EPA-450/3-78-008
January 1978
                         EVALUATION
         OF EMISSION INVENTORY
                   METHODOLOGIES
                         f
         FOR THE RAPS PROGRAM

     .S. ENVIRONMENTAL PROTECTION AGENCY
         Office of Air and Waste Management
      Office of Air Quality Planning and Standards
     Research Triangle Park, North Carolina 27711

-------
                                 EPA-450/3-78-008
          EVALUATION
OF EMISSION INVENTORY
      METHODOLOGIES
FOR  THE RAPS  PROGRAM
                    by

         Ronald E. Ruff and Patricia B. Simmon

                SRI International
              333 Ravenswood Avenue
            Menlo Park, California 94025



             Contract No. 68-02-2047



          EPA Project Officer: Charles Masser



                 Prepared for

       U.S. ENVIRONMENTAL PROTECTION AGENCY
          Office of Air and Waste Management
        Office of Air Quality Planning and Standards
        Research Triangle Park, North Carolina 27711

                 January 1978

-------
This report is  issued  by the  Environmental Protection  Agency to report technical
data of  interest  to  a limited  number of  readers.   Copies  are  available  free of  charge
to  Federal employees,  current  contractors and  grantees, and  nonprofit  organizations -
in limited  quantities —  from  the Library Services  Office  (MD-35),  U.S.  Environmental
Protection  Agency,  Research Triangle  Park,  North  Carolina 27711;  or,  for  a fee,  from
the National Technical  Information  Service, 5285  Port Royal  Road,  Springfield,
Virginia  22161.
This  report was furnished  to the Environmental  Protection  Agency  by SRI  International,
333  Ravenswood  Avenue,  Menlo  Park, California  94025.   The contents of  this report
are reproduced  herein  as received from SRI  International.   The  opinions, findings, and
conclusions expressed are those of  the  author  and  not necessarily  those  of the  Environ-
mental  Protection  Agency.   Mention  of  company or  product  names is not  to be  con-
sidered  as an endorsement  by the  Environmental  Protection Agency.
                         Publication No.  EPA-450/3-78-008

-------
                                     TECHNICAL REPORT DATA
                              (Please read Instructions on tin- reverse before completing)
 1. REPORT NO.
             EPA-450/3-78-008
                               2.
 4. TITLE AND SUBTITLE
  Evaluation of Emission Inventory  Methodologies
  for the RAPS Program
                                                               3. RECIPIfcNT'S ACCESSION NO.
               6. REPORT DATE
                 January 1978
               6. PERFORMING ORGANIZATION CODE
 7. AUTHOR(S)
  Ronald E.  Ruff and
  Patricia B.  Simmon
              8. PERFORMING ORGANIZATION REPORT NO
                 SRI-4331
 9. PERFORMING ORGANIZATION NAME AND ADDRESS
                                                               10. PROGRAM ELEMENT NO.
  SRI International
  333 Ravenswood Ave.
  Menlo Park,  California   94025
               11. CONTRACT/GRANT NO.
                68-02-2047
 12. SPONSORING AGENCY NAME AND ADDRESS
  EPA, Office of Air Quality Planning  and  Standards
  Research Triangle Park,  NC  27711
               13. TYPE OF REPORT AND PERIOD COVERED
                Final
               14. SPONSORING AGENCY CODE
                                200/04
 15. SUPPLEMENTARY NOTES
 16. ABSTRACT
        The general  objective  of the work  described  here is the  evaluation  and
   quantification  of the methodology being developed  and used for  the Regional  Air
   Pollution Study (RAPS) emissions inventory.   Improved emission  methodologies are
   one of the RAPS objectives.   However, they are essential to the realization of one
   of  the other principal objectives—namely, the evaluation of  mathematical  air
   quality simulation models.   The output  of  any such  simulation model  is only as good
   as  the input emissions data  supplied to it.  The thrust of this  work  is to  evaluate
   the  individual  emissions models and relate them to  their application to air quality
  models.
 7.
                                 KEY WORDS AND DOCUMENT ANALYSIS
                   DESCRIPTORS
b.lDENTIFIERS/OPEN ENDED TERMS  |c.  COSATI Field/Croup
  Emission  Methodologies
  Regional  Air Pollution  Study
  Air Pollutants
  Emissions
 8. DISTRIBUTION STATEMENT
  Unlimited
                                                19. SECURITY CLASS (This Report)
                             21. NO. OF PAGES
                                 146
20. SECURITY CLASS (This page)
  Unclassified
22. PRICE
EPA Form 2220-1 (Rev. 4-77)   PREVIOUS EDITION is OBSOLETE

-------
                               CONTENTS

LIST OF ILLUSTRATIONS	    ix
LIST OF TABLES	    xi
ACKNOWLEDGMENT	xiii
   I  SUMMARY AND CONCLUSIONS	     1
      A.   Introduction	     1
      B.   Conclusions 	     2
  II  TECHNICAL BACKGROUND 	     3
      A.   Overview of RAPS Emission Inventory 	     3
      B.   Identification of Important Parameters	     3
      C.   Quantification of Inventory Errors	     4
           1.   Previous Studies  	     4
           2.   Discussion	     5
 III  EMISSION INVENTORY EVALUATION PROCEDURE	     7
      A.   General Sensitivity Analyses Procedure	     7
      B.   Application to the RAPS Inventory	     9
  IV  EVALUATION CRITERIA FOR THE RAPS EMISSION METHODOLOGIES.  .  .    11
      A.   Overview	    11
      B.   Evaluation Criteria 	    14
   V  EVALUATION OF STATIONARY POINT SOURCE EMISSIONS	    17
      A.   Sulfur Dioxide	    19
           1.   Background	    19
           2.   Evaluation	    20
           3.   Large Point Sources	    20
           4.   Small Point Sources	    22
      B.   Nitrogen Oxides	    22
      C.   Carbon Monoxide	    23
      D.   Hydrocarbons	    23
      E. -  Particulates	    24

-------
 VI  EVALUATION OF STATIONARY  AREA  SOURCES EMISSIONS	   27

     A.    Estimation of Spatial Resolution	   29
          1.    Residential  Sources	   29
          2.    Commercial Sources  	   30
          3.    General Comments  	   30

     B.    Estimation of Temporal Resolution  	   30

          1.    Fuel Usage	   30
          2.    Evaporative Hydrocarbon Losses  	   31
          3.    Solid Waste Disposal-Structural Fires	   33

     C.    Discussion of Area Sources  by Pollutant	   33

          1.    Sulfur Dioxide	   33
          2.    Nitrogen Oxides	   34
        .  3.    Carbon Monoxide	   34
          4.    Hydrocarbons	   34
          5.    Particulates	   35
     D.    Summary of Quality of  Stationary Area  Source
          Methodology	   35

VII  EVALUATION OF HIGHWAY SOURCE  EMISSIONS  	   37

     A.    Evaluation of Highway  Source Inventory	   37

     B.    Highway Emission Models  	   38

     C.    FTP Emission Model	   39
     D.    Modal Emission Model	   42

     E.    Emission Model Comparison 	   43

          1.    Model Input Data	   43
          2.    Method of Determination of Composite
               Emission Models	   49
          3.    Results	   49
     F.    FTP Model Sensitivity  Analysis	   59
          1.    Sensitivity to Error in a Single  Input
               Parameter	   60
          2.    Sensitivity to Error in Two Input
               Parameters	   64
          3.    Conclusions	   72

     G.    Modal Model Sensitivity  Analysis	   75

          1.    Sensitivity to Error in a Single  Input
               Parameter	   75
          2.    Sensitivity to Error in Two Input
               Parameters	   77
          3.    Conclusions	   81
                                 vi

-------
VIII  EVALUATION OF OTHER TYPES OF SOURCES  OF EMISSIONS	    83

      A.   Fugitive Dust	    83
      B.   Aircraft	    84
      C.   Off-Highway Mobile Sources	    85
           1.   Lawn and Garden Equipment	    86
           2.   Construction Equipment 	    86
           3.   Industrial Equipment 	    86
           4.   Farm Equipment	    87
           5.   Outboard Motorboats	    87
           6.   Summary	    88

      D.   River Vessels	    88

      F.   Separation of Hydrocarbon Emissions into Classes.  ...    89

  IX  COST BENEFIT ANALYSIS	    91

   X  EFFECTS OF EMISSION PARAMETER INACCURACIES  ON AIR
      QUALITY MODEL PREDICTIONS	    95

      A.   Quantification of Errors in the  Meteorological
           Input Parameters	    95
      B.   Case Study for a Single Source	    98

           1.   Problem Formulation	    98
           2.   Results	   100

  XI  EFFECTS OF RAPS EMISSION INVENTORY ERRORS	113

      A.   Point Source Dominated Inventory (802)	113
           1.   Discussion	113
           2.   Summary	121
      B.   Area Source Dominated Inventory  (CO)	123

           1.   Discussion	123
           2.   Sensitivity	123
           3.   Adequacy of the CO Inventory	124
      C.   Other Primary Pollutants	125
           1.   Total Hydrocarbons 	   125
           2.   Nitrogen Oxides	125
           3.   Particulates	125
      D.   Photochemical Model Implications	126

      E.   Other Inventory Factors 	   127

           1.   Source Heights	127
           2.   Area Grid Size	128

REFERENCES	129
                                  vii

-------
                             ILLUSTRATIONS

 L   Assumed Distribution of Error  	    14
 2   RAPS Grid System	    28
 3   Weekend and Weekday Diurnal  Patterns  for Motor Vehicles  ...    32
 4   Plot of CO Emissions Computed  with  FTP Methodology
     and with Modal Model for Cases with Emissions Less
     Than 900 g/veh-mi	    50
 5   Plot of HC Emissions Computed  with  FTP Methodology
     and with Modal Model for Cases with Emissions Less
     Than 60 g/veh-mi	    51
 6   Plot of NOX Emissions Computed with FTP Methodology
     and with Modal Model for Cases with Emissions Less
     Than 15 g/veh-mi	    52
 7   Plot of CO Emissions Computed  with  FTP Methodology
     and with Modal Model for Cases with Emissions Less
     Than 300 g/veh-mi	    53
 8   Downwind Concentrations from Union  Electric  Company
     Souix Plant at Various Stabilities	101
 9   Sensitivity of Concentrations  to Flow Parameter
     Variation (Souix Plant) 	  105
10   Sensitivity of Concentrations  to Diffusion Flow
     Coefficient Variations (Souix  Plant)	108
                                  ix

-------
                                TABLES
 1   Estimates of RAPS  Source  Emissions	    12

 2   PEDCO Average Precision Data	    14
 3   Values of 9 for Selected  Inventory Errors and
     Confidence Levels  	    18

 4   Maximum Allowable  Error o^ for  Point Sources of
     Various Size	    19

 5   Emission Model Input  Data 	    44

 6   Emission Model Input  Data Used  for Model
     Comparison Runs	    45

 7   Roadway Descriptor Classes	    46

 8   Fraction of Annual Travel by Vehicle Age	    48

 9   Model Comparison Statistics  for CO	    54

10   Model Comparison Statistics  for HC	    55

11   Model Comparison Statistics  for NOV	    56
                                      X
12   Results of Tests for  Significance of Differences
     Between Correlation Coefficients of Various Subsets
     of the Data Sample	    58

13   Values of Input Parameters Assumed in FTP Model
     Single Parameter Sensitivity Analyses  	    61

14   Average Fractional Error  in FTP Emission Factors
     Resulting from Error  in a Single Input Parameter	    62

15   Values of Input Parameters Assumed in FTP Model
     Two Parameter Sensitivity Analysis	    66

16   Average Fractional Error  in FTP Emission Factors
     Resulting from Error  in Both Speed and Volume	    67

17   Average Fractional Error  in FTP Emission Factors
     Resulting from Error  in Both Cold Starts and Volume	    69

18   Average Fractional Error  in FTP Emission Factors
     Resulting from Error  in Both Speed and Percent Cold  Starts.  .    71

19   Input Parameter Error Values that Cause FTP Emission
     Factor Error to Exceed 20%	    73

20   Values of Input Parameters Assumed in Modal Model
     Sensitivity Analysis	    77

21   Average Fractional Error  in Modal Emission Factors
     Resulting from Error  in a Single Input Parameter	    78

                                  xi

-------
22   Average Fractional Error in Modal Emission Factors
     Resulting from Error in Both Volume and  Cold  Start	    79
23   Input Parameter Error Values that Cause  Modal
     Emission Factor Error to Exceed 20% 	    82
24   MRI Estimates of Fugitive Dust Inventory Accuracy  	    84
25   Inventory Summary:  Totals, Accuracy,  and Cost	    92
26   Characteristics of Meteorological Measurements  at
     25 Site Locations in RAPS Areas	    99
27   Flow Parameters for Three Cases	104
28   Typical Errors in Significant Downwind Concentrations  ....   Ill
29   Ground-Level SC^—CDM Reference Case	115
30   Sensitivity of S02 Estimates to Wind Direction  Errors  ....   116
31   Erroneous Wind Speed Distribution 	   116
32   Sensitivity of S02 Estimates to Wind Speed Errors	117
33   Erroneous Atmospheric Stability Distribution	118
34   Sensitivity of SC^ Estimates to Typical  Atmospheric
     Stability Estimation Errors 	   118
35   Sensitivity of S02 Estimates to Worst  Cast Atmospheric
     Stability Estimation Errors 	   119
36   Sensitivity of S02 Estimates to Typical  Random  Errors
     in Emission Rate	120
37   Sensitivity of S02 Estimates to Worst  Case Random  Errors
     in Emission Rate	121
38   Sensitivity of S02 Estimates to Typical  Random Errors
     in Stack Exit Gas Parameters	122
39   Summary 1:  Significant Ground-Level Concentration
     Average Error Caused by Typical Input  Parameter Error  ....   122
40   Summary 2:  Significant Ground-Level Concentration
     Average Error Caused by Worst-Case Input Parameter Error.  .  .   122
41   Area Source Sensitivity Comparison	124
                                  xii

-------
                            ACKNOWLEDGMENT

     The authors wish to express their gratitude to  the  people  who  pro-
vided special assistance during the preparation of  this  report.   The
guidance of the EPA Project Officer, Mr.  Charles Masser  was  vital in
ensuring that our effort was focused toward  EPA objectives.   The  cooper-
ation of Dr. Fred E. Littman and Mr. John Piere of Rockwell  International
was also invaluable, particularly in the  timely transfer of  information
on their efforts.
     Our colleagues at SRI International  were also  instrumental in  this
effort.  Mr. Ronald T. H.  Colis reviewed  the manuscript  in detail.  Dr.
Chandrakant M. Bhumralkar, Dr.  J.  Raul Martinez, and Mr.  Hisao  Shigeishi
contributed extensively to sections of this  report.
                                 xiii

-------
                      I  SUMMARY AND CONCLUSIONS

A.   Introduction
     The overall objective of the work described in this report is the
evaluation and quantification of the methodology being developed and
used for the Regional Air Pollution Study (RAPS) emissions inventory.
Improved emission methodologies are one of the RAPS objectives.  How-
ever, they are essential to the realization of one of the other RAPS
principal objectives:  the evaluation of mathematical air quality simu-
lation models.  The output of any such simulation model is only as good
as the input emissions data supplied to it.   The thrust of this work is
to evaluate the individual emissions models and relate them to their
application to air quality models.   This project encompassed four major
elements:
     •  An in-depth review of emission models and procedures
        used in the total context of the RAPS emissions
        inventory,  to ensure that together they provide a
        well-balanced approach to the overall objective.
     •  An evaluation of the effectiveness of the emission
        models to ensure an accuracy commensurate with the
        needs of the RAPS program.
     •  The identification of possible modifications to the
        emission models and procedures.
     •  Design of an experimental verification program that
        could be conducted in St. Louis.   This program would
        validate the various emission models to fulfill an
        objective of RAPS as well as for more general use.
     During the development of validation procedures for the emissions
modules,  benefit-cost ratios were determined.  The recommended valida-
tion program was developed under this contract and is reported
separately (Shelar et al.,  1976).

-------
B.    Conclusions
     Sections V through VIII contain an evaluation of the various emis-
sion inventory methodologies.  With the exception of the methodologies
for sulfur dioxide, enough data are not generally available to support
objective conclusions.  A number of recommendations are made concerning
possible improvements or validation of various emission estimates.  Areas
that need the most further effort include:
     •  Highway source inventory—validation is needed for both
        model input data and the emission methodology (see
        Section VII).
     •  Point source inventory—additional stack tests are
        warranted for nitrogen oxides  (see Section V.B).
     •  Fugitive dust — further work is needed to validate
        existing results and to separate particles into size
        categories (see Section VIII.A).
     Sections X and XI estimate the effect of emission inventory
errors on air quality model performance.  Analysis shows that the sulfur
dioxide inventory is the most accurate and most suitable for air quality
modeling purposes.  Further validation work is needed to support any
firm conclusions about hydrocarbons and carbon monoxide.  More research
on fugitive dust is required to support particulate modeling efforts.
     Although there are limitations in the accuracy of the RAPS inventory,
the inventory is a definite asset.  More significantly,  it is the most
comprehensive inventory available to support air quality modeling objec-
tives.   Significant improvements could be realized with nominal future
expenditures.  The data system is flexible enough to allow refinements
in input data such as emission factors, fuel consumption,  and land use.

-------
                       II  TECHNICAL BACKGROUND

A.   Overview oi RAPS Emission Inventory
     The RAPS emission inventory is the most comprehensive in existence.
It was assembled primarily to support the development and verification
of air quality models.  Generally,  the previous inventories  tell short
of providing the requisite accuracy and resolution (spatial and temporal),
For the RAPS inventory, the air quality modelers had an input to the
requirements prior to the collection activity (Littman et al., 1974).
     In response to air quality modeling requirements, EPA adopted a
program to acquire the most responsive data and develop the best avail-
able methodology within the budget constraints of the program.  This
report assesses the adequacy of the resulting product.

B.   Identification of Important Parameters
     The inventory is comprised of three general source types:  point,
line,  and area.   For point sources, emission parameters consist of:
     •  Emission rate
     •  Location (UTM coordinates)
     •  Stack parameters
        -  Height
           Diameter
           Gas exit velocity
           Gas exit temperature.
Line source parameters involve only emission rate and coordinates of
the endpoints of each line segment  (link).   Area sources are represented
by a grid,  one kilometer square,  or an even multiple thereof.  In
addition to emission rate, area source parameters are the grid size and
the coordinates given for the lower left (southwest) corner.   Nominal
emission release heights are also needed for both line and area sources.

-------
     Emissions, in this evaluation study, are considered for the primary
pollutants:  sulfur dioxide (SCO , carbon monoxide  (CO), total hydro-
carbons (THC), nitrogen oxides  (NO ), and particulates.  Some discussion
                                  X
addresses the breakdown of hydrocarbons and particulates (size
distributions).
     The temporal resolution of the emission data base is one hour.  The
RAPS data handling system accesses the raw input data  (fuel consumption,
automobile traffic data, and so forth), applies the appropriate emission
factors, and computes the emissions.  Improvements  in  input data and
emission factors (models) can readily be incorporated  into the data
handling software.   In some cases, for large point  sources, raw process
data was compiled hourly from January 1975 through March 1977.  In other
cases, hourly rates were estimated on the basis of available data.

C.   Quantification of Inventory  Errors
     1.   Previous Studies
          The effects of emission inventory errors on  the prediction of
ambient air pollution concentrations have been evaluated in previous
studies.  One of these studies  (Koch et al., 1971), had a significant
bearing on the RAPS emission inventory methodology.  Some of the more
relevant Koch study findings were:
          •  Averaging area source emission rates over areas larger
             than 1 square mile can lead to significant errors in
             estimated pollutant  concentrations.
          •  Treating area sources as emitting from the same height
             does not significantly affect estimated pollutant con-
             centrations.
The Koch study also addressed the sensitivity of certain meteorological
parameters.
          In another study (Hilst, 1970), a steady-state Gaussian model
was demonstrated to be relatively insensitive to random errors in the
specification of source strength.  Hilst concluded that,  among multiple
sources, random errors tend effectively to cancel.  Of course, it was
recognized that systematic emission errors lead to systematic errors
in predicted concentrations.
                                   4

-------
          The SRI study (Littman et al., 1974) addressed the accuracy
requirements for source coordinates.  Their sensitivity analysis showed
that large sources should be located within 10 meters of their true
coordinate.

     2.   Discussion
          While the studies just mentioned (Littman, Koch,  and Hilst)
provide valuable insight toward the specification of inventory para-
meters, they did not address the specific problem associated with RAPS.
Namely, is the RAPS inventory accurate enough such that it does not
significantly hinder efforts in the model development and evaluation
process?  To answer this question,  we must examine the other constraints
that limit the accuracy of prediction:  errors in the meteorological
input data, and errors in the model formulation itself.   Since model
error is subject to improvement over the years,  our approach has been
to quantify errors induced by emission inventory inaccuracies in relation
to those induced by inaccuracies in the meteorological input parameters.

-------
              Ill  EMISSION INVENTORY EVALUATION PROCEDURE

A.   General Sensitivity Analyses Procedure
     Common to the emission inventory evaluation process is the appli-
cation of a sensitivity analysis procedure to the various components of
the RAPS emissions and air quality models.  Sensitivity is formally
defined as the partial derivative of the output of a model to the input
parameter (s) in question.  In the case of complex models, it is more
appropriate to consider incremental changes in output resulting from
incremental changes in input because the determination of the analytical
expressions for partial derivatives becomes too cumbersome.
     One of the primary objectives of this work is to determine the
sensitivity of the air quality model output as a function of various
emission inputs in relation to meteorological inputs.  Quantifying the
potential sources of error (output) in this manner gives the air quality
modeler a better understanding of the limitations on model performance
and identifies those types of input data that most need improvement.
     The air quality model output,  x>  can ^e represented as some function
of emissions inputs,  Q, and meteorological inputs G as follows:
                               ,s'  [G]t,s  +
where x, x'  >  Q,  and G are vectors in time (t) and space (s).  The
    ,1.     GITi
term  represents an error in the formulation,  which is only a mathematical
approximation to the physics and chemistry of the atmosphere.  For our
purposes, it is assumed that the X'   term represents a random error not
 In the notation used here,  the prime will be used to designate errors
 caused by the model formulation; unprimed error terms will designate
 the total error.

-------
related to the Q and G arrays.  The sensitivity of any element in the X
array to any element in either the Q or G arrays is given by
                       oX
                                             8
                                            ) ° )
or                                                                   (2)

                       oXi _ of}[Q]t>s, [G]t>s}
                       FcT =       acT
                         J           J
However, Q's and G's are also only mathematical approximations or
measurements subject to error.  These are expressed as
and                                                                  (3)
The variations  in  the Q    and  G     terms are numbers  that can be used
                      xerr      err
to quantify air quality model errors by  substitution for  incremental
changes  in Q. and  G. in the  discretized  approximation  of  Equation  (2).
      The G   terms can be estimated from reports  of previous meteoro-
           err
logical  studies.   Parameters such  as wind and diffusion coefficients
are approximated  in some manner before they  are used in an air quality
model.   They are  derived from measured data  of questionable adequacy.
In the RAPS study, meteorological  data available for model input are
quite comprehensive.  However,  significant errors  will still arise.
      The Q   terms were estimated after conducting a  separate sensi-
           err
tivity analysis  for the relevant emission model.   The  emission variable
can be expressed  as:

                                                                     (4)
                                   8

-------
where g represents the emission model, and P represents  the  input  para-
meters; Q'   represents the model formulation error.
     The sensitivity analysis for the emissions model  is analogous  to
that for the air quality model.   Errors in the input parameters, P   ,
are used to quantify emissions errors.
B.   Application to the RAPS Inventory
     In Sections V through IX, Q    is estimated for all significant  source
types.   For some large point sources, source test statistics are used
to quantify Q   .   For other sources, we subjectively evaluate the term
on the basis of assumed accuracies.
     Estimation of the G    term is described in Section XI.  Then,  the
                        err                                       '
relative effects of G    and Q   ,  as they impact model pollutant con-
                     err      err
centration predictions, are presented.

-------
      IV  EVALUATION CRITERIA FOR THE RAPS EMISSION METHODOLOGIES

A.   Overview
     Adequate validation data are the key to the successful evaluation of
any emission model methodology.  In the presence of such data, statistical
tests can quantify the model performance.  Ideally, one would like to have
such data for each major source or, at the very least, each major source
type.  Furthermore, ideally, these test data should be comprehensive
enough to constitute a statistically valid sample.
     Unfortunately, validation data rarely exist in sufficient quantity.
Therefore, a truly objective methodology assessment is difficult.  Never-
theless, a "semiobjective" evaluation is possible for the RAPS inventory.
The level of objectivity varies according to the source type and size.
     In this report, various source types are evaluated relative to each
of the major pollutants--SO-, CO, NO , hydrocarbons (HC), and particulates.
                           ^_        X
For the latter two pollutants, reactive versus nonreactive HC and particu-
late size distributions are briefly considered also.  Approximate magni-
     &
tudes  by category for point and area source emissions are presented in
Table 1.  These estimates are used as a guide in the remainder of the
report.
     Emission factors are the basis for converting a source activity level
to pollutant-specific emission rates.  The activity level is described in
parameters that are readily measured (or estimated) for the type of source.
As an example, for power plants, fuel consumption rate is commonly used to
describe an activity level.  To calculate the emission rate for sulfur di-
oxide, one can multiply the activity rate by the fuel sulfur content and
the emission factor.
*
 Courtesy of J. Piere, Rockwell International,
                                   11

-------
                      Table  1
ESTIMATES OF RAPS SOURCE EMISSIONS (TONS/YEAR)--1975
        (Courtesy of Rockwell International)


Area Sources
River Vessels
Fugitive Dust
Unpaved roads
Agricultural tilling
Wind erosion
Construction
Aggregate storage
Unpaved airstrips
Paved roads
Highways
Area sources
Line sources
Railroads
Stationary Residential and Commercial
Residential fuel oil
Residential natural gas
Residential LPG
Residential coal
Commercial fuel oil
Commercial natural gas
Commercial LPG
Commercial coal
Evaporation from HC
Evaporation from HC-automatic tanks
Evaporation from HC-dry cleaning
Structural fires
Solid waste disposal
Off-Highway Mobile
Motorcycles
Lawn and garden equipment
Farm equipment
Construction equipment
Industrial equipment
Outboard motors
Stationary Industrial Sources
Airports
Total Area Sources
Point Sources
Fuel Combustion
Industrial Process
Solid Waste Disposal
Total Point Sources
Grand Total (all sources)
Particulate
(TSP)

196

458,605
60,244
524,583
138,156
937
470
64,464

704
6,278
878

547
432
386
2,079
895
218
0
1,608
0
0
0
306
127

3
32
390
964
734
0
133
65
1,264,434

11,165
31,884
319
43,368
1,307,802
GO
b°x

458

0
0
0
0
0
0
0

302
2,347
2,001

1,573
25
1
9,873
6,676
12
0
5,457
0
0
0
18
14

1
8
277
983
679
45
58
48
30,856

805,262
96,293
227
901,782
932,638
NO
NOX

3,353

0
0
0
0
0
0
0

5,856
67,185
11,960

655
3,453
1,623
312
3,000
2,612
0
529
0
0
0
115
24

2
124
3,287
12,652
8,765
46
149
555
126,257

275,687
13,404
182
289,273
415,530

HC

688

0
0
0
0
0
0
0

12,376
99,990
4,229

164
345
163
2,079
149
174
0
115
3,807
13,650
645
574
102

220
1,530
1,951
1,893
3,056
7,799
106
1,467
157,272

2,422
45,216
136
47,774
205,046

CO

1,319

0
0
0
0
0
0
0

120,081
1,033,770
4,360

273
863
406
9,355
199
435
0
414
0
0
Q
1,625
206

421
11,927
23,148
19,508
66,085
23,474
20
2,969
1,320,858

7,931
130,238
3,180
141,349
1,462,207
                        12

-------
     For point sources, emission factors have been classified into source
classification codes (SCC).   Each SCC represents a particular source type
and, consequently, has its own emission factor.  These factors are based
on pre-RAPS data and have been published in AP-42,* Supplement 5.  As part
of the RAPS program, new source test data were acquired for many of the
point sources in the St. Louis area.  For these tested sources, new emis-
sion factors were established when the test results did not verify the
old (Supplement 5) factors.   Consequently, the RAPS point source inventory
relies on a mixture of emission factors based on (1) stack test results
for the source in question,  or, when not available, (2) AP-42, Supple-
ment 5.
     Extensive evaluation of the AP-42 emission factors was undertaken
by PEDCO Environmental (Gibbs et al., 1974).   Results of this EPA-
commissioned study form one  of the foundations for the analysis presented
in this report.  More recently, Littman et al. (1977), at Rockwell Inter-
national, performed an evaluation for a few of the large point sources in
the RAPS inventory.  Verification studies for the mobile and area source
inventories, however, do not currently exist.
     Some summary statistics from the PEDCO study are presented in Table 2.
Average precision  is a measure of the variation among stack test results
according to pollutant.  For S09, the statistical interpretation is that
about 687o of the results fall within 17.770 of the mean or average: for CO,
about 687=, would fall within  32.170 of the mean.  Therefore, the average
precision can be thought of  as being a measure of the repeatability of
stack measurements.  This repeatability is inversely proportional to the
precision value.  This concept is further illustrated in later sections.
*
 EPA Publication AP-42 with Supplements, "Compilation of Air Pollutant
 Emission Factors."
 Ratio of the standard deviation to the mean of the test results--averaged
 over SCC codes by pollutant.

                                   13

-------
                                Table 2
                     PEDCO AVERAGE PRECISION DATA
Pollutant
Particulate
S00
z
NO
X
HC
CO
Average Precision of Test Data
0.196
0.177

0.134

0.203
0.321
B.    Evaluation Criteria
     The analysis errors in the inventory will be assumed to follow a
Gaussian distribution as shown in Figure 1.  The symbols in the figure
indicate the error parameters of interest.  The distribution of errors
about the true emission rate, T, is subject to bias (b) and variability
Precision, a , is a measure of the variability of the distribution.
                FIGURE 1    ASSUMED DISTRIBUTION OF ERROR
                                   14

-------
     For the ideal emission model,  b and a  are both zero.   Where possible
in the analyses to follow, limits are estimated for these parameters.   The
effect of such errors on the performance of air quality  models  is eval-
uated later in Sections X and XI.
                                  15

-------
            V  EVALUATION OF STATIONARY POINT SOURCE EMISSIONS

     The methodology for stationary point sources is based on the appli-
cation and verirication of standard EPA emission factors.  As described
by Littman (1974), the RAPS inventory is distinguished from previous
inventories by three factors:
     •  Accuracy
     •  Space resolution
     •  Time resolution.
As a quantitative measure of overall accuracy and probable error, consider
the Weighted Sensitivity Analysis Program (Ditto et al., 1973).   Although
this program does not supply any estimates on the absolute accuracy, it
does help evaluate the maximum permissible error of any part of  the in-
ventory, if provided with a maximum permissible error for the whole sys-
tem.  In its evaluation, the program keeps the inventory at an equivalent
level of accuracy and points out areas where accuracy has to be  improved
to provide a desired overall accuracy.  In addition, it also provides an
approach to establish confidence levels for the emission inventory.
     The first step of the method is based on the following linear model :
                                           -
                                    k=l
where
          Q = Total amount of pollutant emitted
      100 0 = Percentage error associated with Q
         Q,  = Amount of pollutant emitted by subclass k
     100 crk,= Percentage error associated with Q .

This linear model is postulated as an appropriate model to analyze  the
propagation of error through the emission inventory.   If each subclass
                                   17

-------
 contributes  to  the  error an amount  proportional to its relative physical
 contribution, it  can be shown  for each k that
                                                                    (6)
The analysis demonstrates that to obtain a predetermined level of pre-
cision  for a source  class, not all subclasses need to be measured with
the same  precision;  the greater the ratio of Q:Q  , the greater the al-
                                                K.
lowable value of a, .  Conversely, a,  approaches the value of 9 as the
ratio' approaches unity.  Equation (6) allows calculation of allowable
inventory error per  category, afc, after specification of allowable total
error,  9.  Confidence in the values for 9 and a,  is quantified by a sec-
ond step as described in the following paragraph.
     By using Chebyshev's inequality, the second step allows one to estab-
lish probabilistically the confidence level for the inventory;  i.e., what
is the probability that the actual overall error in emissions will not
exceed  9.  Ditto et  al. (1973) provide a complete treatment of the theory.
Using results from this reference, Table 3 shows the interrelationships
among overall emission inventory error, confidence level, and 9.   As an
interpretation, Table 3 indicates that, for the emission inventory to be
accurate within 10%  of the true value at a 95% confidence level,  9 must
be 2.24% (or less).
                                Table 3
               VALUES OF 9 FOR SELECTED INVENTORY ERRORS
                         AND CONFIDENCE LEVELS

Inventory Error

5
10
20
Confidence Level
(%)
90
1.58
3.16
6.32
95
1.12
2.24
4.47
99
0.5
1.0
2.0
                                  18

-------
     A 107<> overall inventory accuracy specification was deemed reasonable
by Rockwell International.  (Ideally, the specification should reflect
requirements of the air quality modeling community.)  They used values
from the 1973 NEDS inventory to calculate the allowable error for point
source classes of various sizes, which are given in Table 4.
                                 Table 4
       MAXIMUM ALLOWABLE  ERROR ak FOR POINT SOURCES OF VARIOUS SIZE^
        (Acceptance Interval 10%,  Confidence  Level 95%,  6 = 2.24%)

Pollutant
so2
CO
NO
X
HC
Particulate
Total
Point Source
Emissions, Q
(tons/yr) ^
1,187,296
1,684,794
310,993

78,474
323,952

Allowable Error a, for Point Source of
(100 ton/yr)
244%
290%
125%

63%
127%
(1,000 ton/yr)
77%
92%
40%

20%
40%
(10,000 ton/yr)
24%
29%
— —

--
--
  Reprinted from Littman et al.  (1974).
 t,
  Source:  NEDS December 1973, except for CO, which is given in NEDS at
  2,836,270 before correcting to the value shown in this table.
     The allowable error results served as the foundation for the point
source data collection strategy.  Sulfur dioxide (S0?)  was the pollutant
emphasized initially in the inventory because it occupied the position
of highest priority within the RAPS.   Therefore, S02 emissions will  be
discussed first.
A.   Sulfur Dioxide
     1.   Background
          The SO- point source inventory and its verification are de-
scribed in detail by Littman, Griscom, and Klein (1977).   For sources
emitting at least 1000 tons of SO  per year, hourly fuel  consumption
                                  19

-------
data were acquired.  For smaller point sources (less than 1000 ton/year),
hourly data were derived from annual data plus a detailed operating pat-
tern.  This approach is consistent with the results of the weighted sensi-
tivity analysis.

     2.   Evaluation
          Calculation of the SO- emission rate from the process data is
based on the following equation:

                         Q = W • EF

                         Q = W • EF = W • K • S                     (7)

where
          Q = The emission rate
          W = The fuel consumption rate
         EF = The emission factor (normally EF = K • S)
          S = The percentage of sulfur in the fuel
          K = An emission constant.
For air quality modeling purposes, it is assumed that a temporal resolu-
tion of one hour is sufficient.  For sources emitting at least 1000 ton/
year, W is acquired on an hourly basis.  These large sources are con-
sidered first.

     3.   Large Point Sources
          As a first approximation, the error is related to the product
of S and K.  S is measured directly, but not on an hourly basis.  More
typically it is measured weekly or monthly.  Emission factors have been
experimentally verified.  Based on the PEDCO report (Gibbs et al., 1974),
the precision of measuring the sulfur content is generally less than 0.1,
depending on the type of fuel.  The daily variability of the sulfur con-
tent can be at least that great.
          Assessment of the precision of emission factors can only be
semiquantitative.  While the PEDCO report does present precision and

                                  20

-------
accuracy (bias) calculations for a wide variety of source types, these
calculations are based on source test data that are subject to inaccura-
cies.  In recognition of this, EPA commissioned Rockwell International
to conduct emission factor verification studies in the St.  Louis area.
These have been reported by Littman, Griscom, and Klein (1977).
          The Rockwell International report concluded that some previous
stack tests were suspect, particularly those that relied on stack gas
velocity measurements such as used in EPA Method 2.  Upon reviewing these
measurements, Rockwell found that stack flow values were high by varying
but substantial amounts.  They further verified these conclusions using
fuel consumption rates.
          Based on their most recent mass flow data, Rockwell concluded
that AP-42 emission factors for SCK were reasonably accurate.  In a pri-
vate communication,  Littman estimated the general precision of the RAPS
S0? emission factors to be within 0.15 with no significant bias.  (This
roughly corresponds to a "typical error" of 157o.)  The precision estimate
is supported by the PEDCO analysis (see Table 2).
          The overall precision of the emissions model is the square root
of the sum of the square for the individual precisions based on:  (1)
emission factors^" and (2) the daily variability of the sulfur content
Mathematically,
                  a  = P V(0.15)2  4- (0.15)2  =  0.225                 (8)

where a  is defined in Figure 1.  For purposes of this evaluation, a  will
       e                                                            e
be approximated at 0.2 for future calculations.  Further it is assumed
that the bias, b, is negligible for these major SO  sources.  The Rockwell
studies tend to verify this assumption.
*
 Dr. F. E. Littman, Rockwell International, private communication, 1976,
 The precision of the sulfur content is inherent in the emission factor
 precision.

                                  21

-------
     4.   Small Point Sources
          Statements made in the previous subsection (3) concerning the
accuracy of emission factors and sulfur measurements still apply here.
However, for small point sources, hourly process data are not collected.
Instead, annual data are acquired and modified by a detailed operating
pattern.  Sufficient data are not on hand to validate the accuracy of
these operating patterns.  For the purposes of future analyses, vie will
assume a precision of 0.4 and negligible bias.

B.   Nitrogen Oxides
     In the PEDCO report, the precisions of NO  measurements (by SCO code)
                                              X
are, on the average, comparable  to those for SO^.  However, a substantial
bias appears because emission factors for combustion sources are too high.
The Rockwell verification study  corroborates the PEDCO results.  Their
experimentally obtained  factors  ranged from a low of 7. TL to 12% of the
applicable AP-42 factors.
     The RAPS data handling  system allows for the input of plant specific
emission factors.  Standard  AP-42 emission factors are being used at un-
tested sites.  It is reasonable  to assume that these sites will also be
represented by emission  factors  that are too high.  This will result in
the inventory being biased accordingly.
     The amount of inventory bias is a function of the number of sources
verified, as described above.  If we assume one-half of the emission fac-
tors are corrected, then it  would appear that the other half is biased
by 7.7 to 727», or, on the average, 4070.  The total inventory then will
be biased by one half that amount--20%.
     If we assume precisions comparable to those  for S02, a very tentative
estimate of NO  emission inventory errors follows:
     •  Large point sources:
                             b = +0.2 T.
                             a  = 0.2
                             e
                                   22

-------
     •  Small point sources:
                             b = +0.2 T
                            a  = 0.4
                             e
The impact of such errors on air quality predictions is discussed in Sec-
tions X and XI.  The relative importance of these point source data is
evident in Table 1, which shows the dominant effect of point sources ac-
counting for almost 70% of NO  emissions.
                             X

C.   Carbon Monoxide
     As shown in Table 1, point source CO  emissions account for only a
few percent of the total inventory.  As with NO , CO emissions cannot be
                                               X
verified through material balance calculations.  The Rockwell tests indi-
cate experimental emission factors much lower than those given in AP-42.
The PEDCO report does not contain sufficient data to enable any general
conclusions to be made.  Therefore, we will assume that the precision
and bias are comparable to those for NO .

D.   Hydrocarbons
     Point source emissions make up about  25% of the total HC inventory.
For HC, point sources consist of those emissions released through a stack
or vent, resulting from either fuel combustion or evaporation.
     Rockwell has been delegated the responsibility of measuring HC com-
ponents, both reactive and nonreactive. Primarily, the efforts were in
separating out methane and nonmethane components with a particular emphasis
on accurately measuring the nonmethane components.  Littman, Griscom, and
Seeger (1977) describe a chromatographic technique that is linear with
                                                            3
respect to carbon number and HC concentration up to 3.5 X 10  per carbon
number.  Griscom (1977) also estimated HC  emissions by category.   (The
categorizing of HC components in the inventory is an important input in
photochemical air quality simulation models.)  This topic is discussed
more thoroughly in Sections VIII.F and XI.C.  Our evaluation here is
limited to total hydrocarbons (THC).
                                  23

-------
     The Rockwell approach consists of modifying emission factors for
tested sources only.  Other sources are estimated using AP-42 factors.
The process data are collected hourly for major combustion sources.   For
evaporative sources, which account for about 4070 of the point source
inventory, data are presented annually.  Therefore, the hour-to-hour
variability will be another source of inaccuracy.
     All the above factors must be taken into account in estimating pre-
cision and bias.  For combustible sources, these parameters should be
about the same as for NO  and CO.  For evaporative sources, for which
annual fuel loss data are used, the long-term bias should be negligible,
but the precision should be high because no corrections are made for loss
as a function of time.  Our estimate of the composite of the above cases
is a bias of +0.1 T and a precision of 0.4.

E.   Particulates
     Based on the figures in Table 1, ground level TSP measurements should
be dominated by fugitive dust sources.  However, the dominance is primarily
caused by the mass of large inert particles (diameters above 2 pm).   Other
source types may be dominant emitters of the smaller (resperable) particles
or chemically active particles.  Hence, accurate particulate data from.
nonfugitive dust sources are essential.
     Previous inventories focused on TSP, but the RAPS inventory separates
particulate data by particle size categories.  In general,
where
                    E. = W • EF • (1 - C.) •  F. •  P                 (9)
          E  = Emission rate for particles in category i
           W = Mass flow rate
          EF = Emission factor
          F. = Fraction of particles in category i
          C  = Efficiency of control system for particles in
               category i
           P = Percentage of production subject to control.
                                   24

-------
Rockwell was commissioned to assemble and verify the point source particu-
late inventory (both by TSP and size category).   The work is described by
Littman, Griscom, and Wang (1977).   Some emission factors, by particle
size, were derived from the Rockwell stack test  data.  Others were based
on other factors experimentally derived by Midwest Research Institute
(Weast et al., 1974).
     The frequency categorization and control  device efficiency are two
additional sources of inaccuracy in determining  total emissions.   Suf-
ficient data do not exist to allow us to make  a  reasonable estimate on
the bias and precision of these parameters and,  hence,  the particulate
point source inventory as a whole.
                                  25

-------
          VI  EVALUATION OF STATIONARY AREA SOURCE EMISSIONS

     Unlike the point source inventory, where emission estimates can be
evaluated by testing a relatively small number of sources,  stationary
area sources can be numbered in the hundreds of thousands.  Broadly speak-
ing, there are three basic categories of stationary area sources:   resi-
dential, commercial, and industrial.   The emissions from any  one source
under varying conditions can be estimated within an assumed accuracy.
However, the complete characterization of every small source  is  a  monu-
mental task well beyond the scope (and budget) of the RAPS  survey.   There-
fore, the general approach has been to derive fuel consumption,  land use,
and other algorithms to estimate hourly emission rates per  unit  area.
     The minimum size for an area source is a 1 km by 1 km  square  grid
with larger grid sizes being even multiples.  Figure 2 illustrates  the
RAPS grid system.  In general, areas of heavy emissions and large  spatial
variability are represented by the smaller grids and the more rural areas
(with low emission density) are represented by larger grids.   The RAPS
grid system is consistent with requirements of most air quality  models.
     With the exception of evaporative hydrocarbon emissions, stationary
area source emissions are predominantly the result of demand  fuel  usage.
For commercial and residential sources, the emissions methodology  was
developed by Environmental Science and Engineering, Inc. (Holden,  1975);
the industrial sources were estimated by Rockwell International  (Littman
and Isam, 1977).  These industrial sources are insignificant  and,  for our
purposes, can be neglected.  Therefore, the focus of the section is the
commercial and residential area sources.
     In the methodology, spatial allocations to the RAPS grid system was
based on census data* for residential estimates and land use  data^ for
*
 Bureau of Census, Population Estimates and Projections for 1972-73.
t
 The East-West Gateway Coordinating Council, 1971-72 Existing Land Use
 Update and Analysis.
                                  27

-------
                                                         City of St. Louis
                                                                                      Clinton Co.       Bond Co.
oo
                                       St. Louis Co.
                   St. Charles Co.
                                                                   Monroe Co.

                                                                            4
                                                                 Randolph Co.
St. Clair Co.
                                                    Jefferson Co.

-------
commercial estimates.  Temporal distributions were derived from demand
fuel records.  The 1973 NEDS inventory was the source of annual emissions
estimates.
     The adequacy of any of the techniques is largely a function of the
pollutant and the currency of the basic data.  The basic data are three
to five years old and should be updated.   In the following two subsections,
spatial and temporal distribution methodologies are discussed.   Then each
of the critical pollutants is evaluated separately.

A.   Estimation of Spatial Resolution
     Environmental Science and Engineering Inc. (ESE) distinguished be-
tween residential and commercial land use.  Quantification of residential
areas was based on the U.S. Bureau of Census Fourth Count Computer Summary
Tables, which contain data on:  size and  nativity of families,  education,
employment status, age of home, and fuel  usage for space-heating, water
heating, and cooking for census tracts, county subdivisions,  and counties.
The determination of the spatial distribution of residential  fuel usage
was estimated from the data on these tapes for the tracts in  the St.  Louis
Standard Metropolitan Statistical Area (SMSA).

     1.   Residential Sources
          ESE developed a methodology to  equate the Census data to the
RAPS grid system (Figure 2).  In essence, their procedure involved an
overlay of the RAPS grid system on tract  maps.  Each grid square was as-
signed a visually estimated land area percentage of the total tract.   Al-
lowances were made for special terrain and land use effects (along the
banks of the Mississippi River; near Forest Park;  and so forth).  ESE
then checked results to ensure that 10070  of each tract had been apportioned
among the grid system.
          Specifically, the census data provided information  on the num-
ber of housing units using:  (1) natural  gas, (2)  bottled (LP)  gas, (3)
electricity, (4) fuel oil, (5) coal or coke, (6) wood, (7) other, and
(8) none.  It became possible for ESE to  determine annual tract fuel
usages by the following formula:
                                  29

-------
   Number  of  homes  in  tract heated
   	by fuel type i	      annual  county residential
   Number of homes in county heated   area source use of fuel type i
            by fuel type i

The annual county residential area source fuel usage of each fuel type
was available from the EPA NEDS Stationary Source Fuel Summary Reports
for the respective counties.

     2.   Commercial Sources
          Land use for commercial sources was resolved in a similar manner
as that for residential sources, only the basic reference was different.
In this case, the East-West Gateway Coordinating Council report (previously
cited) became the basic reference.  However, this report did not include
the type of fuel used.  Therefore, this methodology assumes that the rela-
tive fuel type distribution is the same as that for residential space
heating in the same area.

     3.   General Comments
          The ESE technique of spatially distributing area source emissions
is comprehensive and technically sound.  For RAPS purposes, the basic data
sources should be updated and verified.  Although St. Louis is not a rap-
idly growing  area, it  is growing and the spatial distribution of its popu-
lation and commercialization is changing.

B.   Estimation of Temporal Resolution
     For the  sake of discussion, this area source inventory can be divided
into three categories:  demand fuel usage, evaporative hydrocarbon emis-
sions, and solid waste disposal-structural fire activities.  These topics
are discussed in the following paragraphs.

     1.   Fuel Usage
          Data supplied by Laclede Gas Company  formed the basis for the
ESE temporal  distribution algorithm.  The data  itself concerned natural
gas usage, but ESE  generalized the analysis  for other space heating fuels.

                                   30

-------
The Laclede data included over a year of hourly gas flow and a number of
meteorological parameters (wind speed and direction, temperature, and
solar radiation).
          Based on the LaClede statistics, ESE empirically derived flow
equations for natural gas usage as a function of temperature and wind
speed for times of space heating demand.  In this analysis, space heating
was assumed to be in demand when temperatures were below 68°F (20°C).
When temperatures exceed 68°F (20°C), the ESE algorithm projects a base-
line value independent of the meteorological parameters.  Estimates were
also corrected for time of day.   The natural gas algorithm for space
heating is also applied to other fuels.  Therefore, strictly for space
heating, emissions for these fuels are estimated as zero when temperatures
exceed 68°F (20°C).
          The fuel demand algorithms are an effective way of temporally
distributing emissions.  They are based on a large enough data base to
be realistic.   There are bound,  however, to be some inaccuracies.  For
instance, the discontinuity of the equations at 68°F (20°C) is too abrupt
to be real.   Nevertheless, the results should provide good estimates.

     2.   Evaporative Hydrocarbon Losses
          Hydrocarbon emissions  result from evaporation of dry cleaning
fluids, solvents from paint, and gasoline at service stations.  The tem-
poral distribution of dry cleaning emissions was assumed uniform over
normal working hours.  Paint emissions were similarly assumed to be uni-
form.
          Gasoline emissions occur in two ways:  filling losses  from
underground storage tanks, and filling losses and spillage from  the fill-
ing of vehicle tanks.  The former distribution was assumed uniform over
the normal working day.  The latter was assumed to be related to auto-
mobile traffic patterns as estimated by Ludwig and Dabberdt (1972).   In
Figure 3, these diurnal traffic  patterns are illustrated for weekend and
weekday.   Both types of gasoline emissions are adjusted as a function of
month (based on marketing information).
                                  31

-------
       006
       004
    (T
    UJ
    z
    o
    <
    tr
       002
                 I	1	1	1	1	(	T



             WEEKEND MEASURED. BROADWAY AND LOCUST
           I-.—-\
       0.10
    < 008

    O


    V
    <
    o
    UJ
    uj

    $
    
-------
          In Table 1, it can be seen that the most critical  estimate is
that for evaporative losses from automobile gas filling and  spillage.
Appropriately, ESE derived a comprehensive estimate for this case.   Use
of the diurnal traffic patterns is, perhaps, the best  criteria.   However,
a composite pattern from a number of locations should  be used.   In the
analysis, data from one downtown location was used, which may or may not
be representative.

     3.   Solid Waste Disposal—Structural Fires
          Emissions from solid waste disposal through  open burning and
incineration were taken from an inventory prepared by  St. Louis  County.
Temporal allocation was based on an 8:00 a.m. to 5:00  p.m. workday.
Structural fires, on the other hand, occur randomly and, temporally, the
resulting emissions are assumed to be uniformly distributed  over the year.
Generally, emissions in this category are insignificant on an annual
basis.

C.   Discussion of Area Sources by Pollutant
     In the following paragraphs, each of the criteria pollutants is dis-
cussed separately.

     1.   Sulfur Dioxide
          Space heating requirements almost totally account  for  area
source SO^ emissions.  Fuel oil and coal are the primary fuels here.  The
consumption of these fuels can be estimated accurately for a year.   The
unknowns are the fuel sulfur content, the temporal distribution, and the
spatial distribution.  All of these parameters can be  estimated  or modeled
(as discussed previously).
          There are probably compensating factors (offsetting errors) that
should be taken into consideration.  For instance, an  underestimation of
a spatial allocation in one grid may be offset somewhat by an overestima-
tion in an adjacent grid, or fuel with a below average sulfur content from
one area may be offset by an above average sulfur content in another area--
and so forth.
                                   33

-------
          In general, with some checking of the yearly fuel consumption
data and sampling of fuels (for sulfur content) the methodology should
produce negligible annual bias.  The emission factors should be as  ac-
curate as those for point sources.   Nevertheless, the accuracy of the
emission estimate from any particular grid at a specified time is prob-
ably subject to a high degree of variability.  For the sake of future dis-
cussion, we will estimate that the precision is approximately 0.75  and the
bias is negligible.

     2.   Nitrogen Oxides
          The problems in estimating NO  are much the same for area sources
                                       X
as they were for point sources once the effects of spatial and temporal
distribution are factored out.  Spatial allocations are much the same as
they were for SO .  Temporal allocations are also much the same.  The
                A
Laclede Natural Gas data are, if anything, more accurate here since NO
                                                                      X
emissions from natural gas are prominant.
          For the purposes of our evaluation, we will again estimate the
precision at 0.75.  This time we should acknowledge that a bias probably
exists.  The PEDCO report, in analyzing the precision of area sources,
relies heavily on precisions computed for point source emission factors.
Using this rationale, the bias could average on the order of +0.4 T.

     3.   Carbon Monoxide
          For lack of better data, the accuracy of CO emission estimates
should be comparable to those for NO
                                    X

     4.   Hydrocarbons
          Evaporative sources dominate the inventory.  Therefore, the bias
should be negligible for these sources since they are based on fuel bal-
ance estimates.  The precision should be comparable to that for other pol-
lutants.
                                   34

-------
     5.    Particulates
          Except for fugitive dust,  particulate emissions  from area sources
are rather small when compared to other sources.   Emissions  predominantly
are a function of space-heating requirements.   Precision and bias estimates
should be comparable to those for NO  and CO.
                                    X

D.   Summary of Quality of Stationary Area Source Methodology
     The general methodology proposed by ESE should adequately fulfill
the accuracy requirements of the RAPS inventory.   Improvements can be
realized in a number of areas such as:
     •  Improvement of verified emission factors  for point sources
        could be applied to area sources using the same processes.
     •  A random analysis of fuels should be conducted  to  verify
        ash and sulfur content.
     •  Verification of fuel use data-records  from major fuel sup-
        pliers should be checked.
In general, the temporal and spatial allocation of the  emissions  should
be sufficient.  However, the diurnal pattern assumed for evaporative emis-
sions for hydrocarbons should be updated.   This would almost automatically
result from a similar analysis needed in the mobile source inventory.
                                  35

-------
              VII  EVALUATION OF HIGHWAY SOURCE EMISSIONS

A.   Evaluation of Highway Source Inventory
     As shown in Table 1, emissions from conventional highway sources
dominate the inventory for two pollutants (CO and HC) and significantly
effect the third (NO ).   Yet, no St.  Louis field test data (comparable
                    X
to stack test data) exist.  Hence, any evaluation will lack objectivity.
Lack of these field test data is the  most serious deficiency in the entire
RAPS inventory.  To fulfill this need, a field test plan was developed as
part of this contract and is reported separately (Shelar et al.,  1976).
     In the absence of data to corroborate estimated emissions,  we must
rely on a subjective evaluation of the quality of the input data.   Even
in this respect, accuracy of most input parameters is not known.   For
instance, the 58 second-by-second speed and time profiles are basic input
data to the RAPS highway (modal) model.  Though these are based  on actual
measurements, the sample size is relatively small.  Therefore, the repre-
sentivity of these speed and time profiles should be determined  through
further tests.
     The discussion in this section is minimal on explicit treatment of
traffic volume.  The number and type  of vehicles on the road at  any given
time are crucial.  Since emissions are estimated on an hourly basis, the
diurnal traffic variations must be defined throughout the RAPS area.  For
vehicle mix, local vehicle registration data are used.  The resulting mix
(percent) is assumed as  uniform in time and space.  In reality,  vehicle
mix is related to economic and social issues which vary from one part of
the area to another.  On the average, this variability might be negligible.
However, this should be  confirmed by  random field tests.
     The number of vehicles on the road at any specified hour (of the year)
is based on 1975 ADT estimates adjusted by a temporal distribution algor-
ithm.  The traffic volume is adjusted for:
                                  37

-------
     •  Month of the year through evaluation of gasoline sales
        records
     •  Day of the week, weekend versus weekday as shown in Fig-
        ure 3.
     •  Hour of day, as shown in Figure 3.
The latter two factors are based on site specific data that are several
years old.  Such data should be sampled in various parts of the city on
different highway types.  The resulting analysis should then be used to
determine whether one diurnal cycle is sufficient.
     Potential inaccuracies in all model input parameters must be deter-
mined through field test data.  Model performance should also be deter-
mined through field tests.  Only then can an accuracy assessment be made
on the highway source emission inventory.  As the inventory exists now,
it is probably the best of its type in existence.  However, an estimation
of bias and precision, as per Figure 1, is not too meaningful at this
time.  Nevertheless, to carry the analysis further will estimate bounds
on the precision (a ) as 1.0.
     About 807o of the highway emissions are represented by line sources
(links).  The remainder are allocated to grid squares as area sources as
shown in Figure 2.  This allocation of highway emissions to grids and
lines is consistent with requirements of air quality models (line and
area source types).  As part of the quality assurance program, the ac-
curacy of road coordinates were verified through random sampling.

B.   Highway Emission Models
     The previous discussion has focused on the need to verify highway
model input parameters.  The sensitivity of two models is treated in the
next subsection.  Although the modal model is used for the RAPS highway
inventory, we were commissioned to compare its sensitivity and estimates
to those generated by an alternative model—based on the Federal Test
Procedure (FTP).
     Emission models play an important role in the determination of mobile
source emission inventories.  First, the choice of the emission model to
be used must be made, considering both the accuracy of the model and the
                                  38

-------
difficulty and cost of obtaining the input data required by the model.
Second, once a model has been selected,  its sensitivity to errors in the
input data must be assessed to determine the effects of such errors on
emission estimates.  In fact, the model  sensitivity to input may be a
factor in the model selection process.
     For the RAPS evaluation, two emission models were considered:   (1)
the methodology presented in AP-42, Supplement 5, and (2) the modal emis-
sion model (Kunzelman, et al., 1974).  The following subsections briefly
describe these models.  A comparison was made of the emissions produced
by the two models over a wide range of model input.   The results of this
comparison are discussed in Section VII.E.  The results of an analysis
of the sensitivity of each model to errors in the input data are described
in Sections VII.F and VII.G.

C.   FTP Emission Model
     The emission factor methodology described in AP-42, Supplement 5
(referred to here as the FTP emission model), enables prediction of emis-
sion factors for CO, HC, and NO .  These basic emission factors for light-
                               X
duty vehicles (LDV) and light-duty trucks (LDT) were derived from measure-
ments made on a variety of vehicles operating over the Federal Test
Procedure (FTP) driving sequence.  The  conditions present for the FTP
driving cycle tests are:  (1) the cycle  has an average route speed of
19.6 mph, and (2) the ambient air temperature is approximately 75°F
(24°C).  The emission factors presented  in Supplement 5 represent a com-
bination of 20% of the vehicles operating in a cold condition, 2770 in a
hot start-up condition, and 537o in a hot stabilized condition.  The emis-
sion factors vary for each model year and the age of the model year at
the time of interest.
     To compute emission factors for speeds, temperatures, and percentages
of cold and hot starting vehicles other  than those given above, correction
factors are applied to the basic emission factors using the following
equation:
                                  39

-------
                              n
                             ,    c.  m. v.  z.  r.                (10)
                 npstwx     / j   ipn in ips ipt iptwx
                           i=n-12

where
          e       = Composite emission factor for LDV and LDT for
           npstwx     T    ,             ,,
                    calendar year n, pollutant p, average speed s,
                    ambient temperature t, percentage cold opera-
                    tion w, and percentage hot start operation x
                    (g/mi)
             c.   = FTP mean emission factor for the ith model year
                    during calendar year n for pollutant p (g/mi)
              m.  = Fraction of annual travel by the ith model year
               in   ,  .      ,   ,              J               J
                    during calendar year n
             v.   = Speed correction factor for the ith model year
                    for pollutant p and average speed s
             z.   = Temperature correction factor for the ith model
               P    year for pollutant p and ambient temperature t
           r      = Hot/cold vehicle operation correction factor
             P      for the ith model year for pollutant p, ambient
                    temperature t, percentage cold operation w, and
                    percentage hot start operation x.
Speed correction factors apply to speeds between 5 and 45 mph, with the
end point value used for a speed outside the range.  Temperature dependent
correction factors apply to a temperature range of 20 to 80°F (-6.7 to
26.7°C), with the end-point value used for a temperature outside the range.
The above equation applies to both LDV and LDT, although the values of the
parameters differ for the two types of vehicles.
     Heavy-duty gasoline truck (HDG) and heavy-duty diesel truck (HDD)
emission factors are based on the San Antonio Road Route test and assume
100% warmed-up vehicle operation at an average route speed of approximately
18 mph.  To adjust these emission factors for other average route speeds,
a speed correction factor is applied according to the following equation:

                                 n
                      e    =    7    c.  m. v.                     (11)
                       nps     £^j   ipn in ips
                              i=n-12
                                  40

-------
where
          e    = Composite emission factor for HDG and HDD for
            "    calendar year n, pollutant p, and average speed
                 s (g/mi)
          c.    = Test procedure emission factor for pollutant p
            P    for the ith model year in calendar year n (g/mi)
           m.  = Fraction of annual travel by the ith model year
                 during calendar year n
          v.    = Speed correction factor for the ith model year
                 for pollutant p and average speed s.
Equation 11 applies to both HDG and HDD, although the values of the param-
eters differ for the two types of vehicles.
     In addition to exhaust emission factors, the FTP methodology  provides
for computation of evaporative and crankcase hydrocarbon emission  factors
for LDV, LOT, and HDG.  Composite crankcase and evaporative HC emission
factors are determined using the equation:
                                  n
                          f  =    >    h.m.                         (12)
                           n     ^^   i in
                                i=n-12

where
          f  = Combined evaporative and crankcase HC emission fac-
               tor for calendar year n (g/mi)
          h.  = Combined evaporative and crankcase HC emission fac-
               tor for the ith model year (g/mi)
         m   = Fraction of annual travel by  the ith model  year dur-
               ing calendar year n.
Of course, each of the above parameters differ for the three vehicle types,
     Once exhaust and, for hydrocarbons, evaporative and crankcase  emis-
sion factors  have been computed for all types  of  vehicles, the emission
factor for each vehicle type is weighted according to the  percentage of
the total vehicle population of that vehicle type.  The weighted factors
are summed to obtain a composite emission factor.
                                  41

-------
D.   Modal Emission Model
     The automobile exhaust emission modal analysis model was developed
to reproduce light-duty vehicle exhaust emissions over any specified
driving sequence containing accelerations, decelerations, steady-speed
operation, and idling for CO, HC, NO , and C09.  A functional form for
                                    X        ^
the emission rate function for steady-speed cases and for acceleration
and deceleration cases was determined from test data recorded over the
Surveillance Driving Sequence.  The steady speed emission rate function
is given by:

                        es(v) = Sl + S2v + S3v2                    (13)
where
          v = Speed
   S  ,S  ,S  = Constants.

For nonzero accelerations and decelerations, the emission rate function
is given by:
                                              222
         e.(v,a)  = b1 + b2v + b,,a + b.av + b^v  + b,a  + b._v a

                      + b0a2v + b.a2v2                             (14)
                         o       y
where
          a = Acceleration  (or deceleration)
   b1  to bq  = Constants.

     The instantaneous  emission rate  function for a given vehicle and pol-
 lutant is  given  by:

                 e(v,a)  = h(a)e(v) +  [1  = h(a)]e, (v,a)            (15)
                               S                 A

 where  h(a) is a  weighting  function dependent upon acceleration and bounded
 by the values of 0 and  1,  which allow for a smooth, continuous transition
 from steady  speed to acceleration and deceleration emission rate functions,
                                   42

-------
     The emission rate function for  a  group  of  vehicles  has  been deter-
mined by averaging the coefficients  that  make up  the  emission rate  func-
tions of each vehicle in the group.  The  average  vehicle response is
obtained by integrating its  rate function over  the second-by-second speed-
time curve specified by the  driving  sequence.   The emission  response  of
the group during any driving sequence  is  determined by multiplying  the
average vehicle response by  the number of vehicles in the group.  These
emissions are then added to  the emissions computed for the other groups
that form the total vehicle  population on the highway.
     Temperature and percent cold-start vehicle correction factors  origi-
nally used in the modal model were those  given  in AP-42, Supplement 5
for model years prior to and including 1971.  The coefficients,  S , S ,
S-, and b1 to bg, however, have been updated and, as  now used, relate to
vehicles of model years prior to and including  1975.   Light-duty vehicle
modal emissions may be projected to  years beyond  1975 by applying the
ratio of future year emissions to those of the  1975 FTP  emission factors
for LDV for the appropriate  pollutant.

E.   Emission Model Comparison
     Emission predictions resulting  from  the FTP  and  modal models were
compared to determine whether the collection of the detailed input  data
required by the modal model  is warranted  or  whether the  FTP  methodology,
with its somewhat simpler input data,  predicts  emissions with sufficient
accuracy for RAPS purposes.

     1.   Model Input Data
          The input data required by the  two models are  listed in Ta-
ble 5.  To include all possibilities likely  to  be encountered in the RAPS
highway network, the CO, HC, and NO  emission  factors were computed with
                                  X
each model for a wide range  of input conditions.   We  chose the following
input conditions:  two calendar years, three temperatures, three per-
centages of cold-start vehicles (with  associated  percentages of  hot-
starting vehicles), 25 speed and time  profiles  (for the  FTP  model,  cor-
responding average route speeds, computed by taking a simple average of
                                  43

-------
                                Table 5
                       EMISSION MODEL INPUT  DATA
        FTP Model Required Input
Modal Model Required Input
      Pollutant type
      Calendar year
      Ambient temperature
      Percentages cold-starting and
      hot-starting vehicles
      Average route speed

      Fraction of total vehicles of
      each vehicle type
      Fraction of annual travel
      driven by each model year of
      each vehicle type during the
      calendar year of interest
      Traffic volume
Pollutant type
Calendar year
Ambient temperature
Percentage cold-starting
vehicles
Second-by-second speed and
time profile over the route
Fraction of annual travel
driven by LDV of each model
year during 1971

Traffic volume
the second-by-second speeds).  Values of each variable were selected  so
that the range of values for each variable allowed by the emission models
was represented.  Table 6 lists the values of the variables used in emis-
sion model computations.  Emissions were computed for all combinations of
these values, with the exception of those that are not physically realistic,
The exceptions will be discussed later.  The total number of possible cases
for the various combinations of input variable values is 1350.
          The speed and time profiles were based on data obtained by  the
Department of Transportation (DOT).  Of the 25 speed and time profiles
used in the modal computations, 24 are the results of a study made by
Washington University for EPA in which an attempt was made to isolate
"representative" profiles for the various roadway types in the St. Louis
area.  In this study, three roadway type descriptors were defined: road-
way function, average daily traffic (ADT), and volume to capacity (V/C)
ratio^  For each descriptor7~~;classes were definecPas shown' in Table 7.
There are 48 different combinations of classes for the three descriptors
                                  44

-------
                                 Table 6
         EMISSION MODEL INPUT DATA USED FOR MODEL COMPARISON RUNS
       Variable
                                  Values of Variable Used
                             FTP Model
                               Modal Model
  Calendar year
  Temperature °F (°C)
  Percentage cold
  starting and hot
  starting vehicles*
  Pollutant
  Vehicle speed
  Fraction of total
  vehicles of each
  vehicle type
  Fraction of annual
  travel driven by
  each model year
  of each vehicle
  type
1975, 1977
20,50,80 (-6.7,10,26.7)
0/0,50/10,100/0

CO, HC, NO
          X
Average route speed
corresponding to each
of 25 speed/time pro-
files
National average
vehicle type mix

National average
model year mixes
1975, 1977
20,50,80 (-6.7,10,26.7)
0,50,100

CO, HC, NO
          x
                      j.
25 speed/time profiles'
1971 national average1
   Correction factors involving hot-starting vehicles apply only to the
   FTP model.
   See text for description of speed/time profiles used.
   Taken fromAP-42, Supplement 5.   See text and Table 8.
(and thus 48 different roadway types), but only 29 of the types actually
exist in the St. Louis Area.
          In the study, speed/time profiles for both peak and off-peak
hours were chosen that were considered "representative" of each of the 29
roadway types.  The 58 second-by-second profiles chosen are of various
durations in time alrHi~~~c~ove"rr different lengths of roadway.  Examination of
the 58 profiles revealed that for the purpose of emission model comparison
it is possible to eliminate ADT as a descriptor.  In general, for each
combination of functional class of roadway and V/C ratio class, roadways
                                  45

-------
                                Table 7

                       ROADWAY  DESCRIPTOR  CLASSES
                  Descriptor
  Descriptor Classes
        Roadway function
1 = freeway
2 = principal arterial
3 = minor arterial
        Average daily traffic (ADT)
Freeway
  1 = <40,000  vehicles
  2 = 40,000-60,000
  3 = 60,000-80,000
  4 = >80,000

Principal arterial
  1 = < 10,000
  2 = 10,000-20,000
  3 = 20,000-30,000
  4 = >30,000
Minor arterial
  1 = <5,000
  2 = 5,000-10,000
  3 = 10,000-15,000
  4 = > 15,000
        Volume to capacity (V/C) ratio
1
2
3
4
<0.3
0.3-0.6
0.6-0.9
>0.9
exist for only two or three of the ADT classes.  The variation among each
of these sets of two or three profiles was found to be minimal, so one

profile was chosen to represent each combination of roadway function class
and V/C ratio class.  Since there are 3 roadway function classes and 4

V/C ratio classes, 12 peak hour and 12 off-peak hour speed and time pro-

files were selected for use in the model comparison.

          As shown in Table 7, the roadway function classes are freeway,

principal arterial, and minor arterial.  Among the original 58 speed and
time profiles there were no freeway cases that were congested, even the

peak hour case having V/C ratio class 4 and ADT class 4 (see Table 7).

In fact, the lowest average route speed for freeways was 50.0 mph with

a standard deviation of speed over the profile of 1.42 mph.  Since the
                                   46

-------
DOT study in the St.  Louis area did not contain a congested freeway pro-
file during peak travel hours, it was felt one should be added to the data
set used for emission model comparison.  Therefore,  a speed and time pro-
file was fabricated from a travel distance and travel time study conducted
on a section of Interstate 80 near Berkeley,  California.  This was a very
congested case, with  an average route speed of 4.6 mph.
          A further note is in order about the speed and time profiles
for the 29 roadway types.  For most cases, comparison of off-peak hour
profiles and peak hour profiles, with identical roadway  descriptors, shows
only slight differences in average route speed or the standard deviation
of average route speed over the profile.  There are  a few notable excep-
tions.   For many of the roadway types in the  St.  Louis area, variation
between peak and off-peak hour profiles would be expected.  This fact,
as well as the absence of a congested freeway case,  was  discussed with
the investigators who performed the speed and time profile study.   The
explanation given for these factors is that the data sample from which
the 58 "representative" speed and time profiles were chosen is limited.
The entire data sample contained no congested freeway cases, and the cases
chosen as representative are the most representative of  the data that was
available.
          As mentioned above, emissions were  computed using both the FTP
and modal models for  all combinations of 2 calendar  years, 3 temperatures,
3 percentages of cold-starting vehicles, 3 pollutants, and 25 average
route speeds or speed and and time profiles,  with the exception of those
cases that were not physically realistic.
          The cases considered unrealistic are the following:  (1) the
162 cases having 1007o cold-start vehicles on  freeways; (2) the 144 cases
having 10070 cold-start vehicles on principal  arterials;  and (3) the 144
cases having 070 cold-start vehicles on minor  arterials.   Thus, of the
1350 possible combinations, 900 cases (300 for each  pollutant) were con-
sidered realistic and were used in the model  comparison  computations.
          Other input data required by the FTP model include the fractions
of total vehicles of  each vehicle type and the fractions of annual travel
                                  47

-------
driven by each model year of each vehicle type.   The vehicle type mix
used in model calculations is the national average vehicle type mix:
80.4% LDV, 11.8% LOT, 4.6% HDG, and 3.2% HDD.  The national average model
year mixes used in FTP emission computations for 1971,  1975, and 1977,
are listed in Table 8.
                                Table 8
               FRACTION OF ANNUAL TRAVEL BY VEHICLE

Vehicle Age
(years)

1
2
3
4
5
6
7
8
9
10
11
12
13+
Fraction of Annual Travel
Calendar Year
1971
LDV
.116
.135
.125
.122
.106
.086
.083
.072
.051
.037
.023
.012
.033
LOT
.094
.138
.127
.131
.098
.083
.076
.057
.044
.032
.023
.016
.081
HGT
.062
.111
.117
.122
.093
.080
.066
.057
.047
.040
.031
.021
.153
Calendar Years
1975 and 1977
LDV
.112
.143
.130
.121
.108
.094
.079
.063
.047
.032
.019
.013
.039
LOT
.094
.141
.132
.123
.098
.083
.076
.057
.044
.032
.023
.016
.081
HGT
.062
.124
.117
.110
.093
.080
.066
.057
.047
.040
.031
.021
.153
HDT
.096
.169
.168
.164
.110
.080
.067
.048
.034
.018
.011
.007
.029
          Taken from AP-42, Supplement 5.


          The modal model also requires input of the fraction of annual
travel driven by each LDV model year.  Since the coefficients used in
computations with the modal model are derived from calendar year 1971
data, the national average model year mix for 1971, listed in Table 8,
was used as model input.  To project modal emissions to future years (in
this study, years 1975 and 1977), the future year FTP emission factor for
the case being modeled is computed using the vehicle model year mix for
that year.  The 1971 FTP emission factor is also computed, using the 1971
vehicle model year mix.  Then, the ratio of the FTP future year to FTP
                                  48

-------
1971 emissions is applied to the 1971 modal LDV emissions to project the
modal LDV emissions to a future year.

     2.   Method for Determination of Composite Emission
          Factors
          The emission factor methodology in AP-42, Supplement 5, allows
computation of CO, HC, and NO  exhaust along with HC evaporative and
                             X
crankcase emission factors for LDV, LOT, HDG, and HDD.  Each of these
emissions is weighted according to the fraction of the total vehicle
population belonging to that vehicle type;  then the weighted factors are
summed to produced a composite emission factor for the highway.
          The modal emission model allows for computation only of LDV
exhaust emission factors.  To compare the output of the two models, LDV
modal emissions were converted to a composite emission factor.  This was
done using the following method:  (1) the ratio of the composite FTP ex-
haust emission factor to the FTP LDV exhaust emission factor was computed
for each case; (2) the modal LDV exhaust emission factor was multiplied
by this ratio; and (3) for HC, the FTP evaporative and crankcase composite
emission factor was added to the result of  (2) above.  The emission factor
computed using this procedure is considered to be a composite modal emis-
sion factor.  The basic assumptions of the  procedure are:  (1) the func-
tional form of LDT, HDG, and HDD emissions  over the driving sequence is
the same as the functional form of LDV emissions over the sequence; (2)
the ratio of composite exhaust emissions to LDV exhaust emission is the
same for both the FTP and modal models; and (3) for HC, the functional
forms of exhaust emissions and of evaporative and crankcase emissions are
not the same.

     3.   Results
          CO, HC, and NO  emission factors  were computed with the two
                        X
emission models- for each of the 300 cases.   For each pollutant, the 300
modal emission values were plotted against  the corresponding FTP emission
values.  The range of emission values was such that when all 200 points
were plotted on the same figure, individual points could not be

                                  49

-------
distinguished in areas where points were  clustered.  Therefore,  to  gain
sufficient resolution so that individual  points  could  be  distinguished,
the plots presented in Figures 4  to 6  include  only  those  points  within
a limited range.  The range was chosen so the  region where  the majority
of points was clustered could be  shown in detail.   Figure 7  is an en-
largement of a portion of Figure  4.  The  regression lines on Figures 4
to 7 are a best fit to each of the entire 300  point data  sets.
          Various statistical quantities  were  computed from the  emission
factors for each pollutant.  The  quantities  computed include;  (1)  the
correlation coefficient and linear regression  constants  (FTP emission  =
A X Modal emission + B) for the modal  versus FTP emission factors;  and
       900
       800
        700
     2 600
     l
     1
     o 500
     o
     (7>
     (O
     LJ
400
       300
       200
        100
                                   I
                                   CO
                               •    •
                                               \
                                              I
          0     100    200   300   400   500    600    700    800   900
                           MODAL EMISSIONS (G /VEH-MI)
FIGURE 4  PLOT OF CO EMISSIONS COMPUTED WITH FTP METHODOLOGY AND WITH
          MODAL MODEL FOR CASES  WITH EMISSIONS LESS THAN  900 g/veh-mi
                                  50

-------
         60
         50
     2
     I
     i
     o
     V)
     o
     tn
     
-------
          12
          10
     UJ
          8
     VI
     o
     LJ
     a
     c
                   -J    i    I   ,    J    ,
                                       NO
          • •• M • **4kl>>. I •
  • « »«^   • . i*   •tt'v.1.
 ...**.i/!.* •"•* '.•^C*i ""•""*
•i *ri .->-••.» V "A*tIJ.'-
   %     ••

                                  i
                  i
i
                                  6      8      10
                            MODAL EMISSIONS (G /VEH-MI)
                                12
FIGURE 6   PLOT OF NOX EMISSIONS COMPUTED WITH FTP METHODOLOGY AND WITH
           MODAL  MODEL FOR CASES WITH EMISSIONS  LESS THAN 15 g/veh-mi
          The  correlation between the modal and FTP  emission models for
CO and HC is very  high, as shown in Tables 9 and 10.  For  both  pollutants
the correlation  coefficients for all cases and for each  subset  of cases
are greater  than 0.9.   Also, the fits of the regression  lines are good,
as seen in Figures 4 and 5.   The correlations of the  subsets were com-
puted and compared to determine which cases correlated the best and which
cases had the  poorest correlation.   A test was applied to  the correlation
coefficients of  the subsets  which were being compared to determine whether
the differences  between them are statistically significant.   A  test sta-
tistic, z, was computed according to the following equations:
                                   52

-------
        300
        250
     2  200
     I
     r
     UJ



     o
     o
     V)
     CO

     5
        150
        100
         50
                     ~r

                     CO
                        " %'  .— X*   •
                          .  .; ''• • .
                          i. • • «.«••»  •
                         .  ,'.  . ,

                         .'7?, '* '
                              I
                      I
I
                    50       100       150      200

                            MODAL EMISSIONS (G /VEH-MI)
                                       250
                 300
 FIGURE 7   PLOT OF CO  EMISSIONS COMPUTED WITH FTP METHODOLOGY AND WITH

           MODAL MODEL FOR CASES WITH EMISSIONS LESS THAN 300 g/veh-mi
where
                           Z.  = 1/2 In
                              = 1/2 In
                                        1 - r.
                                        1 + r.
                                        1 - r
                                             2 J
                                      Zl -
          Z =
                                                    (16)
                                                    (17)
                                                                     (18)
                                           N2-3
          rl> r2


          N1»N2
= The two correlation  coefficients being compared


= The sizes of the  samples  from which the correla-

  tion coefficients were  computed.


                  53

-------
                                                      Table 9
                                        MODEL COMPARISON  STATISTICS  FOR  CO
Cases
All cases
1975
1977
Freeway
Principal arterial
Minor arterial
Peak hour
Off-peak hoyr
Congested
Noncongested
20° temperature
50° temperature
80° temperature
0% cold starts
50% cold starts
100% cold starts
Number
of Cases
in Sample
300
150
150
108
96
96
156
144
84
216
100
100
100
102
150
48
Correlation
Coefficient
.9714
.9793
.9762
.9874
.9697
.9412
.9748
.9549
.9722
.9354
.9736
.9758
.9847
.9823
.9754
.9390
Linear Regression
Constants
A
.793
.730
.908
.849
.785
.665
.802
.773
.815
.779
.772
.933
.731
.753
.806
.648
B
4.39
7.51
-1.70
0.12
1.70
21.55
1.91
7.42
-3.22
6.66
4.08
2.76
2.05
0.40
4.98
29.51
Modal
Average
94.0
105.3
82.7
75.7
101.7
106.9
112.7
73.7
174.6
62.7
140.9
78.4
62.8
48.7
112.6
132.2
Standard
Deviation
100.6
111.9
86.3
119.8
90.1
82.3
121.6
65.3
145.8
47.8
139.3
69.2
52.3
42.3
117.5
96.9
FTP
Average
78.9
84.4
73.4
64.4
81.5
92.6
92.3
64.4
139.1
55.5
112.8
75.9
48.0
37.0
95.8
115.2
Standard
Deviation
82.1
83.4
80.3
103.0
72.9
58.1
100.0
52.9
122.3
39.8
110.5
66.2
38.9
32.4
97.1
66.9
U1

-------
                                                       Table 10

                                         MODEL COMPARISON STATISTICS FOR HC
Cases
All cases
1975
1977
Freeway
Principal arterial
Minor arterial
Peak hour
Off-peak hour
Congested
Noncongested
20° temperature
50° temperature
80° temperature
0% cold starts
50% cold starts
100% cold starts
Number
of Cases
in Sample
300
150
150
108
96
96
156
144
84
216
100
100
100
102
150
48
Correlation
Coefficient
.9635
.9694
.9582
.9894
.9621
.9253
.9781
.9174
.9803
.9313
.9643
.9690
.9795
.9801
.9784
.9266
Linear Regression
Constants
A
.842
.809
.896
.851
.830
.818
.842
.869
.855
1.04
.825
.880
.703
.757
.824
.810
B
1.55
1.80
1.17
1.14
1.30
2.51
1.38
1.51
..694
.419
2.12
1.71
1.71
1.32
1.91
3.07
Modal
Average
9.10
9.89
8.32
8.21
9.73
9.47
10.28
7.82
14.40
7.04
11.91
8.13
7.27
7.07
9.95
10.77
Standard
Deviation
5.86
6.37
5.19
7.03
5.09
4.95
7.02
3.88
7.94
2.84
7.76
4.22
3.58
3.66
6.66
5.83
FTP
Average
9.21
9.80
8.62
8.13
9.37
10.26
10.04
8.30
13.00
7.73
11.94
8.87
6.81
6.68
10.10
11.79
Standard
Deviation
5.12
5.31
4.85
6.05
4.39
4.37
6.04
3.67
6.92
3.17
6.64
3.83
2.57
2.83
5.60
5.09
Ul
Ul

-------
                                                      Table  11
                                        MODEL COMPARISON STATISTICS FOR NO
Cases
All cases
1975
1977
Freeway
Principal arterial
Minor arterial
Peak hour
Off-peak hour
Congested
Noncongested
20° temperature
50° temperature
80° temperature
0% cold starts
50% cold starts
100% cold starts
Number
of Cases
in Sample
300
150
150
108
96
96
156
144
84
216
100
100
100
102
150
48
Correlation
Coefficient
.3669
.4154
.3108
.5838
.0071
.1547
.3845
.3618
.0640
.6853
.4039
.2251
-.0240
.3451
.1290
-.1420
Linear Regression
Constants
A
.140
.163
.116
.220
.003
.070
.181
.113
.020
.248
.136
.067
-.007
.152
.047
-.058
B
5.00
4.86
5.15
4.78
5.60
5.27
4.71
5.22
5.28
4.56
5.43
5.45
5.42
5.07
5.52
5.71
Modal
Average
6.13
6.26
5.99
6.89
6.47
4.92
6.06
6.20
6.50
5.98
6.67
6.13
5.58
7.05
5.98
4.62
Standard
Deviation
1.65
1.67
1.62
1.47
1.68
1.00
1.40
1.87
1.82
1.55
1.85
1.56
1.30
1.68
1.39
0.87
FTP
Average
5.86
5.88
5.84
6.29
5.62
5.62
5.81
5.91
5.41
6.04
6.34
5.86
5.39
6.14
5.80
5.44
Standard
Deviation
0.63
0.65
0.60
0.55
0.59
0.46
0.66
0.58
0.56
0.56
0.62
0.46
0.36
0.74
0.51
0.36
Cn

-------
The test statistic is then compared with the confidence coefficients cor-
responding to the 95 and 997o confidence levels to determine whether the
differences in the correlation coefficients are significant at the 95 or
997, confidence level.  Table 12 lists the results of the significance
tests.
          The significance tests show that the correlation between emis-
sion model results are equally good for emissions from the 1975 vehicle
mix and the 1977 vehicle mix.  The freeway cases are better correlated
than the principal or minor arterial cases, and the principal arterial
cases are better correlated than the minor arterial cases.  However, peak-
hour cases show better correlation than do nonpeak hour cases, and con-
gested cases are better correlated than noncongested cases.  The congested
and noncongested results seem inconsistent with the correlations found for
the three highway-type subset comparisons.  Since all except one of the
freeway cases are noncongested and over half of the principal arterial
cases are congested and those cases comprise 717o of the congested cases,
it would appear that the principal arterial cases should have better cor-
relation than the freeway cases.  This apparent inconsistency has not been
resolved.  Since the peak-hour cases are better correlated than the off-
peak hour cases and most of the congested cases are for the peak hour,
the peak and off-peak hour and the congested and noncongested results
appear consistent.
          The cases having an ambient air temperature of 80°F (26.7°C)
are better correlated than those having a temperature of 20°F (-6.7°C).
The poorer correlation of the 20°F cases is probably a result of error
introduced by the temperature correction factors of the emission models.
The temperature at which the basic emission factor measurements were made
was near 75°F (25°C).  Thus, for the 80°F cases, the effect of the tem-
perature correction  factors was minimal since the factors  for 80° are
near unity; for the 20° cases, however, the factors are large.
          The correlation of cases having 070 cold-start vehicles is sig-
nificantly better than the correlation of cases having 10070 cold-start
vehicles.  Also, cases with 507o cold-start vehicles show better correla-
tion than cases with 1007» cold-start vehicles.  The explanation for this

                                   57

-------
                                     Table 12
RESULTS OF TESTS FOR SIGNIFICANCE OF DIFFERENCES BETWEEN CORRELATION COEFFICIENTS
                      OF VARIOUS SUBSETS OF THE DATA SAMPLE
Pollutant
HC and CO
HC and CO
HC and CO
HC and CO
HC and CO
HC and CO
HC and CO
HC and CO
HC and CO
HC and CO
HC and CO
HC
CO
Cases for Which Correlation Coefficients
are Being Compared
1975 1977
Freeway Principal arterial
Freeway Minor arterial
Principal arterial Minor arterial
Congested Noncongested
20° 50°
20° 80°
50° 80°
0% cold starts 50% cold starts
0% cold starts 100% cold starts
50% cold starts 100% cold starts
Peak hour Off-peak hour
Peak hour Off-peak hour
Significance
Not significant
Significant at a 99%
confidence level
Significant at a 99%
confidence level
Significant at a 95%
confidence level
Significant at a 99%
confidence level
Not significant
Significant at a 95%
confidence level
Not significant
Not significant
Significant at a 99%
confidence level
Significant at a 99%
confidence level
Significant at a 99%
confidence level
Significant at a 95%
confidence level
Cases Showing
Highest Correlation

Freeway
Freeway
Principal arterial
Congested

80°


0%
50%
Peak hour
Peak hour

-------
is similar to that for temperature:   error is probably introduced by the
correction factors--in this case,  the cold-start correction factors.  The
basic emission factors used in the FTP model assume 207<> cold-start vehicles;
the modal model coefficients, however, are based on 0% cold-start vehicles.
Thus, for percentages of cold-start vehicles greater than the assumed base
levels, the influence of the correction factors is greater and correlation
between the models is reduced.
          It is probable that the  poorer correlation of the 10070 cold-
start cases is a major influence on the correlation of the minor-arterial
cases, compared with the other highway type subsets.  One-half of the minor
arterial cases have 10070 cold-start vehicles; the freeway and principal
arterial cases include only 0 and  507o cold-start cases, respectively.
Since the 1007<> cold-start cases show poorer correlation than do the 0 and
507o cases, the correlation of the  minor arterial cases, compared with
freeway and principal arterial cases, would be reduced.
          Examination of Table 11  reveals that correlation between the
emission models for NO  is poor.  Figure 6 shows that the variation among
                      X
modal NO  emission values for the  test cases is considerably greater than
        X
the variation among FTP emission values for the same test cases.  Table 11
gives the standard deviation of modal emissions for the 300 test cases as
1.65 g/veh-mi and the FTP standard deviation as 0.63 g/veh-mi.  This large
difference is due to the data bases from which the emission models
were derived.  The emission values from which the modal coefficients, de-
scribed in Section VII.D, were computed were not corrected for humidity;
the FTP emission factors were corrected.  It is believed that for NO ,
                                                                    x
the FTP methodology is a more accurate means of computing emissions.

F-   FTP Model Sensitivity Analysis
     The input data required by the FTP methodology are listed in Table 5.
An analysis was performed to assess the error that would be introduced into
the emission factors if there were an error in estimating the input data.
First, the analysis was centered on determination of relative error result-
ing from error in a single input variable.  Then, based on the results of
                                   59

-------
this analysis, relative error resulting from error in two input param-
eters was determined.

     1.   Sensitivity to Error in a Single Input Parameter
          For each pollutant, the sensitivity to error in each of the
following four input parameters was determined:  average route speed,
ambient temperature, percentage of cold-start vehicles, and traffic vol-
ume.  It was assumed that the other input parameters, such as calendar
year, vehicle -type mix, and model year mix, could be accurately assessed.
The national average vehicle type mix and model year mix were used in the
calculations.

          Speed Sensitivity Analysis—In the speed sensitivity analysis,
relative error caused by erroneous speed was computed for each of three
roadway types:  freeways, principal arterials, and minor arterials.  Sev-
eral assumptions were made to distinguish the roadway types.  It was as-
sumed  that average route speeds on freeways vary from 5 to 45 mph; speeds
on principal arterials vary from 5 to 40 mph, and speeds on minor arterials
vary from 5  to 35 mph.  The following were assumed for all roadway types:
(1) a  speed  of 5 mph can be as much as +10 mph in error; (2) a speed of
10 mph can be as much as +10 or -5 mph in error; (3) speeds from 15 to 35
mph can be as much as +10 or -10 mph in error.  In addition, it was as-
sumed  that a  speed of 40 mph on a principal arterial or a freeway can be
as much as +5 or -10 mph in error, and a speed of 45 mph on a freeway can
be as  much as -10 mph in error.
          Emission factors were computed for all combinations of 2 calen-
dar years, 3  temperatures, 3 percentages of cold-start vehicles, and 9
average route speeds.  Table 13 lists the values of  these parameters used
in the analysis.  For each roadway type, the average relative error in
estimating emission  factors caused by errors of +5,  -5,  ±5, +10, -10,
and  ±10 mph  in speed was computed for each pollutant,  considering the
assumptions  listed above.  The results of these computations are shown
in Table 14.  As could be expected, the larger the magnitude of the error
in speed, the greater the relative error in the emission factor.  For HC

                                   60

-------
                               Table 13
        VALUES OF INPUT PARAMETERS ASSUMED IN FTP MODEL SINGLE
                    PARAMETER SENSITIVITY ANALYSIS
        Parameter

      Speed
      Calendar year
      Temperature
      Cold starts
   Units
         Values Assumed
      Temperature

      Calendar year
      Cold starts
      Speed
   Speed Sensitivity Analysis
    mph      5,10,15,20,25,30,35,40,45
             1975,  1977
    °F       20,50,80
    %        0,50,100

Temperature Sensitivity Analysis
    °F       20,21,22,38,39,40,41,42,58,59,60,
             61,62,78,79,80
             1975,  1977
    %        0,50,100
    mph      5,15,25,35,45

 Cold-Start Sensitivity Analysis
      Cold starts

      Calendar year
      Temperature
      Speed
     °F
     mph
0,5,10,15,20,25,30,35,40,45,50,
55,60,65,70,75,80,85,90,95,100
1975, 1977
20,50,80
5,15,25,35,45
and CO, underestimated speeds produce larger  errors  then  do  overestimated
speeds of the same magnitude.  This  is because of  the  exponential  form of
the speed correction factor functions. For NO , the slight  difference in
                                              X
relative error between cases with underestimated and overestimated speeds
of the same magnitude is not significant;  it  results from the  choice  of
cases used to compute the average relative error.  From the  HC and CO re-
sults, it appears that,  for these two emission types,  it  is  better to over-
estimate average route speed than it is to underestimate  it.
               The fractional error  caused by a particular amount  of  error
in speed is least for freeways and greatest  for minor  arterials.   The
                                  61

-------
                         Table 14

AVERAGE FRACTIONAL ERROR IN FTP EMISSION FACTORS RESULTING
          FROM ERROR IN A SINGLE INPUT PARAMETER

Pollutant

HC
CO'
NO
X
HC
CO
NO
X
HC
CO
NO
X

Facility Type

Freeway
Freeway
Freeway
Principal arterial
Principal arterial
Principal arterial
Minor arterial
Minor arterial
Minor arterial

Error in Average Route Speed
(mph)
+10
.24
.37
.08
.24
.37
.08
.24
.37
.08

-10
.35
.69
.08
.40
.77
.09
.45
.85
.09

±10*
.29
.53
.08
.31
.55
.08
.32
.57
.09

+5
.14
.22
.05
.14
.22
.05
.15
.24
.05

-5
.18
.31
.05
.20
.34
.05
.22
.38
.06

±5*
.16
.26
.05
.17
.28
.05
.18
.30
.05

Pollutant
HC
CO
NO
X
Error in Ambient Temperature
(°F)
+2
.02
.03
.00
-2
.02
.03
.01
±2*
.02
.03
.01
+1
.01
.01
.00
-1
.01
.02
.00
±1*
.01
.01
.00
Pollutant
HC
CO
NO
X
Error in Percentage Cold-Starting Vehicles
±5Qt
.37
.77
.06
+20
.16
.37
.02
-20
.13
.24
.02
±20*
.15
.30
.02
+10
.08
.18
.01
-10
.07
.12
.01
±10*
.07
.15
.01
 The average fractional  error  in emission  factors  resulting
 from input parameter error of ±X may  or,may  not be  the  simple
 average of the fraction error caused  by input  parameter error
 of +X and -X,  since in  some cases different  sample  sizes were
 used in computing the +X and  -X errors.

 Errors of +50  and -50 were grouped together  because the sample
 size was small.
                            62

-------
differences arise from the assumptions made about  what average route
speeds occur on the different roadway types.   The  differences in frac-
tional error of emission factors among the roadway types suggest that
the error is smaller when high speeds are misestimated than when the low
speeds are misestimated; the exponential nature of the speed correction
function causes less sensitivity to error at the higher speeds.  The maxi-
mum error computed for HC is 457>, for CO is 85%, and for NO  is 9%.
                                                           X

          Temperature Sensitivity Analysis—To perform an analysis of the
sensitivity of FTP emission factors to temperature, the emission factors
for all combinations of the 2 calendar years, 3 percentages of cold-
starting vehicles, 5 average route speeds, and 16  temperatures listed in
Table 13 were computed.  For each pollutant, the sensitivity of the emis-
sion factors to errors of +2, -2, ±2, +1, -1, and  ±1 degrees Farenheit was
assessed; the results appear in Table 14.  The largest emission factor
error results from the cases with the largest error in temperature.   Tem-
perature can be measured quite accurately, so it is doubtful that it will
be in error by more than 2°F (1°C).  Since the largest relative error
computed for any pollutant is less than three percent, even for 2°F tem-
perature error, the sensitivity of the FTP methodology to erroneous tem-
perature is virtually negligible.

          Cold-Start Vehicle Sensitivity Analysis—Table 13 lists the
values for the 2 calendar years, 3 temperatures, 5 average route speeds,
and 21 percentages of cold-start vehicles used in  the analysis of FTP
emission factor sensitivity to error in estimating the percentage of
cold-start vehicles.  The average relative errors  in emission factors
resulting from cold-start errors of ±50, +20, -20, ±20, +10, -10, and ±107o
were computed and are given in Table 14.  As expected, the larger the mag-
nitude of the error in percent cold-start vehicles, the larger the error
in the emission factor.  For HC and CO, overestimating the percent of
cold-start vehicles produces larger error than underestimating it by the
same magnitude.  This is due to the functional form of the cold-start
correction factor.  On the average, for CO, 777o error in emission factor
is introduced by underestimating or overestimating the percent cold-start
                                   63

-------
vehicles by 5070 or more.  Less than 67, error in NO  emission factors re-
                                                  X
suits from even a 507, error in percent of cold-start vehicles.  The HC
emission factor error is as large as 37% when the cold-start factor is
in error by 507, .

          Traffic Volume Sensitivity Analysis—Since pollutant emissions
are directly proportional to the volume of vehicles on a roadway, an error
in volume of a particular magnitude produces a corresponding error in
emission factor.  This will hold true as long as the roadway is uncongested.
However, if the'volume is in error and the average route speed is therefore
estimated erroneously, the resultant error in emissions would not be di-
rectly proportional to the volume error.  An analysis of error introduced
by congested conditions is considered beyond the scope of this study.

     2.   Sensitivity to Error in Two Input Parameters
          The results of the single parameter sensitivity tests show that
FTP emission factors are relatively insensitive to the error in temperature
measurements likely to be encountered in RAPS.  Therefore, the sensitivity
of emission factors to temperature error in conjunction with error in
another input parameter was not considered.  The two parameter sensitivity
analysis was conducted for:  error in speed and volume; percent cold-
start vehicles and volume; and speed and percent cold-start vehicles.
          Rather than compute emission factors for various combinations
of volume error and speed or cold-start error to obtain average relative
error caused by error in volume and speed or cold starts, an equation was
derived to compute the relative error.  This is possible because of the
direct proportionality of volume error to emission factor error.  Let v
be the correct volume and v ' be the amount by which the volume is in error,
positive or negative.  Let E be the correct emission factor and E  be the
positive or negative amount the emission factor is in error because of
error in speed or percent cold-start vehicles.  Then the relative error
is given by
                         • vE - (v + v ') (E + E ')
                 error =  -      --*
                                   64

-------
Reorganizing and reducing,

                      error = I  ?- + %- I 1 + ^- II                  (20)
But the quantity v /v is the fractional amount the volume is in error.
Therefore, for any fraction of volume error, the total relative error
caused by error in both volume and a second input parameter is a simple
function of the fractional error due to error in the second input
parameter.  Equations with the same form as Eqs. (19) and (20) were used
in the speed and volume and cold-start and volume sensitivity analyses.

          Speed and Volume Sensitivity Analysis--Table 15 lists the values
of speed, calendar year, temperature, and percent cold-start vehicles used
in the speed and volume sensitivity analysis.  This analysis was performed
using speeds ranging from 5 to 45 mph.  Computations were made for volume
error of +20, -20, +10, -10, +5, -5, ±20, ±10, and +5% and for speed error
of +5, -5, ±5, +10, -10, and ±10 mph.  The relative error resulting from
all combinations of volume and speed error classes for each pollutant are
given in Table 16.  The largest error results when the effect of the errors
in both volume and speed produces emission factor errors of the same sign.
          For HC and CO, the largest error occurs when volume is overesti-
mated and speed is underestimated, or when volume is underestimated and
speed is overestimated.  In these cases, when there is an error in either
the speed or volume parameter, the emission factor error will become larger
as the magnitude of the error in the other parameter is increased.  How-
ever, when volume and speed are both either underestimated or overesti-
mated, the emission factor error introduced by error in one of the input
parameters has the opposite sign of the emission factor error caused by
error in the other input parameter.  The net effect is a varying degree
of error cancellation, depending on the magnitude of the error in each in-
put parameter.  For a given amount of error in volume, the larger the mag-
nitude of the error in speed, the larger the emission factor error.  But
for a given amount of error in speed, as the magnitude of the volume error
decreases the emission factor error may either  (1) increase, or  (2)
                                   65

-------
                                Table 15
          VALUES OF  INPUT PARAMETERS ASSUMED IN FTP MODEL TWO
                     PARAMETER SENSITIVITY ANALYSES
        Parameter
Units
          Values Assumed
                  Speed  and Volume  Sensitivity Analysis
      Speed             mph     5,10,15,20,25,30,35,40,45
      Calendar year     --      1975, 1977
      Temperature       °F      20,50,80
      Cold starts       %       0,50,100

               Cold-Start and Volume Sensitivity Analysis
      Cold starts

      Calendar year
      Temperature
      Speed
 °F
 mph
0,5,10,20,25,30,35,40,45,50,55,60,
65,70,75,80,85,90,95,100
1975, 1977
20,50,80
5,15,25,35,45
                Speed and  Cold-Start  Sensitivity Analysis
      Speed
      Cold starts

      Calendar year
      Temperature
 mph     5,10,15,20,25,30,35,40,45,
 %       0,5,10,15,20,25,30,35,40,45,50,55,
         60,65,70,75,80,85,90,95,100
         1975, 1977
 °F      20,50,80
decrease and then increase.  Thus, in some cases, for a given amount of
speed error, the total emission factor error is smaller for a large vol-
ume error than it is for a small volume error.   This occurs because the
emission factor errors related to speed and a larger volume error more
nearly cancel than do the errors related to speed and a smaller volume
error.
          For NO  when speed and volume are both overestimated or under- •
                x       r
estimated, they produce emission factor errors  of the same size.   However,
the emission factor error resulting from underestimated volume and over-
estimated speed, or from overestimated volume and underestimated speed
                                   66

-------
                               Table  16
      AVERAGE FRACTIONAL ERROR IN FTP EMISSION FACTORS RESULTING
                  FROM ERROR IN BOTH SPEED AND  VOLUME


Pollutant

HC





CO





NO
X





Error
in Speed
(mph)
+5
-5
±5
+10
-10
±10
+5
-5
±5
+10
-10
±10
+5
-5
±5
+10
-10
±10
Error in Volume

(%)
+20
.11
.42
.26
.12
.62
.37
.11
.57
.34
.24
1.03
.63
.21
.20
.20
.23
.18
.21
-20
.31
.13
.22
.39
.15
.27
.37
.13
.25
.49
.36
.42
.20
.20
.20
.18
.22
.20
+10
.08
.30
.19
.16
.49
.32
.14
.44
.29
.30
.86
.58
.11
.10
.10
.15
.08
.11
-10
.22
.10
.16
.31
.22
.27
.30
.18
.24
.43
.52
.48
.09
.10
.10
.08
.14
.11
+5
.10
.24
.17
.20
.42
.31
.18
.38
.28
.33
.78
.55
.08
.05
.06
.11
.07
.09
-5
.18
.12
.15
.27
.29
.28
.26
.25
.25
.40
.61
.50
.04
.07
.06
.06
.11
.08
±20*
.21
.27
.24
.26
.39
.32
.24
.35
.30
.37
.69
.53
.20
.20
.20
.21
.20
.20
±10*
.15
.20
.17
.24
.36
.30
.22
.31
.27
.37
.69
.53
.10
.10
.10
.11
.11
.11
±5*
.19
.18
.16
.24
.35
.29
.22
.31
.26
.37
.69
.53
.06
.06
.06
.08
.09
.09
   The average fractional error in emission factors  resulting from input
   parameter error of ±X may or may not be the simple average of the
   fractional error caused by input parameter error  of +X and -X, since
   in some cases different sample sizes were used in computing the +X
   and -X errors.
have opposite signs and have a cancelling effect.   For a given speed error,
larger volume error produced larger emission factor error.   This is also
true for cases in which emission factor errors resulting from speed error
and volume error both have the same sign.  But when these errors have op-
posite signs for a given amount of volume error,  the resultant emission
factor error from'a smaller amount of speed error may be larger than the
emission factor error from a larger amount of speed error.
          The magnitude of the emission factor error resulting from error
in both volume and speed varies among the pollutants.  The maximum error
                                  67

-------
for all three pollutants occurs when speed is underestimated by 10 mph
and volume is overestimated by 207o.  The maximum errors are:  HC, 6270;
CO, 103%; and NO , 23%.
                X

          Cold-Start and Volume Sensitivity Analysis—Table 15 lists the
values of percent of cold-start vehicles, calendar year, temperature, and
speed used in the cold-start and volume sensitivity analysis.  Volume
errors were computed for +20, -20, +10, -10, +5, -5, ±20, ±10, and ±5%;
cold start errors were computed for +10, -10, ±10, +20, -20, ±20, and
±5070.  The average relative errors in emission factors for each pollu-
tant resulting from all combinations of the above classes of volume and
cold-start errors are given in Table 17.
          For HC and CO, the largest emission factor error results when
cold starts and volume are both either underestimated or overestimated.
For these cases, for a given amount of error in one input parameter (cold
starts or volume), the emission factor error will become larger as the
magnitude of the error in one input parameter (volume or cold starts) is
increased.  When cold starts are underestimated and volume is overesti-
mated or when cold starts are overestimated and volume is underestimated,
the emission factor errors resulting from error in volume and from error
in cold  starts partially cancel.  For these cases, for a given amount of
volume error, the emission factor error will become larger as the magni-
tude of  the error in cold starts becomes larger.  However, for a given
amount of error in cold starts, the emission factor error will (1) de-
crease,  (2) decrease and then increase, or (3) increase as the volume
error magnitude decreases.  The reason such effects are observed was dis-
cussed previously in this report.
          For NO  , emission factor errors are additive and therefore larg-
                x'
est when cold start is overestimated and volume is underestimated, or when
cold start is underestimated and volume is overestimated.  For these cases,
for a given error in one of the input parameters being tested, the emis-
sion factor error decreases as the magnitude of the error in the other
input parameter decreases.  When error in cold start and volume are both
either underestimated or overestimated, the emission factor error resulting

                                   68

-------
                               Table  17

       AVERAGE  FRACTIONAL  ERROR IN FTP EMISSION FACTORS RESULTING
               FROM ERROR IN BOTH COLD STARTS AND VOLUME


Pollutant

HC






CO






NO
X






Error in
Cold Start
(7»)
+10
-10
±10*
+20
-20
±20*
±50^
+10
-10
±10*
+20
-20
±20*
±50t
+10
-10
±10*
+20
-20
+20*
±50t
Error in Volume
(°i \
{ ")
+20
.30
.12
.21
.39
.09
.24
.48
.42
.08
.25
.64
.11
.37
.94
.19
.21
.20
.17
.23
.20
.20
-20
.14
,25
.20
.11
.31
.21
.31
.12
.30
.21
.14
.39
.27
.60
.21
.19
.20
.22
.18
.20
.20
+10
.19
.05
.12
.28
.07
.18
.42
.30
.05
.17
.50
.16
.33
.86
.09
.11
.10
.08
.13
.10
.10
-10
.05
.16
.11
.08
.22
.15
.32
.08
.21
.15
.23
.32
.27
.69
.11
.09
.10
.12
.08
.10
.11
+5
.14
.03
.08
.22
.09
.15
.39
.24
.08
.16
.43
.20
.32
.81
.04
.06
.05
.03
.08
.05
.07
-5
.04
.11
.08
.11
.17
.14
.35
.12
.16
.14
.30
.28
.29
.73
.06
.04
,05
.07
.03
.05
.07
±20*
.22
.19
.20
.25
.20
.22
.39
.27
.19
.23
.39
.25
.32
.77
.20
.20
.20
.20
.21
.20
.20
±10*
.12
.10
.11
.18
.14
.16
.37
.19
.13
.16
.37
.24
.30
.77
.10
.10
.10
.10
.10
.10
.11
±5*
.09
.07
.08
.16
.13
.15
.37
.18
.12
.15
.37
.24
.30
.77
.05
.05
.05
.05
.05
.05
.07
   The average fractional error in emission factors resulting from input
   parameter error of ±X may or may not be the simple average of the
   fractional error caused by input parameter error of +X and -X, since
   in some cases different sample sizes were used in computing the +X
   and -X errors.
   Errors of +50 and -50 were grouped together because the sample size
   was small.
from error in one input parameter has the opposite sign to the emission
factor resulting from error in the other input parameter.   For a given
value of error in cold start, the emission factor error decreases as the
magnitude of error in volume decreases.  But for a given amount of error
in volume, the emission error increases as the magnitude of cold-start

error decreases.
                                  69

-------
          The number of cases with cold starts underestimated or overesti-
mated by 50% were few enough that it was felt that these cases should be
combined to be statistically significant.  The behavior of the emission
factor error as the error in volume changes is evident from Table 17.
          The magnitude of emission factor error resulting from error in
both cold start and volume is greatest for CO and least for NO .  The
                                                              X
maximum HC and CO emission factor error occurs when cold starts are ±50%
in error and volume is overestimated by 20%.  The maximum value for HC
is 48% and for CO is 94%.  For NO  the maximum error of 23% occurs when
                                 X
cold starts are underestimated by 207o and volume is overestimated by 20%.
          Speed and Cold-Start Sensitivity Analysis—Table 15 lists the
values of speed, percent cold-start vehicles, calendar year, and tempera-
ture used in the speed and cold-start sensitivity analysis.  Computations
were made for  speed error of +5, -5, ±5, +10, -10, and ±10 mph and for
cold-start error of +10, -10, ±10, +20, -20, ±20, and ±50%.  The average
fractional error in emission factor resulting from all combinations of
the above error classes are given for each pollutant in Table 18.
          The  largest emission factor errors for HC and CO occur when
speed is overestimated and cold  start is underestimated or when speed
is underestimated and cold start is overestimated.  For these cases, for
an error in one input parameter  (speed or cold start), emission factor
increases with increasing error magnitude in the other input parameter.
When speed error and cold-start  error are both underestimated or overesti-
mated, for a given value of cold-start error, emission factor error in-
creases as the magnitude of speed error is increased.  But for a given
amount of speed error, emission  factor error may increase or decrease as
cold-start error magnitude increases.
          For  NO , emission factor error is largest when speed is over-
                X
estimated and  cold start is underestimated, or when speed is underesti-
mated and cold start is overestimated.  For a given amount of error in
speed, emission factor error changes little as cold-start error magnitude
increases.  Since the sample sizes are not large, the differences in emis-
sion factor errors are considered negligible.  For a given value of
                                   70

-------
                                Table 18
       AVERAGE FRACTIONAL ERROR IN FTP EMISSION FACTORS  RESULTING
            FROM ERROR IN BOTH SPEED AND PERCENT COLD  STARTS


Pollutant

HC





CO





NO
X





Error
in Speed
(mph)
+5
-5
±5*
+10
-10
±10*
+5
-5
±5*
+10
-10
±10*
+5
-5
±5*
+10
-10
±10*
Error in Cold Starts
ft \
(/°)
+10
.10
.28
.19
.18
.47
.32
.15
.57
.36
.27
1.02
.65
.04
.06
.05
.08
.09
.08
-10
.19
.12
.16
.29
.28
.28
.31
.19
.25
.44
.51
.47
.06
.04
.05
.09
.08
.08
±10*
.15
.20
.17
.23
.37
.30
.23
.38
.30
.36
.76
.56
.05
.05
.05
.08
.08
.08
+20
.11
.38
.24
.16
.58
.37
.21
.82
.52
.25
1.35
.80
.04
.06
.05
.07
.09
.08
-20
.25
.11
.18
.33
.21
.27
.40
.17
.28
.51
.36
.44
.06
.04
.05
.10
.07
.09
±20*
.18
.24
.21
.25
.40
.32
.30
.49
.40
.38
.86
.62
.05
.05
.05
.08
.08
.08
±50t
.33
.46
.40
.34
.59
.47
.64
1.07
.86
.57
1.47
1.02
.08
.08
.08
.10
.10
.10
       •V
        The average fractional error in emission factors  resulting
        from input parameter error of ±X may or may  not be the simple
        average of the fractional  error caused by input parameter
        error of +X and -X,  since  in some cases different sample sizes
        were used in computing the +X and -X errors.

        Errors of +50 and -50 were grouped together  because the sample
        size was small.
cold-start error, emission factor error increases  as  the magnitude of speed
error is increased.   These two trends  also apply to cases where speed error
and cold-start error are both either underestimated or overestimated.

          The error in CO emission factors resulting from error in speed
and percent of cold-start vehicles is  the largest  error of the three pol-
lutants.  The largest errors in both HC and CO emission factors occur
when cold start is 5070 in error and the speed is underestimated by 10 mph;
                                  71

-------
the maximum values are 59% for HC and 14770 for CO.   NO  error is smaller,
                                                      x                 '
not exceeding 10%.

     3.   Conclusions
          It has been shown that relatively large error in FTP emission
factors are produced when model input parameters are measured or estimated
erroneously.  To illustrate more clearly the effect of input error, an
arbitrary threshold value of 20% was chosen, and a table was made to show
which input parameter error classes or combination of parameter error
classes would produce emission factor error exceeding the threshold value.
Table 19 shows which error classes exceed the threshold value for each
pollutant.
          Cold-start error of 20% or greater produces emission factor
error greater than 20% for CO; for HC, the threshold value is not ex-
ceeded unless cold-start error is about 50%.  NO  emission factor error
                                                X
is less than 2070 even when cold-start error is as large as 50%.
          Error in emission factors caused by temperature error of a mag-
nitude likely to be encountered in RAPS data has been determined to be
negligible; this was discussed in the preceding section.
          For HC and CO, most of the speed-error classes produced errors
exceeding the emission threshold value.  However, NO  emission factors
                                                    X
were in error by less than 207o for all of the speed-error values.
          Error in both volume and speed cause the HC emission factor
error to exceed the threshold value for several combinations of values
of speed error and volume error, as seen in Table 19.  CO emission factors
are slightly more sensitive than the HC factors.  The only error classes
causing NO  emission factor error to exceed the threshold value are those
          X
with 20% volume error.
          The results of the volume and cold-start analysis appear in
Table 19.  They are very similar to those described above for the volume
and speed analysis.
          The HC emission factor error resulting from error in both speed
and cold start exceeds the threshold value for many of the combinations
                                  72

-------
                      Table 19
INPUT PARAMETER ERROR VALUES THAT CAUSE FTP EMISSION
             FACTOR ERROR TO EXCEED 20%

Pollutant

HC
CO
NO
X
Error in Cold Start
(%)
+10




-10




+20

X


-20

X


±50
X
X



Pollutant

HC
CO
NO
X
Error in Minor Arterial Speed
(mph)
+5

X


-5
X
X


+10
X
X


-10
X
X



Pollutant
HC



CO



NO
X

•
Error
in Speed
(mph)
+5
-5
+10
-10
+5
-5
+10
-10
+5
-5
+10
-10
Error in Volume
(/o)
+5

X

X

X
X
X



-5


X
X
X
X
X
X



+10

X

X

X
X
X



-10
X

X
X
X

X
X



+20

X

X

X
X
X
X
X

-20
X

X

X

X
X


X
                        73

-------
Table 19 (Concluded)
Pollutant
HC




CO




NO
X



Error in
Cold Start
(°l \
\i°)
+10
-10
+20
-20
±50
+10
-10
+20
-20
±50
+10
-10
+20
-20
+50
Error in Volume
("l\
\i°)
+5


X

X
X

X

X





-5




X


X
X
X





+10


X

X
X

X

X





-10



X
X

X
X
X
X





+20
X

X

X
X

X

X

X

X

-20

X

X
X

X

X
X
X

X



Pollutant

HC



CO



NO
X


Error
in Speed
(mph)
+5
-5
+10
-10
+5
-5
+10
-10
+5
-5
+10
-10
Error in Cold Start
(°i \
\l°)
+10

X

X

X
X
X




-10


X
X
X

X
X




+20

X

X
X
X
X
X




-20
X

X
X
X

X
X




±50
X
X
X
X
X
X
X
X




        74

-------
of speed error and cold-start error.   CO emission factors are in error by
more than 20% for almost all combinations of speed error and cold-start
error tested.  However, NO  error was under the threshold value for all
                          X
error classes tested.
          In general, CO emission factors are most sensitive to error in
input parameters, and NO  emission factors are the least sensitive to
                        X
errors in input parameters.  HC emissions are somewhat less sensitive
than CO emissions, but are considerably more affected by input parameter
error than are NO  emission factors.
G.   Modal Model Sensitivity Analysis
     The input parameters required by the modal model are listed in Ta-
ble 5.  The modal model sensitivity analysis was centered on assessment
of emission factor error introduced by error in each of three input param-
eters:  temperature, percent of cold-start vehicles, and traffic volume.
Based on the model sensitivity to error in a single input parameter, an
analysis of emission factor error introduced by error in two parameters
was performed.
     It was assumed that calendar year, vehicle type mix, and model year
mix could be accurately assessed.  Assessment of the effects of error in
the second-by-second speed and time profiles was considered beyond the
scope of this study.

     1.   Sensitivity to Error in a Single Input Parameter
          Since modal temperature and cold-start correction factors are
not dependent on vehicle model year, the effect of error in either of
these parameters on the emission factor can be assessed independent of
the value of the uncorrected emission factor.  This is also true for error
in traffic volume.  The following equations illustrate the independence:

                                 I aE - a'E  |                        ._,_
                        error =  I - aE -  I   '                   (21)
                                   75

-------
Reducing,

                                       *3   I
                         error =   1 - —  |                        (22)

where
          a = The correction factor for the correct temperature
              or cold start or the correct volume
         a  = The correction factor for the erroneous temperature
              or cold start or the erroneous volume
          E = The uncorrected modal emission factor.
Equation (22) was used to assess the sensitivity of modal emission factors
to error in temperature, cold start, and volume.

          Temperature-Sensitivity Analysis--All combinations of the values
of temperature and percent of cold-start vehicles listed in Table 20 were
used in the temperature-sensitivity analysis.  For each pollutant, the
sensitivity of the emission factors to errors of +2, -2, ±2, +1, -1, and
±1 degree Farenheit was assessed; the results appear in Table 21.  Since
temperature error is not likely to be greater than 2°F (1°C), and the
maximum emission factor error computed was 3%, it appears that modal model
sensitivity to error in temperature is not significant.

          Cold-Start Sensitivity Analysis—Table 20 lists the percent of
cold-start vehicles and the temperatures used in the modal cold-start
sensitivity analysis.  For each pollutant, the average errors in emission
factor resulting from cold-start error of +10, -10, ±10, +20, -20, ±20,
and ±50% were computed.  In general, the greater the magnitude of the
cold-start error, the greater the error in emission factors.  The maximum
error occurs when cold-start is 50% in error; the error value for HC is
49%, for CO it is 82%, and for NO  it is 13%.

          Traffic Volume Sensitivity Analysis—The effect on emission
factors of error in traffic volume is the same for both the modal and
the FTP emission models.  See the discussion for the FTP model in Sec-
tion VII.F.
                                  76

-------
                               Table 20
           VALUES OF INPUT PARAMETERS ASSUMED IN MODAL MODEL
                          SENSITIVITY ANALYSES
     Parameter   Units                Values Assumed
                   Temperature Sensitivity Analysis
    Temperature    °F    20,21,22,28,29,30,31,32,38,39,40,41,42,48,
                        49,50,51,52,58,59,60,61,62,68,69,70,71,72,
                        78,79,80
    Cold starts    %     0,20,40,60,80,100
                    Cold-Start Sensitivity Analysis
    Cold starts   %     0,5,10,15,20,25,30,35,40,45,50,55,60,65,70,
                        75,80,85,90,95,100
    Temperature   °F    20,30,40,50,60,70,80
               Cold-Start and Volume Sensitivity Analysis
    Cold starts   %     0,5,10,15,20,25,30,35,40,45,50,55,60,65,70,
                        75,80,85,90,95,100
    Temperature   °F    20,30,40,50,60,70,80
     2.   Sensitivity to Error in Two Input Parameters
          As was the case in the FTP model temperature sensitivity analy-
sis, the modal model is relatively insensitive to error in temperature.
Therefore, emission factor sensitivity to temperature error in conjunction
with error in another input parameter was not considered.   Therefore, the
modal model two parameter sensitivity analysis was performed only for
error in both cold start and volume.  The following equation describes
the emission factor error:

                           I  vaE - (v + v ') a 'E i
                   error = |  	>—	 |    .              (23)

Reducing,

                   error = | 1 - ^ (l + ^ )  I                     (24)

                                   77

-------
                                Table  21
      AVERAGE  FRACTIONAL  ERROR  IN MODAL EMISSION FACTORS RESULTING
                 FROM ERROR IN  A SINGLE INPUT  PARAMETER

Pollutant

HC
CO
NO
• x
Error in Ambient Temperature
(°F)
+2
.03
.03
.01

-2
.02
.03
.01

±2*
.03
.03
.01

+1
.01
.02
.00

-1
.01
.01
.00

±1*
.01
.01
.00



Pollutant

HC
CO
NO
X
Error in Percent Cold-Start Vehicles
ft \
('<>)
+10
.11
.20
.03

-10
.08
.12
.03

±10*
.10
.16
.03

+20
.22
.39
.05

-20
.16
.24
.05

±20*
.19
.31
.05

±50t
.49
.82
.13

      The average fractional error in emission factors resulting from
      input parameter error of ±X may or may not be the simple average
      of the fractional error caused by input parameter error of +X
      and -X, since in some cases different sample sizes were used in
      computing the +X and -X errors.
      Errors of +50 and -50 were grouped together because the sample
      size was small.
where
          a = The correction factor for the correct percent of
              cold starts
         a  = The correction factor for the erroneous percent of
              cold starts
          E = The temperature-corrected emission factor
          v = The correct volume
         v' = The amount by which the volume is in error.
Again, the emission factor error is independent of the value of the
temperature-corrected emission factor.
                                   78

-------
          The sensitivity of modal emission factors was  computed for all
combinations of cold-start error of +10,  -10,  ±10,  +20,  -20,  ±20,  and ±50%
and for volume error of +20, -20, +10,  -10, +5,  -5, ±20, ±10,  and  ±5%.
The results of the analysis appear in Table 22.
                               Table 22
     AVERAGE FRACTIONAL ERROR IN MODAL EMISSION FACTORS RESULTING
               FROM ERROR IN BOTH VOLUME AND COLD-START



Pollutant

HC






CO






NO
X






Error in

Cold Start
(%)
+10
-10
±10*
+20
-20
±20*
±50 1
+10
-10
±10*
+20
-20
±20*
±50 1
+10
-10
±10*
+20
-20
±20*
±50t
Error in Volume
/at \
( L )
\'°/
+20
.33
.11
.22
.47
.07
.27
.59
.43
.08
.26
.67
.10
.38
.98
.17
.23
.20
.14
.26
.20
.22
+10
.22
.04
.13
.34
.08
.21
.54
.31
.04
.18
.53
.16
.35
.90
.07
.13
.10
.05
.16
.10
.15
+5
.17
.04
.10
.28
.12
.20
.51
.26
.07
.16
.46
.20
.33
.86
.02
.08
.05
.02
.11
.06
.14
-5
.06
.13
.09
.16
.20
.18
.46
.14
.16
.15
.32
.28
.30
.77
.07
.03
.05
.10
.02
.06
.12
-10
.05
.17
.11
.11
.25
.18
.44
.09
.21
.15
.25
.31
.28
.73
.12
.08
.10
.15
.05
.10
.13
-20
.12
.26
.19
.10
.33
.21
.39
.11
.30
.20
.15
.39
.27
.65
.22
.18
.20
.24
.16
.20
.19
±20*
.23
.19
.21
.28
.20
.24
.49
.27
.19
.23
.41
.24
.33
.82
.20
.21
.20
.19
.21
.20
.20
±10*
.14
.11
.12
.23
.16
.19
.49
.20
.12
.16
.39
.24
.31
.82
.10
.10
.10
.10
.11
.10
.14
±5*
.11
.08
.10
.22
.16
.19
.49
.20
.12
.16
.39
.24
.31
.82
.15
.05
.05
.06
.06
.06
.13
  The average fractional error in emission factors resulting from input
  parameter error of ±X may or may not be the simple average of the
  fractional error caused by input parameter error of +X and -X, since
  in some cases different sample sizes were used in computing the +X
  and -X errors.
  Errors of +50 and -50 were grouped together because the sample size
  was small.
                                  79

-------
          For HC and CO, the largest emission factor error results when
cold start and volume are both either underestimated or overestimated.
For these cases, for a given value of error in one input parameter (cold
start or volume), the emission factor error increases as the magnitude
of the error in the other input parameter is increased.  When cold starts
are underestimated and volume is overestimated or when cold starts are
overestimated and volume is underestimated, for a given value of error
in one input parameter, the emission factor may (1) increase, (2) decrease,
or (3) decrease and then increase as the magnitude of the error in the
other input parameter is increased.  As in the FTP sensitivity analysis,
this is a result of a varying degree of cancellation of the error intro-
duced by error in the two input parameters.
          For NO , the largest emission factor error results when volume
                X
is overestimated and cold starts are underestimated, or when volume is
underestimated and cold starts are overestimated.  For these cases, for
a given amount of error in one input parameter, emission factor error in-
creases as the magnitude of the error in the other input parameter is
increased.
          When errors in cold starts and volume are both either underesti-
mated or overestimated, for a given amount of cold-start error, the emis-
sion factor error increases as the magnitude of volume error increases.
But for a given amount of volume error, the emission factor error may de-
crease or remain constant as the magnitude of cold-start error is in-
creased.
          For HC and CO, the largest emission factor error occurs when
volume is overestimated by 20% and the percent of cold starts is in error
by 50%.  The maximum HC error is 59%, and the maximum CO error is 98%.
The maximum NO  emission factor error of 26% occurs when volume is over-
              x
estimated by 20% and cold starts are underestimated by 20%.
          The sample of cases having cold-start error of 50% was small,
so both underestimated and overestimated cases were combined.  It is
probable that error resulting from a larger sample of cases having either
50% overestimated or 50% underestimated cases alone may be higher than
                                   80

-------
the error computed when all 50% cases are combined.   Therefore, worst-
case error would probably occur for cases with 50% overestimated or under-
estimated cold-start error.

     3.   Conclusions
          As was seen in the preceding sections, emission factor error
introduced by error in some input parameters can be appreciable, partic-
ularly for HC and CO.  There is considerable variation between pollutants
as to the magnitude of error produced by various input parameter errors
and as to what value or pair of values of input parameter errors produce
a given level of emission factor error.  As in the FTP model analysis,
an arbitrary value of emission factor error (20%,) , was selected as a
threshold error value.  Table 23 shows what values of input error will
produce emission factor error exceeding the threshold value.
          For HC and CO, 10% cold-start error does not exceed the thresh-
old, but for the most part, 20 and 5070 cold-start errors do exceed it.
However, as much as 50% cold-start error does not produce more than 20%
NO  emission factor error.  The preceding sections showed that error in
  X
temperature produces negligible emission factor error.
          For HC, any of the chosen values of cold-start error in com-
bination with at least one of the volume error classes exceeds the thresh-
old, and vice versa.  There are even fewer combinations of error classes
that do not produce more than 20% CO emission factor error.  NO  emission
                                                               A
factor error is  less than 20% for most combinations of error classes, ex-
cept for those with 20% volume error.
                                  81

-------
                       Table 23
INPUT PARAMETER ERROR VALUES THAT CAUSE MODAL EMISSION
              FACTOR ERROR TO EXCEED 20%

Pollutant

HC
CO
NO
X
Error in Cold Start
(%)
+10




-10




+20
X
X


-20

X


±50
X
X


Pollutant
HC




CO




NO
X



Error in
Cold Start
(°/\
\'o}
+10
-10
+20
-20
±50
+10
-10
+20
-20
±50
+10
-10
+20
-20
±50
Error in Volume
(%)
+5


X

X
X

X

X





-5




X


X
X
X





+10
X

X

X
X

X

X





-10



X
X

X
X
X
X





+20
X

X

X
X

X

X

X

X
X
-20

X

X
X

X

X
X
X

X


                          82

-------
        VIII EVALUATION OF OTHER TYPES OF SOURCES OF EMISSIONS

A.   Fugitive Dust
     Fugitive dust is perhaps the most difficult type of emission to quan-
tify.  Midwest Research Institute (Cowherd and Guenther, 1976) compiled
the inventory for the following sources:   (1)  unpaved roads, (2) agri-
cultural land tilling, (3) wind erosion of agricultural land, (4) con-
struction sites, (5) aggregate storage piles,  and (6) unpaved airstrips.
Rockwell International (Littman and Isam, 1977) has estimated fugitive
dust emissions from paved roads.
     For each of the over 2000 area sources, data were compiled on annual
emissions of fugitive dust.  Hourly apportioning factors were derived to
account for emission variations by hour of the day, day of the week, and
season of the year.  Total particulate emissions (particles smaller than
30 urn) from categories (1) through (6) listed  above, were about 1,183,000
tons/year.  It was assumed that 2070 of the emissions involved particle
sizes smaller than 5 (J.m.  Therefore, for these smaller particles, TSP
emissions are about 237,000 tons/year.  (The fugitive dust estimated in
Table 1 include all particle sizes.)
     As can be seen in Table 1, fugitive dust  emissions dominate the par-
ticulate inventory.  Dust from unpaved roads and wind erosion of agri-
cultural land is the major source of fugitive  dust emissions.
     Midwest Research Institute (MRI)  estimated relative errors for each
of the source categories.  These are represented in Table 24.  These errors,
corresponding to a 9070 confidence level, were  determined by a progressive
analysis of errors associated with each calculation step.  One could con-
clude that estimates for paved roads should be somewhat better than those
for unpaved roads.  Traffic volumes and vehicle speed should be much better
known for paved roads.  In view of all the uncertainties, the estimates in
Table 4, appear somewhat optimistic, particularly for the "hourly adjust-
ment factors."
                                   83

-------
                               Table 24
           MRI ESTIMATES OF FUGITIVE DUST INVENTORY  ACCURACY
                       (Estimated Relative Error)
Source Category
Unpaved roads
Agricultural tilling
Wind erosion
Construction
Aggregate storage
Unpaved airstrips
Source
Extent
±5%
±15%
±30%
±35%
±25%
±15%
Corrected
Emission
Factor
±20%
±30%
±20%
±30%
±30%
±25%
Hourly
Adjustment
Factor
±15%
±20%
±15%
±20%
±20%
±20%
     In general, the fugitive dust inventory is probably somewhat less ac-
curate than those inventories that depend on fuel consumption data.   Fur-
thermore, information on the size distribution from fugitive dust sources
is virtually nonexistent.  Some field data should be collected to (1)  give
breakdowns by particle size and (2) verify assumptions inherent in the MRI
methodology.

B.   Aircraft
     The aircraft emission inventory was developed by GCA Corporation
(Patterson et al., 1974).  Even in the Lambert Field area, aircraft  (and
support vehicles) account for only modest emissions of the criteria  pol-
lutants.  The maximum emissions found in any grid within the Lambert Field
vicinity is 793 tons per year of carbon monoxide (grid 523).  Emissions
for the other pollutants were less.  For the area as a whole, aircraft
emissions account for a very small portion of the emissions burden (less
than 1%).
     Sources of emissions include airctaft operation, engine maintenance
testing, ground support vehicles, and fuel storage and handling.  Three
types of airports were considered:  municipal, military, and civilian.
For each airport, the number, type, and operating patterns of the aircraft
                                   84

-------
were considered.  Ground support vehicle and fuel  handling emissions are
related to the volume and type of aircraft activity at  each airport.  GCA
applied aircraft emission rate data which was compiled  by Cornell Aero-
nautical Laboratory.  These data are summarized in AP-42.  Various modes
of aircraft operation were considered:   idle, taxi, take-off,  landing,
climb-out, and approach.
     The number and type of aircraft operating out of each airport is
fairly well known.  For Lambert Field,  scheduled airline data  are known
accurately.  GCA gathered data from the Official Airline Guide, the Fed-
eral Aviation Authority, and the airport managers  office.  Similar data
are available from other airports.   The number and types of sources gen-
erally are quite well known.  The validity of the  emission factors would
appear to be the most questionable item.
     In general, the aircraft sources are specified with better accuracy
than other mobile sources, with the possible exception  of railroads.  For
our purposes, we will assume negligible bias and a precision of unity.

C.   Off-Highway Mobile Sources
     The off-highway mobile source emission inventory was compiled by
Rockwell International (Littman and Isam, 1977), using  in part methodolo-
gies developed by Hare and Springer (1974).  Six categories have been
calculated for the 1989 RAPS grid squares.  These  categories included:
motorcycles, lawn and garden equipment, industrial equipment,  construction
equipment, farm equipment, and outboard motorboats.  From Table 1, it can
be seen that all categories, except motorcycles, account for modest quan-
tities of pollutants (CO, NO  and HC) normally attributed to mobile sources
                            X
     Temporal apportionments were made on the basis of  seasonal activity
and normal daily operating estimates.  Emissions were distributed uni-
formly over the operating period.  For instance, industrial equipment was
assumed to operate year round from 8:00 a.m. to 6:00 p.m.; lawn and gar-
den equipment was assumed to operate from April through September, 9:00
a.m. to 7:00 p.m.
                                  85

-------
     The primary problem is lack of accurate background data on which  to
base emission calculations.  Although emission factors are known to be
within typical precisions (0.1 to 0.3), the number, type, and operating
patterns of most off-highway equipment are not known well.  Each type,
except motorcycles, are considered separately below.

     1.   Lawn and Garden Equipment
          The lawn and garden equipment category includes riding mowers,
walking mowers, garden tractors, and motor tillers.  There are registra-
tion data to quantify numbers of these devices.  Therefore, Rockwell
acknowledges that emission figures at the grid level are meant solely  to
give an "order-of-magnitude" calculation.
          The emission estimates were based on U.S. Census data on the
distribution and types of houses, emission factors, an assumed linear
correspondence between one-unit housing structures and small utility
engines, and a seasonal utilization factor.  While this approach does
serve  to quantify emissions, the bias and precision could be quite high.
In absence of any verification data, we will assume a bias of 0.5 and
precision of 2.0.

     2.   Construction Equipment
          The major sources of data on construction equipment are national
figures on units shipped per year, annual usage, total horsepower in use,
load factors, and duty cycles.  Rockwell apportioned national data to the
RAPS area based on a ratio of national-to-state construction volumes and
state-to-county populations.*
          Composite nationwide emission  factors for an assumed distribu-
tion for each of ten construction categories were  compiled for the nation
as a whole and then apportioned to the RAPS area in the manner based on a
ratio  of national-to-state construction volumes.   Grid emissions were
 •
 Illinois and Missouri  construction volumes were not available by the
 county.

                                   86

-------
based on a ratio of grid construction acreage to county construction
acreage.  The accuracy of the construction inventory should be comparable
to that for lawn and garden equipment.

     3.   Industrial Equipment
          The industrial equipment category includes fork lifts,  utility
carts, small tractors and wheel loaders,  quarrying machinery,  portable
generators, and so forth.  Rockwell categorized these devices  into small-
utility and heavy-duty engines.  The methodology for industrial equipment
is similar to that used for construction  equipment.   National  emissions
data serve as the starting point.   County emission estimates were derived
from a ratio of national industrial activity to county industrial activity,
Grid apportionments were based on locations of industrial plants  on a grid
basis.  Again, accuracies are probably comparable to the other off-highway
mobile source inventories.

     4.   Farm Equipment
          Farm equipment includes farm tractors, garden tractors, self-
propelled combines, forest harvesters,  balers, irrigation pumps,  and aux-
iliary engines.  Background data on equipment population was available
through 1969 Census of Agriculture estimates.  Based on EPA usage esti-
mates and emission factors, county emissions were compiled.
          Emissions were then appropriated by grid based on farm acreage
data prepared by MRI in compiling the fugitive dust inventory.  In gen-
eral the farm equipment inventory is more accurate than the construction
and industrial equipment inventory.

     5.   Outboard Motorboats
          From Table 1 it can be seen that emissions are modest for two
pollutants (HC and CO) and negligible for others.  Emission factors are
low for typical outboard operation because a certain portion of the ex-
hausts are removed in the water (exhaust  outlets are normally  below water
level).  Boat registration data was the basis for calculations.  Annual
                                  87

-------
usage and EPA emission factors were applied to compile the regional in-
ventory.  Spatial apportionments were made on the basis of navigable water
ratios  (county-to-state, grid-to-county).  The accuracy of this inventory
is about the same as that for industrial equipment.

     6.   Summary
          As stated in the Rockwell report, this inventory provides "order-
of-magnitude" estimate only.  Also, the applied background data in many
cases is several years old.  Therefore, the overall accuracy is much a
question mark.  For our purposes, we will assume an overall bias of
±0.5 T and precision of 2.

D.   River Vessels
     Emissions from river vessels are dominated by towboat traffic.  Esti-
mates of emissions from towboats have been prepared by the Department of
Transportation (Sturm, 1976).  As shown in Table 1, emissions from river
vessels are almost insignificant, accounting for less than one percent of
the total.  The DOT report accurately quantifies the yearly estimates,
based on current emission factors.  However, the report concludes that
hourly  emissions are not worth the effort to accurately quantify, given
their low contribution to the total.  Therefore, in this report, we will
assume  a bias of T and coefficient of precision of 2.

E.   Railroads
     Railroad emissions are more substantial than river vessel emissions,
but do  not contribute signficantly to the total.  They are large enough
to be observable locally in the vicinity of railroad tracks during active
periods.  Railroad emissions were compiled by Walden Research (Wiltsee
et al., 1977).  Walden developed separate methodologies to compile emis-
sions from two types of rail activity:  line-haul-operations and switch
yard activity.  The study inventoried emissions only from diesel locomo-
tive operations since other type emissions were deemed insignificant.
                                  88

-------
     Features of the Walden methodology include:
     •  Department of Transportation (DOT)  compilations  of:
        - Routing, runtime, and locomotive  information for each
          train in the study area
        - Total active and idle hours and locomotive information
          for each rail yard in the AQCR
        - Interyard transfer routing and runtime.
     •  A system of links simulating the rail network.
     •  Classification of locomotives into  the five engine cate-
        gories specified in AP-42.
     •  Derivation of active and idle load  factors  for switch  and
        road locomotives.
     •  Characterization of a "typical" transfer  engine  used in
        the St. Louis AQCR.
In general, the Walden methodology  appears  thorough in estimating annual
emissions and spatial apportioning.  The main inaccuracy appears to be that
of time distinction.  In general, this problem could be  overcome by  col-
lecting minimal data on diurnal variation.   On an annual basis,  the  rail-
road inventory should be somewhat better than conventional highway sources.
For our purposes, we will assume that bias  is negligible and coefficient
of precision is 0.75.
F,   Separation of Hydrocarbon Emissions into Classes
     Once total hydrocarbon emissions were determined for each source,
a classification scheme was applied to estimate the proportion of these
emissions contributed by the following classes:  nonreactive, paraffins,
olefins, aromatics, and aldehydes (Griscom, 1977).   Emission factors by
hydrocarbon type were established for each source category.   These fac-
tors were based on an earlier study which consisted of compositional anal-
yses on organic emissions in Los Angeles (Trijonis  and Arledge, 1975).   As
Rockwell noted, there were three limitations in applying the Los Angeles
results to St. Louis:
     (1)  The classification scheme (e.g., olefins) required for
          RAPS was somewhat different than that used in the Los
          Angeles study.
                                  89

-------
     (2)  The compositions of petroleum products and solvent  usage
          may be different between Los Angeles and St.  Louis.
     (3)  Certain source types existing in St. Louis were not con-
          sidered in the Los Angeles study.
     In spite of the above limitations, Rockwell was able to  make some
reasonable assumptions by augmenting the Los Angeles results  with those
from a host of other studies.  (See the Griscom report, 1977, cited  above.)
     For point sources, emission factors, by hydrocarbon type, were  de-
veloped for each SCC.  Area sources were generally described  according to
the source types listed in Table 1.  One exception is the highway sources,
which were separated into 1DV, HDG, and HDD classes.  (Separate  factors
for exhaust and evaporative losses were used for LDV and HDG.)  The  ac-
curacy of the emission factors, by hydrocarbon class, cannot  reliably be
estimated since verification data are not available.  In view of the
limitations stated above and assumptions made in both the Griscom and
Trijonis studies, it is clear that the accuracy in the emission  estimates
by class is substantially less than that for total hydrocarbons.
                                   90

-------
                       IX  COST BENEFIT ANALYSIS

     The Lack of field validation data precludes  a completely objective
analysis of cost and benefits.   However, based on the background  and  in-
formation presented in our previous discussion on bias and  coefficient  of
precision, we can make some relative estimates.   These estimates  in turn,
can be used to provide insight  on (1)  the effectiveness of  monies already
spent and (2) where future improvements can best  be applied.
     The emissions inventory is summarized in terms of:  emission totals,
estimated accuracy, and costs.   The totals are derived from Table 1.  The
estimated accuracy relates to combined coefficient of precision and bias
estimates previously discussed  in the various emission evaluations.  The
                                    &
cost estimates were provided by EPA.   The total  inventory  budget, ex-
clusive of the data handling system, was about $870,000.  This was divided
as shown in Table 25.
     The classification of the  inventory by pollutant is  shown in Table 25.
Exclusive of TSP, the point source inventory made up 49.8%  of all the to-
tals; the highway inventory made up 37.970; and so forth.  The accuracy
ranking shows that by far the point source inventory is the most  accurate
with an average coefficient of  precision-bias of  0.275.  (This may be
taken generally to be the sum of the estimated coefficient  and one-half
the bias.)  The most outstanding figure corresponds to the  accuracy of
the highway (mobile source) inventory.  It is about the same  as it is for
those inventory types (airports, railroads, and so forth) contributing
much less to the total inventory budget.  This is the most  outstanding
inconsistency in the accuracy of the inventory.   (Accuracy  should be  pro-
portional to percent of the emission inventory budget.)
 Masser, C. (private communication, July 1977)

                                   91

-------
                                                                   Table  25


                                                INVENTORY  SUMMARY:   TOTALS,  ACCURACY, AND  COST
Inventory
Point Source
Highways
Stationary,
residential and
commercial
Stationary,
industrial
Off highway
mobile
Railroads
River vessels
Airports
Fugitive dust
Percent of Total Emissions
SO
X
96.7
0.3
2.5
_ —
0.2
0.2
--
--
--
NO
X
69.6
17.5
3.0
„ _
6.0
2.9
0.8
0.1
--
CO
9.7
78.9
0.9
— —
9.9
0.3
0.1
0.2
--
HC
23.3
54.8
10.7
__
8.0
2.1
0.3
0.7
--
Particulate*
3.3
0.5
0.5
. _
0.2
0.1
--
--
95.4*
(Avg)4t
49.8
37.9
4.3
0.0
6.0
1.4
0.3
0.3
--
Accuracy Ranking
SO
X
0.20
1.0
0.75
« _
2.0
0.75
2.0
1.0
--
NO
X
0.30
1.0
0.95
— _
2.0
0.75
2.0
1.0
--
CO
0.30
1.0
0.95
_ —
2.0
0.75
2.0
1.0
--
HC
0.30
1.0
0.95
.._
2.0
0.75
2.0
1.0
--
Particulate*
0.'30
1.0
0.95
__
2.0
0.75
2.0
1.0
--
(Avg)4+
0.275
1.0
0.90
- -
2.0
0.75
2.0
1.0
--
Order*
1
4
3
_ _
_ _
2
5
4
--
Percent
Cost
44.4
25.3
6.9
2.6
4.6
2.9
2.9
7.5
3.4
NJ
         For information only;  these figures  are not used  in calculating  averages.


         (Avg)4 denotes the average of the four gaseous  pollutants  (SO  ,  NO  ,  CO, and HC).
        .                                                             XX

         The inventories (e.g., point source, highways)  are  ordered from  1 to  5  from the  most accurate to the least accurate.

-------
     To an extent, the same type of inconsistency appears  in the percent
cost category.  For the smaller inventories,  anywhere from 3.4 to 6.970
of the budget was expended.  This is reasonable since it requires a mini-
mal amount (about $30,000) to quantify any of these smaller inventories.
Although their individual contributions might be negligible, this would
not be known until the study has been completed.  The highway inventory,
however, constitutes almost 4070 of the emissions budget (excluding TSP) ,
but only about 257o of the cost budget.  If we are to assume that the ac-
curacy of the point source inventory is adequate, than a similar accuracy
should be the goal of future highway inventories.
                                   93

-------
              X  EFFECTS OF EMISSION PARAMETER INACCURACIES
                     ON AIR QUALITY MODEL PREDICTIONS
     The objective of the analysis in this section is the characterization
of the effects of emission parameter errors when they are used as input
to air quality simulation models.   In the analysis to follow,  steady-state
Gaussian diffusion models are used.   Of course,  the final RAPS models will
be more sophisticated.  However,  the results generated herein are easily
extended to these more sophisticated models within the accuracy constraints
of the analytical techniques applied.

A.   Quantification of Errors in  the Meteorological Input
     Parameters
     One of the primary objectives of this work  was to determine the sensi-
tivity of the air quality model output as a function of various emission
inputs in relation to the sensitivity as a function of the meteorological
input parameters.  Quantifying potential sources of error in this relative
manner allows the air-quality modeler to gain a  better understanding of
the limitations on model performance and identifies those types of input
data that need to be improved.
     The air quality model output X can be represented as some function
of emissions input Q and meteorological input G.  Thus,
                                                                    (25)
where X, Q, and G are vectors in time (t)  and space (s).   Any element in
the X array is sensitive to variations of  any element in both Q and G
arrays and to errors in the measurement of these elements.  In this sec-
tion we examine the accuracy of measuring  the meteorological elements
that are used to derive meteorological input parameters for the models.
                                  95

-------
     In the sensitivity analyses to follow, Gaussian, plume models are
used.  Accordingly, our discussion here relates to input parameters for
Gaussian plume models.  However, we do not imply here that the Gaussian
plume formulation will appear in the final versions of any of the RAPS
models.  In fact, a RAPS modeling objective is to improve on the Gaussian
model.  Nevertheless, we believe that the utility of the formulation in
our sensitivity analysis will produce representative values.
     Most Gaussian plume models that are used to analyze pollutant con-
centrations in an .urban environment require the following meteorological
parameters as input:
     •  Wind speed and direction
     •  Diffusion parameters, both crosswind (cr ) and vertical (a )
     •  Vertical mixing ceiling (height).
     Major differences among model applications pertain to the degree of
detail of meteorological input data and the averaging time periods of the
data.  For example, whereas some models hypothesize the existence of mean
meteorological conditions, others use the joint frequency distribution of
meteorological parameters, including wind direction, speed, and stability
class.  At least one model (APRAC-1A) uses wind data from airport weather
stations and estimates mixing layer ceilings from the nearest radiosonde
data and maximum afternoon temperature.
     The meteorological measurements needed to determine the meteorologi-
cal input parameters in air quality models may be listed as:
     •  Wind speed and direction
     •  Solar radiation
     •  Cloud cover and types
     •  Vertical temperature profiles/height of mixing layer.
     As mentioned earlier, the wind measurements are used directly as
input parameters in the air quality models; other parameters are derived.
The accuracy and variations of wind consequently have a direct bearing on
the performance of air quality models.
                                   96

-------
     The airflow over a metropolitan area is a critical factor in air
quality since it is the medium that transports many effluents.  Also,
transport occurs throughout the depth of mixed layer;  therefore, wind
through the planetary boundary layer, rather than only the surface wind,
is a relevant parameter.  The variations of wind over  the RAPS area have
been verified by special field experiments.  It has been found that:
     •  Airflow over RAPS area is modified to heights  well above 300
        meters.
     •  Wind speeds, varied by 20 to 30%, and wind direction by 20 to
        30 degrees near the metropolitan area (Ackerman, 1974).
     These findings have a significant bearing on the  sensitivity analy-
sis of air quality models.  If the accuracies of wind  measuring instru-
ments and of pilot balloon observations are considered, the complexity of
performing the analysis becomes more evident.  For example, it is very
difficult to predict the accuracy obtainable from a pilot balloon ascent
in which the most important source of error is the uncertainty in the
rate of ascent of the balloon.  Even if two theodolites are used, no
great precision should be expected, and it has been estimated that even
the best observations at low altitudes do not give a true wind more
closely than ± 2 degrees in direction and ± 1 mps in speed (Middleton and
Spilhaus, 1953).  These errors are comparable to those that occur in
surface-based instruments, especially the errors in measuring the direc-
tion.
     The mixing ceilings are estimated from vertical temperature profile
measurements.  The instruments used to measure temperatures at network
sites, because of their limited range, are inadequate  for this purpose.
If radiosonde temperatures are used (as in the case of APRAC-1A), then
errors in measurements can play a significant role in  the prediction of
concentrations.   Errors caused by lag of a radiosonde  thermometer at
300 m range from 2°C to 7°C as reported by Middleton and Spilhaus (1953,
Figure 185) .
     The estimated diffusion parameters that depend on measurements of
solar radiation and observations of cloud cover can be affected not only
by instrumental errors, but also by human errors in estimating cloud

                                   97

-------
cover.  Other techniques of estimating these parameters depend on wind
or temperature measurements, and they are sensitive to those errors just
described above.
     The meteorological network and instruments used in the RAPS study
are described by Myers and Reagan (1975).  Besides air quality measure-
ments made at 25 sites, meteorological measurements were also made as
part of the data base for the modeling work.  In addition, solar and sky
radiation, temperature gradients, and turbulence measurements were made
at some sites.
     For the radiation measurements, comparisons were made between urban
and rural, and upwind and downwind.  For the turbulence measurements,
values were obtained for different types of surface roughness and thermal
characteristics.  The vertical temperature difference between 5- and 30-
meter levels was measured.  Table 26 shows some of the pertinent character-
istics, including accuracy, of the instruments used for measuring meteoro-
logical elements.  In addition to the measurements made at the network of
stations, special upper air observations were made by pilot balloon and
radiosonde techniques.

B.   Case Study for a Single Point Source
     Prior to considering the RAPS emission inventory as a whole, a single
source sensitivity analysis is presented.  The purpose of this analysis
is to demonstrate how input data errors change the magnitude and spatial
distribution of the predicted air pollution concentrations.  This insight
will facilitate the interpretation of the regional sensitivity analysis
to follow.

     1.   Problem Formulation
          A case study was conducted for one of the Union Electric power
plants in St. Louis.  The Sioux power plant was selected because its
stack flow maximum values are similar to those of a TVA plant for which
more complete characteristics are available.  A steady-state Gaussian
plume model was used to calculate SO- concentrations downwind of the

                                   98

-------
                                                     Table 26
                 CHARACTERISTICS OF METEOROLOGICAL MEASUREMENTS AT 25 SITE LOCATIONS IN RAPS AREAS
     Meteorological
         E1ement
                   Manufacturer
                      Range
                Starting Threshold
                (i,e., sensitivity)
    Accuracy
     Remarks
vO
vO
     Wind speed
Wind direction

Temperature

Temperature
gradient

Dew point



Solar radiation


Turbulence
     Pressure
                 MRI
MRI

MRI

MRI


EG&G



Eppley Lab


R. M. Young Co.
                 MRI
                 0 to 22 mps
0 to 540 deg
-20 to +50°C
-5 to +5°C


-40 to +50°C
                                       0 to 4 cal
                                         r\
                                       cnr/min

                                       0 to 22 mps
                 914 to 1067 mb
                    0.22 mps
                                                           0.3 mps
                    0.22 mps
                    <0.270
±0.17 mps or 1%
whichever is
greater
±2.5 deg

Not known
Not determined


1°C, normal
0.02%
Range in specific
instances is not
adequate because
winds in excess
of 25 mps occur
in St. Louis
                                                                                             Subject to exces-
                                                                                             sive drift and
                                                                                             malfunction
                                                                                             Meaningful data
                                                                                             for only 30 to
                                                                                             40% of the time
                 Operated only
                 during intensive
                 field investiga-
                 tions of RAPS

-------
Sioux plant for various perturbations in meteorological and emission
inputs.
is:
          In the Gaussian formulation, the ground level concentration (X)
                    X =
                        rr a cr U
                           y z
                                 exp -
                                                          (26)
where:
         U = the mean wind speed, which will be allowed to vary between
             2 and 8 m/s
     cr ,a  = the Pasquill (Slade, 1968) horizontal and vertical disper-
             sion coefficients, a function of downwind distance and
             stability class
         y = the lateral distance to the plume centerline
         Q = the source strength
         H = the effective plume height
The plume rise is a function of the stack flow parameters (exit velocity
and temperature) and the prevailing atmospheric conditions.  The plume
rise formulas chosen were proposed by Briggs (1971, 1972) and are cur-
rently being used in EPA point source models (Turner and Busse, 1973).
     2.
Results
          The maximum steady-state downwind concentrations for nine dif-
ferent atmospheric conditions are illustrated in Figure 8.  The concentra-
tions were calculated for maximum plant capacity (Q = 1760 g/h) and a
receptor located in the lateral centerline of the plume (y = 0).   Reduc-
tions in plant load to half capacity and the effects of the reductions on
the downwind concentrations were then calculated for three sets of exit
temperature (T ) and volumetric velocity (V ) conditions:
     (1)   Both Te and Vg were assumed to be unaffected by the load change
          (i.e., their maximum values were used).
     (2)   The maximum value for Te was used; Ve was made directly propor-
          tional to the load (one-half maximum in this case).  This
          technique is referred to as the "proposed RAPS" method.
                                   100

-------
                                     STABILITY INDEX • 3
   0.07
   0.06
   0.05
u
   0.03
   0.02
   0.01
  WIND SPEED
    2 m/s
    5 m/s
    8 m/s
      LINE TYPE
      solid
      small desh
      med dash
           (a)
                         ,   I .... I
       O.I
0.2
0.5      I       2          5
        DISTANCE —  kilometers
to
20
50
     * Note:  The Greek symbol, Xf used throughout the text  was not an allowable output

             character  in the computer-generated plots for  Figures 8,  9, and 10.  Hence,

             the phonetic term "CHI" was used to represent  the concentration of S02

             in ambient air.



   FIGURE  8    DOWNWIND CONCENTRATIONS FROM UNION ELECTRIC COMPANY SOUIX

               PLANT  AT VARIOUS STABILITIES
                                         101

-------
      5TAB1UTY INDEX • 4
0.020

0.015
a
a
£ 0.010
o


0.005

0
1 ' ' 1 ' ' ' ' 1 1 ' ' 1 ' ' ' ' ) I ; '
WIND SPEED UNE TYPE / \
- 2 m/s solid / \ _
5 m/s small dash ' x
8 m/s med dash / \
t i
i /
* /
/ /
• t
~~ / /' *~
/ i
/ /
/ •
.. » i _
i '
* i
/ '
» »
* i
' i
— ' / —
/ •
" <") / / .;
1 .. 1 .... 1 1 • . /^-r'.-.'f f ^^T
.1 0.2 0.5 1 2 5 10 20 5
DISTANCE — kilometers
FIGURE 8   (Continued)
        102

-------
                                 5TASIUTY INDEX » 5
  0.025
  0.020
  0.015
x
o
  0.010
  0.005
WIND SPEED
  2 m/s
  5 m/s
  8 m/s
LINE TYPE
solid
small dash
med dash
        '  (c)
O.I
             0.2
        0.5
   I       2
  DISTANCE -
      5
kilometers
                           FIGURE 8   (Concluded)
     (3)  Both Te and Vg were  assumed  to follow a relationship of the
          form A-L + B, where  L  is  the load,  and A and B are the pro-
          portionality and  offset  conditions, respectively.  The actual
          numbers were taken from  a relationship empirically derived
          for a similar Tennessee  Valley Plant in Kingston, Tennessee
          (Ruff et al., 1976).

          The actual numbers are summarized in Table 27.  The Gaussian

model was run for a few different  atmospheric conditions, using the three
sets of flow parameters shown.   Figure 9 shows the downwind concentrations

that result from different  flow  parameter selection strategies.
                                   103

-------
                                Table 27
                     FLOW PARAMETERS FOR THREE CASES
               (Union Electric, Sioux Plant--5070 Capacity)
Case
Maximum
Proposed RAPS
Actual
(modeled from TVA data)
Ve(m3/s)
483
236
277
Te(°K)
435
435
416
          Clearly, use of the maximum values provides unacceptable errors
                    *
of 100% and greater.   The RAPS technique produces maximum errors on the
order of 15% for this case.  This low figure is a result of offsetting
errors, in that T  is overestimated while V  is underestimated.  If the
                 e                         e
errors from either parameter are considered separately, the concentration
error will increase by a factor of about two.
          To place the errors just described in perspective with those of
meteorological origin, perturbations in the Gaussian diffusion coefficients
(a ,  a ) were considered.  These coefficients are generally derived from
  z   y
primary meteorological measurements, such as vertical temperature or
velocity profiles, solar raidation and wind speed, or wind fluctuations.
While the RAPS data base is rich in such measurements, the current state
of the art permits generalizing such measurements into only six broad
categories of atmospheric stability.  In our analysis, we allow a devia-
tion of one-quarter of the difference between category values as being
representative of a coefficient standard deviation (typical error).  We
feel  such an approach in estimating likely errors is a reasonable one
and allows for improvements in the technique of estimating such parameters.
This  statement holds for both the current Gaussian models and the more
complex K-theory formulations.
 Errors in the context here are comparable to a standard deviation of cal-
 culated values divided by a reference (true) value expressed in percent.
                                   104

-------
  0.035
  0.030
  0.025
|o.020
X 0.015
o
  0.010
  0.005
                               STABILITY  INDEX  •=  3



                               WIND SPEED 2.0 m/s
         i    •      i • rTwi        i

         FLOW PARAVETER    LINE TYPE
            actual           solid  ,  ._
            maximum        small jdash
            proposed        med dash
          (a)
O.I
              0.2
                         1 ,  , , , !
                                  I
0.5      I       2         5

        DISTANCE — kilometers
10
20
50
   FIGURE 9   SENSITIVITY OF CONCENTRATIONS TO FLOW PARAMETER  VARIATION

              (Souix Plant)
                                     105

-------
o
   0.05
   0.04
   0.03
   0.02
   0.01
           (b)
                               STABILITY  INDEX =  3



                               WIND SPEED 5.0 m/s
FLOW PARAMETER
   actual
   maximurn
   proposed
       O.I     0.2
                                 I
                                         I    •          • •
0.5      I       2         5

        DISTANCE — kilometers



     FIGURE 9   (Continued)
                                      10
20
50
                                     106

-------
                             STABILITY  INDEX = 3
                             WIND SPEED  4.0  m/s
  O.W5
  0.040
  0.035
  0.030
aO.025
- 0.020
o
  0.015
  0.010
  0.005
        maximum
        proposed
             |JNE TYPE
            small dash
           mad dash
          (c)
      O.I
0.2
0.5     I       2        5
       DISTANCE — kilometers
10
20
50
                           FIGURE 9   (Concluded)

          The effect of allowing  the  Gaussian  diffusion coefficients to
assume values that deviate from their categorical  standards is shown in
Figure 10.  Consider Figure  10(a):  an interpretation is that the true
values could correspond to stability  indexes of  2.75  or 3.25, which are
represented by  the dashed lines.  However,  owing to  the inherent limita-
tions in determining the stability  class  to that precision, those cases
would be placed in Class 3 (the solid line).   The  differences between
the dashed and  the solid lines are  indicative  of the  errors that result.
These differences are  quite  pronounced in this case.
          It should be noted that the accuracy of  the stability parameters
decreases with  the distance  from  the  source.   Turner  indicated that errors
                                   107

-------
                                    STABILITY INDEX  • 3

                                    WIND SPEED -   5.0  m/s
E
a.
a
X
o
   0.12
   0.10
   0.08
   0.06
   0.04
   0.02
 I           I  • • • •  I

 STABILITY     LINE
 standard      solid
       Edev     srnal
       oerv     meo
           (a)
       O.I
0.2
                                   TYPE

                                  I  dash
                                  dash
0.5      I        2         5

        DISTANCE — kilometers
10
20
50
  FIGURE 10   SENSITIVITY OF CONCENTRATIONS  TO DIFFUSION  FLOW COEFFICIENT

              VARIATIONS (Souix Plant)
                                       108

-------
                                     STABILITY INDEX  • 4
                                     WIND SPEED •   5.0 m/s
  0.035
  0.030
  0.025
|o.020
I 0.015
o
  0.010
  0.005
                I           I
                STABILITY
                standard
                    >5 gev
           (b)
         I
     UNE TYPE
     solid
     small  dash
     meo dash
                                           I
       O.I      0.2
0.5      I        2
        DISTANCE —
      5

kilometers
                              FIGURE  10   (Continued)
                                        109

-------
                                  STABILITY INDEX • 5
                                  WIND SPEED -  5.0 m/s
  0.035
  0.030
  0.025
 |o.020
£ 0.015
o
  0.010
  0.005
 STABILITY
 standard
 •0-25 gev
 •0-25 oev
        r (C)
                                           I  T
     UNE TYPE
     solid.
     smajl
     med
                                       I
       O.I
0.2
0.5     I      2         5

       DISTANCE — kilometers
10
20
50
                           FIGURE 10   (Concluded)






may be severalfold,  in some cases reducing to  factors  of two within a few


hundred meters  of the source.  Our estimates are  conservative compared to


those of Turner (1970).



          The accuracy of the emission factor,  on the  basis of recent


stack tests  by  Rockwell  International appears  to  be  quite good.   Littman

                                              *
estimated the standard error at five percent.   The  daily fluctuation


from the mean of the fuel sulfur content is assumed  to be 10% or less.
 The PEDCO analysis  does not support this optimistic  figure;  however,

 the Rockwell  International stack tests are more  current.
                                    110

-------
This is analyzed and reported on a monthly basis.  Therefore, the calcu-
lation of the emission parameter, Q is normally within 15% of the actual
value.
          The approximate errors for the Union Electric Sioux plant are
summarized in Table 28.  We emphasize that this is a study of only one
case for a few meteorological and emission rate conditions.  Nevertheless,
it illustrates the effects of not altering the flow parameter estimates
with varying plant loads.  We recommend the use of simple models of the
stack exit volumetric flow and temperature.  Such models should be based
on data from the plants, when available.  When the data are not available,
the "proposed RAPS" technique of scaling back the volumetric flow with
constant stack temperature should be used.   By reducing errors in all
emission parameters to within 15%, modelers can focus on improvement on
the atmospheric portion of the models.
                                Table 28
          TYPICAL ERRORS IN SIGNIFICANT DOWNWIND CONCENTRATIONS
                  Sensitivity Parameter
             Emission rate
             Flow parameters (maximum values
             at half capacity)
             Flow parameters (proposed RAPS
             at half capacity)
             Dispersion coefficients
                                                   Error
   15

> 100

   15
> 100
 The PEDCO analysis does not support this optimistic figure;  however,
 the Rockwell International stack tests are more current.
                                  Ill

-------
              XI  EFFECTS OF RAPS EMISSION INVENTORY ERRORS






A.   Point Source Dominated Inventory (S02)




     1 .    Discussion



          As discussed in Section III, the air quality model output, X,


can be represented as some function of emissions inputs, Q, and meteorologi-


cal input G as follows:
                                                                    <27>
where X  x'  ,  Q, and G are vectors in time (t)  and space (s).  The X'
    .      err'    '                                    v               err
    V?

term  represents an error in the formulation that is only a mathematical


approximation to the physics and chemistry of the atmosphere.   For our


purposes, it is assumed that the X'   term represents a random error not


related to the Q and G arrays.   The sensitivity of any element in the X


array to any element in either the Q or G arrays is given by
                       a x.   a f [Q]      [GI.
                       	1 =    I    t,S,    t.S

                       3 Q.         a Q.
or
                       ax    a f [Q]      [G]
                       	i. _ 	!	L i s <	L »t» I                    /oo\
                       a G  ~       a G.              •              (28)
*
 In the notation used here, the prime will be used to designate errors

 due to the model formulation, unprimed error terms will designate the

 total error.
                                   113

-------
The variations in the Q    and G     terms are numbers that can be used to
quantify air quality model errors by substitution for incremental changes
in Q. and G. in  the discretized approximation of Eq.  (28).
          Our sensitivity analysis  to date has consisted of evaluating
individual Q. and G. parameters in  relation  to their effect on X..  The
parameters involved are  summarized  as follows:
          X  = ground level  SO  concentrations
          Q, =  SO  emission estimate  (g/s)

          Q9 =  Stack  gas  exit parameters  (temperature  K; volume flow
                m3/s)
          G, =  Wind direction (degrees)
          G~ =  Wind speed (m/s)
          G_ =  Atmospheric stability  (six classes derived from  the
                atmospheric lapse  rate  in   C/100 m) .
          Using our simulated RAPS  inventory and the CDM (Busse and
Zimmerman,  1973) ,  the following reference set of meteorological param-
eters were  chosen:
     •  Wind direction sectors:   NNW  (70%); NW (30%)
     •  Wind speed average:       U3 (4.43 m/s)
     •  Atmospheric stability:    Class 4.
This set of parameters is typical of winter conditions in the St. Louis
area.  The  ground-level concentrations classified by area, small point,
and large point sources were calculated for six downwind locations cor-
responding  to RAPS monitoring stations.   Results of the reference run
are presented in Table 29.
          First let us examine the  sensitivity of ground-level  concentra-
tion to errors  in  the wind direction  input, G.  If we assume an instrument
and interpolation  error in degrees, we must relate this to a frequency
distribution as required  by the CDM.   It  must also be noted that the CDM
requires uniform wind fields, although more current models might require
complex wind fields.   Nevertheless, our analysis is valid and applicable

-------
                               Table 29
                  GROUND-LEVEL SO- CDM REFERENCE CASE
                                 j£~
                               (in
Coordinates
Station
104
109
110
115
117
123
UTMX
747.31
755.80
747.21
757.11
760.56
777.32
UTMY
4277.30
4279.89
4272.21
4297.80
4272.82
4286.38
Area
Source
33
4
25
1
6
1
Small
Point
5
10
10
2
10
17
Large
Point
0
741
50
148
322
57
Total
38
755
86
151
338
75
in both cases since we are assuming that the true wind for our reference
case is uniform (spatially)  throughout the region.
          In our analysis, we assume that wind direction error will result
in an incorrect wind category 10% of the time.  This is traceable to an
estimated wind direction error of 2.5 degrees.  The assumed erroneous
distribution is:
           •  NNW:  60%
           •  NW:    30%
           •  WNW:  10%
Table 30 summarizes the results of the computer analysis including esti-
mates of ground-level concentrations, individual errors, average error,
and average significant error.  Individual errors are the percentage
difference between the Table 30 totals and those for the reference case
(Table 29).  The average significant error is the mean error for stations
                                 3
with concentrations over 100 |J>g/m .
          A typical erroneous wind speed distribution is estimated in a
manner similar to that described for wind direction above and is sum-
marized in Table 31.  Results for the run below are displayed in Table 32.
          The third meteorological parameter, G-, considered was atmo-
spheric stability.  This parameter is not measured directly and must be
                                  115

-------
                           Table 30
      SENSITIVITY OF SO  ESTIMATES TO WIND DIRECTION ERRORS

                          (in
Station
104
109
110

115

117
123
Area
Source
31
5
24

1

6
1
Small
Point Source
6
10
8

1

9
15
Large
Point Source
7
643
43

128

279
64
Total
44
658
75
*
130
*
294
80
Error
16
13
13

14

13
7
Average error = 13; average significant error = 13.

*
 Used to calculate average significant error (for totals greater
 than 100 jig/m3.
                            Table 31
               ERRONEOUS WIND SPEED DISTRIBUTION
Wind Speed Category Distribution
(7.)
(Mean Wind Speed in mps)
Class
Sector
NNW
NW
2
(2.46)
10
5
3
(4.47)
50
20
4
(6.93)
10
5
Total

70
30
                             116

-------
                                Table  32
            SENSITIVITY  OF  SO   ESTIMATES  TO WIND  SPEED  ERRORS
Station
104

109
110

115

117
123
Area
Source
35

5
27

1

7
1
Small
Point Source
5

11
11

2

10
18
Large
Point Source
0

765
51

145

329
56
Total
40
*
780
88
*
148
*
345
74
7o Error
5

3
2

2

2
1
   Average error  =  3; Average  significant  error =  2.
   *
    Used to calculate average  significant  error.
derived from other measurements.  For this sensitivity analysis, atmo-
spheric stability is assumed to be determined solely by the average
thermal lapse rate.  We emphasize that all assumed errors are based on
conservative estimates of instrument interpolation and interpretation ac-
curacies.  The assumed stability erroneous distribution is shown in
Table 33.  Results for this run are displayed in Table 34.
          So far, we have considered estimates of errors for meteorologi-
cal inputs used with COM.  We have chosen those conditions under which
CDM is most accurate (steady-state conditions, uniform wind fields,
moderate wind speeds) .  Our assumption is that variations in the meteoro-
logical variables occur over a short undefined period and manifest them-
selves in a sequence of discrete steady-state conditions.  Our error esti-
mates have been on the conservative side.  There are occasions when more
severe errors are encountered.  As an example, atmospheric stability could
be completely misclassified.  This condition is illustrated in Table 35,
which indicates the results of misclassifying stability to the next lower
level.  Even if RAPS achieves its objectives, errors such as these will
                                  117

-------
                               Table 33
             ERRONEOUS ATMOSPHERIC STABILITY DISTRIBUTION
Atmospheric Stability Distribution
(*/}
{/o}
Sector
NNW
NW
Class 3
14
6
Class 4
56
24
                               Table 34
                SENSITIVITY OF  S02 ESTIMATES TO TYPICAL
                ATMOSPHERIC STABILITY ESTIMATION ERRORS
Station
104
109
110
115
117
123
Area
Source
30
4
23
1
6
1
Small
Point Source
6
9
9
2
9
15
Large
Point Source
0
647
45
138
277
51
Total
36
*
660
77
*
141
*
292
67
% Error
5
13
10
7
14
11
  Average error = 10; Average significant error = 11.
  *
   Used to calculate average significant error.
still occur, but they should be infrequent.  We choose to consider the
results presented in Tables 30, 32, and 34 as being more typical.
          The sensitivity analysis for the emission parameters, Q-^ and Q_,
depends upon a random error about the mean estimate.  The error follows
a Gaussian distribution.  A standard deviation of error is estimated.   The
model (CDM) uses emission parameter values that vary from the control  case
according to the equation
                                  118

-------
                                Table  35



                    SENSITIVITY  OF  S02 ESTIMATES  TO

          WORST  CASE  ATMOSPHERIC STABILITY  ESTIMATION ERRORS
Station
104
109
110
115
117
123
Area
Source
19
3
16
1
6
1
Small
Point Source
11
4
4
1
3
6
Large
Point Source
0
270
23
99
98
29
Total
30
277"
42
100*
*
107
36
% Error
21
66
51
51
68
52
  Average error = 52; Average significant error = 62.

  *
   Used to calculate average significant error.
                                       Qk+6k
k=l
                                   k=l
                                         (30)
where



          Q' = The actual value of the parameter for source k



          Q  = The estimated emissions value for source k
           K.


          e,  = The error which is randomly selected but follows

               a normal distribution.



          In the first sensitivity analysis for emission rate variations,


Q, standard deviations for the different types of sources were assumed


as follows:



          •  Large point:  0.2Q


          •  Small point:  °-^Qk


          •  Area:  0.75Qk.
                                  119

-------
An interpretation for large point sources is that the standard deviation
for the error, e, , is 20% of the estimated value.  Assuming a normal dis-
tribution, approximately 6870 of the sources will have errors within 2070
of the estimated value, 95% of the sources will have errors with 2 stan-
dard deviations  (4070) , and so forth.  A random number generator is called
for each source  in the CDM run.  The actual value, e, , for that source is
                                                 '  k
a function of the random number, which is part of a normally distributed
set.  The results of  the sensitivity analysis for Q, are given in Table 36.
                                  Table 36
                  SENSITIVITY OF S02 ESTIMATES TO TYPICAL
                      RANDOM ERRORS IN EMISSION RATE
Station
104

109
110

115

117
123
Area
Source
31

4
26

2

7
1
Small
Point Source
6

13
11

1

11
17
Large
Point Source
0

684
47

156

300
60
Total
37
*
701
83
*
159
*
317
79
% Error
3

7
3

5

6
5
   Average  error  =  5; Average  significant  error =  6.
   •&
    Used  to calculate average  significant  error.
          In  the  second  sensitivity analysis  for  emission rates we as-
sumed a worst case  standard  deviation  for  random  error as follows:
          •   Large  point:  0-^Qi,
          •   Small  point:  0.8Q
                               iC
          •   Area:   l«5Qk
The results of this  sensitivity analysis are  presented in Table 37.
                                  120

-------
                               Table 37
              SENSITIVITY OF S02 ESTIMATES TO WORST CASE
                    RANDOM ERRORS IN EMISSION RATE
Station
104
109
110
115
117
123
Area
Source
44
4
35
2
7
1
Small
Point Source
6
4
7
2
6
18
Large
Point Source
0
902
45
121
379
52
Total
49
911*
87
124*
392*
72
% Error
29
21
1
18
16
4
      Average error = 15;  average significant error = 19.
      *
       Used to calculate average significant  error.
          A similar sensitivity analysis was conducted for variations in
Q~, the stack flow parameters (exit temperature and volumetric flow).  In
this analysis we assumed standard deviations for both parameters as fol-
lows :
          •  Temperature:   0.1 T
          •  Volume flow:   0.2 VF
where T and VF represent the estimated values for exit temperature and
volume flow, respectively.
          In this analysis, variations in the Q~ parameters are assumed,
equal for small and large point sources and are not a factor for area
sources.  Results of this sensitivity run are given in Table 38.

     2.   Summary
          The results of our sensitivity analysis for SO,, are summarized
in Tables 39 and 40.
          The actual numbers in the two summaries are representative, but
they are subject to wide variations, depending on the exact meteorology
                                  121

-------
                         Table  38

          SENSITIVITY OF SC>2 ESTIMATES TO TYPICAL
        RANDOM ERRORS  IN STACK  EXIT  GAS  PARAMETERS
Station
104
109
110
115
117
123
Area
Source
33
4
25
1
6
1
Small
Point Source
4
10
10
1
9
17
Large
Point Source
0
714
47
124
312
52
Total
36
728*
82
127*
327*
70
% Error
5
4
5
16
3
7
Average error = 7; average significant error = 8.
*
 Used to calculate average significant error.
                          Table  39

     SUMMARY  1:   SIGNIFICANT GROUND-LEVEL CONCENTRATION
   AVERAGE ERROR CAUSED BY  TYPICAL INPUT PARAMETER ERROR
Parameter
Q1
Q2
Gl
G?
G3
= Emission rate
= Stack exit parameters
= Wind direction
= Wind speed
= Atmospheric stability
Error
6
8
13
2
11
                          Table 40

     SUMMARY  2:   SIGNIFICANT GROUND-LEVEL CONCENTRATION
  AVERAGE ERROR CAUSED BY WORST-CASE INPUT PARAMETER ERROR
Parameter
Q
G3
= Emission rate
= Atmospheric stability
Error
(01 \
(to)
19
62
                            122

-------
and relative orientation of sources.   In any case, the analysis does re-
veal that the accuracy of the S0_ emission estimation methodology is con-
sistent with accuracies for meteorological inputs, even allowing for
reasonable improvements in quantifying the meteorological parameters.
Moreover, our test case assumed that  the model represented the physics
of the atmosphere perfectly.  Therefore, our approach was quite conserva-
tive.
          Our tests also revealed, that, as expected, S0» ground-level
concentrations are primarily caused by large (1000 tons/year and above)
point sources.  These are the sources that are being monitored most
closely; i.e., hourly process data are being collected.  Also, a number
of stack measurements were collected  to verify emission factors.

B.   Area Source Dominated Inventory  (CO)
     1.   Discussion
          Much of the discussion for  point sources (Section XI.A) applies
to area sources also.  The sensitivity analysis for meteorological param-
eters generally applies.  The percentage of error in area source CO esti-
mates caused by variations in meteorological parameters (G.. , G_ , and G_)
will be somewhat less than for a point source dominated inventory.  This
lower percentage of error results from area sources being more uniformly
spread out over relatively large areas and, hence, less sensitive to hori-
zontal variations in the meteorological parameters.  Also, area source
emissions are not as sensitive to plume rise variations resulting from
errors in stack characteristics, Q9.   Therefore, air quality estimates
should be less sensitive to errors from area sources than to errors from
point sources.

2.   Sensitivity
     During the course of this study, the RAPS inventory was unavailable
for use as an input to a comprehensive sensitivity analysis.  Neverthe-
less, we were able to make some estimates based on previous analyses  for
S02, which also tr.eated area sources.

                                   123

-------
          To give some idea of the sensitivity of area dominated sources
to emission estimate inaccuracies, we have reduced previous results from
Tables 29, 36, and 37.  This result appears in Table 41.  As shown, even
for large errors (coefficients of precision of 0.75 T and 1.5 T), the aver-
age observed error in ground-level concentration is between 7 and 297o.
                               Table  41
                   AREA  SOURCE  SENSITIVITY  COMPARISON
Station
104
109
110
115
117
123
Average
Ref
33
4
25
1
6
1
—
a = 0.75 T
e
Value
31
4
26
2
7
1
—
Error
6
0
4
—
17
—
7
CT = 1.5 T
e
Value
44
5
35
2
7
1
--
Error
(7o)
33
25
40
—
17
—
29
     3.   Adequacy of the CO Inventory
          It is tempting to say that errors in the CO inventory should
not result in prediction errors in excess of 307o.  Based on the number
of assumptions in estimating the error's coefficient of precision, this
upper bound of 307» is tenuous at best.  Nevertheless, it is the best esti-
mate that can be made at this time.
          It should also be noted that sensitivity is a function of the
air quality model used.  For our evaluation, CDM was used.  We have as-
sumed that the CDM results are representative of what other models might
yield.  Also, our sensitivity analysis did not use the CO RAPS inventory,
which was not available.  Therefore, the actual numbers (7 and 297o) could
change  somewhat if the  real inventory were used.
                                  124

-------
C.   Other Primary Pollutants
     1.   Total Hydrocarbons
          As in the case for CO, the THC inventory is dominated by high-
way source emissions.  However, THC point sources and stationary area
sources do contribute more than they do for CO.  Nevertheless, the point
sources contribute 23.370 of the total THC emissions (compared with 9. TL
for CO; see Table 25).
          Since we have estimated comparable accuracies for both CO and
THC (see Table 25), it  would appear that that air quality prediction
errors caused by errors in the THC emissions inventory should be about
the same for both pollutants--? to 30%.

     2.   Nitrogen Oxides
          The NO  inventory is dominated by point source emissions, al-
                X
though to a lesser extent than point source emissions dominated the S0_
inventory.  As discussed in Section V.B, the NO  estimates are not as
accurate as the SO- estimates.  Furthermore, the point source data may
have considerable bias — estimated as 20% (see subsection V.B) .  Neglect-
ing this bias, we estimate that random errors in the NO  emission inventory
                                                       X
could typically produce errors in predicted ambient NO  concentrations com-
                                                      X
parable to the worst case SO- estimates, perhaps somewhat less than aver-
age 19% shown in Table 37.

     3.   Particulates
          The TSP inventory consists primarily of fugitive dust sources.
This inventory relies on a totally different methodology than other sources,
The TSP inventory is also much more cumbersome to verify.  Only limited
quantities of field test data exist.  These are not adequate to assess
fully the bias and coefficient of precision of the emission estimates.
This is a current research area that may produce results that can be used
to assess the accuracy  of the RAPS data.  Because of these several factors,
we have not included an estimate on TSP emission inventory inaccuracies.
More work is needed on particle size distribution, and more effort must be
expended in collecting validation data.
                                  125

-------
D.   Photochemical Model Implications
     Hydrocarbons (HC) and nitrogen oxides (NOX) are the two primary pol-
lutants of interest in the context of photochemical oxidants.  Although
estimates of pollutant emissions contain inaccuracies, it is worthwhile
to have an estimate of the magnitude of the errors that might be encoun-
tered when computing secondary pollutants such as 0- and NO- .  This prob-
lem can be treated within the framework of sensitivity analysis since
some kind of transfer function linking secondary and primary pollutants
must be postulated.
     Let S denote the concentration of the secondary pollutant — i.e., 0«
     0-- and let Q   and %o  deno
     2.            HC        X
the transfer function has the form
or N00-- and let Q   and %o  denote the emissions of HC and NO .   Then,
     2.            HC        X                                  X
                   S = f(x, y, z, t, QRC, QNQ ,  G)                 (30)
where x, y, and z denote the spatial coordinates, t denotes the time co-
ordinates, and G denotes generically the weather conditions.   The function
f may be as simple as the linear rollback equation, or as complicated as
a nonlinear chemical diffusion model.
     Estimates of pollutant emissions are generally available as aggre-
gated tonnages over a large area, e.g., a metropolitan area.   One can
postulate a range of error for the aggregated estimate, e.g., ±257o, and
attempt to estimate the effect of this error on S.  In this case localized
effects are ignored and only the chemistry of HC and NO  remains as the
                                                       X
major factor relating emissions and secondary pollutants.
     For ozone, an upper bound on the magnitude of the change in concen-
may be obtained by a simple proportional model:

                              AS  « AQ                             (31)

This applies only if AQ  and AQjqQ  are relatively small (less than 20%)
and their respective signs are the same.  If the signs are different--
e.g. , if E   is low and ENQ  is high — then Eq.  (31) can yield very mis-
leading results for AO^ because of the nonlinearity  of the chemical
interaction between HC and NO .
                             x
                                  126

-------
     Equation (31) also applies to N02 only in the case of A QNQ  and AQ
have the same sign and roughly the same magnitude.  If the size of the
error for N09 emissions differs greatly from the error of the  HC emissions,
then Eq. (31) will yield a low value for the impact of the error on NO-
levels.  For example, if A (}„_  =  -15%  and A Qwr.  = -2%, then Eq. (31) would
                           HC                ux
yield an optimistic (low) value for ANO?.
     In summary, errors in aggregated areawide emissions can be used to
provide pessimistic (large)  upper bounds for 0~ uncertainties  and optimis-
tic (small) lower bounds for N0~  inaccuracies.  This can be done only when
certain conditions associated with the size and sign of the emissions esti-
mates are satisfied.  Highly inaccurate error bounds will be obtained when
the errors in HC and NO  emissions have different signs or their respec-
                       X
tive magnitudes are far apart, or both.  For the RAPS inventory, we cannot
be sure whether the signs are typically different or the same.  One compli-
cation is that HC emissions are dominated by highway sources and NO  is
                                                                   X
dominated by large point sources.  Further  field tests, as described ear-
lier, would alleviate this uncertainty.
E.   Other Inventory Factors
     Previous studies have addressed such issues as grid size (for area
source), resolution of point source coordinates, and source heights (see
Section II).  The resolution of point source coordinates is generally
accepted to be about 10 m.  The rationale was described in a previous SRI
report  (Littman et al., 1974).  In general, Rockwell International has
endeavored to use the 10-m criteria in their compilation.

     1.   Source Heights
          Some discussion on source heights appears in the background
literature (see Section II).  Earlier, in Section XI.A, we have shown the
effect of inaccuracies of source height estimates (including plume rise
calculations) for point sources.  For the other dominant inventory (high-
way sources), the emission heights are well known.
                                  127

-------
     2.    Area Grid Size
          The smallest grid in the RAPS system is 1 km by 1 km.   A previ-
ous sensitivity analysis, described in Section II, indicates that grids
                o
larger than 1 mi  can lead to significant errors in estimated pollutant
concentrations.
          In reality, the required grid size is a function of:   (1)  the
spatial variability of emissions, and (2) the ability of the air quality
model to handle the resulting grid matrix.
          For the problem of handling the grid matrix, with the  advent
of more efficient computers and more advanced models, the 1-km grid will
in the future be larger than desired.  At present, it is consistent with
the current state of the art.
          For economy and efficiency, the 1-km grids are convenient now
because they are consistent and compatible with the census data.  There-
fore, 1-km grids appear to be a reasonable compromise in satisfying RAPS
requirements.  Should smaller grid sizes be required at some future date,
new algorithms could be used to refine the spatial resolution.
                                  128

-------
                               REFERENCES
Ackerman, B., 1974:   "Wind Fields Over the St.  Louis Metropolitan Area,"
     J. of Air Pollution Control Association,  24,  3, 232.

Briggs, G. A., 1972;  "Discussion on Chimney Plumes in Neutral and Stable
     Surroundings,"  Atmospheric Environment.

Briggs, G. A., 1971:   "Some Recent Analyses of  Plume Rise  Observations,"
     Proceedings of  the Second International  Clean Air Congress,  ed.
     H. M. Englund and W. T. Beery, Academic  Press.

Busse, A. D., and J.  R. Zimmerman, 1973:   "User's  Guide for the Climato-
     logical Dispersion Model," EPA Publication EPA-R4-024.

Compilation of Air Pollutant Emission Factors,  Second Edition with Supple-
     ment 5, U.S. EPA Publication No. AP-42,  1975.

Cowherd, C. and C. Guenther, 1976:  "Development of a Methodology and
     Emission Inventory for Fugitive Dust for the  Regional Air Pollution
     Study," Midwest Research Institute,  EPA Report EPA-450/3-76-003.

Ditto, F. H. et al.,  1973:  "Weighted Sensitivity  Analysis of Emission
     Data," Final Report, EPA Contract 68-01-0398.

Gibbs, L. L., C. E.  Zimmer, and J. M. Zoller,  1974:  "Source Inventory
     and Emission Factor Analysis," Vol.  I and  II, Final Report,  EPA
     Contract 68-02-1350.

Griscom, R. W., 1977:   "Point and Area Source Organic Emission Inventory,"
     100% Completion Report, Rockwell International AMC 7010.T0108I-CR.

Hare, C. T., and K.  J. Springer, 1973 and 1974:  "Exhaust  Emissions from
     Uncontrolled Vehicles and Related Equipment Using Internal Combustion
     Engines," Parts 1-7, EPA Contract EHS 70-108

Hilst, G. R., 1970:   "Sensitivities of Air Quality Prediction to Input
     Errors and Uncertainties" (Stern, ed.) ,  Proceedings of Symposium
     on Multiple Source Urban Diffusion Models,  U.S. Environmental Pro-
     tection Agency,  Research Triangle Park,  North Carolina.

Holden, R. E., 1975:   "Residential and Commercial  Area Source Emission
     Inventory Methodology for the Regional Air Pollution  Study," En-
     vironmental Science and Engineering, Inc.,  EPA Report EPA-450/3-75-
     078.
                                   129

-------
Koch, R. C., et al., 1971:  "Validation and Sensitivity Analysis of the
     Gaussian Plume Multiple-Source Urban Diffusion Model," NTIS Publica-
     tion Number PB-206 951, Geomet Incorporated, Rockville, Maryland.

Kunzelman, P., et al., 1974:  "Automobile Exhaust Emission Modal Analysis
     Model," Report No. EPA-460/3-74-005, Calspan Corporation,  Buffalo,
     New York.

Littman, F. E., 1974:  "Regional Air Pollution Study Point Source Metho-
     dology and Inventory," EPA Publication EPA-450/3-74-054.

Littman, F. E., R. W. Griscom, and Otto Klein, 1977a:   "Point Source
     Emission Inventory," Rockwell International, EPA Report EPA-600/4-
     77-014.

Littman, F. E., R. W. Griscom, and G. Seeger, 1977b:  "Hydrocarbon Emis-
     sion Inventory," Rockwell International Report AMC7010.T0108F-FCR,
     RAPS Task Order.108-F.

Littman, F. E., R. W. Griscom, and H. Wang, 1977c:  "Sulfur Compounds
     and Particulate Size Distribution Inventory," Rockwell International,
     EPA Report EPA-600/4-77-017.

Littman, F. E., and K. M. Isam, 1977:  "Off Highway Mobile Source Emis-
     sion Inventory," Rockwell International', EPA Report EPA-600/4-77-041.

Littman, F. E., S. Rubin, K. T. Semrau, and W. F. Dabberdt, 1974:  "A Re-
     gional Air Pollution Study (RAPS) Preliminary Emission Inventory,"
     SRI Project 2579, EPA Report EPA-450/3-74-030.

Ludwig, F. L. and W. F. Dabberdt, 1972:  "Evaluation of the APRAC-IA Urban
     Diffusion Model for Carbon Monoxide," SRI Final Report, Coordinating
     Research Council and Environmental Protection Agency, Contract CAPA-
     3-68 (1-69).

Middleton, W.E.K. and A. F. Spilhaus, 1953:  Meteorological Instruments,
     University of Toronto Press, Canada.

Myers, R. L. and J. A. Reagan, 1975:  "The Regional Air Monitoring System-
     St. Louis, MO.," paper presented at International Conference on Envi-
     ronmental Sensing and Assessment, Las Vegas, Nevada, September 15-19.

Patterson, R. M., R. D. Wang, and F. A. Record, 1974:  "Airport Emission
     Inventory Methodology," GCA Corporation  (Technology Division), EPA
     Contract 68-02-0041  (Task 18).

Ruff, R. E., F. L. Ludwig, and L. S. Gasiorek, 1976:  "An S02 Emission
     Limitation Program," Final Report, SRI Project 4044, Stanford Re-
     search Institute, Menlo Park, California.
                                  130

-------
Shelar, E. S., W.  F.  Dabberdt,  and R.  E.  Ruff,  "Field  Test  Plan for Eval-
     uation of the RAPS Mobile-Source  Emissions Methodology,"  Project Re-
     port, SRI Project 4331.

Slade, D.  H. ,  1968:   Meteorology and Atomic  Energy, U.S.AEG.,  NTIS  (TID-
     24190),  (1968).

Sturm, J.  C.,  1976:   "An Estimation of River Towboat Air  Pollution  in St.
     Louis, Missouri," U.S.  Department of Transportation  Report No. DOT-
     TSC-OST^75-42.

Trijonis,  J.  C. and K. W.  Arledge, 1975:   "Utility of  Reactivity Criteria
     in Organic Emission Control Strategies  for Los Angeles,"  TRW Environ-
     mental Services, EPA Contract 68-02-1735.

Turner, D. B., 1964:   "A Diffusion Model  for an Urban  Area," J. of  Applied
     Meteorology,  3_,  1, 83.

Turner, D. B., 1970:   "Workbook of Atmospheric  Diffusion  Estimates," EPA
     Publication AP-26.

Turner, D. B.  and A.  D. Busse,  1973:   trUser's Guide  to Interactive  Ver-
     sions of Three Point Source Dispersion  Programs:   MAX, PTDIS,  and
     PTMT."

Weast, T.  E.,  L. J.  Shannon, P. G. Gorman, and  C. M. Guenther, 1974:
     "Fine Particulate Emission Inventory and Control  Survey," Midwest
     Research Institute, EPA Report EPA-450/3-74-040.

Wiltsee, K. W., S. B. Khanna, and J.  C. Hanson, 1977:   "Assessment  of
     Railroad Fuel Use and Emissions  for  the Regional  Air Pollution Study,"
     Walden Research, EPA Report EPA-450/3-77-025.
                                  131

-------