Project: RFP # FAAA 2016000120 (FRAPPÉ and DISCOVER-AQ Data Analysis)

Final Report, July 2017

Process-Based and Regional Source Impact Analysis for
FRAPPÉ and DISCOVER-AQ 2014

Principal Investigators: Gabriele Pfister, Frank Flocke
Co-Investigators: Rebecca Hornbrook, John Orlando
Post-Doc: Sojin Lee
With contributions by Jason Schroeder, NASA Langley Research Center

National Center for Atmospheric Research (NCAR)
Atmospheric Chemistry Observations and Modeling Laboratory (ACOM)
3450 Mitchell Lane, Boulder CO 80301

Date: 31 July 2017

 Overview
Two major field campaigns – the National Science Foundation (NSF)/National Center for
Atmospheric Research (NCAR) and State of Colorado Front Range Air Pollution and
Photochemistry Éxperiment (FRAPPÉ) and the 4th deployment of the National Aeronautics and
Space Administration (NASA) Deriving Information on Surface conditions from Column and
Vertically Resolved Observations Relevant to Air Quality (DISCOVER-AQ) – were conducted
jointly in summer 2014 to study summertime ozone pollution in the Northern Colorado Front
Range Metropolitan Area (NFRMA).
The NFRMA is located at an elevation of roughly 1600-1800 m, on the plains just east of the
Central Rocky Mountains. The mix of diverse anthropogenic pollution sources including urban
sources, power plants, large industrial sources, agricultural activities, and oil and gas extraction
contribute with varying degree to the local ozone production. Project RFP # FAAA 2016000120
has been focused on characterizing the contributions of the different photochemical drivers to
surface ozone in the NFRMA through analysis of the FRAPPÉ and DISCOVER-AQ data sets
and accompanying modeling studies.
This final report builds upon work described in the previous reports and presents our findings on
the evaluation and updates of available emission inventories and the assessment of estimated
source contributions. We expect that this report will enable the CDPHE to develop emission
control scenarios to address the ozone non-attainment status of the NFRMA. The results included
in this report are based on analysis using the FRAPPÉ and DISCOVER-AQ observational data
sets, the Sparse Matrix Operating Kernel (SMOKE) modeling system, the Community
Multiscale Air Quality (CMAQ) model, the chemical boxmodel BOXMOX (an extension to the
Kinetic PreProcessor (KPP) and the NASA LaRC Steady State Model. Our derived source
contributions are based on a zero-out approach of the major contributing emission sectors in
CMAQ. Not all analysis and results from the two Interim Reports from December 2016 and
April 2017 are included in here, but only repeated or referenced as needed to support the
discussed analysis and final conclusions.
The main findings from this work are the following:
•

•

•

Typical summer high ozone days exhibit upslope winds during the day characterized by
transport of emissions to the west and into the mountains. Often, we observed at least a
partial separation between air masses originating further north (oil and gas related
emissions) and further south (Denver area urban and industrial emissions); some mixing
can occur, especially if the upslope winds have a northeasterly component.
Specific meteorological conditions such as the Denver Cyclone can have a large
influence on the magnitude of ozone production by more efficiently mixing VOC from
oil and gas sources with NOx emissions in the urban areas. This causes a different, more
local distribution of ozone, concentrated in eastern Boulder, Gilpin, and northern
Jefferson counties.
Mobile sources and oil and gas related emissions are the largest contributors to local
ozone production in the NFRMA. On average, oil and gas emissions show a stronger
influence in the northern part of the NFRMA and the northern foothills, while mobile

Final Report for RFP # FAAA 2016000120

1

 •
•

1

emissions dominate farther south and in the southern foothills. Both sectors contribute, on
average, 30-40% each to total NFRMA ozone production on high ozone days.
Industrial emissions contribute somewhat less to NFRMA ozone on average, but can
dominate locally, as is the case in and around Commerce City.
NOx from power plant emissions as well as from Denver International Airport causes
ozone titration locally due to large NOx emissions in a confined area. Further downwind
the additional NOx can result in net ozone production in the western part of the NFRMA
on high ozone days, albeit over a limited area. Reduction of NOx from these sources will
result in higher ozone locally and slightly lower ozone in areas further downwind.

WRF Meteorological Simulations and CMAQ Modeling Setup

The CMAQ modeling work is conducted using the latest CMAQ version v5.2 beta with the
carbon bond mechanism version 6 (CB6r3) mechanism (we acknowledge Deborah Luecken
(U.S. EPA) for providing the model version ahead of a public release and for help with running
the model. WRF v3.8.1 is run to create hourly meteorological fields for driving CMAQ. In here
we only provide a brief overview of the model setup, for a more detailed description we refer to
Interim Report II (April 2017).
The CMAQ domain is comparable to the domains used in the Denver SIP modeling and includes
a 12 km x 12 km outer domain (d01; EPA CONUS12) covering the Western U.S. and a 4 km x 4
km inner domain (d02; EPA CONUS4) over Colorado. The model top is set at 50 hPa with 37
vertical levels. We use NCAR ECMWF analysis fields (available from NCAR’s Research Data
(RDA), https://rda.ucar.edu/) for meteorological initial and boundary conditions and Real Time
Air Quality Monitoring System (RAQMS) model output (provided by Brad Pierce, NOAA
NESDIS) for chemical initial and boundary conditions.
We applied analysis and observational nudging in the WRF simulation following suggestions
from the Western State Air Quality Modeling Study (WSAQS) (UNC, 2016). In brief, analysis
grid nudging is applied to the outer domain (d01) with nudging for temperature and water vapor
mixing ratio turned off in the boundary layer. Observation nudging was performed for the inner
domain (d02) for winds, temperature and water vapor mixing ratio. For our first model
simulations, we included NCEP ADP Global Upper Air and Surface Observational Weather Data
available from the NCAR Research Data Archive (RDA) website in the observational nudging.
Tests with adding wind, temperature and dew point observations from FRAPPÉ and
DISCOVER-AQ in the observation nudging showed a small but notable improvement, which is
why the latter has been used in the final runs. A discussion on the performance of model
transport and meteorology is given in Section 2.
Emission inputs for CMAQ are based on a set of different inputs which are summarized in Table
1 in Interim Report II (April 2017). We acknowledge Dennis McNally (Alpine Geophysics) for
providing the 2017 onroad mobile inventory data and Dale Wells (CDPHE) for providing the
2014 oil and gas (O&G) emission input data. EPA emission totals were derived from the EPA
2011 v6.3 platform (ftp://ftp.epa.gov/EmisInventory/2011v6/v3platform/reports) and the EPA
2014 Emissions Inventory System (EIS) sector summary (https://www.epa.gov/air-emissionsinventories/2014-national-emissions-inventory-nei-data ;download date: 05/04/2017). The EPA
2017 emission information is derived from the emission summary available from
Final Report for RFP # FAAA 2016000120

2

 ftp://ftp.epa.gov/EmisInventory/2011v6/v3platform/reports/2017ek_county_monthly_report.xlsx
(download date: 03/28/2017). At the time of the study, EPA 2014 SMOKE-ready input files had
not yet been available. The emission inventory derived from these inputs is defined as “a priori”
or “S0” in this document.
In Table 1 we list the emission totals per county and also the county total per sector for the S0
inventory. Onroad mobile emissions account for about 1/3rd of all NFRMA anthropogenic
emissions with largest contribution from the Denver counties. O&G sources amount to the
second largest NOx sources with most of them attributed to Weld County. NOx from industrial
sources is largest in Denver county, where Denver International Airport (DIA), which is
considered an industrial source, represents a major emitter. Cherokee power plant in Adams
County is the largest contributor to CEM emission followed by the Valmont power plant in
Boulder county. For VOC, O&G activities, mostly in Weld County, account for about 50% of all
NFRMA emissions. Industrial sources are the second largest contributor with most emissions
located in the Denver counties.
Table 2 compares the a priori (S0) emission estimates for CO, NOx and VOCs to estimates from
EPA 2011, 2014 and 2017. Highest emission values for all species are found in NEI 2011 as this
inventory does not include more recent emission controls and cleaner technologies considered in
the later inventories. S0 emissions for CO are in line with EPA 2014 and higher than in EPA
2017. For NOx, our inventory is closer to EPA 2017 than EPA 2014 which likely is due to our
inventory also using mobile emissions for a 2017 projection. S0 has the lowest VOC total
emissions from all four inventories with about 12% lower totals compared to EPA 2014. EPA
2017 is notably higher with similar Weld County emissions as EPA NEI 2011.
In Interim report II (April 2017; Tables 1 and 2) we compared CO, NOx and VOC emission
totals for the mobile sector and the oil and gas sector for S0 and the different EPA NEI data sets.
This comparison showed that the mobile emission totals used in our study are overall lower
compared to EPA 2011 and EPA 2014 and either of similar magnitude or slightly higher when
compared to EPA 2017. This is also the case for the major contributing counties of Denver,
Jefferson and Arapahoe. NOx mobile emissions in our study as well as in EPA 2017 are about
half of the estimates in NEI EPA 2011 and 2014 in Denver, Jefferson and Arapahoe counties,
which are the counties with the highest mobile emissions. Our study totals for Weld and Larimer,
in contrast, are notably higher compared to EPA 2017 and in the case of CO closer to EPA 2011
and in the case of NOx and VOC closer to EPA 2014. For the oil and gas sector most of the
contribution comes from Weld county and while Weld county totals for our study are higher in
CO and NOx compared to all other inventories, our VOC totals are lower by 39%, 17% and 37%
compared to EPA 2011, EPA 2014 and EPA 2017, respectively.

Final Report for RFP # FAAA 2016000120

3

 Table 1: Sector emissions for NOx and VOCs for the S0 a priori emissions. Emission totals are listed per county together with the county total.
Offroad
Adams
Arapahoe
Boulder
Broomfield
Clear Creek
Denver
Douglas
Elbert
Gilpin
Jefferson
Larimer
Park
Weld
Total Result

1,311
1,349
880
121
26
1,548
982
88
70
1,434
885
78
1,116
9,886

Onroad
4,350
4,068
2,107
544
455
4,496
2,938
426
219
3,790
4,225
272
6,716
34,605

Nox (tons/year)
O&G
Industrial
1,497
480
24
3
7
8
11
21
19,783
21,835

1,980
1,289
1,591
104
69
5,343
502
52
21
1,739
922
38
1,501
15,151

CEM
5,447
2,067
93
12
1,420
626
9,664

Others
53
77
66
7
1
82
35
3
1
75
69
3
50
522

Anthro. Total Offroad
14,639
7,262
6,735
779
552
11,568
4,456
578
311
7,060
7,542
390
29,791
91,663

1,163
1,853
1,019
98
70
1,419
998
83
34
1,673
1,305
281
917
10,913

Onroad
2,391
2,933
1,467
336
164
3,014
1,588
215
72
2,815
2,341
197
2,572
20,105

VOC (tons/year)
O&G
Industrial
1,173
933
233
85
29
42
6
305
75,665
78,471

6,498
5,481
2,710
636
124
6,682
2,563
255
41
6,068
3,022
132
4,192
38,405

CEM

Others

45
14
5
5
38
18
126

532
774
683
73
16
859
349
31
9
747
720
28
522
5,343

Anthro. Total
11,802
11,975
6,126
1,229
373
12,009
5,498
627
156
11,313
7,731
639
83,886
153,363

Note: The O&G sector includes area and point sources from oil and gas exploration. Industrial combines the SMOKE other nonpoint (nonpt) and non-Integrated
Planning Model (non-IPM) sector.

Table 2: Comparison of CO, NOx and VOC emissions for our a priori (SO) inventory and EPA NEI 2011, 2014 and 2017. Emission totals are
listed per county together with the county total.
Anthropogenic Emission
CO (tons/year)
Nox (tons/year)
VOC (tons/year)
EPA 2011 EPA 2014 This study EPA 2017 EPA 2011 EPA 2014 This study EPA 2017 EPA 2011 EPA 2014 This study EPA 2017
Adams
Arapahoe
Boulder
Broomfield
Clear Creek
Denver
Douglas
Elbert
Gilpin
Jefferson
Larimer
Park
Weld
Total Result

64,200
82,911
43,153
7,550
6,703
85,003
40,635
5,539
1,357
78,715
55,140
5,385
65,222
541,513

55,351
70,066
35,798
6,615
4,407
74,863
36,085
3,955
959
64,137
40,647
3,686
62,041
458,609

52,314
67,645
35,160
6,578
3,922
69,656
34,690
3,761
915
58,867
48,261
3,417
64,472
449,655

49,625
63,302
33,214
5,403
4,097
63,937
35,741
3,827
838
57,756
38,127
3,507
48,628
408,002

Final Report for RFP # FAAA 2016000120

23,538
11,492
8,981
1,347
1,717
18,683
7,297
1,461
450
13,147
10,124
821
30,010
129,067

17,612
10,190
8,431
1,297
1,654
15,403
6,876
989
394
10,734
7,907
549
31,293
113,328

14,639
7,262
6,735
779
552
11,568
4,456
578
311
7,060
7,542
390
29,791
91,663

14,009
7,296
6,902
822
989
11,623
5,089
963
317
8,427
6,745
449
24,487
88,115

16,389
14,328
9,878
2,019
714
14,613
6,804
1,148
258
15,065
23,363
3,820
133,584
241,983

12,523
12,710
6,279
1,326
550
12,688
5,677
737
166
11,372
7,599
707
101,677
174,009

11,802
11,975
6,126
1,229
373
12,009
5,498
627
156
11,313
7,731
639
83,886
153,363

16,215
14,328
8,555
2,019
619
14,613
6,565
1,127
213
14,104
9,897
920
132,978
222,153

4

 2

Model Evaluation – Meteorology & Dynamics

For comparison of modeled wind speed and direction to C-130 aircraft observations, hourly
instantaneous WRF output from the 4 km x 4 km inner domain (d02) was interpolated to the time
and location of the 1-minute merged aircraft data. For comparison to surface sites, model grids
containing the site location were selected and model wind speed and direction at 10 m altitude
above ground level (a.g.l.) were compared to 1-hour averaged surface observations. Only data
are included where both observations and model have available data.
As mentioned earlier, we apply observation nudging in the WRF/CMAQ inner domain and
performed tests whether the addition of FRAPPÉ data improves the model performance of the
wind fields. We included wind, temperature, and dew point observations from C-130 flights,
valid P-3B aircraft wind, temperature and dew point data and available wind and temperature
measurements from CDPHE sites. The evaluation with FRAPPÉ data is not completely
independent as we compare in parts to the same observations the model is nudged to and have
only a limited observational data set for a truly independent evaluation.
Appendix A, Figure A1 shows observed and modeled wind roses for C-130 aircraft data for the
two different nudging methods, defined here as ADP nudging and ADP/FRAPPÉ nudging,
respectively. The changes with ADP/FRAPPÉ nudging are small given that the FRAPPÉ data
add limited temporal and spatial information and the model is only nudged to the degree that the
underlying model dynamics allows. Overall, however, the addition of FRAPPÉ data improves
the agreement with C-130 data. Using the wind criteria applied in our analysis of the emission
evaluation (Interim Report II, Section 2.2 and Section 3 below), we find an increase in the
number of matching cases for the 10-17 LT time window by about 3-5% dependent on species.
The comparison of modeled and measured wind roses is shown in Appendix A, Figure A2 for all
surface sites. Note that sites not operated by CDPHE such as Longs Peak, Trail Ridge Road,
NREL-Golden, Table Mountain, Platteville or Fort Collins (FC) West were not included in the
nudging. In general, the changes between ADP nudging and ADP/FRAPPÉ nudging are small
but noticeable. The model represents wind speeds overall well at NFRMA sites, but is too high at
NREL Golden, CAMP and FC CSU and too low at RF-North, Chatfield Park, and Aurora East.
ADP/FRAPPÉ nudging helps correcting for these biases. As for wind direction, we see that
model results tend to have a stronger upslope component than measured at the surface sites. A
possible reason for this is that the model better represents the regional picture because it does not
resolve local topography or large buildings, which impact many of the NFRMA sites. This is
clearly seen at RF North with the model showing the typical daytime upslope winds, while actual
winds at this site are strongly influenced by nearby canyons and the proximity to the Foothills.
ADP/FRAPPÉ nudging clearly improves the model performance at this site.
The local nature of the surface sites versus the more regional nature of the model is also reflected
when comparing FC CSU and FC West. Both sites are close to each other, but FC CSU is located
in downtown Fort Collins and likely influenced by the urban environment as wind speeds are
also clearly lower compared to FC West. FC West is only little influenced by local topography or
urban disturbances and reflects the regional SE upslope flow. The dominance of a more regional
flow in the model is also seen when comparing La Casa and CAMP in downtown Denver. These
Final Report for RFP # FAAA 2016000120

5

 sites are only 3.5 km apart, but feature a quite different wind pattern in the observations, which is
not resolved by the model.
In addition to grid resolution, thunderstorms might also be contributing to the stronger upslope
component in the model. During FRAPPÉ, thunderstorm activity frequently interrupted flow
patterns and ozone production and the forecast models tended to underpredict thunderstorms and
clouds. We do not have the relevant data from FRAPPÉ available to evaluate the thunderstorm
predictions in WRF/CMAQ and adding an assessment with operational radar data cannot be
conducted with the provided resources. However, we compare the model to measured
downwelling shortwave radiation at three surface sites (see below) which demonstrates that our
WRF/CMAQ simulation also tends to underpredict clouds.
Winds for the north-eastern NFRMA are available from two sites, WC Tower and Platteville.
While Weld County winds show the upslope flow in both observations and model, the Platteville
measurements reveal a highly variable pattern without a dominance of upslope winds. This
suggests that the local topography disturbs the flow patterns at this site, likely because it is
situated in a shallow topographical depression that is not resolved in the model. The two high
altitude sites Longs Peak and Trail Ridge Road, which are considered as being more
representative of the free tropospheric flow regimes, show a strong component of winds from the
southeast in the measurements. This is well reproduced by the model and highlights the
frequently encountered upslope conditions. The model wind speed at the two mountain sites is
biased high which is likely related to the sites not fully representing free tropospheric conditions
due to small scale, local topography and unresolved surface roughness.
In Figure 1 we assess the performance of modeled winds, relative humidity and temperature
using measurements collected from ozone sonde launches in Platteville and FC West. For the
sonde data we do not filter for the time of the day, but all sonde launches took place during
daytime. At Platteville, the model has challenges representing the wind direction at the
lowermost altitudes and also overestimates the wind speed, similar to what has been seen in the
comparison to surface wind observations. Above 0.5 km a.g.l. the model is in close agreement
with measurements in representing the transition to the free tropospheric westerlies. Temperature
and relative humidity are well captured in the model but the model appears biased high at higher
altitudes. For FC West, we find a similar overestimate in relative humidity as for Platteville.
Wind direction and wind speed, which were simulated well in the model when compared to
surface sites, also show good agreement with the sonde data.
Meteorological profiles measured at BAO (Figure 2) demonstrate a good model performance for
winds and temperature for both magnitude and profile shape. The relative humidity tends to be
overestimated at higher altitudes in line with findings from ozone sonde launches at Platteville
and FC West. There are two possible causes that could explain this bias. As will be shown in the
evaluation of vertical ozone profiles and has been discussed in Interim Report II (April 2017),
specifically the last week of July 2014 was characterized by enhanced free tropospheric ozone in
the Front Range, which is underestimated in the model. The origin of this high ozone is not fully
determined, but believed to be due to a combination of large scale transport of ozone, possibly
mixed with stratospheric ozone. The higher relative humidity in the model might suggest an
underrepresentation of stratospheric air. Another possible explanation might be that the model
keeps more moisture in the air while in reality clouds and rain removed moisture from the air.
Final Report for RFP # FAAA 2016000120

6

 Figure 1: Vertical average profiles for measured (black) and modeled (red) wind direction and wind
speed, temperature and relative humidity for ozone sonde launches in Platteville(top) and Fort Collins
West (bottom).

Figure 2: Vertical average profiles for measured (black) and modeled (red) wind direction and wind
speed, relative humidity and temperature from the Atmospheric Emitted Radiance Interferometer (U
Wisconsin) at BAO.

Final Report for RFP # FAAA 2016000120

7

 While the model shows reasonable performance for the wind fields, there are notable
discrepancies that are related to uncertainties in the model transport, but in parts are also due to
the local nature of the measurements. Overall, however, the model is able to represent the
characteristics of NFRMA wind patterns and given confidence in the emission inputs, we can put
trust in using the model for estimating sensitivities of surface ozone to different source types. We
emphasize that our focus of the evaluation of modeled meteorology has mostly been on winds.
Clearly, reasonable model representation of clouds, rain, and thunderstorms is crucial, and the
comparison of the model to solar radiation measurements available for four surface sites during
FRAPPÉ offers indirect evaluation of the model’s ability to represent clouds.
These comparisons are shown in Figures 3 through 6 for Platteville, BAO, Table Mountain and
NREL Golden, respectively. Note that for the first three sites measurements are available for 11
July – 12 August, while for NREL Golden data are only provided for 15 July-10 August. The
latter site also appears to be subject to a time shift in the measurements for the first few days of
the data record. This, however, does not impact the interpretation.
For BAO, Platteville and Table Mountain, which are all located north of the Denver Metro Area,
there were four days with cloud-free conditions (22 July, 2&3 August and 12 August) in both the
observations and the model; these days were also high ozone days and in parts are discussed
later. August 2 and 3 measurements and model also show cloud-free skies at NREL Golden, but
22 July measurements indicate some clouds later in the day while the same days modeled solar
radiation suggests cloud-free conditions. On most other days, the model underpredicts the
amount of clouds, which has implications on the ozone production as will be discussed later.
Day-to-day variability is overall similar at the four sites but localized differences are common.
An example of the influence of localized cloudiness is 19 July, when measured solar radiation
suggests a cloud-free day at Platteville with more clouds building towards BAO and even more
clouds towards Table Mountain located further west and at NREL Golden to the south. The
model does not simulate notable clouds at any of the sites for that day.
A final data set included in the evaluation of the model meteorology and dynamics is the
boundary layer height (PBLH) data retrieved from Micropulse Lidar (MPL) data at FC West,
Platteville and NREL Golden (Figure 7). This is not a true direct comparison, but shown here to
give an indication of the model's ability to reproduce the typical PBLH and day-to-day
variability. The WRF Yonsei University (YSU) boundary layer scheme is a first-order nonlocal
scheme, with a countergradient term and an explicit entrainment term in the turbulence flux
equation. In YSU, the PBLH is defined as the level in which minimum flux exists at the
inversion level. The MPL retrieved PBLH, in contrast, have been obtained from gradients or
variance in the backscatter profile, wavelet covariance, and fits to idealized profiles.
Despite the noise in the data, some unreasonable looking MPL retrievals (e.g. the first four days
at Platteville), and the different definitions of PBLH, we can see that the model overall
represents the day-to-day variability fairly well. Comparing PBLH amongst the three sites for the
same days, it is interesting to note that the PBLH can vary notably, reflecting the local nature of
sites and, in general, the variability in meteorological and dynamical conditions across the
NFRMA.

Final Report for RFP # FAAA 2016000120

8

 Table Mountain

Figure 3: Time series of measured (black) and modeled (red) downwelling shortwave solar radiation
at the surface for Table Mountain.

BAO Tower

Figure 4: as Figure 3 but for BAO.

Final Report for RFP # FAAA 2016000120

9

 Platteville

Figure 5: as Figure 3 but for Platteville.

NREL Golden

Figure 6: as Figure 3 but for NREL Golden.

Final Report for RFP # FAAA 2016000120

10

 (a)

(b)

(c)

Figure 7: Time series of modeled boundary layer height (PBLH) to Micropulse Lidar (MPL) retrievals
at (a) FC West, (b) Platteville and (c) NREL Golden.

3

Evaluation and Constraints of a priori NOx, CO and VOC emissions using
NCAR-C130 Observations

The methodology used to evaluate the a priori (S0) emission inventory has been discussed in
detail in Interim Report II (April 2017), a brief summary is given here. The base run with a priori
emissions in the previous report had originally been performed with ADP observational nudging
only, but has been updated to include also nudging with FRAPPÉ wind observations for this
report. However, the extended observational nudging overall did not notably change the spatial
or temporal distribution of trace gases or the derived conclusions.
For evaluating emissions, we focus on data from the C-130 because of its complete chemical
payload. Three different regions (Figure 8) were chosen based on the dominance of certain
emission sources. The Denver urban area and the non-Denver urban area are dominated by
mobile emissions and, dependent on species, might also possess a large contribution from
industrial emissions, whereas the oil and gas region is mostly dominated by oil and gas activities.
We only filter for data where observations and modeled winds originated from within the same
sector and focus on constraints in the mobile and oil and gas sector. The emission criteria are
defined such that for a measurement data point to be considered, the mean emissions of not only
the model grid boxes containing the actual aircraft observation but also their adjacent upwind
grid boxes must have a contribution of more than 60% from either the mobile or the O&G sector.
Final Report for RFP # FAAA 2016000120

11

 We include only aircraft measurements for 10-17 LT and below 1 km a.g.l., which can be
considered as representative of concentrations within the boundary layer (PBL) and therefore are
most suited for constraining emission sources. P-3 aircraft measurements and surface
measurements are used to support the conclusions derived from the analysis of C-130 data.

Figure 8: The three areas used for
evaluating model emissions shown on a
map of the NFRMA with C-130 flight
tracks colored by altitude above ground
level and P-3 spirals over the six key
ground sites shown in grey.

In Interim Report II (April 2017) we discuss a comprehensive evaluation of the a priori (in the
following referred to as S0) mobile and O&G emissions and derived the following conclusions:
• Ethyne, mostly released from mobile sources, is about a factor of 2 too low in the model while
other mobile tracers and their ratios to each other are predicted fairly well by the model.
• Modeled absolute mixing ratios for the two northern subregions non-Denver urban and O&G
appear underpredicted by about a factor of 2 for all mobile tracers (NOx, CO and toluene).
• All oil and gas VOC species (alkanes and benzene) appear strongly underestimated by the
model, with the exception of ethane.
Based on these results we decided to apply the following adjustments to the S0 inventory: a
factor of 2 increase in all mobile ethyne emissions, a factor of 2 increase in mobile (onroad and
offroad) emissions for the non-Denver urban and O&G region and a factor of 4 increase in O&G
VOC emissions aside from ethane. The newly derived inventory (S1) was tested in the model; S0
and S1 model results are compared in Appendix B, Figures B1 and B2, respectively.
In order to constrain the emissions quantitatively, we show in these Figures the relationship
between measured and modeled species. Only data were used that have been filtered for the
agreement in wind direction and for the relative contribution of the respective emission sector as
discussed above. For NOx, the agreement between S0 and S1 does not change notably as
expected and because of the scatter in the comparison no clear conclusion about a possible
adjustment can be derived. For the non-Denver urban region, we find a good modelmeasurement correlation and S1 clearly shows an improvement over S0. The model is biased
Final Report for RFP # FAAA 2016000120

12

 low in NOx in the O&G region, but because the mobile emissions are only a small contributor in
this region, an increase in the mobile sources does not significantly change the agreement.
Results for CO are similar to NOx, yet due to an instrument issue during the first part of the
campaign and a smaller number of available data points only a very limited set of data is
available for analysis. The increase in ethyne emissions in S1 yields to an overall better
agreement for all three regions. Toluene, which also is mostly an urban tracer is now
overpredicted in Denver in the model, whereas the agreement somewhat improves with S1 in
both the non-Denver and O&G region, but there is still a low bias in the latter.
Benzene is released from both mobile and O&G sources and the model agrees fairly well in the
Denver area in both scenarios and despite a larger scatter in the data shows an improvement in
S1 over S0 for the other two regions. Ethane concentrations, which mostly originate from O&G
sources, are low in the urban areas and only few data points fulfill the emission selection. Since
the emissions were not changed between the S0 and S1 O&G inventories, the agreement is the
same in both simulations and given the larger scatter no clear conclusion on a bias can be
derived. Propane and PAR, which are unique tracers for O&G but clearly show a different
characteristic than ethane, are clearly underestimated in S0 and the four times increase in the
emissions brings the modeled concentrations into better agreement with the measurements.
In addition to comparing measured and modeled absolute concentrations, we also evaluate the
modeled emission ratios with measured ratios. For the more typical urban trace gases we use
ethyne as a reference species and for the more typical O&G trace gases we use propane as a
reference species. Evaluations for S0 and S1 model runs are shown in Appendix B, Figures B3
and B4, respectively.
The ratio of NOx versus ethyne in the urban areas has a weaker correlation in the observations
compared to the model which partly can be explained in that the emission criteria we apply
filters for data points impacted by mobile emission in the model but because of model transport
errors or inaccurate representation of the real-time traffic volume, the filtering might not be as
effective for the observations. Given there are other potentially significant NOx sources in the
urban areas, this will impact the analysis and might explain the discrepancy between modeled
and measured ratios. While the agreement with S0 looks somewhat better, a clear conclusion
cannot be drawn from the graphs. CO, in contrast, shows a tight relationship in the Denver area
in both observations and model and a clear improvement in S1 over S0. Similarly, the agreement
for toluene clearly improves in S1 for both the Denver and non-Denver urban regions. Benzene
versus ethyne emission ratios significantly improve in S1, whereas the agreement degrades over
the non-Denver urban region. The benzene to ethyne relationship in the model is similar for both
the Denver and non-Denver region as a result of our filtering for mobile impacts. The ratio,
however, changes in the observations reflecting uncertainties in the filtering and suggesting that
this might be due to transport of benzene from O&G sources, the emissions of which were
increased in S1.
For the O&G region, the ratios of NOx and CO with propane show large scatter given that both
species are not uniquely emitted from O&G sources, but S1 seems to perform better for NOx
whereas no clear conclusion can be drawn for CO. The increase in O&G VOC but not in ethane
significantly improves the ethane to propane emission ratios in the model. Benzene versus
propane and PAR versus propane ratios over the O&G region change only slightly given that all
Final Report for RFP # FAAA 2016000120

13

 species are increased by the same factor in S1. While the model ratios are not perfect, they are
reasonably well in line with the measurements and the underlying uncertainties in the analysis
(such as model transport errors that are not fully eliminated despite the filtering for wind
directions, the influence of coarse model temporal and spatial resolution, incorrect boundary
layer heights in the model or the limited data sampling available for the analysis which amongst
other restrictions does not allow a filtering for weekends versus weekdays, inaccurate
representation of realtime temporal variability in emissions, etc. ) do not allow room for a further
refinement of the tracer ratios.
The agreement with the C-130 observations using the S1 emissions overall improved the model
performance, but in addition, we tested two other adjustments to the emissions to account for
possible shortcomings found in the S1 inventory:
S2: as S1 but in order to not change the original O&G emission ratios and account for the low
NOx bias in the model over the O&G region, we also increased NOx emissions from the
O&G sector by a factor of 4 over a priori.
S05: as S1 but only onroad mobile emissions in the non-Denver urban and O&G region were
increased by a factor of 2 and O&G NOx and VOCs were increased by a factor of 2 over S0.
Table 3 compares emission totals for the a priori S0 emissions, the first scenario S1 as well as the
two additional scenarios S05 and S2. S2 has highest NOx emission totals for the NFRMA
because of the increase in NOx emissions from O&G on top of the increase in mobile emissions.
S05, even though compared to S1 and S2 only has increased onroad emissions versus increasing
the entire mobile (offroad and onroad) sector in S1 and S2, has the second highest total NOx
emissions because of the increase in O&G NOx. Offroad NOx emissions (Table 1) have been
estimated to be only about 1/3rd of the onroad emissions and about ½ of the O&G emissions.
VOCs did not change between S1 and S2 and totals for S05 are about half of the values for S1
and S2 and about 50% higher than S0. Changes are dominated by the different scaling of O&G
emissions.
Table 3: Total anthropogenic emissions for NOx and VOCs for the a priori emissions (S0) and the
four different constraints S05, S1 and S2.
Anthropogenic Emission
S0
Adams
Arapahoe
Boulder
Broomfield
Clear Creek
Denver
Douglas
Elbert
Gilpin
Jefferson
Larimer
Park
Weld
Total Result

14,639
7,262
6,735
779
552
11,568
4,456
578
311
7,060
7,542
390
29,791
91,663

NOx (tons/year)
S0.5
S1
14,639
7,262
8,689
888
552
11,568
4,456
578
311
7,060
11,637
390
55,331
123,360

14,639
7,262
9,723
912
552
11,568
4,456
578
311
7,060
12,652
390
37,623
107,724

Final Report for RFP # FAAA 2016000120

S2

S0

14,639
7,262
9,723
912
552
11,568
4,456
578
311
7,060
12,652
390
57,406
127,508

11,802
11,975
6,126
1,229
373
12,009
5,498
627
156
11,313
7,731
639
83,886
153,363

VOC (tons/year)
S0.5
S1
11,864
12,054
7,711
1,306
379
12,082
5,498
627
159
11,387
10,267
639
153,881
227,853

11,864
12,054
8,695
1,324
379
12,082
5,498
627
159
11,387
11,511
639
289,271
365,489

S2
11,864
12,054
8,695
1,324
379
12,082
5,498
627
159
11,387
11,511
639
289,271
365,489

14

 Similar to what is shown in Appendix B, Figure B1 for S0, Figures B5 and B6 show the
relationship between measured and modeled concentrations of the C-130 filtered data for the S2
and S05 model results, respectively. NOx in the Denver area does not change much between the
different scenarios; in the non-Denver urban area the best performance is seen with S1 and in the
O&G region the increase in O&G NOx in S2 leads to the best agreement. Little changes are seen
for CO since the CO emissions from O&G were not increased and overall mobile emissions have
a smaller contribution to CO levels. Ethyne performance is best in S1 and S2 where the emission
increase is the largest. Toluene and Benzene in the Denver area change little but all three
scenarios S05, S1 and S2 show a clear improvement over S0 in the other two regions. However,
as mentioned earlier, even the four times increase in O&G still leads to a low model bias. This is
also the case for the other O&G tracers propane and PAR over the O&G region. Ethane, because
the emissions have not changed, looks similar in all model runs.
A similar conclusion can be reached from looking at the emission ratios, which are shown in
Appendix B, Figures B7 and B8 for S2 and S05, respectively. Overall the best agreement is seen
for S2 with the largest increase in O&G VOCs and at the same time an increase in O&G NOx.
All three constrained emission inventories show an improvement over the a priori S0 emissions.
In Appendix B, Figure B9 we look now at a comparison of spatially averaged C-130 aircraft
measurements to modeled values for the trace gases discussed above as well as for ozone. Only
C-130 measurements below 1 km a,g,l. were used for these maps. All selected data have been
averaged over a 0.1° by 0.1° grid, whereas individual 1-minute average data are included in the
scatter plots. These comparisons are not filtered by the level of agreement with model winds and
emission strength and model transport errors will play a more pronounced role in this
comparison. However, these maps allow assessment of the ability of the model to represent the
overall species distributions. We also include scatter plots of the 1-minute measurements and the
model data, but there is a large scatter in these comparisons which, aside from uncertainties in
model physics, transport or chemistry, is due to the model not being able to represent single
measurements solely due to the coarse spatial resolution and the interpolation from hourly model
output to 1-minute values.
The highest measured and modeled NOx and CO concentrations are observed and modeled in the
Denver area. S2 shows the highest NOx values compared to the other model runs and also
compared to observations. Interestingly, a large increase of NOx in S2 is seen in the Denver area,
yet only O&G NOx emission have been increased over S1. This suggests that transport in the
model towards the south is too strong on some of the flight days, which is corroborated by the
highest model CO and toluene concentrations being seen south of Denver. Looking at individual
flights, a large fraction of these high values is from flights on July 27 and 28 when a Denver
Cyclone was established; this is also present in the model but the model does not fully represent
the true spatial placement. This error in transport impacts and challenges the comparison of all
species.
Toluene is overestimated in all simulations in the Denver area, but too low in the north-eastern
NFRMA. Benzene is too low in S0, but appears too high in S1 and S2. For ethane, no clear bias
could be determined from the previous analysis, yet this comparison seems to suggest a high
ethane bias in all model runs. Other oil and gas tracers, while based on the previous analysis

Final Report for RFP # FAAA 2016000120

15

 appearing too low even with a four-fold increase, in this analysis seem to be simulated best in
S05.
The C-130 flights over the NFRMA do not provide a comprehensive coverage in time, but were
conducted on selected days dominated by either a Denver Cyclone or upslope conditions and
many flights were conducted later in the day when frequent thunderstorms interrupted the
transport patterns. This poses additional challenges to a model and in order to provide additional
information on the performance of the different emission scenarios we discuss in Section 4 the
comparison of the model to P-3 measurements and surface data.

3.1 Assessing emission constraints using P-3 and surface observations
This Section examines how well the initial emission adjustment factors derived from the C-130
observations can be reconciled with NASA P-3 and surface observations. The average spatial
distribution of trace gases from the observations and the different model simulations are shown
in Appendix C, Figure C1. In Figure C2 we show comparison plots for the NASA P3 spirals at
the six spiraling sites as well as missed approaches at Greeley airport. We apply the same
filtering criteria as for the C-130 analysis, i.e. we filter for data from 10-17 LT and below 1 km
a.g.l. However, the wind filtering cannot be used because the P3 winds are not available during
spirals. The spiral locations Golden, Chatfield Park and Denver La Casa are located in the
Denver urban sector, the spiral sites Fort Collins West and BAO are located in the non-Denver
urban area and the spiral site Platteville is located in the O&G area. The P-3 measurements
represent more of the small-scale variability compared to the C130, e.g. the P-3 spirals were
flown with a diameter of about 5 km, i.e. of similar order as the model grid size. The repetitive
flight pattern, in exchange, allows for more robust averaging.
The evaluation with P-3 data confirms the overall conclusions derived from evaluation with C130 data and also suggests that transport in the model towards the south is too strong. NOx
overall is too high in S2 and too low in S0, in S05 and S1 the sign of the bias varies spatially. CO
overall is too high independent of emission scenario suggesting a small offset in the model
background value. Toluene has a high model bias in the western flight legs and spirals, but is
biased low in the eastern legs and spirals. Benzene is too low over the O&G region in all
simulations and overpredicted in the Denver area in S1 and S2 while S0 and S05 underpredict
benzene in the Denver area. Ethane data conform the high bias in the ethane emissions as
suggested from spatial analysis of C-130 data.
The spirals included in the P-3 flights provide a reasonably dense data coverage over six core
sites in the NFRMA with different characteristics. For most sites ~20-30 spirals between 10-17
LT are available; over Greeley only 6 missed approaches are available. The measured and
modeled vertical profiles for primary emission species for these sites are shown in Appendix C,
Figure C2. Overall the conclusions derived from analysis of C-130 data and P-3 spatial
distribution is replicated in this evaluation. For all sites and species, we find that during the
spirals a large range of concentration values have been encountered reflecting the large
variability present in the NFRMA even at single locations.
Highest NOx is measured for La Casa with the model underestimating both the median and the
range. This is most likely due to the coarse model resolution not capturing the local nature of this
downtown Denver site and the evaluation with surface sites discussed below will show this
Final Report for RFP # FAAA 2016000120

16

 further. Below we also look more closely at the low bias at NREL Golden and Chatfield. At
BAO and Platteville, the best performance is seen for S05. The median and range for CO
concentrations is overall fairly well captured in the model.
The comparison for toluene reflects the regionally varying bias detected during the above
evaluations and overall, S05 provides the closest agreement with the observations, with toluene
concentrations in the mid-range of the considered model scenarios. Benzene is highest at sites in
the O&G area reflecting the large emissions from O&G sources over urban sources. S1 and S2
represent the median at BAO and Platteville best, but tend to overestimate the range. Also in the
urban areas benzene tends to be underestimated, which in parts might be due to transport of too
low benzene from the O&G region, but also might be due to uncertainties in the model not
representing local urban sources. The range in ethane is overpredicted at most sites in the model
and for BAO and Platteville the model also simulates a higher median in line with previous
results. Modeled ethane at FC West is too low which might suggest incorrect transport in the
model to this site during the conducted spirals but also could be a resolution issue. The
concentrations encountered are on the low end.
For comparison to surface measurements of NOx, CO, benzene and toluene, which are shown in
Appendix C, Figures C3 through C5, we limit the range to 9-18 LT and exclude nighttime. The
PBL during nighttime is very low leading to very high concentrations. Plotting these on the same
scale would make it hard to discern the daytime values from the different data sets. Because of
the low PBL and the limited degree of mixing, nighttime values are also more strongly impacted
by nearby sources which cannot be resolved by the model.
Surface measurements of NOx are available from 8 sites in the NFRMA with some of them also
P-3 spiral sites. This allows to relate the P-3 measurements to the surface. Measurements in
downtown Denver at La Casa and CAMP reflect the large small scale variability in NOx that
also impacts the P-3 comparison. The two sites are close to each other and the measurements
show very different characteristics with concentrations about twice as high at CAMP; model
results, however, are comparable at both sites and the model has a smaller bias at La Casa. At the
I-25 monitor, the model is significantly underestimating the high NOx levels, which can be
expected since this site is right next to a busy interstate highway. Time series at Chatfield show a
different diurnal cycle in the model compared to the observations. Model winds at Chatfield
(Appendix A, Figure A2) are not capturing the north-northwest winds in the observations and
likely the model does not simulate the transport correctly. Modeled NOx levels at NREL Golden
are in reasonable agreement with the observations with a small low bias in all simulations.
At FC West, all model scenarios overestimate NOx and the model seems to predict the drop off
in NOx about an hour earlier than seen in the observations possibly due to a too early change in
wind direction from the general daytime upslope to nighttime downsloping winds. The
overprediction in the model might in part be due to the selection of the model grid. For
comparison with the surface data we select the model grid containing the measurement site. In
the case of FC West, the site is located right at the edge of the grid box. If we compare the
surface observations to a model grid cell just west of the site, modeled NOx concentrations drop
by about 50%. This then would bring the higher NOx scenarios in good agreement with the
observations. NOx at BAO and Platteville is predicted well by the model, except for the S2
scenario which is biased too high.
Final Report for RFP # FAAA 2016000120

17

 The evaluation with surface CO monitors is in parts impaired by the low measurement sensitivity
of some of the sensors. Sites in Denver are subject to similar grid resolution issues as NOx but
are less pronounced because of the long lifetime of CO and fewer large CO sources in the
Denver area. CO at FC CSU is too low in all model simulations and the model does not replicate
the relatively flat diurnal cycle in CO. Such CO behavior is not typical and indicates a local
source influencing the site. A similar feature is seen at NREL Golden with all model scenarios
too low compared to the measurements. CO at Platteville is well simulated in all runs. Measured
and modeled values at this site approach background concentrations given that CO sources are
rather small around this site.
At Platteville, we also have measurements of toluene and benzene available for comparison to
the model (Appendix C, Figure C5). As discussed above, model performance of winds is poor at
this site but there is a large number of O&G facilities surrounding the site and emissions from
O&G reach this site independent of wind direction. For both, benzene and toluene, S05 gives the
best agreement, whereas S0 is too low and S1 and S2 are too high.
The final data set we use for evaluation of the modeled emissions are canister samples with the
Whole Air Sampler (WAS) that were collected from a mobile van, typically nearby larger
emission sources. We note that a few locations in the WAS data file are listed incorrectly (see
Section 9), but this should not bias the results notably if only tracer ratios are considered.
Appendix C, Figures C6 through C9 show the evaluation for the different model scenarios. For
this data set we show the comparison of emission ratios as the measurements represent strong
local sources that might be too diluted in the 4 km x 4 km model resolution. We select samples
that were mostly impacted by either urban sources or sources from O&G activities. To do so we
filter the collected data by region, by the absolute ethane concentrations and by the i-pentane to
n-pentane ratio. The thresholds that were used are indicated in the plots.
For the Denver and non-Denver urban areas, the emission scenarios S05, S1 and S2 typically
show a better agreement than S0 and compare rather well with the measurements. The exception
is the ratio of toluene with ethyne in the Denver area, which degrades in S05, S1 and S2, but also
shows a larger variability in the samples given that industrial sources are also a major source of
toluene. Samples identified as O&G samples show very high concentrations of propane, ethane
and PAR. These samples were collected right next to O&G facilities and related operations such
as condensate tanks, fracking fluid disposal sites, etc. The 4km model grid resolution will not be
able to resolve these types of sources and some of these sources may not be part of the
inventories. Even though the model values are hard to discern in the graphs, the derived slopes
suggest the best agreement in S1 and S2 and the least agreement in S0. The modeled ratios of
PAR to propane have not changed in the different model scenarios. The modeled slope of 3.1 is
too low compared to the measured slope of 11. However, the slope derived from the C-130 data
is 5.2 (Appendix B, Figure B3) suggesting that the WAS samples are influenced by the special
source selection and might not be fully representative for the overall O&G emission ratios.

Final Report for RFP # FAAA 2016000120

18

 4

Model performance of simulated ozone concentrations

In this Section, we provide a comprehensive evaluation of the modeled ozone with
measurements from aircraft, sondes and surface instrumentation. For comparison to the C-130
aircraft data we separate the flights into July and August. During July, two of the flights were
conducted on days with a Denver cyclone and a slight misplacement in the modeled cyclone
center can lead to large uncertainties in the species distribution. Flights in August mostly were
focused on upslope flows, though the predicted upslope flows did not always materialize because
of the occurrence of thunderstorms.
Appendix D, Figures D1 and D2 show the spatial measured and modeled ozone distribution for
the C-130 in July and August, respectively. During July, the highest ozone values were measured
north of Denver, which are generally underestimated in the model. The situation is reversed
around the Denver area, where all model runs typically overestimate the measured values with
S2 showing the largest bias. The model better represents the conditions during the August flights
with the better agreement between measurements and model during periods of high ozone
concentrations along the Foothills. S0 seems to underpredict the measurements, whereas S2
appears on the high side. P-3 measurements (Appendix D, Figure D3) show the highest ozone
concentrations around the Denver area and north of Denver, which are picked up well by the
model. The model tends to underpredict ozone in the eastern part of the flight paths but
represents well the measured values in the western part of the flight paths, aside from S0 which
appears too low in the model.
More insight into the P-3 data evaluation can be gained from the analysis of the individual spirals
(Appendix D, Figure D4), which in the map graphs are only showing up as single points. At most
sites the model has a low bias in ozone, which extends throughout the vertical coverage of the
aircraft data suggesting an overall underestimate in background ozone. This has been discovered
also in WRF-Chem simulations performed in our group (Interim Report I, December 2016) and
by other members of the FRAPPÉ Science Team (Yo Yan Cui, "Using potential vorticity to
specify the effect of stratosphere-troposphere exchange on the vertical distribution of ozone
during the FRAPPE 2014 field campaign", presented at the May 2017 FRAPPÉ/DISCOVER-AQ
Science Team Meeting). It was found that part of the low model bias is caused by the models not
capturing enhanced background ozone, which is most pronounced during the last week of July.
This indicates that the low model bias in ozone in comparison to the C-130 and P-3 aircraft data
can in part be explained by large scale features and is not necessarily due to regional and local
effects. Taking into account the low overall bias, the model agrees reasonably well with
measured ozone at the lowest altitude bin and also represents the site-to-site variability. For
Greeley, where only 6 missed approaches are available, the model bias is the largest. S1 and S2
simulations show the largest range, sometimes exceeding the measured range of ozone values.
Additional datasets for the evaluation of the vertical performance of the model simulations are
available from ozone sonde launches at Platteville and Fort Collins West (Figure 9). For better
comparison with P-3 profiles, we only show the altitude range for 0-5 km a.g.l. Modeled ozone
concentrations at Platteville available from 40 ozone sonde launches are also biased low at
higher altitudes, but the bias is smaller compared to the P-3 profiles and the model is in close
agreement with the observations at the lowermost altitudes. Sonde profiles are available for more
days compared to the P-3 flights, the latter weighted by a large number of flights towards the end
Final Report for RFP # FAAA 2016000120

19

 of July (6 flight days between 22 July-31 July out of 15 total days with flights). In addition,
ozone data from the sondes are collected all the way to the surface whereas the aircraft is only
allowed to fly as low as ~ 0.3 km a.g.l.
At Fort Collins West, twelve ozone sondes were launched. For these the model runs are too low
at the lowest altitude bin except for S2, which is very close in median and quantile
concentrations. The sign of the bias changes with altitude and at the highest altitude bin the
models tend to overestimate the sonde measurements. This is in contrast to the Platteville sondes
comparison and reflects the influence of the different launch frequency on the comparison.

Figure 9: Vertical average profiles for measured (black) and modeled (colored) ozone for ozone sonde
launches in Platteville and Fort Collins West.

Ozone measurements from tethered balloon are available at Chatfield, Fort Collins West and
Denver Golf Course for 3 days each and the comparison to modeled ozone is shown in Figure
10. Tethered balloon measurements represent the near-surface vertical variability and reach up to
about 0.8 km a.g.l. The launches were planned to be conducted on days with high ozone, which
was not always successfully achieved. The model agrees well with the measurements at Denver
Golf Course and is slightly too high at Chatfield and Fort Collins West. The model generally
underestimates the variability in the measured ozone concentrations and also the vertical
variability which has to be expected when comparing the model to a high-frequency small scale
data set.

Final Report for RFP # FAAA 2016000120

20

 Figure 10: Vertical average profiles for measured (black) and modeled (colored) ozone for tethered
balloon launches at FC West, Denver Golf Course and Chatfield Park.

The final data set explored are ozone measurements available during FRAPPÉ from 43 surface
monitors (8 out of them are located outside the NFRMA or the nearby mountains but are still
included for completeness). The average diurnal cycle in measured ozone compared to the
different model scenarios is shown in Appendix D, Figure D5.
For most surface sites the model tends to overestimate surface ozone with S2 showing the
highest ozone concentrations. A major part of this high bias might be explained by the
underrepresentation of clouds in the model as was discussed in Section 2. The reverse sign of the
bias when the model is compared to P-3 measurements and to a lesser degree the ozone sonde
data can be explained in that flights and sonde launches were favored for days when no or little
cloudiness was expected. In addition, a large fraction of the flights took place during a period of
enhanced large scale background. The surface sampling, in contrast, is continuous throughout the
campaign period independent of meteorological conditions. The FRAPPÉ period was
characterized by above average rain and thunderstorm activity, often very localized and more
likely to occur during the afternoon. For many of the sites, e.g. Chatfield, Fort Collins, RF North,
Weld County Tower, Table Mountain or Welby, the morning buildup of ozone is represented
better in the model compared to the afternoon, which might in part be related to this issue. The
uncertainties seen in winds and NOx will also translate into uncertainties in ozone as well as the
local nature of many of the sites.
Most sites in the Denver area showed a low model bias in NOx and in line with this show an
overprediction of ozone. Similar to what has been seen in the comparison of NOx data, for the
two downtown Denver urban sites La Casa and CAMP, the models simulate similar
Final Report for RFP # FAAA 2016000120

21

 concentrations at both sites, whereas ozone measured at Lasa Casa is ~5 ppb higher compared to
CAMP; the latter also showed higher NOx measurements. In the non-Denver urban area, we see
a similarly high model bias for sites situated inside the urban environment compared to the
measurements taken in Boulder or the monitor at Fort Collins CSU. Ozone at Fort Collins West
or RF North, in contrast, is fairly well captured by the model.
Ozone at Weld County tower in the O&G region is well simulated in the model. This site is
located in the southern part of Greeley and with winds typically from the east and southeast
during the day, it does not receive the full impact of the Greeley city emissions. BAO, also in the
O&G region is rather well predicted during the day. S2 model simulations at Platteville show a
narrow diurnal cycle and have the highest peak ozone of all simulations, which is related to this
simulation also having the highest NOx. Note that this is in contrast to P-3 profile measurements
over Platteville, where the model is biased low. At both sites, early morning ozone is
overpredicted by the model. Nighttime NOx values are significantly overperdicted in the model
(not shown here), which likely is related to the model not representing well the nighttime PBL,
and also could explain the low model ozone in the early morning at both sites.
Additional sites had been established during FRAPPÉ in the nearby Foothills alongside Boulder
Canyon (e.g. Sugarloaf, Lost Angels, Coughlin Meadows). These sites are located in a narrow
canyon and we do not expect the model to be able to capture this small-scale topography. Ozone
concentrations at Squaw mountain (3420 m a.s.l.) located in the Foothills West of Denver, and
Mines Peak (3805 m a.s.l.), west of Squaw Mountain at an exposed high altitude location, show
a late afternoon peak in ozone which is too pronounced suggesting that the model is
overpredicting the intensity of upslope transport of NFRMA ozone. Note that the evaluation of
surface winds also indicated a too pronounced upslope in the model. This might be related to the
model not simulating the frequent thunderstorm activity or the buildup of clouds that would
interrupt upslope patterns and this would also lead to a high surface ozone bias in the model.
High altitude measurements are also available from three sites near Niwot Ridge, with the
Tundra Lab (TL) at the highest elevation (3739 m a.s.l.). The model best represents the average
condition at TL, where also the highest ozone values have been measured (concentrations ~50
ppb without consideration of upslope influence), but overestimates the effect of upslope flow.
The model simulates similar values at all sites because with the model spatial resolution the
small-scale topography cannot be resolved, while ozone measured at the two lower Niwot Ridge
sites (~3400 m a.s.l.) is lower compared to TL. Aside from an overall high bias, the model does
simulate the occurrence of upslope events reasonably well at the lower sites.
Further to the north, ozone measurements are available from Longs Peak (2743 m a.s.l.) and
Trail Ridge Road (3498 m a.s.l.). Also for these sites the effect of model resolution is evident.
The comparison with Longs Peak shows similar features as for the lower Niwot Ridge sites; at
Trail Ridge Road, the model is biased low throughout the day and simulates very similar
concentrations as for Longs Peak. Further evaluation of the model compared to surface
measurements is included in Section 8.

Final Report for RFP # FAAA 2016000120

22

 5

Conclusion on Emission Constraints

The above comparisons to the very comprehensive measurements available from FRAPPÉ and
DISCOVER-AQ provides detailed insight into the model performance of the parameters relevant
for understanding all the factors impacting ozone concentrations. A conclusive evaluation of the
different emission inventories is hampered somewhat by uncertainties in the modeled transport
and meteorology, and is challenged even more because of the above normal cloud and
thunderstorm activity during the campaign. Periods of high ozone background ozone, the source
of which is not yet fully understood, further complicate the model assessment of ozone. Model
performance varies strongly from day to day and is sensitive to the spatial and temporal coverage
of the observations.
Despite these complications, various conclusions can be drawn from this evaluation:
•
•
•
•

The S0 inventory overall underestimates all relevant emission species including NOx and
primary VOCs; the updated inventories S05, S1 and S2 generally show an improvement
over S0.
S2 emissions perform best in regard to VOC emissions from O&G activities
Surface ozone is biased high in all scenarios with the largest bias in S2
Free tropospheric ozone during flight days is, on average, too low in the model owing in
large parts to the model not capturing high background ozone during the last week of
July.

For an emissions inventory that can be used in a base simulation and for conducting zero-out
emission scenarios our selection was made for S05. S05 in many ways shows improvements over
S0 and while O&G VOCs might be on the low side in this inventory, it performs better in surface
ozone compared to S2. We also decided for an increase in both NOx and VOC emissions from
the O&G sector so as to not modify the underlying emission ratios and also because the NOx
comparison in Weld County suggested a low model bias in NOx. We note that much of the
observed overestimate of local ozone production when using the S1 and S2 scenarios in the
model could be due to the underestimation of clouds, as discussed in Section 2. The exact impact
of this cannot be quantified, however it might explain a major part of the overall bias given that
the model shows a significantly higher fraction of clear-sky conditions compared to the
measurements.
The selection of S05 as our best estimate (“base case”) scenario is a rather conservative choice,
particularly with respect to the VOC emissions from oil and gas related activities. Thus, in the
following Sections we examine not only the S05 emissions and related zero-out scenarios, but
also discuss model results from the other two scenarios, which will provide insight into the
variability of impacts on Front Range ozone across the plausible range of the emission profiles
considered here.

Final Report for RFP # FAAA 2016000120

23

 6

Zero-Out Emission Scenarios and CMAQ simulations

6.1 S0.5 base case emission scenario
Following the establishment of a best estimate for the base case emission scenario, we now
explore the regional impact of different emission sectors on ozone formation by zeroing out from
various sectors utilizing the CMAQ model. Table 4 lists NOx and total VOC emissions by
county as well as the NFRMA totals for the S05 base case and four zero-out cases where either
mobile emissions (on- and offroad; Smobile), emissions from oil and gas activities (SOG),
industrial emissions (Sind) or EGU emissions (SCEM) are turned off. Zeroing out mobile
emissions expectedly has large impacts on the emissions in all counties, especially on NOx, with
the exception of Weld, where the largest emission impact is from oil and gas on both NOx and
VOC. Industrial as well as mobile emissions have significant, and comparable, impacts on VOC
emissions in all counties except Weld, and EGU emissions predictably are only important for
NOx emissions in Adams, Boulder, and Larimer counties where the generation units are located.
For the NFRMA overall, the largest amount of NOx comes from mobile emissions, followed by
oil and gas and industry, while the largest amount of VOC emissions stem from O&G activities.
Table 4: Anthropogenic emissions for the base case S05 and the different zero-out scenarios. Emission
totals are shown per county as well as the county totals.
Anthropogenic Emission
S0.5
Adams
Arapahoe
Boulder
Broomfield
Clear Creek
Denver
Douglas
Elbert
Gilpin
Jefferson
Larimer
Park
Weld
Total Result

14,639
7,262
8,689
888
552
11,568
4,456
578
311
7,060
11,637
390
55,331
123,360

NOx (tons/year)
Smobile
SOG
8,978
1,845
3,748
114
71
5,525
537
64
22
1,836
2,432
41
41,743
66,955

14,639
7,262
8,689
888
552
11,568
4,456
578
311
7,060
11,637
390
15,765
83,793

SInd

SCEM

S0.5

12,658
5,973
7,098
784
483
6,225
3,954
526
290
5,320
10,715
352
53,830
108,209

9,192
7,262
6,621
888
552
11,475
4,456
578
311
7,048
10,217
390
54,705
113,695

11,864
12,054
7,711
1,306
379
12,082
5,498
627
159
11,387
10,267
639
153,881
227,853

VOC (tons/year)
Smobile
SOG
8,248
7,188
3,640
795
139
7,576
2,912
329
50
6,826
4,084
160
147,650
189,597

11,864
12,054
7,711
1,306
379
12,082
5,498
627
159
11,387
10,267
639
19,375
93,347

SInd

SCEM

5,366
6,572
5,001
670
255
5,400
2,935
372
118
5,319
7,245
506
149,688
189,447

11,819
12,054
7,697
1,306
379
12,077
5,498
627
159
11,383
10,230
639
153,862
227,727

CMAQ model results for zero-out scenarios across the entire NFRMA are summarized in
Figures 11a-d. Shown is the difference for average MDA8 between base case simulations and
zero-out simulations of the four major emission sectors over the NFRMA during the FRAPPÉ
time frame. Figure 11a demonstrates that eliminating emissions from oil and gas activities would
significantly lower the average MDA8 by around 6-8 ppb over the entire northeastern NFRMA
with the largest impact of up to 8 ppb centered over the eastern Boulder county border. This is
expected since the majority of the O&G sources are situated in the western part of Weld county
and transport typically is towards the west on high ozone days. The slower reacting VOCs
emitted from oil and gas activities require time to build up ozone. In addition, NOx availability
increases as air masses travel towards the northern urban corridor and experience these
emissions. As a result, the impact of O&G emissions on average MDA8 reaches quite far west to
the Continental Divide and beyond with a simulated 2-3 ppb increase of average MDA8
attributable to O&G emissions.
Final Report for RFP # FAAA 2016000120

24

 The zero-out scenario for mobile emissions (Figure 11b) displays a similar impact. Due to
prevalent upslope transport on high ozone days the main influence of this emission sector on
average MDA8 also appears offset to the west from the main traffic corridors, and is of about the
same order of magnitude as the oil and gas sector. MDA8 fractions attributable to mobile
emissions are around 6 ppb in central Larimer, Boulder, and Jefferson counties with decreasing
values to the west and east. Influence reaches far west beyond the Continental Divide but is
strongest in Gilpin and Clear Creek counties downwind of the Denver Urban area at the southern
end of the study area, while the oil and gas influence reaches furthest west in western Larimer
county more to the north, including Rocky Mountain National Park. This regional separation of
the influence of the two major emission sectors, caused mainly by typical summer upslope flows,
creates distinctly different photochemical regimes. The southern NFRMA ozone production
regime is generally more VOC limited due to dominant mobile emission sources while many
areas in the northern NFRMA exhibit NOx limited ozone production due to the larger influence
of O&G emissions. This finding is generally corroborated by the BOXMOX and LaRC Steady
State model results discussed in Sections 7.1 and 7.2, respectively, but because of day-to-day
variability and transport of air masses and because the box model does not consider clouds, there
are exceptions to this finding, particularly in the downtown areas of the northern Front Range
urban centers Greeley and Fort Collins.
It should be noted that emissions from oil and gas and mobile sources exhibit very different
NOx/VOC ratios (higher in mobile, lower in oil and gas) which strongly influences the ozone
production efficiency of either zero-out scenario. This cross-dependence needs to be considered
when mitigation strategies are to be derived from the findings above. Locally, and on individual
days, impacts from regulating one emission sector could be very different from the campaign
average results shown in Figure 11.
Given the significant contributions of industrial emissions in almost all NFRMA counties, the
comparably small impact when zeroed out on average ozone MDA8 is somewhat surprising
(Figure 11c). Because of large industrial emissions of NO around Denver International Airport
(DIA) as well as in Commerce City (likely the Suncor power plant), removing these emissions
increases the local average MDA8 ozone in the model, presumably due to less ozone titration.
Outside of the industrial areas between downtown Denver and DIA, where up to 2 ppb of
average MDA8 ozone is attributable to industrial emissions, the impact of removing industrial
emissions is overall very small across the NFRMA. However, the impact of industrial emissions
can be more significant on individual days. Figure 12 shows a matrix of ozone MDA8
contributions from the emission sectors for three selected single days which had high ozone (28
July, 3 and 12 August; see Section 8 for additional analysis of the meteorological conditions
leading to high ozone in the NFRMA and analysis of these days). 28 July is different from the
August days because we observed a Denver Cyclone (DC) on that day. DC conditions tend to
transport O&G emissions into the area west of Denver where NOx contributions from the other
emission sectors can enhance ozone production. It can be clearly seen that on 28 July the
influence of O&G emissions reaches far to the south and west and the injection of VOC from
these emissions causes additional ozone production downwind of the industrial NOx sources,
enhancing ozone in the SW and W sectors of the metro area by 2-4 ppb. The August days shown
are upslope days with a more northeasterly component on the 3rd and a more southeasterly
component on the 12th. The influence of O&G VOC emissions on ozone production extends into
Final Report for RFP # FAAA 2016000120

25

 western Boulder and Gilpin counties on the 3rd and far into NW Larimer County and Rocky
Mountain National Park on the 12th. Ozone enhancements due to the industrial NOx sources is
again 2-4 ppb on both days, but effects are more far reaching well into the mountains. Still, on all
three days, mobile and oil and gas sources dominate the MDA8 ozone enhancements over the
entire NFRMA.
The zero-out scenario for EGU emissions (SCEM) shows average MDA8 ozone to increase in
the vicinity of the power stations (the largest being Cherokee in Commerce City and Valmont
plant in central Boulder county) because of decreased ozone titration with NO (Figure 11d).
Likely because the southern NFRMA already has very high average NOx values (from mobile
and industrial emissions) there is, on average, no indication of ozone production downwind of
the power plants attributable to their NOx output. The example high-ozone days shown in Figure
12 still show initial ozone titration because of the very large, localized NOx emissions from both
power stations. However, under upslope and DC conditions, mixing with VOC emissions and
dilution and loss of NOx to reactive nitrogen reservoir species results in positive contributions of
around 2-4 ppb to the MDA8 on all three days further downwind of the sources. This enhances
ozone in western Boulder and Gilpin counties as well as Clear Creek county on days with a
northeasterly wind component.
(a) S_OG

(b) S_Mobile

(c) S_Ind

(d) S_CEM

MDA8 Difference (ppb)
Figure 11: Contributions (represented as the difference of S05 and the zero-out scenarios) of the four
main emission sectors to MDA8 averaged over the campaign period.

Final Report for RFP # FAAA 2016000120

26

 28 July                      

3 August                  

12 August        

SOG

SMob.

SInd

SCEM

MDA8 Difference (ppb)
Figure 12: Contributions (represented as the difference of S05 and the zero-out scenarios) of the four
main emission sectors to MDA8 on three days exhibiting high ozone values across the NFRM, as
simulated by CMAQ.

In summary, the CMAQ simulations suggest that the major contributors to average MDA8 ozone
values in the NFRMA are O&G and mobile emissions, with the former having a larger influence
on the northern part of the NFRMA and the latter on the southern half of the region, on average.
Both emission sources appear responsible for a 6-10 ppb increase in average MDA8 each, which
is a very sizeable fraction of the total local ozone production over background. Figure 13 shows
the difference between S05 and SBckg (all NFRMA anthropogenic emissions turned off), i.e. the
MDA8 ozone enhancement in the NFRMA over large scale background ozone. Results are
Final Report for RFP # FAAA 2016000120

27

 shown for the average over the entire time period as well as the three selected days that were
discussed above. The average enhancement is around 15 -20 ppbv, varying between 20 and 30
ppb on high ozone days with largest values typically around the larger Denver metro area;
maxima reach up to 40 ppb (as seen on 28 July).
As discussed earlier, evaluation with P-3 aircraft and ozone sonde data suggests that the model
underestimates the ozone background specifically in late July (Section 3.1) which could skew the
average background slightly to the low side. However, while this does not linearly affect the
simulated chemical ozone production, it might still introduce additional uncertainty into this
analysis. The underestimate is expected to be small compared to the magnitude of the local
ozone production, likely less than 1 ppb.
(a) Average

(b) 28 July

(c) 3 August

(d) 12 August

S05-SBckg   MDA8 Difference (ppb)
Figure 13: Difference in MDA8 ozone between S05 and SBckg for an average over the entire campaign
(a), 28 July (b), 3 August (c) and 12 August (d).

6.2 Other emission scenarios
During the process of finding the best adjustment factors for the a priori emissions (S0) we
developed two additional test scenarios, as described in Section 3 (Table 3). The S2 scenario
differs from S05 in two ways: all oil and gas emissions (VOC and NOx) are doubled in S2 over
S05 and the non-road mobile emissions in the northern sectors of the NFRMA are also doubled
compared to a priori, while S05 only doubles the on road mobile emissions in the northern
sectors (see Figure 8, Section 3 for details). Because in Weld and Larimer counties the
contribution to NOx from non-road mobile emissions is small (12 and 3%, respectively) and
Final Report for RFP # FAAA 2016000120

28

 their contribution to total anthropogenic VOC is negligible in Weld county and about 15% in
Larimer county, this scenario can be used as an upper limit to what would happen if oil and gas
emissions again doubled in the NFRMA.
Figure 14 shows the average MDA8 ozone difference between the S2 and the base case (S2-S05)
for the campaign period as well as the calculated MDA8 difference for the three high ozone
days. These results show that the MDA8 ozone increase due to O&G emissions roughly scales
with the emission strength for the current state of the NFRMA pollution levels. The maximum
average ozone increase is up to about 5 ppb for S2 over the base case and the largest increase is
located in south-central Larimer and central Boulder counties. To a lesser extent, ozone increases
everywhere in the NFRMA. This is only slightly smaller than the ozone difference when oil and
gas emissions are zeroed out (see above) and distributed slightly differently. The doubling in
O&G NOx in S02 over S05 leads to a small decrease in surface ozone in south-western Weld
County reflecting the sensitivity of ozone production to the NOx levels.
On high ozone days, the effect on the MDA8 can be much larger with an additional 10 ppb
ozone formed in the western half of the NFRMA on 27 July as a result of doubled O&G
emissions. On August 3 and 12 we find slightly smaller increases of the ozone MDA8 with
additional ozone increases of 5 and 9 ppb, respectively, in the southwest and northwest sections
of the NFRMA. On 12 August, we also see a larger increase in the south-eastern NFRMA. This
is because the winds during the morning were such that they transported Weld County air to the
south-east changing later to upslope flow.
These differences are closely comparable to the simulated changes in ozone when the O&G
emissions are zeroed out, thus demonstrating that increased oil and gas extraction activity in the
region would result in increasing regional ozone, especially in the western sections of the
NFRMA, if emission regulations remained at current levels. Conversely, if the S2 scenario was
to be considered closer to actual conditions, this would mean that oil and gas related emissions
would be the major contributor to ozone in the entire NFRMA.
Figure 15 illustrates the difference in MDA8 ozone between the S1 and S2 scenario and thus
provides keen insight into the strong dependence of the conclusions from the analysis above on
the actual NOx emissions from the oil and gas sector. The two scenarios were discussed and
evaluated above and both contain a fourfold increase of the a priori VOC emissions from oil and
gas activities. S2 also quadruples the NOx emissions from O&G, while S1 leaves the NOx
emissions unchanged from the a priori (S0) level. Section 7 discusses more in depth the
photochemical regimes across the NFRMA but the northwestern NFRMA area, most impacted
by oil and gas emissions, appears strongly NOx limited on average. Thus, lower NOx emissions
from O&G in scenario S1 result mostly in reductions of the campaign average MDA8 ozone of
~2-3 ppb except in central Weld county (Figure 15a), where ozone in S2 is reduced by up to
about 5 ppb. On the high ozone days, CMAQ simulates a further increase of ozone in western
and southwestern Weld county as well as around the large NOx emitters in the Denver area on
July 28 (Figure 15b). However, reduced ozone production occurs in the western sections of the
NFRMA, because the ozone production efficiency drops as a result of lower NOx/VOC ratios as
photochemical processing progresses downwind (Figures 15 c and d).

Final Report for RFP # FAAA 2016000120

29

 (a) Average

(b) 28 July

(c) 3 August

(d) 12 August

S2-S05   MDA8 Difference (ppb)

Figure 14: Difference in MDA8 ozone between S2 and S05 for an average over the entire campaign (a),
28 July (b), 3 August (c) and 12 August (d).
(a) Average
(b) 28 July

(c) 3 August

(d) 12 August

S2-S1   MDA8 Difference (ppb)
Figure 15: as Figure 14 but for the difference in MDA8 ozone between S2 and S1.

Final Report for RFP # FAAA 2016000120

30

 7

Chemical Boxmodeling in support of derived emission sensitivities

7.1 BOXMOX modeling
In Interim Report I (December 2016), we described the findings obtained running the BOXMOX
chemical box model using C-130 observations as input to evaluate the performance of different
simplified mechanisms commonly used in air quality models. Significant differences were
observed between the lumped mechanisms and the Master Chemical Mechanism (MCM version
3.3), especially for some of the more reactive samples collected (i.e., over Commerce City and
the Uintah Basin). We discovered that much of this discrepancy can be attributed to two issues.
First, the attribution of the highly reactive aromatics in most of the lumped mechanisms is done
by including these in a xylene-type species with comparably lower reactivity. Second, some of
the TOGA samples exhibit what is likely a contaminant, causing very high readings of the 1,3,5and 1,2,4-trimethylbenzenes. These very high values were the main reason for the observed
discrepancies in the reactive samples that were shown.

100

80

110
80

Commerce City (MCM)
100 T1)
Commerce City (MOZART

90

Ozone (ppb)

Commerce City (MCM)
Commerce City (MOZART T1)

80

70

60

Ozone
Ozone (ppb)
(ppm)

70
100

Ozone (ppm)

Ozone (ppb)

100

70

6090

90

120
110

100

80
Denver (T1)
Denver (MCM v3.3)
80
70

50

80

60

110

90

Ozone
Ozone(ppb)
(ppb)

110

Ozone (ppb)

120

We corrected for the contaminated TOGA samples and compared again the performance of the
MOZART T1 mechanism with the MCM reference mechanism for a number of samples. Figure
16 demonstrates that the MOZART T1 mechanism compares well with the explicit mechanism
for the most reactive samples collected over Denver and Commerce City after the correction to
the TOGA measurements had been applied. Note that both air masses sampled are capable of
producing around 40 ppb of ozone in 12 hours. Especially the Commerce City sample shows
clear evidence of initial NOx suppression of ozone production. Figure 17 demonstrates that
removal of the EGU emissions from the 120
Commerce City sample would result in enhanced ozone
production there, which confirms the CMAQ results above. Removal of the on-road mobile
emissions from the Denver sample
120 would result in a reduction of ozone by about 10 ppb, again
confirming the results obtained with the110
3D model above. Note that the removed fractions of
120
tracers
in all photochemical box model calculations120
are based on the S05 “base case” emission
110
scenario.

70

100

Denver (T1)
Denver (MCM v3.3)

90

90

80

Denver
(T1)
Denver
(T1)
Denver (MCM v3.3)
Denver (MCM v3.3)

80
70
Denver (T1)
Denver (MCM v3.3)

Denver (T1)
Denver (MCM v3.3)

60
60
20 14 2 4 4
66
10
70
0
20
66
1012 0 1412 70
2 4
4
8 8
10
8 8 10
12
14 12
Hoursof
of continuous
sunlight
Hours
continuous sunlight
Hours
ofofcontinuous
sunlight
Hours
continuous
sunlight
60
0
2
6
8
10on July12
14
Figure
16:
Comparison
of
ozone
simulated
for continuous
(40N,
14 LT
31)
604sunlit conditions
60
50
Hours
of continuous
sunlight
for the MCM
(red)
and
the
MOZART
T1
(blue)
mechanism.
The
two
mechanisms
agree
very
well
for
6
8
10
0
2
4
6
8
10 0
12 2
14 4
Commerce City (left), and within
better
than 10%sunlight
for the Denver sample (right).Hours of continuous sunlight
Hours
of continuous
40

31

Final Report for RFP # FAAA 2016000120
40
0

2

4
6
8
10
Hours of continuous sunlight

12

14

14

12

 120

120

110

110
110

120

60

0

100

90

100
70

6080

80

100

80
Denver (T1)
80
Denver (MCM v3.3)

80

40

20
70

100

90

Ozone
(ppb)
Ozone
(ppb)

70

Ozone(ppb)
(ppm)
Ozone

120 100
80

110

110

Ozone (ppb)

120

90

100
100

Commerce City (MCM)
Commerce City (MOZART T1)
Commerce City -EGU NOx (MCM)
90
Commerce City -EGU NOx (MOZART T1)

Commerce City (MCM)
Commerce City (MOZART T1)
Commerce City -EGU NOx (MCM)
Commerce City -EGU NOx (MOZART T1)

Ozone (ppb)

110

Ozone (ppm)

Ozone (ppb)

100

120

2

70

90

Denver (T1)
Denver (MCM v3.3)

90

80

70

80
Denver (T1)
Denver (MCM
v3.3)
60

60
60
4 4
6
8
10
6
8
10
12
14 0
Hours ofof
continuous
sunlight sunlight
Hours
continuous

12

20 14
70

2

DenverDenver
(T1)(T1)
DenverDenver
(MCM
v3.3)
(MCM
v3.3)
- on-roadmobile
mobile emissions)
DenverDenver
- on-road
emissions)

Denver (T1)
Denver (MCM v3.3)

4 4
810
66
8
12 10
Hours of
sunlight sunlight
Hours
ofcontinuous
continuous

12

14

0 continuous
2
4 conditions
6
8
12 31).14
Figure 17: Comparison of ozone simulated for
sunlit
(40N,
1410LT on July
60 Hours of continuous sunlight
60
Like Figure 15, but with Cherokee Station emissions removed for Commerce City (left) and on road
6
8
0
2
4 Denver6(right). 8
10 0
12 2
14 4
mobile
emissions
removed
for
60

Hours of continuous sunlight

10
Hours of continuous sunlight

Similarly, positive results for the performance of MOZART T1 are obtained in BOXMOX with a
sample collected over Weld County, and thus dominated by emissions of alkanes from oil and
gas40related activities (Figure 18). The MOZART T1 mechanism reproduces well the ozone
0
2
6 mechanism.
8
10
14gas related emissions is
production
obtained
with4the explicit
Removal
of 12
the oil and
Hours
of because
continuous
somewhat difficult in this
case
thesunlight
NOx emissions that can be attributed to oil and gas
activities vary between 20% and 80% of the total across different areas in Weld County. In the
reduced oil and gas case, added to Figure 18, the alkanes were reduced by 90% and NOx was
reduced by 50% over the base case. In general, Figure 18 supports the CMAQ results, also
showing very efficient ozone production over Weld County in excess of 30 ppb and a reduction
of excess ozone by 30% with the oil and gas emissions removed (assuming a 50% reduction in
NOx). A smaller reduction of NOx would have resulted in less reduction of ozone and vice
versa, which is corroborated by the analysis of the difference between the S2 and S1 scenarios at
the end of Section 6 and by results presented in Figure 15.
Due to time constraints and technical difficulties, we were not able to implement the CB6r3
mechanism into BOXMOX. CB6 is the chemical scheme used in CMAQ, but it is very difficult
to integrate into BOXMOX due to its complicated allocation algorithm. We are planning to do
this as soon as possible and amend the results here when they become available. However, the
general speciation and species allocation of CB6 is similar to that of MOZART T1, so that we do
not expect major biases in the CMAQ model due to chemistry.

Final Report for RFP # FAAA 2016000120

32

14

12

 110

120

100

110
90

100

100

Ozone
Ozone (ppb)
(ppm)

90

Ozone (ppb)

Ozone (ppm)

100

80

70

90

Denver (T1)
Denver (MCM v3.3)

90

O3 WeldO3County
MCM v3.3
Weld County MCM v3.3
Weld CountyMOZART
MOZART T1
O3 WeldO3
County
T1
O3 Weld County - O&G reduced
O3 Weld County - O&G reduced

80

80

80

Denver (T1)
Denver (MCM v3.3)

60
0

70

20
70

2

4 4
810
66
8
12 10
Hours of continuous
sunlight
Hours
of continuous
sunlight

14

12

14

Figure
70 18: Comparison of ozone simulated for continuous sunlit conditions (40N, 14 LT on July 31)
60
for the 0
MCM (red)2 and the MOZART
T1
mechanism
County,
4
6 (blue)
8
10for a sample
12 collected
14 over Weld12
0 oil and gas
2 sector.
4 The blue6 dotted lines
8 show10
and dominated by emissions
from
the
the
change
for 14
Hours of continuous sunlightHours of continuous sunlight
MOZART T1 when emissions from oil and gas are removed.

7.2 NASA Langley Research Center steady-state boxmodel results
A second way of examining the chemistry and ozone formation potential of samples
representative of Front Range areas with different emission profiles is the NASA LaRC Steady
State Model. The model is described in Olson et al (1999 and 2006). Briefly, the model uses a
diurnal steady state (DSS) approach with long- lived species constrained to measurements. In the
DSS approach, each input point of in-situ data (here the data was taken from the C-130 WAS
merge, and the model was run using each of the 415 canister samples for which all auxiliary data
was available) is integrated by the model to find an internally self-consistent diurnal cycle for all
computed species to within a given tolerance. Output is then sampled for time and date of the
actual observation. The model runs were performed by Jason Schroeder from NASA LaRC and
we gratefully acknowledge Jason and the Jim Crawford group for providing the results to us. The
model outputs numerous parameters and steady state mixing ratios for radicals and other short
lived species as well as calculated parameters. For this report, we use the calculated
instantaneous ozone production rate in ppb/hr, as well as the total VOC reactivity for OH in 1/s
and the ratio of peroxy radical losses via self-reaction versus reaction with NO. The latter ratio
provides an indication whether the ozone production in a specific area tends to be more NOx or
more VOC limited.
The areas chosen for the model runs are shown in Figure 19. They were chosen to represent both
places dominated by one single source category (such as O&G in the Pawnee and NE Weld
County areas) as well as mixed categories in the urban and industrial areas.

Final Report for RFP # FAAA 2016000120

33

 41.0

Larimer

Weld
Fort Collins

40.5

Latitude, degrees

Greeley

Longmont
Boulder

Pawnee Grasslands
NE Weld Cty
E Weld Cty
West Denver Urban
Denver Urban
Fort Collins / Loveland Urban
Greeley Urban
Commerce City

40.0

Denver

Surface elevation, m

4000

39.5

3000

2000

1000

0

-105.5

-105.0

-104.5

-104.0

-103.5

-103.0

Longitude, degrees

Figure 19: Areas chosen for analysis with the LaRC steady state model. The traces represent the
timing for the WAS samples taken on board the C-130. Sampling time for the canisters is 10-30
seconds, depending on altitude. The steady state model was run for each sample, and runs were
repeated with tracer measurements adjusted for zeroed-out emission sectors.

To examine the contributions of individual emission sectors in each area, the steady state model
runs were repeated with selected emission sectors zeroed out. This was done by reducing the
enhancements of individual chemical tracers over background by the fraction of each tracer
attributable to the emission sector being zeroed out. The latter fractions were determined from
CMAQ S05 sector emissions. Background values were obtained from WAS samples taken over
the Continental Divide during westerly flow.
Several interesting conclusions can be drawn from the steady state model analysis shown in
Figures 20 and 21. Note that the shown Bar-Whisker Plots include the 10th and 90th percentile
and not the quantiles; mean values are indicated by the square symbol. Not surprisingly, the
more remote areas that are largely impacted by oil and gas emissions (i.e., NE- and E Weld
counties) show both relatively high OH reactivity and ozone production rates around 10 ppb/hr
(Figure 20). Both regions are strongly NOx limited (LROx/LNOx > ~0.5 is NOx limited,
LROx/LNOx < ~0.3 indicates more strongly VOC limited conditions), which also is not
surprising given the large hydrocarbon emissions from oil and gas activities. Pawnee, which is
the most remote area with lower density oil and gas activities, shows lower values for PO3 and
OH reactivity and even more strong NOx limitation.
Greeley and Fort Collins stand out as transition areas which are under the influence of both oil
and gas related emissions and transportation emissions. This causes the ozone production there to
Final Report for RFP # FAAA 2016000120

34

 be still NOx limited but very efficient with values over 15 ppb/hr in Greeley. Greeley displays a
VOC mix that is typical for oil and gas emissions but also high NOx, presumably from mixing
with local mobile emissions. This is corroborated by the zero-out scenarios shown in Figure 20,
where the removal of oil and gas and on road mobile emissions by themselves lowers ozone
production rates by only about 20% each.
The Fort Collins / Loveland urban (FCLove) area is more dominated by emissions from on road
mobile sources, but the strongly NOx limited ozone production indicates that transport from oil
and gas areas is also important. Increased NOx emissions from additional traffic expected for the
future would likely, on average, increase ozone in the summer in this area.
The above results strongly corroborate the the CMAQ model results showing a fairly large
influence from oil and gas related emissions in Weld County as well as the Ft. Collins/Loveland
urban corridor (Section 7). In addition, the ozone monitors in the area indicate additional ozone
production during transport between the monitor at CSU near downtown Fort Collins and the
second monitor at Ft. Collins West (Appendix D, Figure D5).
The Denver urban area is clearly separated from the other areas by larger NOx availability as
demonstrated in Figure 20. In fact, the ozone production over the Commerce City (CommCity)
area is, on average, VOC limited and so is the Denver Urban (DenAM) area in the mornings. We
started many C-130 research flights with a leg across Denver within the PBL immediately after
takeoff and these samples have been separated from the afternoon Denver samples mainly for
two reasons. First because the average sample timing in most areas matches better with the
Denver area when the morning cross-city runs are removed from the average, and second
because this is one chance to gain insight in how the ozone production conditions evolve over the
course of the day. Despite the large NOx abundance, Commerce City and DenAM (samples
taken in the Denver area during morning) still show the largest ozone production rates in the
urban area. Zero out scenarios for the CEM sector were not run in the LaRC steady state model
but expected NOx reductions from conversion of the remaining boilers of the Cherokee EGU
will likely increase ozone in this area as strongly corroborated by both the CMAQ EGU zero-out
scenario as well as the BOXMOX model runs for the Commerce City sample. The DenPM
(samples taken in the Denver area during afternoon) samples show reduced ozone production
rates by about a factor of 2 compared to mornings, likely caused by dilution with cleaner air
originating either from above the PBL or transported in from the east. The zero-out scenarios
(Figure 21) show that ozone production in both CommCity and DenAM are dominated by
industrial emissions (large contributions likely from the SunCor refinery, being one of the largest
source of reactive VOC in the area). These emissions likely accumulate at night and dominate
Denver until dilution and continuing traffic emissions make mobile emissions the largest
contributor in the Denver urban area in the afternoon, while Commerce City remains dominated
by industrial emissions throughout the day.

Final Report for RFP # FAAA 2016000120

35

 Figure 20: LaRC steady state model results for the ten areas described above. To illustrate the
dependence on the time of day we show the results for Denver Urban split up into AM and PM
observations. Shown are instantaneous ozone production in ppb/hr (bottom panel), total OH reactivity
in 1/s (center panel) and the ratio of radical losses via self-reaction or reaction with (top panel).

.

Final Report for RFP # FAAA 2016000120

36

 .

Figure 21: Ozone production rates as calculated with the LaRC steady state model for each region.
Shown is the base case for each sample as well as selected scenarios where individual sectors were
zeroed out (see text for more information).

Final Report for RFP # FAAA 2016000120

37

 8

Meteorological Patterns on High and Low Ozone Days

In the summer months, particularly during weak synoptic conditions, the local meteorology in
the NFRMA is mainly controlled by thermally driven, terrain-induced, diurnal flow patterns,
often also referred to as upslope/downslope or mountain-valley winds (Johnson and Toth, 1982;
Toth and Johnson, 1985, and references therein). This has unique consequences for the transport,
mixing and photochemical processing of local emissions (Haagenson, 1978; Greenland, 1980;
Doran, 1996; Baumann et al., 1997; Olson et al., 1997) including the potential of bringing
NFRMA pollution into the pristine mountains (Parrish et al., 1986; Brodin et al., 2010; Brodin et
al., 2011; Benedict et al., 2011; Darrouzet-Nardi et al. 2012). Conversely, stronger frontal
passages can induce outflow of NFRMA pollution to the east, impacting downwind agricultural
areas in the central Great Plains. Such export events, as well as thunderstorms, are the major
mechanisms for cleaning out accumulated pollution in the NFRMA.
To assess how important mountain-valley winds are on high ozone days, we show in Figures 2224 pollution roses for measured and modeled ozone for Longs Peak and Trail Ridge Road. The
data are separated into high and low ozone days, which have been defined as the average ozone
concentration at either FC West, RF-North or NREL Golden for 12-18LT being > 70 ppb (high)
or < 60 ppb (low). The day selection is done separately for observations and model since we are
not evaluating the model for representing the actual conditions on a given day, but for its ability
to represent flow regimes on high ozone days. The results are similar if other nearby Front Range
Foothills sites are used or if we choose a different threshold for high ozone such as 75 ppb.
Note that there are more data points in the observations for low ozone days compared to high
ozone days, which is due to the FRAPPÉ/DAQ time period overall being characterized by few
high pollution episodes. In the model, this is not necessarily the case which is due to the overall
high bias compared to surface ozone sites (Section 4). At FC West, we calculate 21 low and 5
high ozone days from the observations compared to 19 low ozone and 4 high ozone days from
S05, respectively. The number of low and high ozone days amount to 14 and 6 days at RF North
in the observations and to 10 and 12 days in the model. For NREL Golden the respective
numbers of high and low ozone days is 17 days versus 3 days in the observations and 6 versus 15
in the model. The large number of high ozone days at NREL Golden is caused by the high bias in
surface ozone in S05 at this site.
Figures 22-24 show that high surface ozone at the Foothills sites, in general, results in high
ozone at the mountain sites. Ozone pollution roses at Longs Peak and Trail Ridge Road are
shown for 12-20LT to accommodate that the highest ozone concentrations at mountain sites are
typically reached late afternoon/evening. On high ozone days, the measurements show a strong
component of upslope flows with elevated ozone concentrations (> 60 ppb) during the afternoon,
whereas for low ozone days the dominant wind direction is from the west and northeast with
ozone concentrations mostly below 60 ppb. This is in line with findings by Reddy and Pfister
(2016) who state that high ozone days occur mostly on days with upper level high pressure
ridges. In addition to bringing warmer temperatures and fewer clouds, upper level ridges in this
region reduce synoptic winds and thus allow cyclic terrain-driven circulations. It is interesting to
note that at Longs Peak the dominant wind direction on FC West high ozone days is from the SE
compared to the S-SSE for the other two sites.
Final Report for RFP # FAAA 2016000120

38

 The model reflects the measured wind statistics well and on high ozone days suggests
significantly enhanced ozone from the southeast sector. Longs Peak shows a higher variability of
the measured wind direction compared to Trail Ridge Road and the modeled winds at Longs
Peak are from the southeast, similar to measured and modeled winds at Trail Ridge Road,
whereas the observed winds are from the south-southeast. This is explained by the Longs Peak
sites being located in the valley of Cub Creek, which has a northwest to southeast orientation and
likely channels the local winds, whereas the site of Trail Ridge Road is more exposed and
somewhat better captures the regional flow.
In Figure 25 we provide examples of four selected days on which both observations and model
show elevated ozone concentrations in the NFRMA. A common feature on all days is a general
upslope direction even though the exact direction of the upslope flow can vary and locally there
still might occur disturbances in the flow pattern. On all selected days, we find enhanced ozone
at least at one of the high elevation sites located west of the NFRMA.
Another meteorological feature that can lead to high ozone concentrations includes the Denver
Cyclone. A Denver Cyclone forms when a low-level moist, southeasterly flowing air mass meets
the Palmer Divide. If the moist air lifts over the ridge and meets northwesterly winds originating
in the Rocky Mountain foothills, winds may converge to create enhanced cyclonic vorticity.
Dependent on the location of the center of the Cyclone, different types of emissions are injected
into the cyclone and mixed within this flow leading to efficient ozone production. During the
campaign, Denver Cyclone conditions were experienced on two days: July 27 and July 28.
Measured and modeled surface ozone and winds for these days are shown in Figure 26. While
the model represents cyclonic conditions on both days, it incorrectly places the Denver Cyclone
on July 27 too far south misrepresenting the area of high ozone. The distribution on July 28 is
represented better showing the highest ozone in both model and observations in and at the edge
of the Cyclone.

Final Report for RFP # FAAA 2016000120

39

 Longs Peak (FC West > 70 ppb)

Longs Peak (FC West < 60 ppb)
OBS

S05

OBS

Trail Ridge Road(FC West > 70 ppb)

Trail Ridge Road(FC West < 60 ppb)
OBS

S05

S05

OBS

S05

Figure 22: Measured and modeled ozone pollution roses for 12-20 LT for Longs Peak (top) and Trail
Ridge Road (bottom) for days when FC West mean afternoon ozone was < 60 ppb (left columns) or
> 70 ppb (right columns).
Longs Peak (RF North > 70 ppb)

Longs Peak (RF North < 60 ppb)
OBS

S05

OBS

Trail Ridge Road(RF North > 70 ppb)

Trail Ridge Road(RF North < 60 ppb)
OBS

S05

S05

OBS

S05

Figure 23: as Figure 22 but for days when RF North mean surface afternoon ozone was < 60 ppb (left
columns) or < 70 ppb (right columns)

Final Report for RFP # FAAA 2016000120

40

 Longs Peak (NREL Golden > 70 ppb)

Longs Peak (NREL Golden < 60 ppb)
OBS

S05

Trail Ridge Road(NREL Golden < 60 ppb)
OBS

S05

OBS

S05

Trail Ridge Road(NREL Golden > 70 ppb)
OBS

S05

Figure 24: as Figure 22 but for days when NREL Golden mean surface afternoon ozone was < 60 ppb
(left columns) or < 70 ppb (right columns)

Final Report for RFP # FAAA 2016000120

41

 Figure 25: Surface ozone and winds from CMAQ and surface observations for 15 LT and 18 LT on
selected days for which observations and model show elevated ozone concentrations.

Final Report for RFP # FAAA 2016000120

42

 Figure 26: Surface ozone and winds from CMAQ and surface observations for days with a Denver
Cyclone. Results for 15 LT are shown.

9

Other Findings

Canister samples collected with the Whole Air Sampler (WAS) from a mobile van have been
used in Section 3.1 for evaluating the modeled emission ratios. Most of the cans have been taken
near selected emission sources and the data do provide valuable insight into single large emitters.
In Table 5 we list the WAS canister samples that showed the ten highest benzene concentrations.
Note that the location of some of the samples is incorrect in the file uploaded to the NASA Data
Archive. We contacted UC Irvine for providing a correction.
All samples were taken in Weld County as shown in Figure 27. In the same graph, we also show
the location of the highest samples; for some locations, more than one sample was taken. In all
samples, ethyne concentrations are low suggesting low impact from mobile sources. The highest
benzene concentration of 120 ppb (sample 166) was measured near Platteville at a “produced
H2O disposal well” as stated in the canister sampling log sheet. While benzene and also toluene
concentrations are very high, typical products released from oil and gas production (e.g. propane
and ethane) are not among the highest in this sample. The second highest benzene was measured
in samples 206 & 207, which were taken downwind of an oil well waste dumping facility.
Samples 12,13 and 14 with benzene concentrations of ~30-40 ppb also show high methane and
ethane. These were taken next to the Marilyn Compressor Station, where natural gas is
compressed to a specified pressure, thereby allowing it to continue traveling along the pipeline.
The shown samples only provide a snapshot of facilities that where a high release of VOC
concentrations was detected during FRAPPÉ and demonstrate a large variability in the
composition of the emitted VOCs from the different processes used in oil and gas extraction
activities. Some of these sources are not considered in emission inventories nor is the range of
VOC emission ratios.

Table 5: WAS Canister samples for the ten highest benzene concentrations.

Final Report for RFP # FAAA 2016000120

43

 index latitude longitude
CH4_ppmv
Ethane_pptv Ethyne_pptv Propane_pptv Benzene_pptv Toluene_pptv
166 40.2169
-104.7206
1.93
21,810
259
55,296
120,807
105,708
206 40.3197
-104.5678
2.19
60,643
406
158,469
62,666
48,405
14 40.1453
-104.8694
3.45
141,901
144
92,304
42,686
55,206
207 40.3197
-104.5678
2.17
61,376
347
128,046
36,780
37,423
13 40.1453
-104.8694
3.53
353,178
302
198,842
33,673
92,257
12 40.1453
-104.8694
3.41
335,157
277
190,545
30,864
82,111
25 40.2467
-104.8144
3.10
828,036
313
793,001
24,278
29,512
24 40.2467
-104.8144
3.07
538,869
327
635,712
23,564
28,111
68 40.3647
-104.8328
2.34
303,854
134
399,946
11,668
8,046
5 40.0147
-104.8972
2.84
258,838
641
505,596
8,637
7,194

166

206 & 207

12,13 & 14

Figure 27: Location of the ten highest benzene WAS canister samples (top left). and of selected high
benzene canister samples with the sample index listed in each graph.

Final Report for RFP # FAAA 2016000120

44

 10 Summary
We have employed an extensive range of modeling tools to analyze the data set collected during
the FRAPPÉ and DISCOVER-AQ field campaigns. Emphasis was placed on evaluating
available emission data against measurements and determining a best-estimate, adjusted
emissions scenario which could then be used as a base case for simulating possible emission
controls and their impact on photochemical ozone production in the NFRMA.
The WRF/CMAQ model used here captured the general meteorology and resulting tracer
transport and mixing very well, given the challenging, terrain induced flow patterns that are
typical for summer conditions in this area. Planetary Boundary Layer heights were also generally
well simulated, with exceptions on some days. The model did not reproduce well afternoon
cloudiness and development of convective storms, resulting, on average, in an overestimate of
available photochemically active radiation and resulting ozone production. Many of the high
ozone days were simulated very well in the model and the available data on these days allowed
for exhaustive evaluation of all aspects of the analysis.
Box models were used to assess the performance of the lumped chemistry schemes that are used
in time dependent chemical transport models and we found no strong biases caused by the
simplification of the chemistry. Box models and chemical steady state models were also used to
inform about the photochemical ozone production regimes across areas of the NFRMA that are
dominated by different emission sectors and to support independent verification of conclusions
drawn from the WRF/CMAQ zero-out scenarios.
We found that significant adjustments were necessary to the reported emissions from the oil and
gas as well as the mobile emission sectors. We recommend that more measurements are made,
possibly with CDPHE’s new mobile lab, to find possible missing emission sources or identify
discrepancies between reported and actual emissions (see below).
Zero-out scenarios were used in both WRF/CMAQ and the box models to determine the
influence of the different major emission sectors on regional ozone production, suggesting
possible emissions control efforts by the Colorado legislature.
We found that the oil and gas sector as well as the mobile sector are major contributors to
NFRMA ozone, with the oil and gas sector influencing more often the northern part of the
NFRMA while mobile emissions dominate more in the southern metro area, on average. Large,
concentrated NOx emitters such as the power generating stations, DIA, and emissions from some
large industrial plants cause a reduction in ozone close to these sources but contribute to ozone
production farther downwind, affecting ozone in the western suburbs of the NFRMA and the
Foothills during high production days. Prevailing daytime transport to the west during summer
exposes the Foothills and Front Range mountains, including Rocky Mountain National Park and,
occasionally, the adjacent western valleys to high ozone originating from the NFRMA.
Measures resulting in lowering the emissions from oil and gas as well as from mobile sources
could result in substantial reductions of NFRMA ozone. We calculate as much as 10 ppb MDA8
ozone reductions on high ozone days when either sector is zeroed out. These reductions are
somewhat spatially separated on classic upslope days while the occurrence of a Denver Cyclone
can mix the emissions from both sectors more efficiently resulting in large increases in ozone in
Boulder and Gilpin counties. Ozone production in the northern part of the NFRMA is, on
Final Report for RFP # FAAA 2016000120

45

 average, more NOx limited while the southern part of the NFRMA is closer to being VOC
limited in some places. This places emphasis on NOx availability for efficient ozone production
in the Greeley, Loveland and Fort Collins urban areas. NOx emissions inside the DenverJulesburg basin are particularly important for how efficiently ozone is produced before NOx
availability increases due to transport of these air masses across the urban regions to the west.
Current emission assessment indicates that lowering NOx emissions in Weld county would have
a larger effect on lowering ozone than VOC emission controls from the O&G sector. However,
this is only the case if NOx emissions from the mobile sector do not increase in this area, which
could be the case given the increasing push of suburban development into the eastern sections of
the NFRMA. Because the evaluation of the NOx emissions in the region is difficult, mainly
because of the short atmospheric lifetime and the distribution of large point sources, the NOx
budget is one of the most uncertain parameters in this analysis.
Measures resulting in the lowering of mobile emissions would reduce ozone everywhere in the
NFRMA, emphasizing the western and southwestern sectors of the region.
NOx reductions from the EGUs in the area (shutdown of the Valmont plant and conversion of
the last boiler of the Cherokee plant to natural gas) will result in localized increases of ozone (but
also lower the NO2 concentrations there), but will help lower ozone somewhat in the western
suburbs.
Industrial VOC emissions appear less important compared to mobile and O&G on the large
scale, but can be very important locally, such as in Commerce City. Overall reductions in
industrial emissions would only have a small impact on some of the western suburbs, although
somewhat larger than that from the EGU emissions.
The Foothills communities and Wilderness areas in the Mountains and adjacent western valleys
would generally benefit from a reduction of emissions in the NFRMA as this would result in less
advected ozone from the NFRMA as well as less additional production during air mass transport.
We have identified a number of point sources which, at the time of measurement, emitted
alarming amounts of highly reactive and toxic VOC. Some of these sources were not part of the
emission inventory. From the sporadic observations made during FRAPPÉ it is impossible to tell
what fraction these sources contribute to the derived adjustments to the VOC emissions in the oil
and gas sector, for example. Better constrained emission inventories would help State
Implementation Plan (SIP) modeling efforts and greatly aid the process of developing effective
ozone reduction strategies.
Finally, since emissions in the NFRMA are constantly evolving, due to rapid population growth
causing ever increasing commuting miles driven as well as impacting transport of goods in and
out of the area, and due to market driven rapid changes in the oil and gas extraction activity in
the region, repeated aircraft measurements like what was done during FRAPPÉ would greatly
help to detect these trends, now that a baseline has been established with the 2014 campaign.

Final Report for RFP # FAAA 2016000120

46

 11 FRAPPÉ Website, Meetings and Public Outreach
We have been updating the NCAR/ACOM FRAPPÉ Website on a regular basis
(https://www2.acom.ucar.edu/frappe). To the Best of our knowledge the list of publications on
the website represent the latest published papers. Numerous other publications have been
submitted to journals and will be added to the website as soon as they have undergone the peerreview process.
In May 2017, we organized a joint FRAPPÉ and DISCOVER-AQ Science Team Meeting. All
presentations have been made available on the NCAR/ACOM FRAPPÉ website. As part of the
Science Team Meeting and in recognition of the EPA Air Awareness Week NCAR held an Air
Quality Fair at the NCAR Mesa Lab with booth by NCAR, CU, RAQC, EPA, GO3, NPS, Ball
Aerospace and many others.
Presentations have been given to Policy Makers and AQ Managers on the progress of FRAPPÉ
analysis. The more recent presentations include talks at the Colorado Air Quality Control
Commission (AQCC) Spring 2017 Science Forum, the Air & Waste Management Association
(AWMA) in November 2016 and a presentation to the Air Pollution Control Commission
(APCC) in August 2016. The authors will be available to consult with CDPHE if questions arise
with respect to the materials made available and results presented here, and are available to
present and discuss the results and findings from this report at upcoming CDPHE or related
meetings if requested.
More recent public outreach activities included NCAR Earth Explorer Talks at the Boulder Mesa
Lab in January 2017 and the Denver Library in June 2017. A recording of the Boulder Talks is
available from the UCAR Connect Youtube site: https://www.youtube.com/
watch?v=kCPzUFaWwBs)

12 Data and Tool Availability
The tools and data sets used in this report are publicly available.
•
•
•
•
•

BOXMOX: http://boxmodeling.meteo.physik.uni-muenchen.de/index.html
CMAQ: https://www.cmascenter.org/cmaq/
SMOKE: https://www.cmascenter.org/smoke/
FRAPPÉ and DISOVER-AQ Data Set:
https://www-air.larc.nasa.gov/missions/discover-aq/discover-aq.html
WRF Modeling System: http://www2.mmm.ucar.edu/wrf/users/

We offer to make all graphs included in this report available as electronic version. In addition,
we can provide a full set of additional graphs including model comparisons to daily C-130 and
P-3 flights, daily maps of measured and modeled surface ozone and similar. We further will
provide, upon request, all analysis codes (mostly written in NCL and R), as well as the CMAQ a
priori and S0.5 emission inputs, the WRF ADP/FRAPPÉ nudging model output and the S0.5
CMAQ model output or any other data set. Given the large data volume involved, however, we
consider it the most feasible to receive input from the State on the requested set of information
and discuss on the appropriate method for the data exchange, which will be decided on
dependent on the requested amount of data.
Final Report for RFP # FAAA 2016000120

47

 We thank the FRAPPÉ and DISCOVER-AQ Science Team for the collection, quality control and
availability of the observational data sets. Due to the wealth of data set included in this report we
are not able to list each data PI individually. All observations included in this report have been
downloaded from the publicly available NASA FRAPPÉ/DISCOVER AQ Data Archive
(https://www-air.larc.nasa.gov/missions/discover-aq/discover-aq.html). Each data set contains
the contact information for the data PI. We acknowledge the State of Colorado/CDPHE and the
National Science Foundation (NSF) for funding of FRAPPÉ. NASA is acknowledged for
funding DISCOVER-AQ.
References
Baumann, K., E. J. Williams, J. A. Olson, J. W. Harder, and F. C. Fehsenfeld (1997),
Meteorological characteristics and spatial extent of upslope events during the 1993
Tropospheric OH Photochemistry Experiment, J. Geophys. Res., 102(D5), 6199–6213,
doi:10.1029/96JD03251.
Benedict, K. B., Carrico, C. M., Kreidenweis, S. M., Schichtel, B., Malm, W. C. and Collett, J.
L. (2013), A seasonal nitrogen deposition budget for Rocky Mountain National Park.
Ecological Applications, 23: 1156–1169. doi:10.1890/12-1624.1
Brodin, M., D. Helmig, S. Oltmans (2010), Seasonal ozone behavior along an elevation gradient
in the Colorado Front Range Mountains. Atmospheric Environment 44 :5305-5315
Brodin, M., D. Helmig, B. Johnson, and S. Oltmans (2011), Comparison of ozone concentrations
on a surface elevation gradient with balloon-borne ozonesonde measurements. Atmos.
Environ., 45, 5431- 5439.
Darrouzet-Nardi, A., J. Erbland, W. D. Bowman, J. Savarino, and M. W. Williams (2012),
Landscapelevel nitrogen import and export in an ecosystem with complex terrain, Colorado
Front Range. Biogeochemistry, 109, 271-285.
Doran, J.C. (1996), The influence of Canyon Winds on flow fields near Colorado's Front Range,
J. Appl. Meteorology, Volume 35, 587-600.
Greenland, D. (1980), The Climate of Niwot Ridge, Front Range, Colorado, U.S.A., Arctic and
Alpine Research, vol. 21, no. 4, 1989, pp. 380–391., www.jstor.org/stable/1551647.
Haagenson, P. L., (1979), Meteorological and climatological factors affecting Denver air
quality, Atmos. Environ., 13, 79–85.
Johnson, R.H. and J.J. Toth (1982), Topographic effects and weather forecasting in the Colorado
PROFS mesonetwork area, Preprint Volume: 9th Conference on Weather Forecasting and
Analysis, June 28 - July 1, 1982; Seattle, WA, 440-445.
Olson, J. A., K. Baumann, C. J. Volpe, J. W. Harder, E. J. Williams, and G. H.
Mount (1997), Meteorological overview of the 1993 OH Photochemistry Experiment, J.
Geophys. Res., 102(D5), 6187–6197, doi:10.1029/96JD00402.
Olson, J. R., Crawford, J. H., Chen, G., Brune, W. H., Faloona, I. C., Tan, D., Harder, H., and
Martinez, M. (2006), A reevaluation of airborne HOx observations from NASA field
campaigns, J. Geophys Res., 111 D10301, doi:10.1029/2005JD006617.

Final Report for RFP # FAAA 2016000120

48