AMTAtmospheric Measurement TechniquesAMTAtmos. Meas. Tech.1867-8548Copernicus PublicationsGöttingen, Germany10.5194/amt-10-3073-2017Effects of variation in background mixing ratios of N2, O2, and Ar
on the measurement of δ18O–H2O and δ2H–H2O values by cavity ring-down spectroscopyJohnsonJennifer E.jjohnson@carnegiescience.eduhttps://orcid.org/0000-0003-2181-8402RellaChris W.Department of Ecology and Evolutionary Biology, University of
Arizona, Tucson, AZ, USAPicarro, Inc., Santa Clara, CA, USAcurrent address: Department of Global Ecology, Carnegie
Institution, Stanford, CA, USAJennifer E. Johnson (jjohnson@carnegiescience.edu)24August20171083073309111April201724April201730June201713July2017This work is licensed under the Creative Commons Attribution 3.0 Unported License. To view a copy of this licence, visit https://creativecommons.org/licenses/by/3.0/This article is available from https://amt.copernicus.org/articles/10/3073/2017/amt-10-3073-2017.htmlThe full text article is available as a PDF file from https://amt.copernicus.org/articles/10/3073/2017/amt-10-3073-2017.pdf
Cavity ring-down spectrometers have generally been
designed to operate under conditions in which the background gas has a
constant composition. However, there are a number of observational and
experimental situations of interest in which the background gas has a
variable composition. In this study, we examine the effect of background gas
composition on a cavity ring-down spectrometer that measures δ18O–H2O and δ2H–H2O values based on the
amplitude of water isotopologue absorption features around 7184 cm-1
(L2120-i, Picarro, Inc.). For background mixtures balanced with N2, the
apparent δ18O values deviate from true values by -0.50 ± 0.001 ‰ O2 %-1 and
-0.57 ± 0.001 ‰ Ar %-1, and apparent δ2H
values deviate from true values by 0.26 ± 0.004 ‰ O2 %-1 and 0.42 ± 0.004 ‰ Ar
%-1. The artifacts are the result of broadening, narrowing, and
shifting of both the target absorption lines and strong neighboring lines.
While the background-induced isotopic artifacts can largely be corrected
with simple empirical or semi-mechanistic models, neither type of model is
capable of completely correcting the isotopic artifacts to within the
inherent instrument precision. The development of strategies for dynamically
detecting and accommodating background variation in N2, O2, and/or
Ar would facilitate the application of cavity ring-down spectrometers to a
new class of observations and experiments.
Introduction
In most commercially available laser absorption spectrometers, the accuracy
and precision of trace gas measurements are sensitive to the composition of
the background gas. In this paper, we explore this issue in the context of a
class of laser absorption spectrometers that is of increasing importance for
environmental research, the cavity ring-down spectroscopy (CRDS) analyzers.
While the CRDS analyzers can accurately and precisely measure the
concentration and isotopic composition of trace gases in situations where
the background gas has a constant composition, they make substantial
measurement errors in situations where the background gas has a variable
composition (Chen et al., 2010;
Friedrichs et al., 2010; Aemisegger et al., 2012; Becker et al., 2012; Nara et al., 2012; Long et al., 2013; Volkmann and Weiler, 2014). In variable
backgrounds, measurement errors emerge from the interaction between two
factors: first, collisional shifting and broadening of the trace gas
absorption transitions; and second, the spectral acquisition and analysis
strategies employed by the CRDS analyzers (Hendry et al., 2011; Gralher et al., 2016; Sprenger et al., 2017). While the fundamental
collisional effects are qualitatively well understood, their quantitative
impacts on analyzer performance and the strategies needed to overcome those
impacts are both incompletely understood.
To date, background effects on CRDS measurements have been reported in three
different types of situations. First, calibrations for observations of the
unconfined atmosphere: even though the natural levels of variability in
atmospheric N2, O2, and Ar mixing ratios are small (i.e.,
∼ 100 ppmv), large contrasts can occur between the average
composition of the atmosphere and the composition of the mixtures used for
calibration (i.e., ∼ 10 000 ppmv; Chen et al., 2010; Aemisegger et al., 2012; Nara et al., 2012; Long et al., 2013).
Second, observations of confined atmospheres: for trace gas measurements in
lakes, streams, oceans, and soils, the background concentrations of O2
can vary naturally over a wide range because the rates of biological
processes that produce and consume this gas can proceed more rapidly than
the physical processes that control mixing with the unconfined atmosphere
(i.e., ∼ 150 000 ppmv; Friedrichs et al.,
2010; Becker et al., 2012). Third, experiments with active control of
background composition: some measurement techniques utilize N2 dilution
to modulate the concentrations of target trace gases in both confined and
unconfined atmospheric backgrounds (Volkmann and Weiler, 2014;
Gralher et al., 2016).
The fundamental physical mechanisms that give rise to background gas effects
on CRDS measurements are well-understood. The CRDS analyzers use
high-finesse optical cavities to make ultra-sensitive quantitative
absorption measurements based on infrared absorption transitions of various
trace gases (O'Keefe and Deacon, 1988). Two features of the absorption
transitions of the trace gases are affected by collisions with the
background gas: (i) the frequencies of maximum absorption intensity (i.e.,
denoted ν0) and (ii) the shapes of the absorption line profiles around
those central frequencies (i.e., described by I(ν0), the maximum
amplitude at ν0, and I(ν0)/2, the full width at half maximum)
(Demtröder
2014; Hanson et al., 2015). The former effect is termed “shifting”; the
latter effect is termed “broadening” and/or “narrowing”. Due to these
effects, any contrast between the backgrounds used for calibrations versus
observations changes the geometry of the target absorption spectrum and has
the potential to introduce errors into the resulting measurements of trace
gas concentrations and isotope ratios. However, whether or not errors
actually occur in any given CRDS analyzer is a function of the specific
absorption spectra that are targeted and how those spectra are acquired and
interpreted.
In principle, it should be possible to make CRDS measurements that are
completely insensitive to background gas composition by measuring the
integrated absorbance (i.e., the absorption peak area) of any isolated
absorption feature in a given spectrum (Zalicki and Zare, 1995). In
practice, however, most current-generation CRDS analyzers are expected to
exhibit some degree of sensitivity to background gas composition because
they (i) target absorption features that are not completely isolated from
neighboring absorption features; (ii) measure the amplitude, rather than the
area, of the target absorption features; and/or (iii) attempt to optimize
measurement precision by treating lineshape parameters as fixed rather than
free variables (Hodges and Lisak, 2006; Hendry et al., 2011;
Steig et al., 2014). On account of these design constraints, the
susceptibility of different CRDS analyzers to background gas effects is a
function of the identity of the specific absorption features that are
targeted, the spectral acquisition approach that is used to measure those
features, and the spectral analysis techniques that are used to interpret
the measurements. The interactions between these factors make it difficult
to predict how any particular analyzer will respond to background gas
variation. As a result, experimental measurements are necessary to determine
both the quantitative impacts of background gas variation on analyzer
performance and the strategies needed to overcome those impacts.
The overall objective of this study is to characterize how background gas
composition affects measurements of water isotopologues in one
commercially available CRDS analyzer, the L2120-i manufactured by Picarro,
Inc. Three factors make the L2120-i an attractive test bed for studying
background gas effects. First, a number of other types of interference have
been studied in the L2120-i. Previous work has characterized interference
from self-broadening (Schmidt et al., 2010) and from organic
contaminants (Brand et al., 2009; West et al., 2010), tested
algorithms for correcting for organic interference during or after analysis
(Hendry et al., 2011; Schultz
et al., 2011; West et al., 2011; Schmidt et al., 2012; Martín-Gómez et al., 2015; Johnson et al., 2017), and tested peripherals for pyrolyzing or
oxidizing organic contaminants prior to analysis
(Berkelhammer et al., 2013; Martín-Gómez et al.,
2015; Lazarus et al., 2016). Second, the L2120-i has been widely used to
measure δ18O–H2O and δ2H–H2O values in
situations where background variation could be relevant to the calibration
procedures and/or the fundamental measurements. Examples include
applications to measurements of liquid water in precipitation
(Munksgaard et al., 2011), plant water (West et al., 2011),
soil water (Herbstritt et al., 2012), and seawater (Munksgaard et al.,
2012), as well as water vapor in the terrestrial boundary layer
(Berkelhammer et al., 2013) and marine boundary layer (Steen-Larsen et al., 2014). Third, it has recently been shown that the L2120-i measurements
are highly sensitive to the N2/ O2, N2/ CO2, and
CO2/ O2 composition of the background gas and that the magnitude
of the sensitivity is relevant to many observational and experimental
situations (Gralher et al., 2016).
Schematic diagram of the mixing system used for the
experiments. Pure background gases were obtained commercially, mixed with
mass flow controllers and supplied to the inlet of the L2120-i at a slight
overpressure. See text for details.
To evaluate how background gas composition impacts L2120-i measurements and
the strategies needed to correct for those impacts, we carried out a series
of experiments addressing the following questions:
What are the magnitudes of the effects of variation in the mixing ratio of
N2/ O2, N2/ Ar, and O2/ Ar on the apparent δ18O–H2O and δ2H–H2O values measured by the
L2120-i CRDS analyzer?
How are the background effects on apparent δ18O–H2O and
δ2H–H2O values derived from the interaction between the
target spectra and the spectral acquisition and analysis strategies in this
instrument?
Is it practicable to develop post hoc calibrations for this instrument that
accurately account for the effects of background variation in N2,
O2, and/or Ar on the apparent δ18O–H2O and δ2H–H2O values?
Sensitivity of δ18O and δ2H
values to background N2, O2, and Ar mixing ratios. For each
experiment, sensitivity is plotted as the difference between the apparent
and true isotopic composition of the standards (i.e., Δδ18O =δ18Oapparent-δ18Otrue
and Δδ2H =δ2Happarent – δ2Htrue). For each panel, n= 330 measurements of four liquid
standards across a range of injection volumes (i.e., 400–2400 nL, in 11
steps of 200 nL each). Points represent mean values ± SD for
replicates of each standard at each injection volume, and regression slopes
are given by β5 values in Table 2.
MethodsBackground gas mixtures
Background gas streams with various compositions of N2, O2, and Ar
were generated with a mixing system (Fig. 1). The mixing system consisted
of four cylinders of compressed gas, thermal mass flow controllers, and a
back-pressure regulator upstream of the CRDS instrument inlet. Three of the
cylinders contained ultra high-purity N2, O2, and Ar (99.999 %
purity, < 3 ppm H2O, and < 0.5 ppm total hydrocarbon
content (THC); ALPHAGAZ 1, Air Liquide America Specialty Gases LLC, Houston,
TX, USA). The fourth cylinder contained ultra high-purity air (< 1 ppm H2O, < 0.01 ppm THC,
< 0.01 ppm CO, < 0.001 ppm NOx, < 0.001 ppm SO2; Ultrapure Air, Scott-Marrin,
Inc., Riverside, CA, USA) with the N2, O2, and Ar composition of
the natural atmosphere (i.e., 78.1 % N2, 20.9 % O2, 0.9 %
Ar; Brewer et al., 2014; Flores et al., 2015). In the experiments,
background gas mixtures were dynamically mixed from these cylinders with the
mass flow controllers (FC-260 with RO-28, Tylan-Mykrolis, Allen, TX, USA).
The mass flow controllers were calibrated with a bubble flow meter (25 mL
Kimax bubble flow tube, Kimble-Chase, Vineland, NJ, USA) and mixing accuracy
was tested for N2/ O2 and Ar / O2 mixtures with a galvanic
oxygen sensor (MO-200, Apogee Instruments, Logan, UT, USA). With this
system, the composition of each mixture could be controlled to an accuracy
of ±0.1 % of each constituent. The back-pressure regulator was
used to ensure that the mixtures were supplied to the CRDS analyzer inlet at
2.5 psi above atmospheric pressure.
Liquid water standards
All of the measurements in this study were based on four vaporized liquid
standards. The isotopic composition of the standards was initially
established by measurement with a Finnigan Delta S Isotope Ratio Mass
Spectrometer (Thermo Fisher Scientific, West Palm Beach, FL, USA) in the
Environmental Isotope Laboratory, Department of Geosciences, University of
Arizona (Tucson, AZ, USA). For oxygen, samples were equilibrated with
CO2 gas at approximately 15 ∘C in an automated equilibration
device coupled to the mass spectrometer. For hydrogen, samples were reacted
at 750 ∘C with Cr metal using a Finnigan H/Device coupled to the
mass spectrometer. Standardization was based on distilled water standards
referenced to VSMOW2 and SLAP2.
The resulting standards had the following isotopic compositions: (1) δ18O =-3.74 ‰,
δ2H =-15.3 ‰; (2) δ18O =-9.52 ‰, δ2H =-62.2 ‰;
(3) δ18O =-14.18 ‰, δ2H =-102.7 ‰; (4) δ18O =-30.32 ‰, δ2H =-246.7 ‰.
These values were determined with analytical precision of ±0.08 ‰ for δ18O and ±0.9 ‰ for δ2H.
To ensure that the isotopic
composition of the standards remained stable over time, they were stored in
1 L amber glass bottles, with Polyseal cone-lined screw caps, sealed with
Parafilm. For CRDS measurements, 1.5 mL aliquots of each standard were
pipetted into 1.8 mL glass vials with polypropylene screw caps and bonded
PTFE–silicone septa (66020-950 and 46610-700, VWR, Radnor, PA, USA). To
eliminate any effects from diffusive losses through the septa, each vial was
measured within 24 h of being filled and was sampled for a maximum of n= 10 successive injections.
Spectral acquisition
The CRDS analyzer used in these experiments was an L2120-i (Picarro, Inc.,
Santa Clara, CA, USA). The key components of this analyzer are a laser, a
wavelength monitor, an optical cavity, and a photodetector. The laser
targets H2O absorption lines close to 7184 cm-1 (1392 nm). The
specific lines that are utilized are 7183.685 cm-1 (1392.043 nm)
for 1H1H16O, 7183.585 cm-1 (1392.063 nm) for
1H1H18O, and 7183.972 cm-1 (1391.988 nm) for
1H2H16O (Tennyson et al.,
2009, 2010, 2013). Operationally, the
analyzer scans the laser across these features, recording absorption loss as
a function of optical frequency (spectrograms). To generate each frequency
and absorption pair, light from the laser is directed into the optical
cavity, the frequency is determined by the wavelength monitor, and the power
in the cavity is monitored with a photodetector detecting light leaking
through one of the mirrors. The absorption is quantified based on the rate
at which the light intensity decays (i.e., “rings down”) when the laser is
turned off (Crosson, 2008).
Since the absorbance measurements are of gas-phase H2O, a front-end
peripheral must be used to convert liquid-phase standards into the gas phase
(Gupta et al., 2009). For this study, the L2120-i analyzer was
equipped with a V1102-i high-precision vaporizer (Picarro, Inc., Santa
Clara, CA, USA) and autosampler (HTC PAL, Leap Technologies, Carrboro, NC,
USA). To ensure stable performance, the analyzer, vaporizer, and autosampler
were installed in an air-conditioned laboratory where the air temperature
was maintained at 20.0 ± 1.7 ∘C. All of the measurements
were performed in the air carrier mode and with the vaporizer running at
110 ∘C. Injections were made on a 9 min cycle, using a
10 µL
syringe (SGE 10R-C/T-5/0.47C, Trajan Scientific Americas, Inc., Austin, TX,
USA) which was rinsed twice in 1-methyl-2-pyrrolidinone (99.5 %, Acros
Organics, Fisher Scientific, Pittsburgh, PA, USA) before each injection.
Spectral analysis
In the L2120-i, the spectrograms are interpreted with a nonlinear curve
fitting routine based on the Levenberg–Marquardt algorithm. The analysis is
conceptually similar to that utilized in the earlier generation L1102-i
analyzers (Hendry et al., 2011), but some details differ. Briefly,
the fitting routine compares each measured spectrogram to a modeled
spectrogram and adjusts the model parameters in order to minimize the
residual error. The modeled spectrogram represents a mixture of
1H1H16O, 1H1H18O, and 1H2H16O
in a pure water standard (i.e., one that was evaluated in a zero air
background and had an isotopic composition near δ18O = 0 ‰ and δ2H = 0 ‰).
The fitting routine compares this modeled spectrogram to the measured
spectrogram in three stages.
In the first stage, the fitting routine varies the amounts of
1H1H16O, 1H1H18O, and 1H2H16O,
the centration and scale of the frequency axis, the absolute value and slope
of the baseline, the linewidth, and the amounts of several potential organic
contaminants (CH4, C2H6, and MeOH). This fit determines the
observed centration and scale of the frequency axis (“h2o_shift” and “h2o_squish_a”), the observed
linewidth (“h2o_y_eff_a”), and
the linewidth expected for the observed amount of 1H1H16O in
an air background (“h2o_y_eff”). The second
and third stages of fitting then make different assumptions about the
presence of organic contaminants.
In the second stage, the fitting routine assumes that there is no organic
contamination. The centration and scale of the frequency axis are fixed
based on the results of the first stage (“h2o_shift” and
“h2o_squish_a”) and the effective linewidth is
fixed based on the amount of 1H1H16O (“h2o_y_eff”). The free parameters are the amounts of
1H1H16O, 1H1H18O, and 1H2H16O,
as well as the absolute value and slope of the baseline. This fit determines
the reported residuals (“standard_residuals”), baseline
(“standard_base”), baseline slope (“standard_slope”), H2O mixing ratio (from the amplitude of the
1H1H16O peak), and the δ18O and δ2H
values (from the ratios of the amplitudes of the 1H1H16O,
1H1H18O, and 1H2H16O peaks).
In the third stage, the fitting routine allows for the possibility of
organic contamination. Here, the centration and scale of the frequency axis
are fixed based on the results of the first stage (“h2o_shift” and “h2o_squish_a”) and the effective
linewidth is fixed based on the observed linewidth (“h2o_y_eff_a”). The free parameters are the amounts
of the organic contaminants, the amounts of 1H1H16O,
1H1H18O, and 1H2H16O, as well as the absolute
value and slope of the baseline. This fit determines the reported
“organic-corrected” residuals (“organic_res”), baseline
(“organic_base”), baseline slope (“organic_slope”), and δ18O and δ2H values (from the ratios
of the organic-corrected amplitudes of the three peaks).
Experimental designCharacterizing background gas effects
We performed three experiments to characterize the effects of variation in
the mixing ratios of N2/ O2, N2/ Ar, and O2/ Ar,
respectively. In each experiment, we generated five backgrounds from the two
gases (i.e., 0/100, 25/75, 50/50, 75/25, and 100/0; in %). In each
background, we measured the four liquid standards across a range of
injection volumes (i.e., 400–2400 nL, in 11 steps of 200 nL each). For
the most isotopically enriched standard, we performed three replicate
injections at each injection volume; for the other three standards, we
performed a single injection at each injection volume. At each transition
between standards, we inserted 15 additional injections to allow for
full equilibration and eliminate any carryover effects. At each transition
between backgrounds, we inserted four additional injections in UHP N2
to check for instrumental drift. Both the transition injections and the
drift check injections were analyzed solely for quality control and quality
assurance purposes. For the primary analyses of N2, O2, and Ar
effects, the remaining injections yielded a total sample size of n= 330
per experiment and n= 990 across the three experiments.
Evaluating corrections for background gas effects
We performed a fourth experiment to test whether the background effects
observed in the pure gases (N2, O2, Ar) and binary gas mixtures
(N2/ O2, N2/ Ar, and O2/ Ar) could be used to predict the
effects in a ternary mixture representing natural atmospheric composition
(N2/ O2/ Ar). In this experiment, we used the ultra high-purity
whole air as the background and measured the same four liquid standards
across a range of injection volumes (i.e., 600, 1200, 2000, 3000 nL). For
each standard, we performed three replicate injections at each injection
volume. At each transition between standards, we again inserted 15
additional injections to allow for full equilibration and eliminate any
carryover effects. The transition injections were analyzed solely for
quality control and quality assurance purposes, such that the remaining
injections yielded a total sample size of n= 240.
Data analysis
Since the default configuration of the L2120-i software does not write all of
the intermediate spectral parameters to the liquid injection output files,
we retrieved the analyzer's complete raw data files for the duration of
these experiments from the archive directory and calculated the mean and
standard deviation of each parameter over the intervals defined by the
injection peak-picking algorithm. Statistical analyses were then performed
using the open-source statistical software, R (R Core Team, 2017).
Briefly, the data were fit to a series of multivariate linear models using
the “lm” function from base R, and the fit was evaluated in terms of the
residual standard error (RSE), adjusted R2 and F-test P value. More
details of each analysis are provided in the following sections.
Notation
For all samples, the relative abundances of the heavy and light
isotopologues were expressed with the dimensionless isotope ratios:
Rsample=[1H1H18O]/[1H1H16O]or[1H2H16O]/[1H1H16O].
The isotope ratios were normalized relative to the international standard
VSMOW (Vienna Standard Mean Ocean Water):
δ18Oorδ2H(‰)=(Rsample/RVSMOW-1),
where Rsample and RVSMOW represent the ratios of the abundance of
the heavy and light isotopologues in the samples and international standard,
respectively.
To refer to calibrated δ18O and δ2H values as
determined by IRMS, we use the subscript “true” (i.e., δ18Otrue and δ2Htrue). To refer to uncalibrated
δ18O and δ2H values as determined by CRDS, we use
the subscript “apparent” (i.e., δ18Oapparent and δ2Happarent).
Magnitude of background effects
To visualize the effects of variation in the mixing ratios of
N2/ O2, N2/ Ar, and O2/ Ar on the apparent isotopic
composition of H2O, we plotted the background composition against the
difference between the apparent and true isotopic composition of each sample
(i.e., Δδ18O =δ18Oapparent –
δ18Otrue and Δδ2H =δ2Happarent – δ2Htrue). To quantify the
magnitude of the effects of background variation on the apparent isotopic
composition of H2O, we then fit a series of multivariate linear models.
First, we examined the measurements in the pure background gases. For each
pure background, we described variation in the apparent isotopic composition
of water with a multivariate linear model of the following form:
Y=β+X1⋅β1+X2⋅β2+X3⋅β3+X4⋅β4,
where Y is the apparent isotopic composition of water (δ18Oapparent, δ2Happarent; in units
‰), X1 is the true isotopic composition of water
(δ18Otrue, δ2Htrue; in units
‰), X2 is the water mixing ratio (H2O; in
units %), X3 is the inverse of the water mixing ratio (1/X2),
X4 is the square of the water mixing ratio (X22), and β0, β1, β2, β3, and β4
are the regression coefficients. This functional form provides a good
description of the sensitivity of the apparent isotopic composition of water
to the water mixing ratio in the L2120-i and similar analyzers from the same
manufacturer (Rella, 2010; Rella et al., 2015). Next, we examined the
measurements in the binary mixtures. For each binary mixture, we added an
additional term to capture the effects of background variation:
Y=β+X1⋅β1+X2⋅β2+X3⋅β3+X4⋅β4+X5⋅β5,
where X5 is the mixing ratio of either O2 or Ar (in units %)
and β5 is the corresponding regression coefficient. Finally, we
combined all of the binary mixtures into a composite dataset. For this
composite dataset, we added two terms to capture the effects of background
variation:
Y=β+X1⋅β1+X2⋅β2+X3⋅β3+X4⋅β4+X5⋅β5+X6⋅β6,
where X5 and X6 are the mixing ratios of O2 and Ar (in units
%) and β5 and β6 are the corresponding regression
coefficients.
Geometric basis of background effects
To visualize the geometric basis of the background effects, we plotted the
Δδ18O and Δδ2H values against
three parameters calculated during the second stage of fitting
(“standard_residuals”, “standard_base”,
“standard_slope”), three parameters calculated during the
first stage of fitting and included as fixed values during the second stage
(“h2o_shift”, “h2o_squish_a”,
“h2o_y_eff”), and one parameter calculated
during the first stage of fitting and omitted during the second stage
(“h2o_y_eff_a”). To visualize
the interactions between the background composition and the H2O mixing
ratio, we also plotted each parameter against the H2O mixing ratio
(“h2o_ppmv”). We then formulated a series of semi-mechanistic
models to test which of the seven spectral parameters was the best predictor
of the isotopic error terms. Each model described variation in the apparent
isotopic composition of water as
Y=β+X1⋅β1+X2⋅β2+X3⋅β3+X4⋅β4+X7⋅β7,
where X7 is one of the seven spectral parameters and β7 is
the corresponding regression coefficient. Note that this expression is analogous
to Eq. (5) with the exception that one of the spectral parameters (X7)
has been substituted for the mixing ratios of O2 and Ar (X5 and
X6).
Prediction of background effects in a ternary mixture
We used the empirical model described by Eq. (5) and the semi-mechanistic
model described by Eq. (6) derived from the measurements of the standards in
binary mixtures to predict the apparent isotopic composition of the
standards within the ternary gas mixture. To evaluate how water vapor
self-broadening vs. background-broadening affected model performance, we
assessed model skill across the entire range of water vapor mixing ratios
(i.e., n= 240 analyses for 2500 ppmv ≤ H2O ≤ 35 000 ppmv), as
well as within a restricted subset of intermediate-range water vapor mixing
ratios (i.e., n= 116 analyses for 10 000 ppmv ≤ H2O ≤ 25 000 ppmv). To provide a benchmark for evaluating the 1σ precision of
each empirical and semi-mechanistic model, we calculated the long-term
1σ precision of the L2120-i analyzer using an independent dataset
comprised of previous measurements of the same set of standards, across the
same range of water mixing ratios, and in the same type of ultra high-purity
air that was used to test the two models.
Empirical models of the sensitivity of δ18O
and δ2H values to H2O in pure N2, O2, and Ar.
For significant predictors, coefficient estimates are given for β2, β3, and β4 in Eq. (3); see text. Overall
model fit is summarized with the residual standard error (RSE), adjusted
R2, and P value. The abbreviation n.s. means not significant.
Relationship between the spectral residuals, Δδ18O, and Δδ2H values.
For each panel, n= 330 measurements of four liquid standards across a range of injection
volumes (i.e., 400–2400 nL, in 11 steps of 200 nL each). Points
represent mean values ± SD for replicates of each standard at each
injection volume.
ResultsMagnitude of background effects
Across the N2/ O2, N2/ Ar, and O2/ Ar mixing experiments,
the average differences between the apparent and true isotopic composition
of the standards are -18.45 ± 20.02 ‰ for Δδ18O values and 24.5 ± 17.1
for Δδ2H values (i.e., for n= 990; Fig. 2). The variation in Δδ18O and Δδ2H values is partially due to
the variation in the mixing ratio of H2O, and partially due to the
variation in the mixing ratios of N2/ O2, N2/ Ar, and
O2/ Ar. For δ18O values, the range of H2O mixing
ratios that was evaluated has effects of smaller magnitude than the ranges
of N2/ O2, N2/ Ar, and O2/ Ar mixing ratios that were
evaluated (Fig. 2a–c). For δ2H values, both factors have
effects of similar magnitude (Fig. 2d–f).
Empirical models of the sensitivity of δ18O
and δ2H values to each of the binary mixtures of N2,
O2, and Ar. For significant predictors, coefficient estimates are given
for β5 in Eq. (4); see text. Overall model fit is summarized with
the residual standard error (RSE), adjusted R2, and P value.
Within the subsets of measurements made in pure N2, O2, and Ar,
Eq. (3) accounts for the effects of the H2O mixing ratio with overall
precision ranging between 0.37 and 1.16 ‰ for δ18O values and between 2.3 and 3.7 for δ2H values (Table 1). The
structure of the best-fit models varies between isotopologues and between
backgrounds, with all three of the H2O mixing ratio-dependent
parameters significant in some cases and none significant in others (Table 1). Within the models where the coefficient describing the first-order
response to the H2O mixing ratio, β2, has significant
explanatory power, it tends to have a negative sign for δ18O
values (Table 1; models 1 and 3) and a positive sign for δ2H
values (Table 1; Model 4).
Within each of the binary mixtures, Eq. (4) accounts for the combined
effects of the H2O mixing ratio and the N2/ O2, N2/ Ar,
and O2/ Ar mixing ratios with overall precision ranging between
0.33 and 0.62 ‰ for δ18O values and between 3.2 and 5.2 for
δ2H values (Table 2). Within these models, the coefficient
describing the first-order response to the O2 or Ar mixing ratio,
β5, also tends to have a negative sign for δ18O
values (Table 2; models 1, 3, 5) and a positive sign for δ2H
values (Table 2; models 2, 4, 6). For both δ18O and δ2H values, the magnitude of β5 in the O2/ Ar
experiment is equivalent to the difference in the magnitude of β5 in the N2/ O2 experiment versus in the N2/ Ar
experiment (Table 2).
Empirical models of the composite sensitivity of δ18O and δ2H values to all of the binary mixtures of
N2, O2, and Ar. For significant predictors, coefficient estimates
are given for β5 and β6 in Eq. (5); see text. Overall
model fit is summarized with the residual standard error (RSE), adjusted
R2, and P value.
When all three experiments are combined into a single dataset, Eq. (5)
accounts for the combined effects of the H2O mixing ratio and the
N2/ O2, N2/ Ar, and O2/ Ar mixing ratios with overall
precision of 0.62 ‰ for δ18O values and 3.6 ‰ for δ2H values (Table 3). Within these
models, the coefficients describing the first-order response to the O2
and Ar mixing ratios, β5 and β6, have negative signs
for δ18O values (Table 3; Model 1) and positive signs for
δ2H values (Table 3; Model 2). For both δ18O and
δ2H values, the sensitivity to Ar is relatively higher than the
sensitivity to O2 (Table 3; models 1 and 2). In absolute terms, the
apparent δ18O values deviate from true values by -0.50 ± 0.001 ‰ O2 %-1
and -0.57 ± 0.001 ‰ Ar %-1, respectively (Table 3; Model 1).
The apparent δ2H values deviate from true values by 0.26 ± 0.004 ‰ O2 %-1 and
0.42 ± 0.004 ‰ Ar %-1, respectively (Table 3; Model 2).
Relationship between the spectral baseline, Δδ18O, and Δδ2H values. For each panel, n= 330 measurements of four liquid standards across a range of injection
volumes (i.e., 400–2400 nL, in 11 steps of 200 nL each). Points
represent mean values ± SD for replicates of each standard at each
injection volume.
Geometric basis of background effects
Overall, the relationships between the isotopic error terms and the spectral
parameters have two shared features. First, the background composition has
consistent effects across the three experiments. For each spectral fitting
parameter, the patterns observed in the O2/ Ar experiment are equivalent
to the difference in the patterns observed in the N2/ O2 experiment
versus in the N2/ Ar experiment (i.e., for δ18O values,
compare differences between panels (a) and (b) to (c) in Figs. 3–9; for
δ2H values, compare differences between panels (d) and (e) to
(f) in Figs. 3–9). Second, for those spectral parameters that have
significant linear relationships with the isotopic error terms, the relative
sensitivities of the two isotopologues to any given spectral parameter
always have opposing signs. The apparent δ18O values become
more depleted with decreases in the absolute value of the baseline (Fig. 4a–c), increases in the slope of the baseline (Fig. 5a–c), increases
in the frequency scale correction parameter (Fig. 7a–c), and decreases
in the free linewidth parameter (Fig. 9a–c). In contrast, the apparent
δ2H values become more enriched with the analogous changes in
those parameters (Figs. 3–9d–f).
Relationship between the baseline slope, Δδ18O, and Δδ2H values. For each panel, n= 330
measurements of four liquid standards across a range of injection volumes
(i.e., 400–2400 nL, in 11 steps of 200 nL each). Points represent mean
values ± SD for replicates of each standard at each injection
volume.
Beyond these shared features, there is substantial variation between the
spectral parameters in terms of the complexity of their relationships to the
isotopic error terms. The spectral residuals do not have a linear
relationship with the isotopic error terms: although maximum values of the
residuals are usually associated with maximum values of the isotopic error
terms, minimum values of the residuals are associated with the full range of
values of the isotopic error terms (Fig. 3a–f). The absolute values of
the baseline and the baseline slope each have significant linear
relationships with the isotopic error terms but also have large amounts of
variation that are not directly related to the isotopic error terms (Figs. 4–5a–f). The frequency shift parameter does not have significant
relationships with the isotopic error terms (Fig. 6a–f), but the
frequency scale correction parameter does have significant linear
relationships with the isotopic error terms (Fig. 7a–f). Analogously,
the fixed linewidth parameter does not have significant relationships with
the isotopic error terms (Fig. 8a–f), but the free linewidth parameter
does have significant relationships with the isotopic error terms (Fig. 9a–f).
The complexity of the relationships between the spectral parameters and
isotopic error terms is driven by interactions between the background
composition and the H2O mixing ratios (Figs. 10–11). For the spectral
residuals, baseline, and baseline slope, there is a multiplicative
interaction between the background composition and H2O mixing ratio: at
the lowest H2O mixing ratios, variation in background composition has
the smallest effects on the spectral residuals, baseline, and baseline
slope; at the highest H2O mixing ratios, the opposite is true (Fig. 10a–c). For the frequency scale correction and free linewidth parameters,
there is an additive interaction between the background composition and
H2O mixing ratio: regardless of the H2O mixing ratio, variation in
background composition has similar effects on the frequency scale correction
and free linewidth parameters (Fig. 11b, d). For the frequency shift and
fixed linewidth parameters, there is no interaction between the background
composition and H2O mixing ratio: both parameters vary with the
H2O mixing ratio, but those relationships are insensitive to the
background composition (Fig. 11a, c).
Relationship between the frequency shift, Δδ18O, and Δδ2H values. For each panel,
n= 330 measurements of four liquid standards across a range of injection
volumes (i.e., 400–2400 nL, in 11 steps of 200 nL each). Points
represent mean values ± SD for replicates of each standard at each
injection volume.
Relationship between the frequency scale correction,
Δδ18O, and Δδ2H values. For each
panel, n= 330 measurements of four liquid standards across a range of
injection volumes (i.e., 400–2400 nL, in 11 steps of 200 nL each).
Points represent mean values ± SD for replicates of each standard at
each injection volume.
Semi-mechanistic models of the composite sensitivity of
δ18O and δ2H values to all of the binary mixtures
of N2, O2, and Ar. For significant predictors, coefficient
estimates are given for β7 in Eq. (6); see text. Overall model
fit is summarized with the residual standard error (RSE), adjusted R2,
and P value.
In the semi-mechanistic models (Eq. 6), the free linewidth parameter is a
better predictor than any of the other spectral parameters (Table 4). The
models based on the free linewidth parameter account for the combined
effects of the H2O mixing ratio and the N2/ O2, N2/ Ar,
and O2/ Ar mixing ratios with overall precision of 2.54 ‰ for δ18O values and 4.4 ‰
for δ2H values (Table 4). In these
models, the coefficient describing the first-order response to the free
linewidth parameter, β7, has a positive sign for δ18O values (Table 4; Model 7) and a negative sign for δ2H
values (Table 4; Model 14). The second-best predictor is the frequency scale
correction parameter (Table 4). The models based on the frequency scale
correction parameter have overall precision of 6.72 ‰
for δ18O values and 4.9 ‰ for δ2H values (Table 4). For this parameter, the β7
coefficient has a negative sign for δ18O values (Table 4; Model
6) and a positive sign for δ2H values (Table 4; Model 13).
Comparison of empirical and semi-mechanistic models for
predicting δ18O and δ2H values in a ternary
mixture of N2, O2, and Ar. Overall model fit is summarized with
the residual standard error (RSE), adjusted R2, and P value.
Relationship between the fixed linewidth parameter,
Δδ18O, and Δδ2H values. For each
panel, n= 330 measurements of four liquid standards across a range of
injection volumes (i.e., 400–2400 nL, in 11 steps of 200 nL each).
Points represent mean values ± SD for replicates of each standard at
each injection volume.
Prediction of background effects in a ternary mixture
For δ18O, the empirical model predicts the apparent δ18O values with a 1σ
precision of ±0.99 ‰, whereas the semi-mechanistic model predicts the
apparent δ18O values with a 1σ precision of ±1.68 ‰ (n= 240; Table 5). When the test measurements
in the ternary gas mixture are restricted to intermediate mixing ratios in
the range of 10 000–25 000 ppmv H2O, these values improve to ±0.80 and ±1.21 ‰,
respectively (n= 116; Table 5). For δ2H, the empirical model
predicts the apparent δ2H values with a 1σ precision of
±3.1 ‰, whereas the semi-mechanistic model
predicts the apparent δ2H values with a 1σ precision of
±3.0 ‰ (Table 5). For intermediate mixing ratios
in the range of 10 000–25 000 ppmv H2O, these values improve to ±2.0 and ±2.1 ‰,
respectively (n= 116; Table 5). As a benchmark for comparison, the
average long-term 1σ precision of this L2120-i analyzer is ±0.24 ‰ for δ18O values and ±1.4 ‰
for δ2H values across the range of mixing
ratios from 2500 to 35 000 ppmv H2O.
Relationship between the free linewidth parameter,
Δδ18O, and Δδ2H values. For each
panel, n= 330 measurements of four liquid standards across a range of
injection volumes (i.e., 400–2400 nL, in 11 steps of 200 nL each).
Points represent mean values ± SD for replicates of each standard at
each injection volume.
DiscussionWhat are the magnitudes of the effects of variation in the mixing ratio
of N2/ O2, N2/ Ar, and O2/ Ar on the apparent δ18O–H2O and δ2H–H2O values measured by the
L2120-i CRDS analyzer?
Across the range of backgrounds considered in this study, variation in the
N2/ O2, N2/ Ar, and O2/ Ar ratios has substantial effects
on the apparent isotopic composition of water reported by the L2120-i (Fig. 2). For δ18O, combining the long-term 1σ precision of
the analyzer (±0.24 ‰) and the magnitude of the
sensitivities to O2 and Ar relative to N2 (i.e., -0.50 ± 0.001 ‰ O2 %-1 and -0.57 ± 0.001 ‰
Ar %-1; Table 3) implies that variation
over the thresholds of ±0.48 % O2 or ±0.42 % Ar is expected to result in detectable oxygen isotope errors. For
δ2H, combining the long-term 1σ precision of the
analyzer (±1.4 ‰) and the magnitude of the
sensitivities to O2 and Ar relative to N2 (i.e., 0.26 ± 0.004 ‰ O2 %-1 and
0.42 ± 0.004 ‰ Ar %-1; Table 3) implies that variation over
the thresholds of ±5.4 % O2 or ±3.3 % Ar
is expected to result in detectable hydrogen isotope errors.
Relationships between H2O mixing ratio and the
spectral residuals, baseline, and baseline slope parameters. For each panel,
n= 990 measurements of four liquid standards across a range of injection
volumes (i.e., 400–2400 nL, in 11 steps of 200 nL each). Points
represent mean values ± SD for replicates of each standard at each
injection volume.
The only previous measurements available for direct comparison to these
results are those of Gralher et al. (2016). In binary
N2/ O2 mixtures, Gralher et al. (2016) found that a
different L2120-i analyzer exhibited a sensitivity of -0.56 ‰ O2 %-1 for δ18O values and a
sensitivity of 0.42 ‰ O2 %-1 for δ2H values (i.e., see Fig. 2 in that reference). Overall, the Gralher
et al. (2016) values are more similar to our binary N2/ Ar
sensitivities (i.e., for δ18O–H2O, -0.56 ± 0.001 ‰ Ar %-1; for δ2H–H2O,
0.43 ± 0.005 ‰ Ar %-1; Table 2) than our binary
N2/ O2 sensitivities (i.e., for δ18O–H2O, -0.50 ± 0.001 ‰ O2 %-1; for
δ2H–H2O, 0.26 ± 0.009 ‰ O2 %-1; Table 2). This is unexpected, and the responsible mechanisms
are not entirely clear.
Since the Gralher et al. (2016) sensitivities were derived from
measurements across narrow ranges of H2O mixing ratios (i.e.,
∼ 17 000 ppmv) and O2 mixing ratios (i.e., 0–20 %), one
possible explanation is that the wider ranges used in our study could be
responsible for the different sensitivity estimates. Subsetting our dataset
to the range of H2O mixing ratios used by Gralher et al. (2016)
yields N2/ O2 sensitivities equivalent to those reported in Table 2. However, subsetting our dataset to the range of O2 mixing ratios
used by Gralher et al. (2016) does yield N2/ O2
sensitivities in much closer agreement with those authors' results (i.e.,
for δ18O, -0.54 ± 0.003 ‰
O2 %-1; for δ2H, 0.38 ± 0.002 ‰ O2 %-1). This suggests that the
differences in the sensitivity estimates are derived from the different
ranges of N2/ O2 mixing ratios used in the two studies.
How are the background effects on the apparent δ18O–H2O and δ2H–H2O values derived from the
interaction between the target spectra and the spectral acquisition and
analysis strategies in this instrument?
Relationships between H2O mixing ratio and the
frequency shift, frequency scale correction, fixed linewidth, and free
linewidth parameters. For each panel, n= 990 measurements of four liquid
standards across a range of injection volumes (i.e., 400–2400 nL, in 11
steps of 200 nL each). Points represent mean values ± SD for
replicates of each standard at each injection volume.
Direct and indirect effects of background gas
composition on peak amplitude determination. Panels (a) and (b) plot
theoretical spectrograms illustrating how isolated absorption features are
directly broadened by air (solid line) versus N2 (dashed line) or
O2 and Ar (dotted line). Panel (c) plots simulated spectrograms
illustrating how the baseline of the three H2O lines targeted by the
L2120-i is indirectly affected by the strong neighboring lines at lower
frequencies. Simulations were performed using spectraplot.com (Goldenstein
et al., 2017a) with HITRAN/HITEMP data and the following parameters: T= 80 ∘C, P= 35 torr, and L= 1 cm. The thick line is a
simulation at 2.5 % H2O, and the thin line is a simulation at 0.5 % H2O.
Overall, the strongest direct effect of the background gas composition is on
the effective linewidth of the target absorption features. In pure N2
backgrounds, the actual value of the effective linewidth is greater than the
value prescribed for air backgrounds. The spectral analysis algorithm
attempts to compensate for this mis-specification of the peak shape by
decreasing the frequency scale correction parameter (i.e.,
h2o_squish_a), but the peak shape
mis-specification persists in spite of the frequency scale adjustments
(i.e., h2o_y_eff_a > h2o_y_eff). As a result, the amplitudes of the
absorption peaks are systematically overestimated (Fig. 12a). The degree
of overestimation increases in the order 1H2H16O < 1H1H16O < 1H1H18O, with the result
that the Δδ18O values are positive and the Δδ2H values are negative. In pure O2 and Ar backgrounds,
the actual value of the effective linewidth is less than the value
prescribed for air backgrounds. Here, the spectral analysis algorithm
attempts to compensate by increasing the frequency scale correction
parameter (i.e., h2o_squish_a), but the peak
shape mis-specification again persists (i.e., h2o_y_eff_a < h2o_y_eff). In these backgrounds, the amplitudes of the
absorption peaks are systematically underestimated (Fig. 12b). The
degree of underestimation increases in the order 1H2H16O
< 1H1H16O < 1H1H18O, such that
the Δδ18O values are negative and the Δδ2H values are positive. On the one hand, the tendency for the
absorption spectrum to be broader in N2, intermediate in O2, and
narrower in Ar is entirely consistent with the normal behavior of isolated
water vapor absorption lines
(Buldyreva et al., 2011). On the other
hand, it is not entirely clear why the 1H2H16O,
1H1H16O, and 1H1H18O lines exhibit increasing
susceptibility to peak shape mis-specification.
One possible explanation is that the differential errors are derived from
indirect effects of the background gas. The three water vapor absorption
lines that are targeted by the L2120-i are all characterized by relatively
low line strengths, but they sit on the upper “wings” of lower-frequency
water vapor absorption lines that are characterized by much higher
line strengths (i.e., at ν0= 7181.156, 7182.209, and 7182.950 cm-1; (Lisak et al., 2009); Fig. 12c). The broadening,
narrowing, and shifting of these strong off-screen lines appear to be the
major control on variation in the spectral residuals, spectral baseline, and
baseline slope parameters. Specifically, N2-induced increases in the
width of the offscreen lines seem to decrease the baseline slope and
increase the absolute value of the baseline, whereas O2- and Ar-induced
decreases in the width of the offscreen lines seem to increase the baseline
slope and decrease the absolute value of the baseline. Since the target
1H1H18O line is at a lower frequency than the target
1H1H16O line, and the target 1H1H16O line is
in turn at a lower frequency than the target 1H2H16O line,
the baseline perturbations from the off-screen lines increase in the same
rank order as, and could be responsible for, the differential peak shape
mis-specifications (i.e., 1H2H16O < 1H1H16O < 1H1H18O).
However, proximity to the off-screen features is not the only possible
explanation for the differential peak shape mis-specifications. For example,
the 1H2H16O, 1H1H16O, and
1H1H18O lines also vary in line strength in the order
1H2H16O < 1H1H16O < 1H1H18O (Tennyson et al.,
2009, 2010, 2013). As a result, a second
possible explanation is that the differential susceptibility to peak shape
mis-specification is primarily a function of line strength. This
interpretation is supported by the fact that the largest negative Δδ18O errors occur in the most δ18O-enriched
standard (i.e., where the difference in amplitude between the
1H1H16O and 1H1H18O lines is maximized),
whereas the largest positive Δδ2H errors occur in the
most δ2H-depleted standard (i.e., where the difference in
amplitude between 1H2H16O and 1H1H16O is
maximized). Nonetheless, it could also be the case that the differential
susceptibility to peak shape mis-specification is a function of a
combination of several of the above mechanisms, or other undefined
mechanisms. To definitively distinguish among these possibilities, it would
be necessary to have accurate measurements of the broadening, narrowing, and
shifting coefficients for each of the individual lines in the target
spectrum rather than the “effective” coefficients that the L2120-i
calculates for the composite spectrum.
Is it practicable to develop post hoc calibrations for this instrument that
accurately account for the effects of background variation in N2,
O2, and/or Ar on the apparent δ18O–H2O and δ2H–H2O values?
On the one hand, the majority of the background-induced isotope artifacts
can be corrected with either simple empirical or semi-mechanistic models
(Table 5). The success of both types of models is likely a reflection of the
fact that the collisional broadening, narrowing and shifting coefficients of
any given absorption line in a mixed background can all be satisfactorily
described as linear combinations of the corresponding coefficients in pure
backgrounds (Buldyreva et al., 2011).
On the other hand, neither type of model is capable of completely correcting
the isotopic artifacts to within the inherent instrument precision (Table 5). Although the loss of precision for δ2H values is similar
for the semi-mechanistic and empirical corrections, the loss of precision
for δ18O values is slightly greater for the semi-mechanistic
corrections than for the empirical corrections. In combination, these
findings indicate that there are several feasible approaches for post hoc
calibrations of CRDS measurements that accurately account for background
variation in N2, O2, and/or Ar but that all currently tradeoff
with measurement precision. Since the precision of the δ18O and
δ2H measurements in turn controls the precision of the derived
deuterium excess parameter (i.e., d-excess =δ2H -8⋅δ18O), this has important implications for the range of
strategies that can be used to calibrate CRDS analyzers for δ18O, δ2H, and d-excess measurements. Different types of
strategies are likely to be required for measurements (i) in the
atmosphere and (ii) in other settings.
For atmospheric applications, there are likely to be systematic inaccuracies
in δ18O and d-excess values if “synthetic air” is used as a
calibration background without accounting for the fact that these
N2/ O2 mixtures lack Ar and may exhibit cylinder-to-cylinder
variation in N2 versus O2 content. For example, these effects may
explain the cylinder-to-cylinder calibration shifts observed when Air
Liquide's “ALPHAGAZ 1” has been used as a calibration background for
atmospheric observations (e.g., see Aemisegger et al., 2012 and Casado
et al., 2016). To address this issue, previous studies have recommended
performing CRDS calibrations for atmospheric observations in natural air
backgrounds (Chen et al., 2010; Aemisegger et al., 2012;
Nara et al., 2012; Long et al., 2013). The results of the current study
corroborate this approach but indicate that it represents only one of two
alternatives. The other approach is performing calibration measurements in a
background that does not conform to atmospheric composition and using
sensitivity experiments of the sort reported here to develop transfer
functions that translate between the calibration and observation backgrounds
(i.e., similar to Eq. 5). Despite its relatively lower precision, this
approach may nonetheless represent the preferred strategy for applications
where it is difficult or impossible to obtain sufficiently purified natural
air for calibration.
For marine, freshwater, and soil applications, there are likely to be
systematic inaccuracies in δ18O, δ2H, and
d-excess values of liquid and vapor samples if the calibration strategy does
not account for dynamic variation in the O2 content of the measurement
background. For example, marine dissolved oxygen levels range from
supersaturated during high-productivity periods in upwelling zones
(Schmidt and Eggert, 2016) to hypoxic during harmful algal blooms in
coastal zones (O'Boyle et al., 2016) and to anoxic in deep water oxygen
minimum zones (Larsen et al., 2016). To address this type of dynamic
variation in background O2 content, Friedrichs et al. (2010)
and Becker et al. (2012) have demonstrated that linewidth
information from CRDS measurements of CO2 and δ13C–CO2 can be used to both detect and correct for
O2-induced errors. The results of the current study indicate that an
analogous approach can be used with the L2120-i (i.e., based on Eq. 6),
although doing so will further reduce the precision of the δ18O
values. It is likely that the spectroscopically based corrections are less
successful in the L2120-i because the 7183–7184 cm-1 region is
congested, and the fitting algorithm does not perform individual fits on the
target H2O isotopologue peaks. In contrast, the EnviroSense 2050
analyzer used by Friedrichs et al. (2010) and Becker et al. (2012) targeted a relatively uncongested spectral region
(6251–6252 cm-1) and performed individual fits on each of the target
CO2 isotopologue peaks.
Looking forward, the most straightforward approach to overcome the tradeoff
between background stability and measurement precision would be to develop
new spectral acquisition and analysis strategies for CRDS measurements that
can accommodate dynamic variation in the composition of the background gas.
Considering that the integrated absorbance of isolated features in CRDS
spectra is expected to be conserved regardless of the degree of broadening,
narrowing, or shifting induced by the background gas (Zalicki and Zare,
1995), the next generation of CRDS analyzers that quantify absorption based
on peak areas may be less sensitive to background variation than those that
quantify absorption based on peak amplitudes (Steig et al., 2014). A
recent report of the insensitivity of off-axis integrated cavity output
spectroscopy (OA-ICOS) to background variation from 1 to 5 % CO2 is
consistent with this idea (Sprenger et al., 2017). However, while
measurements of integrated absorbance may be necessary for limiting
sensitivity to background effects, they are unlikely to be sufficient for
entirely eliminating sensitivity to background effects. To achieve this
objective, it may be useful to introduce spectral fitting strategies in
which all of the lineshape parameters are treated as free rather than fixed
variables. Such “calibration-free” spectral fitting strategies have been
recently been developed for high-temperature and high-pressure applications
in energy research, and these might serve as models for lower-temperature
and lower-pressure applications in environmental research
(Sun et al., 2013; Goldenstein et al., 2014, 2017b;
Sur et al., 2015).
Conclusions
Due to the sensitivity of the L2120-i to background gas composition, this
model CRDS analyzer is best suited for applications in which the background
O2 and Ar mixing ratios vary by no more than a maximum of 0.5 % and
ideally by less than 0.1 %. Calibration strategies should be designed to
ensure that if there is any contrast between the background used for
calibration and measurement, it is no greater than this threshold. For
observations or experiments in which the background O2 and Ar mixing
ratios vary by more than 0.5 %, the measurements of the L2120-i will
include systematic errors that are derived from the broadening, narrowing,
and shifting of both the target absorption lines and strong neighboring
lines. If the composition of the variable background is known, the errors
can be accurately corrected with the empirical model described by Eq. (5).
If the composition of the variable background is unknown, the errors can
also be accurately corrected with the semi-mechanistic model described by
Eq. (6). In either case, accuracy and precision will be maximized by
calculating the coefficients for Eq. (5) or (6) from a calibration dataset
that encompasses the full range of variation in N2, O2, and/or Ar
mixing ratios, H2O mixing ratios, and δ18O and δ2H values within the unknown observations. Since neither of the post hoc
correction approaches optimize precision, new strategies for dynamically
detecting and accommodating background variation in N2, O2, and/or
Ar are needed in order to capitalize on the possibilities of CRDS
measurements in variable backgrounds.
The data generated in this study are publicly available and can be
downloaded from the Open Science Framework
(Johnson and Rella, 2017; 10.17605/OSF.IO/C7NSG).
JEJ and CWR designed the experiments and JEJ
carried them out. Both authors analyzed the results and JEJ
prepared the manuscript with contributions from CWR.
Jennifer E. Johnson declares that she has no conflict of interest. Chris W. Rella is an employee
of Picarro, Inc.
Acknowledgements
We are grateful to Russell K. Monson (University of Arizona) for encouragement to pursue
these experiments; to Joseph A. Berry (Carnegie Institution) and Christopher B. Field
(Carnegie Institution) for loaning the Picarro L2120-i; to Chris Redondo (University of
Arizona), Chris Burkhart (Air Liquide), Scott Martinez (Air Liquide), and Sam Owens
(Air Liquide) for assistance obtaining compressed gases; and to David Dettman
(University of Arizona), Francina Dominguez (University of Illinois), Chris Eastoe (University of Arizona),
Steve Leavitt (University of Arizona), and Kees Welten (University of California-Berkeley) for
assistance obtaining and calibrating liquid standards. This study was
completed with support from the National Science Foundation through the
Macrosystem Biology Program award no. 1065790 (Russell K. Monson) and the Major
Research Infrastructure Program award no. 1040106 (Christopher B. Field and Joseph A. Berry).
Edited by: William Ward
Reviewed by: Franziska Aemisegger and one anonymous referee
ReferencesAemisegger, F., Sturm, P., Graf, P., Sodemann, H., Pfahl, S., Knohl, A., and Wernli, H.: Measuring variations of δ2H and δ18O
in atmospheric water vapour using two commercial laser-based spectrometers: an instrument characterisation study,
Atmos. Meas. Tech., 5, 1491–1511, 10.5194/amt-5-1491-2012, 2012.Becker, M., Andersen, N., Fiedler, B., Fietzek, P., Körtzinger, A.,
Steinhoff, T., and Friedrichs, G.: Using cavity ringdown spectroscopy for
continuous monitoring of δ13C(CO2) and fCO2 in the
surface ocean, Limnol. Oceanogr.-Meth., 10, 752–766,
10.4319/lom.2012.10.752, 2012.
Berkelhammer, M., Hu, J., Bailey, A., Noone, D., J, S. C., Barnard, H.,
Gochis, D., Hsiao, G., Rahn, T., and Turnipseed, A.: The nocturnal water
cycle in an open-canopy forest, J. Geophys. Res.-Atmos., 118, 10225–10242,
2013.Brand, W. A., Geilmann, H., Crosson, E. R., and Rella, C. W.: Cavity
ring-down spectroscopy versus high-temperature conversion isotope ratio mass
spectrometry; a case study on δ2H and δ18O of pure
water samples and alcohol/water mixtures, Rapid Commun. Mass Spectrom., 23,
1879–1884, 10.1002/rcm.4083, 2009.Brewer, P. J., Brown, R. J. C., Miller, M. N., Miñarro, M. D., Murugan,
A., Milton, M. J. T., and Rhoderick, G. C.: Preparation and Validation of
Fully Synthetic Standard Gas Mixtures with Atmospheric Isotopic Composition
for Global CO2 and CH4 Monitoring, Anal. Chem., 86, 1887–1893,
10.1021/ac403982m, 2014.
Buldyreva, J., Lavrentieva, N., and Starikov, V.: Collisional line broadening
and shifting of atmospheric gases: a practical guide for line shape
modelling by current semi-classical approaches, Imperial College Press,
London, 2011.Casado, M., Landais, A., Masson-Delmotte, V., Genthon, C., Kerstel, E., Kassi, S., Arnaud, L., Picard, G., Prie, F., Cattani, O.,
Steen-Larsen, H.-C., Vignon, E., and Cermak, P.: Continuous measurements of isotopic composition of water vapour on
the East Antarctic Plateau, Atmos. Chem. Phys., 16, 8521–8538, 10.5194/acp-16-8521-2016, 2016.Chen, H., Winderlich, J., Gerbig, C., Hoefer, A., Rella, C. W., Crosson, E. R., Van Pelt, A. D., Steinbach, J., Kolle, O., Beck, V.,
Daube, B. C., Gottlieb, E. W., Chow, V. Y., Santoni, G. W., and Wofsy, S. C.: High-accuracy continuous airborne measurements of
greenhouse gases (CO2 and CH4) using the cavity ring-down spectroscopy (CRDS) technique, Atmos. Meas. Tech., 3, 375–386, 10.5194/amt-3-375-2010, 2010.Crosson, E. R.: A cavity ring-down analyzer for measuring atmospheric levels
of methane, carbon dioxide, and water vapor, Appl. Phys. B, 92, 403–408,
10.1007/s00340-008-3135-y, 2008.
Demtröder, W.: Laser Spectroscopy 1: Basic Principles, 5 Edn., Springer,
Berlin, 2014.Flores, E., Rhoderick, G. C., Viallon, J., Moussay, P., Choteau, T.,
Gameson, L., Guenther, F. R., and Wielgosz, R. I.: Methane Standards Made in
Whole and Synthetic Air Compared by Cavity Ring Down Spectroscopy and Gas
Chromatography with Flame Ionization Detection for Atmospheric Monitoring
Applications, Anal. Chem., 87, 3272–3279, 10.1021/ac5043076, 2015.Friedrichs, G., Bock, J., Temps, F., Fietzek, P., Körtzinger, A., and
Wallace, D. W. R.: Toward continuous monitoring of seawater
13CO2/12CO2 isotope ratio and pCO2: Performance of
cavity ringdown spectroscopy and gas matrix effects, Limnol. Oceanogr.-Meth., 8, 539–551, 10.4319/lom.2010.8.539, 2010.Goldenstein, C. S., Strand, C. L., Schultz, I. A., Sun, K., Jeffries, J. B.,
and Hanson, R. K.: Fitting of calibration-free scanned-wavelength-modulation
spectroscopy spectra for determination of gas properties and absorption
lineshapes, Appl. Opt., 53, 356–12, 10.1364/AO.53.000356, 2014.Goldenstein, C. S., Miller, V. A., Spearrin, R. M., and Strand, C. L.:
SpectraPlot.com: Integrated Spectroscopic Modeling of Atomic and Molecular
Gases, J. Quant. Spectrosc. Ra., 200, 249–257,
10.1016/j.jqsrt.2017.06.007, 2017a.Goldenstein, C. S., Spearrin, R. M., Jeffries, J. B., and Hanson, R. K.:
Infrared laser-absorption sensing for combustion gases, Prog. Energ. Combust., 60, 132–176, 10.1016/j.pecs.2016.12.002, 2017b.Gralher, B., Herbstritt, B., Weiler, M., Wassenaar, L. I., and Stumpp, C.:
Correcting Laser-Based Water Stable Isotope Readings Biased by Carrier Gas
Changes, Environ. Sci. Technol., 50, 7074–7081,
10.1021/acs.est.6b01124, 2016.Gupta, P., Noone, D., Galewsky, J., Sweeney, C., and Vaughn, B. H.:
Demonstration of high-precision continuous measurements of water vapor
isotopologues in laboratory and remote field deployments using
wavelength-scanned cavity ring-down spectroscopy (WS-CRDS) technology, Rapid
Commun. Mass Spectrom., 23, 2534–2542, 10.1002/rcm.4100, 2009.
Hanson, R. K., Spearrin, R. M., and Goldenstein, C. S.: Spectroscopy and
Optical Diagnostics for Gases, Springer, New York, 2015.Hendry, M. J., Richman, B., and Wassenaar, L. I.: Correcting for Methane
Interferences on δ2H and δ18O Measurements in Pore
Water Using H2O(liquid)–H2O(vapor) Equilibration Laser
Spectroscopy, Anal. Chem., 83, 5789–5796, 10.1021/ac201341p, 2011.Herbstritt, B., Gralher, B., and Weiler, M.: Continuous in situ measurements
of stable isotopes in liquid water, Water Resour. Res., 48, W03601,
10.1029/2011WR011369, 2012.Hodges, J. T. and Lisak, D.: Frequency-stabilized cavity ring-down
spectrometer for high-sensitivity measurements of water vapor concentration,
Appl. Phys. B, 85, 375–382, 10.1007/s00340-006-2411-y, 2006.Johnson, J. E. and Rella, C. W.: Dataset for “Effects of variation in background mixing ratios
of N2, O2, and Ar on the measurement of δ18O-H2O and δ2H-H2O values by cavity
ring-down spectroscopy”, 10.17605/OSF.IO/C7NSG, 2017.Johnson, J. E., Hamann, L., Dettman, D. L., Kim-Hak, D., Leavitt, S. W.,
Monson, R. K., and Papuga, S. A.: Performance of induction module-cavity
ring-down spectroscopy (IM-CRDS) for measuring δ18O and δ2H values of soil, stem, and leaf waters, Rapid Commun. Mass
Spectrom., 31, 547–560, 10.1002/rcm.7813, 2017.Larsen, M., Lehner, P., Borisov, S. M., Klimant, I., Fischer, J. P.,
Stewart, F. J., Canfield, D. E., and Glud, R. N.: In situ quantification of
ultra-low O2 concentrations in oxygen minimum zones: Application of
novel optodes, Limnol. Oceanogr.-Meth., 14, 784–800,
10.1002/lom3.10126, 2016.Lazarus, B. E., Germino, M. J., and Vander Veen, J. L.: Online induction
heating for determination of isotope composition of woody stem water with
laser spectrometry: a methods assessment, Isot. Environ.
Healt. S., 52, 309–325, 10.1080/10256016.2016.1141205, 2016.Lisak, D., Havey, D. K., and Hodges, J. T.: Spectroscopic line parameters of
water vapor for rotation-vibration transitions near 7180 cm-1, Phys.
Rev. A, 79, 052507, 10.1103/PhysRevA.79.052507, 2009.Long, D. A., Gameson, L., Truong, G. W., Bielska, K., Cygan, A., Hodges, J.
T., Whetstone, J. R., and van Zee, R. D.: The Effects of Variations in Buffer
Gas Mixing Ratios on Commercial Carbon Dioxide Cavity Ring-Down Spectroscopy
Sensors, J. Atmos. Ocean. Tech., 30, 2604–2609,
10.1175/JTECH-D-13-00039.1, 2013.Martín-Gómez, P., Barbeta, A., Voltas, J., Penuelas, J., Dennis,
K., Palacio, S., Dawson, T. E., and Ferrio, J. P.: Isotope-ratio infrared
spectroscopy: a reliable tool for the investigation of plant-water sources?
New Phytol., 207, 914–927, 10.1111/nph.13376, 2015.Munksgaard, N. C., Wurster, C. M., and Bird, M. I.: Continuous analysis of
δ18O and δD values of water by diffusion sampling
cavity ring-down spectrometry: a novel sampling device for unattended field
monitoring of precipitation, ground and surface waters, Rapid Commun. Mass
Spectrom., 25, 3706–3712, 10.1002/rcm.5282, 2011.Munksgaard, N. C., Wurster, C. M., Bass, A., Zagorskis, I., and Bird, M. I.:
First continuous shipboard δ18O and δD measurements in
sea water by diffusion sampling—cavity ring-down spectrometry, Environ.
Chem. Lett., 10, 301–307, 10.1007/s10311-012-0371-5, 2012.Nara, H., Tanimoto, H., Tohjima, Y., Mukai, H., Nojiri, Y., Katsumata, K., and Rella, C. W.: Effect of air composition (N2, O2, Ar, and H2O)
on CO2 and CH4 measurement by wavelength-scanned cavity ring-down spectroscopy: calibration and measurement strategy,
Atmos. Meas. Tech., 5, 2689–2701, 10.5194/amt-5-2689-2012, 2012.O'Boyle, S., McDermott, G., Silke, J., and Cusack, C.: Potential impact of an
exceptional bloom of Karenia mikimotoi on dissolved oxygen levels in waters off western
Ireland, Harmful Algae, 53, 77–85, 10.1016/j.hal.2015.11.014, 2016.O'Keefe, A. and Deacon, D.: Cavity ring-down optical spectrometer for
absorption measurements using pulsed laser sources, Rev. Sci.
Instrum., 59, 2544–2551, 10.1063/1.1139895, 1988.R Core Team: A language and environment for statistical computing, R
Foundation for Statistical Computing, Vienna, Austria, available at:
http://www.R-project.org/, last access: 18 August 2017.Rella, C. W.: Accurate Greenhouse Gas Measurements in Humid Gas Streams
Using the Picarro G1301 Carbon Dioxide/Methane/Water Vapor Gas Analyzer,
1–18, available at:
https://www.picarro.com/assets/docs/White_Paper_G1301_Water_Vapor_Correction.pdf
(last access: 18 August 2017), 2010.Rella, C. W., Hoffnagle, J., He, Y., and Tajima, S.: Local- and regional-scale measurements of CH4, δ13CH4, and
C2H6 in the Uintah Basin
using a mobile stable isotope analyzer, Atmos. Meas. Tech., 8, 4539–4559, 10.5194/amt-8-4539-2015, 2015.Schmidt, M. and Eggert, A.: Oxygen cycling in the northern Benguela
Upwelling System: Modelling oxygen sources and sinks, Prog.
Oceanogr., 149, 145–173, 10.1016/j.pocean.2016.09.004, 2016.Schmidt, M., Maseyk, K., Lett, C., Biron, P., Richard, P., Bariac, T., and
Seibt, U.: Concentration effects on laser-based δ18O and
δ2H measurements and implications for the calibration of vapour
measurements with liquid standards, Rapid Commun. Mass Spectrom., 24,
3553–3561, 10.1002/rcm.4813, 2010.Schmidt, M., Maseyk, K., Lett, C., Biron, P., Richard, P., Bariac, T., and
Seibt, U.: Reducing and correcting for contamination of ecosystem water
stable isotopes measured by isotope ratio infrared spectroscopy, Rapid
Commun. Mass Spectrom., 26, 141–153, 10.1002/rcm.5317, 2012.Schultz, N. M., Griffis, T. J., Lee, X., and Baker, J. M.: Identification and
correction of spectral contamination in 2H/1H and
18O/16O measured in leaf, stem, and soil water, Rapid Commun. Mass
Spectrom., 25, 3360–3368, 10.1002/rcm.5236, 2011.Sprenger, M., Tetzlaff, D., and Soulsby, C.: No influence of CO2 on
stable isotope analyses of soil waters with OA-ICOS, Rapid Commun. Mass
Spectrom., 31, 430–436, 10.1002/rcm.7815, 2017.Steen-Larsen, H. C., Sveinbjörnsdottir, A. E., Peters, A. J., Masson-Delmotte, V., Guishard, M. P., Hsiao, G., Jouzel, J., Noone, D.,
Warren, J. K., and White, J. W. C.: Climatic controls on water vapor deuterium excess in the marine boundary layer of the North Atlantic
based on 500 days of in situ, continuous measurements, Atmos. Chem. Phys., 14, 7741–7756, 10.5194/acp-14-7741-2014, 2014.Steig, E. J., Gkinis, V., Schauer, A. J., Schoenemann, S. W., Samek, K., Hoffnagle, J., Dennis, K. J.,
and Tan, S. M.: Calibrated high-precision 17O-excess measurements using cavity ring-down spectroscopy
with laser-current-tuned cavity resonance, Atmos. Meas. Tech., 7, 2421–2435, 10.5194/amt-7-2421-2014, 2014.Sun, K., Chao, X., Sur, R., Goldenstein, C. S., Jeffries, J. B., and Hanson,
R. K.: Analysis of calibration-free wavelength-scanned wavelength modulation
spectroscopy for practical gas sensing using tunable diode lasers, Meas.
Sci. Technol., 24, 125203–125212, 10.1088/0957-0233/24/12/125203, 2013.Sur, R., Sun, K., Jeffries, J. B., Socha, J. G., and Hanson, R. K.:
Scanned-wavelength-modulation-spectroscopy sensor for CO, CO2, CH4
and H2O in a high-pressure engineering-scale transport-reactor coal
gasifier, Fuel, 150, 102–111, 10.1016/j.fuel.2015.02.003, 2015.Tennyson, J., Bernath, P. F., Brown, L. R., Campargue, A., Carleer, M. R.,
Császár, A. G., Gamache, R. R., Hodges, J. T., Jenouvrier, A.,
Naumenko, O. V., Polyansky, O. L., Rothman, L. S., Toth, R. A., Vandaele, A.
C., Zobov, N. F., Daumont, L., Fazliev, A. Z., Furtenbacher, T., Gordon, I.
E., Mikhailenko, S. N., and Shirin, S. V.: IUPAC critical evaluation of the
rotational–vibrational spectra of water vapor, Part I – Energy levels and
transition wavenumbers for H217O and H218O, J. Quant. Spectrosc. Ra., 110, 573–596,
10.1016/j.jqsrt.2009.02.014, 2009.Tennyson, J., Bernath, P. F., Brown, L. R., Campargue, A., Császár,
A. G., Daumont, L., Gamache, R. R., Hodges, J. T., Naumenko, O. V.,
Polyansky, O. L., Rothman, L. S., Toth, R. A., Vandaele, A. C., Zobov, N.
F., Fally, S., Fazliev, A. Z., Furtenbacher, T., Gordon, I. E., Hu, S.-M.,
Mikhailenko, S. N., and Voronin, B. A.: IUPAC critical evaluation of the
rotational–vibrational spectra of water vapor, Part II Energy levels and
transition wavenumbers for HD16O, HD17O, and HD18O, J. Quant. Spectrosc. Ra., 111, 2160–2184,
10.1016/j.jqsrt.2010.06.012, 2010.
Tennyson, J., Bernath, P. F., Brown, L. R., Campargue, A., Császár,
A. G., Daumont, L., Gamache, R. R., Hodges, J. T., Naumenko, O. V.,
Polyansky, O. L., Rothman, L. S., Vandaele, A. C., Zobov, N. F., Derzi, Al,
A. R., Fábri, C., Fazliev, A. Z., Furtenbacher, T., Gordon, I. E., Lodi,
L., and Mizus, I. I.: IUPAC critical evaluation of the
rotational–vibrational spectra of water vapor, Part III: Energy levels and
transition wavenumbers for H216O, J. Quant. Spectrosc. Ra., 117, 29–58,
10.1016/j.jqsrt.2012.10.002, 2013.Volkmann, T. H. M. and Weiler, M.: Continual in situ monitoring of pore
water stable isotopes in the subsurface, Hydrol. Earth Syst. Sci., 18,
1819–1833, 10.5194/hess-18-1819-2014, 2014.West, A. G., Goldsmith, G. R., Brooks, P. D., and Dawson, T. E.:
Discrepancies between isotope ratio infrared spectroscopy and isotope ratio
mass spectrometry for the stable isotope analysis of plant and soil waters,
Rapid Commun. Mass Spectrom., 24, 1948–1954, 10.1002/rcm.4597, 2010.West, A. G., Goldsmith, G. R., Matimati, I., and Dawson, T. E.: Spectral
analysis software improves confidence in plant and soil water stable isotope
analyses performed by isotope ratio infrared spectroscopy (IRIS), Rapid
Commun. Mass Spectrom., 25, 2268–2274, 10.1002/rcm.5126, 2011.
Zalicki, P. and Zare, R. N.: Cavity ring-down spectroscopy for quantitative
absorption measurements, J. Chem. Phys., 102, 2708–2717,
1995.