Calibrated high-precision 17 O-excess measurements using cavity ring-down spectroscopy with laser-current-tuned cavity resonance

High-precision analysis of the 17O /16O isotope ratio in water and water vapor is of interest in hydrological, paleoclimate, and atmospheric science applications. Of specific interest is the parameter 17Oexcess ( 117O), a measure of the deviation from a linear relationship between 17O /16O and18O /16O ratios. Conventional analyses of 117O of water are obtained by fluorination of H 2O to O2 that is analyzed by dual-inlet isotope ratio mass spectrometry (IRMS). We describe a new laser spectroscopy instrument for highprecision 117O measurements. The new instrument uses cavity ring-down spectroscopy (CRDS) with laser-currenttuned cavity resonance to achieve reduced measurement drift compared with previous-generation instruments. Liquid water and water-vapor samples can be analyzed with a better than 8 per meg precision for 117O using integration times of less than 30 min. Calibration with respect to accepted water standards demonstrates that both the precision and the accuracy of 117O are competitive with conventional IRMS methods. The new instrument also achieves simultaneous analysis of δ18O, δ17O and δD with precision of < 0.03 ‰, < 0.02 and< 0.2 ‰, respectively, based on repeated calibrated measurements.


Introduction
Measurements of the stable isotope ratios of water are ubiquitous in studies of earth's hydrological cycle and in paleoclimatic applications (Dansgaard, 1964;Dansgaard et al., 1982;Johnsen et al., 1995;Jouzel et al., 2007).Isotopic abundances are reported as deviations of a sample's isotopic ratio relative to that of a reference water, and expressed in the δ notation as where 2 R = n( 2 H)/n( 1 H), 18 R = n( 18 O)/n( 16 O), 17 R = n( 17 O)/n( 16 O), and n refers to isotope abundance.One important innovation was the development by Merlivat and Jouzel (1979) of a theoretical understanding of "deuterium excess": where δD is equivalent to δ 2 H.The deuterium excess is commonly used as a measure of kinetic fractionation processes.For example, deuterium excess variations from ice cores have been used to infer variations in evaporative conditions over the ocean surface areas from which polar precipitation is derived (Johnsen et al., 1989;Petit et al., 1991;Vimeux et al., 2001;Masson-Delmotte et al., 2005).The δ 18 O and δD isotopic values can be experimentally determined via a number of isotope ratio mass spectrometry (IRMS) techniques.For δ 18 O, equilibration with CO 2 has been the standard method for many decades (Cohn and Urey, 1938;McKinney et al., 1950;Epstein, 1953).For δD, reduction of water to H 2 over hot U (Bigeleisen et al., 1952;Vaughn et al., 1998) or Cr (Gehre et al., 1996) has typically E. J. Steig et al.: 17 O-excess measurements by laser spectroscopy been used.Simultaneous determination of δ 18 O and δD was made possible via the development of continuous-flow massspectrometric techniques utilizing conversion of water to CO and H 2 in a pyrolysis furnace (Begley and Scrimgeour, 1997;Gehre et al., 2004).
A recent innovation is the measurement of the difference between δ 18 O and δ 17 O at sufficiently high precision to determine very small deviations from equilibrium.In general, the nuclei mass differences of +1n 0 and +2n 0 (n 0 denotes a neutron) imply that the fractionation factor for δ 17 O between two different phases will be approximately the square root of the fractionation factor for δ 18 O (Urey, 1947;Craig, 1957;Mook, 2000): where λ = 0.5010-0.5305(Kaiser, 2008) and the subscripts "a" and "b" refer to different phases or samples.For isotopic equilibrium, the value of λ will approach θ, given theoretically by the ratio of the partition functions (Q), which in the limit of high temperature approaches a constant value given as follows (Matsuhisa et al., 1978): where m 16 is the atomic mass of 16 O, m 17 is that of 17 O, etc.1 By analyzing a set of meteoric waters, Meijer and Li (1998) estimated the value of λ to be 0.528.Barkan and Luz (2005) used careful water equilibrium experiments to determine an equilibrium value for λ of 0.529, while Barkan and Luz (2007) showed that λ is 0.518 under purely diffusive conditions, in good agreement with theory (Young et al., 2002).Thus, the Meijer and Li (1998) value of 0.528 for meteoric waters reflects the combination of equilibrium and diffusive processes in the hydrological cycle.
Based on these observations, Barkan and Luz (2007) defined the 17 O excess parameter as the deviation from the meteoric water line with slope of 0.528 in ln(δ + 1) space: 17 O = ln(δ 17 O + 1) − 0.528 ln(δ 18 O + 1). (5) Like deuterium excess, 17 O excess is sensitive to kinetic fractionation but, unlike deuterium excess, it is nearly insensitive to temperature and much less sensitive than δD and δ 18 O to equilibrium fractionation during transport and precipitation.Natural variations of 17 O in precipitation are orders of magnitude smaller than variations in δ 18 O and δD and are typically expressed in per meg (10 −6 ) rather than per mil (10 −3 ).
The potential of 17 O in hydrological research is significant because it provides independent information that may be used to disentangle the competing effects of fractionation during evaporation, in transport, and in the formation and deposition of precipitation (Landais et al., 2008;Risi et al., 2010;Schoenemann et al., 2014).It also has applications in atmospheric dynamics because of the importance of supersaturation conditions that, during the formation of cloud ice crystals, impart a distinctive isotope signature to water vapor (e.g., Blossey et al., 2010;Schoenemann et al., 2014).
Compared to the routine nature of δ 18 O and δD analysis, isotopic ratio measurements of 17 O, the second heavy isotope of oxygen in terms of natural abundance, are challenging.The greater abundance of 13 C than 17 O makes the measurement of δ 17 O in CO 2 equilibrated with water by IRMS at m/z = 45 impractical.As a result, the precise measurement of 17 O requires conversion of water to O 2 rather than equilibration with CO 2 or reduction to CO. Meijer and Li (1998) developed an electrolysis method using CuSO 4 .More recently, Baker et al. (2002) used a fluorination method to convert water to O 2 , which was analyzed by continuous-flow IRMS; this approach was updated by Barkan and Luz (2005) for dual-inlet IRMS.
The dual-inlet IRMS method can provide high-precision and high-accuracy 17 O measurements.However, the technique is time consuming, resulting in significantly lower sample throughput when compared to the standard and relatively routine analysis of δ 18 O and δD.The fluorination procedure requires 30 min or more per sample, while the dualinlet mass-spectrometric analysis requires 2-3 h.In practice, multiple samples must be processed because of memory effects in the cobalt-fluoride reagent and other issues that can arise in the vacuum line (e.g., fractionation during gas transfer) (Barkan and Luz, 2005).Moreover, while this method provides the most precise available measurements of 17 O, measurements of individual δ 18 O values by this method are generally less precise than those obtained with other approaches.
In recent years, laser absorption spectroscopy in the nearinfrared and mid-infrared regions has increasingly been used for isotope analysis.An overview of experimental schemes for different molecules and isotopologues can be found in Kerstel (2004).In the case of water, laser absorption spectroscopy constitutes an excellent alternative to mass spectrometry.The main advantage is the ability to perform essentially simultaneous measurements of the water isotopologues directly on a water-vapor sample.As a result, tedious sample preparation and conversion techniques are not necessary.Commercialization of laser absorption spectrometers has recently allowed measurements of water isotope ratios to be performed with high precision and competitive relative accuracy, provided that a valid calibration scheme is applied (Brand et al., 2009;Gupta et al., 2009;Gkinis et al., 2010Gkinis et al., , 2011;;Schmidt et al., 2010;Aemisegger et al., 2012;Kurita et al., 2012;Wassenaar et al., 2012).
The measurement of 17 O / 16 O ratios should in principle not pose any additional challenges when compared to the measurement of 18 O / 16 O and D / H. Provided that the absorption lines of interest are accessible by the laser source with no additional interferences from other molecules, a triple isotope-ratio measurement can be performed, resulting in calibrated values for δ 18 O, δ 17 O and δD.In fact, triple isotope-ratio measurements of water have been presented in the past via the use of various laser sources utilizing different optical and data analysis techniques (Kerstel et al., 1999(Kerstel et al., , 2002(Kerstel et al., , 2006;;Van Trigt et al., 2002;Gianfrani et al., 2003;Wu et al., 2010).However, with the exception of results presented recently by Steig et al. (2013) and Berman et al. (2013), precision has not been sufficient to be useful for applications requiring the detection of the very small natural variations in 17 O.
In this work we report on development of a new cavity ring-down laser absorption spectrometer that provides both high-precision and high-relative-accuracy measurements of 17 O.The instrument we discuss here is a modification, first described by Hsiao et al. (2012) and Steig et al. (2013), of the Picarro Inc. water isotope analyzer model L2130-i.It is now commercially available as model L2140-i.Critical innovations we introduced include the use of two lasers that measure absorption in two different infrared (IR) wavelength regions, and modifications to the spectroscopic measurement technique.We also developed a sample introduction system that permits the continuous introduction of a stable stream of water vapor from a small liquid water sample into the optical cavity.In combination with precise control of the temperature and pressure in the optical cavity of the instrument, data averaging over long integration times results in precision of better than 8 per meg in 17 O.We establish the relative accuracy of our results in comparison with IRMS measurements.This work can also be seen as a demonstration of state-of-theart performance for laser absorption spectroscopy isotope ratio analysis for all four main isotopologues of water (H where δ 18 O assigned SLAP = −55.5 ‰ is the value assigned by the International Atomic Energy Agency (Gonfiantini, 1978;Coplen, 1988).δD is normalized in the same manner, using δD assigned SLAP = −428 ‰.We normalize δ 17 O using There is no IAEA (International Atomic Energy Agency)defined value for δ 17 O assigned SLAP , but Schoenemann et al. ( 2013) recommended that it be defined such that SLAP 17 O is precisely zero.We follow that recommendation here; that is, we define which yields δ 17 O assigned SLAP = −29.6986‰, well within the error of published measurements (Barkan and Luz, 2005;Kusakabe and Matsuhisa, 2008;Lin et al., 2010;Schoenemann et al., 2013) after normalization to the associated δ 18 O values (Schoenemann et al., 2013).
Throughout this paper, reported values of δ 18 O, δ 17 O, δD and 17 O have been normalized as described above unless specifically noted otherwise.Superscripts and subscripts are omitted except where needed for clarity.

17 O analysis with mass spectrometry
IRMS measurements provide the benchmark for comparison with results from analysis of 17 O by CRDS (cavity ringdown spectroscopy).We used IRMS to establish accurate measurements of 17 O of five laboratory working standards and the IAEA reference water GISP (Greenland Ice Sheet Precipitation), relative to VSMOW and SLAP.We also used both IRMS and CRDS measurements to determine the δD and δ 18 O of the same standards; δ 17 O is calculated from the 17 O and δ 18 O data.Table 1 reports the values, updated from those in Schoenemann et al. (2013).We used the method described in Schoenemann et al. (2013) to convert water to O 2 by fluorination, following procedures originally developed by Baker et al. (2002) and Barkan and Luz (2005).A total of 2 µL of water are injected into a nickel column containing CoF 3 heated to 370 • C, converting H 2 O to O 2 , with HF and CoF 2 as byproducts.The O 2 sample is collected in a stainless steel cold finger containing 5A molecular sieve following Abe (2008).To minimize memory effects, a minimum of three injections are made prior to collecting a final sample for measurement.The O 2 sample is analyzed on a ThermoFinnigan MAT 253 dual-inlet mass spectrometer at m/z = 32, 33, and 34 for δ 18 O and δ 17 O, using O 2 gas as a reference.Each mass-spectrometric measurement comprises 90 sample-toreference comparisons.Precise adjustment of both sample and reference gas signals (10 V±100 mV) permits long-term averaging with no measurable drift, so that the analytical precision is given by simple counting statistics: σ/ √ 90, where σ is the standard deviation of the individual sample/reference comparisons.The resulting precision of repeated measurements of O 2 gas is 0.002, 0.004, and 0.0037 ‰ (3.7 per meg) for δ 17 O, δ 18 O, and 17 O, respectively.Reproducibility of the δ 17 O and δ 18 O ratios of water samples is in practice less precise than these numbers indicate, because fractionation can occur during the fluorination process or during the collection of O 2 .However, because this fractionation closely follows the relationship ln(δ 17 O + 1) = 0.528 ln(δ 18 O + 1), the errors largely cancel in the calculation of 17 O (Barkan and Luz, 2005;Schoenemann et al., 2013).The reproducibility of the calibrated 17 O of repeated water samples ranges from 4 to 8 per meg (Schoenemann et al., 2013).

2.3
17 O analysis with cavity ring-down spectroscopy

Instrument design
We used modified versions of a CRDS analyzer designed for δ 18 O and δD, commercially available as model L2130i, manufactured by Picarro Inc.The L2130-i is an update to the water-isotope analyzers originally discussed in Crosson (2008).It uses an Invar (Ni-Fe) optical cavity coupled to a near-infrared laser.Optical resonance is achieved by piezoelectric modifications to the length of the cavity.When the intensity in the cavity reaches a predetermined value, the laser source is turned off and the intensity then decays exponentially.The time constant of this decay is the "ringdown time".The ring-down time depends on the reflectivity of the mirrors, the length of the cavity, the mixing ratio of the gas being measured, and the frequency-dependent absorption coefficient.The frequency is determined with a wavelength monitor constructed on the principle of a solid etalon (Crosson et al., 2006;Tan, 2008).Determination of δ 18 O and δD ratios on the model L2130-i is obtained by measurements of the amplitude of H 2 18 O, H 2 16 O and HDO spectral lines from a laser operating in the area of 7200 cm −1 (wavelength 1389 nm).In a modified version, which we refer to as the L2130-i-C, we added a second laser that provides access to another wavelength region, centered on 7193 cm −1 , where there are strong H 2 17 O and H 2 18 O absorption lines (Fig. 1).Rapid switching between the two lasers allows the measurement of all three isotope ratios essentially simultaneously.About 200-400 ring-down measurements are made per second, and complete spectra covering all four isotopologues are acquired in 0.8 s intervals.
For isotope measurements with the L2130-i or L2130-i-C under normal operating conditions, water vapor in a dry air or N 2 carrier gas flows continuously through the cavity to maintain a cavity pressure of (66.7 ± 0.1) hPa at a temperature of (80 ± 0.01) • C, normally at a H 2 O mixing ratio of 20 mmol mol −1 .The flow rate of 40 cm 3 min −1 (290 K, 10 5 Pa) is maintained by two proportional valves in a feedback loop configuration up-and down-stream of the optical cavity.The spectral peak amplitudes are determined from the least-squares fit of discrete measurements of the absorption (calculated from measurements of the ring-down time) to a model of the continuous absorption spectrum.
The spectroscopic technique utilized for the acquisition and analysis of the spectral region relevant to the measurement of the isotopologues of interest is essentially the same as the one used in the earlier commercially available L2130-i analyzer.One of the main features of this technique is that optical resonance is obtained by dithering the length of the cavity.As discussed in Results (Sect.3), we found that drift on timescales longer than a few minutes limited the achievable precision of 17 O measurements to about 20 per meg; this drift is ascribed to small but detectable drift in the wavelength monitor.
To improve measurement precision, we developed an updated version of the L2130-i-C, hereafter referred to as model L2140-i, which incorporates a different spectroscopic method.As in the L2130-i, a piezoelectric actuator is used to physically move one mirror of the cavity, and the wavelength monitor is used for feedback to the laser-frequency control electronics, thus allowing for rapid tuning to a target frequency.In the new method, though, the length of the optical cavity is kept constant during the acquisition of a spectrum, and resonance is obtained by dithering of the laser frequency by means of laser-current modulation.The frequency for each ring-down measurement is then determined directly from the resonance itself, based on the principle that resonance will occur only at frequencies spaced by integer multiples of the free spectral range (FSR) of the cavity (e.g., Morville et al., 2005).
The target frequency for each spectral region (e.g., that for H 2 17 O) is determined in advance from measurements made at higher frequency resolution and used to tightly constrain the parameters in a spectral model (see below).The fine frequency spacing for a given narrow spectral region is determined only by the FSR.In this way, each ring-down measurement can be unambiguously assigned to a stable and equidistant frequency axis and the spectral line shape fit to a well-defined model; only a few data points are needed to precisely define each spectral peak.This new scheme also yields higher cavity excitation rates -typically 500 ring-downs per second.
The FSR is inversely proportional to the cavity length.The FSR of the L2130-i and L2140-i under normal operating con-ditions is 0.02 cm −1 , and varies by no more than 10 −5 cm −1 owing to the precisely controlled temperature and pressure conditions.The cavity finesse is 44 000.The ring-down time constant for an empty cavity is 22 µs, corresponding to an effective optical path length of 6.7 km.Each ring-down measurement has a frequency resolution of 14 kHz (given by the FSR divided by the cavity finesse).The noise-equivalent absorption spectral density is 2.3×10 −11 cm −1 Hz −1/2 for both the L2130-i and L2130-i-C, and the L2140-i instruments.This corresponds to a noise-equivalent absorption of only 7 × 10 −13 cm −1 for integration times of 10 3 s.

Spectroscopy
The use of laser-current tuning permits greater accuracy in the determination of the width of spectral lines than was achievable with the L2130-i or L2130-i-C instruments.This allows us, with the L2140-i, to use the integrated absorption under the spectral lines, rather than the height of spectral peaks, to determine isotopologue abundances (e.g., Kerstel, 2004;Kerstel et al., 2006;Hodges and Lisak, 2007).
The integrated absorption (cm −1 ) is given by where κ(ν, T , P ) is the molecular monochromatic absorption coefficient (cm 2 ), u is the column density of absorbers (cm −2 ) and ν is the wavenumber (cm −1 ) (Rothman et al., 1996).The integrated absorption is directly related to the absorption strength, S, via where f is the line shape function due to Doppler and pressure spectral line broadening, T is temperature and P is pressure.The integral ∞ 0 f (ν, T , P )dν = 1 and S is independent of pressure (Rothman et al., 1996).The ratios A i /A j for two different absorbing isotopologues i and j -and therefore in principle the isotope ratios -are also independent of pressure.This makes the integrated absorption superior to the spectral peak amplitude used in earlier-generation instruments.In practice, it is convenient to replace the wavenumber, ν, in the integral with the dimensionless detuning x = (ν − ν0 )/σ D , following Varghese and Hanson (1984), where ν0 is the center frequency of the absorption line, and σ D is the Doppler width (half-width of the Gaussian Dopplerbroadening profile at 1/e of the height).
Values of A are obtained by a least-squares fit of the measurements to an empirically determined spectral model.The spectral model describes the measured absorption as the sum of a baseline and molecular absorption lines.Free parameters in the baseline are an offset, slope and quadratic curvature term.The molecular absorption spectrum is modeled as the superposition of Galatry profiles, which describe the shape function, f , for each spectral line.The Galatry profile, G, is given by the real part of the Fourier transform of the correlation function, (Galatry, 1961): where x is the frequency separation from the line center normalized by the Doppler width (as given above), y and z are collisional broadening and narrowing parameters, respectively, and τ is dimensionless time.
The parameters that determine the shape of the lines are obtained from spectra acquired by operating the analyzer in a fine-scan mode where ring-downs are acquired with a frequency spacing much smaller than the line width and using the wavelength monitor to determine the frequency axis.This determines the relationship between the collisional broadening and narrowing parameters, y and z, and the relationship between y for the "normal" water peak (H 2 16 O) and the values of y for each of the isotopologues.The Doppler width is a known function of temperature (e.g., Galatry, 1961) and is therefore a fixed parameter.This leaves three or four free parameters needed to describe absorption for unknown samples in each spectral region: one y parameter and one value for the integrated absorption, A, for each independent isotopologue spectral line of interest (e.g., one each for the H 2 18 O, H 2 16 O and HDO lines in the 7200 cm −1 wavenumber region).

Determination of isotope ratios
For the determination of δ 18 O and δ 17 O, the 18 O / 16 O and 17 O / 16 O ratios are obtained from the ratios of integrated absorptions of the rare isotopologues on the second laser to the integrated absorption of the common isotopologue on the first laser: where H 2 18 O(11), H 2 16 O(2), etc. refer to the absorption lines shown in Fig. 1.
The raw (uncalibrated) δ 18 O and δ 17 O values are then obtained using the usual definition of δ: where the value of R ref is an instrument-specific estimate of the ratio of integrated absorption of H

Sample inlet system
We use two different inlet systems for the introduction of water into the CRDS optical cavity.To obtain measurements of the same water sample continuously over several hours, we use a "custom vaporizer".The custom vaporizer comprises a continuous-flow inlet system similar to that described by Gkinis et al. (2010Gkinis et al. ( , 2011) ) and used previously for δ 18 O and δD.In this design, water is pumped continuously through a capillary and into a stainless steel tee union heated to 170 • C. In our application, the "pump" is a simple air pressure system, with a double needle that is used to puncture septum-sealed vials; air pressure introduced into the vial through a small steel tube pushes water through a fusedsilica capillary and into the heated tee union.Within the tee union, the liquid water is mixed with dry air and exits the tee union as water vapor that is introduced into the CRDS optical cavity through an open split (Fig. 2).Water-vapor mixing ratios as measured by the CRDS analyzer are maintained at a target value (normally 20 mmol mol −1 ) to within better than ±0.1 mmol mol −1 .For discrete injections of water into the CRDS we use a commercial vaporizer available from Picarro Inc. as model A0211 and described by Gupta et al. (2009).The vaporizer, operating at 110 • C, mixes dry carrier gas with 1.8 µL of water, which is injected through a septum.The resulting water vapor is introduced into the optical cavity via a three-way valve after a ≈ 60 s equilibration.Analysis of a single injection pulse takes approximately 120 s, excluding injection, vaporizer purging and equilibration time.Automated sampling from 2 mL vials is accomplished with an autosampler (LEAP Technologies LC PAL).In our experiments, a complete vial analysis consists of 10 repeated injections from the same vial, for a total analysis time of about 1200 s.

Measurement precision and drift
We use the custom vaporizer to obtain CRDS analyses of the isotope ratios of the same water over several hours.The Allan variance statistic provides a convenient way to assess the analytical precision and drift for the resulting long integrations.The Allan variance is defined as (Werle, 2011) where τ m is the integration time and δ j +1 , δ j are the mean values (e.g., δ = δ 18 O or δ 17 O) over neighboring time intervals.Here, we use the "Allan deviation" (σ Allan , square root of the Allan variance) which can be interpreted as an estimate of the achievable reproducibility as a function of integration time.
Figure 3 shows σ Allan for measurements made both with the L2130-i-C instrument using peak amplitudes, and with the L2140-i instrument using laser-current tuning and the integrated absorption measurement.In both cases, σ Allan values for 17 O of < 20 per meg are achieved after integration times of 5×10 2 s, and σ Allan values for δ 18 O, δ 17 O and δD are below 0.03, 0.03 and 0.04 ‰, respectively.However, these values represent the limits with the L2130-i-C; no additional improvements in precision were achieved with longer integration times, and in general σ Allan begins to rise after 10 3 s.In contrast, with the L2140-i, σ Allan values for δ 18 O and δ 17 O improve to < 0.015 ‰, and σ Allan for 17 O is better than 10 per meg after 1200 s (20 min).For δD, the precision is < 0.07 ‰ at 10 3 s, and remains well below 0.1 ‰ for much longer integrations times (> 10 4 s).Both the measurements with the custom vaporizer, and those with the commercial vaporizer, show that long-term drift in 17 O is greatly reduced in the L2140-i.Long-term drift for δ 18 O and δ 17 O is also improved, though not eliminated.We discuss the relationship between drift in δ 18 O, δ 17 O and 17 O in Sect. 4.
Repeated measurements of discrete water injections provide another way to assess measurement precision and drift.Results from running the same water from multiple vials (with 10 discrete 1.8 µL injections per vial), yield statistics comparable to those obtained with the custom vaporizer (Fig. 4).Typical injection-to-injection precision is 20 per meg for 17 O.Averages over 10 repeated injections from each vial result in a total analysis time per vial of 1200 s, corresponding to the integration time at which the Allan deviation data (Fig. 3) show 17 O precision reaching < 10 per

Sensitivity to water-vapor mixing ratio
Laser spectroscopy instruments used for water isotope measurements exhibit dependence of δ 18 O and δD values on the water-vapor mixing ratio (Gkinis et al., 2010), and similar dependence is expected for δ 17 O and 17 O.This dependence arises primarily from the effect of pressure broadening on peak shape.As noted in Sect.2.3.2,use of the integrated absorption in place of peak amplitude in the calculation of isotope ratios with the L2140-i instrument should theoretically eliminate the water-vapor mixing-ratio dependence.We used the custom vaporizer to obtain measurements with the L2140-i over a wide range of water-vapor mixing ratios.Figure 5 shows that there is a significant reduction in the sensitivity of isotope ratios to mixing ratio when using the integrated absorption measurement, as expected.For δ 18 O, sensitivity is reduced from 0.2 ‰ for a 1 mmol mol −1 variation in water-vapor mixing ratiocomparable to that seen in the L2130-i and other earliergeneration instruments -to less than 0.04 ‰/(mmol mol −1 ).Sensitivity for δ 17 O is comparably reduced, from 0.4 ‰ to less than 0.08 ‰/(mmol mol −1 ).Finally, the sensitivity of 17 O to the water-vapor mixing ratio is reduced from > 250 per meg to < 30 per meg/(mmol mol −1 ).The mixingratio sensitivity of δD, however, at about 1 ‰/(mmol mol −1 ) is not significantly changed between earlier models and the L2140-i.This may suggest an incomplete accounting for the structure of the mixing-ratio-dependent spectral baseline, or other aspects of the spectroscopy that are not yet fully characterized.

Calibration to VSMOW and SLAP
We performed two independent types of calibration experiments with the L2140-i.In the first experiment, we analyzed standard waters SLAP2 and VSMOW2, along with reference waters GISP, VW (Vostok Water), WW (West Antarctic Ice Sheet Water) and KD (Kona Deep), and used the two-point calibration lines defined by Eqs. ( 7) and ( 8) to determine the value of the references waters treated as "unknowns".The resulting calibrated δ 18 O and δ 17 O values are then used to calculate 17 O, using Eq. ( 5) (note that the δ 18 O and δ 17 O of VSMOW2 and SLAP2 are indistinguishable from those of VSMOW and SLAP (Lin et al., 2010)).In the second experiment, we analyzed lab reference waters SW (Seattle Water) and WW and used the IRMS δ 18 O and δ 17 O values of PW (Pennsylvania Water) and VW as calibration points.
In both types of calibration experiments, we used the commercial vaporizer and 2 mL vials, from which ten 1.8 µL injections were made.The measurement order was as follows, where the number gives the number of vials for each water sample in parentheses.First experiment: 5 (KD), 5 (VSMOW2), 4 (VW), 5 (SLAP2), 4 (WW), 5 (GISP), 5 (KD), 5 (VSMOW2), 4 (WW), 5 (GISP), 4 (VW), and 5 (SLAP2).Second experiment: 7 (VW), 7 (WW), 7 (SW), 7 (KD), 7 (PW), 7 (VW), 7 (WW), 7 (SW), 7 (KD), and 7 (PW).In the experiment with VSMOW2, SLAP2 and GISP, the use of lab reference waters with similar isotopic composition prior to the IAEA standards was done in order to reduced the potential for instrument memory effects influencing the results.We use data only from the last three vials for each standard or reference water in our calculations, using all 10 injections from each of those vials in the average.We find that the instrument response time for δ 17 O is indistinguishable from that for δ 18 O.This suggests that memory effects should be minimized for 17 O measurements compared with deuterium excess, which can be problematic because the response time for δD is greater than for δ 18 O in most instruments (Aemisegger et al., 2012).Further work is needed, however, to fully characterize the influence of memory on 17 O with the L2140-i.
The results of the calibration experiments are tabulated in Table 2. Figure 6 shows the calibrated mean values and uncertainties in 17 O for the two different types of calibration experiment.The uncertainties are calculated as the standard deviation of the mean (σ/ √ n) based on n repeated measurements.This calculation may underestimate the true uncertainty because it assumes a Gaussian error distribution, which is not supported by the Allan deviation data for long integration times (Fig. 3).However, this is conservative with respect to the calibration experiments: the results show that the 17 O values of the "unknowns" in each experiment with the CRDS are indistinguishable from the values previously determined using IRMS.Note in particular that the CRDS value of the IAEA reference water, GISP (27 ± 4 per meg), calibrated independently, is nearly identical to the IRMS value of 28 ± 2 per meg (Schoenemann et al., 2013).Further, we find that both KD, which is fresh water derived by reverse osmosis from an ocean water sample, and VW, which is a meteoric water sample from the interior of East Antarctica, have indistinguishable 17 O values.
We emphasize that, as with IRMS measurements, data that are referenced to VSMOW but are not normalized on the VSMOW-SLAP scale can result in inconsistent results because of instrument-specific scale compression (or expansion) relative to the defined calibration (see e.g., Coplen, 1988;Schoenemann et al., 2013).In the context of 17 O measurements on water, such scale compression results in a slope differing from the defined value of 0.528 on a plot of ln(δ 17 O + 1) vs. ln(δ 18 O + 1).Also, if the slope is significantly different from 0.528, errors in 17 O will result even if a linear normalization to VSMOW-SLAP is applied.This problem can in principle be addressed using a nonlinear normalization method (Kaiser, 2008); i.e., and similarly for δ 18 O, rather than our linear calculation (Eqs.7-9).However, the nonlinear calibration method cannot effectively remove scale compression due to blank  effects.In our case, as shown in Fig. 7, the slope of ln(δ 17 O + 1) vs. ln(δ 18 O + 1) is 0.5254; the scale compression is therefore 0.995.Use of Eq. ( 20) would result in a difference for the GISP reference water of < 0.0006 ‰ for δ 17 O, < 0.003 ‰ for δ 18 O and < 1.6 per meg for 17 O, all well below measurement uncertainty.Use of the linear normalization from Schoenemann et al. ( 2013) is therefore preferred.Nevertheless, users of L2140-i instruments will need to verify any calibration strategy for their particular application, taking into account the instrument response time, the availability of reference waters of known composition, and the scale compression, which may be different for different instruments.

Discussion
Our results demonstrate that analysis of 17 O using cavity ring-down laser absorption spectroscopy, as implemented in the L2140-i instrument, can be competitive with analyses by mass spectrometry.The reproducibility of repeated individual measurements made over 30 min is better than 8 per meg, similar to the precision reported for IRMS (e.g., Luz and Barkan, 2010;Schoenemann et al., 2013), and calibrated values of reference waters are indistinguishable between the two methods.Achieving 17 O measurements at the < 10 per meg level with CRDS requires relatively long integration times when compared with the more common δ 18 O or δD measurements, which for typical applications require lower  Measured -Calibrated δ (17,18)   O (‰) δ (17,18) O (VSMOW-SLAP) (‰) precision (< 0.1 and < 1 ‰, respectively).Nevertheless, the new method is less time consuming, less labor intensive, and safer than the IRMS method requiring the use of fluorination.Measurements of 17 O with a laser spectroscopy instrument with a different design (off-axis integrated cavity output spectroscopy, or OA-ICOS) were reported recently by Berman et al. (2013).Measurements of the IAEA reference water GISP reported by Berman et al. (2013), when calibrated to VSMOW and SLAP, are somewhat lower than ours (23 ± 2 per meg, compared with our values of 27 ± 4 (CRDS) and 28 ± 2 (IRMS)), but both are compatible within 2σ of most reported IRMS values from the literature; e.g., the weighted average of the most precise previously reported measurements (IRMS only) was 22 ± 11 per meg (Schoenemann et al., 2013).The mean VSMOW-SLAP-normalized value for GISP for all recent measurements from four different laboratories (as reported here, and by Schoenemann et al., 2013 andBerman et al., 2013) is 17 O = 28 ± 3 per meg.
High-precision 17 O measurements are achieved without drift correction on the L2140-i.Indeed, the precision and drift characteristics of the 17 O results are better than would be expected from the simple combination of noise in the δ 18 O and δ 17 O measurements, both of which show evidence of some drift in their Allan deviations (Fig. 3).
The relationship between δ 18 O, δ 17 O and 17 O errors can be understood as a combination of correlated and uncorrelated noise contributions (Schoenemann et al., 2013): where σ xs is the precision of 17 O, σ 18 is the precision of ln(δ 18 O+1), and η 17 is the residual in ln(δ 17 O + 1) from a best-fit line through the data having slope m.In general, the uncorrelated errors (η 17 ) are small.At higher frequencies, m tends towards higher values (Fig. 8).We find that for the 0.8 s averages of ≈ 400 individual ring-down measurements, the slope is 0.82±0.02,or 1.0±0.1 if a "model 2" regression that accounts for variance in both the δ 18 O and δ 17 O measurements is used (e.g., York, 1969) frequency data is indistinguishable from Gaussian, and is consequently reduced as a function of the square root of the integration time.For longer measurement times (integrations of 10 3 s or longer), m is ≈ 0.5, so that the term (m−0.528)σ18 is small.As for IRMS measurements, it is the combination of the very small magnitude of uncorrelated noise, η 17 , combined with m ≈ 0.5 that leads to the very high precision for 17 O measurements, even where the δ 18 O and δ 17 O measurements are comparatively imprecise.
Frequency dependence of the error slope, m, is not observed in IRMS measurements.As discussed in Schoenemann et al. (2013), in both the H 2 O fluorination procedure and in the mass-spectrometer source, likely sources of error will involve some combination of diffusive and equilibrium fractionation processes, both of which will lead to values of m close to 0.5 (e.g., Miller, 2002).That the relationship between δ 18 O and δ 17 O errors in the CRDS also tends towards m ≈ 0.5 at longer integration times suggests that low-frequency drift in these measurements is similarly attributable to fractionation effects, rather than, for example, drift in the optical cavity temperature or other aspects of the CRDS instrument itself.Fractionation of the δ 18 O and δ 17 O values could be associated with diffusion of water vapor, incomplete evaporation, or condensation and re-evaporation during the vaporization process, or possibly in the optical cavity.
These observations suggest that the current practical limit of precision for isotope measurements on the L2140-i is set by the sample introduction system, rather than the CRDS analysis itself.As illustrated in Fig. 9, which compares IRMS and CRDS measurements, the magnitude of η 17 is very small -similar to that obtained with high-precision IRMS -while the magnitude of σ 18 is much smaller than that obtained with IRMS measurements of O 2 prepared by fluorination.This was not the case with our original prototype instrument (L2130-i-C), for which analyzer noise was dominant even for long integration times (Fig. 9).Because the term (m − 0.528)σ 18 is very small, < 1 per meg, for vialaverage measurements, changes to the sample introduction system that would significantly improve 17 O precision will be challenging.These comparisons attest to the significant improvement in the spectroscopic measurements achieved in the L2140-i, as well as to the stability of the water-vapor delivery and minimal amount of fractionation occurring both in the commercial vaporizer and in our custom vaporizer design.
The L2140-i should be useful in a variety of applications, such as the high-resolution analysis of ice core samples using in-line continuous melting systems (Gkinis et al., 2011), or in the measurement of ambient water-vapor mixing ratios in the atmosphere, currently done with laser spectroscopy instruments for δ 18 O and δD (e.g., Noone et al., 2011;Sayres et al., 2009), though such applications have not yet been fully tested.The low sensitivity to water-vapor mixing ratio achieved with the integrated-absorption measurement would be an advantage in such applications, though there is still some sensitivity that may become important for mixingratio variability greater than ±0.1 mmol mol −1 .In the current commercial version of the L2140-i instrument, a watervapor mixing-ratio correction is available in the instrument software that uses a bilinear relationship of the form A(1) corrected = A(1) + a 0 + a 1 A(1)A( 2), where A(1) and A(2) refer to the integrated absorption for peaks 1 and 2 (Fig. 1), and a 0 and a 1 are empirically determined coefficients.The coefficients are determined by varying the water mixing ratio over a large range and then applied to each measurement.A similar correction is applied to A(3), A(11), and A(13).A simple linear correction following instrument-specific empirical measurements such as illustrated in Fig. 5 could be used as an alternative.We note, however, that we have not evaluated the performance of the instrument at low water-vapor mixing ratios (< 18 mmol mol −1 ).
As an example of an application of the L2140-i, we performed a simple experiment in which 42 vials containing identical water, open to the air, were measured sequentially using 10 injections each.Because the vials were open to a relatively low-humidity laboratory atmosphere, evaporation of the vials would be expected to raise the δ 18 O values through time, and the 17 O value should decrease; furthermore, the relationship between ln(δ 17 O + 1) and ln(δ 18 O + 1) would be expected to evolve along a slope intermediate between the equilibrium value (0.529) (Barkan and Luz, 2005) and the value for diffusion into dry air (0.518) (Barkan and Luz, 2007).These features are indeed observed in the experiment: 17 O decreases by 90 per meg (Fig. 10).
The slope of ln(δ 17 O + 1) vs. ln(δ 18 O + 1) is 0.5232 ± 0.0005, distinguishable at > 99% confidence from the "meteoric water line" slope, accounting for scale compression.A simple experiment like this, which was run fully automated over ≈ 60 h, would take many hours of sample preparation time and > 100 h of analysis time using the traditional fluorination and IRMS method.Note also that the progressive low-ering of the 17 O value is clearly detectable from vial to vial at the 1-2 per meg level; this would probably not be possible to observe using the IRMS method.We suggest that the laser spectroscopy method for 17 O could be used in a number of hydrological and atmospheric sciences applications that were previously impractical.

Conclusions
CRDS is commonly used for measurements of the 18 O / 16 O and D / H isotope ratios of water and water vapor, reported as δ 18 O and δD deviations from VSMOW.We have developed a new CRDS instrument that makes possible the additional measurement of the 17 O/ 16 O isotope ratio, and of the small difference, 17 O, between ln(δ 17 O+1) and 0.528ln(δ 18 O + 1), known as the " 17 O excess".The new instrument uses a novel laser-current-tuned cavity resonance method to achieve precision of < 8 per meg for 17 O while simultaneously providing measurements of δ 18 O and δD with a precision competitive with previous-generation instruments.Liquid samples are introduced into the optical cavity using an automated vaporization system that requires no prior sample preparation.Direct analysis of ambient water vapor in air is also possible.Calibration against the IAEA standard waters VSMOW2 and SLAP2 yields calibrated values for the reference water GISP of 27 ± 4 per meg, indistinguishable from the value of 28±2 obtained by Schoenemann et al. (2013) using IRMS.Our results establish CRDS measurements of 17 O of H 2 O as a viable alternative to conventional IRMS methods that require the use of fluorination to convert H 2 O samples to O 2 prior to analysis.

Figure 3 .
Figure 3.Comparison of Allan deviations for water isotope ratios with the L2130-i-C using a conventional wavelength monitor and spectral peak amplitude (green dashed lines), and with the L2140-i using laser-current-tuned cavity resonance and integrated absorption (solid lines).(A) δ 18 O, (B) δ 17 O, (C) δD, (D) 17 O.

Figure 4 .
Figure 4. Isotope ratios from repeated measurements of 2 mL vials of identical water, using integrated absorption on the L2140-i.(A) δ 18 O, (B) δ 17 O, (C) δD, and (D) 17 O.Each dot represents the average of ten 1.8 µL injections from one vial; the vertical error bars show the standard error (σ/ √ n) of the n = 10 individual injections.The standard deviation of all vial means (σ ) is given in each panel.Horizontal dashed lines are shown for reference at ±0.02 ‰ for δ 18 O and δ 17 O, at ±0.2 ‰ for δD, and at ±10 per meg for 17 O.The experiment shown took about 60 h.No drift corrections or other post-measurement adjustments were made to the raw data.

Figure 5 .
Figure5.Comparison of the sensitivity of isotope ratio measurements on the L2140-i CRDS analyzer to the water-vapor mixing ratio using peak amplitude vs. integrated absorption.Note that a mixing ratio of 20 mmol mol −1 is reported by the instrument software as a concentration (20 000 ppm). (A) δ 18 O, (B) δ 17 O and (C)17 O.

Figure 6 .
Figure 6.Comparison of 17 O data from two independent sets of calibrations of reference waters and standards measured by laser spectroscopy on the L2140-i (CRDS, open squares) with previously determined values from mass spectrometry (IRMS, filled circles). 17O data are plotted vs. δ 18 O.Error bars on the CRDS values are the standard deviation of the mean (seeTable 2).Values and error bars (1 standard error) on the IRMS values are from Table 1, updated from Schoenemann et al. (2013).The calibration points VSMOW, SLAP, PW and VW are shown as open circles for reference.

Figure 9 .
Figure 9.Comparison of the ln(δ 17 O + 1) vs. ln(δ 18 O + 1) relationship for residuals (difference of individual analyses from the mean) of measurements of water samples with the L2130-i-C and the L2140-i CRDS instruments, and with IRMS.The slope of 0.528 that defines 17 O is shown for reference.

Figure 10 .
Figure 10.Results of an evaporation experiment in which 2 mL sample vials are left open to the ambient air and are progressively sampled (ten 1.8 µL injections for each vial) over a ≈ 60 h period.(A) δ 17 O vs. ln(δ 18 O + 1), (B) ln(δ 17 O + 1) vs. ln(δ 18 O + 1).Time progress to the right in both panels.Note the gradual deviation of the measurements (open circles) from a slope of 0.528 (line).

Table 1 .
Schoenemann et al. (2013)topic ratios of reference waters analyzed at the University of Washington " *IsoLab". 17values are from long-term average IRMS measurements, updated fromSchoenemann et al. (2013)to reflect the inclusion of additional data.δ 18 O and δD values are from long-term average laser spectroscopy measurements.δ 17 O values are calculated from 17 O and δ 18 O (Eq. 5).δ 17 O calculated from δ 18 O and 17 O.See Schoenemann et al. (2013).b CIAAW values for GISP are δD = −189.73‰ and δ 18 O = −24.78‰ (Gonfiantini et al., 1995).c Provisional measurement.Long-term average data for KD (Kona Deep) are not yet available. a Schematic of custom vaporizer design used for isotope ratio measurements over long integration times.Double lines denote 1/16 inch and 1/32 inch stainless steel tubing (outside diameter).Single lines denote fused-silica capillary (0.3 mm inside diameter exiting the vials, reduced to 0.1 mm where the capillary enters the vaporizer).

Table 2 )
. Values and error bars (1 standard error) on the IRMS values are from Table 1, updated from Schoenemann et al. (2013).The calibration points VSMOW, SLAP, PW and VW are shown as open circles for reference.

Table 2 .
VSMOW-SLAP-normalized 17 O, δ 18 O, δ 17 O and δD values for reference waters determined by CRDS using (a) IAEA standards VSMOW2 and SLAP2 as calibration points and (b) using University of Washington standards PW and VW as calibration points.IRMSmeasured 17 O values are shown for comparison.Precision (±) is the standard deviation of the mean (σ/ √ n).n is the sample size.VSMOW2 and SLAP2 calibration.b PW and VW calibration.Errors take into account uncertainty in calibration points. a . A slope of precisely 1.0 would be expected if, for example, all measurement error were due to noise in the H 2 16 O spectral line, since this measurement is shared equally in the calculation of both δ 18 O and δ 17 O.The noise in the high-, large circles the individual injection means,"+"signs the vial-mean values.The slopes of the 0.8 s and individual injection data are 0.82 ± 0.02 and 0.59 ± 0.02, respectively (± = 2σ ).The slope of the vial-mean data is 0.54 ± 0.03, shown by the line.