trop/geos/ ) for calendar year 2002.


Comparison With Base02 Simulations

For each of the three Plan02 emissions datasets, annual visibility
modeling was performed using the CMAQ model. This was a key aspect of
the QA procedure, since errors in the emissions inventories that might
not be apparent during the emissions QA steps might be more readily
detected in the results from the CMAQ modeling. 

In our initial analysis of the Plan02 scenario, plots were prepared for
QA purposes that compared the Plan02a CMAQ results with the Base02a CMAQ
results for daily and monthly averages. After revising Plan02a to create
Plan02b and Plan02c, additional QA plots were prepared to compare the
CMAQ results of each revised Plan02 case to the previous iteration.
These were prepared as Program for the Analysis and Visualization of
Environmental data (PAVE) spatial plots showing the change in individual
PM2.5 species concentrations as daily, monthly, and annual averages. The
final set of analysis products,  available on the RMC web site, include
PAVE difference plots comparing the CMAQ-predicted annual average
species concentrations from the Plan02c case with those from the Base02b
case. Note that these plots are not useful for visibility planning
purposes, but are being provided to show the magnitudes of changes when
moving from the 2002 Base Case to the 2002 Planning Case—in other
words, from the actual emissions for the year 2002 to the
“typical-year” emissions created for the final Plan02 scenario. The
primary analysis “product” from the Plan02 CMAQ modeling is the use
of its output in combination with the CMAQ output from the 2018 modeling
to develop the visibility progress calculations and glide path plots,
described below. 

2018 Model Simulations

The 2018 future-year base case scenario is referred to as “2018 Base
Case” or “Base18”.  The purpose of the Base18 scenario is to
simulation the air quality representative of conditions in future year
2018 with respect to sources of criteria and particulate matter air
pollutants, taking into consideration growth and controls. Modeling
results based on this emission inventory are used to define the future
year ambient air quality and visibility metrics.

Base18 Simulation Input Data 

Input data used for the 2018 Base Case model simulations consisted of
the same meteorology as for the 2002 Base Case and the Base18 emission
inventories described under the Emissions Modeling section of the TSS.  

The setup of the CMAQ model (including science options, run scripts,
simulation periods, and ancillary data) for the Base18 cases was
identical to that used in the Base02 modeling, as  described in the 2002
MPE report (Tonnesen et al., 2006). In summary, CMAQ v4.5 (released by
EPA in October 2005) was used on the RPO Unified 36-km domain. The
Carbon Bond Mechanism version 4 (CB4) with RADM aqueous chemistry, the
SORGAM organic aerosol algorithm, and all other science algorithms
detailed in Tonnesen et al., 2006 were used. Initial condition (IC) data
for January 1, 2002, were developed using a 15-day spin-up period
(December 16-31, 2001). Boundary condition (BC) data were generated in
an annual simulation of the global-scale GEOS-Chem model that was
completed by Jacob et al. (  HYPERLINK
"http://www-as.harvard.edu/chemistry/trop/geos/" 
http://www-as.harvard.edu/chemistry/trop/geos/ ) for calendar year 2002.


Base18 Simulation Results

The purpose of modeling 2018 visibility is to compare the 2018
visibility predictions to the 2002 typical-year visibility modeling
results, as discussed below. Some improvements in visibility by 2018 are
expected because of reductions in emissions due to currently planned
regulations and technology improvements. A brief summary is provided
here of the comparison between the 2018 and 2002 results using annual
average PAVE spatial plots. The goal of this summary is to convey the
scale and spatial extent of changes in key PM2.5 species from 2002 to
2018. For planning purposes, on the other hand, states and tribes should
focus on the visibility projections and glide path calculations at
individual Class I Areas. 

Figures 1 through 4 show the annual average concentrations for sulfate,
nitrate, PM2.5 and model-reconstructed visibility (in deciviews),
respectively. In each figure, the bottom two plots show the modeled
concentration or deciviews for the Plan02b and Base18b cases, while the
top plot shows the change in visibility calculated as Base18b minus
Plan02b. The Plan02b results are presented here instead of Plan02c
results because these plots had previously been prepared with version B.
As the differences between Plan02b and Plan02c are extremely small, new
plots prepared using Plan02c would be essentially identical to the
results in Figure 1 through 4.

In each of the top plots in the four figures, cool colors indicate areas
in which model-predicted visibility improved from 2002 to 2018, while
warm colors indicate areas where modeled visibility became worse over
that period. Figure 1 shows that reductions in sulfate were largest in
the southwest corner of the WRAP region and in Texas and Oklahoma. This
results from planned SOx emissions reductions in the CENRAP region.
There were smaller reductions in sulfate in the Los Angeles area,
western Washington state, and southern Nevada. There were small
increases of sulfate, mostly in Wyoming, due to growth in SOx emissions.
Most regions of the WRAP domain had low concentrations of sulfate in
2002 and little change in sulfate by 2018.

Figure 2 shows the results for nitrate. In the both 2002 and 2018, the
modeled nitrate was greatest in California, and there were reduction in
nitrate in that state in 2018 because of reductions in mobile-source NOx
emissions. There were small reductions in the Phoenix area as well, also
from reductions in mobile-source NOx emissions. 

Figure 3 shows the comparison of PM2.5 for 2002 and 2018. In most areas
of the WRAP region, changes in PM2.5 were less than 1 (g/m3. Locations
with increases in PM2.5 correspond to areas of increased sulfate (see
Figure 3-1). Areas with the largest reductions in PM2.5 were the areas
in California that had large reductions in modeled nitrate in 2018 (see
Figure 3-2). Results for other species that contribute to PM2.5 are
available on the RMC web site at   HYPERLINK
"http://pah.cert.ucr.edu/aqm/308/cmaq.shtml#base18bvsplan02b" 
http://pah.cert.ucr.edu/aqm/308/cmaq.shtml#base18bvsplan02b .

Figure 4 compares model-reconstructed visibility for 2002 and 2018. Note
that these results are calculated using the modeled relative humidity
(RH), so they differ from the results that use site-specific monthly
average RH. Nonetheless, the results in Figure 4 are indicative of the
direction and magnitude of visibility changes in from 2002 to 2018.
Although the largest improvements are in California and the Pacific
Northwest, there were improvements throughout the WRAP region. The
change in deciviews is more dramatic than the change in PM2.5 mass
(Figure 3) because the visibility in deciviews is a relative metric, so
small mass changes in PM2.5 in good visibility areas can result in large
relative improvements in visibility.



Figure 1. Annual average aerosol sulfate (ASO4) concentration
comparisons between Base18b and Plan02b. Top plot: difference between
the two (Base18b – Plan02b);

bottom left plot: Plan02b results; bottom right plot: Base18b results.



Figure 2. Annual average aerosol nitrate (ANO3) concentration
comparisons between Base18b and Plan02b. Top plot: difference between
the two (Base18b – Plan02b);

bottom left plot: Plan02b results; bottom right plot: Base18b results.



Figure 3. Annual average PM2.5 concentration comparisons between Base18b

and Plan02b. Top plot: difference between the two (Base18b – Plan02b);

bottom left plot: Plan02b results; bottom right plot: Base18b results.



Figure 4. Annual average deciview comparisons between Base18b and
Plan02b.

Top plot: difference between the two (Base18b – Plan02b); bottom left

plot: Plan02b results; bottom right plot: Base18b results.

Visibility Projections

The Regional Haze Rule (RHR) goals include achieving natural visibility
conditions at 156 Federally mandated Class I areas by 2064. In more
specific terms, that RHR goal is defined as (1) visibility improvement
toward natural conditions for the 20% of days that have the worst
visibility (termed “20% worst,” or W20%, visibility days) and
(2) no worsening in visibility for the 20% of days that have the best
visibility (“20% best,” or B20%, visibility days). One component of
the states’ demonstration to EPA that they are making reasonable
progress toward this 2064 goal is the comparison of modeled visibility
projections for the first milestone year of 2018 with what is termed a
uniform rate of progress (URP) goal. As explained in detail below, the
2018 URP goal is obtained by constructing a “linear glide path” (in
deciviews) that has at one end the observed visibility conditions during
the mandated five-year (2000-2004) baseline period and at the other end
natural visibility conditions in 2064; the visibility value that occurs
on the glide path at year 2018 is the URP goal. 

Preliminary WRAP 2018 visibility projections have been made using the
Plan02c and Base18b CMAQ 36-km modeling results, following EPA guidance
that recommends applying the modeling results in a relative sense to
project future-year visibility conditions (U.S. EPA, 2001, 2003a, 2006).
Projections are made using relative response factors (RRFs), which are
defined as the ratio of the future-year modeling results to the
current-year modeling results. The calculated RRFs are applied to the
baseline observed visibility conditions to project future-year observed
visibility. These projections can then be used to assess the
effectiveness of the simulated emission control strategies that were
included in the future-year modeling. The major features of EPA’s
recommended visibility projections are as follows (U.S. EPA, 2003a,b,
2006):

Monitoring data should be used to define current air quality.

Monitored concentrations of PM10 are divided into six major components;
the first five are assumed to be PM2.5 and the sixth is PM2.5-10.

SO4 (sulfate)

NO3 (particulate nitrate)

OC (organic carbon)

EC (elemental carbon)

OF (other fine particulate or soil)

CM (coarse matter).

Models are used in a relative sense to develop RRFs between future and
current predicted concentrations of each component.

Component-specific RRFs are multiplied by current monitored values to
estimate future component concentrations.

Estimates of future component concentrations are consolidated to provide
an estimate of future air quality.

Future estimated air quality is compared with the goal for regional haze
to see whether the simulated control strategy would result in the goal
being met.

It is acceptable to assume that all measured sulfate is in the form of
ammonium sulfate [(NH4)2SO4] and all particulate nitrate is in the form
of ammonium nitrate [NH4NO3].

To facilitate tracking the progress toward visibility goals, two
important visibility parameters are required for each Class I area:

Baseline Conditions: “Baseline Conditions” represent visibility for
the B20% and W20% days for the initial five-year baseline period of the
regional haze program. Baseline Conditions are calculated using
monitoring data collected during the 2000-2004 five-year period and are
the starting point in 2004 for the uniform rate of progress (URP) glide
path to Natural Conditions in 2064 (U.S. EPA, 2003a).

Natural Conditions: “Natural Conditions,” the RHR goal for 2064 for
the Federally mandated Class I areas, represent estimates of natural
visibility conditions for the B20% and W20% days at a given Class I
area.

Baseline Conditions

Baseline Conditions for Class I areas are calculated using fine and
coarse PM concentrations measured at Interagency Monitoring of Protected
Visual Environments (IMPROVE) monitors (Malm et al., 2000). Each Class I
area in the WRAP domain has an associated IMPROVE PM monitor. The
IMPROVE monitors do not measure visibility directly, but instead measure
speciated fine particulate (PM2.5) and total PM2.5 and PM10 mass
concentrations from which visibility is calculated using the IMPROVE
aerosol extinction equation, discussed later. 

Visibility conditions are estimated starting with the IMPROVE 24-h
average PM mass measurements related to six PM components of light
extinction:

Sulfate [(NH4)2SO4]

Particulate nitrate [(NH4NO3]

Organic matter [OMC]

Light-absorbing carbon [LAC] or elemental carbon [EC]

Soil

Coarse matter [CM]

The IMPROVE monitors do not directly measure some of these species, so
assumptions are made as to how the IMPROVE measurements can be adjusted
and combined to obtain these six components. For example, sulfate and
particulate nitrate are assumed to be completely neutralized by ammonium
and only the fine mode (PM2.5) is speciated to obtain sulfate and
nitrate measurements (that is, any coarse-mode sulfate and nitrate in
the real atmosphere may be present in the IMPROVE CM measurement).
Concentrations for the above six components of light extinction in the
IMPROVE aerosol extinction equation are obtained from the IMPROVE
measured species using the formulas shown in Table 1.

Table 1. Definition of IMPROVE components from measured species.

IMPROVE Component	Calculation of Component from IMPROVE Measured Species

Sulfate	1.375 x (3 x S)

Nitrate	1.29 x NO3-

OMC	1.4 x OC

LAC	EC

Soil	(2.2 x Al) + (2.49 x Si) + (1.63 x Ca) + (2.42 x Fe) + (1.94 x Ti)

CM	MT – MF



where

S is elemental sulfur as determined from proton-induced x-ray emissions
(PIXE) analysis of the IMPROVE Module A. To estimate the mass of the
sulfate ion (SO4=), S is multiplied by 3 to account for the presence of
oxygen. If S is missing then the sulfate (SO4) measured by ion
chromatography analysis of Module B is used to replace (3 x S). For the
IMPROVE aerosol extinction calculation, sulfate is assumed to be
completely neutralized by ammonium (1.375 x SO4).

NO3- is the particulate nitrate measured by ion chromatography analysis
of Module B. For the IMPROVE aerosol extinction calculation, it is
assumed to be completely neutralized by ammonium (1.29 x NO3).

The IMPROVE organic carbon (OC) measurements are multiplied by 1.4 to
obtain organic matter (OMC), which adjusts the OC mass for other
elements assumed to be associated with OC.

Elemental carbon (EC) is also referred to as light-absorbing carbon
(LAC).

Soil is determined as a sum of the masses of those elements (measured by
PIXE) predominantly associated with soil (Al, Si, Ca, Fe, K, and Ti),
adjusted to account for oxygen associated with the common oxide forms.
Because K is also a product of the combustion of vegetation, it is
represented in the formula by 0.6 x Fe and is not shown explicitly.

MT and MF are total PM10 and PM2.5 mass, respectively. 

Associated with each PM species is an extinction efficiency that
converts concentrations (in (g/m3) to light extinction (in inverse
megameters, Mm-1), as listed below. Sulfate and nitrate are hygroscopic,
so relative humidity (RH) adjustment factors, f(RH), are used to
increase the particles’ extinction efficiency with increasing RH; this
accounts for the particles’ taking on water and having greater light
scattering. Note that some organic matter (OMC) compounds may also have
hygroscopic properties, but the IMPROVE aerosol extinction equation
assumes OMC is nonhygroscopic.

βSulfate	=	3 x f(RH) x [sulfate]

βNitrate	=	3 x f(RH) x [nitrate]

βOM	=	4 x [OMC]

βEC	=	10 x [EC]

βSoil	=	1 x [soil]

βCM	=	0.6 x [CM]

The total light extinction (βext) is assumed to be the sum of the light
extinctions due to the six PM species listed above plus Rayleigh (blue
sky) background extinction (βRay), which is assumed to be 10 Mm-1. This
is reflected in the IMPROVE extinction equation:

βext 	=	βRay + bSulfate + βNitrate + βEC +βOMC + βSoil + βCM

The total light extinction (βext) in Mm-1 is related to visual range
(VR) in kilometers using the following relationship:

VR	=	3912 / βext

The RHR requires that visibility be expressed in terms of a haze index
(HI) in units of deciview (dv), which is calculated as follows:

HI	=	10 ln(βext/10)

The equations above, with measurements from the associated IMPROVE
monitor, are used to estimate the daily average visibility at each Class
I area for each IMPROVE monitored day. For each year from the 2000-2004
baseline period, these daily average visibility values are then ranked
from highest to lowest. The “worst days” visibility for each of the
five years in the baseline period is defined as the average visibility
across the 20% worst-visibility days (highest deciview values);
similarly, the “best days” visibility is defined as the average
visibility across the 20% best-visibility days (lowest deciview values)
for each year. The Baseline Conditions for the best and worst days are
defined as the five-year average of the B20% visibility days and of the
W20% visibility days, respectively, across the five-year baseline
period. 

The set of equations given above for relating measured PM species to
visibility (light extinction) are referred to as the “Old IMPROVE”
equation. The IMPROVE Steering Committee has developed a “New
IMPROVE” equation that they believe better represents the fit between
measured PM species concentrations and visibility impairment. Although
conceptually similar to the Old IMPROVE equation, the New IMPROVE
equation includes updates to many of the parameters and the addition of
extinctions due to NO2 absorption and sea salt. 2018 visibility
projections and comparisons with the URP glide path goals were performed
using both the New and Old IMPROVE equations. The reader is referred
elsewhere for details on the New IMPROVE extinction equation (e.g., EPA,
2006a,b).

Mapping Model Results to IMPROVE Measurements

As noted above, future-year visibility at Class I areas is projected by
using modeling results in a relative sense to scale current observed
visibility for the B20% and W20% visibility days. This scaling is done
using RRFs, the ratios of future-year modeling results to current-year
results. Each of the six components of light extinction in the IMPROVE
reconstructed mass extinction equation is scaled separately. Because the
modeled species do not exactly match up with the IMPROVE measured PM
species, assumptions must be made to map the modeled PM species to the
IMPROVE measured species for the purpose of projecting visibility
improvements. For example, in the model’s chemistry (which explicitly
simulates ammonium), sulfate may or may not be fully neutralized; the
IMPROVE extinction equation, on the other hand, assumes that observed
sulfate is fully neutralized by ammonium. For the CMAQ v4.5 model
(September 2005 release) used in the WRAP RMC modeling, the mapping of
modeled species to IMPROVE measured PM species is listed in Table 2. 

Table 2. Mapping of CMAQ v4.5 modeled species concentrations

to IMPROVE measured components.

IMPROVE Component	CMAQ V4.3 Species 

Sulfate	1.375 x (ASO4J + ASO4I)

Nitrate	1.29 x (ANO3J + ANO3I)

OMC	AORGAJ + AORGAI + AORGPAJ + AORGPAI + AORGBJ + AORGBI

LAC	AECJ + AECI

Soil	A25J + A25I

CM	ACORS + ASEAS + ASOIL 



Projecting Visibility Changes Using Modeling Results

RRFs calculated from modeling results can be used to project future-year
visibility. For the urrent modeling efforts, RRFs are the ratio of the
2018 modeling results to the 2002 modeling results, and are specific to
each Class I area and each PM species. RRFs are applied to the Baseline
Condition observed PM species levels to project future-year PM levels,
which are then used with the IMPROVE extinction equation listed above to
assess visibility. The following six steps are used to project
future-year visibility for the B20% and W20% visibility days (the
discussion below is for W20% days but also applies to B20% days):

For each Class I area and each monitored day, daily visibility is ranked
using IMPROVE data and IMPROVE extinction equation for each year from
the five-year baseline period (2000-2004) to identify the W20%
visibility days for each year.

Use an air quality model to simulate a base-year period (ideally
2000-2004, but in reality just 2002) and a future year (e.g., 2018),
then apply the resulting information to develop Class-I-area-specific
RRFs for each of the six components of light extinction in the IMPROVE
aerosol extinction equation.

Multiply the RRFs by the measured 24-h PM data for each day from the
W20% days for each year from the five-year baseline period to obtain
projected future-year (2018) 24-h PM concentrations for the W20% days.

Compute the future-year daily extinction using the IMPROVE aerosol
extinction equation and the projected PM concentrations for each of the
W20% days in the five-year baseline from Step 3.

For each of the W20% days within each year of the five-year baseline,
convert the future-year daily extinction to units of deciview and
average the daily deciview values within each of the five years
separately to obtain five years of average deciview visibility for the
W20% days.

Average the five years of average deciview visibility to obtain the
future-year visibility Haze Index estimate that is compared with the
2018 progress goal.

In calculating the RRFs, EPA draft guidance (U.S. EPA, 2001, 2006a)
recommends selecting modeled PM species concentrations “near” the
monitor by taking a spatial average of PM concentrations across a
grid-cell-resolution–dependent NX by NY array of cells centered on the
grid containing the monitor. For the WRAP 36-km CMAQ modeling, the model
estimates for just the grid cell containing the monitor are used (i.e.,
NX=NY=1). 

 For the preliminary 2018 visibility projections, results are presented
only for “Method 1,” which is the recommended approach in EPA’s
draft modeling guidance documents (U.S. EPA, 2001, 2006a). In the Method
1 Average RRF Approach, an average RRF for the W20% days from 2002
(Modeled Worst Days) is obtained for the Plan02c and the Base18b CMAQ
simulations by averaging the PM concentration components across the
Modeled Worst Days and then calculating the (future year):(base year)
ratio of the average PM concentrations. For example, if SO4i,j is the
measured sulfate concentrations at Class I area j for the i=1,…,N 20%
worst visibility days in 2002, then the RRF for sulfate on the W20% days
would be obtained as:

 

For each Class I area and each of the W20% days, the average RRF for
each PM component would be applied to concentrations for the W20% days
from the 2000-2004 baseline period to estimate future-year PM
concentrations for each of the W20% days. Extinction and HI would then
be calculated to obtain the projected future-year visibility conditions
using the procedures given previously. 

Glide Path to Natural Conditions

The presumptive visibility target for 2018 is the URP goal that is
obtained by constructing a linear glide path from the current Baseline
Conditions to Natural Conditions in 2064 (both expressed in deciviews).
For instance, Figure 5 displays an example visibility glide path for the
Grand Canyon National Park (GRCA) Class I area. EPA’s default Natural
Conditions value for the W20% days (U.S. EPA, 2003b), shown as the green
line, is the 2064 visibility goal at GRCA of 6.95 dv. The blue diamonds
at the left of the plot are the annual average current conditions, based
on IMPROVE observations for the W20% days as obtained from the
Visibility Information Exchange Web System (VIEWS) web site (  HYPERLINK
"http://vista.cira.colostate.edu/views/" 
http://vista.cira.colostate.edu/views/ ). These annual average
visibility values for the 20% worst days allow an assessment of trends
and the year-to-year variation in visibility. The Baseline Conditions
are the average of the W20% visibility from 2000-2004, which is the
starting point for the glide path in 2004 (12.04 dv for GRCA). A linear
URP from the Baseline Conditions in 2004 to Natural Conditions in 2064
(sloping pink line with triangles) is assumed, and the value on the
glide path at 2018 is the presumptive URP visibility target that the
modeled 2018 projections are compared against to judge progress. In this
example, the visibility progress goal in 2018 would be 10.85 dv. Meeting
this would require a 1.19 dv reduction in visibility by 2018 to meet
that milestone year’s visibility progress target at the Grand Canyon
National Park. 

Figure 5. Example of URP glide path using IMPROVE data from the Grand
Canyon

National Park for the W20% days and comparison with Base18b visibility
projections.

Preliminary Visibility Projection Results

For all of the WRAP Class I areas, the RMC performed preliminary 2018
visibility projections and compared them to the 2018 URP goals using the
Plan02c and Base18b CMAQ modeling results and the Old and New IMPROVE
equations. As an example, Figure 5 above compares the Base18b visibility
projections with the URP goal based on the glide path for GRCA and the
Old IMPROVE equation. To achieve the 2018 URP goal, the modeled 2018
visibility projection would have to show a 1.19 dv (=12.04-10.85)
reduction. However, the modeled 2018 visibility projection shows only a
0.33 dv (=12.04-11.71) reduction by 2018, which indicates that the
emission controls simulated in case Base18b would not achieve the
modeled URP goal; the 2018 visibility projection achieves only 28% of
the goal (28% = 100 x 0.33/1.19). Figure 6 displays the 2018 visibility
projections for all WRAP Class I areas, using both the Old and New
IMPROVE equations, expressed as a percentage of achieving the URP goal,
with values of 100% or greater achieving the goal. Using the procedures
outlined above, none of the WRAP Class I areas are projected to achieve
their URP goals. There are various reasons for this, such as the
presence of W20% days that are dominated by emissions from sources that
are not controllable, such as wildfires, dust, and/or international
transport. Additional analysis of these results and alternative
projection techniques are currently under study.

Figure 6. 2018 visibility projections at WRAP Class I areas expressed as
a

percent of achieving the 2018 URP goal using the Old and New IMPROVE

equation and the WRAP Base18c CMAQ 36-km modeling results.

PM Source Apportionment

Impairment of visibility in Class I areas is caused by a combination of
local air pollutants and regional pollutants that are transported long
distances. To develop effective visibility improvement strategies, the
WRAP member states and tribes need to know the relative contributions of
local and transported pollutants, and which emissions sources are
significant contributors to visibility impairment at a given Class I
area. 

A variety of modeling and data analysis methods can be used to perform
source apportionment of the PM observed at a given receptor site. Model
sensitivity simulations have been used in which a “base case” model
simulation is performed and then a particular source is “zeroed out”
of the emissions. The importance of that source is assessed by
evaluating the change in pollutants at the receptor site, calculated as
pollutant concentration in the sensitivity case minus that in the base
case. This approach is known as a “brute force” sensitivity because
a separate model run is required for each sensitivity. 

An alternative approach is to implement a mass-tracking algorithm in the
air quality model to explicitly track for a given emissions source the
chemical transformations, transport, and removal of the PM that was
formed from that source. Mass tracking methods have been implemented in
both the CMAQ and CAMx air quality models. Initial work completed by the
RMC during 2004 used the CMAQ Tagged Species Source Apportionment (TSSA)
method. Unfortunately, there were problems with mass conservation in the
version of CMAQ used in that study, and these affected the TSSA results.
A similar algorithm has been implemented in CAMx, the PM Source
Apportionment Technology (PSAT). Comparisons of TSSA and PSAT showed
that the results were qualitatively similar, that is, the relative
ranking of the most significant source contributors were similar for the
two methods. However, the total mass contributions differed. With
separate funding from EPA, UCR has implemented a version of TSSA in the
new CMAQ release (v4.5) that corrects the mass conservation error, but
given the uncertainty of the availability of this update, the CAMx/PSAT
source apportionment method was used for the WRAP modeling analysis. 

The main objective of applying CAMx/PSAT is to evaluate the regional
haze air quality for typical 2002 (Plan02c) and future-year 2018
(Base18b) conditions. These results are used

to assess the contributions of different geographic source regions
(e.g., states) and source categories to current (2002) and future (2018)
visibility impairment at Class I areas, to obtain improved understanding
of (1) the causes of the impairment and (2) which states are included in
the area of influence (AOI) of a given Class I area; and 

to identify the source regions and emissions categories that, if
controlled, would produce the greatest visibility improvements at a
Class I area.

CAMx/PSAT

The PM Source Apportionment Technology performs source apportionment
based on user-defined source groups. A source group is the combination
of a geographic source region and an emissions source category. Examples
of source regions include states, nonattainment areas, and counties.
Examples of source categories include mobile sources, biogenic sources,
and elevated point sources; PSAT can even focus on individual sources.
The user defines a geographic source region map to specify the source
regions of interest. He or she then inputs each source category as
separate, gridded low-level emissions and/or elevated-point-source
emissions. The model then determines each source group by overlaying the
source categories on the source region map. For further information,
please refer to the white paper on the features and capabilities of PSAT
(  HYPERLINK
"http://pah.cert.ucr.edu/aqm/308/reports/PSAT_White_Paper_111405_final_d
raft1.pdf" 
http://pah.cert.ucr.edu/aqm/308/reports/PSAT_White_Paper_111405_final_dr
aft1.pdf ), with additional details available in the CAMx user’s guide
(ENVIRON, 2005;   HYPERLINK "http://www.camx.com"  http://www.camx.com
).

PM source apportionment modeling was performed for aerosol sulfate (SO4)
and aerosol nitrate (NO3) and their related species (e.g., SO2, NO, NO2,
HNO3, NH3, and NH4). The PSAT simulations include 9 tracers, 18 source
regions, and 6 source groups. The computational cost for each of these
species differs because additional tracers must be used to track
chemical conversions of precursors to the secondary PM species SO4, NO3,
NH4, and secondary organic aerosols (SOA). Table 3 summarizes the
computer run time required for each species. The practical implication
of this table for WRAP is that it is much more expensive to perform PSAT
simulations for NO3 and especially for SOA than it is to perform
simulations for other species.

Table 3. Benchmarks for PSAT computational costs for each PM species.

Run time is for one day (01/02/2002) on the WRAP 36-km domain. 

Species	No. of Species Tracers	RAM Memory	Disk Storage per Day	Run Time
with

1 CPU

SO4	2	1.6 GB	1.1 GB	4.7 h/day

NO3	7	1.7 GB	2.6 GB	13.2 h/day

SO4 and NO3 combined	9	1.9 GB	3.3 GB	16.8 h/day

SOA	14	6.8 GB	Not tested	Not tested

Primary PM species	6	1.5 GB	3.0 GB	10.8 h/day

Two annual 36-km CAMx/PSAT model simulations were performed: one with
the Plan02c typical-year baseline case and the other with the Base18b
future-year case. It is expected that the states and tribes will use
these results to assess the sources that contribute to visibility
impairment at each Class I Area, and to guide the choice of emission
control strategies. The RMC web site includes a full set of source
apportionment spatial plots and receptor bar plots for both Plan02b and
Base18b.  These graphical displays of the PSAT results, as well as
additional analyses of these results are available on the TSS under  
HYPERLINK "http://vista.cira.colostate.edu/tss/Tools/ResultsSA.aspx" 
http://vista.cira.colostate.edu/tss/Tools/ResultsSA.aspx 

CAMx/PSAT 2002 and 2018 Setup

PSAT source apportionment simulations for 2002 and 2018 were performed
using CAMx v4.30. Table 4 lists overall specifications for the 2002 PSAT
simulations. The domain setup was identical to the standard WRAP CMAQ
modeling domain. The CAMx/PSAT run-time options are shown in Table 5.
The CAMx/PSAT computational cost for one simulation day with source
tracking for sulfate (SO4) and nitrate (NO3) is approximately 14.5 CPU
hours with an AMD Opteron CPU.  The source regions used in the PSAT
simulations are shown in Figure 7 and Table 4. The six emissions source
groups are described in Table 6.  The development of these emissions
data are described in more detail below. 

The annual PSAT run was divided into four seasons for modeling. The
initial conditions for the first season (January 1 to March 31, 2002)
came from a CENRAP annual simulation. For the other three seasons, we
allowed 15 model spin-up days prior to the beginning of each season.
Based on the chosen set of source regions and groups, with nine tracers,
and with a minimum requirement of 87,000 point sources and a horizontal
domain of 148 by 112 grid cells with 19 vertical layers, the run-time
memory requirement is 1.9 GB. Total disk storage per day is
approximately 3.3 GB. Although the RMC’s computation nodes are
equipped with dual Opteron CPUs with 2 GB of RAM and 1 GB of swap
space, the high run-time memory requirements prevented running PSAT
simulations using the OpenMP shared memory multiprocessing capability
implemented in CAMx.

Table 4. WRAP 2002 CAMx/PSAT specifications. 

WRAP PSAT Specs	Description

Model	CAMx v4.30

OS/compiler	Linux, pgf90 v.6.0-5

CPU type	AMD Opteron with 2 GB of RAM

Source region	18 source regions; see Figure 4.1 and Table 4.4

Emissions source groups	Plan02b, 6 source groups; see Table 4.5

Initial conditions	From CENRAP (camx.v4.30.cenrap36.omp.2001365.inst.2)

Boundary conditions	3-h BC from GEOS-Chem v2

Table 5. WRAP CAMx/PSAT run-time options.

WRAP PSAT specs	Description

Advection solver	PPM

Chemistry parameters	CAMx4.3.chemparam.4_CF

Chemistry solver	CMC

Plume-in-grid	Not used

Probing tool	PSAT

Dry/wet deposition	TRUE (turned on)

Staggered winds	TRUE (turned on)

Table 6. WRAP CAMx/PSAT source regions cross-reference table. 

Source 

Region ID	Source Region Description1	Source 

Region ID	Source Region 

Description1

1	Arizona (AZ)	10	South Dakota (SD)

2	California (CA)	11	Utah (UT)

3	Colorado (CO)	12	Washington (WA)

4	Idaho (ID)	13	Wyoming (WY)

5	Montana (MT)	14	Pacific off-shore & Sea of Cortez (OF)

6	Nevada (NV)	15	CENRAP states (CE)

7	New Mexico (NM)	16	Eastern U.S., Gulf of Mexico, & Atlantic Ocean (EA)

8	North Dakota (ND)	17	Mexico (MX)

9	Oregon (OR)	18	Canada (CN)

1The abbreviations in parentheses are used to identify source regions in
PSAT receptor bar plots.

Figure 7. WRAP CAMx/PSAT source region map. Table 6 defines the source
region IDs.

Table 7. WRAP CAMx/PSAT emissions source groups.

Emissions Source Groups	Low-level Sources	Elevated Sources

1	Low-level point sources (including stationary off-shore)	Elevated
point sources (including stationary off-shore)

2	Anthropogenic wildfires (WRAP only)	Anthropogenic wild fires (WRAP
only)

3	Total mobile (on-road, off-road, including planes, trains, ships
in/near port, off-shore shipping)

	4	Natural emissions (natural fire, WRAP only, biogenics)	Natural
emissions (natural fire, WRAP only, biogenics)

5	Non-WRAP wildfires (elevated fire sources in other RPOs)	Non-WRAP wild
fires (elevated fire sources in other RPOs)

6	Everything else (area sources, all dust, fugitive ammonia,
non-elevated fire sources in other RPOs)

	PSAT Results 

The source apportionment algorithms implemented in CAMx generate output
files in the same format as the standard modeled species concentrations
files. This typically consists of a two-dimensional, gridded dataset of
hourly-average surface concentrations for each source group tracer that
gives the contribution of the tracer to all the surface grid cells in
the model domain for each hour of the simulation. Three-dimensional
instantaneous concentrations are also output for the last two hours of
the simulation, which are used to restart the model. Although there are
options to output hourly 3-D average tracer concentrations, the model is
usually configures to output only the model’s surface layer
concentrations because of the vast disk storage space needed for the 3-D
file output for all the source group contributions. 

The source apportionment model results are typically presented in two
ways :

Spatial plots showing the area of influence of a source group’s PM
species contributions throughout the model domain, either at a given
hourly-average point in time or averaged over some time interval (e.g.,
monthly average). 

Receptor bar plots showing the rank order of source groupings that
contribute to PM species at any given receptor site. These plots also
can be at a particular point in time or averaged over selected time
intervals—for example, the average source contributions for the 20%
worst visibility days. 

If the 3-D tracer output files are saved, it is also possible to prepare
animations of PM species plumes from each of the source groups. However,
these plots are less useful than the others for quantitative analysis,
are expensive to produce, and require saving 3-D hourly output, which is
disk-space intensive. The primary products of the WRAP PSAT modeling
were receptor bar plots showing the emission source groups that
contribute the most to the model grid cells containing each IMPROVE
monitoring site and other receptor sites identified by WRAP.

Model Sensitivity Simulations

A variety of sensitivity simulations were conducted by the RMC as part
of their modeling efforts to support the WRAP in addressing the Regional
Haze Rule requirements.  These sensitivity simulations are described
below. 

2002 Clean Case

There are many natural sources of ambient PM2.5, both direct emissions
of primary PM2.5 (such as windblown dust) and emissions of gaseous
species that undergo photochemical transformation or condensation to
form secondary PM2.5. Natural sources of PM2.5 are of concern because
they represent sources that cannot be controlled. Estimates of natural
haze levels have been developed by EPA for visibility planning purposes
and are described in Guidance for Estimating Natural Visibility
Conditions Under the Regional Haze Rule (U.S. EPA, 2003a). These are the
natural haze levels to be used in glide path calculations, such as those
we performed as part of the visibility projections for 2018. However,
the natural haze levels developed by EPA for glide path calculations
were based on ambient data analysis, not on visibility modeling. This
question thus arises: Would modeled levels of natural haze be consistent
with the values estimated by EPA for visibility planning? If the natural
haze levels calculated by the model were substantially higher than the
levels used for planning purposes, this would make it more difficult for
modeling studies to demonstrate progress in attaining visibility goals,
because the model would predict haze levels that exceeded EPA’s
natural haze levels even if all anthropogenic sources of PM2.5 were
removed from the modeling. The RMC explored this issue by conducting a
CMAQ sensitivity “clean conditions” simulation

There are many uncertainties and unknowns regarding natural emissions.
There have been only limited studies of natural emissions conditions. It
is known that there are very large uncertainties in the categories of
natural emissions included in the WRAP emissions inventories, and that
some categories of natural emissions are not included at all. Also, it
is difficult to know what truly natural emissions would have been like
in the absence of human modifications of the environment. For example,
wildfire emissions are a large source of natural emissions in our
modeling, but how much larger might that source be in the absence of
fire suppression efforts? For all of these reasons, it was decided to
describe this sensitivity simulation as a “clean conditions”
scenario rather than a “natural conditions” scenario. In this
simulation, all anthropogenic emissions were removed from the inventory
and only those emissions that were defined as biogenic in the 2002 base
case (Base02) were included. Thus, this model simulation does not
represent true natural conditions. It indicates instead the lowest haze
levels that could be achieved in the model if all anthropogenic
emissions were zeroed out.

Emission Inventories

The emissions for the clean 2002 sensitivity case were derived from case
Base02a. Because it was a sensitivity analysis to test the impacts of
natural emissions sources on visibility, it is referred to it as
scenario Base02nt, where “nt” refers to natural. The following
emissions categories in Base02nt were included:

Biogenics: Generated in case Base02a by BEIS3.12 using SMOKE.

WRAP Ammonia: The Base02a ammonia emissions for the WRAP region were
developed with a GIS by ENVIRON. The five emissions category modeled
included three anthropogenic sources (domestic animals, livestock, and
fertilizer application) and two natural sources (soils and wildlife).
Only the two natural sources in scenario Base02nt were used.

CENRAP and MRPO Ammonia: To create ammonia inventory files for only
natural sources, we used a list of SCCs representing natural sources to
extract the emissions records of these sources from the monthly
inventory files that were used in Base02a. it was found that there were
no natural ammonia sources in the MRPO monthly inventory files.

Natural Area Sources: The Base02a area-source inventory files included
natural sources, such as wildfires and wild animals. These records were
extracted from the stationary-area-source inventories. Note that the
WRAP area-source files did not include any natural sources.

Natural Fires: Of the five fire categories modeled in Base02a
(wildfires, wildland fire use, non-Federal rangeland prescribed fires,
prescribed fires [which were split into natural and anthropogenic
prescribed for this purpose of this sensitivity], and agricultural
fires), only the categories that represent natural fires (wildfires,
wildland fire use, and natural prescribed fires) were included. 

Windblown Dust: We used the windblown dust inventory that ENVIRON and
the RMC developed for use in case Base02a. Additional details on this
dust inventory are available at   HYPERLINK
"http://www.cert.ucr.edu/aqm/308/wb_dust2002/wb_dust_ii_36k.shtml" 
http://www.cert.ucr.edu/aqm/308/wb_dust2002/wb_dust_ii_36k.shtml . 

The biogenic and windblown dust emissions from the Base02a SMOKE outputs
that are stored at the RMC were used directly. For the fire (including
both point and area fires), natural area, and ammonia emissions, these
data were reprocessed specifically for scenario Base02nt using the same
ancillary data (temporal, chemical, and spatial allocation data) used in
case Base02a. QA plots and documentation for scenario Base02nt are
posted on the RMC web site at   HYPERLINK
"http://pah.cert.ucr.edu/aqm/308/qa_Base02nt36.shtml" 
http://pah.cert.ucr.edu/aqm/308/qa_Base02nt36.shtml . 

Modeling Results

Figure 8 shows the model-reconstructed light extinction in the clean
emissions model simulation. Because the natural fire emissions in the
WRAP states were a major component of the clean emissions, the largest
visibility impairment is in the regions with natural fire emissions.
Contributions to light extinction from natural sources were small in
regions without large fire emissions, as evidenced in the eastern U.S.,
where the extinction was only slightly larger (about 2 Mm-1) than
perfectly clean Rayleigh conditions of 10 Mm-1.

Although there are large uncertainties in the natural emissions, and it
is known that there are missing types of natural emissions, the
components of the natural inventory used in this sensitivity simulation
did contribute to relatively large visibility impairment in regions
where there were large wildfires. Extinction coefficients as large as 90
Mm-1 were simulated in the southern Oregon and northern California
regions; this was most likely a result of the large Biscuit fire in
Oregon, plus contributions from smaller fires and other natural
emissions. These visibility impairment levels exceed the natural
visibility levels specified in the EPA regional haze natural visibility
guidance document. It will thus be more difficult for the modeling to
demonstrate attainment of progress goals in areas of the country subject
to wildfires because of their large contribution to visibility
impairment that is not controllable. In other regions of the country for
which the inventories lacked large natural fire emissions, the modeled
clean visibility was only slightly greater than clean Rayleigh
conditions. Note the model results may be overly optimistic in these
regions because we lack a complete, accurate natural emissions
inventory.

Figure 8. Annual average model-reconstructed “clean conditions”
visibility

as extinction coefficient.

These results are all very tentative because of the large uncertainties
in natural emissions. Considerable effort would be needed to more fully
investigate natural conditions in future modeling studies. It will
always be difficult to determine and quantify “clean conditions”
based on observations because of the pervasive influence of
anthropogenic emissions.

Also as part of this sensitivity analysis, the contributors to organic
carbon aerosols (OC) for the clean conditions scenario wer4e evaluated.
The CMAQ model represents explicitly three classes of organic carbon
aerosols:

AORGPA: Primary anthropogenic OC resulting from direct organic mass
emissions, such as primary organic aerosol (POA).

AORGA: Secondary anthropogenic OC resulting from aromatic VOCs, such as
xylene, toluene, and cresols.

AORGB: Secondary biogenic OC resulting from biogenic VOCs, such as
terpenes. 

Because it was not cost effective to carry out CAMx/PSAT simulations
with OC, the explicit OC results for the clean conditions case were
analyzed, and then compared those results to the Base02b case in an
attempt to infer the relative contributions of biogenic and
anthropogenic VOCs to OC. These results are difficult to interpret for
at least two reasons:

Because of the simplified approach used by CMAQ and the Carbon Bond
Mechanism version 4 (CB4) to represent these species, it is not possible
to accurately classify all emissions into the CMAQ model as either
biogenic or anthropogenic based simply on the species name. Thus, some
biogenic OC might be included with AORGA, and some anthropogenic OC
might be included in AORB. 

Some fire emissions are classified as anthropogenic, but these emissions
might include species such as terpenes that are typically considered
biogenic. Using the analysis approach in which all terpenes are assumed
biogenic then incorrectly causes some anthropogenic emissions to be
labeled biogenic when we use the simplified approach of analyzing OC in
terms of AORGPA, AORGA and AORGB. 

In spite of these difficulties, however, the results should classify the
majority of the emissions correctly as either biogenic or anthropogenic.

For each of the above three components of OC, plots of the annual
average mass in the Base02b case were prepared, and then the
controllable mass was estimated as the difference between the Base02b
case the Base02nt clean emissions scenario. Figure 9 shows the annual
average mass of OC contributed from AORGPA in case Base02b (top) and the
portion of that mass attributed to controllable emissions (bottom).
Comparing these two plots indicates that in the western U.S. there is
considerable AORGPA mass that is not controllable. It is likely that
much of this mass is from fires, since uncontrollable AORGPA mass is
present at the site of large fires in southern Oregon and north of
Tucson, AZ.

Figure 10 shows the annual average mass of secondary OC contributed from
AORGA in the Base02b case (top) and the portion of that mass attributed
to controllable emissions (bottom). These plots indicate that virtually
all of the AORGA mass is controllable, since the bottom plot is almost
identical to the top plot.

Figure 11 shows the annual average mass of OC contributed from AORGPA in
the Base02b case (top) and the portion of that mass attributed to
controllable emissions (bottom). These plots indicate that although most
of the AORGB mass is not controllable, a significant amount of mass is
controllable. It is likely that the controllable AORGB mass results from
VOC oxidation chemistry and the larger amount of biogenic mass that is
oxidized and subsequently condenses to form OC in the Base02b case.
These results indicate that controlling O3 precursor emissions is
effective at reducing a small but significant fraction of the biogenic
OC.



Figure 9. Annual average modeled primary anthropogenic OC (AORGPA) in
Base02b (top) and the portion that is “controllable” primary
anthropogenic OC (bottom).



Figure 10. Annual average modeled secondary anthropogenic OC (AORGA) in
Base02b (top) and the portion that is “controllable” secondary
anthropogenic OC (bottom).



Figure 11. Annual average modeled primary biogenic OC (AORGB) in Base02b
(top)

and the portion that is “controllable” primary biogenic OC (bottom).

It might be difficult for the WRAP states and tribes to use these
results quantitatively in developing emissions control strategies for
visibility SIPs and TIPs. However, the results do provide some insight
into the relative contributions of biogenic and anthropogenic OC as well
as the amount of each that is controllable in the model simulations.

Finally, it is noted that there are uncertainties in the modeled
emissions of anthropogenic VOCs, and larger uncertainties in the modeled
emissions of biogenic VOCs. It is not possible to evaluate the model
performance individually for biogenic and anthropogenic OC because the
OC measurements do not distinguish between those two forms. Instead,
only comparisons of total modeled OC to total measured OC can be made.
Therefore, even when the model achieves good performance for total OC,
it is possible that the model may be overpredicting one component of
total OC and underpredicting the other. The inability to evaluate model
performance for each component of OC increases the uncertainty of the
results described here and illustrated in Figures 9 through 11, so
caution should be used when drawing conclusions about the sources of OC
based on these results.

References

Byun, D.W. and J. K. S. Ching, Eds., 1999: Science algorithms of the EPA
Models-3 Community Multiscale Air Quality (CMAQ) Modeling System. U.S.
Environmental Protection Agency Rep. EPA-600/R-99/030, 727 pp.
[Available from Office of Research and Development, EPA, Washington, DC
20460.

Dudhia, J. 1993. "A Non-hydrostatic Version of the Penn State/NCAR
Mesoscale Model: Validation Tests and Simulation of an Atlantic Cyclone
and Cold Front", Mon. Wea. Rev., Vol. 121. pp. 1493-1513.

Grell, G. A., J. Dudhia, and D. R. Stauffer. 1994. "A Description of the
Fifth Generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note,
NCAR TN-398-STR, 138 pp.

Kemball-Cook, S. et al., 2005: Draft Final Report, Annual 2002 MM5
Meteorological Modeling to Support Regional Haze Modeling of the Western
United States, prepared for the WRAP by ENVIRON International
Corporation, Novato, CA and the University of California at Riverside,
Riverside, CA.   HYPERLINK
"http://pah.cert.ucr.edu/aqm/308/reports/mm5/DrftFnl_2002MM5_FinalWRAP_E
val.pdf" 
http://pah.cert.ucr.edu/aqm/308/reports/mm5/DrftFnl_2002MM5_FinalWRAP_Ev
al.pdf 

Malm, W., M. Pitchford, M. Scruggs, J. Sisler, R. Ames, S. Copeland, K.
Gebhart and D. Day. 2000. Spatial and Seasonal Patterns and Temporal
Variability of Haze and Its Constituents in the United States – Report
III. Cooperative Institute for Research in the Atmosphere, Fort Collins,
Colorado. May.
(http://vista.cira.colostate.edu/Improve/Publications/Rpeorts/2000/2000.
htm).

Malm, W., M. Pitchford, M. Scruggs, J. Sisler, R. Ames, S. Copeland, K.
Gebhart and D. Day. 2000. Spatial and Seasonal Patterns and Temporal
Variability of Haze and its Constituents in the United States.
Cooperative Institute for Research in the Atmosphere, Colorado State
University, Fort Collins, CO. May.

Tonnesen, G. et al., 2006: Final Report for the WRAP 2002 Visibility
Model Performance Evaluation, Prepared for the Western Governors
Association by the WRAP RMC, Riverside, CA.
http://pah.cert.ucr.edu/aqm/308/reports/final/2002_MPE_report_main_body_
FINAL.pdf

U.S. EPA. 2001. “Guidance for Demonstrating Attainment of Air Quality
Goals for PM2.5 and Regional Haze”, Draft Report, U.S. Environmental
Protection Agency, Research Triangle Park, NC.

U.S. EPA. 2003a. “Guidance for Estimating Natural Visibility
Conditions under the Regional Haze Rule.” EPA-454/B-03-005. September
2003.

U.S. EPA. 2003b. “Guidance for Tracking Progress under the Regional
Haze Rule.” U.S. EPA, EPA-454/B-03-004. September 2003. 

U.S. EPA. 2003c. “Revisions to the Guideline on Air Quality Models:
Adoption of a Preferred Long Range Transport Model and Other
Resources”; Final Rule. Fed. Reg./Vol. 68, No. 72/Tuesday April 15,
2003/Rules and Regulations. 40 CFR51.

U.S. EPA. 2005. “Regional Haze Regulations and Guidelines for Best
Available Technology (BART) Determinations”. Fed. Reg./Vol. 70, No.
128/Wed. July 6, 2005, Rules and Regulations, pp. 39104-39172. 40 CFR
Part 51, FRL-7925-9, RIN AJ31.

#

&

†

‰

³

¶

 h

 h

 h	

 h	

 h

葞и摧粨¼᐀

kd

愀Ĥ摧垌H

†

†

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

②朁쥤턫氀჆

最쥤턫氀჆

!

最쥤턫氀჆

!

最쥤턫氀჆

!

最쥤턫氀჆

!

最쥤턫氀჆

!

最쥤턫氀჆

!

最쥤턫氀჆

!

②朁쥤턫氀჆

!

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

②朁쥤턫氀჆

②朁쥤턫氀჆

最쥤턫氀჆

②朁쥤턫氀჆

最쥤턫氀჆

②朁쥤턫氀჆

最쥤턫氀჆

最쥤턫氀჆

②朁쥤턫氀჆

„

>

D

>

Œ

最쥤턫氀჆

nstrating Attainment of Air Quality Goals for Ozone, Pm2.5, and Regional
Haze – Draft 3.2. U.S. Environmental Protection Agency, Office of Air
Quality and Planning Standards, Research Triangle Park, North Carolina.
September. ( HYPERLINK
http://www.epa.gov/scram001/guidance/guide/draft_final-pm-O3-RH.pdf 
http://www.epa.gov/scram001/guidance/guide/draft_final-pm-O3-RH.pdf ).

U.S. EPA. 2006b. Additional Regional Haze Questions. U.S. Environmental
Protections Agency. August 3. (  HYPERLINK
"http://www.wrapair.org/forums/iwg/documents/Q_and_A_for_Regional_Haze_8
-03-06.pdf#search=%22%22New%20IMPROVE%20equation%22%22" 
http://www.wrapair.org/forums/iwg/documents/Q_and_A_for_Regional_Haze_8-
03-06.pdf#search=%22%22New%20IMPROVE%20equation%22%22 

