Introduction
Cavity ring-down spectroscopy (CRDS) is a high-sensitivity laser absorption
technology that is becoming increasingly common for trace gas analysis
(Wang et al., 2008). As well as returning high-resolution mole
fraction measurements (Crosson, 2008), CRDS is used for stable
isotope analysis of CO2, CH4, H2O, and N2O (Crosson
et al., 2002; Dahnke et al., 2001; Kerstel et al., 2006; Sigrist et al.,
2008). Commercial deployment of CRDS has created novel analytical
possibilities with greater stability, precision, instrument portability, and
a lower cost basis compared with many traditional spectroscopic,
chromatographic, and mass spectrometric techniques (Berryman et al.,
2011; Hancock and Orr-Ewing, 2010; Mürtz and Hering, 2010; Picarro,
2009).
Crosson et al. (2002) provide a description of the working principles for
taking isotopic measurements by CRDS. Commonly used in atmospheric research,
isotopic CRDS gas analysers are normally online instruments whereby sample
gas is continuously pumped through an optical cavity. While such continuous
measurement systems are useful for monitoring applications, technical
adaption is necessary for routine handling of small discrete gas samples.
Commercial add-on modules are available for this purpose
(McAlexander et al., 2010; Picarro, 2013), but these are unable
to match the rapidity of conventional methods like gas chromatography (GC)
and isotope ratio mass spectrometry (IRMS).
CRDS analysis with discrete sample throughput and handling comparable to
IRMS could significantly improve a variety of empirical research. For
example, simultaneous high-precision isotope ratio and mole fraction
measurements from isotopic-CO2 CRDS will reduce empirical workload and
increase accuracy of CO2 flux partitioning calculations in soil and
plant respiration experiments (Midwood and Millard, 2011; Snell et al.,
2014). However, realising these benefits requires regular batch analysis of
discrete samples – existing arrangements that couple CRDS instruments
directly to soil headspace chambers are generally constrained to measuring
just one experiment at a time (Albanito et al., 2012; Bai et al., 2011;
Midwood et al., 2008).
Berryman et al. (2011) describe a syringe sample delivery system for isotope
ratio CRDS that allows small air samples (20–30 mL NTP) to be analysed. In
their method, the optical cavity of the CRDS analyser is flushed and
completely evacuated prior to direct sample injection to ensure consistency
and prevent sample-to-sample contamination. Although it is an important
technical innovation with handling and cost advantages over IRMS, the set-up
is limited by slow sample turnover rates (3 h-1).
In this paper, we present a new method for measuring discrete syringed gas
samples (50 mL NTP) by CRDS. Like Berryman et al. (2011),
this method was conceived for isotopic-CO2 CRDS to provide δ13C–CO2 and CO2 mole fraction (xCO2) analysis in soil
respiration studies, but remains general enough to be used in other contexts
and adjusted for other gas species. Instead of evacuating the cavity prior
to sample introduction, our process intersperses samples against background
measurements of a fixed reference air and post-corrects for bias in the
measurements. This results in considerably faster throughput for atmospheric
samples (up to 13 h-1) than the method of Berryman
et al. (2011). Additionally, with precision and discrete sample measurement
rates comparable to automated continuous-flow IRMS, this method further
advances CRDS as an attractive alternative for trace gas isotope analysis.
Materials and methods
Analyser and sampling system
The CRDS instrument adapted for discrete sample measurement was a Picarro
G2131-i isotopic-CO2 gas analyser (Picarro Inc., Santa Clara, CA, USA).
Detailed descriptions of the operation and spectroscopy of the G2131-i and
predecessor units can be found in Dickinson et al. (2017),
Hoffnagle (2015), Rella (2010a, b, c), and Wahl et al. (2006). In brief,
sample air is circulated through a high-reflectivity
optical cavity (35 cm3) at an inlet flow rate of ca. 25 mL min-1
(NTP). Internal controls maintain the cavity at 318.150 ± 0.002 K and
18.67 ± 0.02 kPa. Spectroscopic ring-down time constants are measured
across spectral bands of 12C16O2 and 13C16O2
to determine optical absorption peak heights, which are computed into
13C / 12C isotope ratio and CO2 mole fraction data
(Hoffnagle, 2015). Spectral lines of 12C1H4 and
1H216O are also measured for correcting direct and indirect
spectral interferences from H2O and CH4 on the CO2 bands. The
normal measurement range for the G2131-i is set at 380 to 2000 ppm
xCO2 and natural abundance to +5000 ‰ in δ13C–CO2 (Picarro, 2011).
Schematic diagram of the discrete gas sample measurement system
coupled to the isotopic-CO2 CRDS analyser.
All measurements made by the G2131-i are continually recorded at a rate of
ca. 0.8 Hz; specific data must be extracted from log files for further
treatment. Although discrete sample measurement is thus possible without
special provision – by inletting the G2131-i with 200 to 300 mL (NTP) of sample
from a gasbag or chamber and retrieving the relevant data (Picarro,
2012) – such a procedure is time inefficient and prone to errors from
operator inconsistency. Furthermore, in many research settings it is
impractical or impossible to gather such large samples (e.g. headspace
chamber analyses). By instead applying a controlled procedure for inletting
smaller volumes and software to automatically process the raw data in
real time, a more feasible method of discrete sample measurement was
created.
A schematic of our measurement set-up is shown in Fig. 1. The system was
simple in construction and concept: hermetic sample collection and delivery
was achieved by a high-quality gas-tight syringe with push-button valve and
Luer lock fitting (50 mL, SGE Analytical Sci., Australia). A
low-permeability multilayer foil gasbag (27 L Plastigas, Linde AG, Germany)
functioned as a reservoir for a reference air standard, which was analysed
between individual samples so as to give a baseline for accurate data
delineation. The large, non-pressurised volume of the reservoir meant
pressure-induced mixing and back-flow risks were excluded and allowed
prolonged operation before refilling (> 15 h). Gas-proof
fluorinated-ethylene-propylene (FEP) tubing (Rotilabo, Carl Roth GmbH,
Germany), Luer lock fittings, and Luer lock 3-way valves completed the
set-up. All permanent tube fittings and joins were adhered with Loctite
406/770 (Henkel AG, Germany) to ensure robustness and prevent leakage. The
FEP tubing between the syringe sample inlet point and the CRDS inlet port
(Fig. 1) was minimised (1/8-inch OD × 44 cm, connected to the
1/4-inch CRDS inlet port with reducing ferrule) to decrease mixing and lag
time between sample delivery and measurement.
Sample measurement
The G2131-i and discrete sample measurement system were installed in an
environmentally controlled laboratory (20 ∘C) to ensure stable
operation. Syringed sample measurement was performed as follows: after
instrument start-up and commencement of normal function, reference air
measurement was initiated to establish stable baselines of xCO2 and
δ13C–CO2. When a sample was ready for analysis, the
syringe was connected to the sample inlet point (Fig. 1), and the 3-way
valve manually actuated to stop the flow of reference air and supply the
sample directly to the analyser. Upon opening the syringe valve the gas
sample was drawn into the G2131-i, causing steady, unassisted collapse of
the syringe plunger. Sample evacuation was completed in ca. 2.5 min, after
which the sample inlet point valve was immediately reset and reference air
intake resumed. Once CO2 and δ13C–CO2 readings had
returned to initial baseline levels (thereby safeguarding against
sample-to-sample carryover), the process was repeated for the next sample.
In this way, reference air readings were punctuated by syringe samples to
create “peaks” in the raw data output with a sample-to-sample time of ca.
5 min (Fig. 2). The gas aliquot size for all measurements was nominally 50 mL
NTP. (Analysis of smaller amounts may be possible but 50 mL was assessed as
a minimum for reliable operation. Samples larger than 50 mL would be easily
handled, although adjustment of peak truncation parameters and
re-calibration may be necessary for accurate performance – see below and
Sect. 2.3.)
Example CRDS data feed for syringe samples. Reference air
measurements (ca. 425 ppm x12CO2 and -37 ‰
δ13C–CO2) are interrupted by successive samples to form
consistently identifiable peaks in the data.
(a) Example of raw G2131-i measurement data and
breakdown of events during analysis of a 50 mL syringe sample. Blue segments
are truncated from the sample peak by our software script while red segments
are the extracted measurement data. All timings and thresholds are user
customisable in the software for variation in sample size and equipment.
(b) The most reliable sample end time (detrigger) was established as
the point when measurements returned to half the difference between
peak-maximum (or minimum in the case of samples with lower xCO2 than
reference air) and the baseline value. Grey lines are amplitude-normalised
tailing segments from 23 test samples widely varying in x12CO2. The
broken green curve denotes a generalised logistic function fit to these test
data by non-linear least squares optimisation. Solving the fitted function
determined that 29 ± 2 s elapsed between peak-maximum and half-maximum
irrespective of sample composition.
To achieve unambiguous sample peak identification, distinction in CO2
was required between reference air and sample. In practice this meant a
relative change of ca. 2 % in xCO2 or ca. 5 ‰ in
δ13C–CO2. However, very large differences resulted in
slower sample turnover (see Sect. 3.1). Best throughput was obtained using
reference air that was similar to samples in xCO2 but contrasted in
δ13C–CO2 (ca. 15 ‰ difference). In
this work, dry standard air with 496 ppm xCO2 and -36.1 ‰
δ13C–CO2 was used as the reference
for all formal measurements (NA1, Table 1).
While sample measurement was performed manually (i.e. syringe connection and
disconnection, valve operation), to ensure method consistency we composed a
custom computer software script to manage the process in real time (script
available in the Supplement). Running through the built-in Coordinator
software programme of the G2131-i, our script prompted the user for correct
timing of sample introduction, detected and extracted sample peak data,
monitored reference air values, filtered problem measurements, and recorded
measurement results. The software script isolated individual samples from the
CRDS data stream by using specific events and timings in the measurement
process as cues (e.g. a basic peak recognition algorithm; Fig. 3a). Prior to
the introduction of a sample, a reference air baseline was recorded for 30 s
and averaged. Sample detection (trigger) then occurred when xCO2 or
δ13C–CO2 values deviated from the baseline beyond a fixed
threshold (default: 0.5 % of xCO2 or 2 ‰ in
δ13C–CO2). The sample end (detrigger) was detected when
measurements returned halfway to baseline values (Fig. 3b). By truncating the
sample peak data +80 s from the trigger and -29 s from the detrigger,
ca. 30 s of representative measurement data was obtained for each sample
(Fig. 3a). Upon completion of a sample measurement, the script computed means
and standard deviations (SDs) of all data elements reported by the
G2131-i (i.e. xH2O and xCH4 values together with
xCO2 and δ13C–CO2). These statistics were compiled
along with corresponding baseline measurements, time-stamped, assigned sample
descriptors, and then outputted into a concise results file (see example in
the Supplement). After each detrigger event the software monitored CRDS
readings for a return to initial reference air baseline values before
directing the operator to proceed with the next sample.
In addition to the G2131-i analyser, our method was successfully trialled on
a sister CRDS instrument (Picarro G2201-i). The G2201-i differs from the
G2131-i only in additionally measuring 13C1H4 to give δ13C-CH4 data (Picarro, 2015). To assist method adoption, we
supply software scripts customised for each instrument (Supplement). The
scripts include provision for user adjustment of peak identification and
truncation parameters to suit individual set-ups. A short video
demonstration of the system is also available (see 10.5446/32922).
Bottle measurement data of the standard air used as baseline for
syringe sample measurements (NA1) and the gas standards used in method
calibration (NA2 to LE2). Values are the averages (SDs in parentheses) of
10 min measurements taken for each standard directly inlet to the CRDS
analyser (see Sect. 2.3). Data have been post-corrected as per the
calibration of Dickinson et al. (2017).
Standard ID
x12CO2 (ppm)
x13CO2 (ppm)
xCO2 (ppm)
RCO2 (13CO2/12CO2)a
δ13C–CO2 (‰)b
NA1 (Ref. air)
490.55 (0.13)
5.286 (0.004)
495.84 (0.13)
1.0776 (0.0006)
-36.14 (0.57)
NA2
1024.26 (0.21)
11.137 (0.004)
1035.39 (0.21)
1.0874 (0.0003)
-27.43 (0.28)
ZERO
0.05 (0.04)
0.004 (0.004)
0.05 (0.04)
–
–
HE1
2028.98 (0.47)
25.528 (0.007)
2054.51 (0.47)
1.2582 (0.0004)
+125.35 (0.34)
HE2
2009.15 (0.53)
100.11 (0.02)
2109.26 (0.53)
4.983 (0.001)
+3456.9 (1.1)
TT
1002.18 (0.22)
50.216 (0.008)
1052.40 (0.22)
5.011 (0.001)
+3481.7 (1.1)
LE1
402.24 (0.11)
25.249 (0.005)
427.49 (0.11)
6.277 (0.002)
+4614.5 (1.7)
LE2
398.21 (0.16)
101.24 (0.01)
499.45 (0.16)
25.42 (0.01)
+21 739 (9)
a RCO2 data are scaled by 102 for ease of comprehension.
b δ13C–CO2 values are reported against VPDB (Vienna Pee Dee Belemnite; Werner and Brand, 2001).
Measurement calibration
Conventional CRDS trace gas measurements are affected by signal noise,
gradual instrument drift, and any unaddressed interferences or perturbations
in the underlying spectroscopy (Vogel et al., 2013).
Calibration strategies exist to counter both drift and spectral errors, while
random noise ultimately limits instrument precision (Friedrichs et al.,
2010; Wen et al., 2013). In the present case of small discrete samples,
however, there was additional inaccuracy stemming from the nature
of sample gas delivery into the CRDS analyser.
As discussed in studies by Gkinis et al. (2011) and Stowasser et al. (2012),
stepwise changes to the inlet gas composition (as occur with discrete
samples) do not give rise to correspondingly abrupt jumps in CRDS
measurements and instead result in sigmoid-shaped steps in the data
(Fig. 3b). These smoothed transitions are the combined result of (i) the rate
of gas replenishment in the optical cavity (Stowasser et al., 2014),
(ii) partial mixing (turbulence and diffusion) of gas compositions downstream
of the sample inlet (Gkinis et al., 2011), and (iii) molecular sorption and
desorption on internal surfaces of the cavity and inlet tubes (Friedrichs et
al., 2010). Although reported response times of CRDS
instruments typically range from 1 to 3 min (Picarro, 2011; Sumner et al.,
2011), the actual time required for an optical cavity to completely
transition to a new gas composition can be substantially longer. In testing
the G2131-i, we observed remnants of previous gases persisting with
asymptotical decline for as long as 40 min following very large shifts in
CO2 composition (e.g. |ΔxCO2|> 10 000 ppm or |Δδ13C–CO2|> 5000 ‰). While the error caused by the
residual gases may sometimes be relatively trivial, all measurements that
occur prior to the cavity attaining equilibrium will experience these
“memory effects”.
In our 50 mL syringe samples, memory effects were clearly present, as
evidenced by the asymptotic curvature in the data peaks (Fig. 2). This meant
that reported measurements of syringe samples were biased towards reference
air compared to “true” values that would eventually be determined from
measurements of indeterminately large sample volumes and monitoring for
asymptotic closure. Other researchers have mitigated memory effects by
evacuating the optical cavity before sample introduction (Berryman et al.,
2011), or through several replicate measurements (Gupta et al., 2009; Leffler
and Welker, 2013). Such solutions significantly reduce sample throughput,
however. In this work, we elected to post-correct for reference air
carryover by calibrating our method with bottled gas
standards. More specifically, we compared discrete sample measurements of gas
standards against measurements of the same standard directly inlet to the
G2131-i for prolonged periods (> 1 h). Importantly, syringe
measurements were not calibrated directly against gravimetric values of the
standards – here we were only concerned with isolating and eliminating bias
associated with the syringe sampling method and not with unaccounted
inaccuracies or drift in instrument spectroscopy. This approach facilitates a
more comprehensive examination of memory effects while also providing
flexibility for method adaptation or applying additional error corrections
for specific samples (e.g. adjusting for gas-matrix pressure broadening
effects: Nara et al., 2012).
To this end, seven gravimetric gas standards were used as fixed-source
calibrants (0.05 to 2109 ppm xCO2 and -27.3 to
+21 740 ‰ δ13C–CO2; see Table 1, exact
compositions detailed in Dickinson et al., 2017). Using such a wide range of
CO2 compositions served to improve overall calibration accuracy as well
as demonstrating reliability and applicability of method. Direct measurements
were performed by inletting the bottled standards to the
G2131-i for more than 1 h to ensure the absence of memory effects
before taking formal measurements for 10 min (ca. 460 data points, averages
reported in Table 1). Next, 50 mL syringe samples of the standards were
taken directly from bottles (syringe was pre-flushed several times to
preclude contamination) and measured as described in Sect. 2.2 (8 samples of
each standard for 56 measurements in total – data set in the Supplement).
Before further analysis, due to the high 13C abundance in several gas
standards, all reported CO2 data were adjusted
using the empirical correction described by Dickinson et al. (2017). (In
testing CRDS performance with 13C-enriched CO2, Dickinson et al.,
2017, identified minor but unaccounted spectroscopic cross-talk in
12CO2 measurements at elevated levels of 13CO2, as well
as logical inconsistencies in the G2131-i data output. Their
correction scheme was applied to the present data as a precaution to preclude
any possibility that an underlying instrument error might obfuscate the
memory effects in syringe sample measurements.)
The relationship between syringe and bottle measurements was established by
recognising that the data peaks generated by syringe samples could be
approximated by generalised logistic curves (Fig. 3b, also Gkinis et al.,
2011). From this, together with a constant aliquot size for all syringe
measurements, we were able to predict a simple linear scaling of syringe
values:
syringe=base+(bottle-base)/K,
where “syringe” refers to the measurement value obtained for a syringe sample of a
gas standard, “base” to the baseline measurement of reference air prior to sample
introduction, “bottle” to the direct bottle measurement of the same standard, and
K is a dimensionless empirical constant.
While all CO2 data elements reported by the G2131-i exhibited
reasonably similar sample peak geometry, the empirical constants for
12CO2 and 13CO2 were expected to differ due to
(de)sorption and diffusion-induced isotope fractionation during sample
filling of the optical cavity. Further, theoretical gas mixing considerations
entailed that Eq. (1) would not consistently hold for 13C / 12C
isotope ratio data (RCO2), where a simultaneous change in
total-xCO2 also occurred. Consequently, only x12CO2 and
x13CO2 data were explicitly calibrated, with RCO2 being
subsequently recalculated. (Moreover, only the dry mole fraction data of
12CO2 and 13CO2 were used due to the high likelihood of
different transition equalisation rates for CO2 and H2O. For
explanation of dry and wet mole fraction data see Hoffnagle, 2015; Rella,
2010a; Rella et al., 2013.) Accordingly, the following correction formulae
were derived from Eq. (1):
x12CO2(corrected)=x12CO2(base)+[x12CO2(syringe)-x12CO2(base)]⋅KC12x13CO2(corrected)=x13CO2(base)+[x13CO2(syringe)-x13CO2(base)]⋅KC13.
Total-xCO2, RCO2, and δ13C–CO2 data were then
determined from the resulting corrected values of x12CO2 and
x13CO2:
xCO2=x12CO2(corrected)+x13CO2(corrected)RCO2=x13CO2correctedx12CO2(corrected)δ13C–CO2=RCO2RVPDB-1⋅1000‰.
The correction constants, KC12 and KC13, were found through weighted
least squares analysis (WLS) of Eqs. (2) and (3) with syringe and bottle
measurements of gas standards as input data (i.e. reverse regression of Eq. 1:
bottle measurements substituting for the left-hand sides of Eqs. 2 and 3). To
increase statistical power, RCO2 and total-xCO2 data from bottle
measurements were also incorporated into the analysis with Eqs. (4) and (5),
thereby forming an extended optimisation problem (n=216). In a similar
vein to the WLS approach used by both Dickinson et al. (2017)
and Stowasser et al. (2014) for calibrating CRDS measurements,
residual weights were taken as the reciprocals of the individual summed
variances resulting from the SDs of each syringe sample and bottle
measurement (see Supplement and Table 1). The WLS solution was determined in
R (version 3.2.1, R Core Team, 2015) by general purpose
optimisation using the L-BFGS-B algorithm (Zhu et al., 1997) to yield
the best-fit correction constants for all available CO2 mole fraction
and 13C / 12C isotope ratio data.
Precision and consistency tests
CRDS precision is generally assessed by the variability in repeated
measurements of a homogenous gas source (e.g. the SD of multiple 5 min
analyses; Vogel et al., 2013; Wang et al., 2013). However, the internal
variation in individual measurements can also be used to gauge analytic
resolution (e.g. the SD of data contained in a 10 min measurement – as with
the bottle measurements in Sect. 2.3; also Pang et al., 2016 and Stowasser et
al., 2014). For our case of 50 mL syringe samples, precision was quantified
in both ways: the SD of the 30 s of CRDS data composing each individual
sample (intrasample SD; see Sect. 2.2) and as the statistical dispersion of
replicate samples (intersample SD).
We tested method precision by repeated measurements of a systematic set of
gas mixtures that spanned the normal operational CO2 range of the
G2131-i. Using gas standards as blending sources (Table 1;
Dickinson et al., 2017), 20 unique mixtures with varied CO2 mole fractions
(ca. 300, 600, 1000, 1500, 2000 ppm) and δ13C–CO2 values
(ca. -30, +800, +1750, +2700, +3600 ‰) were
prepared into multilayer foil gasbags (1000 mL Supel Inert, Sigma-Alrich
Corp., St. Louis, MO, USA). (The set of mixtures formed an orthogonal array
of cross combinations of xCO2 and δ13C–CO2 meaning any
interdependency in precision could be identified.) Each mixture was sampled and
measured with the syringe method three times in succession, and results
were analysed for inter- and intra-measurement variation.
Long-term consistency and reliability of our syringe method was assessed by
periodic analysis of standard air (NA2, Table 1) during the course of 9
months of routine instrument use. More than 200 measurements were conducted
and results were examined for precision and drift.
Results and discussion
System operation
Though somewhat labour intensive and requiring continual operator attention,
the syringe sample measurement process was uncomplicated, reliable, and
economical. Sample handling and CRDS operation was non-specialist in
comparison to conventional IRMS. The method was flexible to CO2
composition, successfully handling samples < 0.1 to > 20 000 ppm
xCO2 and -100 to +30 000 ‰ δ13C–CO2.
The only significant methodological constraint observed was a reduction in
sample turnover rate for compositions greater than either 3000 ppm
xCO2 or +4000 ‰ δ13C–CO2. This was because
post-sample reference air measurements took longer to return to pre-sample
baselines due to memory effects, thereby extending the intersample period.
Keeping CO2 levels within G2131-i specifications resulted in a
throughput of ca. 10 samples h-1. Best measurement rates of 12 to
13 samples h-1 occurred when sample CO2 compositions neighboured
the reference air (e.g. within ca. 100 ppm xCO2 and ca. 20 ‰
δ13C–CO2 of reference). These throughput rates are at least a
2-fold improvement over both the methods of Berryman et al. (2011) and
specialty peripheral devices (Picarro, 2013).
Discrepancies between syringe sample and direct bottle
measurements (syringe bias) of gas standards as a function of the syringe
sample peak height (Eq. 7) for (a) x12CO2 and (b)
x13CO2. The WLS-fitted linear models (see Sect. 2.3) are overlaid
for comparison (solid lines, slopes =1-K, Eq. 7), with 95 % confidence
intervals (shaded) and 95 % prediction intervals (dashed lines) as
determined from the standard error estimates of KC12 and KC13
(Sect. 3.2).
Following initial development, the syringe method was incorporated into our
general laboratory practices and during the first year of implementation
more than 10 000 samples were measured. Despite intense instrument usage, we
noticed no changes or adverse impacts on G2131-i function, although
increased external pressure variations caused by frequent syringe
evacuations may conceivably reduce mechanical lifetimes of optical cavity
pressure control valves. Failures occurred in ca. 1 % of measurements,
principally due to operator mistakes but occasionally because of leakage in
sample inlet valve, syringe fault, or complications from the peak
identification algorithm for samples very similar to the reference air (see
Sect. 2.2). Very rarely, minor instabilities in reference air readings
caused false peak detections and baseline return problems, but such
instances were usually identified by the software script and internally
resolved.
Durability of the gas-tight syringes used for sample delivery was excellent,
although regular monitoring and maintenance was important to ensure smooth
sample evacuation during the measurement process. Excessive plunger friction
led to significant “jumpiness” during the collapse of the syringe plunger,
which manifested as small pressure fluctuations in the optical cavity and
increased measurement noise (evidenced by larger intrasample SDs). Careful
cleaning and exact silicone lubrication was carried out every few hundred
samples to ensure uniform plunger operation and prolongation of syringe life.
Syringe push-button and sample inlet point valves also required periodic
attention and were replaced as necessary to pre-empt leaks and breakages.
Precision in syringe sample data for x12CO2 (a, b)
and x13CO2 (c, d) as quantified by standard deviations (a, c)
and relative standard deviations (b, d) for individual
measurements (red) and replicate measurements (blue).
Correction of memory effects
From rearranging Eq. (1), the discrepancy between syringe and bottle
measurements (syringe bias) was predicted to be proportional to the
difference between the syringe value and reference air baseline (sample peak
height):
(syringe-bottle)=(syringe-base)⋅(1-K).
Comparing the actual syringe sample and bottle measurements of gas
standards, we observed a systematic memory effect bias that was indeed
consistent with this postulated relationship (Fig. 4). WLS across all
CO2 data yielded estimates of 1.00341 for KC12 and 1.00440 for
KC13, with a coefficient of determination (r2) of 0.84 (weighted
residuals) for the complete correction model. Standard errors for
KC12 and KC13 estimates were respectively 0.00017 and 0.00014 (see
confidence intervals in Fig. 4). The Pearson's correlation coefficient
between KC12 and KC13 estimates was 0.26. The observed divergence in
correction constants for 12CO2 and 13CO2 was
statistically significant (t-test: P<0.0001) with a larger memory
effect present in 13CO2 measurements. This result corroborates the
expectation of isotope fractionation occurring during gas equalisation in
the CRDS optical cavity, putatively due to surface (de)sorption and
diffusion phenomena.
Having determined KC12 and KC13, syringe CO2 measurements can
be adjusted for bias with Eqs. (2)–(6). Accuracy of these corrections is
very good: the standard errors on KC12 and KC13 add uncertainty to
xCO2 and δ13C–CO2 data of less than 0.02 % of the
difference between the sample and baseline values. For atmospheric samples,
this additional source of error is entirely negligible compared to the
uncertainty deriving from measurement precision and gas standard analytical
accuracy.
While the correction coefficients (KC12 and KC13) found in this work
are unique to our sampling equipment and G2131-i analyser, the equivalent
calibration may be easily performed on replica set-ups. We provide a generic
spreadsheet to post-correct syringe sample CO2 data for any values of KC12
and KC13, and a template for simultaneously applying the syringe
correction with the spectroscopic calibration strategy of
Dickinson et al. (2017) for 13C-enriched samples
(Supplement). Although our work only addresses memory effect bias in
CO2 data, we are confident that the same strategy (Eq. 1) is
straightforwardly applicable to other gas species (and isotopes) that can be
similarly analysed by syringed samples and CRDS (e.g. CH4, H2O,
N2O).
Measurement precision and consistency
Replicate tests provided a practical account of precision afforded by our
discrete sample measurement system. Figure 5 shows inter- and intrasample SDs
and relative SDs for 12CO2 and 13CO2 mole fraction data
(complete data set in the Supplement). The SDs of both species were generally
proportional to their measured values and unaffected by δ13C–CO2 level (i.e. precision in 12CO2 and
13CO2 measurements were mutually independent). Relative SDs for
both isotopologues remained near constant at ≤ 0.05 % across the
tested ranges, however (Fig. 5c, d). Notably, the majority of intrasample SDs
for both x12CO2 and x13CO2 data were found to be in
general agreement with counterpart intersample SDs (see trend lines in
Fig. 5). This means that the SDs reported by our software script for
12CO2 and 13CO2 mole fractions in individual syringe
sample measurements will reasonably approximate the expected precision for
replicated measurements of those samples.
In contrast, inter- and intrasample SDs in 13C / 12C isotope
ratio data were dependent on the δ13C–CO2 level and
CO2 mole fraction, increasing with higher δ13C–CO2
and lower xCO2 (see Fig. S1a, b in the Supplement). The relative SDs of
isotope ratio measurements were unaffected by δ13C–CO2
level but steadily decreased with increasing xCO2 – declining from
between 0.07 and 0.04 % at 300 ppm xCO2 to between 0.03 and 0.015 %
at 2000 ppm (Fig. S1d). One exception was at natural abundance isotope
ratios (δ13C–CO2≈-30 ‰), where intersample relative SDs of RCO2 were steady at ca. 0.015 %
(i.e. 0.15 ‰) across the tested xCO2 range (Fig. S1b).
Somewhat opposing CO2 mole fraction data, intrasample SDs of
isotope ratio data were almost always greater than corresponding
intersample SDs, which largely reflects the summation of variance from the
12CO2 and 13CO2 spectral measurements used to produce
the 13C / 12C ratios. Nevertheless, as with 12CO2 and
13CO2, the SD reported for δ13C–CO2 in
individual syringe sample measurements may be used as a conservative proxy of
δ13C–CO2 replicate precision.
Repeated syringe sample measurements in
(a) x12CO2 and (b) δ13C–CO2 of
standard gas NA2 (Table 1) over a 9-month period (n=200). Error bars
denote ±1 intrasample SD of each individual measurement. Grand means are
the solid black horizontal lines with dotted lines indicating ±1 SD of
all measurements. Ten sample moving averages are shown in red. Histogram
insets on the right depict cumulative distributions of syringe measurements.
Blue dashed lines indicate the direct bottle measurement of NA2 with
blue-shaded areas covering ±1 SD of the bottle measurement.
Consistency of the syringed sample method was established by long-term
repeated analysis of standard air (NA2, Table 1). Figure 6 shows
x12CO2 and δ13C–CO2 data from 200 measurements
covering a 9-month period (data set available in the Supplement). It is
important to note that, because these data were only adjusted for memory
effects inherent to the discrete sample system (i.e. by Eqs. 2–6), they
represent a simultaneous time-series test of instrument stability and
methodological constancy. The measurements averaged 1024.18 ppm in
x12CO2 and -27.35 ‰ in δ13C–CO2 with
respective SDs of 0.50 ppm and 0.33 ‰. The latter SD is larger than
the intersample SD found in replicate measurement testing (0.15 ‰;
see above), likely indicating the presence of instrument drift in the data in
addition to random errors of repeated syringe sampling. While the separate
components of variance cannot be resolved here, moving means (red lines in
Fig. 6) show neither a sustained trend nor method discontinuity, and imply
that reasonable measurement accuracy is possible under typical laboratory
practices without perpetual calibration against gas standards (compare
syringe sample measurements against the direct bottle measurement of NA2,
Fig. 6), corroborating the similar conclusion reached by Friedrichs et
al. (2010). The mean of intrasample SDs in the 200 measurements was 0.42 ppm
for x12CO2 and 0.35 ‰ for δ13C–CO2, both
corresponding well to the aforementioned SDs of all measurements and the
intrasample SDs in the replicate tests. This consistency further supports our
proposition that a single syringe measurement and its intrasample SD can
deliver a similar (although inherently less reliable) statistical estimate to
one generated through multiple sample measurements, potentially making
replicate CRDS analyses unnecessary in research contexts where statistical
uncertainty is not a critical consideration.
In sum, despite the short CRDS analysis period for a syringe sample (ca. 30 s),
and limited number of replicates in performance testing, achieved
measurement precision was excellent. With our system and G2131-i analyser,
replicate sample SDs of ≤ 0.05 % may be expected for 12CO2
and 13CO2 mole fraction data, while resolution in repeated δ13C–CO2 measurements will be ca. 0.15 ‰ at
natural 13C abundance. Moreover, to a first approximation, similar
precisions can be obtained from intrasample SDs of single syringe sample
measurements. These results should be viewed tentatively if adapting our method
to a different model of CRDS analyser, however. Because observed measurement
variation derives from both volatility in the discrete sampling method and
noise inherent to the instrument, matching the precision reported here is
unlikely with lower-performance CRDS units. As a point of reference, when
using the G2131-i for continuous analysis of a homogeneous source, the
intersample SDs for sequential 30 s data segments are ca. 0.15 ‰
in δ13C–CO2 and 0.01 % in both
x12CO2 and x13CO2, while intrasample SDs are
ca. 0.30 ‰ and 0.025 % (Dickinson et al., 2017;
also manufacturer specifications: Picarro, 2011).
Compared to other CRDS discrete sample methods, our results improve upon
the 0.3 ‰ (δ13C–CO2) and 0.3 %
(xCO2) precision attained by Berryman et al. (2011),
although this is probably due to the enhanced spectroscopic sensitivity of
the G2131-i relative to the older G1101-i analyser used in their study.
Additionally, our method delivers precision in δ13C–CO2
that is similar to both the Picarro SSIM2 discrete sample peripheral device
(0.11 ‰, Picarro, 2013) and traditional
continuous-flow IRMS (typically 0.1 ‰), which, by
contrast, are single-purpose instruments that do not also report accurate
CO2 mole fraction measurements. And finally, although finer measurement
resolution is possible with CRDS, the uncertainties deriving from the
precision of our discrete sample measurements will be, in many cases, no worse
than the tolerances on gravimetric gas standards used for instrument
calibrations (cf. Brewer et al., 2014; Dickinson et al., 2017). In such
contexts, applying our method to isotopic and mole fraction analyses of
trace gases should not result in significantly poorer absolute accuracy
compared to other sampling techniques (i.e. uncertainties on gas standards,
rather than measurement precision, may limit overall accuracy).
Potential applications
At present, isotope ratio analysis of fixed trace gas samples is usually
accomplished by IRMS interfaced to autosampling GC systems. Such instruments
require specialised user training and carry high consumable costs, however.
Similarly capable CRDS-based techniques can avoid both these limitations and
represent an advance in stable isotope analysis. Although not suitable for
all sample types, adapting the present generation of CRDS gas analysers for
rapid discrete sample measurement has promising application in contexts
where syringe or flask sampling is frequently performed – especially where
accurate gas mole fraction data are also valuable – such as in ecosystem
respiration and emission studies (cf. Zeeman et al., 2008), analysing
dissolved gases in terrestrial waters (Hope et al., 1995; Loose et al.,
2009), and certain instances of trapped air in ice cores (e.g. Sowers et
al., 2005).
A specific example where our method has immediate relevance is in measuring
CO2 respiration in soil microcosm headspace studies. To date, applying
CRDS gas analysers to such research is mostly achieved through closed-loop
recirculation (Christiansen et al., 2015; Ramlow and Cotrufo, 2017) or
continuous analysis of open chamber systems (Bai et al., 2011; Jassal et
al., 2016). Apart from cost and complexity, these solutions restrict the
number of experiments that can be concurrently measured by a single
instrument. Our system significantly eases this constraint, however. For
instance, assuming a sample turnover of 10 h-1 and conducting four
syringed headspace measurements per microcosm over the course of a 10 h workday,
it is feasible to use one analyser for measuring daily respiration rates in
25 simultaneous experiments. Further, where CO2 flux partitioning by
isotopic analysis is undertaken, achieving sample measurement precision of
ca. 0.05 % in xCO2 and ca. 0.15 ‰ in δ13C–CO2 means that the resulting uncertainties on efflux
partitions will be comparable (if not smaller) to those in studies using
infrared gas analysis and IRMS or IRMS alone (cf. Joos et al., 2008;
Munksgaard et al., 2013).
The primary drawbacks of employing our method for isotopic-CO2
measurements of discrete samples compared to an automated GC-IRMS system are
(i) the larger sampling size, (ii) a more constrained operational xCO2
range, and (iii) the necessity of near-continuous operator presence at the
instrument. However, implementation of smaller CRDS optical cavities
could dramatically decrease the required
sample amount and allow even shorter measurement times (e.g. Stowasser et
al., 2014), while dilution methods and calibration can expand the xCO2
measurement range of CRDS. Similarly, methodological refinements that
integrate automated syringe sampling and valve systems would curtail labour
requirements.
Conclusions and outlook
Discrete sample analysis of trace gases by CRDS is possible through basic
instrument adaptation. We have set forth a scheme for xCO2 and δ13C–CO2 determination of 50 mL syringed samples on a Picarro
G2131-i isotopic-CO2 analyser. With software to manage the measurement
process and compute results data, our method offers substantially faster
analysis of small gas volumes with equal or better precision than comparable
set-ups. Memory effects present in syringe sample measurements can be
accurately compensated by calibration against large-volume measurements of
gravimetric gas standards.
Although CRDS is gaining scientific acceptance for isotopic-CO2
measurement, so far the technology has not seriously challenged IRMS in
discrete gas sample analysis, despite lower running and capital costs,
simpler operation, less measurement drift, and the added benefit of
providing more accurate xCO2 data concurrently with δ13C–CO2.
In achieving comparable precision and sample throughput
to IRMS, our syringe sample method helps to position CRDS as a tenable
competitor for isotopic analysis of discrete samples.
The chief disadvantages of our process compared to IRMS for
isotopic-CO2 analysis are a narrower xCO2 performance range, higher
labour demands, and a comparatively large sample size (50 mL NTP). Method
improvements towards automation may greatly ease user workload, however, and
the development of smaller optical cavities could reduce the sample gas
needed for discrete analysis on future CRDS analysers as well as increasing
sample throughput rates even further.
This system can be applied with any Picarro G2131-i or G2201-i CRDS
analyser, though calibration and tuning of parameters in the software script
may be necessary to account for variations in set-up, sample volume (and
pressure), and reference air composition. Implementation on other CRDS
instruments and conversion for measurements of other trace gases are
anticipated with only minor software amendments.