The main body of the ISE-CUBE system consists of a stack of the two primary modules (analyzer and pump modules) in addition to the cold-trap expansion module; a list giving specifics of individual components can be found in Table  (more details on system composition and construction can be found in Sect. S2 of the Supplement). All three modules use the same plastic container (iM2875 Storm, Pelican Products Inc), allowing for stacking and fastening of the stack, with a footprint of 0.38 m2 (see Fig. ). Each of the three modules is constantly ventilated with a 40 m3h-1 centrifugal fan. A plastic canvas can be placed over the stack for additional weather protection.

Gas transmission within the stack is via an inlet flow assembly (Fig. , dotted red area) composed of approximately 70 cm of 1/4 in. stainless steel tubing (Swagelok Inc.), heated to 60 ∘C with self-regulating heat trace cabling (Thermon Inc.). A main flow of approximately 9 L min-1 (liters per minute) is drawn into the assembly (Fig. b), with most going through the cold-trapping module, using a vacuum pump in the pump module (Fig. b, “Inlet vacuum”). Analyzer flow is split off prior to entering the cold-trapping module. A one-way check valve (6L-CW4S4, Swagelok Inc.) upstream of the analyzer inlet bulkhead (Fig. b, “Check Valve”) prevents reversal of the flow of sample air bound for the analyzer. The external vacuum pump of the CRDS analyzer (N920AP.29.18, KNF DAC GmbH) is also located in the pump module (Fig. b, “CRDS vacuum”). The inlet flow assembly also allows for connection to additional lengths of inlet tubing, such as the profiling module, which enables sampling at adjustable heights (see Sect. ). We will now describe all four modules of the measurement system individually, beginning with the analyzer and pump as the core modules.

We use a Picarro CRDS water isotope analyzer (L2130-i, Picarro Inc., USA) as the central element of the analyzer module (Fig. , middle container). The analyzer module is lined with custom-fit low-density polyethylene (LDPE) foam padding to protect the analyzer. Some padding can be removed to increase ventilation flow for better temperature regulation, depending on ambient conditions during field operations. With a power draw of ∼100 W in steady state, there is substantial heating from the analyzer. Therefore, adequate ventilation is required to keep the analyzer internal temperatures below around 50 ∘C, as prolonged exposure to higher temperatures (above 70 ∘C) can permanently damage electrical components.

The analyzer computer can be controlled from an external laptop through an ethernet (RJ45) cable or via USB-connected monitor and keyboard–mouse combination, with the “data” connectors (Fig. , D). The particular analyzer used here (ser. no.: HIDS2254) is a custom modification of the standard L2130-i, which enables higher flow rates, similar to the analyzer described in . The high flow rate is obtained by replacing the internal, constricting orifice (70 µm) needed for low-flow mode (typical flow rates 0.035 L min-1) with a standard 1/4 in. stainless steel section in addition to using a stronger vacuum pump (N920AP.29.18, KNF DAC GmbH, Germany). High flow rates (about 0.15 L min-1) enable faster analyzer response and 4 Hz sampling, but they also cause more variable pressure inside the measurement cavity. The full impact of this particular configuration will be discussed in more detail below (Sect. ). Sample air is guided from the exterior inlet bulkhead of the module into the analyzer via a 20 cm piece of flexible, 1/4 in. polytetrafluoroethylene (PTFE) tubing. A similar length of 3/8 in., wire-reinforced, PVC tubing (K7160-06, Kuriyama of America Inc.) connects the vacuum port of the analyzer to the exterior bulkhead (Fig. b, “CRDS Vacuum”). The vacuum and electrical bulkheads of the analyzer module then connect to the pump module (Fig. , F: vacuum, B and C: electrical).

The pump module contains the pumps and additional components necessary for operation of the analyzer (Fig. , bottom container). The pump module is connected to the analyzer module with a 3/8 in., wire-reinforced, PVC tube (K7160-06, Kuriyama of America Inc.), providing the vacuum necessary for the analyzer (Fig. b, “CRDS Vacuum”). The inlet vacuum pump (N022AN.18, KNF DAC GmbH, Germany) continuously flushes the inlet tubing, providing the main flow for both the analyzer and the cold-trapping module (see Sect. ). Additionally, a small 300 W uninterruptible power supply (UPS; EL500FR, Eaton) inside the module protects the analyzer from possible power fluctuations and short power breaks. This UPS also provides 12 V power (via a 230 V AC-to-DC adaptor) for the profiling module (Fig. , B). All components in the pump module are strapped to an aluminum support frame, which itself is firmly wedged between the lid and bottom of the case, when the case is shut. Thereby, we could avoid drilling unnecessary mounting holes in the plastic case. Together, the analyzer module and the pump module are the two essential modules for in situ isotopic measurement of water vapor. We will next describe the profiling module, which enables high-resolution vertical profiling of the lowermost levels of the surface layer.

The profiling module allows us to measure near-surface gradients of water vapor (atmospheric composition, more generally), as well as to investigate the exchange processes between surface and atmosphere. The profiling module allows for the acquisition of vertical profiles of the ambient air at any height in a 2 m range (Fig. ). The lean design is expected to cause minimal flow disturbance at the measurement site. The module attaches to the inlet assembly (Fig. , dotted red area) and consists of approximately 4 m of 1/4 in. stainless steel tubing (Swagelok Inc.), heated to 60 ∘C with self-regulating heat trace cable (Thermon Inc.) and surrounded with 2 cm thick foam nitrile insulation. Profiling capabilities are enabled by encasing the final 1.9 m of tubing in an aluminum articulating arm (Fig. ). The base of this arm is then attached to an aluminum mast and tripod with a custom-made steel mount (see Supplement). The tripod serves as the frame for a winch and pulley system to manually control the sampling height (Fig. ). Additional environmental parameters are acquired during profiling from a sensor package mounted on the “head” of the arm, near the air inlet. The height of the inlet is monitored by ultrasonic distance sensors (HC-SR04, SparkFun Electronics; Fig. , red box). Air temperature at inlet sampling height is measured using a temperature probe (VMA324, Velleman; blue rectangle in Fig. b). Both variables are logged onto an SD card via a microcontroller (UNO, Arduino) housed inside a weatherproof container (Fig. a, yellow case at lower left).

An alternate deployment frame was also custom-made for the profiling arm (Fig. S1 in the Supplement). This allowed us to mount the arm on an elevated platform and to reach downwards to the surface of the water (such as at the Fjord deployment site, Sect. ). The arm mounts onto a ladder-like construction with steel U-bolts. Two right-triangular steel supports are mounted on their base legs, perpendicular to the face of the ladder, forming the main structure of the frame. Base and height legs of the triangular supports are approximately 1 m long. A wheel-and-axle crossbeam connects the two supports at the upper point of the height leg; this crossbeam supports the steel cable moving the arm. Any rung of the ladder construction can slide into two steel brackets which are bolted to the platform. While in “normal” orientation, the ladder construction and the head of the inlet arm would be horizontal. However, the entire frame could then tip forwards, pivoting about the secured rung, with the ladder and inlet head both ending up vertical. This allowed us to make measurements 1.0 to 1.5 m further down at the Fjord site, though this did require an additional front-facing ultrasonic sensor. The fine-scale adjustment of the inlet head height was still possible with the winch and cable system and was necessary to account for the tidal height changes of the seawater.

We included a cold-trap module in the ISE-CUBE system, providing the ability to retrieve sample material from the field for subsequent laboratory analysis. This laboratory analysis could also include δ17O for calculation of the 17O-excess. Many commercially available cold-trapping options involve the use of liquid cooling agents, such as ethanol or isopropyl alcohol. demonstrated the feasibility of collecting vapor samples with a Stirling cycle cryocooler as the cryogenic source of a cold trap. Due to the safe transportation and fast installation of such a cold trap, we adopted the basic design of for the ISE-CUBE cold-trap expansion module.

The cryocooler migrates heat away from the tip of a cryogenic “finger” towards the body of the cryocooler, where the heat is radiated away through a radiator fin. By attaching this cryogenic finger to a thermally conducting mass (150 g of brass and aluminum) encircling a glass sampling vial, it directly cools the vial down to -80 ∘C. Incoming water vapor in sample air (Fig. b, “Inlet”) is routed through a combination of 3/8 in. stainless steel tubing and connectors and is then introduced to the cooled glass collection vial connected to the large bore tubing with a custom-made polycarbonate adapter (“PC” in Fig. b). Upon entering the vial, the water vapor is rapidly cooled below its frost point, and it collects on the interior walls of the glass vial. The dried air exits the glass vial via a 1/8 in. length of stainless steel tubing, leading out of the end of the 3/8 in. tubing–connector combination. Finally, this 1/8 in. tubing is connected to the inlet vacuum pump (Figs. b and b, “Exhaust (Vacuum)”), which provides the necessary flow for the cold trap. After the sampling period is complete, the flow is shut off with the needle valve (Fig. , D), and the vial is manually removed, sealed and stored until laboratory analysis, which can be done from the same vial. Also, the relatively small size of the setup easily fits within the standard ISE-CUBE container (Fig. , top box).

We modified the original design of with regard to two aspects, namely the choice of cryocooler and the flow configuration. Firstly, we opted for a more powerful cryocooler (Leon Peters, personal communication, 20 March 2019), enabling faster and more consistent cooling of the sampling vial. Our chosen cryocooler (CryoTel MT, Sunpower Inc.; Fig. , G) takes 5 to 6 min to reach -80 ∘C, at which temperature it has 23 W of cooling power. Secondly, as the system was designed to work in concert with the analyzer and pump modules, we utilized the inlet pump described in Sect.  to provide flow through the collection vial. Accomplishing this required the splitting of flow inside the cold-trapping module into a “sample” and “bypass” line (Fig. b). The sample line allowed incoming, moist air into the glass collection vial, with flow regulation (approximately 0.5 L min-1 or less) via a manual needle valve. The bypass line carried the excess flow (approximately 8.5 L min-1), ensuring that the flushing of the inlet was maintained. Flow rates through the sample and bypass lines were monitored using 1 and 30 L min-1 mass flow meters (TopTrak 822, Sierra Instruments Inc.; Fig. , E and F), respectively. These flow meters recorded onto an SD card via microcontroller (Mega, Arduino; Fig. , K), which in turn allowed for remote monitoring of the flow rates via the external USB bulkhead (Fig. , D). Splitting the lines in such a way allowed the inlet pump to provide flow through both the cold-trap collection vial and the inlet tubing.

The performance of the ISE-CUBE system was evaluated based on a field campaign dataset obtained in challenging Arctic measurement conditions at the scientific settlement of Ny-Ålesund, Svalbard (Fig. S2), during the ISLAS2020 campaign. The principal aim of the ISLAS2020 field experiment was to obtain detailed in situ measurements of isotope fractionation in an Arctic environment, characterized by open fjord water and snow surface. Ny-Ålesund, with the adjacent strong air–sea interaction in the Fram Strait, is a well-suited location to make such observations. In particular, during winter and spring, this region is frequently subject to periods of strong marine cold-air outbreaks, associated with intense evaporation .

In Ny-Ålesund, we deployed at two measurement sites, the first being approximately 300 m south of Ny-Ålesund, on the tundra (78.92117∘ N, 11.91361∘ E). This site was also referred to as the “snow” location and was used from 25–28 February 2020. The second site was located on a concrete pier at the northernmost edge of the settlement (referred to as “Fjord”; 78.92873∘ N, 11.93552∘ E) and was used from 7–14 March 2020. The profiling module utilized the tripod frame while at the Snow site, while it used the tipping frame at the Fjord site, to reach further down towards the surface of the water.

General 2 m air temperature and 10 m wind speed and gust information for the Ny-Ålesund weather station (SN99910) were retrieved from the Norsk Klimaservicesenter (https://seklima.met.no/observations, last access: 3 February 2021). From this station dataset, 20-year climatological conditions were established by considering the hourly dataset of the period between 21 February and 14 March, between 2000 and 2019. The time period from 21 February to 14 March in Ny-Ålesund has a climatological median air temperature for 2000–2019 of -12 ∘C, with 50 % of hourly average temperatures being between -16 and -6 ∘C (Fig. , black). In comparison, the majority of the ISLAS2020 campaign was spent in temperatures spanning -26 to -18 ∘C, with a median value of -20 ∘C (Fig. , blue). The coldest temperatures experienced during the campaign (-30 ∘C) were comparable to the minimum temperature in the 2000–2019 climatology, which was -32.5 ∘C (Fig. , dashed black line). Additionally, have identified February and March 2020 as having some of the strongest marine cold-air outbreaks of the last 42 years. Overall, with several episodes of extreme cold, our deployment was a formidable testing ground for the system.

The ISE-CUBEs produce two main data streams pertaining to the analyzer and profiling modules, with each module internally recording its own respective stream. The cold-trap expansion module does a similar task for its own data stream. An overview of the information contained in these data streams is given in Table . The isotopic data stream generated by the analyzer in the analyzer module is the primary dataset and also has the highest sampling frequency of 4 Hz. The primary environmental parameters observed by the analyzer are humidity (volumetric mixing ratio), δ18O and δD, alongside a multitude of analyzer diagnostic metrics, including temperature and pressure internal components. Before further use, the isotope dataset was calibrated and corrected as described in Sect.  below.

Both the profiling and cold-trap modules measure at 1 Hz via Arduino microprocessors and record parameters to SD cards. The profiling module records the temperature at the inlet head in addition to height distances measured by the ultrasonic sensors. Laboratory calibration of the temperature probe showed a systematic offset of +0.67 K, which is accounted for during post-processing. The cold-trap expansion module monitors and records the flow through the collection vial alongside the remaining flow through the “bypass” line, as well as the temperature inside the module container. The same module also flags whether the cryogenic finger is within 5 K of the target temperature (-80 ∘C). Throughout the campaign, all module timestamps were compared to universal coordinated time, with offsets accounted for and datasets synchronized during processing.

The analysis of module performance that will follow in Sect.  makes use of datasets averaged across multiple time intervals. The technical analysis of analyzer module performance in Sect.  and  uses a 1 s averaged dataset, as does the assessment of the profiling-module sensors in Sect. . However, subsequent evaluation of the profiling module in regards to sample transmission uses the native 4 Hz resolution. Based on our system characterizations evaluated below, our final calibrated isotopic-profiling dataset is averaged over 30 s (Sect.  and ).

Requisite calibrations of the analyzer were performed immediately preceding and following deployment at each measurement site, at Kings Bay Marine Laboratory in Ny-Ålesund. Isotopic measurements are calibrated on the VSMOW2-SLAP2 (Vienna Standard Mean Ocean Water 2 – Standard Light Antarctic Precipitation 2) scale, composed of international, primary standards. The use of the scale allows for the relative ratios of heavy to light isotopes (Rsample) measured in CRDS analyzers to be normalized to a standard (Rstandard), as described in Eq. () : 1δi=RsampleiRstandardi-1⋅1000(‰). The value of the resulting δi (with i representing one of the heavy SWIs) is expressed in per mil (per thousand, ‰). For our calibrations, we employed two secondary standards, whose isotopic values on the VSMOW2-SLAP2 scale have been established in the laboratory. The two secondary standards used were DI (δ18O: -7.68±0.07 ‰ and δD: -49.7±0.4 ‰) and GSM1 (δ18O: -32.90±0.05 ‰ and δD: -261.6±0.3 ‰). Liquid standards were delivered with the Standards Delivery Module (SDM) device (A0101, Picarro Inc.), utilizing a Drierite-filled moisture trap as a source of dry air. Only sufficiently stable calibration signals lasting longer than 10 min are considered for use in calibrating the dataset. As the specific humidity at the deployment site can be well below 3.5 gkg-1 for the time of year, a correction of the isotope composition was applied during post-processing according to the quantified mixing ratio–isotope ratio dependency by , specific to the analyzer used in the field. Several calibrations made during the ISLAS2020 field deployment were between mixing ratios of 3.5 and 0.5 gkg-1 and confirm overall agreement with the mixing ratio–isotope ratio dependency determined in the laboratory.

Calibrations performed in the Kings Bay Marine Laboratory were considered valid only if they passed automatic quality control thresholds, such as humidity variation below 0.3 gkg-1 (500 ppmv). Across 17 valid calibrations, DI measurements had a standard deviation of 0.15 ‰ for δ18O and 0.48 ‰ for δD. For both isotope species, this standard deviation is similar to (or smaller than) the standard deviation typical during any individual calibration. Total measurement drift across the campaign duration for DI was found to be smaller than these standard deviations. The GSM1 standard experienced technical issues with the SDM during multiple calibrations. The standard deviation across the 17 valid calibrations for this standard was 0.18 ‰ (δ18O) and 0.47 ‰ (δD), again of a similar magnitude to the variabilities seen in individual calibrations. As for DI, the total measurement drift across the campaign duration for GSM1 was found to be similar to or smaller than these standard deviations. Overall, these drift values are compatible with the behavior exhibited by the same analyzer during previous use in the lab and field and exceed the manufacturer's typical performance specifications .

We introduce two reference periods for comparison to our analyzer's behavior in the field. The first represents the optimal operating environment for the analyzer, a well-controlled laboratory setting with ambient room temperatures of approximately 20 ∘C. This first period runs from June–July 2020 when the same analyzer was used at FARLAB, University of Bergen, Norway. During this time, the analyzer was routinely sampling standard vapor with mixing ratios down to 0.155 gkg-1, comparable to humidity minimums encountered in the field. The second period more closely resembles a typical deployment, with the instrument sampling exterior air through an inlet tube while installed inside a building with a controlled environment, but is less stringently regulated than a university research laboratory. During this short-term deployment, the same analyzer was installed at the Zeppelin mountain observatory (475 m a.s.l.), 2 km SSW of Ny-Ålesund. The installation spanned from 29 February 2020 to 2 March 2020, with the analyzer measuring outside the ISE-CUBE module. Both reference periods sampled at 1 Hz and at a lower flow rate than we used in the field. Across the three environments, we compare the overall analyzer temperature; the cavity temperature and pressure; and the temperature of the “warm box”, an enclosure housing essential analyzer electronics critical for spectroscopic fitting.

Since 2010, the Alfred Wegener Institute (AWI) has operated the Ny-Ålesund eddy covariance (EC) site , about 300 m south of the settlement and approximately 20 m away from the ISE-CUBE system during the initial deployment phase at Snow. This EC station measures, amongst others, air temperature at 1.0 m, averaged over 30 s. Temperatures are measured by a Thies Clima compact temperature sensor (2.1280.00.160, Adolf Thies GmbH & Co. KG, Germany) inside a ventilated shield (1.1025.55.100, Adolf Thies GmbH & Co. KG, Germany). This temperature was used for field validation of the temperature probe fixed to the head of the profiling module, as described in Appendix .

Using our laboratory and observatory benchmarks, we now detail how the field conditions influenced analyzer performance and thus data quality. We first use the data acquisition system (DAS) temperature (TDAS) measured inside the analyzer housing as a proxy of the overall temperature and condition of the analyzer . Then, we characterize the measurement cavity through its temperature (TC) and pressure (pC). Finally, we study essential analyzer electronics, namely the wavelength monitor (WLM), via the warm-box temperature (TWB).

Based on the assessment of analyzer parameters presented above, measurement conditions within the ISE-CUBE system and in the laboratory differ mostly with respect to the variability of TWB and TDAS. To identify a potential impact of this temperature variability on the vapor isotope measurements, we now compare the variability of the isotopic signal between our reference periods and the field. Since mixing ratio (and thus the amount of molecules in the measurement cavity) is a key factor in the precision of the CRDS measurements, we divide the measurement data into 15 bins from 0.150 to 2.400 gkg-1, 0.150 gkg-1 wide, similar to the bins we used before in Sect. , though twice as wide.

For our field period, we first partition our measured humidities by site; in this way, we ensure that measurement conditions at each site are represented. Then, we identify the distribution of measured humidities at each site according to our bins and normalize it to 100 for the Snow site and 200 for the Fjord site. Then, occurrence in each bin is rounded up to the nearest whole number, such that the smallest bin contains at least two points. Finally, for each site and bin, we identify a corresponding number of 5 min windows with the most stable mixing ratios (i.e., lowest standard deviation of humidity), as categorized by the mean across the 5 min. The maximum standard deviation across all 5 min windows in the field was 0.030 gkg-1, with the median being 0.002 gkg-1. A similar procedure was done for the observatory reference period, though standard deviations were lower than in the field. Within the laboratory benchmark, we focus on 10 distinct usage events conducted for instrument characterization purposes (Fig. ). During these events, the analyzer was subjected to stepwise mixing ratio sequences between 0.20 and 2.25 gkg-1 (Fig. ). The steps usually lasted 5 to 10 min, and the sequences did not necessarily follow a consistent step magnitude. Similar to the field and observatory setup, the most stable 5 min windows were identified using a cut-off threshold of 0.006 gkg-1. These sequences used a standard of a similar depletion (δ18O: -40.02±0.07 ‰ and δD: -307.8±0.8 ‰) as our field measurements. For laboratory, observatory and field, the 1σ 5 min standard deviation over each of these windows was then calculated for both δ18O and δD.

For all three periods and for both δ18O and δD, the measurement precision decreases with decreasing mixing ratio (Fig. a, c). The same is true of the d-excess (Fig. e). In the lowermost bin of 0.150 to 0.300 gkg-1, 5 min standard deviations in the field reach up to 1.0 ‰ for δ18O (Fig. a) and 8 ‰ for δD and the d-excess (Fig. c, e), whereas the same in the laboratory reference are around 0.7 ‰ (δ18O) and 5 ‰ (δD and d-excess). Across all humidities, field bin means (thick colored line) are consistently higher than laboratory and observatory bin means (Fig. a, c, e; black, gray and white lines, respectively). This difference between field and reference periods (Fig. b, d, f) increases at lower humidities, and is consistent between reference periods. A logarithmic fit (Fig. b, d, f; thick colored lines) gives the maximum increase in measurement uncertainty (at the minimum humidity value from the field, 0.197 gkg-1) as 0.15 ‰ for δ18O, 1.35 ‰ for δD and 1.42 ‰ for d-excess. According to this fit, the median humidity value from the field (0.562 gkg-1) would have associated increases of uncertainty in δ18O, δD and d-excess of 0.10 ‰, 0.89 ‰ and 0.93 ‰, respectively. Averaged across humidity bins, this corresponds to a variability increase of 30 % in the field compared to the laboratory benchmark, though the largest relative increase occurs at the higher humidities.

In summary, the field deployment exhibits consistently higher variability for isotopic measurements as compared to the optimal measurement conditions in a well-controlled research laboratory. This higher variability in the field is likely a consequence of the more variable TWB and TDAS but could also be due to the more variable composition of the ambient air used to quantify stability. Additionally, we have not investigated the contribution that our increased flow configuration (Sect. ) might make to this increase in variability. Therefore, we proceed under the premise that our obtained measurements have an increased uncertainty associated with the conditions and the deployment system, as described by the calculated logarithmic fits. Nonetheless, we will show that, though this decreased precision is inherent to the observations, it does not hinder useful measurements, particularly since the measurement precision is quantified.

Evaluation of the profiling-module is divided into sensor performance and sample transmission. Sensor performance details of the ultrasonic distance sensor and the temperature probe at the tip of the inlet are found in Appendix . We now assess the profiling module's ability to transmit samples to the analyzer, including its capacity to resolve isotopic profiles.

The time it takes for the analyzer to react to a step change in humidity at the inlet head is predominantly governed by the length of the inlet tubing and the flow rate through it. With the heated tubing from the inlet tip to the sample port of the analyzer having a diameter of 4.5 mm and a flow rate of about 9 L min-1 in the profiling-module tubing and 0.15 L min-1 in the inlet assembly, the minimum response time would be approximately 6 s. We were unable to conduct any controlled single-step changes during the field deployment, as we lacked a suitable field device to produce a defined vapor isotope stream at sufficiently high flow rates. However, profiling with the articulating arm approximated multiple step changes, albeit without a controlled humidity source. We take the profiling period from the Fjord site on 9 March 2020, which included multiple abrupt humidity changes as we stepped through profiling levels, with some step changes reaching 0.3 gkg-1 (Fig. a). Therefore, we followed the approach of and and applied a fast Fourier transform analysis to the normalized time derivative of the specific humidity observed during this profiling period (Fig. b). We then consider the median across 20 logarithmically spaced bins, between 5.55×10-3 and 2 Hz, of the resulting power spectra as a signal-to-noise ratio (SNR; Fig. b, blue line), with the spectral minimum defining our baseline at approximately 0.4 Hz or 2.5 s. We see that the signal reaches 63 % of its full response after approximately 20 s; however, by 30 s the signal is 80 % of its full response. These statistics are also similar, though less strong, for the low humidities encountered at the Snow site (Fig. S4). Therefore, we conclude that the system is capable of producing a dataset with 30 s resolution, which can be used for analysis purposes.

Next, we consider the minimum duration required at a particular height to resolve the isotopic profile. We again take our profiling period on 9 March 2020, now examining the isotopic measurements (Fig. a, c, e). We apply a fast Fourier transform on the 4 Hz isotopic signal to obtain a power spectrum across a range of frequencies and periodicities (Fig. b, d, f). The spectra median (across 20 logarithmically spaced bins between 8.33×10-4 and 2 Hz; Fig. b, d, f; thick black lines) indicates a baseline extending from 2 Hz until approximately 0.1 Hz (10 s). Frequencies higher (periodicities smaller) than this are considered to be indistinguishable from noise. The 25th and especially the 75th percentiles for these frequencies are also mostly constant (Fig. b, d, f; colored shading). At periods larger than 20 s, the signal begins to emerge from the noise. The d-excess signal takes the longest to emerge, with the median SNR surpassing the 75th percentile of the noise level (Fig. f; dotted line) at 1.5 min. This means that 50 % of the frequencies at this point have a higher SNR than 75 % of the noise. Periodicities smaller than this have less than 2 times as much power as the noise median and cannot be resolved. After 2 min, the median signal reaches 5 dB or approximately 3.16 times the median noise baseline; periodicities larger than this can begin to be resolved. At periodicities of 4 min, medians of all isotopic signals have an SNR larger than or equal to 10 dB. At the less humid Snow site, the median SNR is equal to or above 5 dB after 4 min (Fig. S5). Therefore, across both deployment sites, we determine that maintaining a single height for at least 4 min is necessary to begin to resolve the signal during profiling, though longer durations will yield a higher resolving power.

The cold-trap expansion module was fully integrated into the sample airflow of the ISE-CUBE system during the field tests. Since the cold-trap box was not thermally regulated, the cryocooler unit was operated well below the manufacturers minimum operating guideline of 5 ∘C, reaching -5 ∘C over sustained periods and even down to -20 ∘C occasionally. Nonetheless, the cryocooler reliably provided temperatures of -80±5 ∘C at the base of the vial enclosure for 80 % of the deployment. The remaining 20 % mostly occurred during overnight collections, whereby excessive frost buildup on the exterior of the brass vial enclosure would provide additional insulation, and the vial temperature would drop below -85 ∘C. Vapor collection in a vial lasted between 8 and 16 h during daytime and nighttime, with target flow rates of 0.5 and 0.25 L min-1, respectively. Water vapor was successfully collected during a total of 28 periods, with up to 0.3 mL in a single sample, though some samples collected significantly less.

We now present a comparison between two collected samples and the mass-flow-integrated isotope measurements from the analyzer for the same period (Table ). One sample is from during the Snow deployment, while the other is from the time at the Fjord. While δ18O values are close to being within measurement error, the δD values are quite different. This is likely due to a combination of deficiencies involving sample collection, inconsistent flow regulation and possibly fractionation effects at the low temperatures in the glass sample vial. These fractionation effects might arise from incomplete freeze-out of the vapor, induced by insufficient thermal stability of the glass collection vial. We additionally observed substantial ice crystal formation in the neck of the collection vials, which inhibited and decreased flow during multiple collection periods. This ice formation also compromised sample recovery during vial exchange, causing frozen samples to fall out of the vial during collection. Therefore, the disagreement between sample and analyzer measurements is not unexpected, especially as the collected samples would no longer correspond with the integrated time period being compared.

In summary, we provide here a first proof of concept that a cryocooler-based cold-trap module can be integrated into the ISE-CUBE system. The deficiencies identified could be corrected with a different collection vial and a better flow regulator (i.e., a mass flow controller), although the sample analysis procedure would need to be changed. A more detailed evaluation of the performance of the cold-trap module in terms of sampling efficiency and its suitability for in-field calibration is, however, beyond the scope of this paper and will be detailed in a future publication.

We now present an example for a profiling operation at the Fjord site on the afternoon of 9 March 2020, from 12:57 to 15:30 (UTC). Winds during this period were fairly constant around 7 ms-1, occasionally gusting to 9 ms-1. The entire profiling operation consisted of a sequence of 19 steps across three cycles, performed at heights between 1.5 and 3.5 m above the surface of the water (Fig. a, black line). Throughout the profiling operation, mean air temperature at the inlet and mean specific humidity were anti-correlated with height (Fig. a, cyan and blue lines, respectively). The profiling head was kept at any particular height for 4.5 to 13 min (Fig. a, yellow highlights), with the first 30 s at the level disregarded. For these periods, we calculated the corresponding means for the humidity, temperature, δ18O, δD and d-excess (Fig. b–e, colored markers), alongside the standard errors of the means with a 99 % confidence interval (Fig. b–e, black error bars). We also calculated the observed variability (standard deviation) of the measured quantities (Fig. b–e, gray error bar), including the system uncertainty as calculated in Sect.  (Fig. c–e, gray ticks on error bar). We found the variability to be a similar magnitude for all height levels regardless of sampling duration, with the system uncertainty accounting for approximately half the variability at any particular height.

Across the 2 m profile span, a linear fit through the mean humidities yielded a gradient (with 95 % confidence interval) of -0.13±0.04 (gkg-1)m-1 (Fig. b, blue). The linear fit for air temperature was -0.63±0.17 Km-1 (Fig. b, cyan). This was to be expected, as the surface of the water was a source of both heat and moisture. The isotopic signature of the moisture also exhibited a gradient; δ18O, δD and the d-excess (Fig. c, d, e) all display negative gradients across their linear fits (-0.86±0.49 ‰m-1, -13.4±5.3 ‰m-1 and -6.52±2.17 ‰m-1, respectively). The profiles, especially those for δD and the d-excess, have a strong correlation with height, and the derived linear gradients have a narrow 95 % confidence interval (Fig. b–e; black shading).