Compact and Lightweight Mid-IR Laser Spectrometer for Balloon-borne Water Vapor Measurements in the UTLS

We describe the development, characterization and first field deployments of a quantum cascade laser direct absorption spectrometer (QCLAS) for water vapor measurements in the upper troposphere and lower stratosphere. The instrument is sufficiently small (30× 23× 11 cm3) and lightweight (3.9 kg) to be carried by meteorological balloons and used for frequent soundings in the upper troposphere and lower stratosphere (UTLS). The spectrometer is a fully independent system, operating autonomously for the duration of a balloon flight. To achieve the required robustness, while satisfying stringent mass 5 limitations, the concepts for optics and electronics have been fundamentally reconsidered compared to laboratory-based spectrometers. A significant enhancement of the mechanical and optical stability is achieved by integrating a segmented circular multipass cell. The H2O mixing ratio is retrieved by calibration-free evaluation of the spectral data, i.e., only relying on SItraceable measurements and absorption line parameters. An open-path design reduces the risk of contamination, allows fast response and thus high vertical resolution. Laboratory-based characterization experiments show an agreement within 2 % to 10 reference measurements and a precision of 0.1 % under conditions comparable to the UTLS. The instrument successfully performed two balloon-borne test flights up to 28 km altitude. In the troposphere, the retrieved spectroscopic data was in excellent agreement with the parallel measurements by a frost point hygrometer (CFH). At higher altitude, the quality of the spectral data remained unchanged, but outgassed water vapor within the instrument enclosure was reducing the accuracy of the retrieved water vapor data. Despite this limitation, these test flights demonstrated the successful deployment of a laser spectrometer in the 15 UTLS aboard a low-volume meteorological balloon, with the perspective of future highly resolved, accurate and cost-efficient soundings.

is minimized over the m points of the spectrum under variation of n and the polynomial coefficients p 1 . . . p N of P N = N j=0 p jν j . Clearly, the reconstruction of P N becomes more accurate if the data set includes sections far from the line center, i.e., where σ(ν) ≈ 0. The minimization of X is performed using the Levenberg-Marquardt least-squares algorithm (Press et al., 2007). The spectral line intensity and the broadening parameters are taken from the HITRAN2016 database (Gordon 90 et al., 2017), whereas the actual gas pressure p and temperature T are measured. It is important to note that Eq. 1 establishes a well-defined relation between the (unknown) number density n and the (measured/reconstructed) absorbance ln(I z /I b ). The relation only contains directly measurable quantities (T, p, z, I z ) and molecular properties (σ). This renders direct absorption spectroscopy (DAS) a potentially calibration-free method to determine the number density of a trace gas. The ideal gas law can be applied to calculate the amount of water molecules relative to the number of total molecules, herein expressed as the 95 mixing ratio r = n/n tot . Figure 1(a) illustrates a typical raw transmission signal I z (τ ) as a function of time τ using a quantum cascade laser (QCL) as a light source. The spectral tuning of the QCL is a consequence of the resistive heating of the laser chip exerted by its driving current. Since neither the tuning speed nor the emission intensity are constant over the course of a spectral sweep, the tuning characteristics dν/dt must be determined for any given driving configuration and operating temperature. This is achieved by 100 recording the Airy transmission signal of a Fabry-Pérot etalon, which yields intensity oscillations with equidistant maxima in wavenumber space. Thereby, the time axis of the raw signal is converted into a wavenumber axis ( Fig. 1(b)).

Absorption line selection
One crucial point in the design of a laser absorption spectrometer is the accurate detection of the absorption signal I z /I b , which usually consists of a tiny absorption feature on top of a large background signal. Therefore, the signal-to-noise-ratio (SNR) 105 must be sufficiently high to allow the precise determination of the absorption, even at the lowest encountered abundance. There are mainly two fundamental options to enhance the signal: Either by extending the optical path length (OPL) or by selecting a strong absorption line. A long OPL is most frequently achieved using multipass optics, which is often the size-determining region. H 2 O exhibits an absorption band associated to the bending vibration mode (ν 2 ) around 1600 cm −1 (6 µm) which is about 30 % stronger than the absorption band due to the (a)symmetric stretching mode (ν 1 and ν 3 ) around 3700 cm −1 (2.7 µm) and about 10 times stronger than the first ν 2 +ν 3 combination band at around 5300 cm −1 (1.9 µm). To exploit this advantage and thereby reduce the required OPL, we use a QCL, i.e. a rapidly tunable and powerful semiconductor mid-IR light source (Faist et al., 1994). The target wavelength region has been determined by simulating the transmission spectra of the ν 2 -vibration band 115 of water vapor under conditions found during the balloon flight. The selected spectral window is shown in Fig. 2 highlighted in gray, which also corresponds to the spectral coverage of the QCL tuned by current. This window is advantageous since it contains a strong and isolated absorption line (2 21 ← 2 12 at 1662.809 cm −1 ) which facilitates the access to baseline I b in its vicinity. According to the HITRAN-database, the absorption parameters for this line, especially the intensity, are known with an accuracy better than 2 %. 120

Optical layout
The absorption line selected above is best combined with an OPL of 6 m to ensure the SNR to stay above 20 in the lower stratosphere. Such an optical path length can be achieved with a variety of multipass cells (MPCs). However, the choice of the MPC is critical for optical and mechanical performance: In fact, there is typically a trade-off between well-controlled and 125 interference-free beam folding, compactness and mechanical stability. We addressed this trade-off by the development of the segmented circular multipass cell (SC-MPC) , which is schematically shown in Fig. 3(a). This monolithic cell consists of a rotationally symmetric arrangement of individual mirror segments carved into its inner surface. This makes the MPC highly resistant to thermally induced distortion, while the spherical shape of these segments preserves a confined laser beam even upon multiple reflection. In addition, the SC-MPC geometry is well-suited for open-path applications because 130 the air can freely stream perpendicular to the optical plane that contains the star-like reflection pattern.
The cell designed for this instrument weighs 160 g and contains 57 segments (6 × 6 mm 2 ) which are circularly arranged with a diameter of 108.82 mm. SC-MPCs are tolerant to various input beam shapes, and furthermore, the special curvature of the last segment directly focuses the laser beam onto the detector. Therefore, the collimated laser and the IR detector can be directly attached to the MPC, without the need of additional beam-shaping optics. This enables a compact setup, increases the 135 mechanical stability, and reduces the optical path outside the sampling volume. These features are of general importance for mobile laser spectrometers, as illustrated by Tuzson et al. (2020) for drone based methane detection.

Integration
The instrument incorporates the highly compact optical layout in an open-path configuration, as shown in Fig. 3(a-b). The optics is attached to a lightweight carbon-aramid honeycomb base plate. The air can freely stream through the central funnel that has a minimal diameter of 86 mm. This yields a large flow rate of 30 l s −1 upon 5 m s −1 ascent rate, which helps reducing the influence of self-contamination caused by water vapor desorbing from the surface. The central funnel is extended by a duct of 10 cm length (PTFE) preventing contaminated air from the proximity of the instrument to stream through the measurement zone. PTFE is chosen because of its low porosity and its low outgassing rate under reduced pressure (Weissler and Carlson, 1980). In addition, it is hydrophobic and generally non-adhesive, thus preventing the deposition of hydrometeors or condensate 145 at the inlet. Flexible bellows, which tightly connect the MPC with the instrument's enclosure, inhibit the propagation of external stress onto the optical system, while suppressing the convective exchange of cold outside air with the internal volume.
For thermal reasons, the highly temperature sensitive laser and detector are not in direct contact with the MPC, which is fully exposed to the outside air temperature variation. The laser is mounted on a custom-made aluminum alignment stage that allows high-precision positioning along five axes. Having all necessary degrees of freedom covered at the laser side, the mounting 150 of the detector can be kept simple, since the beam alignment is accomplished by adjustments on the laser. Thus, the detector is enclosed by a 3D printed holder that is directly fixed onto the board. The detector holder, which further acts as thermal insulation, is covered by an aluminum plate that is connected to heat pipes and serves as a heat exchanger.
This efficient and compact construction permits the optical plane to be located only 1.75 cm above the board surface. At this height, the MPC is held in position by 3D printed braces, which are equipped with a heating wire allowing the temperature-155 control of the cell, e.g., to prevent icing or condensation. The instrument is enclosed in a polycarbonate box. Its lid contains two connections for purging the internal volume with dry gas prior to lift-off. In combination with a 4 cm thick insulating layer of expanded polystyrol (XPS), the instrument has a total weight of 3.86 kg.

Thermal management
Thermal control and stabilization are of utmost importance for high-precision laser spectrometers. The large temperature 160 variability encountered during a balloon flight to the stratosphere renders this especially challenging. To avoid laser frequency drifts, the temperature of the active region of the QCL must be within a few mK, while the outside air temperature may change by 80 K. In addition, the electronics' excess heat (∼ 15 W) must be managed. This cannot be done by passive coupling to the outside atmosphere, because of the large temperature span, which would lead to uncontrollable changes in cooling power. In fact, passive coupling would require a variation of the heat transfer coefficient by about one order of magnitude during ascent, 165 in order to keep the heat source at a constant temperature: On ground and above ca. 20 km, efficient cooling is required, while insulation is needed during the rest of the flight to maintain the internal temperature.
Our strategy of thermal management relies on the fact that the instrument has to be stabilized only for a limited amount of time (∆t ≈ 2 h), and the direction of heat flow is reversed during flight. These are ideal conditions for the use of a heat buffer.
More specifically, the instrument is thermally decoupled from the outside air by the insulating XPS layer mounted around the The pronounced variability of the laser temperature around the tropopause (300-100 hPa) is caused by the fitting procedure of the central frequency of the absorption line rather than effective temperature changes. and specific heat between 11 • C and 26 • C) an amount of 416 g PCM is required to fully take up 15 W during 2 h. Assuming 175 a bidirectional use, the amount of PCM can be halved. Depending on heat production and tolerable temperature range, the individual electronic components are equipped with 10-40 g of PCM, encapsulated in custom-made pads. The laser, as the most sensitive device, is additionally stabilized by a PID-controlled TEC, while its heat-sink is also connected to the buffer medium. This combination of active control and increase of thermal inertia successfully limits the variation of the internal temperature during a balloon flight to T 0 ± 10 K and the laser chip to ∆T = 19 mK as shown in Fig. 4.

Laser driving and data acquisition
Most of the custom-developed electronic circuitry boards (PCBs) are integrated on the bottom side of the carbon base plate, as indicated in Fig. 3(d). The core of the electronics system is given by a commercial single board computer 'Red Pitaya' (STEM-Lab, Slovenia). This partly open source hardware features a field programmable gate array (FPGA) and a microcontroller unit (MCU) running a GNU/Linux operating system. The FPGA on Red Pitaya has been reconfigured to provide the functionalities 185 for high-resolution absorption spectroscopy (Liu et al., 2018;Tuzson et al., 2020).
As a light source, we use a distributed feedback quantum cascade laser (DFB-QCL) packaged in a TEC-equipped HHL housing with embedded collimation optics (Alpes Lasers SA, Switzerland). It is operated at a base temperature of 24 • C. The rapid spectral sweeping of a QCL is achieved by periodic modulation of the laser driving current. An especially economic strategy is referred to as 'intermittent continuous wave' (ICW) (Fischer et al., 2014), whereby the driving current is applied in pulses 190 of a few tens of µs duration, followed by a moment of complete shut-down of the laser to re-establish its initial temperature.
In comparison to the generic continuous wave (CW) driving schemes, icw driving drastically reduces the energy consumption and, thus, the production of excess heat. The current ramps are generated by custom-developed analogue electronics (Liu et al., 2018).
For the detection of the laser signal we use a thermoelectrically cooled MCT-detector Vigo System,195 Poland) coupled to a small-footprint preamplifier (SIP-DC-20M) with a bandwidth of 20 MHz. Figure 5 summarizes the dataflow within the instrument. The preamplified raw signal is digitized by a 14-bit analog-digital-converter (ADC) at 125 MS s −1 , which is integrated on Red Pitaya. Since the FPGA also provides the trigger signals for the laser-driving unit, the acquired data can be grouped into single spectra. A dedicated FPGA functionality sums up in real-time a predefined amount of individual spectra -normally a few thousand spectra per second -in order to improve the SNR. After completion of the spectral summa-

220
In a first experiment, the precision of the instrument at constant conditions is assessed using the Allan-Werle-deviation technique (Werle et al., 1993), as shown in Fig. 6(a), where the evolution of precision as a function of the averaging time is quantified. For this measurement, the sampling volume is held at χ = 2.3 %, p = 258.5 hPa, and T = 296 K by continuous purging. To ensure stable operation of the instrument for longer period than the targeted 2 h, an externally driven liquid cooling system is used (UC180, Solid State Cooling Systems, USA). Values for the mixing ratio χ are acquired at 1 Hz, after co-

In-flight test and intercomparison
As a conclusive assessment of the novel QCLAS instrument, two test flights were performed at Meteorological Observatory Lindenberg, Germany, which also hosts the lead center of the GCOS Reference Upper-Air Network (GRUAN) (Bodeker et al., 2016 we focus here on the evaluation and comparison of the first flight, where the complete ascent and descent data sets are available from both instruments. For the spectroscopic retrieval of the H 2 O mixing ratios, the p and T values are used from an attached RS41 radiosonde (Vaisala, Finnland). Taking the values from the standardized and well-characterized radiosonde is preferred over the use of the system-integrated sensors, since temperature measurements are highly delicate under these conditions and strongly depend on the specific integration properties of the T -sensor (Shimizu and Hasebe, 2010). The descent data stops as soon as the ground station has lost the signal to the RS41, even though the QCLAS continues to operate to 5 km altitude. The instrument was recovered using the GPS coordinates received from the radiosonde. of the QCLAS compared to the CFH. Overall, however, the agreement between the two instruments in the troposphere is excellent, which is illustrated by a mean deviation of only 3 %.

Results and discussion
Above ca. 10 km altitude, the QCLAS gradually retrieves higher H 2 O mixing ratios compared to the CFH. This is attributed to the enhanced outgassing of water vapor from the internal surfaces of the instrument over the course of the flight. This water vapor interacts with the laser beam within the short optical path (9 mm) between the laser/detector and the cell entrance and thus, contributes to the total absorption. Although, this internal OPL is only 0.14 % of the total OPL, the corresponding 265 light absorption can still generate a significant offset in the data with increasing altitude. For example, a superposition of the atmospheric absorption due to 5 ppmv water vapor at 20 km altitude (50 hPa) and the internal absorption by 1 % H 2 O over 0.14 % of the total OPL at similar pressure, results in an apparent mixing ratio of 19 ppmv. This bias could not be avoided even by thoroughly flushing the instrument with dry nitrogen prior to lift-off. An attempt to correct this bias by measuring the internal humidity with a low-cost capacitive sensor is shown as the blue curve for the QCLAS descent data. For better comparison 270 to CFH, a 20 s moving average filter is applied to this curve. While this correction removes the relative trend, the sensor fails to deliver plausible absolute values in the stratosphere, where the relative humidity and the pressure is low (RH < 8 %, p < 200 hPa). This problem is to be addressed in the future by a technical adaptation: Either by completely eliminating the internal optical path or by establishing a pressure difference within the instrument with respect to the surrounding atmosphere.
The latter may be achieved by a leak-tight channel around the internal section of the laser beam, to maintain ground pressure 275 in this compartment during the flight. This would allow the spectroscopic disentanglement of the absorption features due to the different pressure broadening. A strategy that is currently applied successfully to distinguish the contribution of residual humidity within the HHL housing of the laser: Figure 8(a) shows an in-flight spectrum recorded at 13.8 km altitude, i.e., at the hygropause. The narrow absorption feature due the atmospheric and instrument-internal humidity at low pressure (141 hPa) is superimposed over a broad feature that originates from residual moisture within the enclosure of the laser at ground pressure.

280
This OPL is of approximately 3 mm corresponding to the distance between the laser chip facet and the window of the HHL housing. Because of the leak-tight sealing of the laser enclosure, the pressure remains constant during flight, therefore allowing the precise disentanglement of these contributions due to their unequal pressure broadening, especially at high altitude. In the evaluation procedure, we account for this effect by simultaneous fitting of the two absorption line profiles at their corresponding pressure.

285
Apart from the offset due to internal humidity, the ascent data shows distinct spikes with an amplitude of up to 40 ppmv on top of a slowly varying signal above the tropopause, i.e., slightly below 10 km altitude. A close-up of this structure is presented as an inset in Fig. 7(a). A Fourier transform analysis indicates preferred periodicities of τ 1 = 6.5 s and τ 2 = 12.4 s ≈ 2 τ 1 .
Assuming a gravitational pendulum as an idealized model describing the instrument's motion below the balloon, the expected periodicity would be τ p = 2π L/g = 13.5 s for the used rope length L = 45 m. In accordance with Jorge et al. (2020), who 290 analyzed the GPS data of balloon soundings, it is thus plausible that these spikes are caused by repeated transitions of the instrument through the wake of the balloon in an oscillating fashion. This hypothesis is strongly supported by the fact that the spikes immediately disappear after the burst of the balloon, indicating that they are balloon-associated rather than originating from the enclosure of the instrument. Possible reasons for the absence of a similar effect in the CFH data are the lower 10 -3 WV in HHL WV in instrument WV in atmospheric transmission data 2x Voigt fit wavenumber (cm -1 ) transmission residuals Figure 8. Representative transmission spectrum recorded during balloon-borne deployment at 13.8 km altitude. The spectrum exhibits two spectroscopically distinguishable features: a broad absorption (gray) due to the enclosed water vapor within the laser housing HHL (OPL of 3 mm) at ca. 1000 hPa and a narrow absorption (blue) at a pressure of 141 hPa. The latter feature is again a superposition of two contributions: the atmospheric absorption (green) within the MPC (OPL of 6 m and T = 215.8 K) as well as the absorption by trapped water vapor within the instrument's enclosure, which is not distinguishable spectroscopically (yellow). P 3 (ν) P 4 (ν) P 5 (ν) P 6 (ν) 0.2 0.3 0.4 0.5 0.6F igure 9. (a) Fitting of the measured transmission data. The choice of the baseline polynomial (red) and the selected range ∆νi to evaluate a measured spectrum need to be optimized. Varying these parameters can influence the retrieved H2O mixing ratios χ, as shown in (b) after repeated evaluation of the same spectrum. Good agreement among the polynomial degrees over broad range of fitting windows is found.
Within the red highlighted region, the standard deviation of all fitting results amounts to 1.1 % relative to the finally chosen configuration of ∆ν = 0.4 cm -1 and P4(ν).
temporal resolution and the longer rope (60 m) for the CFH. The quantitative contribution of these individual factors is difficult 295 to estimate, but will be investigated during future test flights of the QCLAS. Most importantly, however, any balloon-related contamination can be avoided by measuring during descent (Kräuchi et al., 2016), which is perfectly feasible due to our instrument's high gas exchange rate and its high temporal resolution capabilities, allowing for 1 Hz measurement frequency. It is important to note that the higher mixing ratio measured upon descent in the stratosphere is caused by the internally trapped water vapor concentration which continues to increase due to further desorption during the first 3 km of descent.

300
Nevertheless, future experiments should include a detailed investigation of the herein selected absorption line to determine its missing parameters for a more sophisticated line shape model.

Conclusions
This work describes the development of a compact QCL-based direct absorption spectrometer (3.9 kg) for the measurement of water vapor in the UTLS aboard of meteorological balloons. It relies on an open-path segmented circular MPC and in-340 cludes specifically developed hardware, such as laser driver electronics and FPGA-based data acquisition, as well as dedicated controlling software. A tailored thermal stabilization system based on a combination of phase-change material and thermoelectric cooling allows the autonomous operation of the instrument during balloon-carried ascents and subsequent descents on a parachute. The open-path design prevents self-contamination due to tubing and pumps, and it enables a very fast response time, which is confirmed by the identification of individual wake transitions during balloon-borne ascent. Laboratory experi-345 ments show a precision of 0.1 % and excellent agreement with a CRDS instrument, supporting the calibration-free evaluation approach. Comparison to a CFH, flown simultaneously on a separate balloon, reveals a relative average deviation of 3 % in the troposphere. The accuracy of the stratospheric measurements is currently limited by outgassed water vapor within the enclosure of the instrument, which leads to a bias of the measured concentration at high altitude. Apart from this issue, which should be amendable by constructional adaptations, the system demonstrated highly stable operation even in the stratosphere.

350
In conclusion, we successfully measured UTLS water vapor using a lightweight and standalone mid-IR spectrometer, which is a promising candidate for future high-accuracy assessments of UTLS water vapor on a regular basis.