1
Supplementa
l
Information
Appendix
1
2
Rapid deposition
of oxidized
biogenic
compounds
to a temperate forest
3
4
Tran B.
Nguyen
,
1
ǂ
*
John D.
Crounse
,
1
ǂ
A
lex
P.
Teng
,
1
Jason
M. St. Clair
,
1
Fabien
Paulot
,
3
,4
Glenn
5
M. Wolfe
,
5
,
6
and
Paul
O. Wennberg
1,2
*
6
1.
Division of Geological and Planetary Sciences, California Institute of Technology,
7
Pasadena, California, USA
8
2.
Division of Engineering and Applied Science, California Institute of Technology,
9
Pasadena, California, USA
10
3.
Geophysical Fluid Dynamics Labora
tory, National Oceanic and Atmospheric
11
Administration, Princeton, New Jersey, USA
12
4.
Atmospheric and Oceanic Sciences, Princeton University, Princeton, New Jersey, USA
13
5.
Atmospheric Chemistry and Dynamics Lab, NASA Goddard Space Flight Center,
14
Greenbelt
, Maryland, USA
15
6.
Joint Center for Earth Systems Technology, University of Maryland
Baltimore County
,
16
Baltimore, Maryland, USA
17
18
19
ǂ
TBN and JDC contributed equally
20
*
Correspondence to
T.B. Nguyen (
tbn@caltech.edu
) and
P.O. Wennberg (wennberg@
21
caltech.edu)
22
23
24
2
1.
Site and Campaign
1
This research was conducted during the Southern Oxidant and Aerosol Study (SOAS)
2
ground campaign that occurred from
June
–
July 2013
in Alabama and Tennessee. SOAS
3
was part of the larger Southern Atmosphere Study (SAS) campaign, which
encompassed
4
many sampling sites and included multiple measurement platforms (tower, ground, and
5
aircraft). Both SOAS and SAS focused on understanding biosphere
-
atmosphere interactions
6
in the Southeastern United States. The SOAS site was located near Bren
t, Alabama at the
7
Centreville (“CTR”)
Southeastern Aerosol Research and Characterization Study (SEARCH)
8
location managed by the Electric Power Research Institute (
Latitude 32.90289
Longitude
-
9
87.24968
)
for the U.S. Environmental Protection Agency
, and is h
ereinafter referred to as
10
CTR.
During the campaign, CTR experienced typically humid (RH 50
–
80%) and warm (28
11
–
30 °C) conditions in the daytime. Winds were observed from all directions; however
, the
12
predominant winds were southerly during this experiment
. The CTR site was surrounded on
13
three sides (N,
W, and E) by a temperate forest
that is part of the Talladega National Forest
14
and on the
s
outhern side by a grassy
field
. The forest canopy was comprised of needle
-
leaf
15
coniferous (
shortleaf, longleaf, and l
oblolly
p
ine)
and broad
-
leaf deciduous (primarily
o
ak,
16
s
weetgum, and
h
ickory) tree species. The mean canopy height was approximately 10 m. A 20
17
m metal walk
-
up tower was erected
in the field
several meters from the edge of the forest.
18
The instrument used for this work was stationed
at the topmost platform of the
tower with the
19
inlet
facing north
.
The measurement height (z) was approximately 2
2
m, including the
sensor
20
height
s
. Section 2 describ
es the instrument
ation
employed in this work in more detail. The
21
mass spectrometers, pumps, and computers were housed in an insulated enclosure that
was
22
temperature
-
controlled with an HVAC unit to protect the components from precipitation and
23
large tempera
ture swings. The sonic anemometer was mounted on top of the instrument
24
enclosure, extending approximately 2 meters
north. The sonic was collocated laterally but
25
separated longitudinally from the inlet approximately 0.8 m (see Section 5, standard
26
correction
s).
27
28
2.
Measurements
29
Chemical ionization mass spectrometry (CIMS)
: Gas
-
phase compounds were measured
30
with negative
-
ion chemical ionization mass spectrometry (CIMS) using CF
3
O
-
as the reagent
31
3
ion, described in more detail previously
(1
-
3)
. The CF
3
O
-
ionization
is
sensitive
toward acids,
1
hydro
peroxides,
multifunctional
nitrates, and multifunctional compounds. Hereinafter the
2
measurement technique will be referred to simply as “CIMS” for brevity. Analytes
are
3
generally
ionized via two different
mechanisms:
4
AH +
CF
3
O
–
→ A
·
HF
–
+
CF
2
O
(1)
5
M +
CF
3
O
–
→ M·
CF
3
O
–
(2)
6
The fluoride transfer mechanism occurs for acidic analytes AH (e.g., nitric or formic
7
acids), resulting in an ion with
m/z
= MW + 19. The cluster formation mechanism occurs for
8
all other analytes M (e.g., H
2
O
2
, organic nitrates, hydroxy carbonyls, etc.), resulting in an ion
9
with
m/z
= MW + 85. Weakly acidic analytes will be ionized via both mechanisms, but their
10
fluoride tran
sfer ions are used for quantification due to higher sensitivity and fewer
11
interferences. CIMS calibration and analysis methods are presented in Section 3.
12
The mass analyzer was a compact time
-
of
-
flight (TOF, Tofwerk) spectrometer with
13
mass resolving powe
r of 800 m/Δm and a mass accuracy of < 100 ppm. The measurement
14
rate of 10 Hz was employed for this work. Ambient air was sampled through a 3.1 cm inner
15
diameter inlet, comprising of a 43 cm long glass section then a 17 cm glass section that both
16
coated wi
th a layer of Fluoropel hydrocarbon film (to minimize wall
interactions), at a flow
17
rate of 2000 std. L min
-
1
. From the center of the high inlet flow, ~ 180 std. mL min
-
1
was
18
subsampled for the analytical flow, using a moving aperture which was continuousl
y
19
adjusted to maintain 35.0 hPa pressure in a short Fluoropel
-
coated glass flow tube, as
20
described previously
(1)
.
The ambient air stream was diluted a factor of 9
–
11 with dry
21
nitrogen gas before entering the ion
-
molecule flow region, in order to moderat
e the effect of
22
water vapor on ionization sensitivity (Section 3). The mixing ratios reported have been
23
corrected for the dilution factor.
24
An ambient zero background (ambient air that has been scrubbed of reactive volatile
25
compounds by bicarbonate
-
impreg
nated nylon wool and Palladium
-
Alumina catalysts) and a
26
dry zero
background (dry nitrogen from liquid N
2
boil
-
off) were recorded every 30 minutes.
27
Ambient calibration (a total flow of 92 std. mL min
-
1
of dry nitrogen carrying calibration gas
28
joined with th
e ambient zero flow) and dry calibration (the same calibration gas stream
29
joined with the dry zero flow) were recorded every 2 hours. These in
-
field calibration
30
standards are derived from (a) permeation tubes of isotopically
-
labeled formic acid
31
4
(H
13
COOH),
acetic acid (
13
CH
3
13
COOH), and nitric acid (H
15
N
18
O
3
) kept at 50°C, (b)
1
hydrogen peroxide (H
2
O
2
) from a urea hydrogen peroxide ((NH
2
)
2
CO·H
2
O
2
) standard
kept at
2
0°C, and (c) diffusion vials of deuterium
-
labeled methyl hydroperoxide (CD
3
OOH) and
3
peroxyacetic
acid (CH
3
C(O)OOH) kept at 0°C. The calibrations at ambient and dry RH were
4
used to validate laboratory
-
derived water
-
dependent calibration factors, which were
5
performed with a larger water vapor range, for the select compounds.
6
7
S
onic Anemometer
:
A three
-
dimensional ultrasonic anemometer (Campbell Scientific,
8
model CSAT3, hereinafter “sonic”) was used to measure the wind speeds and the speed of
9
sound along three non
-
orthogonal sonic axes. The wind speeds were transformed by the
10
analyzer to the orthogonal
wind velocity components
u
,
v
, and
w
. The effects of wind
11
blowing normal to the sonic path were corrected online. The speed of sound in moist air was
12
converted to the sonic virtual temperature offline by applying a temperature
-
and humidity
-
13
dependent corre
ction function
(4)
. The data from the sonic was sampled at 8 Hz frequency.
14
Weather Station
:
Meteorological conditions were continuously monitored by a weather
15
station (Coastal Environmental Systems Inc, Zeno® 3200) mounted on top of the CIMS
16
enclosure at
the top of the tower. Air temperature (T, °C)
and relative humidity (RH, %) were
17
monitored by the
S1276Z
sensor with ±
0.4
°
C accuracy
for T and ±
3% for
RH. Barometric
18
pressure (P, mBar)
was monitored by the
S1080Z
sensor with ± 0.3 mBar accuracy. Solar
19
radiation (W m
-
2
) was monitored by the LI
-
COR
LI200SZ
pyranometer sensor with ± 10 %
20
accuracy. Wind speed (m s
-
1
) and wind direction (0
–
360°) were measured by the S1146Z
21
cup and vane sensor with accuracy of ± 5 m s
-
1
for wind speed and ± 5° for direction. Wind
22
direction of 180°
is interpreted as
northerly.
23
3.
CIMS
Calibration and
Data
Analysis
24
a.
Sensitivity and water
-
dependence calibrations
The sensitivit
ies
of CIMS to specific
25
analytes
are
controlled by physical
chara
cteristics
such as their dipole moments and
26
polarizability
(5)
. There are
varying degree
s
of water
-
dependence
in the ionization of
27
each compound
, which
becomes more
significant
for those that form weaker clusters with
28
the anion
. For compounds for which aut
hentic standards were commercially available or
29
can be synthesized, a water
-
dependent sensitivity calibration was performed in the
30
5
laboratory prior to ambient measurements. For other compounds, theoretical calculations
1
were performed
to estimate the sensit
ivity, as described in more detail elsewhere
(6)
. The
2
molecular identities, abbreviations used in this work, methods of calibration, and
3
estimated measurement error for each compound are reported in Table S1. Representative
4
traces of
CIMS compounds and ver
tical wind
w
are shown in
Figure S1
, illustrating
5
different propensities toward turbulent transfer for these compounds
.
The mixing ratio of
6
w
ater
is
anti
-
correlated
with H
2
O
2
and ISOPOOH+IEPOX
, visually demonstrating the
7
different directions of their net
flux. Th
e magnitude of the fluctuations, e.g.,
of H
2
O
2
(±
8
50%
)
and
to ISOPN (±
15
%
)
,
are indicative of the magnitude of their biosphere
-
9
atmosphere exchange velocities.
10
Authentic standards were used to characterize
the
dependence of the CIMS sensitivity to
11
water vapor by introducing a gas stream containing a known quantity of the calibrant
12
compound to the CIMS flow region while varying the amount of water vapor that is co
-
13
introduced. Variable water vapor content was achieved by mixing different ratios of dr
y
14
nitrogen with a gas stream containing ~ 3% water vapor. The water vapor fraction was
15
characterized by Fourier
-
transform infrared spectroscopy (FTIR) with a 19 cm pathlength
16
cell. Spectral fitting for water vapor was performed using the HITRAN spectral da
tabase
17
(7)
and a nonlinear fitting
soft
ware NLM4
(8)
. The quantifications of other calibrant
18
gases were performed as follows:
19
1)
HCN was calibrated with a standard gas mixture (6.3 ppmv in N
2
, Scott Specialty
20
Gasses) that was diluted with a known flow ra
te of dry nitrogen.
21
2)
H
2
O
2
was calibrated by flowing dry nitrogen continuously over urea hydrogen
22
peroxide (Aldrich, purity 97%), kept at 0°C. Absolute H
2
O
2
mixing ratio in this
23
stream was quantified by bubbling the equilibrated outflow into ultrapure w
ater (18
24
MΩ, Millipore) for a fixed time. The aqueous H
2
O
2
was quantified by high
-
25
performance liquid chromatography (HPLC) mass spectrometry, as well as a UV
-
26
Visible colorimetric technique
(3)
.
27
3)
Formic acid (Aldrich, 98%) and nitric acid (Aldrich, 70%
in water) were calibrated
28
by flowing dry nitrogen continuously over permeation tubes (Kintec), kept at 50°C.
29
6
The permeation rate was determined by gravimetric analysis for formic acid. Nitric
1
acid permeation was calculated by collecting the outflow in ultr
apure water and
2
analyzing the solution with ion
-
chromatography (IC).
3
4)
PAA (Aldrich, 40% in water) was calibrated by flowing dry nitrogen continuously
4
over custom diffusion vials, kept at 0°C, and collecting the outflow similarly to
the
5
protocol for H
2
O
2
.
The outflow solution was analyzed by HPLC
-
derivatization
6
fluorescence. Additionally, gas
-
phase PAA was analyzed with FTIR using the IR
7
cross section from Orlando et al
(9)
.
8
5)
HMHP was synthesized in the gas phase by the reaction of H
2
O
2
and formaldehyde
9
and characterized by FTIR as described by Fry et al
(10)
.
10
6).
Hydroxyacetone (Aldrich, 90%) was introduced into a
n evacuated
500 mL glass
11
bulb by
monitoring
pressure
increase
, and backfilled with dry N
2
to obtain several
12
ppmv. This mixture was quantified by FTIR using the cross section archived in the
13
Pacific Northwest IR Database
(11)
. The gas in the IR cell was
introduced
into
a
~
14
300 L
Teflon bag
and diluted with zero air.
15
7)
The sum of ISOPOOH and
IEPOX were observed at the product ion with m/z 203
16
as they are isobaric (C
5
H
10
O
3
). IEPOX was synthesized as reported in Bates et al
17
(12)
. A measured weight of a standard solution of IEPOX in water was atomized
18
into a 24 m
3
FEP Teflon bag alongside hydroxy
acetone and toluene as a volume
19
tracer (that was quantified by GC
-
FID), as reported recently
(13)
. The vapor wall
20
loss and solution
-
phase decomposition for this method were characterized to be
21
negligible. The combined sensitivity was then determined from a
photooxidation
22
experiment, with the assumptions of yields as
reported
in Paulot et al
(14)
.
23
8)
ISOPN was synthesized and calibrated as outlined by Lee et al
(15)
. The calibration
24
method relied on
the quantitative
thermal dissociation of the organic nit
rate to NO
2
25
followed by laser
-
induced fluorescence quantification of NO
2
(TDLIF
instrument
).
26
9)
PROPNN, and the sum of MACN + MVKN were generated as part of a high
-
NO
27
isoprene oxidation and separated with a GC column, as described by Lee et al
(15)
.
28
7
The
outflow of the GC column was directed toward CIMS and the TDLIF
1
instruments, and the calibration technique was similar to that of ISOPN.
2
b.
Measurements of water vapor
:
High
-
frequency water vapor mixing ratios for the
3
calculations of latent heat fluxes were
determined by CIMS, using the
double cluster ion
4
(H
2
O)
2
·CF
3
O
–
(
m/z
121).
This ion
remained linear with
respect to the range of
water
5
vapor mixing ratios experienced at SOAS while
the primary
water
ion (H
2
O)·CF
3
O
–
w
as
6
saturated
. The
(H
2
O)
2
·CF
3
O
–
ion
had a considerable temperature dependence that was
7
corrected for as a function of the measured flow tube temperatures. Atmospheric water
8
vapor, as measured by the Zeno weather station at 1 Hz, was used as an absolute
9
calibrant. Atmospheric water vapor was
calculated from the observed barometric
10
pressure, air temperature, and relative humidity for all dates included in the study (Fig.
11
S
2) and the corrected CIMS water vapor measurements compared well to the weather
12
station measurements (
Fig. S3
.)
13
14
4.
Eddy Covar
iance
(EC)
c
alculations
15
a.
Data processing
:
EC fluxes were calculated after the signals
of the compounds measured
16
by CIMS (
푥
, shown in Table S1)
were converted to mixing ratios (usually in pptv) by
17
applying all calibrations and corrections. Given the wind v
elocity vectors (
푢
,
푣
,
푤
) and
18
scalar matrices for species of interest, flux data for virtual temperature (
푤
′
푇
푣
′
),
19
momentum (
푤
′
푢
′
), and CIMS
-
determined compounds including water vapor (
푤
′
푥
′
) were
20
calculated for each
~ 30 minute flux measurement periods, where primes denote
21
deviation from the mean and overbars denote a mean over the flux period.
Large spikes in
22
CIMS signals, caused by electronic or temperature instability, were removed by
23
referencing a “quiet” m/z, whe
re no chemical signal is found, and by visual inspection (<
24
5% of data).
For each period: the 8 Hz wind measurement was interpolated onto the time
25
vector of the CIMS measurement, the sonic wind velocity coordinates were rotated by a
26
two
-
step rotation so th
at
푣
=0 and
푤
=0, the sonic data and CIMS data were detrended
27
using a linear detrending algorithm, the correction of inlet lag of the scalar signal of
푥
′
28
was corrected for lag with respect to
푤
′
by identifying the peak in the lag
-
covariance
29
function (
Fig
. S4
). Lag times were on the order of 0.1
–
1.1 s. The lag time for H
2
O was
30
8
used as the representative lag time for all compounds, which was justified by inspecting
1
the cross covariance spectra for multiple CIMS species for more than 10 daytime flux
2
period
s. Using a representative lag time from a compound
whose scalar vector
has a large
3
covariance with
푤
, was found to decrease the uncertainty of the resulting EC fluxes
4
propagated from the error in locating poorly
-
defined extrema in the cross covariance
5
spe
ctra for periods of lower flux (e.g., nighttime) or for compounds where the covariance
6
is small (e.g., MTNP). The instantaneous fluxes were then averaged over each flux
7
period. Mean concentrations and solar radiance data were averaged over the same
8
periods
.
9
b.
Quality of flux data
10
The EC fluxes were screened according to the following criteria: i. The upward and
11
downward energy fluxes at the surface should be balanced
within the standard deviation
12
of each measurement (~ 15 % in the daytime)
, ii. The spectral analysis should indicate
13
expected behavior of the individual covariances with respect to eddy magnitudes and
14
surface layer theory; iii. The turbulence should be well developed for the day time
15
periods; iv. The stationarity
(16)
and
inter
mittency
(17)
conditions should be met.
16
Condition
i
severely
limited the number of
useable
days
within the campaign for
17
calculations of
EC fluxes from CIMS.
The quality analysis for Conditions
ii
-
iv
below
18
pertains mainly to days where EC fluxes were deem
ed acceptable per Condition
i
.
19
i.
Energy balance closure condition
:
The degree of surface energy balance closure
20
provides an important and objective evaluation of the EC fluxes, as conditions
21
that violate EC flux assumptions (e.g., contamination of vertical f
lux from
22
horizontal wind due to effects of roughness layer inhomogeneity) should affect
23
the turbulent transfer of energy similarly to mass. An external calibration
24
(pyranometer measurements of radiation) was used to constrain the closure
25
condition. The sur
face energy balance can be written as:
26
27
R
n
=
S
H + LE +
S + G + Q (3)
28
29
where R
n
is the net radiation downward,
S
H is the sensible heat flux, LE is the
30
latent heat f
lux, S is the storage heat flux
,
G is the soil heat flux, and Q
31
9
collectively represents all other energy fluxes. Q is generally a small term and was
1
ignored in this work. We also make the assumption that the canopy is closed, so
2
that G can also be ignored. The revised energy balance equation is written as:
3
4
R
n
–
S
=
S
H + LE
(4)
5
6
And the heat fluxes are defined as:
7
8
SH
=
휌
푎
푐
푝
푤
′
푇
푣
′
(
5
)
LE
=
휌
푎
퐿
푣
(
T
)
푤
′
푥
퐻
2
푂
′
(
6
)
9
where
휌
푎
is the density of air (kg m
3
),
푐
푝
heat capacity of air (J kg
-
1
°
C
-
1
) at 1 atm
,
10
퐿
푣
(
T
)
is the la
tent heat of vaporization of air
(
kJ
kg
-
1
) calculated for the
11
temperatures experienced throughout the flux period, and
푥
퐻
2
푂
is the water vapor
12
fraction as determined by CIMS (described in Section 3a). The storage term S was
13
not measured, but was estimat
ed based on measurements at a similar site
(18)
. S is
14
usually small (
-
20
–
50 W m
-
2
) and thus, error in its estimation does not add
15
significantly to the error of the analysis. Our pyranometer measures solar
16
radiation (shortwave radiation downward) whereas
net radiation includes the
17
upwelling and downwelling shortwave and longwave radiation (R
n
= SW
up
+
18
SW
down
+ LW
up
+ LW
down
). As longwave radiation was not measured, R
n
was
19
estimated with a method similar to the treatment of satellite data, by exploiting
20
em
pirical relationships between net radiation and shortwave radiation as
21
developed by Kaminsky et al
(19)
. It was shown that the error in this type of
22
estimation is small (
r
2
= 0.96
–
0.99, root mean square error 18
–
41 W m
-
2
)
23
compared to the typical
magnitude of R
n
(up to 1000 W m
-
2
).
24
One of the largest influence on energy balance closure for our work
25
appeared to be wind direction, as southerly winds were typically associated with
26
poor balance closure and northerly winds, even with contributions from winds
27
originating from the west and
east, were associated with satisfactory balance
28
closure for the individual day (Fig.
S5
). This strong impact by wind direction is
29
likely related to the challenges of EC flux measurements from a walk
-
up tower
30
(e.g., tower and/or instrument enclosure acting
as physical barriers that isolate the
31
10
inlet and sonic from southerly winds) and the change in roughness element or
1
topology due to the transition from the forest in the north, west, and east
2
directions toward the grassy field to the south.
3
4
ii.
Spectral analy
sis
:
The cospectra of
x
and
T
v
with
w
in the frequency domain
5
were
calculated by applying a fast Fourier transform to the covariance matrices.
6
Figure
S6
shows the averaged cospectral densities of several CIMS compounds,
7
chosen for those with higher signal
-
to
-
noise to limit error,
and
T
v
with
w
for the
8
afternoon
local
hour
13
on
all
days included in this flux study.
If the
9
homogeneous fetch is adequate, an underlying assumption for EC, there should be
10
a development of an “inertial sublayer,” where the fluxe
s of conserved scalars
11
(e.g., energy) are roughly independent of vertical height. The spectra show the
12
expected linear falloff in the inertial subrange frequency (
f
> 0.003 Hz) and, for
13
the averaged CIMS compounds, the typical slope is similar to the
f
–
7/3
slope
14
predicted by surface layer theory
(20)
. For virtual temperature, the linear falloffs
15
for some cospectra were shallower than
f
–
7/3
, but never shallower than
f
–
5/3
, such
16
that the average for all days was most similar to an
f
–
6/3
slope.
17
The plot
s of cumulative distribution of co
spectr
al density (ogives) indicate
18
the frequency ranges where most of the flux is captured. Figure
S
7a shows the
19
representative ogives for
w
with several CIMS compounds
and
T
v
, for the
20
afternoon
local
hours 9
–
15 of all
d
ays included in this flux study
and Figure S7b
21
shows the data for JD165.The spectra were normalized to their asymptotic value
22
in the low
-
frequency range. As expected, the ogives approach horizontal
23
asymptotes at
both
ends of the spectrum
. In the high
-
frequ
ency end, there was no
24
more flux at approximately 1 Hz, which indicates that the measurement
25
timescales used in this work (10 Hz for CIMS species and 8 Hz for winds) was
26
sufficiently fast to capture fluxes carried from the smaller eddies (timescales of 10
27
s). In the low
-
frequency end, all of the flux was carried by eddies of frequencies
28
higher than 1x 10
-
3
Hz. This indicates that, as an ensemble, our averaging time of
29
30 minutes was long enough to capture the entirety of the flux. The
푤
′
푥
′
ogive
30
decreases more quickly than the
푤
′
푇
푣
′
ogive. This is perhaps a real characteristic,
31
11
as suggested by an earlier study of acyl peroxynitrate (APN) compounds at
1
BEARPEX
(21)
, because the spectra are similar for compounds whose
instrument
2
time
respon
se
delays were small
(Section 5).
3
Figure S8 shows the
frequency
-
weighted covariance
-
normalized cospectra
4
plotted against
n
=
f
z/U, where z is the measurement height and U is the average
5
wind speed for the day. The spectra for (a) JD165 and (b) all days
included in the
6
study show one distinct maximum corresponding to
n
~ 0.2
–
0.4 and few
7
complex turbulent structures, indicating that most of the flux was carried by
8
eddies of timescale ~ 30 s. The slope for CIMS compounds was steeper than for
9
푇
푣
in the h
igh frequency domain, also observed for APNs at BEARPEX
(21)
.
10
Combined, the spectral analysis for the species reported in this work provides
11
compelling evidence that the calculated EC fluxes
during the daytime hours are
12
accurately represented.
13
14
iii.
Turbulence
:
The criterion requiring well
-
defined turbulence can be accessed by
15
examining the
friction velocity
푢
∗
,
calculated in this work from the measured
16
momentum flux (
휏
=
−
휌
푎
푤
′
푢
′
), where
휌
푎
is the density of air (kg m
-
3
):
17
푢
∗
=
|
휏
/
휌
푎
|
1
/
2
(
7
)
Figure S
9 shows
푢
∗
averaged for all days included in this study and errors are one
18
standard deviation from the mean. The determination of turbulent threshold
19
푢
∗
values is subject to debate and may be site
-
specific. For periods where
20
conditions can be cha
racterized as turbulent, the
calculated EC fluxes are affected
21
by substantially fewer errors than for calm periods. Threshold values of 0.1
–
0.3
22
m s
-
1
have been suggested, and a median value of 0.23 m s
-
1
that was found to be
23
most representative of multip
le sites and years
(22)
was used in our work to
24
qualitatively assess if daytime periods can be characterized as turbulent and
25
further validate the flux data. It was found that, for
days with good energy
26
balance, the criterion was satisfied within the stand
ard deviation of the
27
푢
∗
measurement.
28
29
12
iv.
Stationarity and Intermittency
:
The stationarity test was performed as
1
suggested by Foken and Wichura
(16)
, where flux data
(F)
were calculated for 5
2
min averaging periods in addition to 30 min averaging periods. Whe
n both
3
calculations were averaged into the same 30 min time bin, the resulting
4
stationarity criterion
(
F
S
< 0.3), defined as
F
S
= |F
30min
–
F
5min
|/F
30min
, was
5
satisfied for the majority of points in the day time periods. The stationarity
6
criterion was used as a qualitative assessment and not a discrimination threshold
7
in this work as the available data are sparse. Figure
S
10 shows the stationarity
8
analysi
s for H
2
O
2
and sensible heat on JD165, where the 5 min data averaged
9
similarly to the 30 min data did not significantly alter the result. The intermittency
10
criterion
(17)
, defined as
F
I
= σ
5min
/ F
30min
where σ
5min
is the standard deviation in
11
the 5 minute
data, was satisfied (
F
I
< 1).
12
13
5.
EC flux corrections:
14
Standard
corrections:
C
orrections that are often
recommended for
EC flux
15
calculations
(23)
were
systematically
applied in this work
where appropriate
. Despiking, time
16
lag, and coordinate rotation were
performed
as
discussed in
Section
4a. Cross
-
wind
17
corrections were applied online
within
the CSAT3 sonic
electronics
. The effects of
18
temperature and humidity on the measured air temperature
, e.g., buoyancy,
were removed in
19
the calculation of the virtual tem
perature from the sonic speed of sound.
C
orrections
due to
20
flux attenuation at high frequencies
were unnecessary as the CIMS measurement
was fast
21
enough to capture
all the flux in the high
-
frequency range (Fig.
S
7a
). Webb
-
Pearman
-
22
Leuning (WPL) corrections
(24)
were unnecessary as we measured the mixing ratio not the
23
partial pressure of chemical species, and temperature and humidity corrections
to CIMS
24
sensitivities needed to output count signals to pptv
were applied internally
as a standard
25
procedure
.
Trans
fer functions used to correct for the potential loss of flux due to the
26
separation of the inlet (Fig. S7b) were calculated as suggested by Moore et al
(25)
but were
27
not applied because the error was estimated to be small ( ~ 3 % for CIMS compounds).
28
C
orrections often reserved for open
-
path analyzers
were not performed for the closed
-
system
29
CIMS measurements
.
30
13
Instrumental
time response
correction:
It is challenging to accurately measure
1
c
ompounds
that have
a
propensity to interact with instrument
surfaces in complex ways.
2
HNO
3
is of particular concern for our instrument, and we discuss here correction for the loss
3
of flux owing to a “smearing” effect of the HNO
3
signal in time from the interaction with
4
instrument surfaces. The correction was based
on observed instrumental responses for HNO
3
,
5
and a few other compounds where calibration standards were available in the field during
6
SOAS. An exponential decay curve (
훼
·exp(
-
t/
τ
)) was fit to periods following a pulse of the
7
calibrant gas through the CIM
S flow region, where τ (s) is the inlet time response constant
8
and α is the pre
-
exponential factor. Figure
S
11 shows the decay curves for HNO
3
(τ
= 32, α =
9
0.55), formic acid
(τ
= 0. 94, α = 0.80), H
2
O
2
(
τ
= 1.3,
α = 0.70), and water vapor
(τ
= 0.22, α
10
= 1
.06) pulses. The fitted parameters for HNO
3
can be used to “degrade” signals with fast
11
time responses (e.g., H
2
O
2
) and recalculate the fluxes by applying the smearing perturbation
12
to the measured mixing ratios for each
α and
τ
of interest. The adjusted mix
ing ratio
C(t)
can
13
be written as:
14
퐶
(
푡
)
=
푋
푡
(
1
−
훼
)
+
훼
·
∫
푋
푖
·
exp
−
(
푡
−
푡
푖
)
휏
푡
푡
−
2
휏
·
푑
푡
푖
∫
exp
−
(
푡
−
푡
푖
)
휏
푡
푡
−
2
휏
·
푑
푡
푖
(
8
)
where
푡
푖
is the time in seconds at each iteration
i
,
and
푋
is the observed mixing ratio of the
15
chemical species.
Figure
S
12 shows the results of applying various
τ
, while fixing α,
to the
16
H
2
O
2
mixing ratio vector.
A 1s time delay does not visibly change the observed H
2
O
2
mixing
17
ratio
but longer time responses noticeably degrade the signal such that the high
-
frequency
18
varia
tions are damped. By design, the smearing function conserves signal.
19
We applied the smearing perturbation of HNO
3
(
τ
= 32 s, α = 0.55) to the EC
20
calculations of H
2
O
2
, formic acid, and latent heat
.
We find that mean chemical mixing ratios
21
do not change more than a few percent until time constants approach 1 or more hours (the
22
timescale of diurnal variation); however, fluxes were affected for time constants on the order
23
of seconds.
Figure
S
13, top pan
els, shows that
the H
2
O
2
flux is significantly decreased from a
24
signal degradation in the instrument, the mean concentration remains unperturbed, and, thus,
25
the deposition velocity is significantly suppressed. The ratio of the undamped vs. damped
26
values i
n V
d
and fluxes ranged from 1.5
–
1.8 for multi
-
day analysis of the formic acid, H
2
O
2
,
27
14
and LE flux (Fig
. S13
, bottom panels). An average value of 1.62 was used to correct the
1
HNO
3
flux to the value that would have been measured if the time response constan
t for
2
HNO
3
was τ < 1 s for all of the days included in this study. The mean mixing ratio values
3
were uncorrected. We note that the analysis only includes dampening caused in the CIMS
4
flow tube region, where the majority of the residence time is expected fo
r our instrument. It
5
is possible that the inclusion of the fast flow (~ 2000 L min
-
1
) inlet interaction will require
6
greater dampening corrections; however, we do not have time decay data for compounds
7
through this section of the inlet (the chemical pulse
would need to be at the tip of the inlet).
8
The correction for HNO
3
closes the gap between the measured and modeled V
d
.
9
We further explored the effect of the time constant magnitude on the ratio of the
10
undamped vs. damped fluxes using realistic time con
stants (τ = 1
–
100 s) for CIMS. This
11
time range is also relevant to eddy scales carrying most of the flux we measured, where the
12
loss of flux becomes especially important
. Figure S14
shows that as τ increases, the ratio of
13
damped vs. undamped V
d
decrease
, as expected, but the decrease is not linear (
Fig. S14
14
insert panel
). For τ = 1 s, greater than 98% of the H
2
O
2
flux is conserved and for τ = 180 s,
15
half of the flux is gone which leads to a significant underestimation in the calculated V
d
16
values. Interes
tingly, most of that flux was lost between 1 and 32 seconds. Flux loss due to
17
chemical interactions with surfaces is expected to be important for other “sticky” compounds
18
like NH
3
and IEPOX, where a similar correction may be needed.
19
For most of the CIMS compounds included in this work, time delays are expected to
20
be small (on the order of 1 s) with, perhaps, the exception of IEPOX. Unfortunately, we do
21
not have field calibration sources for all compounds for which to attempt a damping
analysis.
22
For the combined ISOPOOH +IEPOX flux, ISOPOOH comprise greater than 66% of the
23
mixing ratio
signal for most cases
.
H
owever, the combined V
d
may still be underrepresented
24
by our measurements if the IEPOX has time constant close to that of HNO
3
.
25
26
6.
Resistance Model
27
Many compounds detectable by CIMS were observed to have relatively high
28
deposition velocities, suggesting a small or negligible resistance to surface uptake by plant
29
stomatal or non
-
stomatal components such as leaf cuticles. We calculate
the expected
30
15
contributions to deposition using a parameterization of surface deposition suggested by
1
Weseley and Hicks
(29)
, assuming a resistance
-
in
-
series scheme that considers the
2
aerodynamic resistance (R
a
), molecular diffusion resistance (R
b
) and sur
face resistance (R
c
)
3
to the canopy that is parameterized as a large leaf:
4
푉
푑
=
1
푅
푎
+
푅
푏
+
푅
푐
(
9
)
R
a
describes turbulent transfer of mass in the mixed layer to the surface, and can be
5
parameterized by:
6
푅
푎
=
1
푘
푢
∗
[
푘
푈
(
푧
−
푑
)
푢
∗
−
휓
(
푧
퐿
)
]
(
1
0
)
where
푈
(
푧
−
푑
)
is the mean wind speed at a height equal to the measurement height (z) less the
7
displacement height (d, i.e., thickness of the “leaf”),
푢
∗
is the friction velocity (m s
-
1
, see Eq.
8
7),
k
is the dimensionless von Kármán constant (0.4),
휓
(
푧
퐿
)
is a correction function for the
9
sensible heat and momentum fluxes that is dependent on the dimensionless parameter
z/L
10
often used to characterize atmospheric stability. The formulation of
휓
(
푧
퐿
)
for the unstable
11
period (
local
h
= 9
–
15) was used to
obtain the daytime R
a
(30)
. L is the Monin
-
Obukhov
12
length, defined as:
13
퐿
=
푢
∗
3
훩
푘푔
푤
′
훩
′
(
11
)
where
훩
is the mean potential
temperature (K)
,
i.e.,
the temperature of an air parcel
14
transported adiabatically to surface
pressure
as calcula
ted from the measured virtual
15
temperature and atmospheric pressure,
푤
′
훩
′
is the potential temperature flux, and
푔
is the
16
acceleration due to gravity (m s
-
2
).
17
R
b
describes the diffusion of molecules through the quasi
-
laminar layer at the surface
18
of the roughness element.
R
b
was modeled following the parameterization suggested by
19
Jensen and Hummelsh
ø
j
(31, 32)
that included a direct dependence on leaf area index (LA
I)
20
16
and the characteristic le
af thickness scale in the mixed
-
vegetation
canopy (
푙
, taken to be
1
0.001 m
(31)
):
2
푅
푏
=
휈
퐷
푥
푢
∗
[
100
푙
푢
∗
(
퐿퐴퐼
)
2
휈
]
1
/
3
(
12
)
ν
is the viscosity of air (m
2
s
-
1
) at the ambient pressure of the measurement height, and
D
x
is
3
th
e diffusivity of a molecule
x
in air.
D
H2O2
was taken to be 1.56 x 10
-
5
m
2
s
-
1
at 25°C (scaled
4
from measured values at 60°C
(33)
), and
D
x
for other molecules in this study was calculated
5
from
D
H2O2
using Graham’s Law, where
D
x
=
D
H2O2
(MW
H2O2
/MW
x
)
1/2
. Table S4 shows the
6
diffusivity coefficients and other parameters used in the model for chemical species included
7
in this study and the results of the model.
8
LAI was measured
during the SOAS campaign for the CTR site by coauthor G.M.
9
Wolfe with help from C.J. Groff (Purdue University). Measurements were performed using
10
an upward
-
looking light sensor
(
LAI
-
2000
Plant Canopy Analyzer)
that measures th
e diffuse
11
radiation attenuati
on
,
i.e
.
,
the fraction of sky blocked by canopy elements
,
within the canopy
12
at multiple viewing angles. The attenuation
is a function of total leaf area and average leaf
13
angle.
A representative value (LAI = 4.7) was taken as the mean along three transects
at
14
different orientations.
15
The average R
a
value for the site on day 165 (June 15) was calculated to be ~ 8 s m
-
1
16
and R
b
for H
2
O
2
was ~ 12 s m
-
1
. From the R
a
and R
b
values calculated, we find that the
17
residual resistance (1/V
d, meas.
–
R
a
–
R
b
) is
heavily
-
dependent on the solubility (e.g., the
18
Henry’s Law coefficient (
H
)), with an additional dependence on the molecular mass of the
19
individual compounds. That the residual resistance depends on solubility suggests that the
20
bulk of this resistance is du
e to surface uptake. To model the surface resistance, we adjust the
21
original parameterization by Weseley
(27)
. The Weseley parameterization incorporates the
22
dependence of R
c
on molecular mass (e.g., diffusivity) in the stomatal resistance term (
푟
푠
)
23
and o
n
H
in the mesophyll (
푟
푚
) and cuticular (
푟
푐푢푡
) resistance terms. The canopy was
24
assumed to be closed, and thus only the surface resistance from the canopy top was
25
considered, i.e., neglecting contributions from the soil and “lower canopy.”
26
17
푅
푐
=
(
1
(
푟
푠
+
푟
푚
)
+
1
푟
푐푢푡
)
−
1
(
13
)
An adjustment to the Weseley parameterization was necessary because the original scheme
1
overestimate R
c
for H
2
O
2
and HNO
3
in our work (demonstrated by various experimenters and
2
in this work to be less than 5 s m
-
1
(34, 35)
). O
ther evaluations of observed vs. modeled H
2
O
2
3
flux in a forest
(36)
similarly concluded that the original Weseley scheme overestimated R
c
4
for H
2
O
2
. Coefficients in the parameterizations of
푟
푚
and
푟
푐푢푡
were empirically revised to
5
close the
discrepancy. We demonstrate that the revised parameterization successfully predicts
6
R
c
for not only H
2
O
2
and HNO
3
, but also for the series of organic and inorganic compounds
7
studied in this work. Further validation is needed before this revised Weseley sch
eme can be
8
generally applied for all compounds, at all sites, and for all seasons. The resistances to leaf
9
components can be written as:
10
푟
푠
=
(
퐷
퐻
2
푂
퐷
푥
)
푟
퐻
2
푂
(
14
)
푟
푚
=
(
퐻
50
푅푇
+
100
푓
0
)
−
1
(
15
)
푟
푐푢푡
=
(
10
−
4
퐻
/
푅푇
+
푓
0
)
−
1
(
16
)
where
푟
퐻
2
푂
is
the stomatal resistance to the diffusion of water,
H
is the Henry’s Law
11
coefficient (
M atm
-
1
),
R
is the gas constant (atm M
-
1
K
-
1
),
T
(K) is the air temperature, and
푓
0
12
is a reactivity factor
(27)
that is defined as 0 = non
-
reactive, 0.1 = semi
-
reactive and 1 =
13
reactive as ozone. The values of
D
x
,
H
and
푓
0
used in this work are reported in Table S2.
14
Values of
푓
0
were used as suggested by Weseley for available compounds, and set as 0
15
otherwise. T
he
푟
퐻
2
푂
term may be dependent on number variables, such as photosynthetically
-
16
active radiation, CO
2
air mixing ratio and assimilation, water fraction in the leaves and in the
17
air, stomatal density, LAI, and air temperature. In the absence of measurement
s, only
18
daytime values of V
d
were calculated here, and daytime
푟
퐻
2
푂
was assumed to be similar to
19
the value computed at the BEARPEX campaign
(21)
.
20
For
many of
the multifunctional compounds studied in this work,
H
is
not known
;
21
thus,
we
estimate
H
base
d on chemical proxies.
U
ncertainties in
H
estimates
, which
can be >
22
18
100% due to the paucity of measurements
,
are critically important to the R
c
term
for certain
1
ranges of
H
(see Fig. S16 and related discussion)
.
Thus, resistance model results
should be
2
treated as a rough approximation in this work for compounds with unknown
H
. An example
3
of th
e
large uncertainty
in
H
is the trend with increasing carbon number for the family of
4
hydroxy
nitrates. Shepson et al
(37)
found decreasing
H
for hydroxy
nitrates of increasing
5
carbon number, whereas Treves et al
(38)
observed the opposite trend.
H
can be significantly
6
different for different isomers of the same compound; however, the trend of decreasing
H
7
with increasing carbon number
appears to be
more co
nsistent with observations of some
8
chemical families, e.g., in the family of C
1
–
C
9
straight
-
chain aldehydes or C
3
–
C
11
straight
-
9
chain ketones (
(39)
,
and references therein
). This trend
may
be rationalized
because
the
10
addition of a more hydrophobic alky
l group, while conserving the singular hydrophilic group
11
(
e.g., aldehyde
)
sh
ould depress
H
if all else is equal.
12
Neither the Shepson et al nor the Treves et al
H
values can be applied directly,
13
however,
as the nitrates we measured were not exclusively
alkyl
hydroxy
nitrates
, e.g.,
14
ISOPN is a hydroxy
nitrate with an alkenyl group, PROPNN is a nitrate with a ketone or
15
aldehyde group,
and
MACN + MVKN are hydroxy
nitrates with a ketone or aldehyde
16
group
s
.
Thus, we used
the Shepson et al
H
values, and extra
polate
d
to higher carbon numbers
17
when necessary
. The estimation for IN
P
and MTN
P
are especially uncertain. As more
18
measurements become available due to greater availability of synthesized authentic
19
standards, the uncertainties in the resistance model can b
e revised downward. The modeled
20
resistances are shown in
Fig. S1
5
. Modeled R
c
agree
well with previous observations for
21
HNO
3
and H
2
O
2
(0
-
5 s m
-
1
), and ROOH (20
-
40 s m
-
1
)
(34, 35, 40)
. HCN is both observed
22
and calculated to have the largest total resis
tance to deposition. Further, R
c
is the largest
23
component of the summed resistance
for HCN, due to its small
퐻
value.
24
Here w
e
estimate the sensitivity of the resistance model to
parameters such as LAI,
25
u
* and leaf thickness.
A hypothetical ±
200% change
in these input parameters modified R
b
26
by |37
–
58|% and V
d
by ~ |25
–
30|%. This V
d
model error is similar to
standard deviation of
27
the measurements
. At SOAS,
our
measurements of LAI
in particular,
performed from
28
6/18/2013
–
6/20/2013 at viewing angles 90
–
270 degrees at three locations within the forest
29
had a 2
-
sigma uncertainty of 35% (for a confidence level of 95%). This translates to an R
b
30