of 44
Supplementary
Materials
for
Temperature
-
dependent emissions dominate
aerosol and ozone formation in
Los Angeles
Eva Y. Pfannerstill
et al
.
Corresponding author
s
:
Eva Y. Pfannerstill, e
.pfannerstill@fz
-
juelich.de; Allen H. Goldstein, ahg@berkeley.edu
Sci
ence
3
8
4
,
13
24
(
202
4
)
DOI:
10.1126/sci
ence
.
adg8204
The
PDF
file
includes:
Materials and Methods
Figs. S1 to S16
Table S1 and S2
References
Other Supplementary Material
for this manuscript includes the following:
Data S1
2
Ma
terials and Methods
Aircraft
A two
-engine UV-
18A Twin Otter research aircraft was operated by the Naval Postgraduate
School out of the Burbank airport, CA. The aircraft is equipped with micrometeorological
sensors and is capable of flux measurements (
19
). The NPS Twin Otter payload during RECAP-
CA included total temperature measured by a rosemount probe, dew point temperature (chilled
mirror, EdgeTech Inc., USA), barometric, dynamic, and radome-angle pressures based on
barometric and differential transducers (Setra Inc., USA), total air speed, mean wind, slip- and
attack angles measured by a radome flow angle probe, GPS pitch, roll and heading (TANS
Vector platform attitude, Trimble Inc., USA), GPS latitude, longitude, altitude, ground speed and
track (NovAtel, Inc., USA), and latitude, longitude, altitude, ground speed and track, pitch, roll
and heading measured by C-
MIGITS
-III (GPS/INS, Systron, Inc., Canada).
Air was drawn from a 3
-inch isokinetic pipe inlet extending above the nose of the plane.
Ambient air gets diffused from a 2.047 inch ID orifice at the tip (area ratio of about 2) to another
diffuser with an area ratio of 5, resulting in a flow speed inside the tube of about 10% of the
aircraft speed (~ 60 m s
-1
). Vertical wind speed was measured by a five
-hole radome probe with
33º half-angles at the nose of the aircraft. The measured vertical wind speed is unaffected by the
aircraft movement and flow disto
rtion at the nose, as long as corrections based on “Lenschow
maneuvers” are applied (
56
). More detailed descriptions of this particular aircraft can be found
elsewhere (
57
).
The aircraft payload included: 1) Airborne Vocus PTR-
ToF
–MS for VOC fluxes; 2) Laser-
induced Fluorescence instrument for NO
x
fluxes; 3) Picarro cavity-ringdown spectrometer for
H
2
O, CO, CO
2
, and methane.
Flight routes and study regions
Routes were selected to ensure a good coverage of the South Coast Air Basin (Fig. S1),
while keeping individual flight legs as long and straight at a stable altitude of 300-400 m as
possible (low, stable altitude and long legs assure good quality of airborne flux measurements
(
19
)
). R
outes were also made sure to (i
) cover a wide range of temperatures, (ii)
include ships
anchoring in front of the port of Los Angeles, (iii) the region with a high density of warehouses
in the eastern San Bernardino Valley, and (iv
) Environmental Justice Communities around
refineries. Adjustments to the routes were made according to requests from air traffic control in
Los Angeles both during planning and spontaneously during flights. Otherwise, routes in Los
Angeles were repeated along the same tracks during each flight, while changing the flight
direction (either first going into the San Bernardino Valley or to the south). Each Los Angeles
flight included a 12-15 km long stacked racetrack pattern (
19
) flown at 4-
6 different altitudes
within the planetary boundary layer.
Climatology during the field campaign
For an overview of the temperatures and other meteorological parameters during each
flight, see Table S1. Flights were performed between June 1 and 22, 2021, with average
temperatures between 22 and 30°C in the Los Angeles Basin
. There was no precipitation. Flight
days and routes were chosen so that no cloud cover was encountered. The flights were performed
between 11:00 and 17:00 local time to ensure homogeneous turbulent conditions and a high
planetary boundary layer (PBL). Average boundary layer heights in Los Angeles were around
500-700 m above ground, but often much lower at the coast, which made it at times challenging
3
to stay within the PBL especially in that region, so that the flux data coverage is least dense at
the coast of Los Angeles.
Vocus PTR-
ToF
-MS measurements of VOCs
Sampling and instrument operation
Ambient air was sampled via a 90 cm long heated (40°C) 1/4’’ Teflon line through a Teflon filter
from the abovementioned isokinetic inlet (flow speed ca. 6 m/s for 5 m length) with a mass flow
controller at 1.5 L/min. The resulting lag time between the win
d sensor and VOC detection was
around 3 s.
The Vocus proton transfer reaction time of flight mass spectrometer (Vocus PTR-
ToF
-MS,
Aerodyne Inc., Billerica, MA, USA, (
58
)
) was operated at 60°C reactor temperature, 2.0
mbar
reactor pressure, and an E/N of ca. 130 Td. Mass resolution was 4805 ± 283 (average ± standard
deviation). The potential gradient along the focusing ion-molecule reactor (FIMR) was 590 V.
The gradient between skimmer 1 and skimmer 2 was changed once during the campaign from 6
to 9.1 V, which resulted in an improved sensitivity for some VOCs, but stronger fragmentation
for others, both of which effects were taken into account for by calibrating. The reagent water
flow was 20 sccm. With the resulting high water mixing ratio (10%v/v–20%v/v) in the FIMR,
the instrument showed no humidity dependence for sensitivity, which is an advantage in flux
measurements because it eliminates the necessity to correct for humidity differences between
different eddies caused by water fluxes. The high water concentration causes a high primary ion
(H
3
O+),
whose signal is reduced by a big segmented quadrupole (BSQ) that decreases the
transmission of low-mass ions in order not wear out the detector too quickly. However, we kept
the voltage of the BSQ relatively low at 200 V so that signals for low-
mass VOCs
like methanol
were not too strongly reduced.
Mass spectra were recorded at 10 Hz time resolution (or 2 Hz time resolution, for 2 flights out of
9: LA7, LA8) for a mass range of 10-
500 Da. Zero air measurements were conducted several
times during each flight for 1-5 min, during direction changes of th
e aircraft, because data
acquired during turns cannot be used for flux calculations. Ca. 2-4 times during each flight, the
zero air measurement was followed by a pulse of calibration gas of ca. 1-5 min length. These
calibration data were used to confirm that the instrument sensitivity after correction for the zero
air background did not change with the lower inlet pressure at flight altitude and that, thus, the
calibration factors acquired on the ground were applicable to the airborne data.
VOC data treatment and calibration
Raw PTR
-ToF data were processed with Tofware 3.2.3. No dead time correction was
applied in this step. 630 peaks were selected for peak fitting to derive ion counts per second. Ion
counts from in-
flight zero air measurements were interpolated and subtracted
from the ambient
data. The instrument zero at flight altitude was different from the zero on the ground due to
pressure effects that changed the pressure control valve position. Laboratory tests of pressure
effects confirmed that after subtraction of the flight zero, the sensitivities at the altitudes we flew
at were the same as on the ground.
Ground calibrations were conducted every 1-3 days (in total, 19 times) during the campaign
using one of three different gravimetrically prepared multicomponent VOC standards (Apel-
Riemer Environmental Inc., Colorado, USA). The following VOCs were included in the
calibration gas standards: methanol, acetonitrile, acetaldehyde, ethanol, acrolein, dimethyl
sulfide, isoprene, MACR+MVK, benzene, toluene, xylene, p-cresol, 1-,3-,5-trimethylbenzene,
4
D3 siloxane, D4 siloxane, D5 siloxane, D6 siloxane, propanol, butanol, acetone, furan, furfural,
benzaldehyde, 6-
MHO, monoterpenes (mixture of a
- and b
-pinene and limonene), nonanal,
acrylonitrile, methyl ethyl ketone, b-
caryophyllene. For most VOCs, the s
ensitivities were stable
within 25% over the campaign.
For all m/z without a corresponding gas standard, the sensitivities were derived from a root
function (the expected function of a ToF transmission) fitted to reaction rate normalized
sensitivities of non
-fragmenting and non-
clustering gas
-standard calibrat
ed VOCs. The estimated
uncertainty of the calibration for gas
-standard calibrated VOCs was 20%, while it was 54% for
all other VOCs.
VOC identification
Since the PTR
-ToF
-MS method provides exact masses that can be attributed to chemical
formulas, but often not with certainty to molecular structures, measured ions can be mixtures of
several isomers or originate from varying VOCs, depending on the dominant source type.
Gasoline vapor as well as oil and gas emissions include cycloalkanes that fragment on C
5
H
8
H
+
(m/z 69.0699), the ion that is usually attributed to isoprene in PTR
-MS. E.g. Gueneron
et al.
(
59
)
show that several cyclohexanes fragment on C
5
H
8
H
+
, especially at higher E/N, comparable to the
instrument conditions in our study. Pfannerstill
et al.
(
60
)
reported in measurements of oil and
gas emission
-dominated air around the Arabian Peninsula that isoprene measured by GC
-FID
was substantially lower than the C
5
H
8
H
+
signal in PTR
-ToF
-MS. They therefore attributed the
C
5
H
8
H
+
after isoprene subtraction to emissions from oil and gas extraction. Additionally, many
of the longer
-chain aldehydes, such as nonanal, fragment on C
5
H
8
H
+
, too (
61
)
. These aldehydes
can be substantial in urban areas. Both the long-
chain aldehydes and the cycloalkanes also
appear on C
8
H
15
+
(m/z 111.117) and/or C
9
H
17
+
(m/z 125.132).
In order to separate isoprene from interfering fragments of aldehydes and cycloalkanes, we used
the approach recommended by Coggon et al. (
62
)
: From the (isoprene
-free) nonanal calibration
gas standard, the ratio of m/z 69.0699 vs (m/z 111.117 + m/z 125.132) was derived. This ratio
was compared to that seen over oil and gas fields in the San Joaquin Valley, where the m/z
69.0699 is most likely dominated by cycloalkane fragments. Both ratios were the same at ~ 17
(for a gradient between skimmer 1 and skimmer
2 of 6 V) or ~ 45 (at a skimmer gradient of 9.1
V). This isoprene
-free ratio of m/z 69.0699/(m/z 111.117 + m/z 125.132) was used to correct the
isoprene signal:
Isopren
e
corr
=
m69
.0699
[(
푚푚
111.
117
+
푚푚
125
.132
)
∙푁푁푁푁푁푁푁푁푁푁푁푁푁푁
푠푠푁푁푁푁푠푠푠푠
]
For an accurate isoprene flux correction, this equation was applied to the fluxes of m/z 69.0699
and (m/z 111.117 + m/z 125.132) directly, not to the mixing ratios first. Since aldehydes are
long-
lived and represent a relatively high background in urban areas, the correction was rather
large (factor of 2) when applied to the mixing ratios. However, the correction is small in the flux
data (10%) since the fluxes of the aldehydes are small. We conclude that most of the ob
served
flux (i.e. covariance with vertical wind) on m/z 69.0699 was an actual isoprene flux on top of a
high constant background signal of aldehydes that did not covary with the vertical wind.
Similarly, acetaldehyde was corrected for ethanol fragments. Benzene was calibrated on m/z
78.046 to avoid influence of benzaldehyde fragments following Coggon et al. (
62
)
.
The monoterpenes measured at m/z 137.13 may include fragments of monoterpenoids and
monoterpene alcohols whose parents mass is m/z 155.14 (C
10
H
18
O, e.g. eucalyptol (
63
)
, linalool,
5
cineole, terpineol (
64
)). Fragmentation of sesquiterpenes on the monoterpene parent mass (m/z
137.13; (
65
)) is expected to be negligible, since the fraction of sesquiterpenes ending up on this
mass is on the order of 5% and sesquiterpene fluxes and concentrations are an order of
magnitude lower than those of monoterpenes. Consequently, the resulting interference on m/z
137.13 would be below 1%.
We looked for significant correlations within the dataset in an effort to minimize the
influence of unidentified fragments. Any m/z with a r²> 0.97 correlation with another was
examined for potential effects of water clustering or fragmentation. If it made chemical sense,
the respective m/z was identified as a fragment or water cluster and consequently added up with
its parent m/z. This concerned the following protonated m/z: 61.03 (C
2
H
5
O
2
+
) with fragment
43.02 (C
2
H
3
O
+
) and water clusters 79.04 (C
2
H
7
O
3
+
) an
d 97.05 (C
2
H
9
O
4
+)
, 87.04 (C
4
H
7
O
2
+
) with
water cluster 105.05 (C
4
H
9
O
3
+
), 89.02 (C
3
H
5
O
3
+
) with water cluster 107.03 (C
3
H
7
O
4
+
) and
fragment 71.01 (C
3
H
3
O
2
+), 85.02 (C
4
H
5
O
2
+
) with water cluster 103.04 (C4H7O3
+
), 115.07
(C
6
H
11
O
2
+
) with water cluster 133.08 (C
6
H
13
O
3
+
), 115.11 (C
7
H
15
O
+
) with water cluster 133.12
(C
7
H
17
O
2
+
), 143.11 (C
8
H
15
O
2
+
) with fragment water cluster 147.01 (C
7
H
15
O
3
+
), 181.00
(C
7
H
4
ClF
3
H+
) with fragments 179.99 (C
7
H
4
ClF
3
+
) and 160.99 (C
7
H
4
ClF
2
+
), 107.05 (C
7
H
7
O
+
)
with water cluster 125.06 (C
7
H
9
O
2
+
), 121.06 (C
8
H
9
O
+
) with water cluster 139.08 (C
8
H
11
O
2
+
),
140.03 (C
6
H
6
NO
3
+
) with water cluster 158.04 (C
6
H
8
NO
4
+
), 229.18 (C
13
H
25
O
3
+
) with fragments
173.11 (C
9
H
17
O
3
+
) and 191.13 (C
9
H
19
O
4
+
). All addition of fragments and water clusters was
conducted in molar flux units to avoid bias in the mass flux.
All VOCs that are individually depicted in Fig
ures, contribute significantly to the flux, or
cause significant discrepancies with inventories have been gas
-standard calibrated, and/or their
fragmentation is well understood (
61, 62
)
, so that remaining interferences after corrections can
be assumed to be minimal.
WRF
-Chem model simulation
We conducted model simulations over the study period using the Weather Research and
Forecasting model coupled with Chemistry (WRF-Chem v 4.2.2). The model conFig
uration is
described in Li et al.
(
66
)
. We first performed a WRF
-Chem simulation at 12 km horizontal
resolution over the Continental US to provide the initial and boundary condition, and then
performed
a 4 km horizontal resolution nest run over California.
We utilized the
RACM2_Berkeley2.0 chemical mechanism (
67–69
) with the following updates: We us
ed the
TUV photolysis code (phot opt =4) for the calculation of photolysis rates and coupled a newer
SOA VBS scheme (
70
) for better representation of SOA formation. Isopropanol, propylene
glycol, and glycerol are added as new species to represent the VOC chemistry from VCP
emissions as done in Coggon et al. (
10
).
The anthropogenic emissions are provided by the fuel
-based inventory for vehicle
emissions (FIVE), developed by McDonald et al. (
71
) and updated by Harkins et al. (
29
). The
FIVE inventory was further updated to include emissions from volatile chemical products to
create FIVE
-VCP (
10
)
. We
re-speciated the FIVE
-VCP inventory to the updated
RACM2_Berkeley2.0 mechanism. The biogenic emissions are provided by the Biogenic
Emission Inventory System (BEIS) v3.14. It is the default scheme to estimate volatile organic
compounds from vegetation and NO from soil developed by the United States Environmental
Protection Agency (EPA). We updated the BEIS emissions for isoprene and monoterpenes from
6
the urban land cover type based on Scott and Benjamin (
72
) as done in previous modeling work
over Los Angeles (
73
).
The WRF
-Chem outputs used in this study were J(O
1
D), H
2
O, and O
3
for the chemical
vertical divergence correction (see below).
Airborne Eddy Covariance
Flight segment selection
Flight segments for flux calculation were chosen according to the following criteria: gap-
free length of at least 10 km, stable aircraft roll and pitch (within 8°), stable altitude (within
± 50 m). This is to reduce errors (
19, 74
), which become large for short flight segments.
Planetary boundary layer (PBL) height was derived by eye from stark drops in dew point, water
concentration, toluene concentration and temperature during aircraft soundings. Soundings were
conducted at least at the beginning and end of each flight and before each stacked racetrack, but
sometimes much more frequently when the aircraft accidentally left the low PBL near the LA
coast. Datapoints outside the PBL were disregarded for flux calculation.
Continuous wavelet transformation
Lag times were determined for each VOC and each segment by calculating the
covariance and searching for the maximum covariance in a window of 4 s (covariance peak, see
Fig
. S2). Different VOCs have different levels of stickiness, which causes lag times to differ
between compounds
(
75
). Because of our use of a mass flow controller in front of the inlet
pump, pressure changes additionally influenced the lag time. Moreover, as Taipale et al. (
75
)
describe, lag times can vary because pumping speed varies over t
ime, so a variable lag time
assures that the flux is not underestimated. When there was no covariance peak above the noise,
a constant lag time (the lag time of isoprene) was applied for the respective VOC and segment.
The reason for this approach is that it is possible that there was a positive flux during half of the
segment and a negative flux during the other half of the segment, which can be resolved with
wavelet transformation.
Aircraft fluxes for each VOC were determined by continuous wavelet transformation (
76
), which
de-
convolutes the variance within a timeseries along both the frequency and time (distance)
domains. 10 Hz wind and VOC data were aligned using the lag times determined as described
above. Wavelet transformation of the data yielded the local wavele
t co
-spectra for each data
point along the flight track. Integration over all frequencies generates the flux timeseries. A
quality filter removed points containing > 80% spectral power within the cone of influence, the
region in which edge-
effects can lead to spectral artifacts. A moving average of 2 km was
applied to the 10-Hz fluxes to remove artificial emission and deposition that are effects of
turbulence, and sub-
samp
led to 200 m. All flux data points have associated systematic and
random uncertainties. For the three flights where data was recorded only in 2 Hz resolution,
disjunct Airborne Eddy Covariance was applied. Otherwise, these data were treated the same as
the 10 Hz data. A comparison between results of 10 Hz fluxes and the same data averaged to 2
Hz before doing the wavelet transformation showed a very minor high frequency loss, with an
overall reduced average flux of e.g. 0.5 % for isoprene and 0.4% for benzaldehyde. As the
cospectrum derived from of 10 Hz measurements (
Fig
. S2) shows, almost 100% of all flux is at
frequencies below 1 Hz (the Nyquist frequency which can be resolved by 2 Hz sampling, Fig
.
7
S2). This indicates that the eddies were sufficiently large at our flight altitude that no significant
information was lost by 2 Hz sampling. Previous aircraft campaigns operated at an even lower
0.7 s time resolution (
19, 77
)
without a need for correction for high-
frequency losses. The
Nyquist frequency for 10
Hz measurements is 5 Hz (
Fig
. S2).
We note that polar VOCs such as long-
chain OVOCs or siloxanes are prone to losses in the inlet
system, leading to a dampened covariance peak and thus, a potential underestimation of their
flux. However, the cumulative cospectra for most OVOCs compared well
with the modeled
complete cospectra, including sticky ones like cresol, ethanol or methanol. The stickiest among
the gas
-standard-
calibrated VOCs was nonanal, for which the cospectrum suggests around 50%
spectral loss.
Chemical vertical flux divergence correction
Reactive VOCs are partly lost between emission on the ground and observation at 300-
400 m due to reaction with OH and ozone. In order to correct for this loss, gradients of fluxes of
isoprene, trimethylbenzene, and dimethylfurane were used to derive approximate OH
concentrations. The resulting OH for each of these three VOCs and their isomers
covers a certain
range. In order to get OH concentrations for the whole flight track, and not just the racetrack
locations, we use the steady state box model described in the Supplement of Laughner and
Cohen 2019 (
78
)
. Input parameters include the measured NO
x
concentrations, VOC reactivity
(calculated from all measured VOCs, CO, and methane, multiplied by 1.3 to account for
unmeasured species), an organic nitrate branching yield of 0.056 based on the measured VOC
composition, and OH production rates calculated using simulated J(O
1
D), H
2
O and O
3
from
WRF
-Chem. The performance of the model was verified with data from the CalNex campaign,
where direct OH and total OH react
ivity measurements are available (
79
).
The VOCs were then corrected using the OH concentration following:
푑푑푑푑
푑푑푑푑
=
푘푘
푂푂푂푂 +푉푉푂푂 푉푉
[
푂푂푂푂
][
푉푉푂푂푉푉
]
퐹퐹
푠푠
=
푧푧∗
푑푑푑푑
푑푑푑푑
+
퐹퐹
푑푑
,
where F
s
is the flux at the surface and F
z
the flux at flight altitude, z is the flight altitude, k
OH+VOC
is the OH reaction rate of the respective VOC, and [OH] and [VOC] are the concentrations of
hydroxyl radicals and VOC, respectively. The ozone correction was done the same way. OH and
ozone reaction rates used are listed in Data S1. For m/z that could be attributed to several
isomers, we generally used the average reaction rate coefficient of all potential isomers following
Pfannerstill et al. (
60, 80
)
, and if there was no reaction rate coefficient available either from
IUPAC recommended values (
81
)
or the NIST database (
31
), we used the recommended values
from Isaacman
-VanWertz and Aumont (
82
)
for VOCs containing O, N, or O and N atoms.
The speciation of monoterpenes measured as C
10
H
16
H
+
was assumed to be comparable to the
afternoon Los Angeles monoterpene composition in van Rooy
et al.
(
83
)
. The resulting reaction
rate was verified and adjusted to represent (i) the ratios of inferred surface flux (after O
3
and OH
correction) to measured aircraft flux at altitude, with the gradient observed in stacked racetracks,
and (ii) the monoterpene oxidation product/monoterpene ratio with expected yields according to
the reaction rate used. Both meth
ods showed that the assumed combination of monoterpenes
with 44% a
-pinene, 8% camphene, 1% sabinene, 5% b-
pinene, 5% b-
myrcene, 1% 3-
carene,
14% limonene, 10% eucalyptol, 1% phellandrene, 10% ocimene causing an average OH reaction
rate coefficient of 8.52e-
11 cm³ molec
-1
s
-1
and an ozone reaction rate coefficient of
8
1.9e
-17 cm³
molec
-1
s
-1
are reasonable. Eucalyptol was included here despite being a C
10
H
18
O
monoterpenoid since ~90% of it is expected to fragment on C
10
H
16
H
+
(
63
)
. The monoterpenoid
parent masses measured as C
10
H
16
O and C
10
H
18
O contributed only 0.06% and 2% of the total
monoterpene flux, respectively, and therefore the choice of their reaction rate coefficients (here
based on the average of citral and camphor reaction rates for C
10
H
16
O, and on the average of
terpineol, linalool, and citronellal for C
10
H
18
O) did not significantly impact the result.
Generally, the magnitude of the chemical vertical divergence correction depends on the
oxidation rate applied. PTR
-ToF
-MS cannot separate isomers, so the oxidation rates attributed to
each m/z are based on best estimates. However, for most VOCs the chemical vertical divergence
correction was negligibly small (Data S1) since most of them (no matter which isomer) are
longer lived than the transport time between the surface and the point of observation. Therefore,
the only VOCs where the uncertainty of the che
mical composition caused significant uncertainty
in the final flux were the sesquiterpenes and monoterpenes. For a discussion of uncertainties, see
the section “Flux detection limit and uncertainty.” Even though there is uncertainty in the
monoterpene composition and in the resulting chemical vertical divergence correction, a
sensitivity analysis showed that the next step in the analysis – physical vertical divergence
correction – removes most of the bias caused by an over
- or underestimation of the monoter
pene
OH reaction rate coefficients, since it is based on the remaining vertical gradient in the
oxidation-
corrected data (see below). In the sensitivity analysis, a factor of 2 change in
monoterpene reaction rate coefficients led to only a 12% change in average monoterpene flux.
Physical vertical flux divergence correction
Physical vertical flux divergence is caused by horizontal advection and entrainment. In
Los Angeles, strong marine winds and resulting horizontal advection are likely the main source
of physical vertical divergence (
84
). Entrainment causes the flux divergence to be different for
each chemical species. The vertical racetrack data did not show conclusive vertical gradients in
non-
reactive VOCs. We attribute this to impacts of local emissions and the larger uncertainty of
the fluxes on the short (10 km) racetr
ack legs. Therefore, we used data from complete flights to
determine the vertical flux divergence. Since the boundary layer heights varied strongly across
the study domain and from day to day, we covered a wide range of z/zi (flight altitude
normalized by boundary layer height) from 0.2-
1. Due to the generally low, marine
-influenced
boundary layer in Los Angeles, 95% of the data is between z/zi = 0.5 and z/zi = 1 here.
Vertical flux divergence was determined for each VOC and each of the four regions of Los
Angeles separately. At first, for quality reasons, any data points with a random error > 30% or
with a horizontal wind speed > 8 m/s or with a z/zi > 0.8 were excluded. Data above z/zi 0.8 gets
so close to the entrainment layer that the correction is more uncertain here, and in addition, the
correction would become very large (and thus more uncertain) since it exponentially increase
s
when approaching the top of the boundary layer.
Fluxes for each VOC and region were binned into eight different z/zi bins, removing any bins
that contain less than 3% of the data and any data points that do not have a counterpart in the
other bins within 6 km distance (~ footprint size). The vertical divergence slope is determined
from a linear regression of the median flux of each altitude bin vs. z/zi (
Fig
. S4).
The slope is normalized by the intercept, i.e. C= slope/intercept. Then, the surface fluxes can be
calculated from the fluxes at altitude z as:
퐹퐹
0
=
퐹퐹
푑푑
1 +
푉푉⋅
푧푧
푧푧
푖푖
9
C is negative for VOCs that are emitted at the surface. However, C can be positive when VOCs
are deposited at the surface, or for OVOCs that are being formed (through oxidation) while the
air moves from the surface to the point of observation.
The aggregation of data from multiple time periods caused uncertainty in the determination of
the slopes, which is why we use the day-
to-day
-variation in the slopes as a basis for the Monte
Carlo error propagation to determine the uncertainty of the vertic
al divergence correction (see
section “Flux detection limit and uncertainty”).
Data points where the vertical divergence correction was larger than 3 times the median
correction factor of the respective VOC were replaced with “nan” in order to not introduce very
high uncertainties. The vertical divergence correction amounted to a factor of 2.1 ±
1.8 (average
± standard deviation).
Flux detection limit and uncertainty
The flux detection limit was calculated for each VOC and for each flight segment by first
creating a VOC white noise time series following Langford
et al.
(
85
), and then calculating
wavelet fluxes using this white noise time series and the measured wind. If the resulting random
flux was below the random covariance (i.e., covariance at ±220-
240 s lag time), the random
covariance of the respective segment was used
. The overall precision was propagated from the
2σ detection limit and the random uncertainty of
the flux calculation which was calculated
following Lenschow
et al.
(
19, 86
)
. The accuracy was propagated from the uncertainty of the
calibration, the systematic uncertainty of the flux calculation (
86
)
, and the uncertainties of
divergence corrections. The uncertainty of the chemical vertical divergence correction was
estimated to be 20% of the correction applied (a numerical calculation of this uncertainty is
difficult since the sequence of applying thi
s correction before the physical vertical divergence
correction means that any over
- or undercorrection of the oxidative loss should be approximately
eliminated by the physical vertical divergence correction). The uncertainty of the physical
vertical diver
gence correction was estimated using a Monte Carlo uncertainty propagation,
assuming a 17% uncertainty each for the slopes and boundary layer heights, since 17% was the
average day
-by-
day variability in the vertical divergence slopes of benzene. The result
ing
uncertainty of the vertical divergence correction was ~70% for the VOCs tested. The resulting
uncertainties depend on the VOC, ranging from 19% to 67% precision, and from 73% to 100%
accuracy. Total uncertainties range from 75%
-86% for gas
-standard cal
ibrated VOCs, and 90-
170% for the more than 400 VOCs that were calibrated using the theoretical approach. The
average total uncertainty for each species is listed in Data S1.
As a quality filter, any data points with a random flux error > 30% (indicating that the
corresponding legs were too short) or with a horizontal wind speed > 8 m/s were excluded. The
latter criterion aimed at accommodating the range of wind speeds from the vertical divergence
correction. Since only few data points had such high wind speeds, we assume that the physical
vertical divergence correction would be large
r for these data points, and we do not have enough
data to determine a suitable correction.
The largest source of uncertainty was the physical vertical flux divergence correction, which was
large in Los Angeles due to the low boundary layer height and strong horizontal advection by
marine winds (
84
)
. This uncertainty is likely systematic, and a single scale factor for all
measured species, so this uncertainty does not impact the fractional contribution of terpenoids to
the total fluxes that is the main focus of this paper. To determine the uncertaint
y of the fractional
10
contribution, we propagated the other uncertainties. For each m/z, the average uncertainty was
weighted by the average fractional contribution to the total OH reactivity or SOA formation
potential flux, respectively. This way, we obtained the uncertainties from calibration, flux
calculation, and chemical vertical divergence correction contributing to the total average OH
reactivity or SOA potential of the VOC flux (Table S2). Propagated, and including a 4%
contribution of unmeasured alkanes in Los Angeles following the observations of Hansen et al.
(
28
), this results in an OH reactivity flux terpenoid contribution of 56% ± 26%. For the SOA
formation potential, the propagated uncertainty includes 24% for unmeasured alkanes following
Gu et al.’s estimates (
18
), ending up with a terpenoid flux contribution to the SOA potential of
56% ± 33%.”
Footprint calculation and land cover
The footprint describes the contribution of surface regions to the observed airborne flux.
We used an updated version of the KL04+2D footprint (
87–89
) algorithm to derive 90%
footprint contours. The code is accessible online (
55
). This KL04+2D parameterization is
developed from a 1-D backward Lagrangian stochastic particle dispersion model (
90
)
. Metzger
et al. (
91
) implemented a Gaussian cross
-wind distribution function to resolve the dispersion.
perpendicular to the main wind direction. The input parameters include height of the
measurements, standard deviation of horizontal and vertical wind speed, horizontal wind
direction, boundary layer height, surface roughness length and the friction velocity. The
roughness lengths were derived from land cover using the relationship between Los Angeles land
use and roughness lengths described in Burian
et al.
(
92
). 2019 Landcover data were obtained
from the National Land Cover Database (
93
). High-resolution tree cover
data were obtained from
Alex Guenther (UC Irvine), and population density data was acquired from the US census
database
(
94
).
This footprint algorithm was compared with the half-dome footprints (
95
) applied for airborne
VOC fluxes by Misztal
et al.
(
77
), and with the Kljun
et al.
2015 (
96
) algorithm applied for
airborne fluxes by Hannun
et al.
(
97
). Matches with known point sources and VOC flux
increases observed (dairy farms, methanol) were used to check whether an algorithm’s result
explain the observed VOCs. From this comparison, the KL04+2D algorithm showed the best
match (
98
).
The obtained footprints were typically 3-6 km long and 2-
6 km wide at their widest point.
Inventory comparisons
Both inventories used for comparison with the airborne eddy covariance data are in a
4 km x 4 km resolution and were explicitly computed for the time of the RECAP-CA airborne
measurements in June 2021. The California Air Resources Board (CARB) 2021 inventory
includes anthropogenic and biogenic, point and mobile sources. In the CARB inventory, the
mobile sources are estimated from EMission FACtor (EMFAC) v1.0.2 and OFFROAD mobile
source emission models. The stationary sources are estimated based on the reported survey of
facilities within local jurisdiction and the emission factors from California Air Toxics Emission
Factor (CATEF) database. The biogenic emissions in CARB are from MEGAN 3.0 (
29
).
11
An alternative anthropogenic emission inventory is the FIVE
-VCP emission inventory
described in the section “WRF
-Chem simulations”. We obtained the hourly BEIS v3.14 biogenic
VOC emissions at 4 km resolution from the WRF
-Chem simulations (refer to the WRF
-Chem
section above). The FIVE
-VCP inventory (anthropogenic) and the BEIS inventory (biogenic)
were summed up to obtain the complete inventory that is used for WRF
-Chem by NOAA, in this
work named “BEIS+FIVE
-VCP”.
For comparison with the inventory, each footprint (corresponding to a measured flux) was
matched to the inventory grid cells that it overlapped with, if the overlap was > 10% of the area
of the grid cell and the sum of all overlaps amounted to at least 100%. The measured and
inventory data for each grid cell were matched in time. Only for the purpose of plotting maps, an
average of all flyovers was calculated for each grid cell.
A more extended analysis of the inventory comparison results is published elsewhere (
99
)
.
12
Fig
. S1.
(A) Flux footprints (10
th
, 50
th
, and 90
th
percentile of the origins of the measured fluxes). (B)
Flight routes over Los Angeles (without segments that were removed for flux data quality
reasons) and the four regions defined for the data analysis.
13
Fig
. S2.
Spectral quality control. (A) Covariance peak for D5 siloxane as an example for a VOC that is
sticky and has low emissions. The covariance peak was resolved nonetheless. (B) Cospectra of
toluene flux (w’VOC’) and heat flux (w’T’) for one example segment of
flight LA3. (C)
Cumulative cospectrum for toluene (w’VOC’) and heat (w’T’) fluxes for the same example
segment of flight LA3. The cospectral model was derived by optimizing transfer functions from
Lee
et al.
(
100
)
following Misztal
et al.
(
77
)
.
14
Fig
. S3.
Illustration of the steps from raw 10 Hz timeseries to the flux timeseries. (A) VOC (red), here:
monoterpenes (C
10
H
16
), and vertical wind (blue) time series, normalized. (B) Wavelet cross
spectrum. (C) Scale average time series.
15
Fig
. S4.
Vertical divergence for benzene for each region of Los Angeles. The dashed lines denote the
95% confidence interval of the linear regression, and the orange lines the linear fit (where y=
Flux, x = z/zi). z/zi is the flight altitude normalized by the boundary layer height.
16
Fig
. S5.
Maps of the fluxes of six VOCs that were important for SOA formation and OH reactivity
measured along the flight tracks. Data from all nine flights is shown here. Values are 2 km
running averages downsampled to 100 m.
17
Fig
. S6.
Contribution of different VOC classes at high and low temperature
to
(A, B)
mass flux, (
C,
D)
OH reactivity of VOC flux, and (
E, F)
secondary organic aerosol formation potential of VOC
surface flux of all VOCs measured during the RECAP
-CA flights in June 2021. The pie charts
show the median compositions in the high temperature bin (A, C, E) and low temperature bin (B,
D, F). For each 4x4 km² grid cell (see
Fig
. 1 in the main manuscript), “low temperature” was
defined as an ambient temperature (2 m above ground) in the lowest 25% of all flux
measurements conducted over that grid cell, and “high temperature” was defined as an ambient
tem
perature in the upper 25% of all flux measurements conducted over that grid cell. MeOH:
methanol, EtOH: ethanol, SQT: sesquiterpenes.
18
Fig
. S7.
Box charts of the temperature dependence of VOC fluxes separated by region in Los Angeles.
This shows that emission ranges may differ by region due to a different source distribution, but
emissions of these VOCs increase with temperature almost everywhere.
The coastal region is not
shown because of low coverage. The boxes show the interquartile range, the horizontal line the
median, and the whiskers the 1.5x interquartile range.
19
Fig
. S8.
Summed OH reactivity of VOC emissions (r² = 0.71) and SOA formation potential of VOC
emissions (r² = 0.72) as a function of temperature. The linear fit equations are shown in the
Fig
ures.