* O \

KWJ

*1 PRO^^

Technical Support Document (TSD) for
Adoption of the Generic Reaction Set Method
(GRSM) as a Regulatory Non-Default Tier-3 N02
Screening Option

1


-------
2


-------
EPA-454/R-24-005
November 2024

Technical Support Document (TSD) for Adoption of the Generic Reaction Set Method (GRSM) as
a Regulatory Non-Default Tier-3 N02 Screening Option

U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Air Quality Assessment Division
Research Triangle Park, NC

3


-------
Preface

This technical support document (TSD) provides a review of the Generic Reaction Set Method
(GRSM) NO2 option implementation and model performance in AERMOD version 23132. The
purpose of this TSD is to support adoption of GRSM as a new regulatory non-default Tier 3 NO2
screening option in AERMOD. The TSD presents and summarizes GRSM model performance
based on four NO2 model evaluation databases used to determine appropriate application of
NO2 screening options as part of the regulatory default version of AERMOD.

This version of the TSD revises the previous October 2023 version (EPA-454/R-23-005) to
include supplemental performance and sensitivity modeling evaluations in a new Appendix A.
These supplemental evaluations assess and compare refinements made to GRSM in AERMOD
version 23132 that were not included in the GRSM code previously implemented in AERMOD
version 22112. The refinements made to the GRSM code in AERMOD version 23132 include
improvements to model behavior and performance owing to more realistic treatment of
building effects on instantaneous plume spread and ozone entrapment, better accounting of
multiple plume effects, and mitigating GRSM over-predictions in the far-field (e.g., beyond
approximately 0.5 to 3 km for typical point source releases) while enhancing concentrations in
the near-wake and far-wake building downwash zones (e.g., 10's to 100's of meters downwind)
for some stack and building configurations.

All source code related to the formulation of the GRSM option in version 23132 has been
preserved and is unchanged in version 24142. The requirement to include the BETA model
option flag when using the GRSM option in version 23132 has been removed in version 24142.

4


-------
Contents

Preface	4

Contents	5

Figures	6

Appendix Figures	6

Tables	7

1.	Introduction	8

2.	Background	8

2.1 The 3-Tiered Approach for AERMOD N02 Modeling Demonstrations	8

3.	Current Regulatory Status and Features of GRSM	9

4.	GRSM Implementation in AERMOD	9

5.	Model Evaluation of GRSM	10

5.1	Pala'au, Hawaii N02 Database	11

5.2	Empire Abo, Artesia, New Mexico N02 Database	14

5.3	Balko, Oklahoma N02 Database	17

5.4	Denver-Julesburg Basin, Platteville, Colorado N02 Database	20

6.	Summary	23

References	24

Appendix A - GRSM Code Updates and Testing for AERMOD v23132	25

A.l Model Performance Evaluation Results Summary	25

A.2 Sensitivity Modeling Results Comparison	32

5


-------
Figures

Figure 1 - Pala'au N0X Ranked Q-Q Plot	13

Figure 2 - Pala'au N02 Ranked Q-Q Plot	13

Figure 3 - Empire Abo NOx Ranked Q-Q Plot	15

Figure 4 - Empire Abo N02 Ranked Q-Q Plot for the North Monitor	16

Figure 5 - Empire Abo N02 Ranked Q-Q Plot for the South Monitor	16

Figure 6 - Balko NOx and N02 Ranked Q-Q Plot for all monitors	18

Figure 7 - Balko NOx and N02 Ranked Q-Q Plot by monitor	19

Figure 8 - Colorado NOx and N02 Ranked Q-Q Plots for Pad 1	22

Figure 9 - Colorado NOx and N02 Ranked Q-Q Plots for Pad 2	22

Appendix Figures

Figure A.l-1 - Pala'au Monitor Scatter Plot and Ranked Q-Q Plot	26

Figure A.l-2 - Empire Abo - North Monitor Scatter Plot and Ranked Q-Q Plot	27

Figure A.l-3 - Empire Abo - South Monitor Scatter Plot and Ranked Q-Q Plot	27

Figure A.l-4 - Balko - Field (north) Monitor Scatter Plot and Ranked Q-Q Plot	28

Figure A.l-5 - Balko - North Fence Monitor Scatter Plot and Ranked Q-Q Plot	28

Figure A.l-6 - Balko - East Fence Monitor Scatter Plot and Ranked Q-Q Plot	29

Figure A.l-7 - Balko - Tower (southeast) Monitor Scatter Plot and Ranked Q-Q Plot	29

Figure A.l-8 - Colorado - Pad 1 Monitors Scatter Plot and Ranked Q-Q Plot	30

Figure A.l-9 - Colorado - Pad 2 Monitors Scatter Plot and Ranked Q-Q Plot	30

Figure A.2-1 - Sensitivity modeling stack and complex building configuration	33

Figure A.2-2 - GRSM v23132 35-meter Tall Stacks Highest-8th-High 1-hour N02	36

Figure A.2-3 - GRSM v22112 35-meter Tall Stacks Highest-8th-High 1-hour N02	36

Figure A.2-4 - GRSM v23132 Minus v22112 35-meter Tall Stacks Highest-8th-High 1-hour N02	37

Figure A.2-5 - GRSM v23132 50-meter Tall Stacks Highest-8th-High 1-hour N02	37

Figure A.2-6 - GRSM v22112 50-meter Tall Stacks Highest-8th-High 1-hour N02	38

Figure A.2-7 - GRSM v23132 Minus v22112 50-meter Tall Stacks Highest-8th-High 1-hour N02	38

Figure A.2-8 - GRSM v23132 65-meter Tall Stacks Highest-8th-High 1-hour N02	39

Figure A.2-9 - GRSM v22112 65-meter Tall Stacks Highest-8th-High 1-hour N02	39

Figure A.2-10 - GRSM v23132 Minus v22112 65-meter Tall Stacks Highest-8th-High 1-hour N02	40

6


-------
Tables

Table 1 - Pala'au Model Performance Statistics Summary (ng/m3)	14

Table 2 - Empire Abo Model Performance Statistics Summary (ng/m3)	17

Table 3 - Balko Model Performance Statistics Summary (ng/m3)	20

Table 4 - Colorado Model Performance Statistics Summary (ng/m3)	23

Table A.1.1 - GRSM v23132 and v22112 Model Performance Statistics Summary (ng/m3)	31

7


-------
1.	Introduction

The proposed revisions to Appendix W to CFR 40 Part 51—Guideline on Air Quality Models
(Guideline), includes a new version of AERMOD (23132)1. This new version of AERMOD includes
a proposed regulatory non-default Tier 3 NO2 screening option, i.e., the Generic Set Reaction
Method (GRSM; (Carruthers, Stocker, Ellis, Seaton, & Smith, 2017); (Stocker, et al., 2023)). This
Technical Support Document (TSD) reviews the scientific merit, implementation of the GRSM
formulation, and summarizes selected model evaluations to support the application of GRSM as
a Tier 3 NO2 screening option for use as part of the proposed regulatory version of AERMOD.

2.	Background

The chemistry, regulatory status, and performance evaluations of all existing AERMOD NO2
screening options are discussed in the U.S. EPA TSD for N02-Related AERMOD Options and
Modifications (U.S. EPA, 2015, December). This TSD will discuss the chemistry, proposed
regulatory status, and model behavior and performance of GRSM. Following the 2015 TSD,
selected graphical and statistical (U.S. EPA, 1992) comparisons between GRSM and other NO2
regulatory options are presented.

2.1 The 3-Tiered Approach for AERMOD NO2 Modeling Demonstrations
Section 4.2.3.4 of Appendix W (2017) details a 3-tiered approach for evaluating the modeled
impacts of NOx emission sources. These tiers assume increasing levels of conservatism (i.e.,
conservation of air quality as a resource for protecting public health) in the assessment of
hourly and annual average NO2 impacts from point, volume, and area sources for the purposes
of supporting the PSD program, SIP planning, and transportation general conformity. The
3-tiered approach addresses the co-emissions of NO and NO2 and the subsequent conversion of
NO to NO2 in the atmosphere. The tiered levels include:

•	Tier 1 - assuming that all emitted NO is converted to NO2 (full conversion),

•	Tier 2 - using the Ambient Ratio Method 2 (ARM2), which applies an assumed
equilibrium ratio of NO2 to NOx, based on analysis of and correlation with nationwide
hourly observed ambient conditions (Podrez, 2015), and

•	Tier 3 - applying the Ozone Limiting Method (OLM; (Cole & Summerhays, 1979)) and
Plume Volume Molar Ratio (PVMRM; (Hanrahan, P.L., 1999a and 1999b)) screening
options based on site-specific hourly ozone data and source-specific NO2 to NOx in-stack
ratios.

As discussed in section 4.2.3.4(e), regulatory application of Tier 3 screening options shall occur
in consultation with the EPA Regional Office and appropriate reviewing authority.

1 For more information on the proposed revisions to Appendix W and updates to AERMOD, please reference:
https://www.epa.gov/scram/2023-appendix-w-proposed-rule

8


-------
3.	Current Regulatory Status and Features of GRSM

As part of the 2023 proposed revisions to the Guideline, the EPA is proposing to include the
GRSM as a regulatory non-default Tier 3 NO2 screening option in AERMOD version 23132.

Following peer-reviewed publication (Carruthers, Stocker, Ellis, Seaton, & Smith, 2017), GRSM
was added to AERMOD as an alpha option in version 21112 and later updated to a beta option
in version 22112. The GRSM option is proposed to be adopted as a beta option in AERMOD
version 23132, and later advanced to a full regulatory NO2 screening option upon release of the
2024 version of AERMOD.

The primary motivation behind the formulation and development of the GRSM NO2 screening
option is to address photolytic conversion of NO2 to NO and to address the time-of-travel
necessary for NOx plumes to disperse and convert the NO portion of the plume to NO2 via
titration and entrainment of ambient ozone. The current regulatory non-default Tier 3 NO2
screening options, PVMRM and OLM, do not address or provide for treatment of these
photolysis and time-of-travel mechanisms, and have been shown to over-predict for some
source characterizations and model configurations at project source ambient air boundaries
and within the first 1-3 km. (Stocker, et al., 2023) and as presented in this TSD.

4.	GRSM Implementation in AERMOD

The functionality of the GRSM code implementation in AERMOD is similar to that of the
PVMRM and OLM schemes, with exception to some additional input requirements necessary
for treatment of the reverse NO2 photolysis reaction during daytime hours. Modeled source
inputs for GRSM require NO2/NOX in-stack ratios, with similar assumptions as applied to
PVMRM and OLM pursuant to section 4.2.3.4 of the Guideline. Ambient model inputs for GRSM
require hourly ozone concentrations taken from an appropriately representative monitoring
station or selection of monitoring stations for varying upwind sector concentrations. GRSM also
requires hourly NOx concentration inputs to resolve the daytime photolysis of NO2 reaction in
equilibrium with ozone titration conversion of the NO portion of the NOx plume. Hourly NOx
and NO2 concentrations input to AERMOD when using the GRSM method can also vary by
upwind sector concentration, as appropriate. Background NO2 concentrations are accounted
for in the GRSM daytime equilibrium NO2 concentration estimates based on the chemical
reaction balance between ozone entrainment and NO titration, photolysis of NO2 to NO, and
ambient background NO2 participation in titration and photolysis reactions. Nighttime NO2
estimates from GRSM are based on ozone entrainment and titration of available NO in the NOx
plume, and by default, AERMOD sets nighttime ozone concentrations to 40 parts per billion
(ppb) unless the NOMIN03 model option is specified. Note that all hourly ozone and NOx
ambient inputs to GRSM must coincide with the hourly meteorological data records for the

9


-------
period of the modeling analysis (i.e., minimum of 1-year for on-site data, 3 years of prognostic
data, and 5 years of airport data (i.e., meteorological data collected by either the National
Weather Service (NWS) or the Federal Aviation Administration (FAA), typically at airport
locations).

Updates to the GRSM formulation in AERMOD version 22112 were completed in late 2022 to
address more realistic building effects on instantaneous plume spread, accounting for multiple
plume effects on entrainment of ozone, and the tendency of GRSM to over-predict in the near-
field (e.g., beyond approximately 0.5 km to 3 km for typical point source releases). Sensitivity
testing and model performance evaluations of these updates to GRSM in AERMOD version
23132 have shown consistent or improved model behavior and performance. Please see
Appendix A for model evaluation and sensitivity testing results comparisons between GRSM
code implementations in AERMOD versions 22112 and 23132. The model performance
evaluations of GRSM specific to AERMOD version 23132 and compared to other regulatory NO2
options are presented and discussed in the following sections.

5. Model Evaluation of GRSM

Statistical evaluation of GRSM NO2 model performance was conducted based on four source-
oriented ambient ozone and NO2 monitoring databases assuming rural dispersion conditions.
Two legacy (1993) databases reflect the ARM2, OLM, and PVMRM evaluations presented in the
2015 TSD for NO2 modeling options (U.S. EPA, 2015, December). These legacy databases
included 1-year datasets developed for a power plant located on the island of Moloka'i, Pala'au,
Hawaii (Pala'au), and for a gas processing plant located in Artesia, New Mexico (Empire Abo).
Details of the Pala'au and Empire Abo databases are discussed at length in a 2013 technical
report (RTP Environmental Associates, Inc., 2013). The other two evaluation databases were
developed more recently in the 2014-2016 time period and include a 1-year field study at a gas
compressor station facility located near Balko, Oklahoma (Balko), and a six-week field study at
an oil and gas drill rig installation located in the Denver-Julesburg Basin near Platteville,
Colorado (Colorado). The Balko database included ozone and NO2 data collected at four
monitoring stations from December 2015 through December 2016. Details on the development
of the Balko database were published in March of 2020 (Panek, 2020) along with model
observation comparisons with ARM2, PVMRM, and OLM. The Colorado database included data
collected at a total of 12 monitoring locations upwind and downwind of two oil and gas well
drilling pads for a five-week period October 10th through November 16th, 2014. Further details
on the Colorado database are available at EPA's SCRAM website and documented in a separate
TSD (Colorado Field Study Workgroup: ERM, 2020).2 All four model evaluation databases

2 Denver-Julesburg Bason, Colorado N02 Evaluation Database:

https://gaftp.epa.gov/Air/aqmg/SCRAM/models/preferred/aermod/eval databases/denver-iulesburg.zip

10


-------
included site-specific meteorological data collected at the site, re-processed with AERMET
version 23132. Summaries of all databases and performance model evaluations are also
discussed in a technical report authored by the developers of GRSM for AERMOD (Stocker, et
al., 2023).

As discussed previously in Section 4.3 of (RTP Environmental Associates, Inc., 2013), conversion
of NO2 and NOx measurements in ppb to micrograms per cubic meter (|ag/m3) requires careful
consideration of the actual and standard meteorological conditions as well as the separate
contributing constituent components of the NOx plume in terms of NO and NO2 ppb by volume.
Conversion of ppb to |-ig/m3 from actual to standard temperature and pressure conditions
generally increases measurement concentrations by approximately 10% depending on season,
climatology, and elevation. Additionally, this conversion, typically based on the conventional
assumption that all NOx is NO2 with a molecular weight of 46 grams/mole, would increase the
l-ig/m3 measurement estimates by 20-30% for some shorter source-receptor distances,
especially between 10's to 100's of meters and possibly as far away as approximately 3 km
depending on dispersion conditions. The "true" NOx plume at these shorter distances is
composed of mostly NO (e.g., 50-95% NO/NOx by volume) and therefore, would contain less
mass given the 30 g/mole molecular weight of NO, which accounts for the 20-30% conservative
estimates of NOx as NO2 emission inputs typically used in AERMOD NO2 demonstrations. As
such, and based on the most current information on the four field datasets considered in this
TSD, the NOx as NO2 assumption was applied to all NOx emissions inputs, thereby introducing a
conservative bias in the modeled mass emission rates. From a regulatory perspective, the
performance of this conservative NOx as NO2 emissions assumption when compared to actual
measured NOx concentrations was considered because a regulatory modeling result would
need to show some level of performance as it pertains to the Appendix W requirement that the
model does not show bias to underpredict. Therefore, all input emissions assume NOx as NO2
(based on most current understandings of emission factors applied), and any dispersion
performance indicated from NOx modeled compared to NOx measured assumes no change to
the modeled |-ig/m3 concentrations whereas measured NOx represents the actual |-ig/m3
concentrations (at standard temperature and pressure; STP) as would be the case for any
regulatory modeling or monitoring demonstration. Note that AERMOD NOx and NO2
concentrations in |-ig/m3 are calculated internally based on standard temperature and pressure
(i.e., 298.15 K and 1013.25 mb). Chemistry performance was assessed in terms of modeled
l-ig/m3 NO2 at STP compared to measured |-ig/m3 NO2 (after conversion from ppb to |-ig/m3 at
STP).

5.1 Pala'au, Hawaii NO2 Database

The Pala'au hourly NO2 and ozone data were collected at a monitoring station located

11


-------
approximately 220 meters northwest of the facility. Hourly varying ozone data was developed
from the on-site monitoring data (93% complete). The annual ozone substitution value was set
to the default 40 ppb for all NO2 models. A single annual NOx value was set to 2.5994 |-ig/m3 for
the model simulations using GRSM. Background NO2 assumed 0.69838 ppb for all conversion
methods. NOx emissions were assumed to be non-varying for the entire 1993 study period, and
included six diesel engines and one combustion turbine with emission rates ranging from an
average of 12.6 Ib/hr to a maximum of 27 Ib/hr; total NOx emissions of 88.3 Ib/hr. All sources
assumed NO2/NOX in-stack ratios of 10%. Stack heights at the power plant were relatively short
and range from 24-32 feet and were modeled assuming flat terrain. All NO2 model outputs
were based on 1-hour averages as predicted at the single monitor receptor location.

AERMOD performs well at Pala'au as indicated in the Q-Q ranked plot shown in Figure 1 where
the modeled versus observed NOx concentrations track the 1:1 line throughout the ranked
distribution. As previously discussed, modeled mass emission rates assumed all NOx was NO2,
thus, introducing a conservative emission estimate bias that could be influencing the
agreement between observed and modeled NOx concentrations. Another emissions
uncertainty for Pala'au, which could inadvertently bias model-observation agreement during
non-operating periods, is the non-varying emission rates assumed for the 1-year evaluation
period. However, given the proximity of the monitoring station located 220 meters northwest
of the power plant, and the relatively consistent distribution of the NOx concentrations
throughout the monitoring period, altogether, these factors would indicate that the power
plant operated continuously at a normal demand load for the entire year. Note that no filtering
of the NOx observations was conducted (e.g., by downwind sector) to determine the final set of
7,856 model-observation pairings shown in Figure 1.

Figure 2 shows a ranked Q-Q plot of modeled versus observed NO2 concentrations for modeling
scenarios that use the proposed GRSM NO2 Tier 3 option as well as the other Tier 2 (ARM2) and
3 (OLM, PVMRM) regulatory NO2 options available in AERMOD. The ARM2 option performs as
intended the most conservatively, whereas OLM becomes the less conservative option by
comparison. PVMRM shows some slight underprediction whereas GRSM maintains a slightly
conservative performance trend just above the 1:1 line for most of the ranked distribution.
GRSM performs consistently compared to the other AERMOD NO2 options and shows no
unacceptable bias to underpredict peak concentrations for the Pala'au database.

12


-------
1993 Palaau Database
All Wind Directions

g200

o

2
C£

LLl

< 100

Station
° Monitor (220m)

0	100 200 300 400

NOX Observed (ug/m3)

Figure 1 - Pala'au NOx Ranked Q-Q Plot

1993 Palaau Database
All Wind Directions

N02_0ption
° ARM2
^ OLMALL
~ PVMRM
+¦ GRSM

50	100 150 200

N02 Observed (ug/m3)

Figure 2 - Pala'au N02 Ranked Q-Q Plot

13


-------
Table 1 shows a summary of fractional bias (FB) and robust highest concentration (RHC) model
performance statistics for the NOx and NO2 model option scenarios evaluated for Pala'au. The
FB shows decreasing conservative agreement between observations and model outputs for
NOx, ARM2, OLM, PVMRM, and GRSM model options; note that negative FB indicates a
conservative bias for the option, or overprediction. The RHC ratio and RHC FB results show
similar conservative hierarchy across the NO2 option evaluations, with increasing conservatism
shown for GRSM, PVMRM, OLM, ARM2, and full conversion NOx runs.

Table 1 - Pala'au Model Performance Statistics Summary (ng/m3)

Model Opt.

FB

RHC_Obs

RHC_Mod

RHC_ratio

RHC_FB

NOx

-1.01763

456.5125

471.9136

1.033737

-0.03318

ARM2

-1.22483

90.9536

237.2565

2.608545

-0.89152

OLM

-1.00706

90.9536

103.8257

1.141523

-0.13217

PVMRM

-0.79509

90.9536

98.17854

1.079436

-0.0764

GRSM

-0.90174

90.9536

82.87393

0.911167

0.092962

5.2 Empire Abo, Artesia, New Mexico NO2 Database

The Empire Abo hourly NO2, ozone, and meteorological data were collected at two monitoring
stations located approximately 1.6 km northeast (north station) and 2.4 km southeast of the
gas processing facility from June 11, 1993 through June 10, 1994. Model inputs for hourly
ozone and background NO2 and NOx were based on two wind sectors starting at 100 and 280
degrees, which AERMOD interprets as downwind or flow vector wind directions. The first
sector (winds blowing towards 100-280 degrees) used upwind hourly ozone and NO2
concentrations from the north station, while the second sector (winds blowing towards 280-
100) used upwind data from the south station. Substitution values for missing hourly ozone
and NOx data were taken from season-hourly varying maximum, while NO2 season-hourly
values were developed from highest-3rd-high observed values. The highest-3rd-high was
selected for NO2 substitution values in order to reflect a median value between unreasonably
high maximum NO2 values and the 1-hour NAAQS highest-8th-high. Similar to Pala'au, NOx
emissions for Empire Abo were assumed to be non-varying for the entire study period, and
included 21 combustion sources with emission rates ranging from an average of 20.2 Ib/hr to a
maximum of 69.4 Ib/hr, and with a facility total of 423 Ib/hr. All sources assumed NO2/NOX in-
stack ratios of 20%. Stack heights at the power plant from dominant sources averaged about
30 feet and all sources were modeled assuming flat terrain. All NOx and NO2 model outputs
were based on 1-hour averages as predicted at the north and south monitor receptor locations.

As shown in the ranked Q-Q plot in Figure 3, modeled NOx concentrations at the north and
south monitors tend to overpredict; note the north monitor shows some underprediction for

14


-------
the lower half of the ranked distribution. As previously mentioned, the overpredictions may be
in part due to the NOx mass emission rates that assume all NOx has converted to the mass of
NO2. Given the 1.6 km and 2.4 km distances to the north and south monitors, respectively, this
assumption may be valid for most worst-case scenarios; however, the ambient monitoring data
at these stations indicates the inner quartile range of the ambient NCh/NOx ratios varies
between 66-86%. The non-varying hourly emissions from Empire Abo dispersed over these
longer distances may also play a role in overestimating NOx concentrations. Note that pre-
filtering of the NOx observations was not conducted (e.g., by downwind sector, or other
parameter) to determine the final set of 16,547 model-observation pairings shown in Figure 3.

Figures 4 and 5 show ranked Q-Q plots of modeled versus observed NO2 concentrations at the
north and south monitors, respectively, for modeling scenarios that use the proposed GRSM
NO2 Tier 3 option as well as the otherTier 2 (ARM2) and 3 (OLM, PVMRM) regulatory NO2
screening options available in AERMOD. Similar to the results at Pala'au, the ARM2 option
performs the most conservatively, whereas OLM and GRSM modeled concentrations track
closely together and are more conservative than PVMRM. GRSM performs consistently
compared to the other AERMOD NO2 options and shows no unacceptable bias to underpredict
peak concentrations for the Empire Abo database.

June 1993 - June 1994 Empire Abo Database
North (1.6km) and South (2.4km) Monitors

Figure 3 - Empire Abo NOx Ranked Q-Q Plot

15


-------
June 1993 - June 1994 Empire Abo Database
North Monitoring Station (1.6km)

N02_Option
• ARM2
*¦ OLM
¦ PVMRM
-I- GRSM

50	100	150	200

N02 Observed (ug/m3)

Figure 4 - Empire Abo NOz Ranked Q-Q Plot for the North Monitor

June 1993 - June 1994 Empire Abo Database
South Monitoring Station (2.4km)

150

o~r
E

zs

CN1

2 100

>

Q
O

(T

LLI 50
<

.ALL





• '







"N =32516

/









/,
1 jr.











t jfo# /





















t M

Jft









N02_0ption

•	ARM2

*	OLM

¦ PVMRM
-I- GRSM

50	100	150

N02 Observed (ug/m3)

Figure 5 - Empire Abo N02 Ranked Q-Q Plot for the South Monitor

16


-------
Table 2 shows summary model performance statistics for NOx and NO2 at the Empire Abo north
and south monitors. The FB shows conservative agreement between observations and model
outputs. The RHC ratio and RHC FB show conservative bias for the north monitor and south
monitor, with the most conservatism shown for NOx and ARM2, whereas OLM, PVMRM, and
GRSM display consistent conservative bias for the north and south monitors.

Table 2 - Empire Abo Model Performance Statistics Summary (pg/m3)

Station

Model Opt.

N

FB

RHC_Obs

RHC_Mod

RHC_ratio

RHC_FB

North
(1.6km)

NOx

8418

0.519675

356.2944

477.27

1.339538

-0.29026

ARM2

8418

0.289203

130.2695

204.0459

1.566337

-0.44136

OLM

8418

0.292575

130.2695

152.6254

1.171613

-0.15805

PVMRM

8418

0.354584

130.2695

141.475

1.086018

-0.08247

GRSM

8418

0.314298

130.2695

152.9922

1.174428

-0.16044

South
(2.4km)

NOx

8129

0.364385

323.4253

392.8127

1.214539

-0.19375

ARM2

8129

-0.00239

72.57853

172.0086

2.369966

-0.81304

OLM

8129

0.000409

72.57853

128.2316

1.766798

-0.55429

PVMRM

8129

0.045474

72.57853

121.1714

1.669522

-0.5016

GRSM

8129

0.01555

72.57853

129.4047

1.782962

-0.56268

5.3 Balko, Oklahoma NO2 Database

The 1-year of hourly NO2 observations records for each of the four monitoring stations were
reduced by excluding values collected during hours when NOx concentrations were below 20
ppb and when the downwind direction from the source to the monitoring receptor location was
more than approximately 20-30 degrees (i.e., assuming a 40-60-degree downwind sector of
influence). As such, non-missing hourly modeling results were paired in time with the reduced
observations (total N pairs = 1742) to generate ranked Q-Q plots and summary statistics. In
brief, the monitoring stations, distances, and downwind directions from the sources were:

Field 425 m, 360 deg; North Fence (NF) 140 m, 360 deg; East Fence (EF) 101 m, 68 deg; and
Tower 66 m, 246 deg. NOx emissions at Balko were dominated by two of the large 2-stroke
cycle lean-burn natural gas-fired engines with combined maximum hourly NOx emission rates
on the order of 120 Ib/hr and NO2/NOX in-stack ratios of 10%. Relatively short stacks for these
units were modeled at 10 m and 20 m, with adjacent dominant building heights at 11-13 m. For
further details on the Balko field study and model configurations see (Panek, 2020).

The NOx and NO2 Q-Q plots shown in Figure 6 represent all model-observation data pairs from
the four monitoring stations. Both NOx and NO2 predictions using OLM and GRSM trend below
the 1:1 line at the upper end of the concentration distributions, but above the 1:2 line. NO2
predictions using ARM2 performs above the 2:1 line through the first half of the ranked

17


-------
distribution, converging to the 1:1 line at the upper end of the distribution. The negative bias
shown in the NOx performance suggests dispersion assumptions such as downwash and stack
parameters could be refined for some monitoring locations. The negative bias in NOx is largely
driven by the model performance and higher observed NOx impacts at the North Field (NF)
monitor, which is well within the near wake zone of two adjacent end-to-end, long buildings.
As shown in Figure 7, all other monitor locations show more conservative, or positive bias for
NOx.

The ranked Q-Q plot panels for NOx and NO2 shown in Figure 7 are presented for each monitor
location. NOx model predictions at the Field and East Fence show conservative bias with peak
value data pairs falling between the 2:1 and 1:1 lines. NO2 model predictions at these stations
ranging from 175 |-ig/m3 to 200 |-ig/m3 show a consistent conservative hierarchy across the NO2
options decreasing in order of the ARM2, OLM, PVMRM, and GRSM options. The NOx
predictions at the meteorological Tower monitor fall mostly along the 1:1 line; however, PVRM
is the least conservative performing NO2 option at this predominantly southwesterly, upwind
location. NOx predictions at the North Fence follow the 1:1 line with negative bias trends
starting at 750 |-ig/m3 and ending above the 1:2 line at about 1600 |-ig/m3. The negative bias at
the upper part of the distribution is most likely influenced by uncertainties in source and
building downwash characterizations at what is a relative short downwind 140-meter distance
from the dominant stack. The conservative hierarchy shown for NO2 predictions at the North
Fence is similar to the other monitor locations; however, the overly conservative PVMRM
predictions for the last three data pairs suggest further uncertainties in instantaneous plume
and building downwash formulations coded for PVMRM. GRSM does not mimic this behavior.

Balko	Balko

ALL Winds	ALL Winds

chem
o ARM2

OLMGRPALL
PVMRM
A GRSM

N =/6968

300-

200-

100

0 100 200 300

N02_ugm3

CO

E

§> 1500

x
x

X
CO
IN
>

I

Q
O

a:

LU
<

1000

500

CD
13

I

X
X
X
CO
CM
>

I

o
o

chem

• NOx

Figure 6 - Balko NOx and N02 Ranked Q-Q Plot for all monitors

500 1000 1500
NOx_ugm3

18


-------
CO



E

o
o

00

CD



13



1

X

600-

X



X



CO

400-

C\J

>



n



o

200-





rr



LU

0-

<



Field

N

/

567 /

/ 9



f/



jr.-''
// /•
m **





0 200 400 600 800
NOx_ugm3

chem

• NOx

chem

• NOx

CO

o>

3

I

X
X
X
CO
CM
>

I

D
O

K
LU
<

CO

O)
=3

I

X
X
X
CO
CM
>

I

Q

O

a:

LU

<

150

100-

50-

0 50 100 150
N02_ugm3

300-
200-
100-

NF

N =,$?76

/ y
/ /

'



\

\

\

**

chem
o ARM2

OLMGRPALL
PVMRM
A GRSM

chem
o ARM2

OLMGRPALL
PVMRM
A GRSM

0 100 200 300

N02_ugm3

CO

E

§> 1500

x
x
X
CO
CM
>

Q

0

01
LU
<

1000-

500

0 500 1000 1500
NOx_ugm3

N02_ugm3

CO

E 1200
lib

13

X

x
x
CO
CN
>

I

Q

o
or

LU
<

600

300-

300 600 900 1200
NOx_ugm3

N =?'232
/

800

co
E

~CT>

X1 600
X
X
CO

CM 400

D
O

200

Figure 7 - Balko NOx and N02 Ranked Q-Q Plot by monitor

Tower

chem
• NOx

0 200 400 600 800
NOx_ugm3

150

O)
13

X
X
X

CO 100
CM
>

0 50 100 150
N02_ugm3

Tower

chem

o

ARM2

~

OLMGRPALL

o

PVMRM

A

GRSM

19


-------
Table 3 provides FB and RHC statistics calculated for all monitor locations and for all NO2 model
options evaluated including NOx, ARM2, OLM, PVMRM, and GRSM. The table is sorted by
model option and decreasing RHC ratio values. In all, the GRSM RHC ratio and RHC FB results
indicate more consistent, logical model behavior when compared with modeled NOx
performance at the four monitors. With exception to the uncertainties discussed at the NF
monitor, GRSM performance statistics show less conservative bias than the other NO2 model
options.

Table 3 - Balko Model Performance Statistics Summary (ng/m3)

Station

Model Opt.

N

FB

RHC_Obs

RHC_Mod

RHC_ratio

RHC_FB

EF

NOx

232

-0.30893

785.1347

1449.214

1.845816

-0.59443

Field

NOx

567

-0.33405

481.5579

637.3523

1.323521

-0.27848

Tower

NOx

149

0.050238

825.8741

719.8355

0.871604

0.137204

ALL

NOx

1742

-0.13172

1884.88

1162.402

0.616698

0.474179

NF

NOx

794

0.030388

1884.88

1069.941

0.567644

0.551599

EF

ARM2

232

-0.5797

106.2721

210.4881

1.980653

-0.65801

Tower

ARM2

149

-0.23291

121.7104

211.9201

1.741183

-0.54078

Field

ARM2

567

-0.67069

104.2684

171.3581

1.643433

-0.48682

ALL

ARM2

1742

-0.44312

220.7673

199.6457

0.904326

0.100481

NF

ARM2

794

-0.28016

220.7673

191.5234

0.867535

0.141861

EF

OLMGRPALL

232

-0.51368

106.2721

164.5996

1.548851

-0.43067

Field

OLMGRPALL

567

-0.59014

104.2684

130.0644

1.2474

-0.22017

Tower

OLMGRPALL

149

-0.16588

121.7104

137.5235

1.129924

-0.122

ALL

OLMGRPALL

1742

-0.34007

220.7673

172.0811

0.779468

0.247863

NF

OLMGRPALL

794

-0.14346

220.7673

169.5039

0.767794

0.262707

EF

PVMRM

232

-0.14628

106.2721

192.0816

1.807451

-0.57522

NF

PVMRM

794

0.17654

220.7673

269.9143

1.222619

-0.20032

ALL

PVMRM

1742

0.048435

220.7673

259.2444

1.174288

-0.16032

Field

PVMRM

567

-0.11716

104.2684

110.9297

1.063887

-0.06191

Tower

PVMRM

149

0.299099

121.7104

92.84917

0.76287

0.269028

EF

GRSM

232

-0.11145

106.2721

147.6259

1.389132

-0.32575

Field

GRSM

567

-0.26831

104.2684

105.8487

1.015156

-0.01504

Tower

GRSM

149

0.217114

121.7104

104.3644

0.857481

0.153454

ALL

GRSM

1742

0.034885

220.7673

142.8789

0.647192

0.428375

NF

GRSM

794

0.259959

220.7673

138.3798

0.626813

0.458796

5.4 Denver-Julesburg Basin, PIattevilie, Colorado NO2 Database

The Colorado NO2 database is comprised of twelve monitors deployed for roughly six weeks
(October 10 - November 16, 2014) at two adjacent oil and gas drilling installations, Pads 1 and

20


-------
2 outside Platteville, CO. At Pad 1, the six upwind (southeast) and six downwind (northwest)
monitors were positioned around the pad on which the emission sources included a drill rig,
five generators, and one small boiler. Similarly, at Pad 2, six monitors upwind and six
downwind were located around the pad with the same emission sources. The monitors were
re-positioned at Pad 2 on three separate occasions to capture NOx emissions during changing
wind patterns for the last three weeks of the monitoring period. Monitor locations for both
Pad 1 and 2 were placed approximately 50-100 m away from the drill rig and support
generators. Hourly varying NOx emissions were modeled for the five diesel-fired support
generators and standby boiler at Pads 1 and 2 during the six-week study period. NOx hourly
emission rate totals at both pads range from roughly 10 Ib/hr to 20 Ib/hr with stack release
heights at 18 ft just at or above the 18 ft high drill rig and 30 total adjacent and nearby
buildings. Non-missing hourly modeling results were paired in time with the available
observations at Pads 1 and 2 (total N pairs = 1473) to generate ranked Q-Q plots and summary
statistics. All five generators show relatively equal contribution to total NOx emissions when
operating at Pads 1 and 2. The standby boiler contributions to total NOx emissions appear to be
negligible. The generator operating loads varied between approximately 50-100% load
throughout the study periods at Pad 1 and 2. For further details on the Colorado field study
monitor locations, hourly emission rates and operating scenarios, background hourly ozone and
NOx data wind sector filtering, and other model configurations see (Colorado Field Study
Workgroup: ERM, 2020). Note that there were more than a dozen small adjacent buildings
located within downwash near wake zones that extend between source and monitor locations
at Pads 1 and 2. Downwash effects from these collections of buildings as well as non-varying in-
stack NO2/NOX ratios assumed for the five generators likely influence model performance
biases and uncertainties for this database.

Figures 8 and 9 show ranked Q-Q plots of AERMOD versus observed NOx and NO2
concentrations at Pads 1 and 2, respectively. NOx model predictions at Pad 1 compare well
with observations with exception to the last three data pairs showing some underprediction.
At Pad 2, roughly half of the upper distribution of NOx predictions show negative bias trending
toward the 1:2 line, suggesting AERMOD may be overestimating dispersion; however, NOx
emissions and/or NO2/NOX ratios inputs may be underestimated perhaps related to
uncertainties in the assumed non-varying in-stack NO2/NOX ratios during varying genset
operating loads. Uncertainties in source characterization of building downwash may also be
contributing to the model estimates at both Pads 1 and 2, especially attributable to movement
of monitoring equipment at three different times around Pad 2. The conservative bias
hierarchy for NO2 options at Pads 1 and 2 is similar to NO2 option performance discussed for
the other three databases with exception to PVMRM. The unreasonable conservative bias
shown for PVMRM at Pads 1 and 2 as it compares to ARM2 may be related to similar model

21


-------
uncertainties discussed for the North Fence monitor at Balko, where enhanced downwash and
entrainment effects on the ensemble plume are overestimated by PVMRM in the immediate
vicinity of recirculation cavities and near wake downwash zones.

NOX Observed (ug/m3)	N02 Observed (ug/m3)

Figure 8 - Colorado NOx and M02 Ranked Q-Q Plots for Pad 1

CO

g 1000

O)
D

CNJ
CO

750

500

CO
CN
>

O 250

a:

LU
<

o

Pad2

N =/753 /

/



/



/

/ +

/

/

-

/ /
/ /



' / „











chem

• NOx

0 250 500 750 1000
NOX Observed (ug/m3)

o>
d

CO

CO
CNJ
>
D
O
2
DC
IXJ
<

300-

200-

100-

0-

Pad2



N =/^012 /

' /
' /L

' A

/

/

/ A
/ M

tJi ^
/



chem
¦ ARM2

•	OLMGRPALL
± PVMRM

~	GRSM

0 100 200 300
N02 Observed (ug/m3)

Figure 9 - Colorado NOx and N02 Ranked Q-Q Plots for Pad 2

Table 4 provides summary FB and RHC statistics for all NO2 modeled options at Pads 1, 2, and
both Pads 1 and 2, sorted by model option and conservative RHC ratio. GRSM shows refined
performance consistent with the ARM2 and OLM, with underpredictions most likely
attributable to source characterization uncertainties. No extreme underprediction or
overprediction is indicated in the RHC fractional bias values shown for GRSM at Pads 1 or 2. In
general, performance for all NO2 options seems more degraded at Pad 2 as compared to Pad 1.

22


-------
Table 4 - Colorado Model Performance Statistics Summary (ng/m3)

Pad

Model Opt.

N

FB

RHC_Obs

RHC_Mod

RHC_ratio

RHC_FB

Padl

NOx

720

0.315616

840.7064

734.6718

0.873874

0.134615

ALL

NOx

1473

0.405245

1250.35

731.5125

0.585046

0.523586

Pad2

NOx

753

0.490945

1289.554

743.2198

0.576338

0.537526

Padl

ARM2

720

-0.02909

196.3466

185.0584

0.942509

0.059192

Pad2

ARM2

753

0.181117

388.8654

253.4506

0.65177

0.421645

ALL

ARM2

1473

0.07837

388.8343

247.755

0.637174

0.443235

Padl

PVMRM

720

0.423036

196.3466

216.9619

1.104995

-0.09976

ALL

PVMRM

1473

0.438573

388.8343

337.1309

0.86703

0.142441

Pad2

PVMRM

753

0.45343

388.8654

335.9631

0.863957

0.145972

Padl

OLMGRPALL

720

0.081058

196.3466

119.3968

0.608092

0.48742

Pad2

OLMGRPALL

753

0.294795

388.8654

188.1758

0.48391

0.695581

ALL

OLMGRPALL

1473

0.19032

388.8343

184.8962

0.475514

0.71092

Padl

GRSM

720

0.525857

196.3466

121.2005

0.617279

0.473291

Pad2

GRSM

753

0.723

388.8654

180.1383

0.463241

0.733658

ALL

GRSM

1473

0.626637

388.8343

157.2032

0.404293

0.848408

6. Summary

Four NO2 model evaluation databases were used to assess the comparative model behavior
and statistical performance of GRSM. These databases represent a broad range of NOx source
characterizations, ozone and NOx model inputs, ozone and NO2 monitoring networks, and local
and regional NOx chemistry and meteorological environments. All database evaluations
included comparisons between GRSM and all existing AERMOD Tier 1 (NOx), Tier 2 (ARM2), and
Tier 3 (OLM and PVMRM) AERMOD NO2 regulatory screening options. Based on the ranked
Q-Q plots showing NOx and NO2 model versus observation concentrations, and with exception
to previously discussed uncertainties and degraded model performance at Pad 2 for Colorado,
GRSM performs within a factor of +/- 2 range of the NO2 observations, and thus, demonstrates
no unacceptable under or over prediction biases. The statistical summaries of RHC and
fractional biases for all NO2 databases further demonstrate GRSM behaves and performs
consistently in comparison with the other existing AERMOD NO2 screening options.

23


-------
References

Carruthers, D. S., Stocker, J. R., Ellis, A., Seaton, M. D., & Smith, S. E. (2017). Evaluation of an explicit NOx
chemistry method in AERMOD. Journal of the Air & Waste Management Association, 67:6, 702-
712, DOI: 10.1080/10962247.2017.1280096.

Cole, H. S., & Summerhays, J. E. (1979). A Review of Techniques Available for Estimating Short-Term N02
concentrations. J. Air Poll. Cont. Assoc., 29:8, 812-817. doi:10.1080/00022470.1979.10470866

Colorado Field Study Workgroup: ERM, A. A.-W. (2020). 2014 Colorado Oil and Gas Drill Rig Field Study
Model Evaluation Database - Technical Support Document. Platteville.

Hanrahan, P. L. (1999a). The Plume Volume Molar Ratio Method for Determining N02/NOx Ratios in
Modeling - Part I: Methodology. J. Air & Waste Manage. Assoc., 1324-1331.

Hanrahan, P. L. (1999b). The Plume Volume Molar Ratio Method for Determining N02/NOx Ratios in
Modeling - Part II: Evaluation Studies. J. Air & Waste Manage. Assoc., 1332-1338.

Panek, J. A. (2020). PRCI ambient N02 AERMOD performance assessment and model improvement

project: Modeled to observed comparison. Journal of the Air & Waste Management Association,
Vol. 70, No. 5, 504-521.

Podrez, M. (2015). An update to the ambient ratio method for 1-h N02 air quality standards dispersion
modeling. Atm. Env., 163-170.

RTP Environmental Associates, Inc. (2013, March 3). Ambient Ratio Method Version 2 (ARM2)for use
with AERMOD for 1-hr N02 Modeling. Retrieved from

http://www.epa.gov/ttn/scram/models/aermod/ARM2_Development_and_Evaluation_Report-
September_20_2013.pdf

Stocker, J. R. (2022). #M023 Development and Development of GRSM for N02 Conversion in AERMOD.
Guideline on Air Quality Models: Developing the Future (p. Conference Proceedings). Durham,
NC: Air & Waste Management Association.

Stocker, J. R., Seaton, M. D., Smith, S. E., O'Neill, J., Johnson, K., Jackson, R., & Carruthers, D. (2023).

Evaluation of the Generic Reaction Set Method for N02 conversion in AERMOD. The modification
of AERMOD to include ADMS chemistry. Cambridge Environmental Research Consultants (CERC)
Technical Report.

U.S. EPA. (1992). Protocol for Determining the Best Performing Model. EPA-454/R-92-025. Research
Triangle Park, NC: Office of Air Quality Planning and Standards.

U.S. EPA. (2015, December). Technical support document (TSD) for N02-related AERMOD modifications.
Publication No. EPA-454/B-15-004. Research Triangle Park, North Carolina: Office of Air Quality
Planning and Standards.

24


-------
Appendix A - GRSM Code Updates and Testing for AERMOD v23132

The GRSM code implemented as a beta option in AERMOD version 22112 (GRSM22112) was
updated in AERMOD version 23132 (GRSM23132) to improve the performance and model
behavior as dependent upon the following calculations:

•	Instantaneous plume spread

•	Building effects on plume spread in near-to-far wake downwash zones

•	Multiple plume effects combined with building effects

This appendix provides clarifying supporting information on the improved performance and
model sensitivities to these calculations as implemented in GRSM23132. The motivation for
code revisions to GRSM in AERMOD version 23132 includes accounting for building effects on
enhanced dispersion and entrainment of ozone and subsequent reaction with modeled NOx
emissions. Treatment of building effects addresses overpredictions for some source and
downwash configurations at receptors located between near and far-wake dispersion zones out
to approximately l-3km downwind where building effected turbulence intensities approach
ambient, non-building turbulence levels. Further discussion and details on the specific
mathematical formulation changes and physics can be referenced in (Stocker J. R., 2022).

A.l Model Performance Evaluation Results Summary

The GRSM23132 code updates were compared to the previous GRSM22112 code for the four
model performance evaluation NO2 databases covered in this TSD. Please refer to the text in
the main body of the TSD for discussion of the specific details on each NO2 database. This
section provides side-by-side model performance comparisons between the two GRSM
formulations in terms of scatter plots, ranked Q-Q plots (Figures A.1-1 through A.l-9), and a
model performance statistical summary shown in Table A.1-1. Each scatter and Q-Q plot show
model results from GRSM23132 and GRSM22112 for each of the four NO2 databases to support
implementation of the updated code implemented in AERMOD version 23132. Note that all
four NO2 databases include building downwash with monitor receptors at distances varying
between 50m and 2.4 km, thus, providing a basis for comparing updates to the building effects
formulations in GRSM23132.

The scatter plot and Q-Q plot for the Pala'au database are shown in Figure A.1-1. The scatter
plot shows model-observation data pairs that are paired in time and space. This provides some
indication of the general agreement between modeled and observed data while also illustrating
combined uncertainties in terms of overall performance of dispersion and NO2 chemistry
predictions. The scatter plot for Pala'au indicates a wider range of model predictions with
GRSM23132, with the slope of the linear model increasing closer to unity than shown for
GRSM22112 concentration pairs, suggesting GRSM22112 agreement with observed NO2 values
may be lacking in statistical range. The larger statistical range of GRSM23132 concentrations
carries forward to the ranked Q-Q plot, where ranked model-observation values are unpaired in
time and space, and indicates better agreement between GRSM23132 predictions and
observations, whereas GRSM22112 slightly underpredicts by comparison, especially at the
upper range of the concentration distribution. Note the Pala'au monitor is located within the

25


-------
near wake building downwash zone and improved agreement with GRSM23132 predictions at
that location indicates treatment of building effects is a contributing factor. Table A.1-1
statistics indicate more favorable RHC and FB results for GRSM23132 further indicating
improved performance from the treatment of building effects on near wake zone dispersion
and NO2 chemistry.

o
00

E o

O)

O
z
Q

O

o
¦¦a-

cr

LU

<

Figure Error! No text of specified style in document..1-1 - Pala'au Monitor Scatter Plot and

Ranked Q-Q Plot

Figures A.1-2 and A.1-3 show the scatter plot and Q-Q plot for the Empire Abo NO2 database at
the North and South Monitoring stations located 1.6km northeast and 2.4km southwest,
respectively. Very little difference in scatter and Q-Q plot data pairs are shown for GRSM23132
and GRSM22112 NO2 predictions. This indicates that there is no significant difference between
the performance of the GRSM code versions for the non-varying emission sources and building
downwash characterizations modeled for Empire Abo at distances extending in the near field 1-
3km. Table A.1-1 indicates a slight increase in fractional bias for GRSM23132. Therefore, the
Empire Abo database was not found to be overly sensitive to the code updates in GRSM23132.

Pala'au - Northwest Monitor (220m)

Scatter Plot

Ranked

0	20	40	60	80

Observed N02 (|.ig/m3)

0	20	40	60	80

Observed N02 (^g/m3)

O -

GRSM23132
y = 0.69 x + 1.98
R2 = 0.394

-H-- GRSM22112
y = 0.56 x +1.91
R2 = 0.467

0 GRSM23132
+ GRSM22112

26


-------
Empire Abo - North Monitor (1,6km)

Scatter Plot

Ranked Q-Q Plot

Observed N02 (ng/m )

Observed N02 (ng/m )

Figure Error! No text of specified style in document..1-2 - Empire Abo - North Monitor Scatter

Plot and Ranked Q-Q Plot

Empire Abo - South Monitor (2.4km)

Scatter Plot	Ranked Q-Q Plot

Figure Error! No text of specified style in document..1-3 - Empire Abo - South Monitor Scatter

Plot and Ranked Q-Q Plot

Scatter plot and Q-Q plot performance comparisons between GRSM23132 and GRSM22112 for the
Balko natural gas compressor station N02 database are shown in Figures A.l-4, A.l-5, A.l-6, and A.l-7
for the Field, North Fence (IMF), East Fence (EF), and Tower monitoring stations, respectively. Generally,
the scatter plots indicate a slight increase in GRSM23132 predictions compared to GRSM22112;
however, Q-Q plots show slight improvements in performance in terms of GRSM23132 predictions
approaching one-to-one agreement between model-observation data pairs at the upper-end of the

27


-------
concentration distribution values shown, most notably for the NF and EF monitoring receptor locations
where near wake building downwash effects are expected to significantly influence model performance.
This is consistent with the overall improved performance shown in Table A.l-1 for RHCs, FBs, RHC ratios,
and RHC FBs at the NF, EF, and Tower receptor locations.

Balko - Field (425 m)

	Scatter Plot			Ranked Q-Q Plot	

o

O

o

E

D)

O
CO

O
CD

O

o

CNJ

o

CN

o
o

E

D)

o

00

o
CD

o

"3-

o

CNJ

° GRSM23132
+ GRSM22112

Observed N02 (ng/m )

Observed N02 (fig/m )

Figure Error! No text of specified style in document..1-4 - Balko - Field (north) Monitor Scatter

Plot and Ranked Q-Q Plot

Balko - North Fence (140 m)

o
o

CN

Scatter Plot

GRSM23132
y = 0.39 x+ 14.95
R2 = 0.287

- + - GRSM22112
y = 0.38 x+ 15.61
R2 = 0.279

150

Observed N02 (ng/m3)

o
o

CM

¦o O

E in

O
z

a
O
5
cc

LLI
<

° GRSM23132
+ GRSM22112

Ranked Q-Q Plot

150

Observed N02 (ng/m3)

Figure Error! No text of specified style in document..1-5 - Balko - North Fence Monitor Scatter

Plot and Ranked Q-Q Plot

28


-------
Figure Error! No text of specified style in document..1-6 - Balko - East Fence Monitor Scatter

Plot and Ranked Q-Q Plot

Balko - Tower (66 m)

Scatter Plot	Ranked Q-Q Plot

Figure Error! No text of specified style in document..1-7 - Balko - Tower (southeast) Monitor

Scatter Plot and Ranked Q-Q Plot

Model performance improvements for the Colorado oil and gas well pad N02 database are indicated in
comparison scatter plot and Q-Q plot Figures A.l-8 and A.l-9 for Pad 1 and Pad 2, respectively.
GRSM23132 shows less tendency for underpredictions at the upper end of the concentration
distribution in scatter plots and Q-Q plots for both Pad 1 and Pad 2. Statistics shown in Table A.l-1
indicate the most improvement of all four databases evaluated for Pad 1 and Pad 2 in terms of RHCs, FB,
RHC FBs, and RHC ratios. The twelve monitor receptors at both Pad 1 and Pad 2 were located well within
the cavity and near wake downwash zones of the well pad structures and short stacks modeled, and as

29


-------
GRSM23132
y = 0.42 x + 8.03
R2 = 0.404

O
o h
CvJ

o GRSM23132
+ GRSM22112

GRSM22112
y = 0.36 x + 7.67
R2 = 0.438

such, model performance improvements for these receptors support implementation of GRSM23132
treatment of building effects over the more simplistic treatments provided by GRSM22112.

Colorado - Pad 1 (50~100m)

Scatter Plot

Ranked

o -

50	100 150

Observed N02 (ng/m3)

50	100 150 200

Observed N02 (fjg/m3)

o -

Figure Error! No text of specified style in document..1-8 - Colorado - Pad 1 Monitors Scatter

Plot and Ranked Q-Q Plot

O
O
CO

O
2

Q

O

o
o

CNJ

01
LU
<

Figure Error! No text of specified style in document..1-9 - Colorado - Pad 2 Monitors Scatter

Plot and Ranked Q-Q Plot

-®- GRSM23132
y= 0.26 x+ 13.01
R2 = 0.308

-+- GRSM22112
y = 0.24 x+ 12.09
R2 = 0.336

Colorado - Pad 2

Scatter Plot	

(50-100m)

Ranked Q-Q Plot

0	100	200	300

Observed NOa (|ig/m3)

0	100	200	300

Observed N02 (ng/m3)

° GRSM23132
+ GRSM22112

O -

O

o

E

"Bi

6

z
Q
O
2
Q1
LU
<

30


-------
Table A.l-1 - GRSM v23132 and v22112 Model Performance Statistics Summary (|ig/m3)

Statistics

Pala'au

Empire Abo

Balko

Colorado

Northwest

North

South

Field

NF

EF

Tower

Pad 1

Pad 2

N

7856

8418

8129

576

803

238

149

720

753

FB V23132

-0.902

0.314

0.016

-0.297

0.285

-0.157

0.196

0.526

0.723

FB V22112

-0.890

0.316

0.016

-0.243

0.277

-0.136

0.217

0.604

0.755

RHC Obs

91.0

130.3

72.6

98.3

217.7

106.3

121.7

196.3

388.9

RHC V23132

82.9

153.0

129.4

118.8

146.3

138.8

118.7

121.2

180.1

RHC V22112

64.1

151.4

128.6

105.8

138.4

147.6

104.4

94.1

153.4

RHC Ratio v23132

0.911

1.174

1.783

1.209

0.672

1.306

0.975

0.617

0.463

RHC Ratio v22112

0.704

1.163

1.773

1.077

0.636

1.389

0.857

0.479

0.394

RHC FB V23132

0.093

-0.160

-0.563

-0.189

0.392

-0.266

0.025

0.473

0.734

RHC_FB_v22112

0.347

-0.150

-0.557

-0.074

0.446

-0.326

0.153

0.704

0.868

In summary, comparisons between GRSM23132 and GRSM22112 for all four NO2 databases, as
shown in scatter plots, Q-Q plots, and statistical metrics, indicate that improvements in
modeled NO2 predictions range between a few micrograms to tens of micrograms per cubic
meter in favor of GRSM23132 implementation as a Tier 3 screening option. Performance
improvements were the most pronounced for the Pala'au and Colorado databases, suggesting
the previous GRSM22112 formulation does not adequately account for building effects on
enhanced ozone entrainment and downward mixing of NOx plumes in the cavity and near wake
building downwash zones. Improvements from the GRSM23132 treatment of building effects
was only slightly indicated for the Balko database, which suggests uncertainties in
characterization of building downwash and emissions (for the single dominant short stack) play
a role in model-observation comparisons independent of the GRSM23132 treatments. In
conclusion, the code updates in GRSM23132 show improved performance over the previous
GRSM22112 version, and therefore, the updated GRSM23132 code was implemented in
AERMOD version 23132.

31


-------
A.2 Sensitivity Modeling Results Comparison

Model behavior and sensitivity comparisons between GRSM23132 and GRSM22112 were
conducted for five stacks modeled at release heights of 35-meters, 50-meters, and 65-meters
with emissions modeled at 1,000 tpy (28.8 g/s) from each stack. The five stacks were positioned
north-to-south centered along two adjacent multi-tiered buildings, thus, configured to illustrate
differences between GRSM23132 and GRSM22112 treatments of building effected
instantaneous ensemble plume dispersion and ground-level concentrations in the near-wake
(or building downwash cavity), far-wake (adjacent to the near-wake) zones, and subsequent
transition to the near-field distances between 1km and 3km downwind. Figure A.2-1 shows the
positioning and multi-tier building complex modeled for the 35-meter, 50-meter, and 65-meter
sensitivity scenarios; note the buildings and stacks represent a hypothetical installation and are
located and centered at a coastal North Carolina (NC) airport for GRSM sensitivity modeling
purposes only. The two adjacent buildings stretch end-to-end 196 meters with varying tier
heights. The north building tier heights range 18-29 meters, and the south building tier heights
range 11-18 meters. Stack parameters modeled for the 35-meter, 50-meter, and 65-meter
scenarios include an exit temperature of 311 K, exit velocity of 7.35 m/s, and stack diameter of
5 meters. BPIPPRM (version 04274) was used to process building tier and stack location data to
generate building downwash parameters for input to AERMOD. The building and stack
configurations for the three scenarios represent stack height to building height ratios ranging
between values of 1.2 and 6.5. This range of stack and building height combinations were
included to assess model formulation sensitivities to building downwash and ozone
entrainment behaviors under varying meteorological conditions and at varying receptor
distances.

Modeled receptors were located and oriented around the stacks and buildings to better
identify model formulation sensitivities relevant to the building effects updates in GRSM23132.
As such, receptors were positioned at 10-meter spacing surrounding the perimeter of the two
buildings at the following distances: 10m, 15m, 20m, 30m, 50m, 75m, 110m, 170m, 250m, and
380m. Fine gridded receptors with 100-meter spacing extending 500m-3km. Medium gridded
receptors were modeled with 500-meter spacing extending 3km-15km. Elevations for modeled
receptors were based on 1-arc-second USGS terrain data and AERMAP (version 18081) default
processing options; receptor elevations indicate relatively flat terrain and are consistent with
the coastal study location.

AERMET meteorological inputs to AERMOD were taken from a previous air quality modeling
study given it was readily available and that the scope of the GRSM sensitivity focuses on
building effects on NO2 chemistry and dispersion in the near-wake, far-wake, and near field.
AERMET (version 19191) raw data inputs included hourly surface temperature, hourly surface
wind data, and twice-daily upper air data from a coastal NC airport for a 3-year modeling period
2015-2017. Other AERMET raw data and pre-processing included ASOS 1-minute wind data
processed with AERMINUTE (version 15272), 1992 National Landcover Cover Dataset (NLCD)
processed with AERSURFACE (version 13016) to generate surface parameters, and AERMET

32


-------
regulatory default run options.

GRSM NO2 mode! input options were configured the same for all scenarios. Hourly ozone for
the 2015-2017 modeling period was developed from an isolated rural monitoring station
located in Lee County, NC, The AERMOD default ozone value of 0.060 ppm was used for
substitution of missing hourly ozone values. Hourly NOx values were derived from the same Lee
County rural monitoring station, and based on an equilibrium ratio of 0.9 NCb/NOx, given that
only hourly MO2 data was available; i.e., hourly NOx values were calculated equal to hourly NO2
divided by the 0.9 NO^/NOx equilibrium ratio. Note the Lee County monitor was located in a
farm field greater than 10-20km away from any significant sources of mobile or stationary NOx
emissions sources. Season-diurnal-hourly varying NO2 concentrations were developed based on
the 3-year average, highest-3rd-highest seasonal values from the Lee County hourly NO2 values
2015-2017. An in-stack NO2/NQX emission ratio of 10% was modeled for all five stacks for the
three sensitivity modeling scenarios.

Figure Error! No text of specified style in document..2-1 - Sensitivity modeling stack and
complex building configuration (graphics shown in Google Earth© and created with Lakes
Environmental Software, Inc. AERMOD View©). Note: stacks shown are 30-meters tall.

Figures A.2-2 through A.2-10 show model concentration contours and difference plots from
GRSM23132 and GRSM22112 and for the three sensitivity modeling scenarios. The model
results are represented with concentration isopleths derived from the highest-8th-highest 1-
hour NO2 modeled concentrations and correspond with the model design value used in 1-hour
NO2 NAAQS modeling demonstrations (using standard kriging in Surfer version 27.1.229 ®,
Golden Software, LLC). Fine and medium gridded receptors are visible as small "+" symbols in
each contour plot. Notably, ground level concentrations for the 35-meter stack scenario show

33


-------
modeled exceedances of the 1-hour NO2 NAAQS of 188 ng/m3 (indicated by a red isopleth) with
both GRSM23132 and GRSM22112, whereas modeled concentrations for the 50-meter and 65-
meter scenarios show a decrease, relative to the 35-meter scenario, ranging approximately
20%-50% (well below the NAAQS) at all receptors independent of the GRSM code version.
Further discussion of model results for each scenario continues below.

GRSM23112 predictions for the 35-meter stack scenario results depicted in Figure A.2-2 range
between approximately 10 |-ig/m3 and 200 |-ig/m3 and show approximately 20 |-ig/m3 higher
concentrations at receptors located within near-wake zone distances (~10-100 meters) as
compared to GRSM22112. Similarly, GRSM22112 maximum concentrations depicted in Figure
A.2-3 range between 10 |-ig/m3 and 200 |-ig/m3, but in contrast are roughly 20 ng/m3 higher at
receptor distances extending from the far-wake (~100-1000 meters) to the near field and out to
l-3km. These differences are spatially isolated and shown in Figure A.2-4 and indicate the
GRSM23132 treatment of multiple building effected plumes lowers concentrations in the far
wake and near field while enhancing concentrations by equal measure in the near-wake
downwash zones. The 35-meter stack scenario stack height to building height ratio values
range from 1.2 to 3.5, and thus, NOx plume size, ozone entrainment, and dispersion is heavily
influenced by building downwash as indicated by the maximum concentrations at near and far-
wake receptors. And notably, the GRSM23132 building effects treatment for the 35-meter
stack and building configuration shows equal sensitivity to modeled concentrations in the near-
wake, far-wake, and near field. The 35-meter stack scenario stack and building configuration,
and thus, the heavy influence of downwash, is most comparable to the Pala'au, Colorado, and
Balko NO2 evaluation database configurations as well as the distance to near-wake receptor
monitor locations. As such, the enhanced concentration model behavior for near-wake
receptors shown by the 35-meter stack scenario compliments the improved model
performance for similarly configured NO2 databases while maintaining decreases or no change
in modeled concentrations at far-wake and near field receptors.

The 50-meter stack scenario shows less pronounced spatial differences between GRSM code
versions than what is shown for the 35-meter stack scenario results at far-wake and near field
receptor distances; however, enhanced concentrations at near-wake receptors are relatively
more sensitive to GRSM23132 building treatments for the 50-meter stack scenario, albeit at
concentrations approximately about 50% lower than those shown for the 35-meter stack
scenario. Figure A.2-5 shows the GRSM23132 concentrations are nearly the same as
GRSM22112 concentrations shown in Figure A.2-6 in the near field and far-wake downwash
zones, and near-wake concentrations modeled with GRSM23132 are enhanced by the building
effects treatment formulation. Concentration differences shown in Figure A.2-7 indicate
GRSM23132 concentrations for the 50-meter stack scenario are enhanced in the near-wake
downwash zone with increases on the order of 50 ng/m3. The enhanced concentrations
predicted by GRSM23132 in the near-wake zone is complimented by similar enhanced
concentrations modeled for several NO2 model performance databases with similar building
effected monitor receptor exposure locations where improved model performance, as
previously noted, was shown for Pala'au, Colorado, and some Balko monitors. The sensitivity of
the 50-meter stack concentrations to GRSM23132 building effects treatments in the near-wake

34


-------
zone indicates vertical dispersion becomes relatively more important for taller stacks, or model
setup configurations with stack height to building height ratios greater than approximately 1.5;
however, this enhanced sensitivity is compensated by lower near-wake concentrations
predicted from elevated plume releases and dampened building downwash effects on plume
sizes and entrainment of ozone. The converse is observed for stack height to building height
ratios below approximately 1.5, where the building effects treatment formulation behavior
lowers the sensitivity of increased concentrations in the near-wake while decreasing
concentrations due to enhanced lateral dispersion and plume sizes at receptors in the far-wake
and near-field.

The 65-meter stack scenario shows model sensitivities similar to the 50-meter stack scenario
results, with enhanced concentrations in the near-wake and less differences at far-wake and
near-field receptors between GRSM code versions. Again, this is most likely attributed to the
more elevated release height of the plume relative to the building height and the GRSM23132
building effects formulation, with the 65-meter stack height to building height ratios ranging
between values of approximately 2.1 and 6.5. Figure A.2-8 shows the GRSM23132
concentrations are nearly the same as GRSM22112 concentrations shown in Figure A.2-9 in the
near field and far-wake downwash zones, and enhanced GRSM23132 hourly concentrations
increasing on the order of 50 ng/m3 within the first hundred meters of the near and far-wake.
Figure A.2-10 confirms that differences between GRSM23132 and GRSM22112 are relatively
unchanged in the far-wake and near field and increased in the near-wake. Near-wake receptor
performance improvements indicated by sensitivities shown for the 65-meter stack scenario
are less certain than for the 35-meter and 50-meter stack scenarios given that the stack height
to building height ratios for all NO2 databases are closer to values on the order of 1.1 or less,
and thus, are not comparable to the 65-meter stack scenario.

In summary, the code refinements evaluated and implemented in GRSM23132 are supported
by improved model performance discussed in the previous section and expected code behavior
shown by the sensitivity modeling analyses discussed here. GRSM23132 NO2 predictions show
the building effects treatments are less apparent for the taller 50-meter and 65-meter stack
scenarios with respect to modeled impacts at receptor distances in the far wake and near field;
however, notable enhanced concentrations are shown in the near-wake. The building effects
treatments are most pronounced for the 35-meter stack scenario at all receptor distances with
lower concentrations modeled at receptors in the far-wake and near field, and equally higher
concentrations modeled at receptors in the near-wake building downwash zone. The enhanced
higher concentrations modeled for the 35-meter stack scenario at near-wake receptors
compliments the improved model performance shown for similarly configured NO2 databases,
and therefore, supports this model behavior. The near-wake behavior of GRSM23132 building
effects treatment shows enhanced dispersion and entrainment of ozone that likely contributes
to enhanced NO2 concentrations for shorter stacks and lower stack-to-building height ratios of
approximately 1.5 or less. Ultimately, the behavior of the building effects treatment in
GRSM23132 is shown to be sensitive to varying model inputs for stack and building heights,
improves model performance, and is consistent with the motivation and expected range of
predicted NO2 concentrations at receptors located at distances heavily influenced by building

35


-------
downwash.

3885000

3880000-

3875000

3870000

(Mg/m

300000 305000 310000 315000 320000 325000

Figure Error! No text of specified style in document..2-2 - GRSM v23132 35-meter Tall Stacks

Highest-8th-High 1-hour NO2

36


-------
3895000

3890000

3885000

3880000

3875000-1

3870000

(Hg/m

300000 305000 310000 315000 320000 325000

Figure Error! No text of specified style in document..2-3 - GRSM v22112 35-meter Tall Stacks

Highest-8th-High 1-hour NO2

3895000

3890000

3885000

3880000

3875000

3870000





—

18



14



10

—

6



2



-2

1

6

i

-10



-»

¦

-18

(f-ig/m3

300000 305000 310000 315000 320000 325000

37


-------
Figure Error! No text of specified style in document..2-4 - GRSM v23132 Minus v22112 35-
meter Tall Stacks Highest-8th-High 1-hour NO2

3895000

3890000

3885000

3880000

3875000

3870000





—

100

—

90

—

80

—

70

-

60

-

50

1

40

1

30

L

20

¦

10

(|ag/m3

300000 305000 310000 315000 320000 325000

Figure Error! No text of specified style in document..2-5 - GRSM v23132 50-meter Tall Stacks

Highest-8,h-High 1-hour NO2

38


-------
3875000

3870000

10
(l-ig/m3

300000 305000 310000 315000 320000 325000

Figure Error! No text of specified style in document..2-6 - GRSM v22112 50-meter Tall Stacks

Highest-8th-High 1-hour NO2

3895000

3890000

3885000-

3880000-

3895000

3890000

3885000

3880000

3875000

3870000

50
45
40
35
30
25

15

10

t

-5
(Hg/m3)

300000 305000 310000 315000 320000 325000

39


-------
Figure Error! No text of specified style in document..2-7 - GRSM v23132 Minus v22112 50-
meter Tall Stacks Highest-8th-High 1-hour NO2

3895000

3890000

3885000

3880000

3875000

3870000

—

80

—

75

—

70

—

65

—

60



55



50

—

45



40

—

35

n

30

-

25

-

20

-

15

¦

10

(Mg/m

300000 305000 310000 315000 320000 325000

Figure Error! No text of specified style in document..2-8 - GRSM v23132 65-meter Tall Stacks

Highest-8th-High 1-hour NO2

40


-------
3875000

3870000

300000 305000 310000 315000 320000 325000

Figure Error! No text of specified style in document..2-9 - GRSM v22112 65-meter Tall Stacks

Highest-8th-High 1-hour NO2

3895000

3890000

3885000-

3880000-

—

80

—

75

—

70

—

65

—

60

—

55

—

50



45



40

¦

35

n

30

¦

25

r

20

-

15

¦

10

(Mg/m3

41


-------
3895000

3890000

3885000

3880000

3875000

3870000

—

48

—

43

—

38

—

33

-

28

-

23



18



r





¦

-2

(f-ig/m3

300000 305000 310000 315000 320000 325000

Figure Error! No text of specified style in document..2-10 - GRSM v23132 Minus v22112 65-

meter Tall Stacks Highest-8tb-High 1-hour NO2

42


-------
United States	Office of Air Quality Planning and Standards	Publication No. EPA-454/R-24-005

Environmental Protection	Air Quality Assessment Division	November 2024

Agency	Research Triangle Park, NC

43


-------