DAQ-2024-004526

1

1
The Salt Lake regional Smoke, Ozone and Aerosol Study (SAMOZA) 2
Final Report, September 29, 2023 3
Principal Investigators: Daniel Jaffe (UW), Lu Hu (UMt) and Seth Lyman (USU) 4
5
Executive summary 6
The Salt Lake City region is one of approximately 50 metropolitan regions around the U.S. that 7
do not meet the 2015 8-hour ozone (O3) National Ambient Air Quality Standard (NAAQS). To 8
better understand the causes of high O3 days in the region, a group of scientists from the 9
University of Washington, Utah State University and the University of Montana developed and 10
proposed the Salt Lake regional Smoke, Ozone and Aerosol Study (SAMOZA). The primary 11
goals of SAMOZA are: 12
1. Make observations of a suite of VOCs, including many oxygenated VOCs by Proton 13
Transfer Reaction Mass Spectrometry (PTR-MS) and the 2,4-dinitrophenylhydrazine 14
(DNPH) cartridge method. 15
2. Evaluate whether UDAQ O3 measurements show a positive bias during smoke events. 16
3. Quantify the range of concentrations of NOx, VOCs, CO and PM2.5 on smoke-influenced vs 17
non-smoke days. 18
4. Conduct photochemical modeling and statistical modeling/machine learning analyses to 19
improve our understanding of the sources of O3 and PM2.5 photochemistry (NOx vs VOC 20
sensitivity) on both smoke-influenced and non-smoke days during the summer of 2022. 21
Key results: 22
i. We found no significant difference in the O3 measurements from the “scrubber-less” UV 23
instrument compared to the standard O3 measurements made by UDAQ with a Teledyne 24
T400 instrument at PM2.5 concentrations up to 60 µg m-3. 25
ii. For formaldehyde (CH2O), which was measured by two different methods, there is a 26
generally good correlation in the data from the two methods, but the PTR-MS measurements 27
are approximately 50% greater than the DNPH measurements on smoky days. The cause 28
for this difference is not yet known. 29
2

iii. On days with smoke, we found that PM2.5, CO, O3 and nearly all VOCs were significantly 30
enhanced. On average, NOx was also enhanced on days with smoke, but this was 31
complicated by day of week effects on NOx concentrations (higher on weekdays). 32
iv. Photochemical modeling of O3 production rates at the Utah Tech Center for both smoke 33
influenced and no smoke days demonstrates a strong sensitivity to VOC concentrations and 34
less sensitivity to NOx. For non-smoke days, reductions in VOCs of ~30% result in 35
significantly reduced O3 production. Reductions in NOx of ~60% are needed to get a 36
significant reduction in O3 production for non-smoke days. 37
v. The photochemical modeling shows that formaldehyde and other oxygenated VOC, along 38
with alkenes, were the most important O3 precursors. 39
vi. Generalized Additive Modeling (GAM) gave similar MDA8 O3 enhancements on smoky 40
days as the photochemical modeling. Analysis of the GAM results show that 19-31% of 41
the smoke days have GAM residuals that exceed the EPA (2015) criteria for statistical 42
analysis of O3 data, and thus this method could be used as support for exceptional event 43
cases for those days. 44
45
3

Introduction 46
Surface ozone (O3) is formed from photochemical reactions of nitrogen oxides (NOx = NO + 47
NO2) and volatile organic compounds (VOCs). O3 has serious health impacts up to and including 48
premature mortality (Bell et al., 2004; Di et al., 2017). The Northern Wasatch Front/Salt Lake 49
City metropolitan region is one of approximately 50 regions in the U.S. that are considered non-50
attainment areas (NAAs) for the 2015 O3 standard (https://www.epa.gov/air-trends/air-quality-51
design-values). The National Ambient Air Quality Standard (or NAAQS) for O3 is currently 70 52
ppb and is based on the three-year mean of the annual fourth highest, maximum daily 8-hour 53
average (MDA8) O3 concentration. 54
In the Western U.S. there are several important challenges to meeting the standard. First 55
background O3, defined here as the distribution of concentrations that are observed in rural areas 56
of the western U.S., is high due to the combined influences of stratospheric intrusions, a deep 57
mixed layer and increasing area of wildfires burned each year (Jaffe et al 2018; 2020). 58
Observations and models suggest that Nevada and Utah have some of the highest concentrations 59
of background O3 in the U.S. with a much larger contribution from the stratosphere compared to 60
long distance anthropogenic sources (Langford et al 2017; Mathur et al 2022). In addition to 61
these sources, wildfires emit O3 precursors and can have substantial impacts on surface O3 62
concentrations (Gong et al 2017; Buysse et al 2019; McClure and Jaffe-2018; Jaffe et al 2018; 63
2022; Rickly et al 2023; Permar et al 2023). 64
Specifically for the Northern Wasatch Front/Salt Lake City (SLC) region, concentrations of 65
nitrogen oxides have been declining for the last decade, but the fourth highest O3 MDA8 has 66
been essentially stagnant over this time. In a study of national O3 trends at 40 U.S. non-67
attainment areas, Jaffe et al (2022) found two things that were somewhat unusual for the SLC 68
region. First, the relationship between annual fourth highest MDA8 O3 and annual mean NO2 69
was amongst the weakest of any of the sites considered. Second, while NO2 concentrations have 70
declined and displayed the typical pattern of higher values on weekdays, the pattern of enhanced 71
O3 remains relatively insensitive to the day of week. Similar relationships were also seen at 72
other western U.S. sites and these patterns were attributed to the interannual variations in the 73
influence from wildfires and stratospheric intrusions. At the same time, it is important to 74
recognize that local emission sources also impact O3 and, absent emissions from human and 75
industrial sources, O3 concentrations would rarely exceed the NAAQS levels (Jaffe et al 2020). 76
4

77
Project goals 78
Given this background, a group of principal investigators from the University of Washington 79
(UW), Utah State University (USU) and the University of Montana (UMt) developed the 80
SAMOZA plan. The over-arching goal of SAMOZA is to improve our understanding of O3 and 81
PM2.5 in the SLC region during summer. Specific goals are: 82
1. Make observations of a suite of VOCs, including many oxygenated VOCs by Proton 83
Transfer Reaction Mass Spectrometry (PTR-MS) and the 2,4-dinitrophenylhydrazine 84
(DNPH) cartridge method. 85
2. Evaluate whether UDAQ O3 measurements show a positive bias during smoke events. 86
3. Quantify the range of concentrations of NOx, VOCs, CO and PM2.5 on smoke-influenced 87
vs non-smoke days. 88
4. Conduct photochemical modeling and statistical modeling/machine learning analyses to 89
improve our understanding of the sources of O3 and PM2.5 photochemistry (NOx vs VOC 90
sensitivity) on both smoke-influenced and non-smoke days during the summer of 2022. 91
Long-term context 92
Figure 1 shows the long-term pattern of O3 and NO2 at the Hawthorne monitoring site near 93
downtown SLC. Figure 2 shows the same O3 data along with the number of “smoke days” in 94
each year. Smoke days are defined as days with satellite observed overhead smoke (from the 95
NOAA HMS product (Rolph et al 2009; Kaulfus et al 2017) and surface PM2.5 greater than the 96
mean+1 SD of days with no overhead HMS smoke. The mean and SD for surface PM2.5 with no 97
overhead smoke is calculated from the daily mean values for May-Sept for each individual year. 98
5

Figure 1. Annual fourth 99
highest MDA8 O3 and 100
average May-September daily 101
1-hour maximum NO2 and 102
daytime mean NO2 (0700-103
1400 local standard time) for 104
the Hawthorne monitoring 105
site. 106
107
108
109
110
111
112
113
114
115
116
Figure 2. Number of O3 117
exceedance days (MDA8>70 118
ppb) each year at Hawthorne 119
site (left axis) number of 120
exceedance days with smoke 121
(left axis) and number of 122
smoke days each year (right 123
axis). 124
125
126
127
128
129
130
131
132
Figure 2 suggests that in some years, such as 2021 and possibly 2017, strong influence from 133
smoke likely enhanced O3 and contributed to the elevated fourth highest MDA8 in those years. 134
The SAMOZA experimental period (summer 2022), appears to have taken place during a 135
relatively lower fourth highest MDA8 O3 value (72 ppb) and number of smoke days (13 days), 136
compared to the past decade (76 ppb and 18 days, respectively). At Hawthorne, there were 4 137
exceedance days in 2022, two of which had smoke, while at the UTC site, there were 6 138
exceedance days in 2022. 139
140
6

Methods 141
Measurements for the SAMOZA study were conducted from August 1-October 1, 2023 at the 142
Utah Department of Environmental Quality Technical Support Center (hereafter simply UTC), 143
located approximately four miles east of the Salt Lake City International Airport at 40.78 °, 144
-111.94°. This site was chosen due to available space and power, along with the fact that other 145
key observations were already being made there. The SAMOZA measurements included the 146
following: 147
a. O3 using a scrubber-less UV instrument (2B Technologies, Model 211). 148
In addition to the standard UDAQ measurements of O3 at UTC, the SAMOZA team 149
measured ambient O3 concentrations using a 2B Technologies Model 211 O3 monitor (Boulder, 150
CO), a dual-beamed 254 nm photometer at 1-minute resolution. This instrument uses the reaction 151
between ambient O3 and NO generated in situ by upstream photolysis of added nitrous oxide 152
(N2O) to quantify ozone by UV photometry without the issues affecting conventional O3 153
scrubbers. The instrument was calibrated daily during the campaign against a reference 154
photometer (2B Technologies model 306) that was itself cross referenced to an O3 standard from 155
the NOAA Global Monitoring Lab (Birks, et al. 2018). The O3 monitor (and the CO instrument) 156
shared the same sampling line as the PTR-MS described below. Resolution of the O3 monitor is 157
0.1 ppb, with a limit of detection (2σ) of 1.0 ppb for a 10 s average. 158
b. Proton Transfer Reaction Mass Spectrometric (PTR-MS) measurements of 159
Volatile Organic Compounds (VOCs). 160
Ambient VOCs were measured using proton-transfer-reaction time-of-flight mass 161
spectrometry (PTR-ToF-MS 4000, Ionicon Analytik GmbH, Innsbruck, Austria). The conditions 162
in the drift tube were held constant during the campaign at 3.00 mbar, 60°C, and 815V, which 163
made for an electric field of 135 Td. The PTR-MS was located on the second floor of the 164
Technical Support Center. The sampling inlet was made from perfluoroalkoxy (PFA) tubing and 165
was situated on the roof of the building, ~20m above ground level. The air was subsampled by 166
the PTR-MS through ~100 cm of 1/16” (1.59 mm) OD polyetheretherketone (PEEK) tubing 167
maintained at 60°C. Ions from m/z 19 to 400 were measured once every minute. Instrument 168
background was taken approximately every 2½ hours by measuring VOC-free air generated by 169
ambient air passing through a heated catalytic converter (375°C, platinum beads, 1wt% Pt: 170
Sigma Aldrich). 171
7

Calibrations were performed for 25 species in two compressed gas standard cylinders (stated 172
accuracy 5% at ~1 ppmv; Apel-Riemer Environmental, Inc., Miami, FL; Permar et al., 2021). 173
One cylinder with 10 species was used every other day for the duration of the campaign and was 174
calibrated via dynamic dilution followed by addition of gas to the VOC-free air described above. 175
A second standard gas cylinder containing 15 species was used every other day for the first three 176
weeks of the campaign. Six-point calibrations were performed between 1 and 7 ppb. Only those 177
calibrations with an R2 above 0.998 and with sensitivities for the same species within 10% 178
during the campaign were used. From quadrature addition of individual errors including 179
calibration and mass flow controllers in the instrument, uncertainty for these species is <15%. In 180
addition, D5 Siloxane was calibrated with a gas standard in June 2022 before the campaign 181
(stated accuracy 5% at ~1ppm; Apel-Riemer Environmental, Inc., Miami, FL) using dilution as 182
described above. Uncertainty for this species is <15%. 183
Formaldehyde was calibrated after the campaign using a gas standard in a compressed 184
cylinder (stated accuracy 5% at ~2ppmv; Airgas USA LLC, Plumsteadville, PA) diluted with a 185
zero-air generator (7000 Zero Air Generator, Environics, Tolland, Connecticut). Gases were 186
mixed in a Liquid Calibration Unit (Ionicon Analytik GmbH, Innsbruck, Austria) and water was 187
introduced to find the dependence of sensitivity on changing humidity. Formic acid and acetic 188
acid were calibrated before the campaign using liquid standards evaporated and diluted with 189
zero-air in the same Liquid Calibration Unit. Water vapor was used to find humidity dependence 190
of sensitivity as described above. Uncertainty for these species is estimated at 40%, with the 191
major source of error being instrument drift over time. 192
Sensitivity for maleic anhydride was estimated using the method by Sekimoto et al. (2017) 193
from its molecular dipole moment and polarizability. The procedure for the calculation was 194
further refined in a previous work (Permar et al., 2021). The uncertainty for this species is 195
estimated to be 50%. 196
Mass spectra were first analyzed with Ionicon’s PTR-Viewer software (Version 3.4, Ionicon 197
Analytik). One-minute mass calibrations performed during the campaign were refined using 4 198
ion peaks: m/z 29.9974 [NO+], 59.0491 [C3H6OH+], 203.943 [C6H4IH+], and 330.848 199
[C6H4I2H+]. Ion masses were assigned molecular formulae using a peak list included with the 200
software which was compared and adjusted according to a library of previously published PTR-201
MS mass peaks (Pagonis et al., 2019). Ion counts for each peak in the list were calculated by the 202
8

PTR-Viewer software through a baseline correction as well as a correction for mass 203
discrimination in the time-of-flight. The calculated ion counts were then exported for further 204
processing in R. Instrument background was linearly interpolated and subtracted from the data. 205
Each ion was normalized to the primary ion [H3O+] and water cluster ion [(H2O)H3O+]. 206
Normalized counts were converted to mixing ratios using the sensitivities found during 207
calibration. 208
c. DNPH measurements of carbonyl species 209
We collected carbonyl samples by pulling ambient air through 2,4-dinitrophenylhydrazine 210
(DNPH) cartridges (Waters WAT037500) with potassium iodide cartridges (WAT054420) 211
upstream to remove ozone. The sample path upstream of the cartridges was composed entirely 212
of PFA and PTFE Teflon, with a PTFE filter (5 μm pore) upstream of the sample line to filter 213
particles (5 μm pore size). The DNPH cartridges were installed in automatic sampling trays with 214
solenoid valves to control flow through each cartridge. A pump provided flow, and a mass flo w 215
meter measured the flow rate. Flow through cartridges was about 1 L min-1. A Campbell 216
Scientific CR1000 data logger controlled the system and recorded sample flow rates. We 217
collected three 3-h DNPH samples per day from 1 August 2023 through 3 October 2023. Daily 218
sampling times were 9:30-12:30, 12:30-15:30, and 22:30-0:30 local standard time. We replaced 219
the DNPH cartridges in the sampling trays weekly. After sampling, the cartridges were kept 220
refrigerated storage and transport. One pair of samples was collected simultaneously on two 221
different trays each week as a duplicate. Field blanks were collected weekly by installing DNPH 222
cartridges in a sampling tray and immediately removing them. 223
We eluted cartridges within 14 days of sampling and analyzed the eluent within 30 days. To 224
elute DNPH cartridge samples, we flushed cartridges with 5 mL of a solution of 75% acetonitrile 225
and 25% dimethyl sulfoxide (percent by volume). We collected the solution into 5 mL 226
volumetric flasks and brought the flasks to a volume of 5 mL using 0.5–1 mL of the 227
acetonitrile/dimethyl sulfoxide solution. Finally, we pipetted a 1.6 mL aliquot from the 5 mL 228
flask into two 2 mL autosampler vials for analysis by high-performance liquid chromatography 229
(HPLC). The second vial was kept as a spare in case of contamination or equipment failure. 230
We used a commercial standard mixture (M-1004; AccuStandard) of derivatized carbonyls 231
in acetonitrile for calibration. We analyzed samples with a Shimadzu Nexera-i LC-2040C 3d 232
Plus HPLC and a Shimadzu Shim-Pack Velox C18 column. We used a mixture of acetonitrile, 233
9

tetrahydrofuran, and water as the eluent. We calibrated the instrument on each analysis day with 234
a 5-point calibration curve. We ran at least 1 additional calibration standard at t he beginning and 235
end of each analysis batch to check for retention time drift or other errors. Target compounds 236
analyzed are formaldehyde, acetaldehyde, acetone, acrolein, propionaldehyde, crotonaldehyde, 237
2-butanone, methacrolein, n-butyraldehyde, valeraldehyde, m-tolualdehyde, and hexaldehyde. 238
More information about the methods used is available in Lyman et al. (2021). 239
Compounds in the laboratory blanks were 0.1 ± 0.1 ppb (average ± 95% confidence 240
interval), and field blanks were 0.2 ± 0.2 ppbv in air. (The average volume of air sampled by 241
field samples was applied to blank samples to convert blank results to units of ppbv in ambient 242
air.) All samples were blank-corrected. Compounds in duplicate samples were 20 ± 5% 243
different. Calibration recovery was 103 ± 2%. Detection limits 0.1 to 0.2 ppb. 244
We also received three DNPH cartridges loaded with carbonyls from Eastern Research 245
Group (ERG), an independent laboratory that performs DNPH cartridge analysis for the U.S. 246
Environmental Protection Agency and others. Excluding crotonaldehyde, our analytical results 247
were 14 ± 4% higher than the results from ERG. Our crotonaldehyde results were 105 ± 30% 248
higher, perhaps indicating an error in peak integration for crotonaldehyde. 249
d. CO using a gas chromatography (GC) with a reducing compound photometer 250
In addition to the standard UDAQ measurements of CO at UTC, the SAMOZA team made 251
cconcurrent CO measurements using chromatography (GC) with a reducing compound 252
photometer (Peak Performer 1; Peak Laboratories LLC., USA). CO eluting from the GC column 253
pass directly into a heated mercuric oxide bed, resulting in liberated mercury vapor, which is 254
subsequently measured via UV light absorption in the photometer cell. Compressed ultra-high 255
purity air was used as the carrier gas. Multi-point calibrations are carried out before and after the 256
campaign by dilution of a ppmv-level standard (Scott Specialty Gases, USA; stated accuracy ±2 257
%) into UHP air. The detection limit for CO is 300 pptv and the time resolution of the data was 258
3-minutes. 259
Meteorology, NOx and PM2.5 data were collected by the Utah Department of Environmental 260
Quality at the UTC site and these were used in our analysis. 261
262
10

Models. 263
In addition to these observations several other modeling tools were used including the 264
FOAM box model (Wolfe et al 2016; Ninneman and Jaffe 2021), a machine learning/statistical 265
modeling (Gong et al 2017) and Positive Matrix Factorization. Each of these will be described 266
along with the specific results in the results section below. 267
268
Results 269
Overview 270
Meteorological data were collected at the measurement site by the Utah Department of 271
Environmental Quality. Relative humidity and ambient temperature were measured 272
simultaneously with an electronic thin film air temperature and relative humidity sensor. Wind 273
direction and speed were measured with 2D-ultrasonic anemometer transducers. All 274
meteorological instruments were situated on a tower on the UDEQ building at the UTC. During 275
the SAMOZA campaign, daily maximum temperatures (DMT) averaged 30.8°±4.8oC (mean ± 276
standard deviation). Temperatures were warmer for the first part of the campaign through 277
September 8th (mean DMT of 33.3oC) when temperatures started to cool down for the remainder 278
of the campaign (mean DMT of 26.4oC). Winds came mostly from the southeast, with some 279
influence from the northwest. The highest wind speed seen was 19m/s, while the average was 280
around 6 m/s. 281
282
11

283
Figure 3. Time series of selected trace gas concentrations measured at the UTC site during the 284
SAMOZA campaign (8/1/2022-9/30/2022). 285
286
As mentioned above, the summer of 2022 was a relatively modest year for smoke, with only 287
about 10 days with identified smoke. At the UTC site there were 6 exceedances days 288
(MDA8>70 ppb), five of which occurred during the SAMOZA experimental period (Aug.-Sept. 289
2022. Two of these exceedance days were associated with smoke (9/7/22 and 9/11/22). Four of 290
the smokiest days were Sept. 9-12, 2022, with the highest daily mean PM2.5 values observed on 291
Sept. 10th (33.8 µg m-3). Sept 11th also had smoke, with a daily mean PM2.5 value of 25.3 µg m-3 292
and an O3 MDA8 of 80 ppb. During this time, smoke covered a large portion of the western 293
U.S., with large fires burning in Idaho, Oregon, Washington and California. 294
We use the NOAA Hazard Mapping System-Fire and Smoke Product (hereafter simply 295
HMS) as an indicator of overhead smoke (https://www.ospo.noaa.gov/Products/land/hms.html). 296
This satellite product provides mapping of fire locations and smoke extent for North America on 297
a daily basis. However, as many have pointed out (e.g., Kaulfus et al., 2017, etc.), the smoke 298
extent maps are indicative of overhead smoke and not necessarily of surface smoke. Figure 4 299
shows the hourly distribution of PM2.5 at UTC for August-September of 2022, binned by daily 300
HMS smoke detections, where HMS=0 indicates no overhead smoke and HMS=1 indicates 301
overhead smoke detected. The standard deviation (SD) within each hourly bin is only shown for 302
the HMS=0 data. The SDs within each hour for the HMS=1 data are larger, in the range of 9 -13 303
12

µg m-3. The overall means (using the daily average data) for the HMS=0 and 1 data are 6.2 and 304
11.9 µg m-3 and there were 42 and 19 days in each category, respectively. 305
Binning the data by only the HMS status will include some time periods with overhead 306
smoke, but minimal influence at the surface. For this reason, we add a PM2.5 criteria to identify 307
“smoke days” at the surface. For this we use the mean (6.23 µg m -3) and 1 SD (1.85 µg m-3) of 308
the PM2.5 concentrations on HMS=0 days, such that smoke days are defined as those with 309
HMS=1 and the daily mean PM2.5 > 8.1 µg m-3. For August-September 2022, there were 10 days 310
that met the criteria as a “smoke day” and 51 that were deemed a “no-smoke day”. Figure 5 311
shows the diurnal pattern of PM2.5 data binned by the smoke day criteria and Table 1 shows a 312
summary of the SAMOZA data binned by the smoke/no smoke categories. 313
There are several duplicate measurements. CO was measured using both the standard 314
UDAQ instrument as well as one provided by SAMOZA. While the agreement between these 315
two measurements is good (the correlation coefficient of hourly data is 0.86), we do see that the 316
SAMOZA data are biased high compared to the UDAQ measurements, as evidenced by both the 317
mean values (see Table 1) and the correlation slope of 1.15 (SAMOZA CO vs UDAQ CO). 318
Given that this has little policy implication, we do not investigate the cause of this bias further. 319
320
Figure 4. PM2.5 321
measured at UTC in 322
August-September 323
2022, binned by HMS 324
smoke detections. 325
Errors bars are 1 326
standard deviation 327
and are only shown 328
on the HMS=0 data 329
for clarity. This 330
includes 19 days with 331
HMS=1. 332
333
334
335
13

336
Figure 5. PM2.5 337
measured at UTC in 338
August-September 339
2022, binned by 340
“smoke day” criteria 341
(mean+1 SD of daily 342
mean PM2.5 on 343
HMS=0 days). 344
Errors bars are 1 345
standard deviation 346
and are only shown 347
for the no smoke 348
data. This includes 349
10 days identified as 350
“smoke days” 351
352
353
354
14

355
Table 1. Summary of data during the SAMOZA experiment (August-September, 2022). All data 356
are in ppb except PM2.5 (µg/m3) and ∑VOCs as C (ppbC). 357

Daily
PM2.5
MDA8
O3a COa COb O3a O3b NOx
All data 8.0 56.9 182 314 37.4 38.7 15.8
No smoke (n=51
days) 6.2 55.4 160 284 37.0 38.2 15.0
Smoke days
(n=10 days) 17.3 65.0 293 446 39.4 41.1 20.5

Formal-
dehydec
Ace-
tonec
Iso-
prenec ∑VOCsd ∑VOCs as
C (ppbC)d
Formal-
dehydee
Ace-
tonee
All data 4.3 3.4 0.4 15.6 37.9 3.6 4.0
No smoke (n=51
days) 3.7 3.2 0.3 14.2 34.6 3.4 3.8
Smoke days
(n=10 days) 7.1 4.4 0.5 22.1 53.2 4.6 5.0
aThis column reports data from the UDAQ instrument. 358
bThis column reports data from the SAMOZA instrument. 359
cData from PTRMS instrument. 360
dThis includes these VOCs as reported by the PTRMS instrument (Formaldehyde, Propyne, 361
Acetonitrile, Acetaldehyde, Formic Acid, Butenes, Acetone, Isoprene, MVK_MACR, MEK, Benzene and 362
Toluene. See PTRMS dataset for full list of compounds measured. 363
eData from DNPH cartridge method. 364
365
Next we can examine the diurnal cycle of various pollutants as a function of smoke/non-366
smoke status. Figure 6 shows the average diurnal cycle for CO, O3, formaldehyde, NOx, 367
∑VOCs and ∑VOCs as ppbC. 368
369
15

370
371
372
373
Figure 6: Diurnal cycles of CO, O3, formaldehyde, NOx, ∑VOCs (ppb) and ∑VOCs (as ppbC). 374
Error bars show one standard deviation on the no smoke data. 375
376
In general, the smoke/non-smoke patterns are consistent with previous work (e.g. Buysse et 377
al 2017; Ninneman and Jaffe 2021). Some of the key patterns are: 378
1. O3 increases more rapidly during the morning hours and reaches higher peak values for 379
smoky conditions compared to non-smoky conditions. 380
16

2. CO, PM2.5 and VOCs are all significantly elevated on smoky days, compared to non-381
smoky days. 382
3. Emissions from traffic are seen on all days, as evidenced by rapid increases in CO, NOx 383
and VOCs in early morning hours. 384
The most surprising result is that NOx concentrations appear to be greater on smoky days, 385
which differs slightly from previous work. Buysse et al (2019) found that NOx was enhanced 386
on some days at some locations during smoke, but not at other locations. We found that 387
NOx was higher on average, but the pattern is not very robust, given the high variability in 388
NOx concentrations. Figure 7 compares the NOx diurnal cycle on two smoke days with data 389
from all nonsmoke days. 390
391
Figure 7: Average 392
diurnal cycle for 393
NOx for all non-394
smoky days and 395
the diurnal pattern 396
for September 11th 397
and 12th, 2022. 398
The MDA8 O3 399
values were 80 and 400
68 ppb for 401
September 11th and 402
12th, respectively. 403
404
405
406
407
We note that Sept. 11th was a Sunday, and while there was a surprising level of NOx overnight, 408
by morning the levels had returned to typical values for a non-smoky day. In contrast, Sept. 409
12th, a Monday, had very high NOx levels which persisted until past noon. The pattern of O3 on 410
these two days is counter to the NOx values, with a higher MDA8 on Sunday the 11th (80 ppb), 411
compared to Monday the 12th (68 ppb), suggesting that the high NOx levels on the 12th, may have 412
suppressed O3 formation. Figure 8 shows the diurnal pattern for all non-smoky days vs day of 413
week. A clear pattern is evident with highest NOx concentrations on Monday and Tuesday and 414
lowest values on Saturday and Sunday. 415
17

Figure 8: Mean diurnal 416
cycle for NOx for all non-417
smoky days. 418
419
420
421
422
423
424
425
426
Table 2 shows the data sorted by exceedance status. An exceedance day is one with an 427
MDA8>70 ppb. 428
429
Table 2. Mean values for all days and exceedance days by smoke/no-smoke status for Aug.-Sept. 430
2022. 431
No-smoke Smoke
ALL STUDY DAYS
Count (days) 51 10
Average MDA8 (ppb) 55.3 65.0
Average NOx (ppb) 15.0 20.5
Average ∑VOCs (ppb) 14.2 22.1
Average PM25 (µg/m3) 6.1 17.3
O3 EXCEEDANCE DAYS
Count ( days) 3 2
Average MDA8 (ppb) 76.0 75.5
Average NOx (ppb) 20 22.6
Average ∑VOCs (ppb) 23.1 26.8
Average daily PM25
(µg/m3) 8.5 17.8
432
Binning the data this way shows that both NOx and VOCs are, on average, enhanced on 433
exceedance days by 33 and 63%, respectively, for non-smoke days. 434
18

Comparison of O3 measurements by standard UV and scrubber-less UV 435
Because previous work has shown that some UV O3 instruments can exhibit significant 436
positive biases in smoke (e.g. Long et al 2021; Bernays et al 2022), one of the SAMOZA goals 437
was to compare the standard UV O3 measurements (made with a Teledyne T400 instrument) 438
with a scrubber-less UV that has previously been shown to have little to no bias in smoke (Long 439
et al 2021). Figure 8 shows a comparison of hourly averages measured with the two instruments 440
at UTC and Figures 9 and show the difference in the two observations as a function of PM2.5 441
(Figure 9) and CO (Figure 10). 442
443
Figure 8. Comparison of the 444
hourly data from the standard 445
UDAQ Teledyne T400 446
instrument with the SAMOZA 447
observations using a TwoB 448
Technologies model 211 449
scrubber-less instrument. 450
451
452
453
454
Figure 9. Difference between 455
the two O3 measurements as a 456
function of PM2.5 concentrations. 457
458
459
460
461
462
Figure 10. Difference between 463
the two O3 measurements as a 464
function of CO concentrations. 465
466
467
468
469
470
471
19

During SAMOZA, moderate smoke impacted the region on a number of days, with the highest 472
hourly PM2.5 value of 58 µg m-3 on Sept. 10th , 2022. As seen in Figures 7, 8 and 9, this level of smoke 473
causes no detectable bias in the standard UV measurements at UTC. Long et al (2021) reports the O3 bias 474
for several standard UV instruments in terms of ppb of bias per ppm of CO in smoke. Generally the bias 475
for instruments that include an internal drying system was found to be much smaller then instruments 476
without a drying system. For the “undried” instruments the observed bias was 16.5-24.0 ppb per ppm of 477
CO, compared to 1-3 ppb per ppm of CO for the systems with drying. Given the observations during 478
SAMOZA, and despite having only moderate smoke levels, we can conclude that a bias of this magnitude 479
would have been observable. So based on the SAMOZA results, we see no significant bias in smoke with 480
the standard UDAQ O3 measurements up to at least 1 ppm of CO. 481
482
Photochemical box modeling 483
Numerous studies have demonstrated that photochemical box models are useful tools for 484
investigating O3 formation in smoke plumes (e.g., Mason et al., 2006; Alvarado et al., 2015; 485
Müller et al., 2016; Coggon et al., 2019; Ninneman and Jaffe, 2021; Rickly et al., 2023). Critical 486
to the success of these studies was their ability to include observations of the key chemical 487
species in the model and use a chemical mechanism that sufficiently accounted for the complex 488
chemistry that occurs in smoke plumes. In this section, we apply a photochemical box model, 489
constrained to the observations, to examine the VOC and NOx sensitivity on four high O3 days in 490
2022. 491
The observed hourly NOx, speciated VOCs, pressure temperature and RH were used to 492
constrain the Framework for 0-D Atmospheric Modeling (F0AM) photochemical box model 493
(Wolfe et al., 2016). We used version 3.3.1 of the Master Chemical Mechanism (MCM v3.3.1; 494
http://mcm.york.ac.uk) to drive the chemistry in the model (e.g. Jenkin et al., 2015). For each 495
case study day, a three day simulation was completed to investigate photochemical O3 production 496
at UTC, where only the results from the third day were considered, to allow for model spin -up. 497
The model used an integration time step of 10 minutes. 498
F0AM was constrained to the measured VOCs listed in Table 3. Total methylfurans were 499
assumed to consist entirely of 2-methylfuran, consistent with Coggon et al. (2019). Since the 500
PTR-MS only measured total concentrations of butenes, C8–C10 aromatics, and monoterpenes, 501
the distribution among individual species was estimated using data collected at the SLC 502
Hawthorne site. In addition, F0AM was constrained to observed total NOx concentrations at 503
20

each time step, while the model chemistry determined the NO/NO2 ratio. For all simulations, O3 504
was unconstrained and initialized with the concentrations measured at 0:00 LST. 505
506
Table 3. List of VOCs used to constrain the model. The unitalicized parameters were measured at 507
UTC by the PTR-MS. Speciation for the italicized parameters were estimated based on data from 508
the Hawthorne sites, as described in Ninneman et al 2023. 509
510
VOC Class Parameters
Aldehydes Formaldehyde and acetaldehyde
Alkenes Total butenes a (1-butene, cis-2-butene, trans-2-butene)
Aromatics Benzene, toluene, total C8 aromatics a (ethylbenzene, m-xylene, o-xylene,
p-xylene, styrene), total C9 aromatics a (1,2,3-trimethylbenzene, 1,2,4-
trimethylbenzene, 1,3,5-trimethylbenzene, isopropylbenzene, m-
ethyltoluene, o-ethyltoluene, p-ethyltoluene, n-propylbenzene), and total
C10 aromatics a (m-diethylbenzene)
Alcohols Methanol and ethanol
Ketones Acetone, methyl vinyl ketone, and methyl ethyl ketone
Biogenic VOCs
(BVOCs)
Isoprene and total monoterpenes a (alpha-pinene, beta-pinene)
Furans Furan, total methylfurans (2-methylfuran), furfural, and methylfurfural
Other Acetonitrile
511
On each case study day, we used the daily 25th-percentile concentration for odd oxygen (Ox 512
= NO2 + O3) to prescribe fixed background O3 concentrations in the model. Since UTC is a high-513
NOx site, Ox was used instead of O3 to determine background O3 concentrations because Ox is 514
unaffected by NO titration. The resulting daily 25th -percentile Ox values were 49.0, 42.8, 49.8, 515
and 52.8 ppb on 4 August and 3, 11, and 12 September, respectively. These values are consistent 516
with the relatively high background O3 concentrations reported for many parts of the western 517
U.S. (Jaffe et al 2020). 518
NO2 photolysis rates (JNO2) were estimated using an equation developed by Trebs et al. 519
(2009): 520
521
JNO2 = (1 + α) × ((B1 × SR) + (B2 × SR2)) (1) 522
523
where JNO2 has units of s−1, α is the surface albedo, B1 and B2 are polynomial coefficients with 524
values of 1.47 × 10−5 W−1 m2 s−1 and −4.84 × 10−9 W−2 m4 s−1, respectively, and SR is the 525
measured solar radiation in W m−2. We assumed an α of 0.15, which is a typical value for an 526
urban area. Photolysis rates for all other parameters were computed as a function of the 527
21

overhead O3 column (300 DU), α (0.15), elevation of UTC (1286 m a.s.l.), and solar zenith angle 528
using model-provided lookup tables that are described elsewhere (Wolfe, 2020). Then, the 529
model-calculated photolysis rates were scaled in reference to the JNO2 values calculated from 530
equation 1. This was done to account for smoke impacts on photolysis for all species. 531
The heterogeneous uptake of hydroxyl radical (OH) and hydroperoxyl radical (HO2) onto 532
aerosols was included based on previous studies (Tang et al., 2014;Lindsay et al., 2022): 533
534
dX / dt = −0.25 × γ × c(X) × PM2.5 × SAspecific × Xg × 10−6 (2) 535
536
where dX / dt is the heterogeneous loss rate of OH or HO2 in ppb s−1, γ is the aerosol uptake 537
coefficient, c(X) is the average molecular speed of OH or HO2 in m s−1, PM2.5 has units of µg 538
m−3, SAspecific is the specific aerosol surface area in m2 g−1, and Xg is the concentration of OH or 539
HO2 in ppb that is in the gas phase. γ was assumed to be 0.2, following Jacob (2000) and Slade 540
and Knopf (2014). Molecular weight and temperature were used to determine c(X). SAspecific was 541
assumed to be 4 m2 g−1, following Lindsay et al. (2022). 542
A first-order dilution rate (Kdil) was used to account for mixing of background O3 into the 543
model domain. This was done by varying Kdil until the best fit between modeled and observed 544
afternoon O3 was achieved, consistent with previous work (McDuffie et al., 2016; Ninneman et 545
al., 2020; Ninneman and Jaffe, 2021; Rickly et al., 2023). The Kdil values that led to the best fit 546
between modeled and measured afternoon O3 on 4 August and 3, 11, and 12 September were 1.3 547
× 10−4, 1.5 × 10−4, 9.0 × 10−5, and 1.6 × 10−4 s−1, respectively. 548
Model calculated instantaneous O3 production rates (PO3) were used to investigate O3 549
formation at UTC on the case study days. PO3 was calculated using eq. 3: 550
551
PO3 =kHO2+NO [HO2 ][NO]+∑kRiO2+NOi [Ri O2 ][NO] (3) 552
553
where kHO2+NO and kRiO2+NO are the rate constants for the reactions of HO2 with NO and speciated 554
organic peroxy radicals (RiO2) with NO, respectively. Unlike the net O3 production, PO3 does not 555
consider other processes that are important in the O3 budget, including chemical loss, dry and 556
wet deposition, and advection. A series of model sensitivity tests were completed to examine the 557
impact of anthropogenic VOCs, NOx, and/or temperature on PO3 and O3 at UTC. Except for 558
22

isoprene and monoterpenes, all measured VOCs were considered to be anthropogenic. 559
Many/most of the VOCs are also enhanced during wildfire smoke. For the sensitivity tests, we 560
only varied the initial values for anthropogenic VOCs, NOx, and/or temperature. The other model 561
inputs were left unchanged. 562
Figure 11 and Table 4 show the key observations on the four case study days. MDA8 O3 563
concentrations on three out of the four days – 4 August, 3 September, and 11 September – 564
exceeded the O3 standard of 70 ppb. Even though there was smoke overhead on 3 September, the 565
24 h PM2.5 concentration was only 8.5 µg m−3. As a result, we conclude that smoke negligibly 566
impacted surface concentrations on 3 September. Hourly O3 and PM2.5 were uncorrelated on 11–567
12 September when morning O3 concentrations rapidly increased from 15 to 84 ppb and 13 to 66 568
ppb on 11 and 12 September, respectively. This indicates that the observed O3 was mainly due to 569
in-situ photochemical production, rather than transport from the smoke plume. Figure 11, shows 570
that the daytime median concentrations of ∑VOCs were similar on all four days, although there 571
were higher aldehydes on Sept. 3rd , compared to other dates. NOx concentrations for three of the 572
days were similar, but much higher on September 12th, a Monday. Conditions were favorable for 573
O3 production on the case study days, with daily maximum temperatures (Tmax) exceeding 30 °C, 574
although temperatures were warmer on August 3rd and September 4th, compared to the other 575
days. 576
577
Figure 11. Hour-578
averaged 579
observations of 580
(a) O3 (ppb), (b) 581
PM2.5 (µg m−3), (c) 582
NOx (ppb), and 583
(d) ∑VOCs (ppb) 584
on 4 August, 3 585
September, and 586
11 –12 September. 587
588
589
590
591
592
593
594
595
596
23

Table 4. Observed values of maximum daily 8 h average (MDA8) O3, 24 h average PM2.5, daytime 597
(6:00–17:00 LST) median NOx, daytime median ∑VOCs, and daily maximum temperature (Tmax) at 598
UTC on the selected case study days. 599
600
Date Classification MDA8 O3
(ppb)
24 h PM2.5
(µg m−3)
NOx
(ppb)
∑VOCs
(ppb)
Tmax
(°C)
4 August Non-smoky weekday 75 10.7 12.7 32.6 34.9
3 September Weekend day
(minimal smoke influence)
76 8.5 10.6 34.2 37.3
11 September Smoky weekend day 80 26.0 10.8 28.5 30.1
12 September Smoky weekday 68 21.5 17.3 34.5 32.5
601
602
Figure 12. 603
Modeled O3 604
production rates 605
(PO3) and O3 606
concentrations for 607
the four case 608
study days. 609
610
611
612
613
614
615
616
617
618
619
620
621
Daytime values of modeled PO3 and O3 for the base simulations are shown in Figure 12. 622
Across the four days, peak afternoon PO3 and O3 concentrations ranged from approximately 16 to 623
23 ppb h−1 and 82 to 95 ppb, respectively. Several items in Figure 12 stand out. First, PO3 and O3 624
concentrations increased more rapidly on the morning of 3 September compared to 4 August and 625
11 –12 September. This was likely due in part to higher morning concentrations of reactive VOCs 626
on 3 September (Figure 11), especially formaldehyde, which was very high on the morning of 627
Sept. 3rd. The rapid increase in formaldehyde preceded a rapid increase in O3, as shown in 628
Figure 13, which suggests that formaldehyde contributed to the high O3 levels seen on that day 629
and the photochemical modeling supports that conclusion (Figure 12). 630
24

631
Figure 13. Observed 632
concentrations of O3 633
and HCHO on 3 634
September. 635
636
637
638
639
640
641
642
643
644
On Sept. 3rd, from 0-10am LST winds at UTC and back-trajectories were from the easterly 645
to southerly directions (ca 90 o -160o) which is the general direction of downtown SLC. While 646
no individual source could be identified to explain the rapid rise in formaldehyde on that day, it 647
seems likely that this increase was due to a source within the urban region. Identifying and 648
controlling this source would likely lead to lower O3 concentrations in the region. We 649
recommend that future studies in SLC (eg. the 2024 SLC-Summer Ozone Study) could be used 650
to more accurately identify oxygenated VOCs, which are important O3 precursors. 651
We also used the photochemical model to examine possible emission reductions of NOx and 652
VOCs. Figure 14 shows the model calculated O3 production and O3 concentrations when 653
anthropogenic VOCs and/or NOx were reduced by various amounts for each of the 4 days. 654
655
656
25

Figure 14. Model-predicted 657
sensitivity to reductions in 658
anthropogenic VOCs and 659
NOx at UTC. Top row 660
shows PO3 and bottom row 661
shows O3 concentrations for 662
August 4th, September 3rd, 663
11th and 12th, 2022. For 664
these tests all VOCs were 665
considered anthropogenic 666
except isoprene and 667
monoterpenes. For the two 668
smoky days (September 11th 669
and 12th), smoke VOCs were 670
first reduced by setting these 671
equal to the mean for non-672
smoky weekends (Sept. 11th) 673
or non-smoky weekdays 674
(Sept. 12th). Then the 675
anthropogenic VOCs or 676
NOx were reduced by the 677
indicated %. 678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
26

The results presented in Figure 14 show that O3 is most sensitive to VOC reductions and less 694
sensitive to NOx reductions at UTC. Consequently, it is important to determine which VOCs had 695
the greatest impact on O3 production. To do this, we calculated the observed daytime VOC 696
reactivity for each VOC class (VOCRclass; s−1) using eq. 4: 697
698
VOCRclass = ∑i (kOH+VOCi × [VOCi]) (4), 699
700
where kOH+VOCi is the reaction rate coefficient with respect to OH in units of molecule−1 cm3 s−1 701
and [VOCi] is the concentration of the individual VOCs in molecules cm−3. Figure 15 702
demonstrates that aldehydes, alkenes, aromatics, and biogenic VOCs (BVOCs) accounted for 703
most of the observed daytime VOC reactivity on the case study days. We note that alkanes were 704
not measured at UTC as part of SAMOZA. However model sensitivity studies using alkane 705
concentrations measured at the Hawthorne site found very little impact on the VOC reactivity or 706
O3 production (details in Ninneman et al, manuscript in progress). So we conclude these four 707
VOC classes had the most important influence on O3 formation at UTC. Reducing 708
anthropogenic emissions of aldehydes, alkenes, and aromatics would likely lead to lower O3 709
concentrations for the SLC metropolitan area. 710
711
712
Figure 15. 713
Observed 714
daytime VOC 715
reactivity on (a) 716
4 August, (b) 3 717
September, (c) 718
11 September, 719
and (d) 12 720
September. 721
722
723
724
725
726
727
728
729
27

Figure 14 shows that removing smoke VOCs has a strong impact on O3 production. Table 1 730
shows that the sum of measured VOCs increased an average of 56% on smoky days, compared 731
to non-smoke days. However some VOCs increased much more, like formaldehyde, which 732
increased 91% on smoke days, compared to non-smoke days. Figure 16 shows the sensitivity of 733
O3 production to VOCs on the two smoky days September 11th and 12th. This figure shows that 734
the strong enhancement in O3 production on these two dates is largely driven by high levels of 735
formaldehyde and other oxygenated VOCs in smoke on those days. The lower O3 production 736
and O3 concentrations calculated by the model on Monday September 12th, compared to the 737
Sunday 11th, were likely associated with NOx suppression, due to the high concentrations of 738
NOx seen on that day (as seen in Figures 6 and 7). 739
740
741
Figure 16. Impact of VOCs on 742
model calculated PO3 (top row; 743
panels (a) and (b)) and O3 744
(bottom row; panels (c) and (d)) 745
on 11 September (first column) 746
and 12 September (second 747
column), two smoke influenced 748
days. Red line shows base 749
results where all VOCs and NOx 750
were constrained by observed 751
values. Black line shows model 752
results when all VOCs except 753
aldehydes were reduced to the 754
values seen on non-smoky 755
weekend (Sept 11th) or non-756
smoky weekdays (Sept 12th). 757
Blue line shows model results 758
when all VOCs, including 759
aldehydes, were reduced to the 760
values seen on non-smoky 761
weekend (Sept 11th) or non-762
smoky weekday (Sept 12th). 763
764
Because UTC is one of the highest-NOx sites in the SLC metropolitan area, we want to consider 765
how these results will apply to other parts of the SLC region. During August–September 2022, 766
the observed mean 24 h average NO2 concentration at UTC was 13.4 ppb, which was 56% 767
greater than the regional mean of 8.6 ppb measured by UDAQ at 9 sites in the SLC Core Based 768
Statistical Area (CBSA) . On the two case study days with little to no smoke influence – 4 769
28

August and 3 September – the NOx sensitivity tests indicated that NOx reductions of 75% or 770
greater are needed to noticeably reduce O3 concentrations at UTC (Figure 14). Based on the 771
observed mean 24 h average NO2 concentrations at UTC versus the entire region from August–772
September 2022, a 75% reduction in NOx at UTC corresponds to an approximately 60% 773
reduction in NOx regionwide. Further details and sensitivity tests using the photochemical box 774
model will be presented in a scientific publication that is currently in-progress (Ninneman et al: 775
Investigation of Ozone Formation Chemistry During the Salt Lake Regional Smoke, Ozone, an d 776
Aerosol Study (SAMOZA)). 777
778
a. Generalized Additive Model 779
GAMs are a type of machine learning/statistical model that uses observations to train a 780
dataset to predict a specific parameter. GAMS are particularly useful for air quality applications 781
as they can incorporate linear, non-linear and categorical variables to predict the O3 MDA8. 782
Typical predictors include meteorological variables, such as the daily maximum temperature or 783
geopotential height, day of week, back-trajectory distance and direction, surface chemical 784
measurements (e.g., NOx and VOCs) and satellite observations. Our group has used GAMs to 785
quantify the additional O3 associated with smoke in numerous urban areas, including SLC (Gong 786
et al., 2017; Jaffe, 2021). In addition, we have applied this approach in several successful 787
exceptional event demonstrations to quantify the influence of smoke on the O3 MDA8 (LDEQ, 788
2018; TCEQ, 2017). For SAMOZA, we improved the GAM results from our previous work and 789
extended the analysis through September 2022. 790
We used data from four sites in SLC. For two sites, Hawthorne, and Bountful Viewmont, 791
we used data for May-September 2006-2022. For Erda and Herriman data are only available for 792
2015-2022. For each site, the data were split into smoke and no-smoke days, using the same 793
criteria as given above (overhead HMS smoke plus PM2.5 > mean+1 SD of annual May-794
September mean for non-HMS days). For the GAM model evaluation, the no-smoke data were 795
further split into training (90%) and test data (10%), where the test data are rotated through the 796
whole dataset and the model evaluation is repeated ten times (10-fold cross validation). 797
Since the distribution of MDA8 O3 was close to a normal distribution, Gaussian 798
distributions and the identity link function were used in this study. Penalized cubic regression 799
splines with ten basis functions (i.e., k = 10) was used. For the analysis, we used the “gam” 800
29

function in the “R” software mgcv package with univariate smooths and/or bivariate smooths 801
(i.e., in interactions). Note that the number of “k” for “YEAR” smooth term was set to 6 for the 802
HW and BT sites, and 3 for the HR and ER sites. This approach, based on suggestions in 803
previous study (Walker et al., 2022), was employed to avoid overfitting with respect to the 804
temporal trends in the data. When increasing the k value, overfitting occurs at some sites 805
(particularly, HR and ER sites), resulting in high R2, but bias occurs for certain parts of the data. 806
We tested GAM models through a variety of combinations and then determined the final version 807
of the GAM model by considering main effects and interactive influencing factors as shown in 808
the Lee et al (to be submitted). In addition, we conducted 10-fold cross validation on the GAM 809
model to evaluate its performance. This involved partitioning the data into training data (90% of 810
the total) and test data (10% of total), which is repeated 10 times. Table 5 shows a list of the 811
predictors used and Table 6 shows the form for the final GAM model chosen. 812
813
Table 5. List of predictors used in the GAM modeling for SAMOZA 814
815
Source* NO. Parameter Unit Description
1
1 DOW - Day of week (factor, from Mon to Sun)
2 DOY - Day of year (from 1 to 365/366)
3 YEAR - Year (from 2006 to 2022)
2
4 Tmax ºF Daily maximum temperature at SLC airport
(40.77º N, 111.96º W)
5 RH % Daily average relative humidity at SLC airport
(40.77º N, 111.96º W)
6 DewP ºF Daily average dew point at SLC airport (40.77º
N, 111.96º W)
3
7 TPW kg m–2 Daily average total precipitable water in the
entire atmospheric column
8 T700 K Daily average air temperature at 700 hPa
9 MW10m m s–1 Daily average meridional wind at 10 m above
the ground
10 ZW10m m s–1 Daily average zonal wind at 10 m above the
ground
11 MW700 m s–1 Daily average meridional wind at 700 hPa
12 ZW700 m s–1 Daily average zonal wind at 700 hPa
30

4 13 TM1000 ºF Morning temperature in lowest 1000 m
5
14 NO2VCD molec.
cm–2
OMI NO2 Vertical Column Density (VCD) in
the range of 40.25–41.25º N and 111.50–
112.25º W
15 CLFR - OMI Cloud fraction in the range of 40.25–
41.25º N and 111.50–112.25º W
6
16 TrajDist km
Endpoint distance (point to point) after 12 hours
of transport for a back trajectory initialized at
1pm local time
17 TrajDir deg
Endpoint direction (point to point) after 12
hours of transport for a back trajectory
initialized at 1pm local time
816
817
818
Table 6. The final version of the GAM model in this study 819
𝑀𝐶𝐴8 𝑂3 =
𝛽0 +𝐶𝑂𝑊+𝑟(𝐶𝑒𝑤𝑂,𝑅𝐻)+𝑟𝑒(𝑇𝑟𝑎𝑖𝐶𝑖𝑟𝑟,𝑇𝑟𝑎𝑖𝐶𝑖𝑟)+∑𝑟(𝑤𝑗)𝑚
𝑗=1 +∑𝑟𝑖(𝑊𝑗)𝑚
𝑗=1 +𝑒 (6)
𝑤𝑗: YEAR, DOY, Tmax, TPW, T700, MW10m, ZW10m, MW700, ZW700, TM1000,
NO2VCD, and CLFR
𝑊𝑗: (YEAR, Tmax), (DOY, Tmax), (DOY, RH), (DOY, DewP), (Tmax, MW10m), and
(Tmax, MW700)
s(): The function of modeling the main effects only, or the main effects and interactions
between covariates (the same basis function is used for each covariate)
te(): The function of modeling the main effects and interactions between covariates (different
basis function is used for each covariate)
ti(): The function of modeling interactions between covariates without including the main
effects
820
Figure 17 shows the results from the GAMs for the four sites. R2 for the four sites range 821
0.59-0.66. Figure 18 shows the residuals (observed – GAM predicted) for each site as a function 822
of the prediction value and smoke/no smoke condition and Tables 7 and 8 show statistics on the 823
residuals for the training, test and smoke datasets. As we have seen previously, the GAM results 824
are unbiased for the non-smoke days, but have a significant positive bias on smoke days, 825
associated with the greater than expected amount of O3 for the specific meteorological 826
conditions. 827
828
31

829
830
Figure 17. Observed MDA8 O3 vs. Predicted MDA8 O3 using GAM models. The solid line 831
represents 1:1 line, and the dashed line represents the NAAQS standard (i.e., 70 ppb). The small 832
open circle represents the daily points corresponding to observed and predicted MDA8 O3, and the 833
large open circle represents the overall mean of observed and predicted MDA8 O3 for both no-834
smoke and smoke days, while the error bars represent the standard deviation. (HW=Hawthorne, 835
BT=Bountiful, HR=Herriman and ER=Erda). 836
837
838
32

839
840
841
Figure 18. Residuals vs. Predicted 842
MDA8 O3 (i.e., the model fit) for no-843
smoke days (top) and smoke days 844
(bottom) at all sites. Boxes and 845
whiskers represent the 25th–75th 846
percentiles and 1.5 times interquartile 847
range (1.5IQR), respectively; squares 848
indicate means and horizontal lines 849
within boxes indicate medians; the 850
dashed red line indicates the average 851
97.5th percentile of residuals on no-852
smoke days for all sites (10.9 ppb). 853
854
855
33

Table 7. Summary of GAM results using 10-fold cross validation. 856
Dataset Site R2 Residuals (ppb)
Training data
HW 0.66 0.0 ± 5.8
BT 0.66 0.0 ± 5.8
HR 0.65 0.0 ± 5.0
ER 0.60 0.0 ± 5.3
Test data
HW 0.61 –0.01 ± 6.2
BT 0.60 –0.01 ± 6.2
HR 0.57 0.02 ± 5.5
ER 0.49 0.00 ± 5.9
857
Table 8. GAM residuals statistics for smoke and non-smoke days. 858
Site Smoke day
residuals (ppb)
Smoke day positive
residuals (ppb)
95th percentile of
residuals (no
smoke, ppb)
97.5th percentile of
residuals (no
smoke, ppb)
HW 4.4 ± 8.3 8.2 ± 5.9 9.5 11.5
BT 4.2 ± 8.1 7.6 ± 5.5 9.4 11.5
HR 7.1 ± 8.0 9.5 ± 6.5 8.1 10.3
ER 5.8 ± 8.2 8.8 ± 6.5 8.3 10.5
Avg. 5.1 ± 8.2 8.5 ± 6.1 8.8 ± 0.7 10.9 ± 0.7
859
Table 9. Percentage of days exceeding the EPA (2015) threshold (97.5th percentile of residuals (10.9 860
ppb) to support exceptional event cases for smoke and non -smoke days and contribution to the 861
MDA8 using the EPA 2015 methodology. 862
Site
Smoke days No-smoke days
N (%) MDA8 O3
(ppb)
MDA8 O3
contribution
(ppb)
N (%) MDA8 O3
(ppb)
MDA8 O3
contribution
(ppb)
HW 22 74.3 ± 7.9 4.6 ± 3.7 3 71.2 ± 9.1 2.7 ± 2.5
BT 19 74.9 ± 7.5 4.2 ± 3.5 3 70.9 ± 9.2 3.0 ± 2.7
HR 31 74.5 ± 7.8 5.4 ± 4.5 2 71.0 ± 7.9 2.5 ± 2.2
ER 22 69.9 ± 7.0 6.5 ± 4.6 2 70.6 ± 8.4 2.9 ± 2.0
Avg. - 73.6 ± 7.8 5.1 ± 4.2 - 71.0 ± 8.8 2.8 ± 2.4
1EPA, U. S., “Guidance on the Preparation of Exceptional Events Demonstrations for Wildfire 863
Events that May Influence Ozone Concentrations”, 2015. Retrieved from: 864
https://www.epa.gov/sites/production/files/2015-11/documents/o3_draft_wildfire_guidance.pdf. 865
866
867
868
869
870
34

The EPA suggests that states use a 97.5th percentile criteria of residuals to determine smoke 871
impacts on the MDA8 (U.S. EPA 2015). Table 9 shows that of the non-smoke days, only 2-3% 872
of these days meet this criteria, which is expected from the statistical distribution. In contrast, 873
we find that 19-31% of the smoke days, exceed this criteria and thus the GAM results could be 874
used to support exceptional event documentation for these days. Table 9 also shows that the 875
average O3 contribution for these days is 5.1 ppb, calculated using the guidance provided in U.S. 876
EPA (2015). Further details and sensitivity tests using the machine learning/Generalized 877
Additive Modeling will be presented in a scientific publication that is currently in -progress (Lee 878
et al: Evaluating the impact of wildfire smoke on ozone concentrations using a Generalized 879
Additive Model in Salt Lake City, Utah, USA, 2006–2022). 880
Comparisons of PTRMS and DNPH data 881
We compared two methods to identify carbonyls: proton transfer reaction mass spectrometry 882
(PTR-MS) and collection on 2,4-dinitrophenylhydrazine (DNPH)-coated cartridges followed by 883
analysis by high-performance liquid chromatography (HPLC). PTR-MS and DNPH-HPLC 884
methods quantified four compounds in common: formaldehyde, acetaldehyde, acetone, and 2-885
butanone. Correlation analysis of the two methods indicates a high positive correlation in 886
acetone, formaldehyde, acetaldehyde, and 2-butanone (r2 = 0.83, 0.72, 0.69, and 0.65, 887
respectively). However, the slopes of the correlations were 1.03, 0.34, 0.41, and 0.32, 888
respectively, indicating that the DNPH-HPLC method resulted in similar amounts of acetone—889
but less formaldehyde, acetaldehyde, and 2-butanone—than the PTR-MS method (Figures 19-890
22). 891
Non-target compounds can interfere with target compounds with the same mass in the PTR-MS 892
method, resulting in a high bias (this is a known problem for aldehydes; Vlasenko et al. (2010)). 893
For the DNPH-HPLC method, interference from atmospheric constituents can generate low and 894
high biases (Ho et al., 2013). We used potassium iodide cartridges to eliminate interference from 895
ozone, but other interferents are possible. 896
897
35

898
Figure 19. Comparison of acetone measurements from PTR-MS and DNPH-HPLC 899
methods. The 1:1 line is also shown. 900
901
Figure 20. Comparison of formaldehyde measurements from PTR-MS and DNPH-HPLC 902
methods. The 1:1 line is also shown. 903
36

904
Figure 21. Comparison of acetaldehyde measurements from PTR-MS and DNPH-HPLC 905
methods. The 1:1 line is also shown. 906
907
Figure 22. Comparison of 2-butanone measurements from PTR-MS and DNPH-HPLC 908
methods. The 1:1 line is also shown. 909
910
PTR-MS measurements of carbonyls were used in the modeling work described above. When 911
DNPH measurement-derived values for the four carbonyls shown in Figures 16-19 were used in 912
the model, 1-hr maximum modeled ozone decreased by 2, 14, and 9% for August 4, September 913
11, and September 12, respectively. PTR-MS and DNPH measurements were more different on 914
smoke days than non-smoke days (Table 1), so the change in carbonyls was greater on 915
37

September 11 and 12 than August 4. The greater difference between the two methods on smoke 916
days could be caused by interference in one or both measurement systems. The cause of these 917
differences is still under investigation by the SAMOZA team. 918
919
When all modeled carbonyls, not just the ones measured in common by PTR-MS and DNPH, 920
were set to match DNPH-derived values, 1-hr maximum modeled ozone changed by less than 921
1% relative to the change to just the four carbonyls in Figures 16-19, likely because mixing 922
ratios of those carbonyls were very low (0.5 ppb or less). 923
924
e. Positive Matrix Factorization 925
Positive matrix factorization (PMF) is an analysis technique that often affords clues about source 926
apportionment in an air shed. We performed a PMF analysis on the concentration data of the ten 927
compounds listed in Table 10. All available hourly average data for the dates 2022-08-01 to 928
2022-09-30 were included in the analysis. 929
930
The analysis is based on the following mathematical model of the data. Let 𝐶𝑠𝑐 represent the 931
concentration of compound c at time t. We assume there are a number of sources or factors, each 932
one possessing its own temporally-uniform concentration signature. Let 𝑊𝑠𝑐 represent the mole 933
fraction of compound c in source s. As a mole fraction, 𝑊𝑠𝑐 is unitless and its sum over 934
compounds is 1: ∑𝑊𝑠𝑐=1𝑐. We assume that 𝑇𝑠𝑠 concentration units of source s are present at 935
time t, so that 𝑇𝑠𝑠𝑊𝑠𝑐 represents the concentration of compound c derived from source s at time t. 936
Summing over all sources yields the modeled concentration of compound c at time t: 937
938
𝐶′𝑠𝑐=∑𝑇𝑠𝑠𝑊𝑠𝑐𝑠 (7) 939
940
The prime is used to distinguish modeled and measured concentrations. In our work, 𝑇𝑠𝑠, 𝐶𝑠𝑐, 941
and 𝐶′𝑠𝑐 are in ppb units. Note that this equation represents a matrix product. Therefore, the 942
procedure consists of determining two initially unknown matrices 𝑇𝑠𝑠 and 𝑊𝑠𝑐 that optimize the 943
fit between 𝐶𝑠𝑐 and 𝐶′𝑠𝑐. [Paatero & Tapper (1994)] Because negative values of 𝑇𝑠𝑠 and 𝑊𝑠𝑐 are 944
physically excluded, the optimization must be carried out subject to the constraints 𝑇𝑠𝑠≥0 and 945
𝑊𝑠𝑐≥0. Hence the expression “positive matrix factorization.” 946
38

947
We used the EPA Positive Matrix Factorization 5.0 tool [Norris et al. 2023] t o perform the 948
optimization. The number of sources is a variable in the calculation; we used the default value of 949
6. We performed about 20 independent runs with unique random seeds to verify that the analysis 950
converged consistently to the same solution. 951
952
Table 10. Compounds included in the PMF analysis. 953
acetonitrile isoprene
acetone acetaldehyde
benzene methanol
butenes methyl vinyl ketone + methacrolein (MVK + MACR)
formaldehyde toluene
954
Scatter plots for all pairs of compounds were examined. Some of these display interesting 955
structure related to diurnal trends in the concentrations. Figure 23 shows such a plot for the 956
concentrations of toluene and acetone. The data are separated into eight sets by month, August 957
or September, and by time of day, predawn, morning, afternoon, or evening. Least -squares lines 958
for each set are also shown, colored to indicate the time of day, while solid and dashed lines 959
correspond to the months of August and September, respectively. Afternoon data pile up near 960
the base of the plot, and the least-squares slopes of the predawn data are about nine times larger 961
than the slopes of the afternoon data. As explained fully below, the most logical explanation is 962
photochemical formation of some compounds under the afternoon sun. 963
964
39

965
Figure 23. Scatter plot of toluene vs. acetone concentrations. Eight different datasets, 966
corresponding to the two months of August and September, and to predawn, morning, afternoon 967
and evening time periods, are shown. Least-squares lines corresponding to each dataset are also 968
shown. Solid and dashed lines correspond, respectively, to August and September data. 969
970
The six sources obtained from the PMF calculation are summarized as pie charts in Figure 24. 971
Sources A, B, and E are dominated by formaldehyde, methanol, and acetone, respectively. 972
Acetaldehyde is a majority component of source C. Figure 25 confirms the quality of fit 973
between between 𝐶𝑠𝑐 and 𝐶′𝑠𝑐 for two selected days. Comparable fits are seen for all 61 days. 974
975
40

976
Figure 24. The PMF analysis identified six sources, labeled A through F, with the indicated 977
speciation profiles. 978
979
980
41

981
Figure 25. Comparison between measured (open circles) and modeled (crosses) concentrations for 982
two different days. The measured and modeled concentrations agree to within the resolution of this 983
plot. 984
985
986
42

Figure 26 displays the average concentration contributed by each of the six sources at each hour 987
of the day, normalized relative to each maximum. All six sources show a dip in the afternoon, 988
not surprising because we expect dilution resulting from mixing of the atmosphere. However, 989
sources C, D, and F dip to values around 10% or 20% of their maxima, while B, A, and E dip to 990
about 30%, 50%, and 80%, respectively, of theirs. If dilution caused by meteorology were the 991
sole explanation for the dips, we would expect all six dips to be of comparable size. Rather, it 992
appears that compounds dominated by sources A and E, and probably B, are created in processes 993
that are stronger in the afternoon and that partially compensate for the meteorological dilution. 994
Presumably, some compounds are formed photochemically under the afternoon sun. 995
996
997
Figure 26. Average concentration of each of the six sources at each hour of the day, normalized 998
relative to the maximum. 999
1000
1001
43

1002
Figure 27. Average concentrations of ten compounds at each hour of the day, normalized relative 1003
to their maximum. 1004
1005
1006
44

1007
Figure 27 displays diurnal variations in the concentrations of the ten compounds. Toluene and 1008
benzene are dominated by source D, and the butenes by sources D and F, indicating little or no 1009
secondary formation. Acetone and formaldehyde are dominated, respectively, by source E and 1010
A, implying secondary formation. Indeed, there is independent evidence for secondary 1011
formation of formaldehyde and acetone. That of formaldehyde is documented in the box model 1012
results given by Ninneman et al (in preparation), while Hu et al. [2013] estimate that about 50% 1013
of the North American acetone budget is contributed by secondary formation. The MVK + 1014
MACR signal has important contributions from sources A and E, but since these compounds are 1015
known to form biogenically, we may in fact be seeing a biogenic production that intensifies in 1016
the afternoon. Interestingly, one might expect to see secondary acetaldehyde formation in 1017
parallel with that of formaldehyde, but we see no strong evidence for it. 1018
1019
Discussion and implications 1020
The SAMOZA team measured a suite of VOCs, CO and O3 (using a novel “scrubber-less” 1021
method) at the UTC site in SLC during August-September 2022. Along with the standard 1022
UDAQ observations, the SAMOZA data have been used to support a variety of analyses. The 1023
main conclusions from SAMOZA are as follows: 1024
1025
1. We found no evidence for bias in smoke from the standard O3 measurements made by 1026
UDAQ using a Teledyne T400 instrument at PM2.5 concentrations up to 60 µg m-3; 1027
2. Formaldehyde (CH2O) and other aldehydes are key O3 precursors. We measured 1028
formaldehyde by two different methods and these showed generally good correlation, but the 1029
PTR-MS measurements are approximately 50% greater than the DNPH measurements on smoky 1030
days. The cause for this difference is not yet known. 1031
3. There appear to be primary sources of formaldehyde in the SLC urban region. Identifying 1032
and controlling these sources could lead to significant reductions in regional O3; 1033
4. On days with smoke, we found that PM2.5, CO, O3 and nearly all VOCs were significantly 1034
enhanced. While all VOCs contribute to the increase O3 production on smoke days, aldehydes 1035
are the strongest contributor. 1036
45

5. Photochemical modeling of O3 production rates at the Utah Tech Center demonstrates a 1037
strong sensitivity to VOC concentrations and less sensitivity to NOx. For non-smoke days, 1038
reductions in VOCs of ~30% would result in significantly reduced O3 production, potentially 1039
meeting the O3 standard. Reductions in NOx of ~60% are needed to get a significant reduction in 1040
O3 production for non-smoke days. VOCs with the greatest contribution to O3 production are 1041
oxygenated VOCs, along with alkenes. 1042
6. Generalized Additive Modeling (GAM) gave similar MDA8 O3 enhancements on smoky 1043
days as the photochemical modeling. Analysis of the GAM results show that 19-31% of the 1044
smoke days have model residuals that exceed the EPA (2015) criteria for statistical analysis of 1045
O3 data, and thus this method could be used as support for exceptional event cases for those 1046
days. 1047
1048
Complete SAMOZA publication list (all papers are currently in progress and expected to 1049
be submitted by the end of 2023: 1050
We note that several of the analyses results presented here are still in progress and thus 1051
should be considered preliminary. Further details and refined analyses will be presented in 1052
several scientific publications: 1053
1054
Cope E., et al., Sources of VOCs in SLC. 1055
Lee H., et al: Evaluating the impact of wildfire smoke on ozone concentrations using a 1056
Generalized Additive Model in Salt Lake City, Utah, USA, 2006–2022). In-review for 1057
the J.Air Waste Management Association. 1058
Jaffe D.A., et al: An Overview of the Salt Lake Smoke, Ozone and Aerosol Experiment 1059
(SAMOZA). In-review for the J.Air Waste Management Association. 1060
Ninneman M., et al: Investigation of Ozone Formation Chemistry During the Salt Lake Regional 1061
Smoke, Ozone, and Aerosol Study (SAMOZA) 1062
1063
1064
Data availability statement 1065
Final SAMOZA data has been archived at the University of Washington ResearchWorks archive: 1066
http://hdl.handle.net/1773/50049 1067
1068
Results and R codes for the GAM analysis have been supplied to the Utah Division of Air 1069
Quality 1070
1071
46

Acknowledgements 1072
The SAMOZA team consisted of faculty, students and post-doctoral fellows from the University 1073
of Washington, University of Montana and Utah State University. In addition to the P rincipal 1074
Investigators (Jaffe, Hu and Lyman) the team included: 1075
Matt Ninneman, Linh Nguyen and Haebum Lee (University of Washington) 1076
Colleen Jones, Trevor O’Neil and Marc Mansfield (Utah State University) 1077
Damien Ketcherside, Lixu Jin and Emily Cope (University of Montana) 1078
1079
We thank the entire team for their contributions to this project ! 1080
1081
SAMOZA could not have happened without funding from the Utah Division of Air Quality 1082
through a Science for Solutions grant to the three universities. We further acknowledge support 1083
from several industry partners including: 1084
Rio Tinto Kennecott Utah Copper, LLC. 1085
Tesoro Refining and Marketing Co. 1086
Holly Frontier Woods Cross Refining LLC. 1087
Big West Oil LLC. 1088
Chevron USA Inc. 1089
1090
The Utah Division of Air Quality reviewed the draft report and made suggestions for the 1091
final version. The industry partners had no input on the experimental design or 1092
interpretation of the data and results presented in this report. 1093
1094
References cited 1095
Alvarado, M.J.; Lonsdale, C.R.; Yokelson, R.J.; Akagi, S.K.; Coe, H.; Craven, J.S.; Fischer, E.V.; 1096
McMeeking, G.R.; Seinfeld, J.H.; Soni, T.; Taylor, J.W.; Weise, D.R.; Wold, C.E. Investigating 1097
the links between ozone and organic aerosol chemistry in a biomass burning plume from a 1098
prescribed fire in California chaparral. Atmos. Chem. Phys. 2015, 15, 6667–6688. 1099
Bell, M. L., McDermott, A., Zeger, S. L., Samet, J. M., & Dominici, F. Ozone and short -term mortality 1100
in 95 US urban communities, 1987-2000. JAMA, 292(19), 2372–2378. 1101
https://doi.org/10.1001/jama.292.19.2372, 2004. 1102
Bernays, N., Jaffe, D. A., Petropavlovskikh, I., and Effertz, P.: Comment on “Comparison of ozone 1103
measurement methods in biomass burning smoke: an evaluation under field and laboratory 1104
conditions” by Long et al., Atmos. Meas. Tech., 15, 3189–3192, https://doi.org/10.5194/amt-15-1105
3189-2022, 2022. 1106
47

Birks, J. W., Andersen, P. C., Williford, C. J., Turnipseed, A. A., Strunk, S. E., Ennis, C. A., & Mattson, 1107
E. (2018). Folded tubular photometer for atmospheric measurements of NO2 and NO. 1108
Atmospheric Measurement Techniques, 11, 2821– 2835. https://doi.org/10.5194/amt-11-2821-1109
2018 1110
Buysse, C.E., Kaulfus, A., Nair, U., and Jaffe D.A.: 2019. Relationships between particulate matter, 1111
ozone, and nitrogen oxides during urban smoke events in the western US. Environ Sci Technol 1112
53, 21, 12519-12528, doi: 10.1021/acs.est.9b05241. 1113
Coggon, M.M.; Lim, C.Y.; Koss, A.R.; Sekimoto, K.; Yuan, B.; Gilman, J.B.; Hagan, D.H.; Selimovic, 1114
V.; Zarzana, K.J.; Brown, S.S.; Roberts, J.M.; Müller, M.; Yokelson, R.; Wisthaler, A.; 1115
Krechmer, J.E.; Jimenez, J.L.; Cappa, C.; Kroll, J.H.; de Gouw, J.; Warneke, C. OH chemistry of 1116
non-methane organic gases (NMOGs) emitted from laboratory and ambient biomass burning 1117
smoke: evaluating the influence of furans and oxygenated aromatics on ozone and secondary 1118
NMOG formation. Atmos. Chem. Phys. 2019, 19, 14875–14899. 1119
Di, Q., Dai, L. Z., Wang, Y., Zanobetti, A., Choirat, C., Schwartz, J. D., & Dominici, F. Association of 1120
short-term exposure to air pollution with mortality in older adults. Journal of the American 1121
Medical Association, 318(24), 2446–2456. https://doi.org/10.1001/jama.2017.17923, 2017. 1122
Gong X., A. Kaulfus, U. Nair and D. A. Jaffe, Quantifying O3 Impacts in Urban Areas Due to Wildfires 1123
Using a Generalized Additive Model, Environmental Science and Technology, 2017, 51 (22), 1124
13216–13223. https://doi.org/10.1021/acs.est.7b03130. 1125
Ho, S. S. H., Ip, H. S. S., Ho, K. F., Dai, W.-T., Cao, J., and Ng, L. P. T.: concerns on the use of ozone 1126
scrubbers for gaseous carbonyl measurement by DNPH-coated silica gel cartridge, Aerosol Air 1127
Qual. Res., 2013, 13, 1151-1160. 1128
Hu, L., Millet, D.B., Kim, S.Y., Wells, K.C., Griffiths, T., Fischer, E., Helmig, D. Hueber, J., and Curtis, 1129
A., North American acetone sources determined from tall tower measurements and inverse 1130
modeling, Atmos. Chem. Phys. 2013, 13, 3379-3392. 1131
Jacob, D.J. Heterogeneous chemistry and tropospheric ozone. Atmos. Environ. 2000, 34, 2131–2159. 1132
Jaffe D.A., Fiore A.M. and Keating, T.J. Importance of Background O3 for Air Quality Management. EM. 1133
November 2020. 1134
Jaffe D. Evaluation of Ozone Patterns and Trends in 8 Major Metropolitan Areas in the U.S. Final project 1135
report for CRC Project A-124, Coordinating Research Council, Alpharetta, GA, March 2021. 1136
Available at: http://crcao.org/wp-content/uploads/2021/04/CRC-Project-A-124-Final-1137
Report_Mar2021.pdf 1138
Jenkin, M.E.; Young, J.C.; Rickard, A.R. The MCM v3.3.1 degradation scheme for isoprene. Atmos. 1139
Chem. Phys. 2015, 15, 11433–11459. 1140
Kaulfus, A.S., Nair, U., Jaffe, D.A., Christopher, S.A., and Goodrick, S.. Biomass burning smoke 1141
climatology of the United States: Implications for particulate matter air quality, Environmental 1142
Science & Technology 50, 11731-11741, doi: 10.1021/acs.est.7b03292, 2017. 1143
Langford, A. O., Alvarez, R. J., II, Brioude, J., Fine, R., Gustin, M. S., Lin, M. Y., et al. Entrainment of 1144
stratospheric air and Asian pollution by the convective boundary layer in the southwestern U.S. 1145
Journal of Geophysical Research: Atmospheres, 122, 1312–1337. 1146
https://doi.org/10.1002/2016JD025987, 2017. 1147
Lindsay, A.J.; Anderson, D.C.; Wernis, R.A.; Liang, Y.; Goldstein, A.H.; Herndon, S.C.; Roscioli, J.R.; 1148
Dyroff, C.; Fortner, E.C.; Croteau, P.L.; Majluf, F.; Krechmer, J.E.; Yacovitch, T.I.; Knighton, 1149
W.B.; Wood, E.C. Ground-based investigation of HOx and ozone chemistry in biomass burning 1150
plumes in rural Idaho. Atmos. Chem. Phys. 2022, 22, 4909–4928. 1151
Long, R. W., Whitehill, A., Habel, A., Urbanski, S., Halliday, H., Colón, M., Kaushik, S., and Landis, M. 1152
S., 2021. Comparison of ozone measurement methods in biomass burning smoke: an evaluation 1153
under field and laboratory conditions, Atmos. Meas. Tech., 14, 1783–1800, 1154
https://doi.org/10.5194/amt-14-1783-2021. 1155
Louisiana Department of Environmental Quality (LDEQ), 2018. Louisiana Exceptional Event of 1156
September 14, 2017: Analysis of Atmospheric Processes Associated with the Ozone Exceedance 1157
48

and Supporting Data. Available at https://www.epa.gov/sites/production/files/2018-1158
08/documents/ldeq_ee_demonstration_final_w_appendices.pdf. 1159
Lyman, S. N., Holmes, M., Tran, H., Tran, T., and O’Neil, T.: High ethylene and propylene in an area 1160
dominated by oil production, Atmosphere 2021, 12, 1. 1161
Mason, S.A.; Trentmann, J.; Winterrath, T.; Yokelson, R.J.; Christian, T.J.; Carlson, L.J.; Warner, T.R.; 1162
Wolfe, L.C.; Andreae, M.O. Intercomparison of Two Box Models of the Chemical Evolution in 1163
Biomass-Burning Smoke Plumes. J. Atmos. Chem. 2006, 55, 273–297. 1164
Mathur, R., Kang, D., Napelenok, S. L., Xing, J., Hogrefe, C., Sarwar, G., et al. (2022). How have 1165
divergent global emission trends influenced long-range transported ozone to North America? 1166
Journal of Geophysical Research: Atmospheres, 127, e2022JD036926. 1167
https://doi.org/10.1029/2022JD036926. 1168
McClure C.D. and Jaffe D.A. Investigation of High Ozone Events due to Wildfire Smoke in an Urban 1169
Area. Atmos. Envir. https://doi.org/10.1016/j.atmosenv.2018.09.021, 2018. 1170
McDuffie, E.E.; Edwards, P.M.; Gilman, J.B.; Lerner, B.M.; Dubé, W.P.; Trainer, M.; Wolfe, D.E.; 1171
Angevine, W.M.; deGouw, J.; Williams, E.J.; Tevlin, A.G.; Murphy, J.G.; Fischer, E.V.; 1172
McKeen, S.; Ryerson, T.B.; Peischl, J.; Holloway, J.S.; Aikin, K.; Langford, A.O.; Senff, C.J.; 1173
Alvarez II, R.J.; Hall, S.R.; Ullmann, K.; Lantz, K.O.; Brown, S.S. Influence of oil and gas 1174
emissions on summertime ozone in the Colorado Northern Front Range. J. Geophys. Res.: Atmos. 1175
2016, 121, 8712–8729. 1176
Müller, M.; Anderson, B.E.; Beyersdorf, A.J.; Crawford, J.H.; Diskin, G.S.; Eichler, P.; Fried, A.; 1177
Keutsch, F.N.; Mikoviny, T.; Thornhill, K.L.; Walega, J.G.; Weinheimer, A.J.; Yang, M.; 1178
Yokelson, R.J.; Wisthaler, A. In situ measurements and modeling of reactive trace gases in a 1179
small biomass burning plume. Atmos. Chem. Phys. 2016, 16, 3813–3824. 1180
Ninneman, M.; Jaffe, D.A. The impact of wildfire smoke on ozone production in an urban area: Insights 1181
from field observations and photochemical box modeling. Atmos. Environ. 2021, 267, 118764. 1182
Ninneman, M.; Lu, S.; Zhou, X.; Schwab, J. On the Importance of Surface-Enhanced Renoxification as 1183
an Oxides of Nitrogen Source in Rural and Urban New York State. ACS Earth Space Chem. 1184
2020, 4, 1985–1992. 1185
Norris, G., Duvall, R., Brown S., Bai, S., EPA Positive Matrix Factorization (PMF) 5.0 Fundamentals and 1186
User Guide. 2023, https://www.epa.gov/sites/default/files/2015-1187
02/documents/pmf_5.0_user_guide.pdf. 1188
Paatero, P., Tapper, U., “Positive matrix factorization: A non-negative factor model with optimal 1189
utilization of error estimates of data values, Environmetrics 1994, 5, 111-126. 1190
Permar, W., L. Jin, Q. Peng, K. O’Dell, E. Lill, V. Selimovic, R. J. Yokelson , R. S. Hornbrook, A. J. 1191
Hills, E. C. Apel, I-T. Ku, Y. Zhou, B. C. Sive, A. P. Sullivan, J. L. Collett Jr., B. B. Palm, J. A. 1192
Thornton, F. Flocke, E. V. Fischer, L. Hu, Atmospheric OH reactivity in the western United 1193
States determined from comprehensive gas-phase measurements during WE-CAN, Environ. sci.: 1194
Atmos., 2023, https://doi.org/10.1039/D2EA00063F. 1195
Rickly, P.S.; Coggon, M.M.; Aikin, K.C.; Alvarez II, R.J.; Baidar, S.; Gilman, J.B.; Gkatzelis, G.I.; 1196
Harkins, C.; He, J.; Lamplugh, A.; Langford, A.O.; McDonald, B.C.; Peischl, J.; Robinson, M.A.; 1197
Rollins, A.W.; Schwantes, R.H.; Senff, C.J.; Warneke, C.; Brown, S.S. Influence of Wildfire on 1198
Urban Ozone: An Observationally Constrained Box Modeling Study at a Site in the Colorado 1199
Front Range. Environ. Sci. Technol. 2023, 57, 1257–1267. 1200
Rolph, G. D.; Draxler, R. R.; Stein, A. F.; Taylor, A.; Ruminski, M. G.; Kondragunta, S.; Zeng, J.; 1201
Huang, H.-C.; Manikin, G.; Mcqueen, J. T.; et al. Description and Verification of the NOAA 1202
Smoke Forecasting System: The 2007 Fire Season. Weather Forecast., 24, 361−378, 2009. 1203
Sekimoto, K.; Li, S.-M.; Yuan, B.; Koss, A.; Coggon, M.; Warneke, C.; de Gouw, J. Calculation of the 1204
Sensitivity of Proton-Transfer-Reaction Mass Spectrometry (PTR-MS) for Organic Trace Gases 1205
Using Molecular Properties. International Journal of Mass Spectrometry 2017, 421, 71–94. 1206
https://doi.org/10.1016/j.ijms.2017.04.006. 1207
49

Slade, J.H.; Knopf, D.A. Multiphase OH oxidation kinetics of organic aerosol: The role of parti cle phase 1208
state and relative humidity. Geophys. Res. Lett. 2014, 41, 5297–5306. 1209
Tang, M.J.; Cox, R.A.; Kalberer, M. Compilation and evaluation of gas phase diffusion coefficients of 1210
reactive trace gases in the atmosphere: volume 1. Inorganic compounds. Atmos. Chem. Phys. 1211
2014, 14, 9233–9247. 1212
Texas Commission of the Environment (TCEQ), 2017. El Paso Ozone Exceptional Event: June 21, 2015, 1213
Addendum 4: Update, May 17, 2017. 1214
Trebs, I.; Bohn, B.; Ammann, C.; Rummel, U.; Blumthaler, M.; Kӧnigstedt, R.; Meixner, F.X.; Fan, S.; 1215
Andreae, M.O. Relationship between the NO2 photolysis frequency and the solar global 1216
irradiance. Atmos. Meas. Tech. 2009, 2, 725–739. 1217
U.S. EPA, “Guidance on the Preparation of Exceptional Events Demonstrations for Wildfire Events that 1218
May Influence Ozone Concentrations”, 2015. Retrieved from: 1219
https://www.epa.gov/sites/production/files/2015-11/documents/o3_draft_wildfire_guidance.pdf. 1220
Vlasenko, A., Macdonald, A., Sjostedt, S., and Abbatt, J.: Formaldehyde measurements by Proton 1221
transfer reaction–Mass Spectrometry (PTR-MS): correction for humidity effects, Atmos. Meas. 1222
Tech. 2010, 3, 1055-1062. 1223
Walker S.E., S. Solberg, P. Schneider and C. Guerreiro, The AirGAM 2022r1 air quality trend and 1224
prediction model, Geoscientific Model Development, 2023, 16, 573–595. 1225
https://doi.org/10.5194/gmd-16-573-2023. 1226
Wolfe, G.M., Marvin, M.R., Roberts, S.J., Travis, K.R., Liao, J., 2016. The Framework for 0 -D 1227
atmospheric modeling (F0AM) v3.1. Geosci. Model Dev 9, 3309–3319. 1228
https://doi.org/10.5194/gmd-9-3309-2016. 1229
1230
1231
1232