Continuous and simultaneous measurement of triple-oxygen and hydrogen isotopes of liquid and vapor during evaporation experiments

Rationale: Oxygen and hydrogen isotopes are important tools for studying the modern and past hydrological cycle. Previous evaporation experiments used episodic measurement of liquid and/or vapor or did not measure all isotopologues of water. Here, we describe an evaporation experimental system that allows all isotopologues of liquid and water vapor to be measured simultaneously and near-continuously at high precision using cavity ring-down laser spectroscopy (CRDS). Methods: Evaporating liquid is periodically sampled from a closed recirculating loop by a syringe pump that delivers a constant supply of water to the vaporizer, achieving a water vapor concentration of 20,000 ppmV H 2 O (±132, 1 σ ). Vapor is sampled directly from the evaporation chamber. Isotope ratios are measured simultaneously with a Picarro L2140- i CRDS instrument. Results: For liquid measurements, Allan variance analysis indicates an optimum data collection window of 34 min for oxygen isotopes and 27 min for hydrogen isotopes. During these periods, the mean standard error is ±0.0081 ‰ for δ 17 O values, ±0.0081 ‰ for δ 18 O values, and ±0.019 ‰ for δ 2 H values. For the derived parameters 17 O-excess and d-excess, the standard error of the mean is 5.8 per meg and 0.07 ‰ , respectively. For the vapor phase a 12.5 min data window for all isotopologues results in a mean


| INTRODUCTION
The movement of water in the hydrological cycle is largely dominated by evaporation and condensation processes. Oxygen and hydrogen stable isotopes ( 1 H, 2 H, 16 O, 17 O, and 18 O) of water undergo fractionation during evaporation and precipitation as a consequence of both kinetic and equilibrium effects. 1 A variety of evaporation experiments have been conducted in an attempt to understand and quantify these effects. [2][3][4][5][6][7][8][9] In all previous experimental designs, sampling of the liquid undergoing evaporation and its resulting vapor has been episodic and relatively infrequent. The liquid and the vapor have been sampled at discrete times during the course of the experiment, resulting in a relatively small number of measurements. Incident vapor has been collected for measurement by freezing or trapping, 10  Oxygen isotopes are typically measured by isotope exchange with CO 2 and hydrogen isotopes by reduction with a metal (Ni, Cr, or U). 11 where λ is estimated to be 0.528 for unevaporated natural waters. 16 Using multiple isotopes to trace fractionation processes provides distinct advantages for both testing theoretical models and their application to hydrologic problems.
Here, we report an experimental design that permits the measurement of all stable isotopes of liquid water and resulting vapor upon evaporation at a sampling frequency that is more than an order of magnitude greater than most previous experiments. This  The method alternates between analysis of the liquid (from the syringe pumps) and vapor sampled from the evaporation chamber.
The vapor from the evaporation chamber enters the L2140-i, bypasses the vaporizer and flows directly to the analyzer cavity, thereby minimizing memory effects between the liquid and vapor phaseswhich have significantly different isotopic ratios. In the SDM, one syringe pump is utilized to monitor the evaporated water, and the other is connected to a standards delivery bag to monitor instrument drift over the course of an experiment. For correct δ 17 O measurement of the vapor from the evaporation chamber, it is essential that the water-free air used as the carrier gas should be the same for the glove box and the SDM for proper background correction. The SDM air is supplied from a glove bag that is filled by the same dry gas type that is supplied to the evaporation chamber. A schematic of the system is given in Figure 1 and photographs of the system in a lab context are provided in section SI1 (supporting information). If these criteria are met, linear regressions are applied to each isotope ratio, and the gradient, γ, is recorded throughout the length of the experiment. To correct the experimental data and initial and final standards the following calculation is applied:

| Instrumental drift
where δ* dc is the drift-corrected value, δ* m is the measured value, and the time is in minutes (min). It is only after the drift corrections have been made that the data are calibrated to the VSMOW-SLAP scale.
F I G U R E 1 Experimental schematic. Dashed box A show the sampling and measurement part of the system, with water recirculation driven by the peristaltic pump, and sampling and injection to the vaporizer carried out by the SDM. The true layout minimizes the distances between syringe pump 1 and the needle housing and the heated vapor line to the vaporizer. Dashed box B shows the experimental part of the system with water evaporating from a dish set on a balance. Relative humidity is recorded by the probe mounted on a ring stand, which is set 10 cm above the surface of the water, and controlled by adjusting the metering valve. The box is sealed and liquid, vapor, and data cables exit the box through a silicone plug. The design of dashed box B can be adapted as required

| Calibration procedure
The Picarro water isotope analyzer is calibrated using two in-house standards (MPB Enr and JRW; values in Table 1) where δ* cal is the calibrated isotope ratio of interest, and δ* meas dc is the (drift-corrected) measured isotope value of interest. We plot the VSMOW-SLAP calibrated isotope ratio of the two standards (x-axis) on the SLAP-VSMOW scale versus the measured and drift-corrected isotope ratios (y-axis) and calculate a slope (m* corr ) and intercept (c* corr ).

| Experimental sequence
Prior to the start of an experiment, the evaporation chamber is purged and dried overnight to <5% RH (which corresponds to

| Stability and Allan variance
To determine long-term system stability and select the ideal length of time for injection, Allan variance was calculated by repeatedly measuring isotope ratios of a large, sealed volume of isotopically homogeneous water for 42 h. 18

| Liquid measurements
Measurements of the evaporating water are made by sampling a loop containing recirculating liquid from the evaporation pan.
Memory from the previous sample (which can be a liquid or vapor) is a potential problem that can significantly affect the precision of isotopic measurements in CRDS systems. 17 Memory effects are greatest at the beginning of each liquid injection where a small amount of water remains in the system from the previous sample.
At the target pump rate of 0.05 μL s −1 , data for the initial 4 min of each liquid injection are discarded. To ensure that memory effects have been overcome, we then disregard the next 2 min of injection data for the oxygen isotopologues (34 min of usable data). Since memory effects are more severe for hydrogen isotopes, we ignore the next 13 min of data (27 min of usable data) in order to take advantage of the shorter integration period required to achieve the Allan variance minimum for δ 2 H values. More detail about the length of these windows is given in section SI3 (supporting information).
The precision of isotope measurements on the liquid water undergoing evaporation is estimated in Table 2

| Evaporation experiments
The evolution of the isotopic composition of water was studied as it where A and B are constants derived using least-squares regression.
Constants for all three experiments are given in section SI4 (supporting information). For any value of f during an evaporation experiment, the fractionation factor, *α evap , is given by: where *α evap is the fractionation factor for the isotope of interest, *, and *R w and *R v are the isotopic ratios of the evaporating liquid and measured vapor, respectively. 22 With both the liquid and the vapor measured, the fractionation factor can be calculated continuously over the course of an experiment. For the experiment shown, the fractionation factor decreases throughout the course of the experiment by $5% for both oxygen isotopes and increases by $3% for 2 H, when converted to their 1000Ln(*α evap ) approximations ( Figure 5). In fact, the observed change in the fractionation factor during our experiments was insignificant at the 2σ confidence level, when errors are propagated from the isotopic ratio measurements (Tables 2 and 3; Figure 5).
In addition to evaporative trends as a function of the remaining T A B L E 3 The average gradient and standard deviations, σ, and standard error of the mean, SEM = σ= ffiffiffi n p , of vapor measurement periods during a typical experiment with $20,000 ppmV H 2 O. 3σ IR data removal and window length optimization have been applied. n = $500 data points for each measurement window. Number of measurements = 216. *per meg min −1

Measurement
Av. measurement gradient, ‰ min −1 (1σ) Av. measurement σ Av. measurement SEM ) of the experiment as evaporation proceeds. The horizontal lines are the errors calculated for *α evap at 1σ (thin grey) and 2σ (thick grey) significance. These errors are calculated by propagating the error for the measurements of isotope ratios in the liquid and vapor (Tables 2  and 3). Periodic gaps in data are when the drift standard is introduced into the system. Each point in the liquid evolution is the average of continuous data over 34 min (see section 3.2) and each vapor data point represents a 12 min integration window (see section 3. 3) The slopes and confidence limits produced using the new method are shown in Table 4 Figure 6). These relationships are consistent with predictions of increasing slope of the evaporation line as RH increases.  and $5% in the 100Ln(*α evap ) form, respectively (Table 5). These small decreases in the fractionation factor are probably the result of back equilibration between the vapor in the chamber and the evaporating pan, permitted by the long residence times in the box. As such, it is expected that the largest negative offset would be for the experiment in which the residence time in the box is the longest and the RH is highest. However, the fractionation factors for the oxygen isotopes for the 55.5% RH experiment increase by 2.5% and 2.1% as evaporation progresses (Table 5). The increase in oxygen isotope The mol fractions of water between the liquid and vapor phase of each experiment at initiation and termination, assuming that conditions are homogeneous within the box. Also shown is the percentage change *α evap of each experiment (in the form 1000Ln(*α evap )) The sampling and measuring routine described above is adaptable to almost any two-phase liquid and vapor system. There may also be cases where the user might want to continuously monitor the liquid only, and not the vapor. For example, using a different carrier gas (other than N 2 or dry air) that is incompatible with the Picarro CRDS instrument. 5 It may also be desirable to add water vapor with a known isotopic composition to the chamber so that the vapor is not simply produced by the evaporating water, while still controlling other variables.

| Modelling evaporative trends
Critical to the experimental setup is its ability to recreate evaporative trends that can be used to model real-world systems or interpret paleohydrologic data, particularly the derived hydrological tracers 17 O-excess and deuterium excess.
The model trajectory for each of the isotope species is described by a simple Rayleigh fractionation 22 : where *R is the instantaneous isotope ratio of interest, *R o is the initial isotope ratio, f is the fraction remaining, and *α evap is the fractionation factor. When all the moisture in the vapor is derived from the evaporating water body, *α evap can be calculated using 3 : where *α eq is the temperature-dependent kinetic fractionation factor 2 and *α diff is the diffusional fractionation factor, which is dependent on the turbulence immediately above the surface of the evaporating pan 3 : F I G U R E 7 Liquid and vapor evolution for the derived parameters d-excess (A-C) and 17 O-excess (D-F) for each experiment, with the turbulence parameter, X, as calculated by Equation 9. Experiments conducted at A, D, 26.6% and B, E, 34.2% RH show very close agreement between experimental data and model predictions. The 55.5% RH experiment (C, F) shows a slight deviation between predicted values of the d-excess in both the liquid and the vapor and the experimental data, and this is discussed in the main text The turbulence parameter, X, varies from 1 in purely diffusional regimes to 0 when the flow is entirely turbulent. 9 where θ diff = 0.5185. 2 α diff is calculated using 5 : where T is the temperature in C. Using Equations 10 and 11, model

| FUTURE WORK
Understanding the isotopic evolution of water as it undergoes evaporation is fundamental for interpretation of modern and paleohydrologic data. 29,30 For example, the isotope mass balance is often used to estimate the relative inputs and losses of water in modern lakes, and is one of the few methods available for constraining evaporation. Stable isotopes of minerals formed in lakes are often used to reconstruct past changes in hydrology and climate.
Evaporation experiments are a basic but useful tool to understand how the isotopic composition of water evolves as it undergoes evaporation. Previous work has been successful at revealing fundamental isotopic relationships, but experimental designs have not always been ideal for monitoring more complex evaporative trends. [2][3][4][5][6][7][8][9] Our study outlines a sampling and measuring strategy for evaporation experiments that provides highly precise results at a sampling rate that allows for the determination of isotopic trends with very narrow confidence limits about the slopes of these trends. Although

ACKNOWLEDGMENTS
The authors thank Robert Mulvaney of the British Antarctic Survey for discussions about continuous flow analysis and loaning them the equipment to get started. They are also grateful to Fernando Gázquez who carefully read the manuscript and suggested improvements. They also thank two anonymous reviewers for their comments which greatly improved the paper. The work was supported by the ERC WHIM project (#339694) to DAH and the NERC DTP studentship (NE/L002507/1) awarded to MPB.

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1002/rcm.9078.