The Chicago Water Isotope Spectrometer (ChiWIS-lab): A tunable diode laser spectrometer for chamber-based measurements of water vapor isotopic evolution during cirrus formation

The last decade has seen increasing interest in the use of variations in water vapor isotopic composition to trace atmospheric processes. Remote atmospheric measurements (by solar occultation Fourier transform^1–4 or infrared emission spectroscopy^5,6) have been used to investigate water transport by convection and large-scale atmospheric circulation.^7,8 In situ measurements (by Lyman-α photofragment fluorescence^9,10 or laser-based infrared absorption spectroscopy¹¹) have been used to study detrainment of ice from tropical deep convection and mixing of air masses.^12,13 However, isotopic tracers have not, to date, played a large role in understanding cirrus formation and evolution. Cirrus research using isotopic tracers poses particular difficulties in the real atmosphere: in situ instruments cannot easily follow the evolution of a single cirrus cloud during its lifetime, remote instrumentation typically cannot resolve the thin vertical extent of cirrus (∼1.5 km),¹⁴ and some cirrus formation occurs at altitudes above the range of commonly available commercial or research aircraft. These factors suggest a role for chamber-based measurements of water isotopic composition in areas such as ice nucleation and growth in cirrus (ice) clouds.

Isotopologues of water can be a source of information about cirrus processes because the “vapor pressure isotope effect” produces changes in the isotopic composition of both ice and vapor during ice growth. In equilibrium conditions, deposited ice is enriched in heavy isotopologues compared to the vapor it originates from. Instantaneously,

where R_i is the ratio of isotopologues (e.g., [HDO]/[H₂O]) in the newly deposited ice, R_$v$ is the equivalent ratio in the vapor, and α is the degree of preference for condensation of the heavy isotopologue, known as the fractionation factor. For the temperature range relevant to cirrus (190–235 K), expected values of α are between 1.22 and 1.47 for the HDO/H₂O system and between 1.022 and 1.036 for $H218$O/H₂O.^15,16 As cirrus forms, ice deposition produces progressive isotopic depletion of the vapor surrounding the crystals. If diffusion in ice is negligible on the relevant timescales and diffusion in the vapor phase is not a limiting factor, vapor composition evolves according to the Rayleigh distillation equation,¹⁷ 

where R_$vo$ is the starting and R_$v$ is the final bulk vapor isotopic ratio and (1 − f) is the fraction of original water molecules converted from vapor to solid phase.

Chamber-based measurements can monitor the evolving isotopic composition of vapor during ice formation and provide information on both the fundamental isotopic physics of ice formation and specific cirrus processes. Fundamental physics that can be addressed include measurements of the equilibrium fractionation factor α between water vapor and ice and kinetic modifications to fractionation during diffusion-limited ice formation.¹⁸ Isotopic tracers can be applied to questions such as homogeneous vs heterogeneous nucleation, potential formation of metastable ice at low temperatures, surface inhibitions to growth that are hypothesized to play a role in maintaining supersaturation within ice clouds,^19,20 and glaciation processes as liquid droplets are converted to ice.²¹ 

These motivations drove the design of the Chicago Water Isotope Spectrometer (ChiWIS). The new instrument is a mid-infrared (2.6 µm) laser-based infrared spectrometer designed for simultaneous measurement of H₂O and a heavier isotopologue (either HDO or $H218$O) in water vapor, intended for use at the AIDA (Aerosol Interaction and Dynamics in the Atmosphere) cloud chamber at the Karlsruhe Institute of Technology in Karlsruhe, Germany. Tunable diode laser absorption spectroscopy (TDLAS) offers a potentially precise and rapid method to measure water vapor, and hygrometers referenced to absolute standards have been shown in the literature to be highly accurate.²² The AIDA chamber allows formation of cirrus over the full range of relevant atmospheric conditions, down to the coldest temperatures observed at the tropical tropopause (185 K).²³ To minimize instrument complexity and reduce contamination artifacts, the spectrometer is designed with an open-path configuration in the AIDA chamber rather than extractive sampling. This design precludes the measurement of solid-phase water, so the optimal experimental use of this spectrometer is in conjunction with extractive instruments that can also sample ice particles.

To date, the instrument has participated in the 2011–2013 IsoCloud campaigns focused on ice growth in cold cirrus clouds and the 2013 AquaVIT-II Water Vapor Validation Intercomparison and Testing campaign, an intercomparison exercise between in situ water vapor instruments. During the IsoCloud campaign, the instrument was configured to simultaneously measure H₂O and HDO; resulting measurements are described in the works of Lamb et al.²⁴ and Clouser et al.²⁵ In many AquaVIT-II runs, the instrument was set to measure H₂O only at higher sensitivity.

This paper is organized as follows: Sec. II discusses the science requirements driving instrument and experimental design choices, Sec. III describes the instrument in detail, Sec. IV describes the operation of the instrument, Sec. V covers data analysis and processing, and Sec. VI describes demonstrated instrument performance.

Requirements on ChiWIS performance are stringent because of three factors: the scarcity of the target water species, the nonlinearity of water saturation vapor pressure with temperature, and the scale of expected changes in the isotopic ratio during cirrus formation. These factors place constraints on sensitivity, dynamic range, precision, and accuracy. Because water vapor saturation pressure falls off roughly exponentially as temperature decreases,²⁶ the water vapor content in the chamber can vary by two orders of magnitude over cirrus-relevant temperature ranges from 600 parts per million volume (ppmv) at 233 K to 1.6 ppm at 190 K (referenced to 200 hPa air pressure). The heavier isotopologues are scarcer by four orders of magnitude in standard mean ocean water ([HDO]/[H₂O]-SMOW = 3.107 · 10⁻⁴),²⁷ leaving them at part per billion levels in the cold upper troposphere where high-altitude cirrus forms. While expected changes during cirrus formation are large relative to those of many isotopic systems (of order 10%), the resulting required measurement precision of ∼1% is demanding by the standards of in situ atmospheric instruments. The simultaneous requirements that the spectrometer accurately measure H₂O at its highest concentration (without saturation effects) and resolve the lowest HDO content to within 1% means that the dynamic range of the absorption measurements must exceed 10⁴. These constraints are discussed in detail below. ChiWIS development considerations involve meeting these criteria with the simplest possible instrument design.

ChiWIS spectral region choice is strongly constrained by the need to make measurements in cold conditions where water concentrations are low (H₂O < 10 ppmv, HDO <10 ppbv). These conditions mandate the use of one of the strong fundamental mid-IR absorption bands. Commercial instruments measuring water vapor isotopologues at higher concentrations commonly make use of near-IR overtone and combination bands, whose absorption line strengths are an order of magnitude weaker. We use the 2.6 µm fundamental, which is accessible by recent generations of distributed feedback (DFB) lasers and allows the use of detectors with relatively high efficiency.

The spectral region chosen for ChiWIS water vapor isotopic measurements is shown in Fig. 1. H₂O and HDO measurements are made using the lines at 3789.635 cm⁻¹ and 3788.337 cm⁻¹, respectively, and spectroscopic parameters for these lines are used in subsequent sensitivity calculations. The HDO line is among the strongest available in the band with clear wing space on both sides, which facilitates the line fitting process, and the H₂O line lies close enough to be accessible within the typical diode laser current tuning range as needed for the simultaneous measurement of both absorption features in a single scan (see Kühnreich et al.²⁸ for details of a multiplexed TDLAS instrument designed to measure H₂O and $H218$O in the IsoCloud campaigns). This spectral region does present some complications at higher pressures, since a nearby group of two H₂O lines at ∼3789.26 cm⁻¹ can make fitting the target H₂O line difficult. The spectral region is also not useable in the real atmosphere because of contamination from strong absorptions of CH₄ and N₂O, as well as HF and H₂S, but these species are absent (<0.5 ppmv) in the synthetic air used at the AIDA facility.

Obtaining science-quality isotopic measurements for cirrus studies is made easier by the fact that the HDO/H₂O system has one of the strongest vapor pressure isotope effects seen in nature, with α − 1 ∼ 0.5 at the coldest upper tropospheric temperatures. That is, the[HDO]/[H₂O] ratio in vapor-deposited ice is 50% larger than that of the vapor from which it forms. This strong preferential condensation produces substantial isotopic changes in water vapor during cirrus formation, allowing a relaxed instrumental precision specification of ∼1%. That requirement is still too stringent for a typical open-path absorption spectrometer in the real atmosphere but is accessible in a laboratory context given the option of isotopically doping the chamber water vapor. In this section, we use a numerical simulation of chamber cloud formation to derive a precision requirement for science-quality measurements and from that derive the HDO doping level needed to ensure the necessary signal-to-noise ratio.

In a typical cirrus experiment, air in the AIDA cloud chamber is cooled by expansion, which drops the pressure from ∼300 hPa to 200 hPa over ∼8 min and produces a temperature drop of ∼6 K. (Cooling magnitude is limited by the flux of heat from the chamber walls; only the first few minutes of a pump-down are near-adiabatic.) Figure 2 shows a simulation of chamber conditions during expansion and ice formation: water vapor drops by ∼50% and vapor isotopic composition by ∼10%.

We use this simulation to determine the requisite measurement precision to achieve a specified scientific objective, measurement of the effective fractionation factor α to within 1%. To represent real-world instrumental performance, we add white noise to the simulated isotopic composition and then retrieve the fractionation factor by fitting a smooth curve to this noisy “signal,” letting the α value float. With 1% relative root-mean-square (rms) noise and 300 repetitions of this simulation-and-fitting cycle, we determine that in 90% of the cases, the fitted α value is within 0.3% of the original 1.30, beyond the measurement requirement. However, true precision in derived values would likely be worse since this method ignores chamber inhomogeneities present in the real experiment that can cause low-frequency signal fluctuations and any non-stationary optical fringes. We therefore take an [HDO]/H₂O] ratio measurement precision of 1% as the requirement to obtain 1% relative error in the derived α.

A simple calculation shows that 1% precision in [HDO]/[H₂O] ratio is plausible only if the HDO content of the sample is enhanced (Fig. 3). Since the weak HDO line at 3788.337 cm⁻¹ is the hardest-to-measure feature in the spectrum, the accuracy of the experiment depends most strongly on that measurement. With a sample at 200 K temperature, 200 hPa pressure, and a water content of 8 ppmv at natural HDO abundance, every meter of beam path contributes 1.2 × 10⁻⁶ to the relative peak absorption (bottom solid line in Fig. 3). Doping of the HDO concentration produces stronger absorption; middle and top solid lines show 5× and 20× natural abundance. The criterion for science-quality measurements at 1% precision is that the signal be ∼100 times the per-scan noise-equivalent absorption, which we conservatively take as 10⁻⁴. In Fig. 3, solid lines of the signal must intersect the horizontal dashed line showing 100× this per-scan noise-equivalent absorption. For chamber air at natural abundance, this condition is not possible with the 256-m maximum usable path length of the existing AIDA White cell; the intersection falls at the 8.3 km path length. An isotopic enhancement of 20× likely allows fulfilling scientific requirements, especially given that the line-fitting process usually improves the signal-to-noise ratio.

In addition to the scientific goals discussed in the previous sections, the design of the spectrometer must fulfill other technical requirements as well. Although the primary purpose of the instrument was taking measurements in connection with the AIDA chamber at KIT in Germany, the entire development and part of the testing happened in the Chicago laboratory. In order to be adaptable to both environments and to facilitate transportation, a modular design was desirable. The spectrometer setup is therefore split into three units: the laser head, the sample cell, and the detector module. In the following subsections, the two instrument layouts, the laser head, and detector module are described, followed by details of the instrument’s operation, electronics, and data acquisition.

The two instrument layouts reflect the physical constraints and requirements of the AIDA facility (Fig. 4) and Chicago laboratory (Fig. 5). At the AIDA facility, the laser head and the detector module were mounted on the wall of the cloud chamber in the space between the chamber and the insulating wall where the cooling air was circulated (Fig. 4). The beam crossed the chamber wall through two separate windows placed one above the other: a 25.4 mm (1 in.) anti-reflection coated sapphire wedge at the bottom acting as the entrance window and a 38.1 mm (1.5 in.) uncoated N-BK7 wedge on the top as the detector window. The entering and exiting beams were surrounded by flanged tubes connecting the purge boxes of the laser head and the detector to the chamber wall. In order to eliminate changes in the alignment due to temperature variations, both purge boxes were well-insulated and temperature stabilized at 15 ± 0.5 °C.

The White cell mounted on the AIDA chamber is used to achieve 256.52 m of path length in the ∼4 m diameter chamber. White cells^30–32 employ a concave “field mirror” at one end of the cavity and two concave “far mirrors” at the other end to form a non-reentrant multi-pass optical cavity. The beam enters the cell next to the field mirror and is reflected multiple times between the far mirrors and the field mirror before leaving the cell through an aperture, which is also next to the field mirror. The (slightly divergent) beam always interacts with the center of the far mirrors in a large spot. The far mirrors focus it on the field mirror in an array of consecutive, different small spots. In the White cell, the entry point and the exit point are on the same side of the field mirror, one below the other. The separation between the field mirror and the far mirrors is 3.762 m. The cell employs a pair of fold-back mirrors (not shown in Fig. 4) that send the beam back into the cell after the first set of passes. In this manner, the spots on the field mirror are organized in a four row by eight column pattern yielding a total of 68 passes across the chamber.

In the Chicago laboratory configuration, the spectrometer is a top-of-the-optical-table setup employing a Herriott cell³³ (Fig. 5). The sample cell consists of a 1.1 m long cylindrical vacuum vessel with an inner diameter of 102 mm. Many elements of the cell follow the design of Sayres et al.,¹¹ such as the sample feed system that injects the gas mixture through multiple off-centered points right in front of the mirrors in order to create turbulent flow and prevent the development of stagnant gas regions inside the cell.

The Herriott cavity is made of two spherical mirrors of radii of curvature of 1.68 m placed 1.00 m apart inside the vacuum vessel. They are made of Zerodur glass and coated with a highly reflective dielectric structure designed for the 2.64 µm wavelength. The laser beam traverses the cavity 86 times, making a circular spot pattern on the mirrors with a diameter of roughly 91 mm. The beam enters and exits the cavity through the same slit cut into one of the mirrors. The slit has the dimensions of 16 mm in the radial direction and 6.4 mm across.

The beam entry port into the vacuum vessel is closed off by a 9.5 mm thick slightly wedged fused silica window. The material of the window has been deliberately chosen to be absorptive at 2.6 µm so that it provides the ∼5× attenuation necessary to reduce the beam intensity to levels that can be handled by the detector.

The laser head fulfills three functions: it provides the properly focused beam, monitors the output power, and occasionally makes wavelength measurements. In addition to these goals, the design had to satisfy other requirements as well:

• simple main beam path, i.e., few transmissive optical elements in order to keep reflections and therefore fringing low,

• secondary beam path should provide power and frequency referencing without feeding back into the laser,

• co-aligned visible beam for easy realignment of the downstream optics, especially the multipass White cell,

• compatibility with the existing hardware at AIDA, and

• portability for easy transportation, i.e., high mechanical stability, compactness, and low weight.

The main elements of the laser head (Fig. 6) are the diode laser, collimation optics, power-monitoring detector, beam delay circuit for interferometry, visible laser, and the necessary refractive and reflective optics for beam handling. In order to meet the stiffness requirement, all of these components were mounted on a 12.7 mm (1/2 in.) thick optical breadboard cut to 254 mm × 190.5 mm. Because of space limitations imposed by other elements mounted on the AIDA chamber, the beam exit had to be as close as possible to the upper left corner of the laser head.

The spectrometer and its high sensitivity were made possible by a relatively new technology, distributed feedback (DFB) tunable diode lasers. The infrared radiation source used in ChiWIS is such a diode laser, custom-made by Nanoplus GmbH for the specified spectral region. At its maximum current of 150 mA, it delivers an optical power of 7 mW. It is equipped with a Peltier stack inside the laser can, which regulates the chip temperature to a precision of a few mK.

The entire spectrum that can be reached with the laser is the one shown in Fig. 1. The experimental spectrum was recorded by temperature tuning, i.e., by keeping the current at a constant 100 mA and slowly stepping the temperature of the laser from 29 °C to 44 °C. After removing the baseline and using the proper frequency scale, the resulting plot is identical to Fig. 1 within the resolution of the picture. The spectrum swept in a single current ramp is ∼2 cm⁻¹, or about 1.4 nm, wide (see the inset of Fig. 1). The limitation comes from the maximum current at the top of the ramp, which can be applied to the diode laser (150 mA).

The spectral purity of the laser has been studied in an experiment employing very low pressure Doppler broadened spectra. The laser line proved so narrow that the experiment was unable to measure it, establishing the lower limit of the coherence length at about 50 m.

The main beam path was kept as simple as possible: it starts at the laser mount (upper right corner of Fig. 6), passes through the beam splitter, which decouples the reference beam, and then gets sent into the sample cell by the adjustable “exit mirror.” The beam exits the laser head almost perpendicular to the breadboard through a hole cut in the upper left corner of the board.

The laser mount (labeled “laser and collimation lens” in Fig. 6) was custom designed to house the diode laser and the collimation lens. Made of aluminum, it provides mechanical stability and the necessary heat sinking to the laser can. The collimation lens is mounted inside the laser mount on a miniature dovetail translation stage (Newport) that permits a 9 mm adjustment of the laser-to-lens distance. The dual-aspheric collimation lens was custom designed by Physikalisch-Technische Bundesanstalt and made by Phoenix Optical Glass and features an 8 mm diameter, ZnSe substrate and diamond turned aspherical shape (some circular marks visible on its surface). It was anti-reflection coated (LohnStar Optics) on both sides for the 2.64 µm wavelength. Optimal collimation is achieved when the front of the lens is 3 mm away from the laser chip. This lens is adjusted to provide the proper beam collimation after the laser head is mounted in place since the Herriott cell requires a beam focused to the center of the cell, while the White cell works optimally with the input beam focused in the field mirror’s plane.

Two secondary elements in the main beam path are the coupling mirror of the visible light and a splitter that decouples the reference beam. The coupling mirror is mounted on a motorized flipper mount (Newport 8893-K with two mirror tilts), which keeps it out of the beam during normal spectrometric measurements. When “flipped in,” the mirror co-aligns the beam of a visible diode laser (Thorlabs CPS 186) with the axis of the original infrared beam (infrared beam is blocked). This visible beam aids the alignment of any downstream optics and is especially useful for quickly realigning the White cell. The visible laser is mounted on a 12.7 mm (1/2 in.) mirror mount that provides the remaining two degrees of freedom (tilts) necessary for the beam co-alignment. The next element in the main beam path is the beam splitter that decouples the secondary beam. The splitter is a 3° sapphire wedge that was antireflection coated on only one side—the other side provides the 5% reflectivity necessary for the decoupling.

Past the beam splitter, the mirror labeled “exit mirror” sends the beam almost perpendicularly toward the breadboard through the 19 mm diameter hole cut into it. The two tilt adjustments of the mirror allow the beam to be sent onto one of the far mirrors of the White cell. The mirror needs to be adjusted only once since the White cell is able to accept slight angular drifts of its input beam. In the Chicago setup, the breadboard is laid flat on the optical table, so the exit mirror is replaced by a set of two mirrors at 45°, raising the beam and then sending it into the Herriott cell.

The lower half of the optical layout is a power monitor that can be converted into an interferometer to provide a frequency reference. The decoupled part of the beam passes through a second beam splitter (an uncoated 3° wedge made of N-BK7) and then is focused by an off-axis paraboloid mirror (Thorlabs MPD127127-90-M01) onto the detector (Hamamatsu P10090-21). When the detector operates as a power monitor, this is the only beam it receives—the part of the beam that was reflected by the second beam splitter is blocked by a beam stop.

When frequency monitoring is desired, the beam stop is flipped out of the beam, letting it go around in a rectangle-shaped path hitting the “back” of the second beam splitter, and into the detector. This second beam has a delay of 551 mm compared to the first one hitting the detector, causing interference. When the frequency of the laser radiation is scanned by either temperature or current tuning, the detector signal shows a fringe structure. Because the part of the beam decoupled by the second beam splitter and sent around the delay circuit is small, the fringe contrast is 1%, which is enough for a good fringe retrieval and frequency measurement. This frequency referencing method was found to be more stable than a germanium etalon, especially when the temperature of the entire board is constrained within 2 °C. Other advantages are the higher fringe density (adjustable to any value without significant cost) and practically no feedback into the laser.

When mounted on the AIDA chamber, the breadboard is held by the flanged tube that surrounds the beam going into the chamber. The flange is attached to the back of the breadboard and supports its entire weight. The flange incorporates a 25.4 mm (1 in.) diameter anti-reflection coated sapphire window that seals the chamber. The thermal shield of the laser head (a heated and insulated aluminum enclosure) has its own mount, separated from that of the breadboard. This design choice was made so that the optics were not affected by the weight of the thermal shield and the various tensions that may appear due to temperature variations and circulation of the cooling air. The thermal shield surrounds part of the flange tube so that the flange attached to the breadboard and its incorporated window are inside the thermally stabilized volume. A heated copper collar mounted on the tube blocks the tube’s heat conduction from the thermally stabilized region to the chamber wall. The volume of the thermal enclosure is continuously purged with dry nitrogen.

The main elements of the detector unit are the collection optics and the detector element. As a design requirement, the collection optics need to collect the entire incoming beam onto the active area of the detector; otherwise, any beam position instability will cause parts of the beam to enter or leave the detector surface leading to amplitude fluctuations of the signal. This section starts with a short characterization of two types of beam instability followed by our approach to address these challenges in the design of the detector module.

Beam jitter, that is, rapid sideways shifts of the beam entering the detector module, is caused by vibrations altering the alignment of the White cell and fluctuating inhomogeneities locally changing the index of refraction of the gas mixture. Mirror vibrations result in a jitter amplitude that exponentially increases with the number of reflections. The index of refraction inhomogeneities, on the other hand, randomly affects sections of the beam path, resulting in a dependence on the total path length similar to that of random-walk. An analysis of the positions of the reflection spots on the White cell field mirror (Fig. 7) shows that jitter amplitude increases with the reflection number, but the curvature of the trend line (right hand side of Fig. 7) suggests a milder-than-linear dependence on the reflection number. Adding the observation that jitter is greatly reduced when the AIDA chamber is evacuated (but the pumps are still running, causing vibration) leads to the conclusion that most of the beam jitter is caused by the index of refraction fluctuations.

Another beam instability, due to the White cell’s sensitivity to pressure changes, causes the beam to slowly walk distances on the order of a cm during pump-downs from 300 hPa to 100 hPa.

To address all of the above issues, the detector module was designed with large collection optics equipped with five degrees of freedom for precise alignment, and two video cameras monitoring the incoming beam position and beam focus on the detector.

The main element inside the module is an indium arsenide diode detector (Hamamatsu P10090-21) with a circular active surface 1 mm in diameter. The can of the detector contains a two-stage Peltier cooler that cools the chip to −40 °C, almost doubling the detector’s sensitivity and significantly improving its linearity.

The beam collection optics (Fig. 8) consist of a 90° off-axis paraboloid mirror (Opti-Forms) with a protected gold surface and 38.1 mm (1.5 in.) effective focal length. Due to its 44.45 mm (1.75 in.) diameter, it can capture the beam for the entire range of pressure changes occurring at the AIDA facility and for the largest beam position jitters.

The detector sits on an xyz translation stage secured to the mount of the paraboloid mirror. The stage was made by combining a Thorlabs LM1XY translation mount and a Newport MT-X miniature translation stage. This system adjusts the relative position of the detector with respect to the paraboloid mirror. It is adjusted only once, on an optical bench, so that the detector is in the focal point of the paraboloid mirror. This alignment makes sure that the image of the beam is not subject to coma distortion. In order to avoid high power densities and signal distortion, the detector is positioned slightly closer to the mirror than the focal point so that the beam forms a spot that is about one fourth the diameter of the detector. One of the two cameras is looking at the detector’s surface making this alignment process easier. The detector is tilted 7° by the aluminum wedge it sits on to avoid back reflections into the White cell from it and its window.

After the detector unit has been mounted on the AIDA chamber wall, the paraboloid mirror needs to be adjusted so that its axis is along the incoming beam. To achieve this, the mirror + detector assembly was mounted on a two-tilt lens mount (Newport 8808), allowing the adjustment of the mirror’s orientation without affecting the relative position of the detector with respect to the mirror. Aiming the system is aided by the video camera pointed at the detector. The adjustment is done when the beam is focused to the same spot as it was on the optical bench.

The other camera and a grid screen allow the user to observe the incoming beam position during field operation. Whenever the pressure changes inside the chamber, the White cell mirrors undergo slight tilts, which cause its output beam (which is the same as the input beam of the detector) to shift by up to 2 cm. This necessitates readjusting the White cell mirrors every time a significant change in pressure occurs. To aid this, a grid screen has been mounted on the detector module. It is a 4-cm diameter translucent acrylic disk with a set of grid lines drawn on its surface, mounted on a rotary actuator (Testco H-1079-032). Normally, the disc is out of the beam, but when needed, the actuator lowers it into the beam, which has been switched to visible inside the laser head. A second video camera (not shown in the schematic) inside the detector module is pointed at the grid screen. When at rest, the beam position can be determined with a precision of 0.5 mm. While watching the camera image, the operator can adjust the White cell mirrors until the beam returns to the spot on the grid that has previously been determined as the optimal “starting” position.

The volume of the detector module is continuously purged with dry nitrogen. The walls of the module are thermally stabilized and well insulated. The box is attached to the flanged tube that surrounds the incoming beam and holds the entire weight of the detector module. The opposite side of the box carries all of the electrical feedthroughs, among which are the two video signals and the unamplified detector signal.

When recording the spectra, the frequency tuning of the laser is achieved by increasing the laser current in a ramp waveform. For increased compatibility with APicT,³² an already existing AIDA facility TDL spectrometer, the laser controller electronics replicates the design of that instrument. The current is delivered by a laser controller unit (Thorlabs Pro 8000), but the waveform is generated by a general purpose function generator (Agilent 33220A). Both units are interfaced to the data acquisition computer. The interface program makes the initial settings of the two units, after which they function independently.

The Thorlabs Pro 8000 laser controller unit incorporates an LDC 8002 laser current control module and a TED 8020 temperature stabilization unit. The latter implements a Proportional Integral Differential (PID) control algorithm capable of stabilizing the temperature of the laser chip to ∼0.005 °C. Since the laser is operated above room temperature (usually 30–43 °C), most of the time, the Peltier element acts as a heater, delivering between 1 mW and 45 mW.

The shape of the ramp (Fig. 9) is generated by the Agilent 33220A function generator, which works in “arbitrary waveform” mode. The output of the function generator is fed into the LDC 8002 laser current controller, which modulates the laser current accordingly.

The waveform resembles a sawtooth shape, but with small improvements to increase efficiency. The beginning of the waveform has a 0.2 ms long laser-off time (i.e., no laser current), which has two purposes: to let the laser chip cool down in order to “rewind” its frequency and to provide the analysis software with the opportunity to measure the amplifier offsets and other stray lights captured by the detector (e.g., radiation of other instruments mounted on AIDA scattered by the cloud). Operating the laser at currents just under the lasing threshold still produces some LED-mode radiation that can compromise a true offset measurement. The current is then raised abruptly to 54 mA, which is well above the lasing threshold. (Starting the ramp at the threshold would produce a weak beam and low signal-to-noise ratio.)

The actual ramp is 5.523 ms long and ends at a maximum point of 135 mA. To minimize laser aging, the height of the ramp is lowered to 110 mA whenever the laser is operated at 43 °C in order to access the group of strong absorption lines at the lower end of the laser spectrum. The waveform ends with another laser-off section. The reason for dividing the off-time of the laser into a leading and a trailing part is to facilitate the data acquisition: the acquisition starts at the beginning of the waveform, captures the first off-time and the ramp itself, and then stops right after the maximum point allowing time for data transfer before the next cycle begins.

The lower two curves in Fig. 9 are the frequency reference fringes corresponding to two slightly different waveforms: the blue simple ramp and the red preheated (“spiked”) one. The fringe data were recorded by the detector mounted on the laser head breadboard, and the ramps have been removed during data processing. The simple version shows a high density of fringes right after the laser is turned on. This is caused by the abrupt current onset and the temperature of the laser chip increasing rapidly until dynamic equilibrium with the electric current is reached. That data segment could not be used in the analysis due to poor frequency retrieval. Pumping even more current in the form of a spike at the beginning of the ramp (see “spiked” ramp and lower fringe trace) accelerates the increase in the chip temperature and results in an earlier onset of equilibrium right at the beginning of the actual ramp. The duration of the spike is 0.042 ms and is 6.75 mA high.

As a precautionary measure, the rising and falling edges of the waveform were given finite slopes of 304 A/s and 214 A/s, respectively. The waveform is repeated at a rate of 139.8 Hz, which was chosen to match the sampling rates of other instruments at AIDA.

The detector signals are amplified before they are carried through AIDA’s insulating wall to the acquisition computer. Each signal (one from the “spectrum” detector in the detector module and the other from the power reference detector in the laser head) is amplified by its own DLPCA-200 low-noise transimpedance preamplifier (Femto). At the gain of 10 kOhm, the frequency cutoff is at 500 kHz (−3 dB point). Combining with the 0.3 µs time constant of the parallel RC system comprising the detector and the input capacitance of the preamplifier still allow for practically distortion-free amplification of the absorption line shapes. The preamplifiers have been mounted on their own temperature-controlled enclosure (to minimize offset drifts) close to the detectors and connected to them through 0.45 m long double shielded coaxial cables. The amplified signals are carried to the digital acquisition (DAQ) board by another pair of 2 m long double shielded coaxial cables.

The computer hosting the DAQ board is a Dell Optiplex 790 featuring an Intel Core i5 central processing unit (CPU) running at 3.1 GHz and Windows 7 Enterprise operating system. The computer is dedicated to data acquisition only.

The DAQ board is a National Instruments PCIe-6259 card, acquiring from two channels at a sampling period of 2.39 µs and a depth of 16 bits. The board operates in retriggerable continuous acquisition mode in which the acquisition is always on, but the sampling clock consists of trains of 2500 pulses occurring at each trigger. (The trigger signal comes from the function generator, marking the start of the ramp waveform.)

The acquisition software was written in National Instruments’ LabVIEW 2011 employing an external dynamic link library (DLL) written in C. The software has been split into two programs: one that manages the acquisition board and saves the data and another that displays and/or analyses the data. The DAQ manager is responsible for transferring the data to the host computer whenever 2500 samples have been acquired. The tasks of storing 2800 such “scans” of data in memory (about 27 MB) and then saving them to the hard drive have been assigned to the external DLL. Each set of 2800 scans is written to a separate file. The hard drive is an enterprise series Western Digital equipped with 64 MB of buffer memory capable of receiving the entire dataset in such a short time that no ramps are missed during disk writes.

Splitting the software into a DAQ manager and a display unit allows for better usage of the computer’s resources and for higher flexibility. The display program is set to low execution priority level, allowing the DAQ manager to access resources exactly when it needs them. Also, several different versions of the data display program have been written, each specialized in a specific task: displaying raw data for checking the integrity of the current ramp, real-time fringe analysis for fringe fighting purposes, and simple fitting of the absorption lines for real-time estimates of gas mixing ratios. The operator can switch from one to another as needed.

Data analysis is split in two steps: data preparation and fitting. First, 140 scans (∼1 s worth of data) are averaged together in order to reduce random noise, electrical perturbations coming from outside, and fluctuations due to high frequency beam jitter. Then, the amplifier offset is retrieved from the short laser-off signal preceding the ramp and subtracted from the entire scan. This procedure is applied to both the “spectrum” data (coming from the main detector) and the “monitor” data (from the power monitor detector inside the laser head). Then, the processed “spectrum” data are divided by the “monitor” data. Finally, the useful part of the ramp is isolated and normalized to 1.

The tuning curve (i.e., the function that relates the wavenumber to the acquisition sample number) is obtained from data segments in which the interferometer on the laser head has been turned on. Processing such data involves extracting the fringes (lower plots in Fig. 9), normalizing to the [0, 1] interval, and then fitting them with the following function:

where s is the acquisition sample number and $fs$ is the fringe order number modeled as

Here, the parameters p₁–p₇ are fit parameters. Finally, the fringe order number $fs$ is converted into wavenumbers based on the location of the two known absorption peaks (HITRAN²⁹ line center positions and pressure shifts are used). The tuning curve is calculated at least once for each day of data recording and it is repeated whenever the spectral range has been shifted (by changing the laser temperature) in order to record other absorption lines.

Spectra are fit using the ICOSfit software package,³⁴ which uses the above described tuning curve, a polynomial baseline fit (up to third order), and the HITRAN²⁹ line parameters to fit the experimental data to Voigt profiles. In usual practice, at least two of the four parameters of the Voigt profile (center position and amplitude) are floated. At high concentration, the Doppler and Lorentz widths can be floated as well, but at low concentrations, low signal-to-noise ratios or baseline behavior lead to unacceptable line width fits, so line widths are locked. The main H₂O line and two nearby H₂O lines are fit simultaneously, while the HDO line is fit in a separate run.

Although the H₂O line at 3789.635 cm⁻¹ is the deepest feature in the scanned spectrum, it is sometimes perturbed by significant fringing and noise present on the low end of the ramp. One of its wings is affected by the two neighboring H₂O lines at 3789.278 cm⁻¹ and 3789.244 cm⁻¹, which are hard to fit as they form a close-spaced doublet and are strongly temperature dependent. They make identifying the baseline especially difficult at high pressures (500 hPa and above) when the lines are very broad. The other wing has only a small $H218$O line at 3789.889 cm⁻¹, but it is included in the fit for completeness. When fitting this set of four lines, the baseline is modeled as a second or third order polynomial. When the signal-to-noise ratio is better than about 50, the difference between the Voigt and the real line profiles becomes visible in the form of a “W” shaped residual at the site of the absorption lines. However, compared to other sources of error and uncertainty, line shape misspecification is not a major contributor at the pressures and temperatures studied in the IsoCloud campaigns.

Fitting the HDO line is more difficult because of the small absorption peak height and an unidentified, 0.05% tall bump in the baseline perturbing the low-frequency wing of the line profile (Fig. 10). The bump is likely of instrumental origin and attempts to subtract it in software have been successful only in measurements recorded with highly stabilized laboratory setups. Because the baseline is otherwise relatively flat, only first or second order polynomial is used to model it. The noise in the residual is mostly fringing with a spectrum made of three-four frequencies, often showing a beating pattern in time space. The fringes passing through the site of the HDO line will “pull” the fit both in frequency and amplitude, causing ripples on the final concentration vs time curve.

In the context of the vapor phase water measurements made by ChiWIS, we take “precision” to be the expected fluctuation (or repeatability) of the measured concentration of a stable gas mixture and the “accuracy” to be the observed deviation of the reported concentration from the real value. In the following, precision will be discussed first, and then, an attempt will be made to characterize accuracy. We will make a distinction between two quantities that describe precision, which are often used interchangeably: the estimated detection limit and the actual measurement precision.

The estimated detection limit χ_min is derived from the noise present on the scanned spectrum by converting it, using the inverse of Beer’s law, into an equivalent minimum detectable mixing ratio,

Here, the mixing ratio χ is the ratio of the partial pressure of water vapor to the total pressure. The relative noise rel. σ_noise is the root-mean-square of the noise (including fringes) divided by the total detector signal and frac. absorption is the fractional absorption depth observed at the center of the line. The detection limit is a useful quantity in trace gas applications but claiming that it also represents the precision of the measurement of a large peak is questionable, because baseline effects can become a bigger factor than noise.

The measurement precision σ_χ is obtained after performing the entire fitting process for a measurement of a constant sample, and it is defined as the standard deviation of the fluctuations and drifts on the resulting mixing-ratio vs time curve. This quantity serves the purposes of isotopic studies better since their goal is to measure small relative variations of the mixing ratio.

Both of these measures of precision are a function of the integration time of a measurement, which affects the noise on the detector signal. They also depend on the choice of the absorption line, its line width, and the total pressure. Increasing the pressure broadens the lines and therefore increases the estimated detection limit. Broader lines may also worsen the measurement precision by degrading fit quality. In addition, pressure and temperature may have indirect effects on precision measures by altering optical fringing, which can be sensitive to slight deformations of the chamber geometry. These effects mean that providing a list of detection limits and measurement precisions for the most frequent scenarios is preferable over quoting a single value. In Table I, we show ChiWIS detection limits and measurement uncertainties for 1 s integration for some of the most frequent conditions under which the instrument was operated.

At low mixing ratios (rows 1–2 of the Table I), the limiting factor is noise and fringes on the scanned spectrum. The fitting process acts similarly to an integration of the fringes, reducing their effect, so that the realized measurement precision σ_χ is more than a factor of 3 better than the detection limit χ_min. From the relative precisions σ_χ/χ of H₂O and HDO (0.28% and 0.49%, respectively), it can be inferred that the relative precision of the [HDO]/[H₂O] ratio is better than 6‰. During pump-downs, however, small changes in the White cell alignment can cause additional intermittent etalon fringes, degrading the detection limit by as much as a factor of 2.

At higher mixing ratios (rows 5–8 of the Table I), and stronger absorption lines (row 9), the measurement precision σ_χ is worse than the detection limit χ_min because the limiting factor is the discrepancy between the Voigt profile and the real one. Because the fitting process tries to compensate for the difference using the degrees of freedom of the baseline polynomial, slight fluctuations of the baseline will therefore cause fluctuations in the resulting concentration. In this regime, the measurement precision becomes proportional to the mixing ratio (see σ_χ/χ in Table I).

The dependence of either of these measures of precision on integration time can be assessed by an Allan variance plot,³⁵ which provides a graphical representation of the noise characteristics of the instrument and the characteristic timescale of long-term drift. In Fig. 11, we show the dependence of ChiWIS measurement precision on integration time T using data taken during field operation of the instrument (28 min of 1 Hz data taken during a period of relatively constant mixing ratio during the IsoCloud04 campaign on 23 March 2013). Basing the Allan plot on field measurements allows evaluating the combined effects of instrument performance and various processes in the AIDA chamber, including wall outgassing/uptake and inhomogeneities due to the chamber’s mixing fans. The plot therefore provides an approximate upper bound on the ChiWIS instrument’s intrinsic detection limit. (Note that it is not possible to conduct this exercise for HDO or δD because isotopic exchange with the walls means that there are no periods of similarly stable HDO mixing ratio during the IsoCloud campaigns.)

Figure 11 shows two important characteristics of the combined instrument-chamber system. First, the benefits of signal averaging are less than those expected in data with a purely Gaussian noise distribution (the T⁻¹ line in dashed red) by a factor of ∼0.5. The discrepancy is likely due to the presence of fringing in the raw spectra (see Fig. 10). Second, maximum instrument precision is achieved with about 70 s of averaging, after which secular drift in measured concentration becomes important. This limit could reflect intrinsic drift of the ChiWIS instrument, but, in this context, more likely reflects real variation in the AIDA chamber mixing ratio due to uptake and outgassing on the chamber walls. It is important to note that longer averaging is not actually useful scientifically in most experiments in the AIDA chamber, since cooling is so rapid that the entire period of cloud formation is only a few minutes (Fig. 12).

ChiWIS precision is comparable to that of other TDL spectrometers used for measurements of water vapor or water vapor isotopic composition. Table II shows instrument characteristics for the two H₂O spectrometers used at the AIDA chamber, as well as ChiWIS and an HDO/H₂O instrument used in high-altitude aircraft campaigns. The design characteristics of these instruments vary widely, with the two isotopic instruments restricted to relatively weak H₂O lines and therefore compensating with long path lengths. However, instrument detection limits are similar, since all instruments are designed for scientific measurements in cold and dry air characteristic of the uppermost troposphere. ChiWIS H₂O precision is comparable to that of the mature AIDA facility instrument APicT, and HDO precision is within a factor of 1.6 of the airborne Harvard Integrated Cavity Output Spectrometer (ICOS) instrument. In actual operation, isotopic doping of chamber water means that ChiWIS measurement precision on δD exceeds that of high-altitude airborne instruments.

ChiWIS measurement accuracy is a function of uncertainty in all parameters used in the conversion of integrated absorption to concentration (e.g., HITRAN line parameters), as well as experimental condition (e.g., chamber measurements of temperature and pressure). In measurements shown here, the largest single contribution to systematic uncertainty is the uncertainty in line intensities at ∼5%.²⁹ Line strength uncertainty is lower in the most recent release of HITRAN, and Clouser et al.²⁵ found a combined systematic uncertainty across all HITRAN 2016 spectroscopic parameters and chamber factors of 4.5%. Comparison of H₂O measurements to known water contents or against trusted instruments produces agreement within these ranges. (HDO measurements are more difficult to directly calibrate.)

AIDA experiments allow the comparison of ChiWIS H₂O measurements to estimated saturation vapor pressure over ice during cloud formation. Figure 12 shows an example with measured water vapor stabilizing (with fluctuations) around a value that matches estimation saturation vapor pressure over ice to 1%. For a more extensive analysis of saturation measurements in AIDA experiments, see the study of Clouser et al.²⁵ in which 20 heterogeneous nucleation ice formation experiments show measured water vapor concentrations on average 1.4% below estimated saturation values, well within assumed systematic uncertainty, with a scatter of 0.3% among individual experiments.

We also compare ChiWIS measurements to simultaneous measurements of water vapor from two other facility instruments. The inter-comparisons shown here involve 26 experiments each with a stable air mixture inside the AIDA chamber, with well-characterized temperature and constant water vapor concentration at slightly sub-saturated conditions, without the presence of ice crystals that might produce back-scatter of laser light and thus erroneous sampling. Figure 13 shows a comparison between ChiWIS and SP-APicT, a single-pass (4.1 m path length) in situ TDL spectrometer.³⁶ SP-APicT is based on the design of and comparable in performance to APicT, a well-established AIDA facility instrument.³⁷ Its target vapor mixing ratio range is 10 ppmv and higher (referenced to 200 hPa air), which translates to chamber temperatures above 200 K if the air is saturated. Agreement between instruments in Fig. 13 can be expressed as an instrumental offset of 2.5% and a scatter around it with a standard deviation of 0.8%, within the systematic uncertainties of the two instruments. Some of this scatter is driven by measurements at the lowest temperatures (below 210 K) where SP-APicT begins losing signal.

Figure 14 shows a comparison over the same experiments between ChiWIS and a commercial precision hygrometer manufactured by MBW Calibration Ltd (“Dew Point Mirror 373LX”). The offset is larger, 3.8%, and the scatter is significantly larger, σ = 1.5%, suggesting precision limitations in the dew point hygrometer, especially at lower temperatures. The two comparisons show slight trends with temperature but in opposite directions. Note that both comparisons include one highly deviating point at 233.5 K, which appears low in both plots compared to the spread of the other four points at that temperature. This point is clearly due to an inaccurate measurement by the Chicago spectrometer but is included since the deviation could not be attributed to any specific cause. Without that point, athough, the comparison against the dew point hygrometer shows only a very mild trend line, varying only 0.66% across the entire humidity range of 15–700 ppmv (referenced to 200 hPa).

These three comparisons constrain the accuracy of the spectrometer within a range of 1%–4%, similar to estimated systematic uncertainty but substantially larger than the estimated relative precision of 0.3% for even the worst-case H₂O measurement (see Table I). Measurement accuracy could then be potentially improved with a more reliable water standard that would allow directly calibrating spectroscopic parameters. For the purposes of fractionation factor measurement, note that absolute accuracy is not needed as the quantity of interest is a change in the ratio of two measured species. The test measurements presented in Table I suggest that isotopic ratios can be determined to a precision of 5.6‰ or better.

ChiWIS is designed to aid cirrus cloud research by accurately measuring the isotopic ratio of HDO/H₂O during ice crystal growth. The 2.6 µm tunable diode laser absorption spectrometer is designed to work with either of two multipass cells: a 256-m path length White cell inside the AIDA cloud chamber at the Karlsruhe Institute of Technology and an 86 m path length extractive Herriott cell at the University of Chicago. The design includes a laser head with a simple primary beam path and a secondary beam for power monitoring and frequency referencing. The detector unit employs wide optics and a sophisticated aiming mechanism for efficient and robust beam detection.

Instrumental precision and accuracy are appropriate for science-quality measurements and match the instrument design goals. Tests using stable mixtures with water contents between 12 ppmv and 560 ppmv (referenced to 200 hPa) revealed relative precisions for a 1-s ChiWIS H₂O measurement ranging from 0.3% to 0.08% at the drier and wetter parts of the instrumental operating range. These values correspond to detection limits of 33 ppbv and 440 ppbv, respectively, for a reference pressure of 200 hPa. For HDO measured in isotopically enhanced conditions (12–17 times natural abundance), the relative precision ranges between 0.5% and 0.14% for 1-s measurements, corresponding to 0.23 ppbv and 4.1 ppbv, respectively, at 200 hPa. The relative precision of the [HDO]/[H₂O] ratio—the final science target—is 5.6‰ in the driest conditions and 1.6‰ in the wettest. Instrumental accuracy in water vapor is limited by systematic uncertainties to 4.5%, and comparison to facility water instruments or to expected saturation vapor pressures produces agreement to ∼1%–4%.

The Chicago Water Isotope Spectrometer made measurements during the IsoCloud campaigns at the AIDA cloud chamber of sufficient precision and accuracy for further understanding of cold cirrus and the conditions in which they form. The instrument provided direct measurements of the HDO/H₂O fractionation factor²⁴ between vapor and ice at the cirrus-relevant temperatures found in the coldest parts of the upper troposphere and allowed the derivation of strong constraints on the magnitude of any anomalous supersaturation during cirrus formation.²⁵ The instrument expands the instrumental capability of the AIDA cloud chamber by enabling simultaneous measurements of HDO and H₂O and can continue to provide scientific value during future measurement campaigns.

This work was funded by an International Collaboration in Chemistry grant, jointly between the National Science Foundation (NSF) and the Deutsche Forschungsgemeinschaft (DFG), and by the Camille and Henry Dreyfus Foundation. L.C.S. acknowledges the support through the Camille and Henry Dreyfus Postdoctoral Program in Environmental Chemistry fellowship in 2011 and K.D.L. acknowledges the support by the Department of Defense through the National Defense Science and Engineering Graduate Fellowship Program and support from the NSF through the NSF Graduate Research Fellowship. B.W.C. acknowledges support by the NSF through the Partnerships in International Research and Education (PIRE) program (Grant No. OISE-1743753).