Ny-Ålesund is located at the southern coast of Kongsfjord on Brøggerhalvøya, a peninsula on the northwest coast of Spitsbergen (78.9° N, 11.9° E), the largest island of the high Arctic Svalbard Archipelago, Norway (Fig. 1a). Located in the North Atlantic sector of the Arctic, the island's climate with a mean annual air temperature of -8.7°C over land areas (Sval-Imp 1971–2000, based on Schuler, 2018, in Hanssen-Bauer et al., 2019) is comparatively mild due to the influence of ocean and atmospheric circulation patterns like Gulf Stream and Arctic Oscillation that facilitate the advection of warm-moist air masses and rapid weather changes (Maturilli and Kayser, 2017; Yeo et al., 2018).

Ny-Ålesund is particularly interesting for investigating melt effects in snow and firn, because its coastal Arctic climate is changing dramatically. Annual mean air temperature has risen by 0.71 °C per decade since 1971, and winters warmed at even higher rates (+1.35°C per decade, 1971–2017, Hanssen-Bauer et al., 2019). This has generally led to shorter winters (Pedersen et al., 2022) and earlier starts of the melting season in Svalbard, but also more frequent rainfall (Peeters et al., 2019), severe ROS (Vikhamar-Schuler et al., 2016) and melt events during winter (Graham et al., 2017). Consequently, the seasonal snowpack adding up to 40 % to the ∼60% spatial glacier coverage of Svalbard through the annual cycle (Gallet et al., 2019) is changing. The Arctic snow period is shortening 2–4 d per decade, and heightened ice layer formation in the snowpack due to ROS events (Brown et al., 2017) is an urgent issue for the local fauna, as their access to sub-snow grasses and moss as essential food source is diminishing.

In this context of rapid environmental and snowpack changes, the topography of Brøggerhalvøya and glaciers near Ny-Ålesund offers a well-suited study region that allows study of the influence of ROS-induced melt events on an Arctic snowpack at a range of elevation (from sea level to >380ma.s.l.) and temperature conditions with easy site access. Year-round international scientific research at the Ny-Ålesund Research Station (https://nyalesundresearch.no/, last access: 27 September 2026) further facilitates interdisciplinary collaboration and continuous observational records (Maturilli et al., 2019; Pedersen et al., 2022), including, e.g., meteorology and physical snow properties (Scoto et al., 2023).

By integrating structural and chemical snow measurements in a new experimental melt study, this collaboration of British Antarctic Survey (BAS), Norwegian Polar Institute (NPI) and the Italian Institute of Polar Sciences of the National Research Council of Italy (CNR-ISP) addresses current knowledge gaps regarding the melt effects for the cycling of snow contaminants in Svalbard (Gallet et al., 2019). Hosted by the UK's Natural Environment Research Council (NERC) Arctic Station, this project feeds into the Glaciology flagship, i.e. one of the four priority programmes for research and monitoring at the Ny-Ålesund Research Station (https://nyalesundresearch.no/research-and-monitoring/glaciology/, last access: 27 September 2026). Percolation experiments are unprecedented at this polar site and have the potential to provide insights into meltwater percolation in cold polar firn that are relevant for interpreting ice cores from Svalbard and the polar regions more widely.

To investigate the structural and chemical imprint of ROS-induced melt events on the Svalbard snowpack before the major onset of the melting season, we conducted a series of 17 in-situ percolation tracer experiments in 9 snow pits near Ny-Ålesund from 21 until 31 March 2023 (Fig. 1b). The field campaign was initiated with an overview of avalanche-probe-based total snow thickness measurements (±0.5cm accuracy, provided as snow height above ground) on the local slopes near sea level on 18 March 2023, which informed the subsequent snow pit and experiment site selection near the Gruvebadet Atmospheric Laboratory approximately one kilometre south of Ny-Ålesund (Fig. 3a). To ensure sampling of the unaltered local snowpack, we chose study sites with pristine snow surfaces away from snowmobile tracks and traffic routes. In addition, aiming to identify a locally representative study site each day, we took >25 snow thickness measurements on a 10m×5m grid and dug the snow pit at a position with average snow height within this grid.

The in-situ experiments provided a comprehensive picture of the pre- and post-melt preservation including the percolation process near Ny-Ålesund. They are based on a combination of visual, structural, temperature and chemical measurements conducted before (Sect. 2.2.1), during (Sect. 2.2.2), and after tracer application (Sect. 2.2.3). To be applicable in environmentally protected areas like Brøggerhalvøya, the concept only draws upon natural substances, and to not interfere with other ongoing observations in the area, we only apply Bluetooth-/Wifi-free measurement techniques. To ensure methodological consistency, all experimental procedures and analytical steps were conducted by the same individual.

To investigate stable water isotope signatures in the snowpack before versus after ROS-induced melting, we sampled at least one of the fresh snow pit fronts in several affected (blue) and unaffected dry (white) parts using 50 mL polypropylene centrifuge tubes with strong screw caps. Sampling generally included the near-surface melt layer and snow underneath but also addressed more complex cases including PFPs and discontinuous melt lenses. For example, samples were taken 1.5 cm to the left and right of a PFP as well as within (23 March 2023). Based on the already decreasing temperatures in the snow pit, we thereby assume that meltwater movement in the snowpack has ceased, and the sampling conditions capture the post-percolation steady state. This combination allows us to assess SWI alteration of near-surface snow layers in the context of pre- and post-melt structural features.

The discrete snow samples taken before and after melt experiments for SWI evaluation were stored in a freezer at the NERC Arctic Station until shipment to the British Antarctic Survey in Cambridge, UK, on 4–5 April 2023, where they were kept at -25°C until analysis. To prevent cross-contamination between samples during transport, all screw tops were sealed using Parafilm, sample tubes double-bagged in batches of 10 and kept upright within the transport box. While leakage was observed in ∼15% of the samples despite these measures, we expect the material remaining within sample tubes to be unaffected.

The SWI measurements of oxygen (δ18O) and hydrogen (δD) were conducted using a Picarro L2130-i setup based on cavity ring-down spectroscopy (Gkinis et al., 2011; Gupta et al., 2009) located at BAS in June 2023. For every sample, seven injections were measured, of which the first three were removed for memory effects and the average of the last four makes up the respective SWI signature. δ18O comes at an instrument accuracy of 0.2 ‰ and δD at 0.6 ‰. The measurements are presented in relation to the Vienna Standard Mean Ocean Water (V-SMOW), and average 1σ precision of the here conducted measurements lies at 0.1 ‰ for δ18O and 0.3 ‰ for δD. All mean values in this study are reported including standard deviation (±2σ).

When comparing δ18O signatures before and after experiments, the alteration due to melt percolation and refreezing has been identified based on the variance defined by nearest-neighbour pairs. To ensure a direct comparison is appropriate, only post-experiment samples taken ±1cm of the corresponding pre-melting SWI sample have been considered here.

To validate the in-situ observations, we use a set of automated nivometric station (ANS) measurements at Gruvebadet (78.917° N, 11.895° E). Scoto et al. (2023) have provided a detailed introduction to the ANS setup, which among other sensors includes total snow thickness (provided as snow height above ground), liquid water content (LWC), 2 m air temperature, snow temperature at 10, 40 and 55 cm above ground, and the snow surface temperature (Table 1). Here, we focus on the named records over the period 13–31 March 2023 at 10 min resolution.

The preliminary snow thickness sampling in the Ny-Ålesund area on 18 March 2023 indicated an average snow height of 68.3 cm (80 measurements, ≥1m apart, ±75.6cm standard deviation). The more detailed 251 snow thickness measurements taken during the selection of snow pit sites average 59.0±24.3cm, covering the range 36.0–95.0 cm (Fig. 3). These conditions are represented by the nine snow pits of this study, which average 55.7±22.3cm in snow height (43.0–72.5 cm).

Snow temperatures before the experiments range from -29.0°C to -4.1°C and average -12.1°C. The daily mean snowpack temperature lies at -12.2°C, warming from -13.3°C to -8.0°C during the field campaign. At the start of fieldwork, the vertical temperature profile exhibits a strong temperature difference (ΔT) between the top and bottom 10 cm (ΔT=-19.5°C on 21 March 2023), which is erased at the end (ΔT=+1.0°C on 31 March 2023, Fig. 4a). All the while, the bottom snow temperature (0–5 cm snow height) remains largely constant, averaging -6.7±1.5°C (Fig. 4a).

The assessments of snow stratigraphy reveal an average of 10±2 layers in the current seasonal snowpack near Ny-Ålesund in March 2023. A detailed depiction of all snow pits is given in Appendix C. In many profiles, dense wind slabs with fine rounded grains alternate with coarser-grained faceted and hoar crystals, which are especially dominant in the lower parts of the snowpack (Fig. 4c).

Furthermore, eight out of nine snow pits already contain melt-affected layers like melt-cluster grains, melt–freeze crusts or cohesive ice layers prior to the experiments (mean=2±2, e.g. Fig. 4c and d). These 20 melt-affected layers are clustered slightly at 4–5 depth horizons (Fig. 4b), of which the upper ones at ∼30–34 and ∼47cm snow height are most distinct (Fig. 4b). Snow density measurements range 106–470 kgm-3 and average 306±160kgm-3 when considering all snow pits, and densification trends with depth are weak to insignificant (Appendix C).

Snow thickness measurements by the ANS show stable conditions during the field campaign (21–31 March 2023) with surface snow height at approximately 69.5±8.8cm and two snowfall events on 22–23 March 2023 (∼5cm snow) and 30–31 March 2023 (∼13cm snow). During the same period, 2 m air temperature ranged from -24.7 to -2.8°C (mean=-14.2±9.2°C). Starting with a cold spell below -22°C on 20 March 2023, the temperature fluctuates from -18 to -12°C until 29 March 2023 and then quickly rises to -3°C on 31 March 2023 (Fig. 5).

The temperature profile of the snowpack recorded by the ANS shows large gradients between the snow surface and lower parts of the profile, especially during cold spells like 20–21 March 2023, when snow surface temperature dropped distinctly below air temperature. However, the snowpack reaches almost isothermal conditions (-7.5°C) on 30 March 2023. The liquid water content (LWC, Appendix C) is generally low in this dry snowpack and within the range of residual water content (3 %–6 %, Fierz et al., 2009). The highest LWC of 5 % is registered 10 cm above ground, while the near-surface snow 60 cm above ground holds <0.5% LWC at maximum during cold conditions. In the lead up to the fieldwork campaign during February 2023, air temperature reached values above freezing on at least three occasions (T>0°C: 5–6, 14–15, and 25–26 February 2023), and two sub-zero yet warm periods (-3°C<T<0°C: 21–24 February and 1–2 March 2023) could have also enabled changes to the snow structure.

The presented ROS experiments are of urgent relevance, as Svalbard holds key Arctic ice-core drilling sites and ROS-driven melt events are already becoming more frequent in its coastal region up to a few hundred meters of altitude, so that a deeper understanding of their impact on polar snow and firn, including SWI climate signatures, is needed. Our experiment approach seizes various elements from previous field studies: working in situ, using a SWI-tracer, and comparing before and after ROS percolation (Juras et al., 2016, 2017; Wong et al., 2013). At the same time, we kept ROS tracer application in the field simple to facilitate experiments with limited logistical, financial or technological capabilities. Moure et al. (2023) have recently reiterated the need for well-controlled experiments including in-depth assessments of snowpack properties, i.e. snow temperature, grain type, and melt-flux rates, to further improve modelling of meltwater flow through porous snow. We agree that laboratory experiments are irreplaceable to ensure repeatable ambient conditions and reproducibility of percolation behaviour.

In this context, our field-based experiments with detailed structural and chemical analyses pre- and post-experiment can serve as an analogue counterpart. Simulating the percolation process in situ bears the large advantage that the stratigraphic layering of the snowpack under investigation remains unchanged. Thus, its influence on local percolation depth can be quantified more reliably, e.g. by including depths, hardness and grain-size differences between stratigraphic units as independent variables during MLRA. There are some studies dealing with melt-induced chemical alteration or structural changes through melt (e.g. Katsushima et al., 2013, 2020; Lee et al., 2010a, b; Spolaor et al., 2021), but this study is one of only few to investigate percolation within the most recent annual snow layer visually, structurally, and chemically at the same time (Juras et al., 2016; Waldner et al., 2004; Wong et al., 2013).

Following, we briefly discuss (1) environmental conditions, (2) ROS setup, and (3) ROS intensity to delineate the transferability of the here obtained experiment results.

Aiming to capture warming snow conditions prior to the onset of the annual snowmelt, the fieldwork campaign has been timed for mid-to-end March. Successful in conducting the percolation experiments before a severely snowpack-depleting melt event in spring 2023, cold ambient temperatures during fieldwork mainly depict a non-ripe snowpack (21–29 March 2023) and less of the ripe snow conditions typical of spring (30–31 March 2023). The cold had to be compensated for by warming the tracer liquid (average=+11.7°C) to prevent freezing within the sprinkler. To monitor the impact of this heat source on flow progression, tracer temperature was recorded immediately before application and included as an independent factor for percolation depth during regression analysis. Future studies should aim to repeat the in-situ percolation tracer experiments over a longer period including warmer conditions to assess the range of melt effects as previously conducted in alpine snow (Juras et al., 2017). Further, keeping liquid temperature steady would reduce the number of changing parameters.

To examine whether snow characteristics before the tracer experiments are typical and representative of the Arctic snowpack in March 2023, we discuss the measured snow properties in the nivometric context. While individually matching melt-affected layers with meteorological events is beyond the scope of this study, the number of melt events is similar to and therefore expected to explain the stratigraphic clusters of melt horizons in the Ny-Ålesund snowpack. More specifically, the melt phases mid- and end-February appear to correlate with melt-affected layers around 30–34 and 47 cm depth. Layers of faceted and hoar crystals have developed through constructive metamorphism within the snow, facilitated by the strong temperature gradient between permafrost and near-surface snow exposed to variable ambient temperatures. This is corroborated by the density profiles, in which the lack of densification can be attributed to de-densifying temperature gradient metamorphism. Together with the observed fine-grained wind slabs, the snow profiles show typical characteristics of a high Arctic maritime snowpack (Eckerstorfer and Christiansen, 2011). Furthermore, the quick warming of seasonal snow and change to isothermal conditions towards the end of this study is an example of rapid temperature transitions, which occur frequently around Ny-Ålesund (Yeo et al., 2018). Comparing pre-experiment surface snow height in the vicinity of Ny-Ålesund (59.0±24.3cm) and the surface snow height recorded by the automatic nivometric station at Gruvebadet (69.5±8.8cm) during the fieldwork period, our measurements indicate slightly less seasonal accumulation on the Gruvebadet slope than at the nivometric station. Given that wind-driven redistribution is known to cause short-distance snow height variability, we nevertheless consider the sampling conditions representative of the Ny-Ålesund snowpack in March 2023.

The in-situ experiments show a range of melt features with structural heterogeneities on a millimetre-to-decimetre scale (Sect. 3.3.1), revealing complex modulations of the percolation process. The experiments clearly illustrate how various characteristics can result from single events. Saturated centres of melt lenses are often surrounded by undersaturated wet snow (Fig. 6e), and an example experiment time-lapse video (31 March 2023, Moser et al., 2025c) shows how this could result from stepwise advances of meltwater along the thermo-hydrological percolation front.

In addition, we for the first time report internal layering of melt lenses in situ on a millimetre scale as a consequence of a single percolation experiment (Fig. 6h and i). Lab experiments and modelling of percolation by Avanzi et al. (2016) showed spatial heterogeneities on a centimetre scale and highlighted the role of capillary boundaries for preferential flow, but to our knowledge, this study is the first to show millimetre-scale heterogeneities developing in situ. Such internal layering confirms that not all of the snow properties are eradicated through melt infiltration, but remnants of pre-experiment snow structure can remain within melt-affected sections. This finding is in agreement with 3D ice μCT scans revealing firn remnants within melt sections (Moser, 2025). Therefore, it needs to be re-emphasized that high-resolution measurements of snow and firn structure are crucial in supplementing meltwater flow modelling to improve our understanding of melt feature formation.

The spatial heterogeneities of melt lenses and PFPs on a centimetre-to-decimetre scale and internal layering on a millimetre scale (Fig. 6) also demonstrate how challenging it is to structurally differentiate individual melt events or melt-intensive seasons in firn and ice cores (e.g., Moser et al., 2024). Some melt features lack both lateral continuity and vertical homogeneity, which limits the ability to extrapolate melt conditions from a single 9 cm diameter profile. Moreover, the number of ice layers in the snowpack will not necessarily be equal to the number of ROS events that have occurred during the season (e.g., Pfeffer and Humphrey, 1998). Repeated percolation experiments at firn core drilling sites could help to create a site-specific framework of expected melt structures and substantiate their environmental interpretation.

Here, we use the structural characteristics of melt features and MLRA (Appendix E) to explore the local factors that determine the percolation process and refreezing conditions.

Firstly, multiple linear regression analysis of the experiment results indicates that snow surface temperature, the location of a first hard layer within the snowpack, grain size differences at the first and second stratigraphic boundaries, liquid temperature and volume can explain ∼63% of average percolation depth variance of all front cuts (adjusted R2=0.629, n=47). More specifically, surface snow temperature (p<0.01), the grain size difference at the first stratigraphic transition (p<0.05) and liquid volume (p<0.01) are positively related to average percolation depth. For the snowpack context, this means that higher snow temperatures allow for deeper meltwater infiltration, fine-to-coarse transitions retard vertical flow, and more intense ROS events reach deeper parts of the annual layer. While quantifying the relative contributions for individual results is hindered due to the multi-dimensional variability of in-situ conditions, the cases discussed here exemplify constraints identified by MLRA.

Already the first imprint of the percolation tracer on surface snow differs visually depending on the snow temperature. In cold snow like on 21 March 2023 (-25°C), we observe distinct bubble-free surface melt layers as well as confined preferential flow fingers and melt lenses (Fig. 6a and e). In warmer snow like on 31 March 2023 (-7°C), there is no refrozen surface melt layer but wider-spread, partly undersaturated percolation that causes snow compaction and a lowering of the surface in the 40cm×20cm experiment area (Fig. 6b, f, and g). This suggests that there is a shift from preferential, spatially heterogeneous flow to more homogeneous matrix flow, which is dependent on the dynamic evolution and interplay of air and snow temperature. Both of these primary flow regimes have been previously documented (Moser et al., 2024, and references therein), and the experiments in this study highlight the importance of refreezing capacity for percolation behaviour. At the same time, Waldner et al. (2004) conducted a series of percolation experiments at 0 °C, reported both preferential and matrix flow, and showed that stratigraphy and grain size distribution within the snowpack are major factors, too.

Analysing the average percolation depths in the context of near-surface snow stratigraphy shows that the in-situ observations are similarly valuable to explain meltwater progression and retardation here. More specifically, the position, hardness and grain-size difference of stratigraphic boundaries can be strong internal constraints for local percolation depth in the experiments beyond the influence of snow temperature. For example, average snowpack temperature was similar on 29 March 2023 to previous days (24 March 2023: -12.5°C; 29 March 2023: -12.7°C) and surface snow comparatively warm (Fig. 4; 24 March 2023=-20.2°C; 29 March 2023=-15.7°C). However, the melt tracer was retained largely within the topmost (1.5 cm), fine-grained (average diameter ∅= 0.2 mm) stratigraphic unit during the first 250 mL experiment and above the first hard layer (3 cm depth) during a subsequent 725 mL experiment (Fig. 6c). Instead of progressing vertically, the tracer spread laterally beyond the experiment boundaries. In these cases, stratigraphy seemed to override snow temperature as a constraining factor and partially suppress the influence of liquid volume. Furthermore, they show that vertical flow retardation and ponding occur not only at high-density boundaries (e.g. 22 March 2023) but at fine-to-coarse transitions due to capillarity (e.g. Waldner et al., 2004, and references therein).

Placed in the wider literature context, the identified major percolation constraints snow temperature, stratigraphy and melt rate are consistent with previous experimental, modelling, and geophysical findings. The here reported features partly resemble the spectrum of flow conditions documented by Bøggild (2000) in tracer dye experiments in Greenland's Tasersuaq basin: near-surface non-preferential infiltration, sharply defined flow fingers, and poorly confined flow fingers transitioning to matrix flow. Although not directly comparable, their study also suggests time-dependent components for both thermal and gravitational retention. Pfeffer and Humphrey (1998) numerically modelled percolation and refreezing of melt and point out the joint importance of “initial temperature, grain-size and density contrast across the stratigraphic transition, water-input rate and minimum impermeable-layer thickness”. In addition, ground-penetrating radar measurements in melt-affected Greenland firn by Culberg et al. (2021) generally stress the importance of the regional thermal regime for melt layer prominence but also the modulating effect of more heterogeneous stratigraphy on percolation behaviour.

After all, questions about the shift from preferential, spatially heterogeneous flow to more homogeneous matrix flow have previously been raised for wet snow, e.g., by Hirashima et al. (2019) and Avanzi et al. (2016). The in-situ observations made here reinforce arguments for a snow temperature-driven regime change and that melt features not only reflect positive air temperatures but a dynamic interplay of air and snowpack, thereby agreeing with existing literature (Moser et al., 2024) and diverse melt structures in firn cores (MacDonell et al., 2021). They also underline the great value of detailed and continuous snowpack assessments as designed by Scoto et al. (2023) for reconstructing meltwater flow and quantifying potential alteration of extracted climate records.

The fact that SWI signatures before and after experiments differ distinctly shows that the study approach is successful and melt percolation is chemically traceable (Sect. 3.3.2). Moreover, the δ18O enrichment in refrozen melt layers generally aligns with broader literature (e.g. Beria et al., 2018; Moser et al., 2024, and references therein).

Examining the individual percolation experiments as case studies, it becomes evident that SWI changes are highly localised and confined to the melt-affected lenses and percolation features on a centimetre scale (Fig. 7). For example, a PFP (Fig. 7b; 23 March 2023) carries an enriched δ18O signature of -15.8‰, but two neighbouring samples taken 1.5 cm left and right of it show pre-experiment values (-19.5‰; -19.5‰). Similarly, the δ18O signature between the visible surface melt layer and a deeper melt lens remains unaltered (-19.7‰).

To further investigate how localized SWI alteration is, SWI samples were taken 1 cm, 3–3.5, 4.5, 6.5–7 and 10 cm below the surface in the left and right part of the experiment area, respectively (Fig. 7c; 27 March 2023). While the effect of melt on δ18O is laterally uniform within the surface melt layer at 1 cm depth, differences are apparent at 3–3.5 cm depth, where the left sample is taken from a melt lens (-19.3‰) and the right is unaffected with a δ18O signature similar to pre-experiment background (-22.6‰, Fig. 7c). Comparing this with a melt-affected sample on the right at 4.5 cm depth (-18.7‰, Fig. 7c) reveals a vertical difference of +3.9‰ between non-lens and melt lens sample material. Thus, meltwater percolation not only leads to mixing and SWI enrichment in parts of the snowpack but also introduces small-scale spatial variability (millimetre-to-centimetre), which need to be considered when interpreting SWI records from melt-affected profiles (Fig. 7c).

The presence of distinct SWI signatures above and below ice layers further supports the idea that refrozen melt can temporarily stabilise the surface near Ny-Ålesund. By preventing wind-driven redistribution and mixing, they can foster distinct stratification and improve preservation of the SWI record beneath and between melt horizons. This complements previous studies that mention improved SWI preservation, e.g. Van Der Wel et al. (2011) describing melt layers as diffusion-limiting.

To quantitatively assess how meltwater flow influences SWI signatures, it is crucial to consider whether and how the sampling location within the flow path affects the results. In an experiment on 30 March 2023, both δ18O and δD exhibit a clear tracer imprint in the surface melt at 3 cm depth, indicating active percolation. However, in the lower flow finger and melt lens, tracer presence is evident as a δD increase, while δ18O remains unchanged compared to pre-experiment values. This stability in δ18O suggests that meltwater movement may have slowed or ceased at the percolation front, preventing further isotopic alteration. Consequently, SWI changes appear to occur progressively, with the degree of modification depending on melt intensity and whether the sample is taken from a pick-up zone, a transition or liquid flow area, or a refreezing regime.

Since melt layers influence both the retention and redistribution of isotopic signals, they shape the sub-annual resolution of SWI proxy records. To fully capture these effects, it is essential to link the structural impact of refrozen melt to its chemical imprint (Westhoff, 2021), ensuring a more robust interpretation of melt-affected isotope records. Overall, the spatial variability in SWI alteration underscores the need for careful sampling strategies in melt-affected snow and firn. Melt percolation introduces complexities that must be considered when using SWI records for climate reconstructions.