United States
        Environmental Protection
        Agency
          Office of Air Quality
          Planning and Standards
          Research Triangle Park, NC 27711
EPA-450/4-92-Qlla
June 1992
        Air
& EPA
GUIDELINE FOR REGULATORY
APPLICATION OF THE
URBAN AIRSHED MODEL
FOR AREAWIDE CARBON
MONOXIDE
        VOLUME I:  TECHNICAL
                    REPORT

-------

-------
                         EPA-450/4-92-0113
GUIDELINE FOR REGULATORY
APPLICATION OF THE
URBAN AIRSHED MODEL
FOR AREAWIDE CARBON
MONOXIDE

VOLUME I: TECHNICAL REPORT
     U.S. Environmental Protection Agency
  Office of Air Quality Planning and Standards
       Technical Support Division
     Research Triangle Park, NC 27711
                     U.fc. Environmental Protection Agency
          June 1992     Region 5, Library (PL-12J)
                     77 West Jackson Boulevard, 12th Floor
                     Chicago, IL 60604-3590

-------
                                      Notice
This report has been funded by the United States Environmental Protection Agency (EPA) under
contract 68D00124 to Systems Applications International (SAI). Thomas N. Braverman served
as the EPA work assignment manager.  Any mention of trade names or commercial products is
not intended to constitute endorsement or recommendation for use.
                                         ii

-------
                                  CONTENTS

                                                                       Page
NOTICE  	   ii
TABLES  	   vi
ABBREVIATIONS  	  vii
ACKNOWLEDGMENTS 	   ix
1 INTRODUCTION	_	    1
  1.1 Background  	    3
  1.2 Application of the UAM  	    5
2 MODELING PROTOCOL	    7
  2.1 Protocol Development Process	    8
  2.2 Contents of Protocol Document	    9
3 DOMAIN AND DATA BASE ISSUES	   11
  3.1 Overview of Model Inputs  	   11
  3.2 Episode Selection	   14
  3.3 Selection of Modeling Domain  and Resolution	   16
      3.3.1  Domain Definition	   16
      3.3.2  Horizontal Grid Cell Size  	   17
      3.3.3  Number of Vertical Layers   	   17
      3.3.4  Time Span 	   18
  3.4 Preparation of Meteorological Inputs	   19
      3.4.1  Surface  Roughness and Deposition (TERRAIN)	   19
      3.4.2  Diffusion Break (DIFFBREAK)   	   20
      3.4.3  Top of the Modeling Domain (REGIONTOP)	   21
      3.4.4  Winds (WINDS)	   22
                                       iii

-------
  3.5 Preparation of Air Quality Inputs	   25
      3.5.1  Air Quality (AIRQUALITY)	   25
      3.5.2  Top Concentration (TOPCONC)  	   26
      3.5.3  Boundary Conditions (BOUNDARY)	   26
  3.6 Performance Evaluation Data	   29
  3.7 Emission Inventory	   30
      3.7.1  Use of Surrogate Factors to Spatially Grid Area Sources  	   32
      3.7.2  Development of Mobile Source Inventories	   32
      3.7.3  Episode-Specific Adjustments	   34
      3.7.4  Consistency with National Inventories	   35
4 DATA QUALITY ASSURANCE .AND MODEL DIAGNOSTIC ANALYSES  . .   37
  4.1 Step 1:  Quality Assurance Testing of Component Fields	   38
  4.2 Step 2:  Diagnostic Testing of the Base Case Meteorological Epiodes	   40
  4.3 Step 3:  Additional Base Meteorological Episode Sensitivity Testing	   41
5 MODEL PERFORMANCE EVALUATION		   45
  5.1 Performance Measures	   46
     .5.1.1  Graphical Performance Procedures	  ,	   46
      5.1.2  Statistical Performance Measures	   47
  5.2 Assessment of Model Performance	   48
6 ATTAINMENT DEMONSTRATION	   53
  6.1 Developing Attainment Year Base Case Emission Inventories	   53
  6.2 Developing Future Year Emission Control Strategies	   54
  6.3 Performing Attainment-Year Simulations to Assess Various Control Strategies   56
  6.4 Using Modeling Results in the  Attainment Demonstration  	   56
  6.5 Exceptions to  Guidance Document  	   57
REFERENCES  	   59
                                        IV

-------
                      VOLUME E. APPENDIXES


APPENDIX A: RECOMMENDED MODELING PROTOCOL CONTENTS

APPENDIX B: PERFORMANCE MEASURE FORMULATIONS

APPENDIX C: TECHNICAL DISCUSSION OF UAM MODEL INPUTS WITH
           EXAMPLE APPLICATION

-------
                                       TABLES








2-1 Example table of contents for protocol document  	    10




3-1 Overview of the 13 Urban Airshed Model input files	    12




3-2 Recommended background concentrations for carbon monoxide  	    27
                                          VI

-------

-------
                                 ABBREVIATIONS
AIRS
CAAA
CARB
CMSA
CO
DWM
FTP
LAV
MSA
NAAQS
NTIS
NWS
OAQPS
OMS
ORD
RWC
SAI
SCRAM BBS

SEP
TAZ
TCM
Aerometric Infonnation Retrieval System
Clean Air Act Amendments
California Air Resources Board
Consolidated Metropolitan Statistical Area
Carbon Monoxide
Diagnostic Wind Model (UAM preprocessor program)
U.S. Environmental Protection Agency
Emissions Preprocessor System for the UAM
Federal Implementation Plan
Link Attribute Value
Metropolitan Statistical Area
National Ambient Air Quality Standard(s)
National Technical Information Service
National Weather Service
EPA Office of Air Quality Planning and Standards
EPA Office of Mobile Sources
EPA Office of Research and Development
Residential Wood Combustion
Systems Applications International
EPA Support Center for Regulatory Air Models
Bulletin Board System
State Implementation Plan
Transportation Analysis Zones
Transportation Control Measures
                                     vii

-------
UAM                   Urban Airshed Model
USGS                   United States Geologic Survey
VMT                   Vehicle Miles Traveled
                                     viii

-------
                              ACKNOWLEDGMENTS
Much of the format, style, and, where applicable, content of this guidance document was
borrowed from .the EPA guidance document titled "Guideline for Regulatory Application of the
Urban Airshed Model." The principal authors of that document were Mr. Dennis C. Doll (U.S.
Environmental Protection Agency (EPA), Office of Air Quality Planning and Standards
(OAQPS)), Dr. Richard D. Scheffe (EPA, OAQPS), Dr. Edwin L. Meyer (EPA, OAQPS), and
Mr. Shao-Hang Chu (EPA, OAQPS).

Mr. Edward L. Carr (Systems Applications International (SAI)) is the principal contributor to
this document. Significant contributions were also made by Julie L. Fieber and Robert C.
Kessler (SAI).  In addition, the document was reviewed and commented upon by Mr. Thomas
N. Braverman (EPA, OAQPS), Mr. Jay L. Haney (SAI), Mr. Henry Hogo (South Coast Air
Quality Management District), Dr. Robert G.  Ireson (SAI), Mr. William Ryan (EPA, Region
X), and Mr. Robert Wilson (EPA, Region X).
                                    ix

-------
                                   CHAPTER 1
                                INTRODUCTION

      Under the Clean Air Act Amendments of 1990, over 40 urban areas are classified as
nonattainment with respect to the 8-hour National Ambient Air Quality Standard (NAAQS) for
carbon monoxide (CO).  Of these urban areas, those whose 8-hour average design value exceeds
12.7 ppm are recommended to use an urban areawide model to address attainment of the CO
NAAQS of 9.0 ppm in the revision of their State Implementation Plan (SIP).  SIP revisions
demonstrating attainment of the CO NAAQS are required by the Clean Air Act Amendments of
1990.  The Urban Airshed Model (UAM) has been identified as an effective tool for evaluating
emission control requirements needed to attain and maintain the CO NAAQS.

      The purpose of this document is to provide guidance in the procedures used to apply the
UAM for  CO  SIP attainment demonstrations and to ensure national consistency  in model
applications.

      Methodologies and procedures used in the preparation of UAM inputs and applications
to carbon monoxide regulatory issues are addressed in this report.  This document also describes
recommendations for the preparation  of all input files  needed to exercise the UAM.  The
technical explanation and rationale for the development of UAM inputs are discussed in
Appendix C.

      The UAM source code is maintained and distributed by the Source Receptor Analysis
Branch, Technical Support Division, of the EPA Office of Air Quality Planning and Standards
(OAQPS).  Users will be informed of  modifications or enhancements to the UAM through the
Support Center for Regulatory Air Models Bulletin Board System (SCRAM BBS). Additionally,

-------
the UAM source code, user's guide, and test case data base are available from the National
Technical Information Service (NTIS)(703-737-4600).

       Steps needed to conduct an urban-scale modeling study consist of the following:

       1.     Establish a protocol for the modeling study in which candidate modeling episodes
             are identified.

       2.     Compile air quality, meteorological, and emissions data to develop UAM input
             files for each meteorological episode to be used in the attainment demonstration
             model simulations.

       3.     Execute the model for each meteorological episode.

       4.     Conduct diagnostic analyses on each meteorological episode simulation.  The
             principal purpose of diagnostic analyses  is to ensure  that the model properly
             characterizes physical phenomena  (e.g.,  wind  fields,  spatial and temporal
             emission patterns) instrumental in leading to observed  CO concentrations.  The
             visible product is enhanced model performance (i.e., better spatial and temporal
             agreement with observed data).  Diagnostic model simulations are intended to
             uncover potential model input data gaps that, when  corrected, may  lead to
             improved model performance.

       5.     Exercise the  model  for  each meteorological episode  and  use a  series of
             performance measures to  determine overall model performance in  replicating
             observed CO concentrations and patterns.

       6.     For each  meteorological episode, estimate emissions and air quality  for the
             projected attainment year required under the CAAA. Perform model simulations

-------
             for each episode to determine whether the CO NAAQS can be  met in the
             attainment year.

       7.     If the model simulations for the attainment year do not show attainment for each
             modeled episode, develop additional emission control measures on selected source
             categories.

       8.     Perform model simulations for the emission control measures to demonstrate
             attainment of the CO NAAQS for each meteorological episode. If the control
             measures do  not show attainment, repeat steps 7 and 8 as an iterative process
             until attainment is shown for each modeled episode.

       This report is divided into six sections. Section 1 provides background information about
air quality problems related to high CO. concentrations and describes the philosophies underlying
UAM areawide CO applications.  Section 2 describes the modeling protocol and development
processes.  Section  3 discusses domain and data base issues.  Section 4 discusses data quality
assurance and model diagnostic analysis.  Section 5 discusses model performance evaluation and
Section 6 describes the methodology to be followed in attainment demonstration.

1.1  Background

       High 8-hour CO concentrations in urban areas often result from periods of high emissions
(the afternoon  and evening  mobile traffic peak)  coinciding  with  adverse meteorological
conditions  (low  wind speeds and  poor  vertical dispersion).   Under  such  conditions,
concentrations can increase rapidly near heavy traffic areas.  The greatest problem in achieving
ambient 8-hour CO standards in an urban area is controlling the neighborhood-scale buildup and
persistence (over 8 hours or more)  of elevated CO  concentrations.  Under meteorological
conditions conducive to high concentrations, CO emissions become trapped in a shallow, stable
surface layer caused by the radiational cooling of the air next to the ground.  When winds are
light and variable,  this high-concentration air mass may extend over a large portion of an

-------
urbanized area depending upon topography and emission distribution and can remain relatively
stationary or meander during nighttime hours, with the result that one or several areas may
experience high CO concentrations throughout the night and early morning hours.  Maximum
8-hour average  CO concentrations  result from the combined effect of microscale  and
neighborhood-scale processes.   For some urban areas, sites on the downwind edge of high
emission density regions can experience extended periods of high concentrations even if wind
speeds are moderate.

       Carbon monoxide emissions result from  incomplete combustion of fossil fuels which
include residential wood combustion and industrial emissions. However, motor vehicle exhaust
accounts for most of the CO emissions in urban areas.  To reduce motor vehicle emissions
injected into the atmosphere, either the emission rates of individual vehicles must be reduced or
traffic conditions (volumes, speeds, stops and idling periods) must be modified. To estimate the
effects of these control measures, both microscale and neighborhood-scale modeling must be
conducted.  Areas with significant stationary sources as determined under Section 187c (Laxton,
1991a) should perform modeling following the techniques cited in the Guideline on Air Quality
Models (Revised) (EPA,  1986).

       Previous UAM modeling studies of CO have been performed in Phoenix (Haney, 1988;
Causley  et al., 1991) and Denver (Anderson et al.,  1977; Rogers,  1986).  In  the studies
conducted in Denver, only neighborhood-scale modeling was performed.  In later studies,
neighborhood-scale and microscale processes were modeled  separately,  and the results of each
were added to estimate the total concentrations at selected roadway  intersections to assess
attainment.  This approach allowed use of the UAM to describe the accumulation of emissions
over  several hours and kilometers within a three-dimensional modeling grid, as well as the
separate estimation of roadway impacts within a few hundred meters of the roadway intersection.
The combined model results were then used to evaluate the  effectiveness of control measures.
Procedures for determining the roadway intersection concentration for CO SIP applications are
addressed in the guidance document Guideline for Modeling Carbon from Roadway Intersections
(EPA, 1992a).

-------
1.2 Application of the UAM

       Conditions that often are conducive to high areawide 8-hour CO concentrations (low wind
speeds, stable conditions) are the same conditions for which the steady-state assumption inherent
in Gaussian  formulated  models  is  invalid.   The steady-state  assumption results in  the
accumulation of large errors due to the inability of the Gaussian models to gradually build
carbon monoxide  concentrations from hour to hour.  The grid cell and time step modeling
approach of the UAM is more  appropriate for CO modeling because it is capable of handling
low wind speeds and stable conditions by allowing concentrations to accumulate over time.

       A number of past studies involving  the  UAM relied on intensive and  expensive
monitoring studies to provide information to adequately prepare model inputs.  Recent studies
have shown that routine meteorological and air quality observations are adequate for UAM SIP
applications.  This set of procedures (known as PLANR, or Procedures for Low-cost Airshed
applications to Nonattainment Regions) was applied to determine its feasibility in future SIP
efforts (Morris et al., 1990). Although the PLANR methodology was initially confined to ozone
nonattainment problems, the same general methodology can be applied equally well to carbon
monoxide nonattainment areas.  The method relies on the  ability of the UAM to accurately
predict hourly concentrations within a certain distance and time using routine data gathered from
the monitoring station measuring the highest carbon monoxide values.  Although the PLANR
method may prove to be applicable to most CO nonattainment areas, some areas with complex
terrain and/or meteorology may still require more intensive data for successful application of the
UAM.

-------
                                     CHAPTER 2
                              MODELING PROTOCOL

       The UAM modeling domain may encompass multiple geopolitical boundaries (counties,
cities, and states) with a potentially large regulated community. Therefore, the development of
a modeling protocol is recommended to (1) promote technical credibility, (2) encourage the
participation of all interested parties, (3) provide for consensus building among all interested
parties concerning modeling issues, and  (4) provide documentation for technical decisions made
in applying the model as well as the procedures followed in reaching these decisions.

       The protocol should detail and formalize  procedures for  conducting all phases of the
modeling study, such as (1) describing the background and objectives of the study, (2) creating
a  schedule and  organizational structure  for the study, (3) developing the  input data, (4)
conducting diagnostic and model performance evaluations, (5) interpreting modeling results, (6)
describing procedures for using  the UAM and roadway  intersection models to demonstrate
whether proposed strategies  are  sufficient to attain  the CO NAAQS,  and  (7) producing
documentation and data analyses that must be submitted for EPA  regional office review and
approval.

       All  issues concerning the modeling study must be  thoroughly  addressed during the
protocol development.  Thus, modifications to the protocol as the study progresses should not
be needed unless unforeseen procedural and/or technical issues are encountered.  All parties
involved  in the  study  should agree to protocol  modifications through the modeling policy
oversight committee, if applicable (see below).   It is especially  important that the state/local
agencies and EPA regional office(s)  overseeing the study concur  on protocol modifications.

-------
2.1    Protocol Development Process

       The state agency responsible for developing the CO State Implementation Plan (SIP) is
usually the  lead  agency responsible  for  developing  the modeling protocol.   For domains
encompassing parts of more than one  state, the responsible state agencies need to develop the
modeling protocol jointly.  Since the protocol should describe the modeling policy and technical
objectives of the study, input will be required from  various  EPA and state/local personnel
dealing with regulatory policy issues and from others with modeling expertise.   In some cases
a modeling policy oversight and technical committee will need to be organized to address these
issues.  The composition and responsibilities of the committee should be defined in the modeling
protocol.

       Responsibilities of the modeling policy oversight committee will be, at a minimum, to
set the objectives of the study, set the schedule, determine resource needs, and implement any
modifications to the protocol as  the modeling study proceeds.  The committee should include
representatives  from the appropriate EPA regional office(s),  state/local agencies, the regulated
community,  and public interest groups. It is important that appropriate policy-oriented personnel
be identified for membership on the committee.

       Responsibilities of the technical committee will  be, at a minimum, to develop the
protocol's technical specifications  concerning emission inventories,  meteorological data, air
quality data, data quality assurance,  emission control strategies,  model  diagnostic analyses,
model performance evaluation procedures, and interpretation of model results.  The technical
committee should include appropriate technically oriented  members from the  EPA regional
office(s), state/local agencies, the regulated community, and  public interest groups.

       The modeling protocol must be  submitted to the appropriate EPA regional modeling
contact for review and approval. The EPA regional modeling contact should be a member of
the policy oversight and/or technical committee so that rapid review and approval of the protocol
is assured.

-------
      Recommendations
      A protocol document is  recommended for each UAM application used for a CO
      attainment demonstration.  This protocol should describe the methods and procedures to
      be used for conducting the CO modeling study.

      Additionally, it is suggested that both a policy oversight committee and a  technical
      committee be established to develop the modeling protocol.   The composition and
      responsibilities  of the committees should be defined in the protocol.

      The modeling protocol and any modifications to it should be agreed upon by all parties
      involved in the  study through the policy oversight committee.  It is especially important
      that the state/local agency  participants  and  EPA regional  office(s)  overseeing  the
      modeling study concur on any protocol modifications.  Protocol modifications should be
      documented for subsequent public review.

      The modeling protocol must be  submitted to the appropriate EPA regional modeling
      contact for review and approval.


2.2   Contents of Protocol Document
      Recommendations

      It is recommended that the applicable components listed in Table 2-1 be included in the
      protocol document for each attainment demonstration modeling study.  A description of
      each component is presented in Appendix A.

-------
                                 TABLE 2-1

     EXAMPLE TABLE OF CONTENTS FOR PROTOCOL DOCUMENT

1.      UAM Modeling Study Design
              Background and Objectives
              Schedules
              Deliverables
              Management Structure/Technical Committees
              Participating Organizations
              Relationship to Planning/Strategy Groups

2.      Domain and Data Base Issues
              Data Bases:
                     • Air quality
                     •Meteorology
              Base Meteorological Episode Selection
              Modeling Domain
              Horizontal Grid Resolution
              Number of Vertical Layers
              Emission Inventory
              Specification of Initial arid Boundary Conditions
              Wind Field Specification
              Inversion Depth
              Sources of Other Input Data

3.      Quality Assurance and Diagnostic Analyses
              Quality Assurance  Tests of Input Components
              Diagnostic Tests of Base Case Simulation
              Test Results/Input  Modifications

4.      Model Performance Evaluation
              Performance Evaluation Tests

5.      Roadway Intersection Modeling
              Selection Methodology for Intersections Modeled
              Modeling Methodology

6.      Attainment Demonstrations
              Identification of Attainment-Year Mandated Control Measures
              Methodologies for  Generating Control Strategy Emission Inventories
              Procedures for Attainment Demonstration

7.      Submittal Procedures
              Data Analysis Review
              Documentation Review and Approval
                                     10

-------
                                    CHAPTERS
                        DOMAIN AND DATA BASE ISSUES

      Described in this chapter are the following topics:  episode selection, domain selection,
meteorological inputs, air quality data,  and emissions inventories. Choices made in each topic
area are often interrelated. Accordingly,  decisions  concerning a particular topic area probably
will be based on consideration of several areas.  In several topic areas, recommendations are
made  concerning minimum requirements for data  availability and  modeling resolution.  To
reduce uncertainties in modeling inputs  and outputs, users are encouraged  to  exceed these
minimum recommendations whenever possible.

3.1  Overview of Model Inputs

      This section provides  an overview of the 13 input files required to exercise the Urban
Airshed Model (UAM).  Table 3-1 presents a list of these input files and indicates the amount
of effort required to create each one. A more complete reference for the UAM is contained in
the UAM User's Manual (EPA, 1990).

      Of the 13 input files, two are considered "universal" in that they do not vary from one
simulation to another. The first file, CHEMPARAM, run in the unreactive mode for carbon
monoxide modeling, contains the list of species  to be modeled.  Other unreactive species may
be simulated for little additional cost if the emission inventory data are available.  The other
input file, SIMCONTROL, contains parameters controlling  the UAM simulation. In general,
the date and time will vary for different meteorological episodes but not for the same episode.
                                         11

-------















J
1^
H—
*•*
0.
_c
«••

5
«M
O
£
0)
1
V
6
i
m
W
T
CO
<
H
|
2g
•a •*-
•2 o
4> _
o£ , ^-a| -a^
f|« |f3| | 1,| S|*J Jl
ill! lllil If ^11 ,I£|J fill
ill! !!!!! ill f ! I!
Ill }!ii! LHHt lf.?P SHi'
^•§^s ossll53 •e-S's^M. s§o£"s s-sas
^:5 .sr TS t* * X3
-i^l§ If-ii! Ill l^ HtfifB g^l^
^^«-3) 3 § B 'i - gTV~ >^S & 1 3 f •g'o ^S53^
•3o^^9? a«°a.2 »«« c?l Z7<22^^cj -2 = >%S


Cfl . 4)
•"• w> e -S *o
. § § -a 2 i
«« .2 ^ c o
CrT "3 C dl c*»
•J-J CQ S «/ «O
O JS „, ^ O. S3 T3
•s ^ S ^ P ja 13
2 E o ep ~ o tg
•fa •— 'a c S. E
S S S '-a 5 «g "S
gl §•§ * S ^
8-S 8 a 1 § "2
T3 »- §2 -2P3 ^ -o
•S a UlS to. 2 oj e "o
'5 ^ 8 ts is e E S .2 &,
•= * -2 o e o <5 " M M
81 H It li 1 1
t/t fl1* cA ?* U ^ft ^^ ^* *^1
111 |l i! i! 1
6Sl ll 11 11 I S


H ^ ^
S 2 u § g
^ Q 52^
§ 1 8 1 § Q
oi g o S ^ z
< PQ P U fc %

-------






















•s
•o
j3
"o
O

X .S





*
<
a
a
u,
u.
s






required by UAM. Recommend sensitivit)
test by repeating simulations while varying
DIFFBREAK height. Use of tethersondes,





























acoustic sounders, tower data, etc., if
available, is preferred.























di
1
S
•o
o
E
60
Prepared using surface temperature data
collected at NWS and air quality monitorin
networks




.- Q-
si




«
D
i
TEMPE
*\
1
£
O)
•8
E

Based on land use information. Deposition
factors and surface roughness values have
been developed for urban-scale model

e
_o
"tn
a

u
'i
Q.
tn
G
O
gj
O
S

S
o
<£S
_C

S-n
- «
2 2
c 13
SI





S
<
«
<
CHEMP

13

"3
's

Most parameters do not vary between
simulations.
•o
_o
'C

-------
       Of the other 11 UAM input files, the most time-consuming and resource-intensive to
create  is the low-level carbon  monoxide emission file,  EMISSIONS.  This  file is usually
prepared concurrently with the preparation of the meteorological and air quality files.  State/local
agency participants using the UAM for areawide CO modeling should concentrate their efforts
on the emissions inventory development, especially the mobile source component, preparation
of the 3-dimensional wind fields, ground-level inversion height, and specification of thickness
of model layers.
3.2 Episode Selection

       This subsection describes the criteria to be "considered in selecting carbon monoxide (CO)
meteorological episodes appropriate for modeling.  The modeling protocol  should include a
complete discussion regarding episode selection.  A trained meteorologist familiar with local and
regional weather patterns should be consulted in the episode selection process.  The following
approach is recommended for selecting episodes for use in modeling:

       1.   Use the most recent years (1988 to present) of CO monitoring data as the period
           from which to select candidate modeling episodes.

       2.   Select as candidate modeling episodes the three highest non-overlapping 8-hour CO
           episodes from each year as determined by the highest monitored concentration per
           episode.

       3.   Examine the meteorological conditions for each candidate modeling episode.

       4.   Determine the different types of meteorological regimes for the candidate episodes.
           Conditions resulting in dissimilar source-receptor configurations should be the prime
           consideration in distinguishing different meteorological regimes. In areas dominated
                                          14

-------
           by mobile source CO emissions and stagnation conditions, there may be only one
           meteorological regime.  In areas with major point sources of CO and/or significant
           wood-burning CO emissions in addition to mobile source CO emissions, there may
           be  multiple  meteorological  regimes  because  of  dissimilar  source-receptor
           configurations.

       5.   Rank each  candidate episode within each meteorological regime according to the
           magnitude of the peak 8-hour CO concentration.

       6.   Model a minimum of one episode for each meteorological  regime.   Select the
           episode(s) for modeling from among the three highest ranked episodes from each
           meteorological regime.   In  general,  the  highest ranked  episode  within the
   	    meteorological regime should be selected for modeling.  However, there  may be
           circumstances in which the second or third ranked episode is more appropriate for
           modeling, e.g. (1) the air quality and meteorological data base is much better than
           that of the  first ranked episode,  (2) the second or third ranked episode is from a
           later year that includes controls that were not in effect in the year  of the highest
           ranked episode; consequently, the use of the second or third ranked episode would
           be more conservative, and (3) the CO  pattern for the highest  ranked episode is
           distinctly different  from all other candidate episodes  (i.e.,  weekend instead of
           weekday).

       States may want to consider a procedure other than the one outlined in Steps  1-6 for
selecting modeling episodes.  Any such procedure should be described in the modeling protocol
and approved by the appropriate EPA Regional Office.

       Recommendations
       The modeling protocol should include a complete discussion of the choice of modeling
       episodes.  It  is recommended that episodes selected for modeling be from the four most
       recent years, 1988  to present. A trained meteorologist should assist in determining the
       different types of meteorological regimes under which 8-hour exceedances occur during
                                          15

-------
       the four-year period. At a minimum, one episode for each meteorological regime should
       be modeled.
3.3  Selection of Modeling Domain and Resolution

       3.3.1   Domain definition

       The size  and location of the modeling domain define  the data requirements for the
modeling.  Definition of the modeling domain  (both horizontal and vertical dimensions and
duration of simulation) depends upon the meteorology of the particular episode to be modeled.
Criteria that play an important role in determining the modeling domain include:

       Stagnation period and duration of elevated CO levels
       Location  of major current and future emission sources
       Wind flow pattern
       Size of recirculation pattern
       Available aerometric data
       Preassumed emission inventory region

Generally, the domain should be set as large as feasible in order to reduce the dependence of
predictions on uncertain boundary concentrations and to provide flexibility in simulating different
meteorological episodes.   It is generally much easier to subsequently reduce the size  of a
modeled area than it is to subsequently increase  it.

       Recommendations
       It is recommended that the domain be set large enough to encompass all major current
       and future emission  sources, and reduce the dependence of predictions on uncertain
       boundary concentrations.
                                          16

-------
       3.3.2   Horizontal grid cell size

       The horizontal dimension of each model grid square is based upon (1) the sensitivity of
predicted concentrations to horizontal grid size, (2) the resolution of observed meteorological
and air quality data and/or estimated emissions data, and (3) limitations imposed by other
considerations such as a  required minimum domain size.  Air  parcel trajectories can  be
performed using wind observations taken on each day of the candidate carbon monoxide episodes
to determine the horizontal limits of the modeling domain.

       Relatively high horizontal grid cell resolution should be used in UAM carbon monoxide
modeling applications.  It is recommended that at a minimum the horizontal resolution should
be no greater than 2x2 km.  Model applications with  finer grid resolution lead to slightly
higher peak values because artificial  dilution  is kept to a minimum.   Larger (or coarser)
horizontal grid resolutions (> 2 km) are not recommended because (1) artificial dilution will
result for point emissions sources, and (2) loss  of spatial  resolution will result in less effective
evaluation of control strategy  effectiveness.

       Recommendations
       It is recommended that, at a  minimum, horizontal grid cell  resolution should be  no
       greater than 2x2 km. Smaller grid cell sizes are encouraged  because they allow more
       accurate gridding of area and mobile sources.  Additionally, point sources are better
       characterized by smaller grid cell sizes.

       3.3.3   Number of vertical layers

       In most UAM CO applications, nearly all CO emissions are confined to the near surface;
as a result, model applications  will need at a minimum two vertical layers:   a lower layer
extending  from the surface to  the diffusion break, and an upper layer extending from the
diffusion break to the top of the model domain. However, in some cases additional model layers
may be warranted.  At a minimum, these situations would include (1) multi-day elevated CO
events in which carryover from the previous day's emissions suspended aloft is believed to play
                                         17

-------
a significant role; (2) episodes where vertical wind shear is significant, resulting in differential
transport of CO emissions within each model layer; (3) episodes where elevated point source
emissions are significant, resulting in emissions being  suspended aloft under stable nighttime
conditions and not mixing downward until the surface-based inversion is fully eroded; and (4)
episodes where the vertical CO concentration profile changes rapidly below the diffusion break.

       To accurately describe the vertical structure, it  is recommended that data be collected
describing both the vertical temperature and CO concentration profile.  The number of vertical
layers below the inversion base should characterize the  vertical CO concentration profile.  The
'modeling  protocol document  should include a complete discussion regarding the basis for
choosing the number of vertical layers.

       Recommendations
       Based on previous UAM CO applications, it is recommended  that a minimum of two
       vertical model layers be used in the modeling study. However, additional layers may be
       warranted in cases  where meteorology  or emission source type justifies  the use of
       additional layers. The modeling protocol document should discuss the basis for choosing
       the number of vertical layers.   It is recommended that 20 m be used as the minimum
       depth of the vertical layers below the diffusion  break  and 20 m for  the vertical layers
       above the diffusion break.

       3.3.4  Time span

       The time span of the UAM simulations should be at least 18 hours. Longer time spans
may be required to examine episodes where carryover from earlier emissions is believed to play
a role in producing high carbon monoxide concentrations. For example, an 8-hour exceedance
ending in mid morning may include  substantial contributions from the previous evening. These
emissions must be modeled to properly treat future changes in daily emissions. If an episode
is selected  that shows persistent  recirculation  of materials or lengthy stagnation,  longer
simulation periods may have to be defined.

       A  primary  reason for extending the simulation period  is to reduce the influence of the
                                           18

-------
initial conditions on the predicted carbon monoxide concentration.  The starting time of the
simulation should be during the mid-afternoon hours when ambient concentrations are still low,
just prior to the development of the ground-level inversion.  The ending time usually  extends
into the mid-morning hours.

       Recom mendation s
       Simulations should extend for at least  18 hours.  Longer time spans  may be preferable
       when one day's exceedance can be traced to emissions from the previous day.
3.4  Preparation of Meteorological Inputs

       The  availability of meteorological  data  varies widely  among prospective  modeling
domains.  Also, a variety of techniques are available for developing wind fields, temperature
fields, and mixing heights.  Although high resolution and confidence for all meteorological data
are  desirable,  time and  resource  constraints  force  a compromise  between desirable and
acceptable methods.   Historically, measured meteorological data have been spatially and
temporally interpolated for most UAM applications. More recently, diagnostic and prognostic
meteorological modeling techniques have been explored as possible means to develop input fields
(particularly  wind  fields) for  air  quality models.    Procedures  for  preparation  of  the
meteorological inputs are presented in this section.  As shown in Table 3-1, with the exception
of the  EMISSIONS file, the WIND file is the most resource-intensive file to prepare.

       3.4.1 Surface roughness and deposition (TERRAIN)

       The TERRAIN file contains (1) surface roughness lengths  and (2) deposition factors.
The surface  roughness lengths are used in computation of vertical exchange coefficients.  Both
the roughness lengths  and deposition  factors  are  used in computing deposition of gaseous
species.
                                           19

-------
Because deposition of CO is negligible for most CO UAM applications, detailed representation
of surface characteristics is not crucial.  The magnitudes of vertical exchange coefficients are
unimportant,  given that maximum CO concentrations occur at the surface under very stable
conditions when mixing between UAM layers is minimal. It may be sufficient to assume values
of surface roughness and deposition factors typical of urban land use over an entire CO modeling
domain.  Appendix C details a methodology for developing surface roughness and deposition
factors.

       Recommendations
       Previous applications have found that using surface roughness and deposition  factors
       typical of  "urban" land use over the entire  CO modeling domain is sufficient because
       deposition of CO is  slight.
       It is recommended that, unless important urban versus rural land use areas are contained
       within the  modeling  domain, a surface roughness length of 0.5 m and a deposition factor
       of 0.3 be used.

       3.4.2   Diffusion break (DIFFBREAK)

In UAM terminology, the diffusion break (DIFFBREAK) is the height at which the upper and
lower  layers  are  divided.  The value of DIFFBREAK affects the way many of the other
parameters are used by the UAM.

       Predictions from the UAM have been shown to be fairly sensitive to the diffusion break
field.  Therefore, the temporal variations in the diffusion break field  over the UAM domain
should be  described as realistically as possible.   The  UAM modeling  system contains  a
methodology for  deriving the diffusion break based on surface temperatures, vertical sounding
measurements of  temperature, and cloud cover (EPA, 1990). Appendix C contains a technical
discussion of important considerations for determining the diffusion break for CO applications.
However, because of the diversity of techniques and data bases that may be available on a case-
by-case basis, a specific procedure for deriving the diffusion break field cannot be recommended
in all cases.
                                          20

-------
      Recommendations
      It is recommended that, at a minimum, the techniques described in the UAM User's
      Guide and/or Appendix C be used in establishing the diffusion break field in the domain.

      The choice of upper-air stations to be used in the diffusion break calculations should be
      based on prevailing wind fields and location of the upper-air stations relative to the UAM
      domain.  If there are no upper-air stations within the domain,  stations outside the domain
      may need to be used.  An experienced meteorologist should be consulted on the selection
      of upper-air stations for use in determining the diffusion break.

      The techniques for generating the diffusion break field should be described in the
      Appendix C protocol document.  Techniques  other than that described in the UAM
      User's Guide or should be documented and justified.

      Because the assignment of DIFFBREAK in CO UAM applications is relatively uncertain,
      it is recommended that the sensitivity of CO predictions to DIFFBREAK assumptions be
      assessed  by repeating the UAM simulations with  DIFFBREAK values (1) decreased by
      50 percent;  and (2) increased by 100 percent.
       3.4.3  Top of the modeling domain (REGIQNTOP)


       A spatially  and temporally constant top of the  modeling region  (REGIONTOP)  is

recommended for virtually all UAM applications.  Most 8-hour CO violations of the National

Ambient Air Quality Standards (NAAQS) occur during nighttime hours following the evening

traffic  rush.   Thus,  it is recommended  that for  most CO applications  (nighttime) the

REGIONTOP be set at 200 m.  However, this value may be low when:  (1) episodes extend

over several days, (2) high CO events are not associated with strong surface-based inversions,
or (3) elevated point source CO contributions are significant.  In these cases the REGIONTOP

should be specified above the highest DIFFBREAK height by at least the depth of one upper-
layer cell.


       Recommendations

       For most CO applications (nighttime) set the REGIONTOP to 200 m.
                                         21

-------
       3.4.4  Winds (WINDS)

       Methodologies to construct wind fields for UAM applications have historically fallen into
three categories:

       1.   Objective analyses that interpolate observed surface and aloft data throughout the
           modeling domain

       2.   Diagnostic wind models in which physical constraints are used in conjunction with
           objective analyses to determine the wind field

       3.   Prognostic models based on numerical solution of the governing equations for mass,
           momentum, energy, and  moisture conservation along with numerical solutions for
           thermodynamic processes

       Objective  analysis.  These procedures  generally involve straightforward interpolative
techniques.  They have the advantage of being relatively simple and inexpensive to use.  The
primary disadvantages are that these analyses contain limited  physical concepts, and results are
highly dependent upon the temporal  and spatial resolution of the observed values.  Thus, in
domains  containing sparse  observational  data or  complex  topography,  results  may  be
unsatisfactory.

       Diagnostic wind models.  These models improve mass consistency for the flow fields.
This may be addressed through parameterizations for terrain blocking effects and upslope and
downslope flows, as in the UAM Diagnostic Wind Model (EPA,  1990).  Diagnostic models
generally require minimal computer resources and can produce a three-dimensional wind field;
however, they need representative observational data to generate features such as land and sea
breezes.  Appendix C contains a detailed technical discussion of the use of the DWM for CO
model applications.
                                          22

-------
       Prognostic models. These models simulate relevant atmospheric physical processes while
requiring minimal observational data.  Prognostic models require specification of the synoptic-
scale flow.  Reliability of these approaches is usually enhanced if sufficient observations are
available to  "nudge"  solutions closer to observations.   Since  these  models  can simulate
temperature fields in  addition to the wind  field,  it is  possible  to determine  stabilities  and
inversion depth,  thus eliminating the need to generate these from sparse observational data.
Another significant advantage is that interdependencies of various meteorological inputs with one
another are  considered in prognostic  models.   A major disadvantage  is  the extensive
computational resources needed to run a prognostic model.   Additionally, the  availability of
evaluated models and expertise needed to apply them for general application with grid  models
is limited.

       Selection of a specific technique for generating the domain wind field depends largely on
(1)  the spatial and temporal  resolution  of  surface and  upper-air observations,  (2) available
modeling expertise  in  applying alternative meteorological models, and (3) available computer
resources.   However,  some guidelines on preferences for generating  the wind fields are as
follows.

       The development of a wind field for each modeling episode depends upon ground-level
and elevated wind observation data.  It  is preferred that a surface-based monitoring network
report  wind  speed and direction as  hourly averages because  an  hour  is the  time  period
commensurate with most UAM concentration output analyses. The surface monitoring network
should be broad and dense so that diagnostic models can depict major features of the wind field.
Data representing vertical profiles of wind speed and direction are required in order to establish
upper-level  wind fields.  Preferably, data  should provide  adequate  spatial (horizontal)  and
temporal resolution.  Results  of UAM -applications are often criticized because of inadequate
meteorological data,  and  lack of sufficient  meteorological data  often  prevents definitive
diagnostic analyses.   Thus, the need for adequate meteorological data cannot be overstated.
Appendix C contains a detailed technical discussion about the data needs for the DWM for CO
applications.
                                           23

-------
       Time  and/or resource constraints may preclude consideration of new meteorological
monitoring stations.  Thus, it  is likely  that the base  case  to be used  in the attainment
demonstration will be from a historical episode for which  model performance has been deemed

acceptable.

       Recommendations

       It is recommended that the DWM be used to generate the UAM gridded wind fields.  As
       discussed in Appendix C, the DWM modeling domain should encompass terrain features
       outside the UAM modeling domain which may affect the DWM representation of the
       flow field.  The use of other techniques for deriving the wind field,  such as prognostic
       wind models or other objective techniques, may be employed on a case-by-case basis,
       subject to approval from the appropriate EPA regional office.

       Meteorological data  routinely  available for a UAM modeling  demonstration usually
       consist of National Weather Service (NWS) hourly surface and upper-air observations
       (for winds  aloft).   If these data are the  only data  available for use in a modeling
       demonstration,  they may  have to  suffice.    However,  the NWS  data consist  of
       observations  made over very  short periods rather than hourly averaged values.   An
       assumption that wind velocity measured over a very short period persists unaltered over
       an hour may lead to an overestimate of transport. Therefore, whenever possible, hourly
       averaged meteorological data  (e.g., from  an intensive field study) should be used.
       Additional meteorological data may be available from  other sources in the domain (e.g.,
       an on-site meteorological monitoring program  at an industrial facility). These data may
       be used to  supplement  the NWS data, provided the data have been adequately quality
       assured.   Additionally, the EPA guideline entitled  On-Site  Meteorological  Program
       Guidance for Regulatory Modeling Application (EPA,  1987) should be consulted to
       assess whether the supplementary data reflect proper siting of meteorological instruments
       and appropriate data reduction procedures.

       In planning a special field  study  to  provide a more  spatially and temporally dense
       meteorological data  base, the number of surface meteorological  monitoring stations
       should be sufficient to describe the predominant wind flow features within the modeling
       domain.  An experienced meteorologist familiar with local climatic patterns should be
       consulted concerning the location and suitability of the  surface meteorological stations.
       Vertical sounders or profilers  are highly encouraged  in a special field study to resolve
       winds aloft and vertical temperature gradients.  Any special field study and monitoring
       program  should  be planned in consultation with  the appropriate  EPA regional office
       before implementing the study.
                                           24

-------
       3.5 Preparation of Air Quality Inputs

       Ambient air quality data are generally used to specify initial- and boundary-condition
concentrations.   In the  UAM, these functions are performed by  the Air  Quality,  Top
Concentration and Boundary Condition files. In addition, air quality data are needed to diagnose
problems  in  setting  up model  applications and  assessing  model  performance for  the
meteorological episodes being considered in the attainment demonstration.  A lean air quality
data base  may introduce significant uncertainties in characterizing model performance.  This
section discusses the procedures for preparing air quality inputs  needed by the UAM.

       3.5.1   Air quality (AIROUALITY)

       The air quality (AIRQUALITY) file specifies the initial  pollutant concentrations  to be
modeled by the UAM.  The concentrations are specified for the full three-dimensional grid.

       A  routine monitoring data base usually does not,  by  itself, make a complete, three-
dimensional specification of initial conditions possible. For instance, spatial coverage provided
by monitoring stations is generally incomplete, with most stations located within urban centers
and therefore  strongly influenced by local emissions sources.  Monitors located in rural areas,
which can give an indication of appropriate local background concentrations, are rare.  As a
rule, only surface measurements are available, so  no direct  indicator of the vertical distribution
of carbon monoxide will be  available.    The problem,  therefore,  is  to specify the three-
dimensional concentration field from a very sparse data set. A detailed technical discussion is
presented  in Appendix C for developing initial conditions from a sparse data set.

       Recommendations
       At the inflow boundaries, air quality data at the  surface and aloft should be used
       whenever available to specify the initial boundary conditions.
       For values near the surrounding perimeter  of the model domain, default values as shown
                                           25

-------
       in Table 3-2 may be used where necessary.  Selection of these values should be based
       on how pristine the surrounding areas are considered to be.  For example, if the area
       near the boundary can be classified as rural, then the low CO value should be used; if
       that area is better classified as suburban, then the middle value should be used.
       To diminish dependence on  arbitrary specification of  initial conditions, a simulation
       should begin prior to the buildup of high carbon monoxide concentrations.
       3.5.2  Top concentration (TOPCONC)

       The recommended  method  for  setting the top  concentrations in cells  above  the
DIFFBREAK is to set them equal to the observed concentration at the top of the region. When
top concentration  is not known,  the concentration  should be  no higher than the middle
concentration level of Table 3-2. If measurements are available indicating higher concentrations
exist at the top of the  region, they can be used, but the user should verify that they  are
representative of more than just a local value before using them for the entire grid.

       In the absence of upper air concentration data, the values in TOPCONC would be the low
values  from Table 3-2  for a relatively remote area.   If there is evidence that considerable
recirculation of carbon monoxide pollutants may occur, these values should be raised to the mid-
level values.  The mid-level values should also be used if the region is surrounded by other
urban/suburban areas that would elevate the background pollutant concentrations.

       Recommendations
       It is recommended that  the concentration at the top of the region  be  used to set
       concentrations in cells above the DIFFBREAK.  If no data are available for upper air
       concentrations, values should be set by  referring to Table  3-2, considering the level of
       urbanization and the importance of recirculation.

       3.5.3  Boundary conditions (BOUNDARY)

       The BOUNDARY  file  contains both the definition of the physical boundaries of the
                                          26

-------
TABLE 3-2. Recommended background
concentrations for carbon monoxide
(Killusetal., 1982).	

              Area      Concentration
 Value     Classification      (ppm)

 Low         Rural           0.1

 Middle     Suburban         0.2

 High         Urban           0.5
                       27

-------
region to be modeled and the concentrations along each of the lateral boundaries.  When there
are no available data near  the lateral boundaries of the region, specification of the boundary
concentrations can be difficult.

       Two approaches for specifying boundary conditions for UAM simulations are as follows:
(1) use objective/interpolation techniques with a  sufficient amount of measured data (i.e., data
from an intensive field program) and (2) use default background values and expand the upwind
modeling domain to mitigate uncertainties due to paucity of measurements.

       Ideally, the preferred technique would be based on an intensive field program.  However,
this approach is seldom feasible for historical episodes.  Presented next are recommendations
for implementing  each technique just  identified for deriving boundary conditions, including
discussion  of the advantages and  disadvantages  of each technique.   A  detailed  technical
discussion for developing boundary conditions for a  sparse data set is presented in Appendix C.

       Use of measured data. All sources of air quality data  for a particular modeling domain
should be evaluated for applicability in establishing  boundary conditions.   Unfortunately,  most
ongoing monitoring programs have been designed  (understandably  so) with a  receptor-based
orientation.  While available monitoring data are  useful for evaluating model performance, they
usually are not adequate  for establishing boundary concentrations.

       Use of default values - Some urban areas may lack adequate data suitable for establishing
boundary  conditions.  Section 3.3 on domain selection and Chapter 4 on diagnostic analyses
recommend constructing domains large enough  to minimize  the sensitivity  of inner core and
downwind concentrations to assumed boundary conditions.

       Boundary-condition concentrations are influenced by large- and  small-scale  weather
patterns and emissions distributions  that are unique to each modeling domain.  Thus, case-
specific attributes should be used  in estimating  these concentrations  whenever feasible.   For
example, boundary concentrations of regions where residential wood combustion  is significant
                                           28

-------
are likely  to  be higher  during  very cold  conditions than during  periods  of more mild

temperatures.


       Recommendations
       To develop boundary conditions, it is recommended that one or more monitoring stations
       be sited upwind of the central urban area along prevailing wind trajectories that give rise
       to carbon monoxide exceedances.

       At the  inflow boundaries, air quality data at the surface and aloft should be used
       whenever available to specify the boundary conditions.

       Those  having to  use default values  should  plan to perform diagnostic/sensitivity
       simulations  (see  Chapter 4)  to  evaluate the sensitivity  of domain-interior  model
       predictions to the boundary conditions.

       Table  3-2 lists  the  recommended default boundary  values depending on  the area
       classification.  When using default values,  the boundary of the domain should extend as
       far upwind as practicable.


3.6  Performance  Evaluation Data


       Ambient carbon monoxide measurements are needed to diagnose problems in  setting up

model applications  and assessing model performance for the meteorological  episodes  being
considered in the attainment demonstration.   A lean air quality  data base  may  introduce

significant uncertainties in characterizing model performance. The Technical Committee should
agree on the adequacy of the existing  data for assessing model performance.


       Recommendations
       For lean data bases the Technical Committee should scrutinize in detail the adequacy of
       the  data base to ensure that model performance that appears to be acceptable has  not
       actually resulted from compensating errors  in the data bases.   Additional diagnostic
       analyses may be necessary for lean data bases  from historical episodes.
                                           29

-------
3.7    Emission Inventory

       The credibility of UAM applications  is directly tied to formulating the best possible
emission inputs.  Model performance may hinge on how well emissions are estimated.  Also,
in the attainment demonstration, modeling results are used to determine emission scenarios that
lead to improved air quality levels consistent with the NAAQS.  A faulty emission inventory
could lead to erroneous conclusions about the extent of needed controls.

       Carbon monoxide  emissions  in  most regions  are primarily from  motor  vehicles.
However, in some  regions emissions from residential wood  combustion  may  contribute
significantly to the total carbon monoxide emissions. Stationary sources are composed of either
point sources, which are sizeable stationary emission sources  at specific locations, or area
sources, which are emissions from stationary and non-roadway  mobile sources that are too small
and/or too numerous to be included in the point source inventory (e.g., wood stoves).  For most,
but not all, areas of the country, the peak CO season is the wintertime months.  Therefore, the
focus of CO emission inventory development is the on-road vehicle emissions for wintertime CO
episodic conditions.  However, discussion will begin with an overview of information sources
that are  available  to explain the development of stationary and  area source CO emission
inventories suitable  for use  with UAM.

       Much of the information required to assemble CO modeling inventories will have already
been assembled for the base year inventory, which is required for all CO  nonattainment areas
under the Clean Air Act Amendments of 1990 (CAAA).   Specific information on base year
inventory requirements is contained in  Emission Inventory Requirements for Carbon Monoxide
State Implementation Plans  (EPA, 1991b).  The base year inventory for the CO SIP submittals
due November 1992 will be from the  1990 base year.   Other documents (EPA, 199Ic; EPA,
1988; Laxton, 199Ib)  describe the development of emission inventories  from raw data,  and
specifically address  questions regarding the methodology to spatially and temporally resolve CO
emission estimates contained in the base year inventory so that they can be used with the UAM.
                                          30

-------
       For use in regulatory applications of the UAM, the base year modeling inventory will
have to undergo several adjustments.  First, the inventory needs to be adjusted to be consistent
with meteorological conditions  during  each  selected episode  (i.e.,  "base  year  day-specific
emissions").  Second,  the resulting "base year day-specific emissions"  should be adjusted to
reflect control programs and activity levels prevailing during the year(s) of selected episodes.
For example,  if a selected episode occurred  in 1988, the "base year day-specific emissions"
would be further adjusted to reflect controls and activity levels prevailing in  1988. This latter
adjustment is  needed   to provide  an  estimate of  emissions  most suitable  for evaluating
performance of the UAM.

       As noted in Chapter 1, once the UAM's performance has been evaluated and the model
has been determined to perform satisfactorily, it is used to derive control strategies to attain the
NAAQS. This requires another adjustment to  the "base year day-specific emissions" that entails
use of growth factors,  ongoing control  programs and retirement rates for obsolete sources of
emissions to project "base year day-specific emissions" to the years by which the CAAA specify
that the NAAQS must be attained.  The resulting attainment year modeling inventory is used as
a starting point from which to  construct a control strategy inventory,  which  is obtained  by
superimposing  additional  control measures  on sources of emissions in  the attainment  year
modeling inventory.

       In summary, a base  year  modeling  inventory is  first  adjusted to  evaluate UAM
performance.  The base year modeling  inventory is  then readjusted  to reflect emissions most
likely to  occur at the time the CAAA require attainment of the NAAQS.

       Two emission files drive the UAM~a file of emissions  that are  injected into the first,
surface-based vertical layer, and a file of elevated point source emissions that are  injected into
vertical layers above ground level.  The UAM Emissions  Preprocessing System (EPS) (EPA,
1992b) reads county-level area- and point-source files and performs three  major functions: (1)
area sources and point  sources are allocated to grid cells; (2) temporal profiles are assigned to
source categories; and  (3) point  sources with  effective plume heights greater than  a prescribed
                                           31

-------
cutoff level are assigned to the elevated point source file, and the remaining point sources are
assigned to the surface-layer emissions file.

       Emission inventories are  developed for stationary area,  stationary point, and off-road
emission inventories that are appropriate for the UAM. The following issues arise in developing
emission input data:  (1) use of surrogate factors to grid area sources, (2) treatment of mobile
sources,  (3)  episodic  adjustment  of inventories  to  day-specific modeling  inputs, and (4)
consistency with national inventories.

       3.7.1  Use of surrogate factors to spatially grid area sources

       Area source emission data, including motor vehicle emission data, are often supplied on
a county basis. Spatial allocation of county-level emission estimates to grid cells is performed
for each identified area source category and requires use of surrogate distribution factors such
as population distribution, land use, and road type. The UAM  EPS (EPA, 1992b) contains  a
program that uses gridded surrogate factors to allocate county-level emissions data to the grid
cell size of the modeling domain.

       Recommendations
       It is recommended  that the emission  inventory guidance  document (EPA,  199 Ic) be
       consulted  for  alternative surrogate  factor  choices  and  sources of  information for
       assimilating surrogate data.  The EPA is currently developing a utility to provide gridded
       surrogate  data.  States will be notified of the availability of gridded  surrogate data
       through the EPA regional offices.

       3.7.2 Development of mobile source inventories

       Under the requirements  of the CAAA,  all CO nonattainment areas  are required  to
develop base year inventories of mobile sources.  These represent base year emission levels for
a typical operating day during the designated peak CO season.  As part of the development of
a  base year inventory, estimates of VMT  must be developed  from traffic  ground counts

                                           32

-------
consistent with the existing Highway Performance Monitoring System (HPMS). Projections of
vehicle miles traveled (VMT) growth must be developed for moderate CO nonattainment areas
through historically based extrapolation techniques if better methods are not locally available.
Non-exempt serious areas must use a transportation demand model to project VMT growth
(EPA, 1991a).  These VMT estimates will  serve as  the basis of the modeling inventory.
Guidance for making projections is detailed in Procedures for Preparing Emissions Projections
(EPA, 199 If).

       The mobile source CO emission are complex  and difficult to prepare because they are
strongly affected by the ambient temperature and motor vehicle speed. There are two primary
ways to prepare a mobile source inventory.   One relies upon county-wide estimates of VMT,
which are multiplied by a "composite" emission factor (expressed as grams pollutant emitted per
mile of travel).   The emission  factor is referred to as "composite" in that it is a  weighted
average value that is representative of the average vehicle operated under average conditions
(Ireson et al., 1991).  The other method relies upon the output of transportation models. These
models can provide roadway link-specific estimates of distance,  traffic volumes, and speed, and
estimates of trip ends for specific transportation analysis zones (TAZ), which are combined with
"composite" emission factors to arrive at regional mobile source emission inventories. Examples
of commonly used travel demand models are the Urban Transportation Planning System (UTPS),
which is maintained by the U.S. Department of Transportation, and commercial travel demand
modeling software packages such as TRANPLAN (UAG, 1990) and  MINUTP (Comsis Corp.,
1991). These models have traditionally been used in regional transportation planning offices as
planning tools for roadway expansions, highway development, land use, etc.  However, the
information provided by travel demand models, coupled with  EPA's MOBILE  model (EPA,
199Id), can be used to produce better estimates of mobile source emissions rather than estimates
based on county-wide VMT estimates.

       Recommendations
       Use  of  output from a  transportation  demand model  is the preferred approach for
       estimating vehicle  activity  levels  and emission factors  because  this method  allows

                                          33

-------
       resolution of variations in speed and vehicle miles traveled (VMT) among different grids
       over hourly time slices.  The transportation demand model approach  is  the most
       appropriate for addressing the inner urban core of modeling domains.  Peripheral,  less
       dense traffic areas can be treated by disaggregating county-level estimates of VMT. Care
       should  be taken not to double count  mobile source emissions as part of both the
       transportation demand model approach and county level disaggregation approach (EPA,
       1992b).  Exceptions to these recommendations should be considered by the Technical
       Committee on  a  case-by-case  basis.    Justification for  more  extensive use  of
       disaggregating  county-level  emissions  should  be  sought in  discussions  with  the
       appropriate EPA regional office.
       Appendix C contains a detailed technical discussion of procedures specific to the use of

EPA's latest MOBILE model to develop emission factors for area-wide CO modeling.


       3.7.3  Episode-specific adjustments


       Both motor vehicle and residential wood combustion emissions are sensitive to ambient

temperature.  Thus, it is important for modeling inventories to reflect episode-specific ambient

temperature.  In addition, known episode-specific events such as changes in process operations

for point sources  affect emissions rates  and should be  reflected  in  the episode  modeling

inventory.


       Recommendations

       Mobile-source emissions should be adjusted for episode-specific temperatures.  Emission
       factors for deriving episode-specific  mobile-source emissions  should use  the latest
       MOBILE model.  Use of models other than the latest EPA MOBILE model  should be
       reviewed by the Technical Committee on a case-by-case basis, and is subject to approval
       by the EPA regional office.

       If available,  episode-specific  operating rates for  point sources  are  preferable for
       estimating temporal point-source emissions. Procedures for temporally adjusting point
       and area sources are also provided  in the emission inventory guidance document.
                                           34

-------
       3.7.4  Consistency with national inventories

       Comparisons should be made between the modeling inventory and the 1990 SIP and RFP
tracking emission inventories reported in the EPA Aerometric  Information Retrieval System
(AIRS). Although these inventories will not be identical, such a check can be considered part
of the quality assurance process.  Major inconsistencies should be noted and documented.

       Recom mendation s
       For an acceptable attainment demonstration,  documentation should be provided that
       shows that the modeling emission inventory is consistent with the emission inventory
       being reported in AIRS in accordance with applicable guidance and  regulations.
                                         35

-------

-------
                                     CHAPTER 4
      DATA QUALITY ASSURANCE AND MODEL DIAGNOSTIC ANALYSES

      This chapter provides general guidance for quality assurance testing of component data
input  fields and diagnostic testing of base case episodes.   These analyses  are  designed  to
establish and  improve reliability of the input data and proper functioning of the model.

      Although the UAM has been evaluated on a number of historical data bases, measures
of model behavior with respect to observed data are necessary for new applications.  Model
developers and users perform diagnostic tests to uncover potential input data gaps that, when
corrected, may lead to unproved treatment of model processes. Regulators need some indication
that the model captures the key features of the base meteorological episodes being applied in the
model simulations in order to have confidence in  the model's ability to predict future carbon
monoxide (1) after applying projected  growth and planned emission controls, and (2) after
applying alternative emission control  strategies.

      Important prerequisites for a model performance  evaluation (see  Chapter 5) are (1)
quality assurance testing of model inputs and (2) diagnostic testing of the base meteorological
episode  simulation to ensure that the model is functioning properly and that apparently accurate
model results are being obtained for the right reasons. For example, quality assurance testing
of input data helps to ensure  that  apparently  good model results have  not resulted from
compensating errors in input data.

      An excellent compilation  of model performance  evaluation  techniques, including
diagnostic tests and related issues,  is contained in  Tesche et al. (1990). Although the Tesche
study was developed for photochemical modeling, much of what is presented is applicable to
                                          37

-------
area-wide CO  modeling.   The  study  by Tesche also serves as the basis for the model
performance evaluation described in Chapter 5 of the work reported here.  Various graphical
and numerical measures described in the following paragraphs are treated in detail in Tesche et
al. (1990).   Two graphical displays used for both quality assurance and diagnostic testing are
mapping and time-series plots.

       Mapping is a two- or three-dimensional spatial display of values illustrated with various
contouring and  tiling methods.  These displays may depict political boundaries and monitoring
site locations as well.   Mapping  capability is a multipurpose tool applicable for all forms of
gridded data, such as future-year emission control strategy results and most input data fields
(e.g.,  gridded wind fields, temperatures, and emission densities).  Point- source locations may
also  be depicted to ensure that they are properly located.  Spatial displays of predicted and
observed carbon monoxide patterns are particularly useful as part; of a model  performance
evaluation.

       Time-series  plots  display  hourly and eight-hourly measured  and  predicted carbon
monoxide values for specific locations such as monitoring sites.  Time-series plots provide an
overview of the temporal performance of the model predictions. Comparison of time-series plots
across multiple monitoring sites provides an indication of spatial response.   These plots may
provide insights to  carbon monoxide prediction patterns and also to data base inconsistencies
requiring further investigation.

       The following sections describe steps recommended for conducting diagnostic testing of
each base case  meteorological episode simulation.

4.1    Step 1;   Quality Assurance Testing of Component Fields

       Starting with initial, quality-assured data, input data are developed for use in various
UAM preprocessors. The first stage of diagnostic testing should focus on assessing the accuracy
of major UAM input fields produced by the  UAM preprocessors.  Generally, the testing is
                                           38

-------
qualitative  in  nature and  based on comparing visual displays of preprocessor outputs  with
patterns exhibited by the observed data.  Prior to conducting a base case meteorological episode
simulation, individual air quality, meteorological, and emissions fields should be reviewed for
consistency and obvious omission errors.  Both spatial and temporal characteristics of the data
should be  evaluated.  These checks  may  be only cursory,  but errors uncovered by this
component testing might be extremely difficult to diagnose later in the modeling process, when
errors could arise from any subset of the data inputs.  Examples of component testing include
the following:

       Air Quality:  Check for correct order of magnitude, especially when using background
                    values

       Emissions:    Plot various source types by grid cell and review major source locations
                    with local emissions patterns; check major highway routes; generally, look
                    for obvious omission errors; plot CO by grid cell and cross-check  with
                    source  distribution  for  logical  patterns, such  as  high  motor vehicle
                    emissions near the urban core

       Meteorology: Plot surface and elevated wind vectors  and compare with  monitoring
                    stations and  weather maps  for consistent patterns;   compare  diffusion
                    break heights with sounding data; check  temperature fields

       In quality assurance testing of component input fields, the emphasis is on capturing large
errors before performing model simulations.

       Recommendations
       It is recommended that quality assurance  testing of the  air quality,  emissions,  and
       meteorological data input files be conducted before proceeding to diagnostic testing of
       the base case meteorological episodes. At a minimum, emissions data should be quality
       assured by looking at emission distribution maps  and known  source locations and
       emission strengths.
                                           39

-------
4.2    Step 2;  Diagnostic Testing of the Base Case Meteorological Episodes


       After confidence in the accuracy of UAM input fields has been achieved, the UAM

should be exercised for each base case meteorological episode.  The initial run is termed a

diagnostic simulation because review of initial base case simulations may uncover additional

input errors requiring correction before an acceptable set of base case inputs can be derived.

During this stage of the process, emphasis is placed on assessing the model's ability to correctly

depict  area-wide  distribution  and   the  timing of observed  carbon  monoxide  maxima.

Accordingly, visual methods such as mapping and time-series plotting, using measured data as
reference marks, may be used to assess model behavior.


       Recommendations

       To aid in interpreting simulation results, it is recommended that predicted and observed
       carbon monoxide concentration maps be constructed for each base meteorological episode
       simulation. Concentration maps present spatial information on the structure of the carbon
       monoxide cloud.  Maps at 1-hour intervals should be constructed over the modeling
       period.

       Consideration  should also be given  to  constructing a map that depicts the highest
       predicted  1-hour and 8-hour maximum carbon  monoxide  value for each grid  cell.
       Examples of various mapping techniques are described in Tesche et al. (1990).

       It is also recommended that the method used to predict concentrations used in time-series
       plots be consistent with the method for deriving predicted concentrations for the model
       performance evaluation described in  Chapter 5.  Time-series of both  1- and 8-hour
       average periods should be constructed.

       Other methods for deriving predicted concentrations for time-series comparisons may be
       judged  appropriate by the Technical  Committee; some suggestions are  contained in
       Chapter 5. These methods should be described in the Modeling Protocol.
                                          40

-------
4.3  Step 3;  Additional Base Meteorological Episode Sensitivity Testing

       In addition to running the  base meteorological  episode  diagnostic simulation,  other
episode diagnostic simulations that perturb levels of emissions, initial and boundary conditions,
and meteorological inputs may provide valuable information for identifying critical input areas
and  ensuring proper  domain  and episode selection.   The  following  simulations,  which are
equivalent to sensitivity tests on major model inputs, illustrate the utility of this exercise.

       1.    Zero emissions.  To indicate levels of sensitivity to emissions, all emissions are
             set to zero and the resulting predicted concentrations are compared with the base
             meteorological episode predictions that include emissions.  A lack of substantial
             sensitivity may indicate a need to reexamine the selection of episodes or domains.
             Variations can be performed by zeroing out emission subsets,  such as mobile-
             source emissions, and individual source  categories.

       2.    Zero boundary concentrations. Inflow concentrations at the lateral boundaries and
             top of  the  modeling domain are reduced to zero or  low background levels.
             Sensitivity of concentrations in  the inner core and downwind portions of the
             modeling domain provide a measure of the boundary conditions' influence.  This
             simulation provides assurance that the upwind extent of the domain is adequate.

       3.    Zero initial concentrations.  Initial concentrations for all grid cells are reduced to
             zero or low background levels. Sensitivity of concentrations within the modeling
             domain provides a measure of the initial conditions' influence.  Changes of less
             than  a  few percent  indicate  that  the  initial conditions are  not dominating
             concentration estimates with the domain.
                                           41

-------
\      4.     Diffusion break and wind speed variations. Much uncertainty is associated with
             the diffusion break and wind  speeds, and simulated  concentrations are often
             sensitive to these inputs.  Simulations that test the sensitivity of model estimates
             to variations in wind speed and/or diffusion break provide bounds on some of the
             uncertainty resulting from these parameters. Large sensitivity values may suggest
             that future model applications will need improvement in the meteorological data
             bases.

       Certain  numerical measures,  which  are  recommended in the  discussion  of model
performance evaluation in Chapter 5, are also useful diagnostic tools. For example, consistent
underpredictions usually produce numerical values greater than zero using these measures (see
Appendix B).  This phenomenon could be due to various factors, such as overstatement of wind
speeds or diffusion break heights, or underestimation of emissions or the number of vertical
layers.  Modelers are encouraged  to  use  numerical as well as  graphical techniques in the
diagnostic process.

       The diagnostic analyses described in this chapter are considered to be a starting point for
a specific modeling study.    Diagnostic tests discussed in Tesche  et al.  (1990) should  be
considered whenever possible.

       Recommendations
       Diagnostic testing of the model should begin with quality assurance testing on input data
       files  (Section 4.1). Diagnostic testing of each base meteorological episode should follow
       (Section 4.2).  Additional diagnostic sensitivity tests for the base episode should also be
       considered (Section  4.3),  including using  zero  emissions  and/or  zero  boundary
       conditions, zero initial conditions, and  varying diffusion break and wind speed estimates.
                                                                                    •
       Agreement should be  obtained among  members of the Technical Committee concerning
       input field modifications arising from the quality assurance testing. These modifications
       should be based on scientific or physical reasoning and not just on  what will improve
       model performance. All changes to the data that result from the diagnostic testing should
       be documented and justified.
                                           42

-------
In addition, all diagnostic steps should be documented to avoid misinterpretation of model
performance  results.   After  confidence  is  gained that  the  simulation  is  based  on
reasonable interpretations of observed data,  and model concentration fields generally
behave  both  spatially and temporally with  known carbon monoxide distribution, a
performance  evaluation  based on  numerical measures  is conducted  for  each base
meteorological episode (see Chapter 5).

-------
                                     CHAPTERS
                      MODEL PERFORMANCE EVALUATION

       The purpose of the UAM application is to simulate a historical CO episode in order to
estimate the area-wide contribution to concentrations measured at neighborhood-scale monitors.
Once the model has adequately simulated observed CO magnitudes and spatial and temporal
patterns for a historical day, greater confidence can be placed in its ability to provide reliable
estimates of concentrations under the same meteorological conditions for a variety'of emission
scenarios.  Thus UAM applications can serve as predictors of future concentrations and assist
in the formulation of control strategies for attaining CO air quality standards.

       Comparison of model predictions with observed values may lead to uncertainty, especially
if samplings are from locations not representative of area-wide concentrations. Furthermore, the
UAM output represents a volumetric (typically, 8 x 107 m3 ) concentration, whereas air quality
data represent point locations that may or may not represent the same volume. Because of these
uncertainties, specification of rigid rejection/acceptance criteria has not been generally supported
by model developers or decision makers participating in previous modeling efforts.   Instead,
performance measures based on past modeling applications are thought to provide a reasonable
benchmark for acceptable model performance.

       Poor performance may necessitate (1) delaying model applications until further diagnostic
testing and  quality assurance checks are reflected in the input data base, or (2) selecting another
meteorological episode for modeling.  Cases  where good model performance is shown should
also be reviewed because compensating errors can induce spurious agreement among observed
and predicted values.
                                        45

-------
5.1    Performance Measures

       This section describes recommended graphical and statistical performance measures for
carbon monoxide predictions.

       The measures used in the performance evaluation should include both qualitative (e.g.,
graphical) and  quantitative  (e.g., statistical)  analyses.   Statistical measures may provide a
meaningful test of model performance for dense monitoring network!!, such as those for special
field studies. However, for some routine monitoring networks where coverage may be sparse,
statistical measures may provide a distorted view of model performanee,  especially for paired
values.

       Tesche et al. (1990) provides detailed descriptions of graphical and statistical measures
available for assessing model performance.  Although these  methods were  formulated primarily
for photochemical grid models, most of the  model performance measures are applicable to area-
wide carbon monoxide modeling. The Technical Committee should consult the Tesche study
when formulating model performance evaluation methods, and may want to use it  for developing
additional performance evaluation procedures other than those recommended in this guidance
document.

       5.1.1 Graphical performance procedures

       Graphical displays can provide important information on  qualitative relationships between
predicted and observed concentrations.  At  a minimum, the following graphical displays should
be developed for each meteorological episode:  time-series plots and ground-level isopleths.

       Time-series plots.  The time-series plot, developed for each monitoring station in the
modeling domain,  depicts the hourly predicted and observed concentrations for  the simulation
period.  The time series reveals the model's ability to reproduce the peak prediction,  the
                                        46

-------
presence of any significant bias within the diurnal cycle, and a comparison of the timing of the

predicted and observed maxima.


       Ground-level isopleths or tile maps.  Ground-level isopleths or tile maps display the

spatial distribution of predicted concentrations at a selected hour. Isopleths of predicted maxima

may also be constructed.  The isopleths provide information on the magnitude and location of

predicted carbon monoxide concentration.  Superimposing observed hourly or  daily  maximum

concentrations on the predicted isopleths  reveals  information  on the spatial  alignment  of

predicted and observed concentrations.


       Recommendations
       At a minimum, the following graphical displays are recommended  in the evaluation of
       each meteorological episode:

       Time-series plots of predicted* and observed hourly carbon monoxide values should be
       constructed for each  simulation period for each  monitoring  station where data  are
       available.

       Ground-level isopleths or tile maps of the spatial distribution of predicted concentrations
       should be  constructed for selected hours.  Also,  ground-level isopleths or tile maps of
       the  carbon monoxide maxima  should be constructed.   The corresponding observed
       concentrations should be superimposed on the predicted concentration isopleths to analyze
       spatial patterns and carbon monoxide magnitudes.

       Additional graphical displays such as  scatterplots of predictions and observations may
       also be used  to assess model performance.  The graphical displays to be used in the
       model performance evaluation should be described  in the modeling protocol.


       5.1.2 Statistical performance  measures


Statistical measures  provide a  useful  measure  of model performance  for  spatially  dense
monitoring networks; however, for routine urban area CO monitoring networks, the typically
    *For this purpose, the predicted value is the weighted average of the predictions from the
four grid cells nearest to the monitoring station.  The four-cell weighted average is derived
from bilinear interpolation as described in EPA (1991e).
                                         47

-------
sparse coverage may result in a statistically distorted view of model performance.  However,
on the basis of UAM applications in past area-wide CO modeling, it is recommended that the

following three statistical  criteria  be  applied  to  all  neighborhood-scale monitors  (and,  if
applicable, roadway intersection monitors showing persistently high CO values during low traffic
volumes):


      Recommendations


      It is recommended that, at a minimum the following three formulations be applied as

      measures for model performance evaluation.

      1.  Unpaired (time or space)  highest 8-hour prediction accuracy.  This measure
      quantifies  the difference between the highest observed 8-hour value and the
      highest predicted 8-hour value over all hours and monitoring locations.

      2.  Average absolute error in 8-hour peak prediction accuracy paired ftime and
      space') values greater  than 5.0 ppm .  This measure quantifies the difference
      between the highest observed 8-hour value and the highest predicted 8-hour value
      at the time and location of each observed maximum.

      3.  Average absolute error in the predicted time of the 8-hour peak concentration.
      paired by  station values greater than  5.0 ppm  .  This measure quantifies the
      difference between the highest observed 8-hour value and the highest predicted
      8-hour value at the location of each observed maximum within a window of time.
Additional statistical measures may also be applied.  Other available measures are listed in the

Guideline for Regulatory Application of the Urban Airshed Model (EPA, 1991e).


5.2  Assessment of Model Performance


       As noted, both graphical and statistical performance measures should be used for the

performance evaluation.   However, statistical  measures should be used with caution  when

interpreting results derived from sparse monitoring networks. The Technical Committee should

consider the monitoring network design in interpreting statistical measures.
                                         48

-------
       Although only a limited number of UAM area-wide CO modeling evaluations have been
conducted, the following statistical  performance  measures should be  achievable  for SIP
applications:

       •      Unpaired (time or space)  highest 8-hour prediction accuracy:  ± 30-35
             percent

       •      Average absolute error in 8-hour peak prediction accuracy for paired
             values (time and space)  > 5.0 ppm:  25-30 percent

       •      Average  absolute error in  the  predicted  time  of the  8-hour peak
             concentration, paired by station >  5.0 ppm: 2 hours

In general, performance results that fall within these ranges would be  acceptable.  However,
caution is urged in using these ranges as the sole basis for determining the acceptability of model
performance. These ranges were derived from limited past model performance evaluations with
varying densities  of air quality and meteorological monitoring  networks and corresponding
variations in the quality and quantity of aerometric model input data. In some cases, they reflect
use of earlier versions of the UAM.  Thus, these ranges should be used  in conjunction with the
graphical procedures to assess overall model performance.

       If statistical results are worse than the above ranges and graphical analyses also indicate
poor model performance, users should consider choosing an alternative  meteorological episode
for modeling.  Performance evaluations should be done on other candidate episodes to identify
those that might result in better model performance.

       If statistical results are worse than the above ranges for any of  the three statistics, but
graphical analyses generally indicate acceptable model performance, simulation results used for
attainment demonstration should be applied with caution.  Users may consider conducting
                                        49

-------
performance evaluations on other candidate episodes to identify any that might yield improved
model performance.

       When performance is less than expected, the assumptions and data used as model inputs
should be assessed, and probable causes of poor performance should be examined before the
model is applied to attainment demonstrations. If performance is poorer than expected using the
existing aerometric data as model input and the existing data base is considered inadequate, the
responsible regulatory agency should consider an enhanced monitoring program to improve the
aerometric data base for future attainment demonstration modeling studies.

       Decisions regarding the acceptability of using a modeling episode with poor performance
should be reviewed and approved by the EPA regional office. Supporting documentation should
include a discussion as to why performance was poorer than expected, and the potential adverse
effects of poor model performance  on  control strategy evaluations.

       Recommendations
       It is  recommended that the model performance for each meteorological episode  be
       assessed as follows:
       1.  The graphical performance procedures specified in Section 5.1.1 should be conducted
       for each meteorological episode. To assess model performance, the Technical Committee
       should  review the time-series plots and ground-level isopleth plots.
       2.  The statistical performance measures specified in Section 5.1.2 should also be derived
       and evaluated for each meteorological episode. When interpreting these measures, the
       monitoring network density and design should be considered.  Caution is urged  when
       interpreting the statistical measures  for a sparse monitoring network.
       It is recommended  that the statistical performance  measures be compared with the
       following ranges:
              •   Unpaired highest 8-hour prediction accuracy: ±30-35 percent
              •   Average absolute error in 8-hour peak prediction accuracy for paired (time
                  and space) values > 5.0 ppm:  25-30 percent
              •   Average  absolute  error   in  the predicted  time of  the  8-hour   peak
                  concentration, paired by stations > 5.0 ppm:  2 hours
                                        50

-------
If all of these statistical measures are within  the  ranges  shown, and  the  graphical
performance procedures also are interpreted to yield acceptable results, then the model
is judged to be performing acceptably.

If any of the  statistical measures are worse than the above ranges,  or the  graphical
procedures are interpreted to  yield unacceptable performance,  users should consider
choosing  an  alternative highly ranked  meteorological  episode  for the attainment
demonstration. Performance evaluations should be conducted on a prospective alternative
episode to determine whether it yields improved model performance.
                                51

-------
                                    CHAPTER 6
                        ATTAINMENT DEMONSTRATION

      The primary purpose for conducting area-wide and roadway intersection modeling is to
demonstrate the effectiveness of control strategies in attaining the National Ambient Air Quality
Standard (NAAQS) for CO. Such demonstration of effectiveness consists of four parts:

      1.     Development  of attainment-year base case emission inventories;

      2.     Development  of future-year emission control strategies;

      3.     Performing attainment year model simulations to assess control strategies;

      4.     Use of modeling results in the attainment demonstration.

6.1 Developing Attainment Year Base Case Emission Inventories

      Base year inventories and initial and boundary concentrations must be projected to the
future attainment year. The future modeling year is a function of the attainment dates required
in the CAAA of 1990. Projections of base year inventories reflect the net effect of existing
required controls and growth projections for all source types.

The methodology for creating future year emission inventories is contained in Procedures for
the Preparation of Emission Projections (EPA, 1991f), which covers development of emission
projections for stationary and mobile sources.  When transportation model outputs are used to
develop  the mobile source portion of the inventory through the use of a mobile source emission
                                         53

-------
model (e.g., TRFCONV,  Causley, 1992), questions of which base inventory to use  for
developing vehicle  emissions  growth estimates must be resolved.   Appendix C provides a
technical discussion of procedures for developing vehicle emissions growth estimates.

       Projection to future year boundary and initial concentrations is usually accomplished by
applying factors that are the ratio of the future year emission totals to the base year totals.  An
exception to this occurs when low values have been specified for all boundary conditions.  These
conditions would not  be expected to become even cleaner, so they  would remain the same.
Before applying the  factors, a background value that is assumed to be unaffected by the emission
controls being considered is  subtracted.  Then the factor is applied to the remaining values and
the background is added back  on.  If the concentration is already at or below the background
value, it is  left unchanged.    The background  values  used  are  normally  the  mid-level
concentrations of 0.2  ppm.  In a few cases, there might be some question  about whether the
boundary conditions should be  scaled.  For example, if the boundary concentrations are affected
by an upwind urban area, will control measures similar to those being applied in the current
simulation be in effect in the  upwind area?  If so, should the current simulation reflect  the
change due to local emission controls or changes due  to all controls?  If upwind controls will
be in effect, boundary concentrations should be reduced to reflect control changes; simulation
results should reflect all changes, not only local ones.

       Recommendations
       It is recommended that the EPA guidance document entitled  Procedures for Preparing
       Emissions Projections  EPA (1991f) be  consulted  for  developing  attainment-year
       inventories.  The guidance document provides procedures for projecting point-source,
       area-source,  mobile-source, and addresses projections of spatial and  temporal, changes
       between the  base year inventory  and the attainment-year inventory.

6.2   Developing Future Year Emission Control Stratc

       Numerous future year emission control strategies can be developed and simulated using
the area-wide model and the roadway intersection model. Eventually,  a modeling analysis must
                                           54

-------
be submitted for approval for a SIP demonstration. The effectiveness of a given set of control
measures in reducing CO is an important factor in selecting the final strategy.

       Prior studies have used a number of control measures to ascertain an effective control
strategy for attainment.  Because of the low reactivity of CO in the wintertime, there is a near
linear relationship between domain-wide predicted changes in CO emissions and the domain-wide
predicted changes in CO concentrations for future year simulations.  However, depending upon
the control strategy employed, highly nonlinear changes may occur within each model grid cell.
For example, control strategies which  employ the use of such transportation control measures
(TCM's) as a parking management program (PMP) will reduce CO emissions locally and within
and near a model grid cell, but will have much less effect elsewhere in the  modeling  domain.

       Prior studies have typically used  a progression of control strategy scenarios in the
modeling to ascertain an effective strategy  for attainment.  A suggested logical progression is
the following:

       1.     Simulate the CAAA and other mandated control measures for the attainment year
             to determine if these measures are sufficient to demonstrate attainment.

       2.     If mandated controls  are  insufficient  to  demonstrate attainment, then  the
             approximate emission-reduction targets may  be estimated by the use of linear
             rollback.

       3.     Once an approximate target range is ascertained in steps  1 and 2, simulate control
             strategies that reflect source-specific or  source-category-specific control measures
             that realize the approximate emission reductions identified as sufficient to reduce
             maximum 8-hour CO concentration to  9.0 ppm or less.

       4.     Adjust the strategy chosen in step 3 until it is sufficient to demonstrate attainment
             of the CO NAAQS, as described in Section 6.4.
                                          55

-------
       Recommendations
       Find the emission reduction target needed to establish attainment by using linear rollback
       of emissions.  Using this as an  estimate,  the level of emission  reductions needed for
       control strategies can be assessed.  Adjustments to the strategy should continue until it
       is sufficient to demonstrate attainment. An outline of procedures for deriving control
       strategies for evaluation in the attainment demonstration must be  specified in the model
       protocol.
6.3  Performing Attainment-Year Simulations to Assess Various Control Strategies

      Many graphical display and numerical procedures are available for illustrating the effects
of alternative emission control strategies on predicted concentrations of carbon monoxide. For
example, the emission levels in the control strategies are often compared with the attainment-
year base emissions.  Also of interest are comparisons with the inventory derived for purposes
of model performance evaluations and corresponding base-case UAM results.  Difference maps
are extremely useful for illustrating changes in the daily 8-hour maximum carbon monoxide
predictions throughout the modeling domain.

      Recommendations
      The primary focus of the carbon monoxide attainment demonstration is on the maximum
      8-hour concentration predicted at each location in the modeling domain.  However, in
      some cases the scope of the attainment demonstration should be broader to assess the
      effects on the subdomain and temporal impacts.

6.4  Using Modeling Results in the Attainment Demonstration

      To demonstrate attainment of the carbon monoxide NAAQS, the combined results from
the areawide and roadway intersection modeling should show no predicted 8-hour maximum
carbon monoxide concentrations greater than 9.0 ppm anywhere in the modeling domain for the
episode  modeled.  Procedures for combining  the area wide and roadway intersection modeling
are given in EPA (1992c). Alternative methods for demonstrating attainment must be approved
by the appropriate EPA regional office on a case-by-case basis.
                                          56

-------
       The attainment test described in the preceding paragraph is consistent with the flexibility
allowed in the choice of episode day (see Section 3.2) and reflects concerns over the difficulty
of accurately estimating emissions inputs to the model.

       Recommendations
       To demonstrate attainment of the carbon monoxide NAAQS, the combined results of
       areawide and roadway intersection modeling should show no predicted 8-hour maximum
       carbon monoxide concentrations greater than 9.0 ppm anywhere in the modeling domain
       for the episode modeled.
       States may opt to conduct more comprehensive statistical testing of the modeling results
       for the attainment demonstration. Any alternative methods for attainment demonstration
       must be  approved by the appropriate EPA regional office on a case-by-case basis.  Any
       optional  methods should be  agreed upon  during the development of  the modeling
       protocol.

6.5    Exceptions to Guidance Document

       It is not  possible in a general guidance document like this to anticipate all contingencies
associated with  developing an attainment demonstration study.  The modeling policy oversight
and technical committees responsible for a specific modeling study may propose an alternative
modeling approach provided that (1) the modeling protocol requires consensus on the proposed
alternative  approach within the Technical Committee, and (2) justification for the proposed
approach is documented.  Application of any  alternative grid modeling approach  must first
receive concurrence in  writing from the responsible EPA regional office(s).
                                         57

-------

-------
                                     References
Anderson, G. E., S. R. Hayes, M. J. Hillyer, J. P. Kfflus, and P. V. Mundkur.  1977.  "Air
       Quality in the Denver Metropolitan Region 1974-2000." Systems Applications
       International, San Rafael, CA (EPA-908-1-77-022).

Causley, M. C., and L. L. Duvall.  1992.  "TRFCONV User's Guide."  Systems
       Applications International, San Rafael, CA (SYSAPP-92/021).

Causley, M. C., J.  L. Haney,  R. G. Ireson, J. G. Heiken, H. Tunggal, A. B.
       Hudischewskyj, and S. B. Shepard.  1991.  "Carbon Monoxide Air Quality
       Modeling of the Phoenix Metropolitan Area in Support of the Federal
       Implementation Plan, Volume I:  Main Report." Systems Applications
       International, San Rafael, CA (SYSAPP-91/105a).

Comsis Corporation.  1991. MINUTP Technical User  Manual. Comsis Corporation, Silver
       Spring, MD

Haney, J. L., and R. G. Ireson. 1988.  "Application of the Urban Airshed Model for
       Carbon Monoxide (CO) in Phoenix, Arizona."  Systems Applications International,
       San Rafael, CA (SYSAPP-88/083).  Presented at the 81st Annual Meeting of the Air
       Pollution Control Association,  Dallas,  TX, June 20-24, 1988.

Ireson, R. G., J.  L. Fieber, and M. C. Causley.  1991. "Generating Detailed Emissions
       Forecasts Using Regional Transportation Models: Current Capabilities and Issues."
       Systems Applications, Int., San Rafael, California. Paper presented at the American
       Society of Civil Engineer's Conference on "Transportation Planning and Air
       Quality", Santa Barbara, CA, July 28-31,  1991.

Killus, J. P., G. Z. Whitten, and R. G. Johnson.  1982.  "Modeling of Simulated
       Photochemical Smog with Kinetic Mechanisms." Systems  Applications International,
       San Rafael, CA (Publication no.  82172).

Laxton, W. G.  1991a.  EPA Memorandum.   "Guidance for Determining Significant
       Stationary Sources of Carbon Monoxide."  U.S. Environmental Protection Agency,
       Research Triangle Park, NC.
                                       59

-------
Laxton, W. G.  1991b.  EPA Memorandum.  "Issues Associated with the 1990 Base Year
       Emission Inventory and Modeling."  U.S. Environmental Protection Agency,
       Research Triangle Park, NC.

Morris, R. E., T. C. Myers, and E. L. Carr.  1990.  Urban Airshed model Study of Five
       Cities: Evaluation of Base Case Model Performance for the Cities of St. Louis and
       Philadelphia Using Rich and Sparse Meteorological Inputs. EPA-450/4-90-006C,
       U.S. Environmental Protection Agency.

Rogers, F.  1986.  "Airshed Modeling of Denver for Carbon Monoxide:  A Comprehensive
       Overview." Colorado Department of Health, Denver, CO.

Schere, K. L., and K. L. Demerjian.   1977.  Calculation of Selected Photolytic Rate
       Constants over a Diurnal Range. U.S. Environmental Protection Agency (EPA-   •
       600/4-77-015).

Siler, W., and Fishman.  1981. Distribution  of ozone and carbon monoxide in the free
       atmosphere. JGR. 86(C8):7255-7265.

Tesche, T. W., P. Georgopoulos, F. L. Lurman,  and P. M. Roth. 1990.  Improvement of
       Procedures for Evaluating Photochemical Models.   Draft final report.  California Air
       Resources Board, Sacramento,  CA.

Urban  Analysis Group.  1990.  TRANPLAN Version 7.0  User Manual.  Urban Analysis
       Group, Danville, CA.

U.S. Environmental Protection Agency.  1981a.  Procedures for Emission Inventory
       Preparation. Volume I:  Emission Inventory Fundamentals. EPA-450/4-81-
       026a, Office of Air Quality Planning and Standards, Research Triangle Park,
       NC.

U.S. Environmental Protection Agency.  1981b.  Procedures for Emission Inventory
       Preparation. Volume II: Point Sources.  EPA-450/4-81-026b, Office of Air
       Quality Planning and Standards, Research Triangle Park, NC.

U.S. Environmental Protection Agency.  1981c.  Procedures for Emission Inventory
       Preparation. Volume m: Area Sources.  EPA-450/4-81-026c, Office of Air
       Quality Planning and Standards, Research Triangle Park, NC.

U.S. Environmental Protection Agency.  1981d.  Procedures for Emission Inventory
       Preparation. Volume IV: Mobile Sources (Revised). EPA-450/4-81-026d,
       Office of Air Quality Planning and Standards, Research Triangle Park, NC.
                                        60

-------
U.S. Environmental Protection Agency.  1981e. Procedures for Emission Inventory
       Preparation. Volume V:  Bibliography. EPA-450/4-81-026e, Office of Air
       Quality Planning and Standards, Research Triangle Park, NC.

U.S. Environmental Protection Agency.  1986. Guideline on Air Quality Models (Revised).
       Research Triangle Park, NC (EPA-450/2-78-027R).

U.S. Environmental Protection Agency.  1987, Qn-Site Meteorological Program Guidance
       for Regulatory Modeling Applications.  EPA-450/4-87-013, Office of Air Quality
       Planning and Standards, Research Triangle Park, NC.

U.S. Environmental Protection Agency.  1988. Procedures for the Preparation of
       Emission Inventories for Precursors of Ozone. Volume I.  EPA-450/4-88-
       021, Office of Air Quality Planning and Standards, Research Triangle Park,
       NC.

U.S. Environmental Protection Agency.  1990. User's Guide for the Urban Airshed Model.
       Volumes I-V.  EPA-450/4-90-007 (NTIS No:  PB91-131243).

U.S. Environmental Protection Agency.  1991a. CAA Section 187 VMT Projection
       and Tracking Guidance.  Ann Arbor, MI.

U.S. Environmental Protection Agency.  1991b. Emission Inventory Requirements
       for Carbon Monoxide State Implementation Plans. EPA-450/4-91-011,
       Office of Air Quality Planning and Standards, Research Triangle Park, NC.

U.S. Environmental Protection Agency.  1991c. Procedures for the Preparation of
       Emission Inventories for Carbon Monoxide and Precursors of Ozone.
       Volume n:  Emission Inventory Requirements for Photochemical Air Quality
       Simulation Models.  EPA-450/4-91-014, Office of Air Quality Planning and
       Standards, Research Triangle Park, NC.

U.S.. Environmental Protection Agency.  1991d. User's Guide to MOBILE4.1.  Ann
       Arbor, MI (July 1991).

U.S. Environmental Protection Agency.  1991e. Guideline for Regulatory Application of the
       Urban Airshed Model. Office of Air Quality Planning and Standards, Research
       Triangle Park,  NC.

U.S. Environmental Protection Agency.  1991f. Procedures for Preparing Emissions
       Projections.  U.S. Environmental Protection Agency, Research Triangle Park, NC
       (July  1991).
                                        61

-------
U.S. Environmental Protection Agency.  1992a.  Guideline for Modeling Carbon Monoxide
       from Roadway Intersections.  Research Triangle Park, NC (to be issued in August
       1992).

U.S. Environmental Protection Agency.  19925.  User's Guide for the Urban Airshed
       Model.  Volume IV:  User's Manual for the Emissions Preprocessor System 2.0.
       EPA-450/4-90-007D(R). Office of Air Quality Planning and Standards, Research
       Triangle Park, NC.
                                         62

-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO. 2.
EPA-450/4-92-011a
4. TITLE AND SUBTITLE
Guideline for Regulatory Application of the Urban
Airshed Model for Areawide Carbon Monoxide
Volute Tt TeH-m-iral Report
7. AUTHOR(S)
Edward L. Carr, Julie L. Fieber and Robert C. Kessler
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Systems Applications International
San Rafael, California 94903
12. SPONSORING AGENCY NAME AND ADDRESS
Office of Air Quality Planning and Standards
U.S. Environmental Protection Agency
Research Triangle Park, N.C. 27711
3. RECIPIENT'S ACCESSION NO.
5. REPORT DATE
June 1992
6. PERFORMING ORGANIZATION CODE
8. PERFORMING ORGANIZATION REPORT N
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
13. TYPE OF REPORT AND PERIOD COVERE
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
  State implementation Plan  (SIP) revisions demonstrating attainment of'the
  Carbon Monoxide (CO) National Ambient Air Quality Standards (NAAQS)  are required
  under the Clean Air Act Amendments of 1990.  Urban  areawide modeling and
  intersection modeling are recommended to address attainment of the CO NAAQS.  The
  Urban Airshed Model  (UAM) has been identified as an effective urban areawide
  model for evaluating emission control requirements  needed to  attain the CO NAAQS.
  The purpose of this document is to provide guidance in the procedures used to apply
  the UAM for CO SIP attainment demonstrations.
17. KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS
Atmospheric Dispersion Models
Carbon Monoxide (CO)
State Implementation Plan (SIP)
Clean Air Act Amendments
Urban Areawide Modeling
18. DISTRIBUTION STATEMENT
b. IDENTIFIERS/OPEN ENDED TERMS

19. SECURITY CLASS (This Report!
20. SECURITY CLASS (This page)
c. COSATl Field/Group

21. NO. OF PAGES
65
22. PRICE
EPA Form 2220-1 (R«v. 4-77)   PREVIOUS ECITION is OBSOLETE

-------