In order to achieve the campaign objectives, a network of measurement sites with SWI sampling in water vapour and precipitation, aerosol measurements, and meteorological observations was established along the coast of Norway. The core measurement location “Coast” for water isotope, aerosol and meteorology measurements was located at 150 m distance from the shoreline at the base of the north-facing slope of Andhauet mountain on Andøya (Fig. a, Sect. ). To study vertical isotope gradients, as well as effects of cloud microphysics and below-cloud exchange on precipitation, water vapour isotope measurements and precipitation sampling were conducted at the mountain site ALOMAR (Fig. a, Sect. ). To cover the vertical gradient between these two measurement sites, precipitation collection boxes and an additional automatic weather stations (AWS) were placed at four locations at different elevations (Fig. b, blue squares, Sect. ). In-situ measurements at site Coast were complemented with radiosondes and remote sensing instrumentation located at the nearby town of Andenes, covering a larger part of the atmospheric column (Fig. a, Sect. ).

Additionally, in order to assess the horizontal representativeness and spatial gradients in precipitation isotopes, precipitation sampling was performed along a 100 km-long surface transect from Andenes, reaching across the Vesterålen archipelago towards the South (Fig. b, Sect. ). To further assess the horizontal variability in SWI signals, including the identification of Lagrangian matches during air mass transport, two additional water vapour isotope measurement sites were established in the town of Tromsø, Norway (120 km northeast of Andøya, Sect. ) and in Bergen, Norway (1100 km southwest of Andøya, Fig. a, Sect. ).

In addition to discrete sampling at regular intervals at the stations described above, higher-frequency sampling was conducted during intense observing periods (IOPs) depending on the prevailing meteorological conditions (Sect. ). In the following sub-sections, we describe each of the sampling sites in more detail. The key measurement equipment used during the campaign at all locations is listed in Table , and the up-times and availability of the different datasets during the campaign are described in Sect. . The dataset for the individual types of measurements and sites is archived as a dataset bundle at 10.1594/PANGAEA.984616 .

Site ALOMAR (Arctic Lidar Observatory for Middle Atmosphere Research) is an observatory located on the top of Ramnan Mountain at an elevation of 379 m a.s.l. and ∼3 km southeast of the town of Andenes, Norway (Fig. b). During the ISLAS2021 campaign, a water isotope CRDS analyser (L2130-i, Ser. No. HIDS2254, Picarro Inc., Sunnyvale, USA) was installed in the hatch control room on the roof top of the ALOMAR main building (Fig. ). The analyser sampled ambient air from a 6 m long inlet line heated to 60 °C, that was flushed at a flow rate of about 5 L min⁻¹ by a manifold pump (N622, KNF GmbH, Germany), resulting in an average time delay before ambient signals arrived at the CRDS of about 20 s. The inlet was shielded from precipitation by a heated metal bowl, mounted at about 2.5 m above the platform level (385 m a.s.l. and 12 m a.g.l). During high wind speeds, snow could occasionally be lofted from surrounding structures and enter the inlet line, but would evaporate completely before reaching the analyser. An SDM (Picarro Inc., Sunnyvale, USA) and vaporiser (Part No. A0211, Picarro Inc., USA) were installed for calibration purposes. Dry air for the calibration vapour generation was produced from a molecular sieve (MT-400, VWR Inc., USA). Next to the inlet, a TinyTag logger (Ser. No. 917160, TGS-4505) with a small screen was installed to measure air temperature and relative humidity at a 2 min time interval.

Snow and rain samples were collected on the platform level at ALOMAR using a snow sampling box and a precipitation collector (Fig. b). Sampling frequency was increased during several IOPs (see Sect. ). A rain collector (Palmex Inc., Croatia) was mounted to the railing close to the room housing the CRDS analyser. Data from a permanently installed MRR2 (Metek GmbH, Germany) were retrieved for the ISLAS2021 campaign period. The MRR was configured to report data at a 10 s time interval, with height bins from 35 to 1085 m until 14:30 UTC on 23 March 2021, and from 100 to 3100 m thereafter.

ALOMAR has been used for routine aerosol and cloud observations of the middle atmosphere since 1996 and for intensive measurement campaigns e.g.,. During limited precipitation-free conditions, and in coordination with air traffic control from the nearby airport, the rooftop hatch was opened by an operator for lidar measurements Fig. c, . The lidar utilised here is a system designed for measuring attenuated backscatter at three wavelengths (1064, 532, and 355 nm) and volume depolarisation ratio at one wavelength (532 nm) in the troposphere. lidar measurements were strongly constrained by the weather conditions, allowing for four valid measurement periods. The total duration of lidar measurements during the ISLAS2021 campaign was ca. 16.5 h, and contained high, middle and low clouds, periods of clear sky and volcanic aerosol, presumably from an ongoing Icelandic eruption (Table ). Since operating the lidar required to open a large hatch of the ALOMAR building and the presence of an operator, measurements were limited to selected precipitation-free periods.

Andøya meteorological station is located on the north-eastern part of Andøya island, 4.4 km from Andøya Space (69.3152° N, 16.1309° E, 3 m a.s.l., WMO-number: 1010). In addition to the near-surface measurements from an AWS, a ceilometer (CHM15k Nimbus, Lufft GmbH, Germany) obtained backscatter profiles and cloud layer heights continuously during the campaign. The ceilometer operates at a wavelength of 1064 nm and provides data with 15 s interval within an altitude range of 5 to 15000 m. Furthermore, an automatic sonde launcher at Andøya meteorological station released a total of 84 radiosondes between 23:03 UTC on 28 February and 17:03 UTC on 31 March 2021. Thereby, the regular twice-daily sounding interval (11:00 and 23:00 UTC) was increased to three to four times a day from 19 to 31 March 2021 (at approximately 05:00, 11:00, 17:00 and 23:00 UTC) during the ISLAS2021 campaign period.

Surface snow and bulk precipitation were collected along a vertical transect between the sites Coast and ALOMAR (Fig. , blue boxes and yellow markers). Three sampling boxes (V1, V2, V3; Table ) were placed between 100 and 300 m a.s.l. near the mountain road leading up to ALOMAR. One TinyTag (Ser. No. 920032, TGS-4505) was installed approximately half-way up along the slope of Ramnan mountain, near the site of box V2. The vertical profile was complemented by collection of surface snow at site V0 (25 m a.s.l.), and the regular precipitation collections at ALOMAR  V4) and Coast.

Precipitation and surface snow were also collected along a horizontal transect from Andenes across Vesterålen towards the Norwegian main land. In total 5 sampling boxes (H1, H2, H2A, H3, H4) were installed for bulk precipitation sampling. Snow surface samples were collected at sites H1 to H4 as well as at four additional locations (H0, H1A, H1B, H3A, Fig. b). The locations cover a distance of approximately 100 km from the north coast of Andøya to the south coast of Hinnøya, with the aim to identify potential isotopic signals from isotopic distillation across the coastal mountains, and to quantify the spatial representativeness of precipitation isotopes measured at Andenes. Boxes were placed in an open area or on the upper part of a sloping area to minimise the collection of blowing snow.

At each location, box samples (consisting of solid or liquid precipitation, or a mixture) were collected using sampling bags and a plastic spoon as described in Sect. . After sample collection, the boxes were emptied and dried with a paper towel. When solid precipitation had accumulated since the previous visit, additional surface snow samples were collected with a spoon and sampling bag from a location within a few metres of the box. At locations H0, H1A, H1B and H3A, only surface snow samples were collected. Boxes H1, H2, H3, and H4 were installed on 18 March and H2A on 23 March 2021. Boxes were if possible cleared ahead of a new IOP to obtain a clean signal without drifting surface snow. The sampling interval and potential for resulting post-depositional effects are described in Sect. . On several occasions, a small meteorological probe (iMet XQ-2, InterMet systems Inc., USA) was mounted outside a car window to obtain horizontal transects of air temperature and relative humidity between Andenes and the horizontal transect sites.

One set of water vapour isotope measurement equipment was originally planned to be installed on a research vessel for underway sea water and water vapour measurements. However, sanitary restrictions due to COVID-19 required on short notice to repurpose the instrumentation to a land-based water vapour measurement station. As an alternative measure, a water vapour isotope measurement station was set up at the town of Tromsø, located ∼120 km to the north-east of Andenes (Fig. b). Situated on an island in the fjord Straumsfjorden, the town is shielded from the open ocean to the west and north by mountains with elevations exceeding 1000 m. An ambient air inlet, protected with a heated precipitation shield was installed at 56 m a.s.l. on the roof of Natural Science building of the University of Tromsø (UiT, 69.6819° N, 18.9777° E), near a web camera and AWS owned by UiT (Fig. b, blue square). The inlet line (ca. 6 m PTFE) was heated to 60 °C with self-regulating heating tape (Thermon Inc., USA) and flushed continuously with an inlet pump (N622, KNF GmbH, Germany). A portable weather station (Kestrel 5000L, Nielsen-Kellerman Co., USA) was installed near the inlet on the roof (Fig. c). The indoor installation was set up in a rooftop instrument room, and consisted of a water vapour isotope analyser (L2140-i, Ser. No. HKDS2039, Picarro Inc., USA) and a Continuous Water Sampler (CWS, Part No. A0217, Picarro Inc., USA) used here for instrument calibration. After setup, the analyser by mistake partly sampled room air through an open split connecting the CWS and the Picarro in the first half of the campaign (until 20 March 2021). On 21 March 2021, the leak was fixed by disconnecting the CWS. The CRDS analyser thereafter continuously sampled air from the flushed inlet line.

Another sampling station was set up in the city of Bergen, located in the south-western part of Norway (Fig. ). While Bergen is generally more influenced by mid-latitude weather systems, the site was located either upstream or downstream of the sampling sites in Northern Norway on several occasions. Continuous water vapour isotope measurements during the campaign were performed at the roof of Geophysical Institute, University of Bergen (60.3837° N, 5.3319° E, 56 m a.s.l.) using the setup described in . In short, a CRDS analyser (L2140-i, Ser. No. HKDS2038, Picarro Inc., Sunnyvalye, USA) continuously sampled from a heated inlet (60 °C) shielded from precipitation at the instrument tower of the building, and flushed with a flow rate of 5 L min⁻¹ by a manifold pump (N622, KNF GmbH, Germany). Measurements of air temperature, relative humidity, pressure and total precipitation close to the air inlet were performed using a hotplate pluviometer (TPS-3100, Yankee Inc., USA) and an AWS (Anderaa, Norway).

In addition, measurements of air temperature, relative humidity, and precipitation were provided by the AWS Bergen-Florida (WMO-number 1317), located at 16 m a.s.l. in the garden of the Geophysical Institute. Previous studies showed that the precipitation measured by the rain gauge at the AWS is ∼10 % lower than measured by the pluviometer at the tower . Precipitation sampling for SWI analysis was conducted during the ISLAS2021 campaign at a location 1.3 km north-east of the Geophysical institute (60.3872° N, 5.3537° E, 143 m a.s.l.) with a manual rain collector for event-based sampling.

We now first describe the general weather conditions encountered during the campaign. The measurement period of the ISLAS2021 campaign was characterised by large synoptic variability. A general classification into weather events associated with warm-air advection from mid-latitudes, and cold-air advection from the Arctic was employed to distinguish between different IOPs (Table ). A positive CAO index, defined as the difference between the potential temperature at sea level and at 850 hPa is used to delineate regions dominated by arctic air masses, and associated with large heat fluxes . We use the percent area coverage with positive mCAO conditions in two domains, a box just offshore of Andenes (69–70° N, 14–17° E), and a larger box including the Vesterålen archipelago and Tromsø (67–70° N, 12–20° E), to quantify regional mCAO conditions near the sampling sites (Fig. c, black and cyan lines). Sea level pressure (SLP), wind speed, and precipitation rate further illustrate the synoptic variability (Fig. a and b).

The first IOP, termed IOP0 since it took already place before all instrumentation was completely operational, lasted from 16 to 17 March 2021. At that time, a mCAO extended from the Barents Sea towards Andenes (not shown). During IOP0, the coldest air temperatures in Andenes during the measurement campaign were observed (below -11 °C; Fig. a) and the CAO index reached 5.1 K (Fig. c, red line). A high-pressure system over Svalbard and the Norwegian Sea directed the flow of arctic air towards Andenes (Fig. a). The high-pressure system subsequently moved eastward during IOP1 (18 to 19 March 2021), and a large mid-latitude cyclone moved into the Norwegian sea, with its core marked by integrated water vapour above 8 kg m⁻² east of Svalbard (Fig. a, shading). At that time, the CAO index had decreased, and precipitation from the warm sector of this system reached Andenes (Fig. b and c). During IOP2, a rapid passage of narrow fronts associated with a short-wave system originating over Greenland occurred within 24 h, as seen from the minimum in SLP on 20 March 2021 of about 988 hPa (Fig. a, black line).

The most pronounced mCAO both in spatial coverage and CAO index magnitude (maximum value was 5.3 K) was encountered during IOP3 from the 21 to 22 March 2021 (Fig. c). Intense showers, wind gusts, and temperature variations were observed as individual convective cells passed over the observing site Coast during that period (Fig. c). IOP4, starting on 22 March 2021, was associated with the passage of a large frontal system that progressed poleward into the Barents Sea (Fig. d). This IOP4 was characterised by warmer air temperatures of up to 5 °C, intense precipitation of up to 3 mm h⁻¹, and a lower CAO index (Fig. a and b). As the mid-latitude cyclone had moved poleward, an intense cyclone developed on the trailing system. During IOP5 on 23–24 March 2021, the site Coast was hit directly by the rapidly intensifying cyclone (“atmospheric bomb”), reflected in a minimum SLP of 970 hPa (Fig. e). This event was associated with the largest accumulated amount of precipitation during ISLAS2021, and winds of up to 20 ms⁻¹ at site Coast (Fig. b). As the cyclone moved away towards the east, it gave way to colder air reaching Andenes, initiating IOP6 that was associated with a short period of mCAO conditions with convective cells and snow showers (Fig. f). During IOP7 on 29 March 2021, a mesoscale cyclone moving northward along the coast of Norway brought warm air masses and light rain to Andenes from its narrow frontal band (Fig. c).

With the first installations starting on 15 March 2021 at Tromsø and site Coast, the continuous measurements became operational across all sites on 16 March (Fig. a). The CRDS analysers did not record ambient vapour measurements during daily calibration periods. The CRDS analyser in Tromsø partly sampled room air in the period 15 to 23 March 2021, leading to a strongly muted signal of ambient-air isotope variations (light red shading). As no personnel was present in the room during the measurement period, and a ventilation provided continuous exchange of ambient air into the room, we decided to retain the time period as part of the dataset, but denoted with a quality flag. During 28 to 30 March 2021, the analyser did not record measurements. The CRDS at site Coast was disconnected from its inlet for performance tests during 26–27 March 2021. Data from the MRR at ALOMAR became available at 100 m vertical resolution during 19 March 2021, changing from 35 m vertical resolution before. The MRR at Coast, the Parsivel² disdrometer and the ceilometer delivered data throughout the campaign except for a few short interruptions. Disassembly started on 30 March 2021, and was completed during the following day.

Across the measurement network, at least two CRDS were operating during 99.2 % (158.8 h) of the campaign duration of 160.1 h, at least three analysers were measuring during 85.7 % (137.2 h), and all four analysers were operating during 23.5 % (37.6 h). When including the period of the inlet leakage in Tromsø as measurement time in the calculation, the percentage when all four CRDS were measuring increases to 58.3 % (93.2 h). The most complete coverage of isotope measurements was obtained from 23 to 27 March 2021. The opening of the lidar hatch at ALOMAR, that was located approximately 5 m away from the inlet, did not produce a measurable imprint on the water vapour isotope signal. From regular and additional radiosonde launches, a total of 57 balloon ascents are available during the campaign period. The radiosondes measured wind speed and direction using GPS, as well as air pressure from a silicon capacitor, air temperature from a resistive sensor, and relative humidity from a humicap sensor. All data is reported at a 2 s time interval.

Discrete sampling of precipitation, and other discrete measurements, were organised into the sequence of IOPs corresponding to pronounced changes in the prevailing meteorological conditions (Sect.  and colour bars on top of Fig. a). The total of 137 precipitation samples taken at site Coast were collected mainly during IOPs 3, 4, 5, and 7 (Table b). Precipitation at site ALOMAR was only collected at high resolution during IOP4 and IOP5. The total of 56 precipitation samples from 8 vertical transects from the boxes and locations V0 to V4 were mostly from IOP3. A total of 54 precipitation samples were collected on the 8 horizontal transects (T1–T8, Fig. b). The typical duration until sampling was 1 d or less for Transects T2–T5, and two days or more for T6–T8, increasing the potential for post-deposition effects to modify the precipitation signature. The most detailed horizontal and vertical sampling was carried out during IOP3 (Transect T5). The severe weather during IOP5 only allowed for collection of the total precipitation from the horizontal transect at the end of the event (T7). Bergen precipitation (30 samples) was collected mostly on a daily basis, but also included higher frequency sampling when the mCAO of IOP3 arrived in Bergen on 22 March 2021. Sea water was collected on a daily basis, except for two samples collected on 22 March 2021, providing a total of 13 samples. INPs were analysed for 52 precipitation samples, up to 5 times per day.

Water vapour isotope measurements from all analysers were processed with an overall similar routine with small specific adjustments. First, raw isotope measurements were corrected for the mixing ratio–isotope ratio dependency of each specific analyser using a correction curve determined in the laboratory . The CRDS analyser at site Coast suffered from unusually strong mixing ratio–isotope ratio dependency, which was corrected using the dependency presented in Appendix . Next, data was normalised to VSMOW-SLAP scale using long-term calibration coefficients specific to each analyser, as described in Appendix  and the following paragraphs. The calibrated water vapour isotope data, along with ambient pressure, water vapour mixing ratio, and several instrument parameters were then averaged (5 min time resolution for site Coast, 2 min for site ALOMAR and Tromsø, 10 min for site Bergen), and combined with the corresponding meteorological measurements from meteorological sensors mounted near the inlets. Specific humidity calculated from the meteorological data was thereby used to time-reference the CRDS measurements.

The CRDS analysers installed at ALOMAR, Tromsø, and in Bergen were each calibrated using instrument-specific long-term calibration coefficients (Table ). These coefficients were obtained from repeated measurements of known secondary standards normalised to VSMOW-SLAP scale over at least several months from a combination of SDM and liquid injection measurements in a controlled laboratory environment. On-site calibrations during the campaign were used to check the validity of the calibration line in terms of slope and offset of the calibration curve. The rationale behind this approach to calibration rather than, e.g., interpolating from one calibration to the next within a measurement interval of 23 h, is that uncertainty introduced by the calibration system in a field setup is similar to the analyser uncertainty, for example due to less reliable dry air provision to calibration units. In addition, the same type of analyser as used here has been observed to have negligible drift over months up to years . The field calibrations are then used as part of calculating the combined uncertainty (see Sect. ). For the CRDS at site Coast, no reliable long-term calibration had been established at the time of the campaign, and the water vapour isotope measurements were therefore calibrated using the average of all available calibrations during the campaign period.

More specifically, for the CRDS analyser installed at ALOMAR (Ser. No. HIDS2254), calibration checks were performed daily with two secondary standards (Table , DI and GSM1). Due to the standard bag being almost empty, standard GSM1 was replaced by standard GLW on 17:15 UTC on 21 March 2021. Drift was less than or smaller than the uncertainty of the calibrations. The CRDS analyser at site Coast (Ser. No. HIDS2380) was calibrated at a 23 h interval using standards EVAP2 and GLW. The CRDS analyser in Tromsø (Ser. No. HKDS2039) was calibrated using a CWS. The CWS was used, since the analyser originally was intended to be deployed on a research vessel. Calibrations were performed daily except for the period of 24 to 30 March 2021 when the CWS was disconnected. Thereby, the CWS supplied three secondary standards in the sequence DIX, GLX and MYRK for 20 min each, with the last 10 min being retained. While the mixing ratio of the retained periods was sufficiently stable (standard deviation between 56 and 346 ppmv), the average humidity of each calibration step varied, requiring correction using linear isotope-humidity dependency derived for this specific analyser. Drift was smaller than calibration uncertainty. The CRDS analyser in Bergen (Ser. No. HKDS2038) was calibrated using four to six manual injections of secondary standards DI2 and GLW at five days during the campaign period. The average of the last three satisfactory injections confirmed consistency with the long-term calibration parameters.

To compute the uncertainty budget for vapour isotope measurements, we adopt here an approach that is also commonly used for liquid water sample isotope analysis . The combined uncertainty is thereby obtained from the squared sum of calibration standards uncertainty, the uncertainty field calibration, and the uncertainty of the sample measurements, each weighted by the sensitivity to the calibration slope. Thereby, the uncertainty of each sample measurement is estimated from the standard deviation during an averaging interval (Eq. ). Since the CRDS at site Coast had an about four times larger standard deviation for both isotopes than the other CRDS, a 24-point moving average (corresponding to a ∼20 s time window) was applied to the δ18O and δD measurements, before recalculating the d-excess. While some finer-scale structures where thus lost from the time series, this filtering brought the estimate of the measurement uncertainty to a range comparable to the other analysers.

Components of the uncertainty budget for the four CRDS analysers are summarised in Table . The uncertainty of the calibration standards can be retrieved from Table . The combined (total) uncertainty ut presented here is calculated for an the average isotope composition and mixing ratio over the entire field period. Uncertainty will increase with lower mixing ratio (larger analytical uncertainty) and towards the upper and lower end of the calibration curve (sensitivities). The uncertainty budget is generally dominated by the measurement uncertainty (35 %–90 %), while analytical uncertainty is largest for sites Coast and Tromsø (∼20 %–40 %, not shown). It should be noted that the combined uncertainty of the d-excess is ∼2 ‰ãcross all sites. Together with the use of common reference standards and processing steps, this enables meaningful comparisons across the CRDS network. In addition, a bias correction was required for the CRDS at site Coast, as detailed in Sect. .

The freshwater (rain, snow and surface snow) and seawater samples were analysed at The Facility for Advanced Isotopic Research and Monitoring of Weather, Climate and Biogeochemical Cycling (FARLAB) at the University of Bergen. Prior to analysis, the samples were filtered and transferred to 2 mL GC-vials (ThermoSci 2-SVW Chromacol). Analysis procedures followed the scheme utilised at FARLAB described in . In short, each batch of about 20 samples was supplemented with secondary laboratory standards DI2, GLW, FIN and EVAP2 in use at FARLAB, which had been calibrated to the VSMOW-SLAP scale against primary standards available from the International Atomic Energy Agency (Table ). Similarly to the procedure described in , 12 injections and 6 injections were done for each standard and sample, respectively. For the sea water samples, a salt liner was installed in the vaporiser. During the run, the water was transferred from the vials to the vaporiser using an autosampler, and high-purity-grade N₂ (nitrogen 5.0, purity > 99.999 %; Praxair Norge AS, H₂O mixing ratio < 5 ppmv) was used as matrix gas.

After the analysis, each run was calibrated and corrected for memory effects and isotope ratio–mixing ratio dependency corrections for each individual analyser using the software FLIIMP FARLAB liquid water isotope measurement processor. Samples were corrected for drift and memory, and normalised for VSMOW-SLAP scale using laboratory standards. The uncertainty of the calibrated samples is calculated based on the assigned uncertainty of the isotopically heavy and light standard with respect to VSMOW-SLAP, the uncertainty of measured values of the standards, and the uncertainty of the sample approximated by the standard deviation of repeated measurements or by long-term reproducibility. Long-term reproducibility at FARLAB for measurement of drift standard DI2 has been estimated to 0.052 ‰ for δ18O and 0.446 ‰ for δD, respectively .

After collection of aerosol samples at the site Coast, the ice-nucleating ability of aerosols was quantified in situ using the home-built drop freezing setup DRINCO , based on the design of and . DRINCO uses a webcam to monitor the freezing of 50 µL aliquots of sample pipetted into a 96-well PCR tray that is partially submerged in a temperature controlled ethanol bath (FP51, Julabo). The webcam captures the freezing progression of the aliquots at 0.25 °C intervals while the ethanol bath is cooled at a rate of 1 °C min⁻¹. An aliquot is identified as frozen based on the amount of light that is transmitted through a well, with a sharp decrease in light transmission after freezing due to the enhanced light scattering in ice relative to water. The result of the experiment is a frozen fraction (FF) for each 0.25 °C interval between -30 and -2 °C. The frozen fraction was then converted to an INP concentration per temperature (INPair(T)) using Poisson counting statistics described by and calculated as: 2INPair(T)=-ln⁡(1-FF(T))⋅VsampleVdropletVair, where Vdroplet is the size of the aliquot in each well (50 µL), Vsample is the volume of water in the Coriolis sampling cone at the end of the sampling period and Vair is the volume of air sampled by the Coriolis during the sample.

All of the aerosol and INP concentrations were normalised to (std L)⁻¹ by using the inlet temperature as measured and recorded by a Type K thermocouple and datalogger (EL-GFX-TC, Lascar Electronics datalogger), respectively, and ambient pressure measurements from the Norwegian Meteorological Institute site located in the town of Andenes (see Sect. ).

When accounting for the 12 m³ of air sampled with the Coriolis impinger (Sect. ) and a residual cone volume of approximately 15 mL, the minimum detection limit of INPs was ∼2.5×10-4 std L⁻¹ at temperatures warmer than -17 °C. Due to the low-end cutoff size of the Coriolis (∼0.5 µm), it is expected that the INP concentration reflects both biological and mineralogical INPs that are typically larger than this size.

Within the ISLAS2021 campaign setup, a network of stable isotope analysers was deployed on distances of a few km (sites Coast and ALOMAR), to 120 km (site Tromsø), to 1100 km (site Bergen). If the analysers are calibrated consistently, covariances, offsets, and time shifts between the different sites can be interpreted in terms of processes and meteorological influences, and thus inform about spatial representativeness of this kind of measurements. The specific humidity from sites Coast (Fig. a, black line) and ALOMAR (blue line) shows a very large degree of similarity throughout the campaign. There are a few occasions during IOP1, IOP4 and IOP6 where site Coast appears to encounter drier conditions. The specific humidity at Tromsø (red line) still appears similar, for example during IOP5, but also has periods with large differences (e.g., more humid during IOP6), in line with expectations for the larger distance between sites. Specific humidity at site Bergen is substantially higher throughout the campaign (Fig. a, green line), except for short periods after IOP1, at the start of IOP3, and on 28 March 2021. While possibly coincidental, some of the increases and decreases appear to lead or lag compared to the measurements from Northern Norway. Detailed trajectory analysis in forthcoming studies will enable identification of any Lagrangian matches between Bergen and Andenes during this period.

Variations in specific humidity at the sites Coast, ALOMAR and Tromsø were also frequently reflected in the δD (Fig. b). During some episodes, there were marked deviations from this rule, such as during IOP5 (purple bar), where δD dropped markedly. Furthermore, offsets between Tromsø and the Andenes measurements become apparent in δD, again during IOP5 (23–24 March 2021) and during 27 March 2021, with Tromsø lagging by 3–6 h. Another interesting observation is that δD measured in Bergen co-varied with the other sites during some periods, such as parts of IOP1, IOP2 and IOP3, despite the more humid conditions. In these situations, the isotopic signal may contain information about distillation or evaporation effects during the ∼1000 km long transport path.

The δ18O in general shows a very similar relation between all sites as δD (Fig. c). However, the site Coast was on average 1.65 ‰ more depleted than site ALOMA ‰. Given the large correction for mixing ratio dependency that had to be applied to the δ18O, and the larger measurement uncertainty of that particular analyser, several lines of evidence were investigated to identify if this bias was real or an artefact of the calibration procedure. Computing the d-excess from the measurements, the site Coast would at times reach 52.1 ‰, with an average of 27.5 ‰. All other sites only reached a d-excess of up to 24.2 ‰ (ALOMAR), 34.3 ‰ (Tromsø), and 15.1 ‰ (Bergen). Furthermore, correspondence to the GMWL was investigated in a δD–δ18O correlation plot (Fig. a). Periods with high d-excess were far from equilibrium, and would have to have evaporated from very low relative humidity with respect to sea surface temperature. Even though it may be plausible to expect a higher d-excess at site Coast, which is closest to evaporation conditions, an overall lower δ18O than at ALOMAR, but a higher δD, was deemed implausible given the large correction of the raw δ18O measurement signal. Therefore, the median offset between site ALOMAR and Coast was calculated for the entire campaign, and then used to bias correct the δ18O of site Coast by 1.65 ‰ (Fig. b, black line). The uncertainty of the bias was estimated as 0.06 ‰, using the squared sum of the standard error of the mean scaled by π/2 for all δ18O measurements at ALOMAR and Coast, respectively. Combining the bias uncertainty with the total analytical uncertainty changed the final uncertainty estimate only marginally from 0.22 ‰ to 0.23 ‰ (Table ). The dataset on the data repository already includes the bias correction, such that it does not need to be applied by the data users.

This bias correction reduced the difference in average δ18O to -0.01 ‰, resulting in an average d-excess at site Coast of 14.3 ‰, and a maximum of 39.9 ‰ (Fig. d, black line). Compared to other sites, mCAO periods (IOP0, IOP1 and IOP3) still showed highest d-excess at site Coast after bias correction. Otherwise, a strong correspondence can be observed for the d-excess from the 4 sites during many situations, such as IOP5 and IOP6. During IOP3, all sites show an increase in the d-excess over the course of the mCAO event. Even the d-excess affected by room air measured in Tromsø matches well with the overall pattern observed at the other sites, indicating that despite the delayed and mixed signal, the d-excess from this time period still contains qualitative information over a time scale of hours to days.

A common framework to identify the relevance of mixing and Rayleigh fractionation processes in vapour isotope measurements is the δD–q mixing diagram . For the ISLAS2021 dataset, the mixing diagram shows a complex pattern (Fig. b). The most depleted and driest data points in the lower left quadrant are obtained from sites Coast and ALOMAR. Site Tromsø was at an intermediate range of water vapour mixing ratios (with the first half of the dataset not included here), while site Bergen is clearly at a regime that is more humid and less depleted in δD than the other network sites. Some mixing lines, with their typical logarithmic shape are evident in the measurements from Coast and ALOMAR (Fig. b, labels A and B). Some mixing lines are also evident in the Tromsø measurements. In addition, there are several vertical patterns evident in the diagram (labels C and D). These vertically oriented features have been observed previously in arctic water vapour isotope measurements . In the ISLAS2021 dataset, the vertical variations appear to reflect depletion during long-range transport, likely being a signal from cloud-level altitudes that is transferred to the vapour below cloud base by downdrafts and below-cloud exchange processes . These features in the δD–q diagram warrant further investigation in forthcoming studies.

In summary, we confirm that interpretable vapour isotope measurements have been made at four measurement locations over a scale of up to 1000 km that show connections between the evaporation, transport, and condensation history of different air masses during the campaign.

Precipitation at Andenes was distributed unevenly in time during the ISLAS2021 campaign (Fig. a). All IOPs were focused at weather events associated with more or less distinct precipitation periods. According to the Parsivel² disdrometer, the most intense precipitation was recorded early in the morning of 19 March 2021 during IOP1 (12 mm h⁻¹), followed by the evening of 24 March 2021 during IOP5 (7.2 mm h⁻¹). Precipitation recorded by the rain gauge at site Andenes, 4 km from site Coast, records precipitation amounts that are generally similar in overall amount and timing. The disdrometer and the rain gauge have different uncertainties that can contribute to such differences. Wind-related undercatch is a common problem for rain gauges , whereas the Parsivel² overestimates precipitation rates for solid precipitation (snow and melting snow). Both effects are likely to contribute to discrepancies between the precipitation measurements, in particular during IOP1.

During several IOPs, precipitation was collected at up to 10 min intervals for water isotope analysis (Sect. ). It should be noted that the uncertainties in precipitation amount measurements do not transfer uncertainty towards the water isotope information in the precipitation samples unless one is interested in the amount-weighted isotope signal, for example in hydrological applications. In order to facilitate comparison with the vapour isotope measurements, we calculated the water vapour in isotopic equilibrium with the precipitation for the average air temperature of the precipitation interval at the respective sites. This so-called equilibrium vapour is denoted by δD_p,eq, and obtained from computing the vapour in equilibrium with the precipitation at a given temperature using established isotopic fractionation factors .

The high-resolution sampling revealed large short-term variations in the isotopic composition. During IOP0, the δD_p,eq varied between -160 ‰ and -30 ‰ (Fig. b, red bars). Distinct variations, albeit at lower magnitudes, were also observed during IOP1, IOP2, IOP5 and IOP7. The smaller variation in the precipitation suggests that the surface vapour follows the precipitation d-excess as a result of (often limited) below-cloud exchange processes . The vapour isotope composition measured at site Coast δD showed variations corresponding to the SWI in precipitation, albeit at a smaller amplitude (Fig. b, grey line). During IOP3 to IOP5, precipitation was also collected at ALOMAR at high resolution. Here, the δD_p,eq shows variations similar to those observed at Coast (Fig. b, blue bars). From 19 to 29 March 2021, precipitation was also collected at Bergen at high time resolution, with largest rainfalls during IOP2 to IOP4 (Fig. b, green bars). A comparison between the precipitation d-excess (not the equilibrium d-excess) with the water vapour d-excess at site Coast shows an astonishing degree of correspondence for several IOPs (Fig. c, red bars and grey line). For example, the transitions from IOP2 to IOP3, as well as from IOP4 to IOP5 match closely in terms of timing and magnitude. We consider this as support for the bias correction performed on the water vapour δ18O. The d-excess in precipitation from ALOMAR and Bergen is within 10 ‰ or less from the d-excess at site Coast.

During the campaign period, the INP concentration at site Coast varied between 6.7×10-4 and 3×10-2 (std L)⁻¹ at -15 °C and 5.8×10-3 and 1.5×10-1 (std L)⁻¹ at -20 °C (Fig. a). The highest INP concentrations were observed during the intense cyclone (IOP5) coinciding with the highest wind speeds and heaviest precipitation rates. Although, it should be noted that no clear relationship between INP concentration and wind speed was observed . Meanwhile, the lowest INP concentrations were observed during 27–28 March 2021, which was a period characterised by above-freezing temperatures, wind speeds between 5 and 10 m s⁻¹ and intermittent precipitation.

More generally, the INP concentrations observed during ISLAS2021 are similar to previous studies in the Norwegian Arctic. As shown by , similar INP concentrations were observed during cold-air outbreaks on Andøya and during the fall and spring in Ny-Ålesund e.g.,. Even though these sites lie on opposite ends of CAOs and WAIs, the concentrations of INPs are comparable. Similarly to , who connected the isotopic fractionation of water vapour and precipitation to cloud microphysical processes, the observations presented here could in future work be exploited by linking the collocated INP and water isotope measurements to study differences in the ice-nucleating ability of different moisture and aerosol sources.

When comparing the INP concentration with the aerosol size distribution as measured by the APS, there is no clear relationship between periods of elevated INP concentrations and generally larger aerosol particles (Fig. b). This is consistent with the lack of relationship observed between the INP concentration and the aerosol concentration larger than 0.5 µm as measured by the OPC . The concentration of aerosol particles larger than 0.7 µm varied between 0.008 and 36.7 cm⁻³ with the highest concentrations occurring during high wind speeds in the afternoon of 19 March 2021 and the lowest concentrations occurring on the morning of 24 March 2021 during IOP5 (Fig. b). More generally, relatively clean periods were observed in conjunction with precipitation as expected due to wet-scavenging e.g.,.

During the campaign period, a total of 8 horizontal transects of precipitation samples have been collected. These transect samples enable dataset users to assess the horizontal representativeness of the precipitation isotope measurements at sites Coast and ALOMAR within a range of up to 100 km. Due to their distance from the site Coast, the samples in the boxes were exposed to the atmosphere for up to several days before being collected. On some occasions, the box samples were also supplemented by surface snow samples collected nearby or at additional locations (Fig. ). Since the sampling locations are roughly oriented in N–S orientation (Fig. b), the transect results are displayed using the latitude of the sampling locations as horizontal axis (Fig. ).

The precipitation δD from boxes varies substantially between events, much more than between collection sites for the same event (Fig. a, coloured x). The average precipitation isotopes measured at site Coast for the corresponding period roughly agrees with the transect samples for most transects (coloured +). However, marked differences are noted for transect T3 and T4. These two transects were collected back-to-back during subsequent IOPs, and may contain spillover from the previous events due to delays in collecting the boxes. The correspondence to site coast is further highlighted by a difference plot (Fig. c), which confirms that all transects but T3 and T4 are within about 25 ‰ from the observations at site Coast. Thereby, events T8 and T6 show a slight tendency towards more depleted values, whereas T7 is less depleted.

The d-excess from the sampling boxes shows a more narrow distribution than δD with values between 5 ‰–10 ‰, and with more scatter at the more distant boxes (Fig. b, coloured x). This range of values is generally consistent with the average d-excess from corresponding precipitation at site Coast (coloured +). The difference plot in d-excess shows that the box samples are about 0 ‰ to 10 ‰ lower in d-excess, possibly indicating evaporation due to longer exposure times (Fig. d). However, the d-excess is substantially higher in all surface snow samples, compared to the corresponding box samples (Fig. b and d, coloured dots). The cause of this positive bias of about 20 ‰ or more in the surface snow samples is currently unclear, and may be related to mixing and exchange processes with the snow pack. Thus, while these transect samples deserve to be investigated more on a case-by-case basis, the fact that inter-event variations of up to 80 ‰ in δD are substantially larger than for the samples from a given transect (2 ‰–20 ‰) clearly indicates the possibility to determine the representativeness of Coast precipitation isotope measurements for the 100 km long transect across the Vesterålen archipelago.

The water vapour isotope measurements at sites Coast and ALOMAR were made at a horizontal distance of ∼2 km, and at an elevation difference of 364 m (Fig. a). Given sufficient accuracy and precision of the measurements, this setup enables the quantification of vertical gradients in the lower atmosphere during the campaign. As there is a temporal offset between the air masses arriving at the Coast and ALOMAR of several minutes (see below), we use 10 min averages to assess if a measurable gradient is present between the two locations. For δD, the probability density function leans to the left, showing a predominance of a negative gradient (Fig. a). The maximum is at -2 ‰, resulting in a gradient of ∼-0.57 ‰ (100 m)⁻¹. In the large majority of cases, the difference is within ±5 ‰. For the d-excess, the gradient between ALOMAR and Coast Δd=dALOMAR-dCoast (‰) is more pronounced, and predominantly negative (Fig. b). The maximum of the probability density function is at -5.2 ‰ corresponding to a d-excess gradient of 1.4 ‰ (100 m)⁻¹, with a secondary maximum close to zero. Due to both the offset applied to the δ18O data, and relatively low measurement precision of the analyser at site Coast, the vertical gradients of the d-excess are associated with larger uncertainty. Nonetheless, a gradient clearly emerges from the measurement uncertainty. Classification of the 10 min periods into IOP categories (Fig. , shading) shows that the strongest negative gradients are associated with mCAO periods when surface fluxes and non-equilibrium fractionation are strongest . In comparison, the gradients are substantially smaller for most of the time during cases dominated by mid-latitude air advection. It will thus be possible to further utilise this dataset for finding how weather events are associated with more or less mixing, and stronger or weaker surface evaporation. We note that the gradients found here are quantitatively very similar when averaging the data at a 30 min time interval, confirming the robustness of the found gradient characteristics (not shown).