A Dual Inlet System for Laser Spectroscopy of Triple Oxygen Isotopes in Carbonate‐Derived and Air CO2

Analyzing the triple oxygen isotope (Δ′17O) composition of carbonates and air CO2 can provide valuable information about Earth system processes. However, accurately measuring the abundance of the rare 17O‐bearing CO2 isotopologue using isotope ratio mass spectrometry presents significant challenges. Consequently, alternative approaches, such as laser spectroscopy, have been developed. Here, we describe an adaptable dual inlet system for a tunable infrared laser direct absorption spectrometer (TILDAS) that maintains stable instrumental conditions for subsequent sample and reference measurements. We report ∆′17O measurements on three types of samples: reference CO2, CO2 derived from the acid digestion of carbonates, and air CO2. The external repeatability (±1σ) for reference‐sample‐reference bracketing measurements is generally better than ±10 ppm, close to the average internal error of ±6 ppm. Our results demonstrate that laser spectroscopy is a capable technique for measuring triple oxygen isotopes with the precision required to resolve variations in the ∆′17O values of air CO2 and to use the ∆′17O of carbonates for paleothermometry.

Variations in the triple oxygen isotope ratios are expressed using the δ-notation (Equation 1 for δ 18 O in ‰; a similar equation can be written for δ 17 O; McKinney et al., 1950) and the Δ′ 17 O value (in ppm, Equation 2; Matsuhisa et al., 1978;Miller, 2002): 18  Supporting Information may be found in the online version of this article.
While several studies have been published on the triple oxygen isotope composition of water (e.g., Aron et al., 2021;Barkan & Luz, 2005) and silicate rock samples (e.g., Pack & Herwartz, 2014), few studies have published high-precision triple oxygen isotope data on CO 2 .This is because the isobaric interference from 13 C and 17 O substitutions within the CO 2 molecule hinders the direct isotope ratio mass spectrometric (IRMS) measurement of δ 17 O.The isotopologues 16 O 13 C 18 O and 17 O 12 C 18 O have nearly identical masses, and common IRMS techniques cannot resolve between the two (cf., Craig, 1957;Saenger et al., 2021).
Various alternative protocols have been developed to obtain the δ 17 O value of CO 2 .These include the high-temperature fluorination of CO 2 where ∆′ 17 O is analyzed on the resulting O 2 (Bhattacharya & Thiemens, 1989;Wostbrock, Brand, et al., 2020;Wostbrock, Cano, & Sharp, 2020), and high-temperature equilibration of CO 2 with CeO 2 or CuO (Assonov & Brenninkmeijer, 2001;Hofmann & Pack, 2010;Horváth et al., 2012;Kawagucci et al., 2005;Mahata et al., 2012;Mrozek et al., 2015), where the 17 O-anomaly is determined by mass-balance calculations.Barkan and Luz's (2012) approach includes the isotopic equilibration of CO 2 and H 2 O followed by the fluorination of H 2 O to O 2 .In the protocols of Brenninkmeijer and Röckmann (1998) and Passey et al. (2014), CO 2 is first reduced with H 2 to CH 4 and H 2 O, and then the δ 17 O value of the water is analyzed using CoF 3 fluorination.Ellis and Passey (2023) built upon this methanation-fluorination technique, but the initial step of their method involves the high-temperature conversion of the oxygen in the analyzed material, for example, carbonates, to CO. Isotope exchange between CO 2 and O 2 over hot platinum is another established protocol to obtain high-precision Δ′ 17 O data (Adnew et al., 2022;Mahata et al., 2013).Limitations of these methods are that they are laborious and involve multiple fractionation steps, which can affect the accuracy of the results.
Alternatively, the δ 17 O of CO 2 can be analyzed on the O + fragment ions using high-resolution gas source IRMS, which yields accurate and precise data but is time-consuming and costly (Adnew et al., 2019).Multi-collector secondary ion mass spectrometry of carbonates offers a high spatial resolution, but the precision for δ 18 O and δ 17 O is currently limited to ±0.5‰ (Bouden et al., 2021).
Laser absorption spectroscopy determines isotopologue abundances by detecting the absorbance of laser light by the analyte gas (McManus et al., 2006).Laser spectroscopy provides a quick and cost-effective approach for quantifying isotopologue ratios directly in CO 2 gas and has the potential to replace IRMS for many typical applications (Nelson et al., 2008;Prokhorov et al., 2019;Stoltmann et al., 2017;Tuzson et al., 2008;Wang et al., 2020;Yanay et al., 2022).
The first high-precision optical absorption measurements of 17 O anomalies in CO 2 were made using cavity-ring-down-spectroscopy (Stoltmann et al., 2017).Sakai et al. (2017)  Ensuring identical analytical conditions for both sample and reference measurements is crucial in laser spectroscopy to minimize analytical bias.This is because the absorbance spectra-from which the isotope ratios are calculated-are influenced by factors such as temperature, pressure, and the partial pressure of the constituents in the analyte gas.The aim of this study was to present a cost-effective and fully automated inlet system for TILDAS that can measure the ∆′ 17 O of carbonate-derived and air CO 2 within 10 ppm.across a ca.50-nm-wide spectral window, and the transmitted light was recorded with a thermoelectrically cooled photovoltaic HgCdTe infrared detector (Teledyne Judson; J19).Spectral data were recorded at 1 Hz.

The Dual Inlet System
The single-laser setup resolved absorption peaks from the 16 O 12 C 16 O ("626"), 16 O 12 C 17 O ("627"), and 16 O 12 C 18 O ("628") CO 2 isotopologues.This shorthand notation for the isotopologues is widely used for laser-specific applications and is based on the spectroscopic high-resolution transmission molecular absorption database (HITRAN, Gordon et al., 2022).
The spectral fitting of our TILDAS instrument (see Section 3) was optimized for ca.420 ppmv CO 2 in CO 2 -free dry air to allow measurements of air CO 2 (cf., Bowling et al., 2003;Sturm et al., 2013).Consequently, prior to measurement, pure CO 2 samples had to be mixed with a collision gas.Because CO 2 -free dry air was not readily available, we attempted to use other collision gases.First, we used pure Ar (Tyczka Industrie-Gase; Ar 4.6), which led to large spectral misfits.Switching to pure N 2 (Tyczka Industrie-Gase; N 2 3.5) showed a visible improvement in the spectral fit.Finally, on 13 September 2022, we changed to artificial, CO 2 -free dry air (Air Products; 1.0 vol.%Ar, 20.9 vol.% O 2 , 78.1 vol.%N 2 ).
The housing of the TILDAS was flushed with nitrogen to avoid the absorption of ambient CO 2 in the laser path.The dry purging gas was extracted from the headspace of a liquid nitrogen tank and was warmed up before being introduced to the TILDAS at a flow rate of ca. 2 L min −1 .

Hardware
The heart of the custom-built inlet system is a modified dual inlet system of a Finnigan MAT 251 mass spectrometer.The block containing the two bellows and the valves was extended with electropolished ¼-inch stainless steel tubings (Swagelok; 6L-T4-S-035-6M-E1) and pneumatically actuated bellows-sealed valves (Swagelok; SS-4BK-1C).We added a third bellow to the inlet system, constructed from a bellows-sealed corner valve with ISO-KF 16 connectors.The system is connected to a dry scroll pump (Edwards; nXDS6i) and a turbomolecular pump (Pfeiffer Vacuum; HiPace 80) via an ISO-KF 25 flexible stainless-steel hose, allowing it to reach a ≤10 −4 mbar vacuum in the setup (Figure 1).
The inlet system is operated by a Raspberry Pi mini-computer (Raspberry Pi 4 Model B Rev 1.1 4GB).Besides communicating with the Windows PC of the TILDAS, the Raspberry Pi controls an Arduino (Arduino MEGA 2560 Rev3) microcontroller, which interfaces with the valves, relays, the stepper motors for bellow positioning, and connected sensors (Figure S1 in Supporting Information S1).On receiving a command from the Raspberry Pi, the Arduino opens and closes the pneumatic valves of the inlet system using Darlington drivers (STMicro; ULN2803A) and solenoid valve actuators (SMC; V100 series).The Arduino expands and compresses the bellows using NEMA17 stepper motors controlled by DRV8825 (Texas Instruments) drivers.The peristaltic pump (Grothen; G528, 12 V, 3 × 5 mm) of the air inlet system is turned on and off using a 5 V relay.
Two capacitance manometers (MKS Instruments; Baratron 623H, 10 Torr) are used to measure the CO 2 pressure in the bellows, while the pressure of the collision gas (gauge A on Figure 1) is measured by a third manometer (MKS Instruments; Baratron 626B, 1,000 mbar).Because the manometers produce a 0-10 V output signal that is read by the Arduino's 0-5 V analog in ports, only pressures below 5 Torr and 500 mbar are recorded, respectively.To increase the resolution of the readings, pressure data are averaged over 1 s intervals.The pressure in the absorption cell is measured using the TILDAS's built-in capacitance manometer (MKS Instruments; Baratron 722B, 100 Torr).An active Pirani gauge measures the vacuum in the inlet system (Edwards; APG100-XLC, 10 −4 ≤ p ≤ 10 3 mbar).
For thermal stability during the measurements, the TILDAS and its peripherals, including the chiller (Solid State Cooling Systems; Oasis Three) and the inlet system, are placed in a 5-cm-thick styrofoam-insulated wooden box.The temperature inside the box is measured using an SHTC3 sensor (Adafruit Industries) connected to the Arduino via I 2 C, and regulated to 32.00°C (±0.01°C; ±1σ of the temperature data over a ca.2.5-hr-long measurement; Figures S3g and S6g in Supporting Information S1) using two 80 mm housing fans, which blow cool (ca.25°C) laboratory air into and ventilate warm air out of the box, respectively (see Section 4.1).Specifically, the speed (i.e., the current) of the fans is PID-controlled by the Arduino and an adjustable power supply (Manson, HCS-3304).The ambient temperature outside the housing was monitored using a BME688 (Pimoroni) sensor connected to the Raspberry Pi via I 2 C.

Software
The inlet system is operated by open-source software.The measurement procedure is controlled via the front panel (Figure 1), a JavaScript web application running on the Raspberry Pi (Figure 2).An Apache HTTP server provides the browser front end.We run the web application within the browser of the Raspberry Pi and access it via Microsoft Remote Desktop.
Information exchange between the JavaScript web application and the hardware interfaces is done in two steps via a Python (v.3.7) and a PHP (v.8.2) script (Figure 2).A continuously looping Python program (ca. 12 Hz) receives a status string from the Arduino (valve status, position of the bellows, pressure data, etc.) and sends back commands (actuating valves, moving bellows, etc.) via serial.The PHP script exchanges data with the Python script via shared variables, and the JavaScript web application via AJAX calls.
The Aerodyne Research TDLWintel software on the TILDAS PC controls the laser spectrometer and performs the spectra fitting.The TILDAS PC is run headless but can be accessed via the Remote Desktop if necessary.The TDLWintel software sends status strings containing the measured isotopologue mixing ratios and the cell pressure to the Raspberry Pi via serial.It receives commands, that is, to start or stop saving spectral data, from the TILDAS through TCP via a PHP script (Figure 2).Clock synchronization between the TILDAS PC and the Raspberry Pi was achieved using a shared local time server.
When a measurement is finished, the saved STC and STR files (containing the mixing ratios, among other information) are copied from the TILDAS PC to the Raspberry Pi.A Python script reads these files and uses them for data reduction (Figure 2).Finally, all data files and the evaluated results are uploaded to a MySQL database, which can be accessed online.

Gas Mixing and Inlet
The entire gas mixing and admission procedure is automated.The sequence of commands is stored in CSV files read by the JavaScript web application.For carbonate-derived CO 2 samples, a manifold containing the sample CO 2 gas is connected to the inlet system via an UltraTorr connector (behind V22; Figure 1).The pneumatic valves of the manifold are actuated automatically during sample bellow refill.To refill the reference bellows, aliquots of the reference CO 2 are taken automatically from tanks placed inside the housing and connected to the inlet system via stainless-steel capillaries.The pure CO 2 gas (sample and reference) is mixed with the collision gas before expansion into the absorption cell.First, the sample or reference CO 2 is expanded into a small volume (for the reference, the cross-piece enclosed by V3, V5, V6, V7; and for the sample V9, V11, V12, V13; Figure 1).The pressure in these volumes is adjusted by moving bellows X and Y, respectively.The collision gas is first expanded to the volume enclosed by V19 and V20 before an aliquot is expanded to the volume between valves V07, V08, V13, V14, V15, and V16.The amount of collision gas is adjusted by moving bellow Z.The pure CO 2 and the collision gas are mixed turbulently by first expanding the collision gas into the cross-piece holding the CO 2 (opening V7 for reference or V13 for sample) and then expanding the analyte to the absorption cell.With this procedure, the mixing ratio (pCO 2 ; here, the concentration of the "626" isotopologue; ca.420 ppmv) can be adjusted to better than ±0.5 ppmv for reference-sample pairs (Figures S3c, S6c, and S11c in Supporting Information S1).Finally, the pressure in the cell was adjusted using bellow Z to ca. 42.100 Torr.For a single measurement, sample and reference analyte pressures are matched within ±0.002 Torr (Figures S3d, S6d, and S9d in Supporting Information S1).The mixing and gas changeover takes approximately 4 min.
The inlet system allows the automated sampling of ambient air by pumping it from outside the laboratory using a peristaltic pump throughout a 6 mm polyurethane tube at a rate of approximately 0.2 L min −1 (V32; Figure 1).Water vapor has a considerable peak-broadening effect (Tan et al., 2019).Since air humidity is highly variable, it is necessary to dry the air to maintain a consistent analyte composition across multiple measurements.For this reason, the air is passed through a 1-cm-inner-diameter and 20-cm-long glass tube containing a drying agent and subsequently a 0.5 μm filter.Although we used coarse anhydrous magnesium perchlorate during the test measurements, other methods, such as a Nafion gas dryer, may be more suitable for drying the sample without affecting its isotopic composition (cf., Paul et al., 2020).To flush the line before sampling, fresh air is pumped through the polyurethane tubing and the drying agent.Then, a ca. 1 L volume between V30 and V21 was filled with air for 5 min.By opening V15 and V21, the volume between valves V07, V08, V13, V14, and V16 is slowly filled to a pre-set target pressure through a crimped capillary.The pCO 2 of the air sample is recorded after it is expanded into the optical cell, which allows automatic adjustment of the mixing ratio of the reference gas during the following measurement cycle.

Sample Material
To test the repeatability and stability of the instrument, we report data from two pure CO 2 gases, a speleothem calcite sample, and air CO 2 .Pure CO 2 gases were taken from gas bottles connected to the inlet system (Figure 1).The "light CO 2 " (δ 18 O = −0.92‰VSMOW was produced by equilibrating CO 2 with isotopically light Antarctic precipitates.The "heavy CO 2 " (δ 18 O = 77.03‰VSMOW) was equilibrated with isotopically heavy water that remained after evaporating tap water.
CO 2 was extracted from a speleothem calcite sample offline using acid digestion with phosphoric acid at 25°C.For each extraction, ca. 2 mg of carbonate powder reacted overnight with >103% phosphoric acid in an inverse Y-shaped reactor vessel (McCrea, 1950).The liberated CO 2 was purified using a −80°C ethanol slush trap and frozen out in the sample manifold.
Air for the automated air CO 2 measurements was collected from the fourth floor balcony of the stable isotope laboratory in Göttingen, Germany.

Data Reduction
During an approximately 4-min-long measurement cycle, the oxygen isotope ratios of the analyte are calculated every second.Specifically, the TDLWintel software calculates mole fractions (X; also referred to as mixing ratios) of the three investigated molecular species ("626," "627," and "628") from the absorption spectra through spectral fitting (McManus et al., 2006).For the spectral fitting, the concentration of the "free-path CO 2 " species in the TDLWintel was set to zero.The isotope ratios (R) are calculated from the mole fractions (Equation 3): (3) To calculate isotope deltas, the isotope composition of the sample CO 2 was compared with the isotope composition of the reference.We tested two data reduction approaches to address instrumental drift: "reference-sample-reference bracketing" and "smooth drift correction."In the case of each approach, for a replicate analysis, we measure reference and sample cycles consecutively multiple times.The first "dummy" measurement cycle is always discarded because it often deviates from the subsequent reference cycles.A 3σ outlier test is performed on the ca.240 data points in each measurement cycle.
The traditional way of calculating isotope ratios is to bracket sample measurement cycles between reference cycles, with quick changeovers in-between (McKinney et al., 1950).A replicate measurement consists of an odd number of gas analysis cycles, with the odd cycles reserved for reference gas measurements and the even cycles used for sample gas measurements (Figures S4 and S10 in Supporting Information S1).This technique accounts for the instabilities of the instrument by comparing the isotope ratio R (Equation 3) of the sample with the mean isotope ratio of a reference that is measured, presumably, under the same conditions immediately before and after the sample: For our "bracketing" analyses, we measured a total of n = 21 cycles.The internal error of the "bracketing" replicate is calculated as the 1 standard error of the Δ′ 17 O values of the cycles (Figures S5 and S11 in Supporting Information S1).
The drift of the instrument, however, may follow a smooth path, for example, in the form of a sinusoid function.
While the bracketing approach can deal with such a situation, it is not the correct mathematical description.Instead, the reference and sample would follow a parallel trend with a constant offset in the intercept.One could approximate the sinusoid drift by fitting polynomial models with identical coefficients but unique intercepts to all sample and reference data points.The difference in the intercepts of the two polynomial fit models provides an apparent isotope fractionation factor (α) from which the raw isotope ratios are calculated.
For a "smooth drift" replicate, we measured two sample cycles interspersed with reference cycles and fitted a second-order polynomial model on the data (Figures S6 and S7 in Supporting Information S1).Our initial testing showed that measuring more than two sample cycles hindered a satisfactory polynomial fit of any degree, due to the rapid and non-linear drift of the measurements (see Section 4.1).The internal error of the "smooth drift" measurements was calculated by taking the root mean squared standard deviations of the differences in Δ′ 17 O between the polynomial models and the reference or sample data points.
The isotope compositions referenced against a working gas are typically recalculated relative to an international standard.However, the δ 17 O value of our working reference gas is unknown.As a working reference gas, we use a commercial CO 2 (Linde Gas; CO 2 4.5) with a δ 18 O value of 27.87‰ VSMOW.Although a nominal Δ′ 17 O value of −90 ppm was assigned to our working reference based on high-resolution gas source mass spectrometry (cf., Adnew et al., 2019), we only report unreferenced and unscaled isotope ratios.

Effect of Temperature Instability
Initially, we placed the TILDAS and the inlet system in an air-conditioned laboratory with a temperature instability of ca.1.5°C h −1 .This led to a considerable 1,400 ppm fluctuation in the measured Δ′ 17 O values over an 8-hr interval (Figure 3).The 1.5°C h −1 variation in ambient room temperature caused by the air conditioning (Figure 3a) was inherited by the coolant and cell temperatures, varying, in turn, by ca.0.06°C h −1 (Figure 3b).Although the coolant temperature was in phase with the ambient room temperature, the cell temperature showed about 20 min lag because of the buffering effect of the hardware around the cell's temperature sensor (Figure 3b).Previous studies have also noted the detrimental effects of unstable cell and electronics hardware temperatures on the repeatability of the Δ′ 17 O measurements by laser spectroscopy (Bowling et al., 2003;Hare et al., 2022;Stoltmann et al., 2017).
Although the intensities of the absorption lines resembled the variability in the cell temperature (Figure 3c), the absorbance of the three isotopologues exhibited different sensitivities to temperature changes (Figure S2 in Supporting Information S1).This irregular behavior resulted in uncorrelated variations in δ 17 O and δ 18 O (Figure 3d), which funneled into the large approximately 1,400 ppm variations in Δ′ 17 O (Figure 3e).The observed non-linear short-scale drift could not be handled by either correction method.Attempts to apply a cell temperature-based correction proved challenging, likely because the measured temperature does not precisely match the actual gas temperature.
Instead, we aimed to minimize temperature fluctuations.Placing the TILDAS and the inlet system in an insulated box equipped with active temperature control (see Section 2.2) led to stable cell temperatures (±2 mK over a ca.2.5-hr-long measurement) and consequently a dramatically reduced drift in the isotope ratios (Figures S3,S6,and S9 in Supporting Information S1).The remaining drift could be effectively managed by frequent sample-reference changeovers.

Effect of pCO 2 Mismatch
We noticed that a mismatch between the reference and the sample analyte's pCO 2 -introduced by inaccurate mixing of CO 2 and collision gas-influenced the Δ′ 17 O measurements.Perdue et al. (2022) also investigated the impact of pCO 2 mismatch but found no decipherable effect within ±25 ppmv (see their Figure 8).To investigate this for our setup, a series of heavy and light CO 2 measurements were taken with intentionally introduced mixing ratio mismatch while maintaining the overall cell pressure constant.The data were evaluated using the "bracketing" approach.
Our results show a clear negative correlation between the measured Δ′ 17 O values and the magnitude of the pCO 2 mismatch, with a slope of ca.−6 ppm ppmv −1 (Figure 4).For the heavy and light CO 2 measurements, where the sample and reference pCO 2 were precisely matched, the average internal error was approximately ±5 ppm (Figure 4 insets).Consequently, a pCO 2 mismatch larger than ±1 ppmv would introduce a bias larger than the internal error.
We opted to minimize the pCO 2 mismatch during gas mixing rather than correcting the measured data.Thanks to our fully automated inlet system, achieving pressure matching at a level where pCO 2 mismatches are negli- 10.1029/2023GC010976 9 of 13 gible (i.e., better than ±0.5 ppmv) was a straightforward task involving just a few lines of code.Additionally, since pressure matching takes place simultaneously with the measurement, this procedure does not introduce any delays into the measurement sequence.

Analysis of Pure CO 2 Mixed With Collision Gas
To demonstrate the long-term repeatability of each data reduction approach, we report Δ′ 17 O data from two measurement periods.We tested the "smooth drift" approach first as it offered a short measurement time for a single replicate analysis (ca.40 min; Figures S6 and S7 in Supporting Information S1).The internal error of the "smooth drift" measurements was ca.±4 ppm, whereas the external repeatability (±1σ) over a ca.two-week period was ca.±15 ppm (Figure S8 in Supporting Information S1).For the "bracketing" measurements (e.g., Figures S3-S5 in Supporting Information S1), the external repeatability of the data (±1σ) over a ca.one-month period was ±9 ppm, while the mean internal error was ±6 ppm (Figures 5a and 5b).The reason why the "smooth drift" approach performs worse than "bracketing" is probably because the instrument, despite the temperature stabilization, drifts at a faster rate than what a polynomial fit can accurately track.Despite the large difference in the δ 18 O values between the light and heavy CO 2, neither data reduction approach shows a resolvable drift in the Δ′ 17 O values (Figures 5a and 5b; Figure S8 in Supporting Information S1).This suggests that if instrumental conditions (e.g., cell temperature) are stable, the scale compression (i.e., the difference in the δ-values) of the instrument does not change rapidly.Once their isotope composition is accurately determined, the light and heavy reference CO 2 can be used for scaling the data.
Because the "bracketing" approach resulted in better external repeatability, we used it for evaluating speleothem measurements (Figure 5c).A total of 14 analyses of carbonate-derived CO 2 were performed, obtained from seven acid digestions, that is, each aliquot was analyzed twice.The analyses yielded an external repeatability of ±8 ppm (±1σ).

Analyses of Air CO 2
The concentration of atmospheric CO 2 can fluctuate by as much as 100 ppmv due to variations in urban CO 2 fluxes and vegetation uptake (e.g., George et al., 2007;Hofmann et al., 2017).Consequently, continuous monitoring requires adjusting the pCO 2 of the reference gas mixture to match the ambient air pCO 2 (Figure 6a).To achieve this, we used a modified "bracketing" measurement procedure, where the amount of working gas mixed with collision gas was automatically adjusted based on the pCO2 of the first air cycle.The first three measurement cycles (dummy and two adjustment cycles) were discarded (Figures S9-S11 in Supporting Information S1).
The mean internal error of the air CO 2 Δ′ 17 O measurements was similar to that of pure CO 2 mixed with collision gas, that is, ±6 ppm (Figure 6b).Compared to CeO 2 equilibration (Hofmann et al., 2017) and high-temperature fluorination (Thiemens et al., 2014), laser spectroscopy offers a three-fold improvement in precision.In addition, rapid, consecutive measurements of air CO 2 can potentially resolve daily variations in its ∆' 17 O value, opening new avenues for atmospheric research (Figure 6).

Conclusions
We present an adaptable, fully automated dual inlet system for tunable infrared laser direct absorption spectroscopy (TILDAS) that uses inexpensive hardware and open-source software.The setup implements the identical treatment of sample and reference by accurately mixing the analyte CO 2 with collision gas before each analysis cycle, and thereby allows high-precision (1σ ≤ 10 ppm) triple oxygen isotope (∆′ 17 O) measurements of pure CO 2 (e.g., from acid digestion of carbonates) and air CO 2 .By matching the reference gas mixing ratios to that of ambient air, the setup enables the continuous monitoring of the triple oxygen isotope composition of air CO 2 .Our study shows that laser spectroscopy can generate high-precision data for Earth system applications, provided that sample and reference measurements are conducted under identical conditions.

Figure 1 .
Figure 1.The front panel with a schematic diagram of the fully automated dual inlet system.The image is a screenshot of the graphical user interface, a website that runs within any internet browser.

Figure 2 .
Figure 2.An overview of the relationships between the dual inlet software and hardware.

Figure 3 .
Figure 3.The effect of changing ambient temperatures on the measured isotope ratios.The fluctuating room temperature (a, ΔT ≈ 1.5°C) is mirrored (b, ΔT ≈ 0.06°C) in the coolant and with a lag (∆t = 20 min, blue arrows) in the cell temperatures.Variations in the cell temperatures led to uncorrelated variations in the measured mixing ratios (c) and the δ-values (d).The Δ′ 17 O (e) roughly follows the changing cell temperature and shows variations of approximately 1,400 ppm.

Figure 4 .
Figure 4. Sensitivity of the Δ′ 17 O values to the mismatch between the sample and reference analyte's pCO 2 .

Figure 5 .
Figure5.Triple oxygen isotope measurements evaluated using the "bracketing" approach.∆′ 17 O data from two CO 2 gases (a), (b) and a carbonate (c), relative to the working reference gas.

Figure 6 .
Figure 6.Results of automated air CO 2 measurements.(a) Cell pCO 2 averaged for the eight sample and nine reference cycles in each replicate measurement.(b) Δ′ 17 O values of air CO 2 , referenced against the working gas.(c) δ 18 O values of air CO 2 , referenced against the working gas.