Supplementary Materials for
Ozone chemistry in western U.S. wildfire plumes
Lu Xu*, John D. Crounse, Krystal T. Vasquez, Hannah Allen, Paul O. Wennberg*,
Ilann Bourgeois, Steven S. Brown, Pedro Campuzano-Jost, Matthew M. Coggon,
James H. Crawford, Joshua P. DiGangi, Glenn S. Diskin, Alan Fried, Emily M. Gargulinski,
Jessica B. Gilman, Georgios I. Gkatzelis, Hongyu Guo, Johnathan W. Hair, Samuel R. Hall,
Hannah A. Halliday, Thomas F. Hanisco, Reem A. Hannun, Christopher D. Holmes, L. Gregory Huey,
Jose L. Jimenez, Aaron Lamplugh, Young Ro Lee, Jin Liao, Jakob Lindaas, J. Andrew Neuman,
John B. Nowak, Jeff Peischl, David A. Peterson, Felix Piel, Dirk Richter, Pamela S. Rickly,
Michael A. Robinson, Andrew W. Rollins, Thomas B. Ryerson, Kanako Sekimoto,
Vanessa Selimovic, Taylor Shingler, Amber J. Soja, Jason M. St. Clair, David J. Tanner,
Kirk Ullmann, Patrick R. Veres, James Walega, Carsten Warneke, Rebecca A. Washenfelder,
Petter Weibring, Armin Wisthaler, Glenn M. Wolfe, Caroline C. Womack, Robert J. Yokelson
*Corresponding author. Email: lu.xu@noaa.gov (L.X.); wennberg@caltech.edu (P.O.W.)
Published 8 December 2021,
Sci. Adv.
7
, eabl3648 (2021)
DOI: 10.1126/sciadv.abl3648
This PDF file includes:
Supplementary Text
Figs. S1 to S38
Tables S1 to S7
References
1 FIREX-AQ
The Fire Influence on Regional to Global Environments and Air Quality (FIREX-AQ) campaign
focused on characterizing fire emissions and investigating their environment impacts. During
FIREX-AQ, the NASA DC-8 aircraft conducted 22 research flights between July 24 and September
5, 2019. From July 24 to August 16, the aircraft was based in Boise, Idaho to study the wildfires
in the western U.S. From August 21 to September 5, the aircraft was based in Salina, Kansas to
characterize prescribed fires, including agricultural burning, forest understory burning, etc. This
study focuses on the O
x
formation in western wildfire plumes and utilizes measurements from
13 research flights that sampled 10 distinct fires, including repeated visits to the same fire (i.e.,
Williams Flats fire in Washington state) on different days. The related flight tracks are shown in
Figure S1. The aircraft typically took off between 15:00 and 17:00 (Mountain Daylight Time,
GMT - 6) and flight duration was around 7 h. The majority of the wildfire plumes sampled in
FIREX-AQ were not mixed with smoke from other fires. Thus, the smoke sampled could be
traced back to a definitive source.
The sampling strategy typically measured the smoke plume close to the wildfire to characterize
the primary emissions followed by multiple crosswind plume transects downwind of the fire.
We define the set of such crosswind transects through the smoke plume as a "fire ladder". In
some flights, the same wildfire was sampled with more than one fire ladder at a different time of
day. The aircraft true air speed was about 150 m s
-1
, which yields a horizontal resolution of 150
m for the in situ 1Hz measurements. The crosswind plume width varies significantly with fires
and downwind distance. Because of air traffic control, the minimal distance to fire is restricted
and ranges from 2 to 20 km. The furthest sampling distance varies with fires and it typically
ranges from 60 to 200 km.
Previous analyses on similar fire ladders typically analyzed the plume evolution as a function
Figure S1: The flight tracks of the NASA DC-8 for sampling wildfires in the western U.S.
during FIREX-AQ.
of smoke physical age, which is computed from downwind distance and wind speed, in a
pseudo-Lagrangian framework (
6, 7, 9
). The potentially changing fire conditions over time
requires that the aircraft follows the same airmass when transecting the plume at different
downwind distances in order to validate this analysis approach. However, this requirement
is often challenged by the aircraft navigation artifacts and complex plume dispersion. For
example, Figure S3 in Wiggins et al. (
31
) illustrates that the equivalent aircraft speed along
the plume length (i.e., downwind distance between successive transects/sampling time interval)
is considerably faster than the wind speed in FIREX-AQ. In other words, the smoke sampled
in different transects was emitted at different times, with samples further downwind being
emitted earlier in the day. Even if the plumes were sampled in a pseudo-Lagrangian fashion,
the complex plume dispersion could still cause artifacts, as a crosswind transect only takes a
snapshot of plume composition at a given altitude but not the plume 3D structure. For example,
while one transect crosses the dense portion of the plume, another transect may only skim
the top edge of the plume. Contrasting these two transects inevitably introduces uncertainty
in examining plume evolution. Analysis based on a single transect, which samples smoke
emitted under similar conditions, is less impacted by fluctuations in fire emissions over time.
The different extent of photochemical processing between the plume center and edges results
in different plume compositions across each crosswind transect. This phenomenon has been
observed in previous studies on both power plant and wildfire plumes (
14,17,32–35
). Here, we
apply single transect analysis to quantitatively investigate the O
x
chemistry in wildfire plumes.
The sampled fires were representative of those occurring in coniferous- and chaparral-
dominated ecosystems of the western U.S. More details about the sampled fuel types are documented
in FIREX-AQ data archive. According to previous elemental analysis of fuels characteristic of
the western US, the nitrogen content is in the range 0.23-1.28% (
36
).
The modified combustion efficiency (MCE) indicates the relative abundance of flaming and
smoldering combustion. MCE is calculated for each transect using Eqn.S1:
MCE =
1
m
CO/CO
2
+ 1
(S1)
where m
CO/CO
2
represents the slope of the York regression between the mixing ratio of CO and
CO
2
for each transect. The MCE in this study ranges from 0.85 to 0.95, with a median value of
0.91 (Figure S2). More than 90% of transects have MCE smaller than 0.92. It is noteworthy that
the temperate wildfires generally burn at lower efficiencies than the laboratory simulations (
36
).
Smoke age is estimated from airmass trajectory analysis. Upwind trajectories are initialized
at the aircraft location every 5 s and computed in HYSPLIT (
37
) with three sets of meteorological
data: HRRR, NAM CONUS nest, and GFS. Trajectories that are grossly inconsistent with
smoke transport direction observed by geostationary satellite imagers (GOES-16 and GOES-
17; GOES-R Algorithm Working Group, 2017) are excluded from further analysis. The smoke
age is determined from the point where the upwind trajectory makes its closest approach to the
Figure S2: The histogram of MCE for wildfire plume transects in this study.
fire source, plus time for the buoyant plume rise from the surface to the trajectory altitude at
7 m s
-1
. We average the age estimates from the three meteorological datasets (excluding any
that are inconsistent with observed transport) to form a single best estimate of the smoke age.
Uncertainty in each age estimate is computed from the spread between meteorological datasets,
errors in model wind speed, and additional uncertainties in the emission location within large-
area fires and plume rise speed. Median age uncertainties are 27%.
The fire radiative power (FRP) is an important indicator of fire strength. The detailed
procedure to obtain the diurnal cycle of FRP for the wildfires specific to the FIREX-AQ campaign
is discussed in Wiggins et al. (
31
) and briefly described below. FRP is obtained from both
GOES-16 and -17 Advanced Baseline Instruments (ABI). GOES data included are within 2
pixels (4km) of the final GeoMAC final fire perimeter. We chose a two-pixel distance to
optimize for 90% of the energy sensed by the ABI instrument, based on the Point Response
Function. We calculate the raw FRP by summing up FRP for each 5-min period. Then, we
modify the raw FRP to represent a more realistic diurnal fire cycle (denoted as pseudo-raw
FRP) in the following procedure. First, an adjusted FRP is generated by subtracting 5% from
each 5-minute total FRP period throughout the day. Second, the subtracted 5% is redistributed
equally across quiescent FRP times, which are defined as periods with no FRP observations for
a given 5-minute period. Third, the “quiescent mean” value, defined as 5% total daily FRP /
number of quiescent 5-minute periods, is added to any remaining 5-minute interval where the
adjusted FRP is less than the “quiescent mean”. Fourth, the transition between quiescent and
active FRP is smoothed. Finally, the resultant product provides a continuous function for the
allocation of emissions across the diurnal cycle, based on FRP variability.
2 Instrumentation
The NASA DC-8 aircraft hosted an unprecedented atmospheric chemistry payload and provided
comprehensive in situ characterization of fire emissions and their atmospheric evolution. The
instruments of interest to this study are summarized in Table S1. Species included in this
analysis are
∼
80 VOCs, CH
4
, C
2
H
6
, HCHO, CHOCHO, O
3
, NO, NO
2
, HNO
3
, peroxyacylnitrates
(PANs), particulate nitrate (pNO
3
), oxygenated aromatics, and hydroxyperoxides and hydroxynitrates
produced from the oxidation of ethene and propene. All the measurements are 1Hz, except
the discrete whole air samples (iWAS). As discussed in Section S7, we generate pseudo 1Hz
estimates of all VOCs measured by iWAS as a group by interpolating between the discrete
observations using the 1Hz C8-aromatics measurement of PTR-ToF-MS.
The extensive instrument payload on the NASA DC8 allows the characterization of several
critical species with different techniques. When duplicate measurements are available, the
measurement is selected based on an integrated consideration of measurement precision and
accuracy, data coverage, and instrument response time. The NO concentration is measured
by both laser-induced fluorescence (LIF) (
38
) and chemiluminescence (CL) (
33
) methods. Two
measurements agree within a 20% calibration uncertainty. The LIF NO is selected because it has
better precision (i.e., 1 ppt for LIF-NO vs. 6 ppt for CL-NO) and data coverage (nearly 100%
for LIF-NO vs.
∼
40% for CL-NO ). NO
2
concentration is measured by three instruments:
nonresonant laser-induced fluorescence (CANOE) (
39
), chemiluminescence (CL) (
33
), and
airborne cavity enhanced spectrometer (ACES) (
40
). Given that the three measurements agree
well (
∼
10%), for each flight, we select the measurement with the largest coverage of plume
data. HONO is measured by two instruments: I
–
CIMS (
41
) and (ACES) (
40
). In general, two
measurements show high correlation on flights when the HONO concentration is much larger
than the ACES measurement precision (600 ppt), but the correlation slope ranges from 0.9 to
1.8, depending on flight. The reason for the lack of agreement between two techniques is under
investigation. HONO measured by I
–
CIMS is selected because of its better precision (3 ppt
for I
–
CIMS vs. 600 ppt for ACES). Using ACES HONO does not alter any conclusions in this
study. Phenol is measured by two instruments: CF
3
O
–
CIMS and PTR. The inter-comparison
shows a good correlation, but a poor quantitative agreement. CF
3
O
–
CIMS measurement is
lower than PTR measurement by a factor of 3-4. This difference is also under investigation
and likely caused by absolute calibration of phenol standards. The main usage of phenol is to
calculate OH exposure and the systematic bias between the CF
3
O
–
CIMS and PTR measurements
does not affect the calculation, because the calculation relies on the loss fraction of phenol,
instead of its absolute concentration. The HCHO is measured by two instruments: the laser-
induced fluorescence instrument (ISAF) (
42
) and the compact atmospheric multi-species spectrometer
(CAMS) (
43
). ISAF HCHO, which is used in this analysis, is 27% lower than CAMS HCHO.
The source of the systematic bias has not been resolved at this time, but likely caused by the
absolute calibration of HCHO standards (
44
). Using CAMS HCHO increases the slope of York
fit in Figure 7 from 1.12 to 1.15. HCN is measured by several instruments, including PTR,
CF
3
O
–
CIMS, I
–
CIMS, and NCAR trace organic gas analyzer (TOGA) (
29
). The instrument
comparison shows complex results. In this study, we use HCN measured by CF
3
O
–
CIMS
and the HCN mixing ratios are determined from the relative signals arising from ambient HCN
(H
12
C
14
N) and continuously added standard addition of isotopically labeled H
13
C
15
N. In this
study, HCN is mainly used together with CO (i.e.,
Δ
HCN/
Δ
CO) to diagnose the variation of
fire emissions across individual transects, so that the uncertainty in the absolute concentration
of HCN does not affect this analysis.
Measurement
Technique
Accuracy
Reference
Note
O
3
gas phase chemiluminescence
2%
(
33
)
NO
y
gas phase chemiluminescence
10-15%
(
33
)
NO
2
nonresonant laser-induced fluorescence
10%
(
39
)
a
gas phase chemiluminescence
7%
(
33
)
airborne cavity enhanced spectrometer
4%
(
40
)
NO
laser-induced fluorescence
9%
(
38
)
phenol & HNO
3
& HCN
CIMS (CF
3
O
-
)
25%
(
45
)
OVOCs
CIMS (CF
3
O
-
)
25%
(
45
)
b
PANs
CIMS (I
-
)
20-30%
(
46
)
c
pNO
3
HR-ToF-AMS
34%
(
47–49
)
VOCs
whole air sampler with offline GC-MS
6-17%
(
50
)
d
VOCs
PTR-ToF-MS
25%
(
51,52
)
e
HCHO
laser-induced fluorescence
10%
(
42
)
CO
diode laser spectrometer
2-7%
(
53
)
CH
4
diode laser spectrometer
1%
(
53
)
C
2
H
6
absorption spectrometer
2%
(
43
)
CHOCHO
airborne cavity enhanced spectrometer
4%
(
40
)
HONO
CIMS (I
-
)
15%
(
41
)
Photolysis frequencies
actinic flux spectroradiometry
12-20%
(
54
)
H
2
O
diode laser hygrometer
5%
(
55
)
a
The instrument in this study was operated at wavelength 532 nm, which is different from that in the reference.
b
OVOCs measured by CF
3
O
–
CIMS include phenol, hydroxynitrates and hydroperoxide produced from the
OH-initiated oxidation of ethene and propene.
c
Four peroxy acyl nitrates (PANs) are measured by I– CIMS, which are peroxyacetyl nitrate (PAN),
peroxylpropionyl nitrate (PPN), peroxylbutyryl nitrate (PBN), and peroxyacryloyl nitrate (APAN).
d
The full list of VOCs measured by the NOAA integrated whole air sampler is shown in Table S3.
e
The full list of VOCs measured by PTR-ToF-MS is shown in Table S4.
Table S1: Instrumentation List
3 OH exposure
3.1 Calculation of OH exposure
The time-integrated exposure of the fire emissions to OH (i.e., OH exposure) is estimated using
the observed ratio between two VOCs (
56–58
). For a pair of VOC species to be selected in
the OH exposure calculation, they should satisfy the following criteria: (1) they arise from
the same source; (2) their dominant atmospheric fate is reaction with OH; and (3) the lifetime
of at least one VOC should be comparable to the time scale of interest. In addition, the single
transect analysis requires that the VOCs are measured at high temporal resolution (at least 1Hz).
Among
∼
80 VOCs quantified in this study, few satisfy the above criteria. For example, the
most widely used toluene/benzene ratio in the literature is not applicable here, because their
lifetimes (i.e.,
∼
10 days and
∼
1.8 days and for benzene and toluene, respectively) are much
longer than the smoke evolution time (<12 h) in this study. Furans have lifetimes on a similar
scale as smoke age, but they are sticky and subject to partitioning delay in sampling lines. After
a comprehensive search and evaluation, we utilize the phenol/benzene ratio to estimate OH
exposure, because both species are directly emitted from fire, dominantly react with OH, and
are reliably measured by CF
3
O
–
CIMS and PTR-ToF-MS, respectively, with 1Hz frequency.
Phenol has a lifetime of
∼
8 h under typical atmospheric OH concentration, well matched to
the maximum smoke transport time in this study. A minor fraction of phenol is consumed by
nitrate radical for these daylight conditions. It is estimated that even in the plume center of the
densest plume where the phenol+nitrate radical reaction is the most important, more than 90%
of phenol is consumed by OH (
59
).
The equation to calculate OH exposure is similar to previous studies, with slight modifications
to account for the secondary production of phenol from benzene oxidation. In the derivation
below, we start from simplified conditions and then generalize the expression to ambient conditions.
The major reactions of benzene and phenol in the fire plume are represented as follows:
B + OH
k
B+OH
−−−−→
Y
P
×
P
P + OH
k
P+OH
−−−−→
products
, where B and P represent benzene and phenol, respectively, Y
P
represents the phenol yield
from OH-initiated oxidation of benzene. Assuming no dilution, the change rates of benzene
and phenol concentrations are governed by
d[B]
dt
= –k
B+OH
[B][OH]
d[P]
dt
= Y
P
k
B+OH
[B][OH] – k
P+OH
[P][OH]
Further, the above equations are integrated to obtain
[B]
t
M
= [B]
t
E
e
–k
B+OH
∫
t
M
t
E
[OH]
·
dt
(S2)
[P]
t
M
= Y
P
[B]
t
E
k
B+OH
k
P+OH
– k
B+OH
(e
–k
B+OH
∫
t
M
t
E
[OH]
·
dt
–e
–k
P+OH
∫
t
M
t
E
[OH]
·
dt
)+[P]
t
E
e
–k
P+OH
∫
t
M
t
E
[OH]
·
dt
(S3)
where t
E
and t
M
represent the time of emission and of measurement, respectively. Equations
S2 and S3 are combined and rearranged to obtain
[P]
t
M
[B]
t
M
– Y
P
k
B+OH
k
P+OH
– k
B+OH
= (
[P]
t
E
[B]
t
E
– Y
P
k
B+OH
k
P+OH
– k
B+OH
)e
(k
B+OH
–k
P+OH
)
∫
t
M
t
E
[OH]
·
dt
(S4)
Rearrange Eqn.S4 to obtain an expression for OH exposure
∫
t
M
t
E
[OH]
·
dt =
ln(
[P]
t
E
[B]
t
E
– Y
P
k
B+OH
k
P+OH
–k
B+OH
) – ln(
[P]
t
M
[B]
t
M
– Y
P
k
B+OH
k
P+OH
–k
B+OH
)
k
P+OH
– k
B+OH
(S5)
We then substitute
[P]
t
E
[B]
t
E
with the excess ratio
Δ
[P]
t
E
Δ
[B]
t
E
to account for dilution and finally arrive at
Eqn.S6.
∫
t
M
t
E
[OH]
·
dt =
ln
(
Δ
[P]
t
E
Δ
[B]
t
E
– Y
P
k
B+OH
k
P+OH
–k
B+OH
)
– ln
(
Δ
[P]
t
M
Δ
[B]
t
M
– Y
P
k
B+OH
k
P+OH
–k
B+OH
)
k
P+OH
– k
B+OH
(S6)
In Eqn.S6, k
B+OH
and k
P+OH
are calculated at sampling temperature. Y
P
is 0.53, which
is obtained at room temperature (
60
), as the temperature-dependent value is unknown. The
impact of secondary production of phenol on estimated OH exposure is minor, because of the
slow consumption of benzene and the secondary production being much smaller than the large
amount of primary phenol emission. Nonetheless, we do account for the small amount of phenol
production in the OH exposure calculation. The initial [P]/[B] excess ratio (i.e.,
Δ
[P]
t
E
Δ
[B]
t
E
) is
represented by the 95th percentile of
Δ
[P]
Δ
[B]
from the transects close to fire. The
Δ
[P]
t
E
Δ
[B]
t
E
value
does not affect the O
x
chemical closure analysis (i.e., Eqn.3) because that analysis utilizes
the difference in OH exposure across each transect (denoted as
Δ
OH exposure) and any error
cancels in the calculation. For the 25 transects involved in the analysis, the
Δ
OH exposure
ranges from 0.73 to 2.51
×
10
10
molecule cm
–3
s. The
Δ
OH exposure is equivalent to 2 to 7 h
transport time, assuming a constant OH concentration of 1
×
10
6
molecule cm
–3
.
Besides chemical loss, another factor influencing ratio between two VOCs is atmospheric
mixing (
56,57
). When the influence of mixing on the ratio of two VOCs is taken into account,
the expression for OH exposure is shown in Eqn.S7, following the derivation in McKeen et
al. (
57
).
∫
t
2
t
1
[OH]
·
dt =
1
k
Y
– k
X
ln
[Y]
t
1
–
D[Y]
bkg
D+k
Y+OH
[OH]
[X]
t
1
–
D[X]
bkg
D+k
X+OH
[OH]
– ln
[Y]
t
2
–
D[Y]
bkg
D+k
Y+OH
[OH]
[X]
t
2
–
D[X]
bkg
D+k
X+OH
[OH]
(S7)
In Eqn.S7, D is a coefficient to describe the atmospheric mixing. Eqn.S7 is difficult to solve
analytically. To reduce the interference of atmospheric mixing on estimated OH exposure, we
estimate the OH exposure only for the measurements when the phenol and benzene concentrations
are at least 10 times higher than ambient background levels. Another approximation worth
noting is that the
D[X]
bkg
D+k
X+OH
[OH]
term is replaced with [X]
bkg
.
0
0.5
1
1.5
2
2.5
OH exposure (
10
10
molecule cm
-3
s)
-0.4
-0.2
0
0.2
0.4
ln (furfural / benzene)
Transect 108
1Hz data
measured slope = -0.26 (ODR fit)
theory decay slope = -0.34
Figure S3: Example plot to illustrate the evaluation of OH exposure estimated from the
phenol/benzene ratio against the observed furfural/benzene ratio.
The uncertainty of OH exposure derived from the phenol/benzene ratio is evaluated based
on the measured decay of other pairs of VOCs. One example of such evaluation is shown
in Figure S3. This figure displays the measured furfural/benzene ratio as a function of OH
exposure (inferred from phenol/benzene ratio) for one plume transect. The measured furfural is
assumed to be the 2-furfural isomer, because it is 21 times more abundant than 3-furfural based
on GC measurements in the 2016 FIREX FireLab study (
15
). On the one hand, the theoretical
decay rate of furfural/benzene ratio is calculated based on their reaction rate coefficients with
OH (i.e., (k
benzene+OH
– k
furfural+OH
)
×
10
10
), which is denoted as "theoretical decay rate" and
is -0.34 for this transect. On the other hand, the decay rate can also be calculated by fitting the
observed furfural/benzene ratio vs. OH exposure, which is denoted as "measured decay rate"
and is -0.26 for this transect. By comparing the decay rates from two methods, the measured
decay rate is slower than the theoretical one, suggesting the OH exposure is overestimated. We
apply this analysis to all transects for three pairs of VOCs: phenol/HCN, furfural/benzene, and
styrene/benzene. Figure S4 shows the histogram of the ratio of measured/theoretical decay rate
for each pair and a Gaussian distribution is fitted to the histogram. Regarding phenol/HCN,
the mode of the histogram (based on Gaussian fit) is 0.97, suggesting that the phenol/benzene-
derived OH exposure can reasonably reproduce the phenol/HCN decay. This agreement also
suggests the the secondary production of phenol is small, because replacing the slow VOC
benzene with HCN has small impacts on the estimated OH exposure. Regarding the furfural/benzene
ratio, the mode of the histogram is 0.79, suggesting the phenol/benzene-derived OH exposure
is overestimated by
∼
30%. Regarding styrene/benzene ratio, the mode of the histogram is 1.34,
suggesting measured decay rate is faster than theoretical decay rate by
∼
25%. This is expected
because styrene has significant additional loss via reaction with O
3
. Overall, we estimate that
the uncertainty in the calculated OH exposure based on the phenol/benzene ratio is
∼
30%.
Figure S4: The histogram of the ratio of measured to theoretical decay rate. The analysis
illustrated in Figure S3 is applied to all plume transects, but only transects with r
2
of the linear
fit greater than 0.2 are included in the histogram. The histogram is fitted with Gaussian curve.
3.2 HO
x
sources
Dividing the OH exposure by the smoke age provides an estimate of the average OH concentration
that the smoke experiences from emission to measurement. The OH concentration derived using
this method for the flight on 8/3/2019 is shown in Figure S5. At 1 h after smoke emission, the
average OH concentration on the plume edge is
∼
4
×
10
6
molecules cm
–3
, 40 times higher
than that in the plume center where attenuated actinic flux hinders the photolysis of HONO. The
difference between plume center and edges gradually diminishes as smoke ages. We caution
that the estimated OH concentration depends on the emission ratio of phenol to benzene, but
this uncertainty is unlikely to substantially alter the observed evolution of OH concentration.
0
5
10
Smoke Age (h)
10
5
10
6
[OH]
(molecules cm
-3
)
0
0.5
1
1.5
2
2.5
3
3.5
4
HONO/
CO (ppt/ppb)
Figure S5: The evolution of OH concentration as a function of smoke age for the second fire
ladder on 8/3/2019 flight. The smoke age is estimated based on backtrajectory analysis.
The instantaneous production rate of HO
x
(P
HOx
) is calculated using Eqn.S8, by following
Peng et al. (
14
). The photolysis rates and concentrations are obtained from in situ measurements,
and
φ
OH
represents the OH yield from O(
1
D)+H
2
O (Eqn.S9). The relative contribution of
HONO photolysis to total HO
x
production is calculated using Eqn.S10.
0
5
10
Smoke Age (h)
-0.5
0
0.5
1
1.5
2
2.5
OH exposure
(
10
10
molecules cm
-3
s)
0
0.5
1
1.5
2
2.5
3
3.5
4
HONO/
CO (ppt/ppb)
Figure S6: The OH exposure quickly increases to
∼
1
×
10
10
molecule cm
-3
s in 2 h. Beyond 2
h, coinciding with a depletion of HONO, it takes nearly 8 h to gain another
∼
1
×
10
10
molecule
cm
-3
s increment in OH exposure. The black line is provided as a visual aid.
The relative contributions of HONO photolysis and HCHO photolysis to P
HO
x
change over
time. Figure S7 displays the P
HO
x
enhancement in the plume relative to the clear sky as a
function of HO
x
production rate from HONO photolysis (i.e., P
HO
x
,HONO
) for the flight on
8/3/2019. When P
HO
x
,HONO
is near its maximum value of 4 ppt s
–1
, the P
HO
x
in the plume
is 300 times faster than that outside the plume and the HO
x
production rate from HCHO
photolysis (i.e., P
HO
x
,HCHO
) makes a minor contribution to P
HO
x
. As P
HO
x
,HONO
decreases,
the P
HO
x
enhancement decreases and P
HO
x
,HCHO
becomes more important. For example, when
P
HO
x
,HONO
is about 0.01 ppt s
–1
, higher P
HO
x
,HCHO
leads to a factor of 10 enhancement in
P
HO
x
.
P
HO
x
= j
HONO
[HONO] + 2
×
j
HCHO
[HCHO] + j
CH
3
CHO
[CH
3
CHO] +
φ
OH
j
O
1
D
[O
3
] (S8)
φ
OH
=
2k
O
1
D+H
2
O
[H
2
O]
k
O
1
D+H
2
O
[H
2
O] + k
O
1
D+O
2
[O
2
] + k
O
1
D+N
2
[N
2
]
(S9)
f
jHONO
=
j
HONO
[HONO]
P
HO
x
(S10)
10
-2
10
0
P
HO
x
,HONO
10
0
10
1
10
2
P
HO
x
plume/clear sky
10
-3
10
-2
10
-1
10
0
10
1
P
HO
x
,HCHO
Figure S7: The enhancement of production rate of HO
x
(P
HO
x
) as a function of HO
x
production
rate from HONO photolysis (i.e., P
HO
x
,HONO
). The color represents the HO
x
production rate
from HCHO photolysis (i.e., P
HO
x
,HCHO
). The measurements are from the second fire ladder
on 8/3/2019 flight. The P
HO
x
in clear sky is on the order of 10
-2
ppt s
-1
.
3.3 HONO evolution
The observed contribution of HONO photolysis to the total HO
x
production rate (denoted
f
jHONO
) varies greatly between fires. Figure S8 compares the maximum f
jHONO
(denoted as
f
jHONO,max
) from the transect that encountered the highest [CO] between different fires. The
highest [CO] was typically sampled in the transect closest to fire, which is typically 20 km
to fire source. The minimal distance to fire is restricted by air traffic control to maintain an
adequate distance between the DC8 and other aircraft supporting fire monitoring and fighting.
Despite similar downwind distance, the f
jHONO,max
varies from 0.2 to 1 between different
fires as shown in Figure S8, indicating that the relative contribution of HONO photolysis to
HO
x
production is reduced to a larger extent in some fires. Such fire-to-fire variability is not
driven by MCE as there is no correlation between f
jHONO,max
and MCE. Instead, the plume
optical properties likely play a more important role. As a proxy for plume optical extinction,
we use the reduction of HONO photolysis in the densest part of the plume (i.e., top 1% CO
concentration) relative to the background photolysis rate determined outside the plume, which
is calculated as 1 -
jHONO
plume
jHONO
background
. The f
jHONO,max
shows a positive dependence on jHONO
reduction (Figure S8), implying that the contribution of HONO photolysis to the total HO
x
production rate is larger in optically thicker plumes. This trend is likely caused by that in
optically thin plumes (i.e., small jHONO reduction), rapid HONO photolysis occurs in the
immediate vicinity of fires (i.e., lifetime
∼
15 minute in daytime background air), resulting in
large HONO depletion when such plumes are intercepted by aircraft at the minimal distance. As
a result, the observed f
jHONO,max
is small for optically thin plumes. This observation suggests
that the most active photochemistry in optically thin plumes may not have been captured by
the aircraft sampling. The implication is that the emission ratios of reactive trace species
measured in the field can be lower than that from laboratory studies which characterized the fire
emissions with minimal photochemical processing. This finding may also cause, to some extent,
the variability in OA evolution observed in the literature (
61
). For example, in optically thin
plumes, the active photochemistry producing SOA near the fire may not be captured by some
airborne measurements. Using the
Δ
[OA]/
Δ
[CO] measured in transects closest to fire as the
baseline could lead to observed net decreases in OA as smoke ages, because the photochemistry
is already weakened at this point and dilution-induced evaporation drives the OA concentration.
4 Transect selection
A plume transect is defined as a period between the aircraft entering and exiting the smoke
plume. As this study focuses on near-field concentrated plumes, the start and stop times for
0
0.2
0.4
0.6
0.8
1
jHONO reduction
0
0.2
0.4
0.6
0.8
1
f
jHONO,max
0
10
20
30
40
50
Downwind distance (km)
Figure S8: The relationship between f
jHONO,max
and jHONO reduction in plume. jHONO
reduction = 1 -
jHONO
plume
jHONO
background
. The jHONO
background
is obtained immediately before and
after plume interception. The jHONO
plume
is represented by the transect minimal. Each data
point represents the properties of one transect that encounters the largest CO enhancement in
sampling one fire.
plume transects are readily identified according to the enhancement of wildfire tracers, such
as CO and HCN. We identify 253 plume transects of wildfires in the western US during the
campaign. These transects are indexed consecutively from 1 to 253. Because of the complex
plume structure, plume transport, and variable fuels on the ground, the fire conditions and
emissions may not be stationary across some transects. Therefore, transects suitable for the
STA need to be scrutinized, which is carried out in the following procedure and outlined in
Figure S9.
The first step is to exclude some measurements within a plume transect, because of the
complex dynamics and geometry of smoke plume. For example, aircraft sampling at a constant
altitude, which is a common strategy, could miss the dense portion of the plume, when one side
of the plume is elevated higher than the other side, resulting from the spatially heterogeneous
Figure S9: Flowchart to select transects suitable for the O
x
chemical closure analysis. The
numbers in circles represent the number of transects. Green and organge circles represent
transects for which the expression is true and false, respectively. Note that the selection criteria
vary with the analysis applied to single transect as described in the text. RONO
2
stands for
hydroxynitrates.
updraft velocity beneath the plume. One case in point is shown in Figure S10a. The contour
displays the vertical profile of aerosol backscatter coefficient at 532nm (
β
532nm
) as a function
of time. The concentrated portion of the plume is elevated higher when the aircraft exists the
plume than when it enters the plume. For further illustration, Figure S10b and S10c compare
the vertical profile of
β
532nm
at two different times. At t1, the
β
532nm
measured right above
the aircraft is close to that right below the aircraft, indicating the aircraft is close to the most
concentrated portion of the plume. However, at t2, the
β
532nm
measured right above the aircraft
is substantially larger than that right below the aircraft, indicating the aircraft merely touches
the lower edge of the plume. The forward camera on the aircraft (Figure S10e) clearly shows
that the thick plume is above the aircraft. After t2, the aircraft encountered another air parcel,
which has up to 20 ppbv HCN but CO
2
/CO different from that before t2. In light of these
observations, we utilize the vertical profile of
β
532nm
to exclude the measurements when the
aircraft sampling deviates from the plume core. Specifically, the measurement at a certain time
is excluded when the
β
532nm
measured right above or right below the aircraft is less than 5% of
the max
β
532nm
in that vertical profile.
Next, we identify the plume transects that experience fluctuating fire emissions, which is
likely influenced by the spatial variation in burning conditions. This analysis is based on the
variations in
Δ
HCN/
Δ
CO across individual transects. This metric is chosen because both HCN
and CO are stable plume tracers and because the emission ratio of HCN to CO depends on MCE
and fuel types.
Δ
CO/
Δ
CO
2
, which is directly related to MCE, is not chosen because the high
and varying ambient CO
2
concentration complicates the analysis. For each plume transect, we
fit the
Δ
HCN/
Δ
CO distribution using Gaussian Mixture Model (GMM) with two modes. Three
example transects are shown in Figure S11. If the two modes are close (i.e., difference < 0.3)
and the standard deviation of
Δ
HCN/
Δ
CO is smaller than 10% of the mean of
Δ
HCN/
Δ
CO,
we assume fire emissions are relatively constant across such transect (Figure S11a). Otherwise,
if the two modes are close (i.e., difference < 0.3) but the standard deviation is larger than 10%
of the mean, we assume the transect experienced varying fire emissions and such transects are
excluded from STA. This criterion excludes 94 transects. If the two modes are far apart (i.e.,
difference > 0.3, Figure S11b and c), only the mode with larger fraction of data is kept.
After identifying the transects or measurements within a transect that have relatively stable
fire emissions, additional criteria are applied to select transects, depending on the analysis
performed on the STA. In the combined analysis based on the O
x
conceptual model and STA
(denoted as O
x
chemical closure analysis), we further scrutinize the transects according to the
following set of stringent criteria.
1.
We restrict analysis to transects that sample fresh smoke (< 12 h since emission). This