Introduction

Snow on sea ice is the critical interface between the atmosphere and underlying ice, playing a crucial role in the sea-ice system’s energy fluxes1. Accurate measurements and understanding of internal snowpack processes are essential, yet the remoteness of most Arctic sea-ice regions limits data collection and knowledge of these processes. Snow on sea ice is exposed to large temperature gradients (ΔT) due to the cold atmosphere and the relatively warmer ice-ocean interface, sustained by oceanic heat flux. These temperature gradients drive water vapor transport and mass flux within the snowpack, leading to substantial internal snow recrystallization through metamorphic processes that reorganize the snow crystals into new structures. This is known as temperature gradient metamorphism (TGM)2,3,4, and a metamorphosed snow grain type, called depth hoar, accounts for 20–50% of the Arctic snowpack5,6,7,8.

Internal snowpack water vapor transport is known to occur through both diffusion and convection; however, the contribution of the latter remains in discussion9,10,11,12. This is due to the dependency of convection on the snow density (ice volume fraction) remaining small enough to trigger convection circulation in the air pockets13. Complicating matters further, there is no definitive answer to the value of the vapor diffusion coefficient in snow14,15. Therefore, modeling and quantifying this vapor transport in the snowpack is complex but essential for understanding internal snow metamorphism and stratigraphy evolution16.

The central Arctic snowpack is heavily influenced by i) wind redistribution, which erodes and deposits the snow17,18 and ii) wind compaction, densifying the upper snowpack. These windy conditions prevent the preservation of atmospheric snow precipitation events in the snowpack stratigraphy, unlike a layered snowpack commonly found in the Alps. Typically, simplistic modeled Arctic snowpacks have upper wind-packed layers with low permeability composed of well-sintered rounded grains and lower metamorphosed depth-hoar layers. However, large spatial heterogeneity and correlation length scales of 20–40 m19, make the in situ stratigraphy datasets challenging to interpret.

Results from fieldwork on sea ice during the 1997/98 SHEBA expedition measured the depth-hoar development in the lower part of the snowpack and hinted towards water vapor transport originating from the ice substrate, which could serve as a potential moisture source5. This mass flux had been previously proven during field measurements of snow above a soil substrate, where the vapor flux out of the soil into the snow was measured by trapping and weighing the moisture that accumulated under an impermeable sheet20. They found the flux averaged 2.6 × 10−7 kg m−2 s−1. Similarly9, measured an upward vapor flux of 2.8–3.5 × 10−7 kg m−2 s−1 from a soil substrate into the snowpack and ref. 4 suggested soil moisture contributes to ~5% of the snow-water equivalent (SWE).

Laboratory and modeling experiments have applied temperature gradients (ΔT > 50 K m−1) to shallow snowpacks, resulting in substantial structural transformations13,21,22. One study observed a snow mass turnover of 60 % per day14. These temperature gradients can potentially drive mass from the ice substrate into the snow23. A controlled laboratory experiment measured the mass incorporation and microstructural changes in snow above an ice lens24. They exposed the snow to ΔT = 100 K m−1 and observed the rate of the ice lens sublimation. The rate of ice mass flux from the sublimating ice lens into the snow in this laboratory setting was 1.2 × 10−6 kg m−2 s−1 24. Figure 1a, shows an annotated micro-computed tomography (microCT) image from these laboratory experiments alongside conceptual isotopic enrichment pie charts, which are explained later in the stable water isotope section. In this microCT image, we see the development of depth hoar after 21 days alongside the sublimation of the ice substrate. Under these conditions (–11 ∘C  < snow-surface temperature (Ts) < –16 ∘C, snow-ice interface temperature (Ti) = –3 ∘C, snow height (hs) = 100 mm), the downward movement of the ice lens over time (t, time in days) is given in Equation (1), where zil − zil,0 represents the thickness of sublimation of the sea-ice surface (ice thickness equivalent, ITE given in mm)24. The correlation coefficient and p value of this relationship were 0.92 and 1.92 × 10−47.

$${{\rm{ITE}}}={z}_{il}-{z}_{il,0}=0.11t+0.07$$
(1)

Stable water isotopes (δ2H and δ18O) are geochemical tracers, elucidating the historical processes of snow, such as vapor origins, re-crystallization, and atmospheric conditions, e.g., temperature25,26. Throughout this study, the ratio of the heavy to light isotopes is given as the deviation from Vienna Standard Mean Ocean Water (V-SMOW), as δ2H and δ18O in permil (‰).

Fig. 1: Conceptual model of sea ice sublimation and the isotopic two-source mixing model leading to 18O enrichment of the lower snowpack.
Fig. 1: Conceptual model of sea ice sublimation and the isotopic two-source mixing model leading to 18O enrichment of the lower snowpack.
Full size image

a The upper image shows an annotated microCT sample that highlights the processes of mass redistribution from the substrate into the snow from a previous lab experiment, where the underlying ice lens is now representing a sea-ice substrate. This microCT sample was first published by Mareike Wiese24https://www.research-collection.ethz.ch/entities/publication/133e13d0-e64f-413a-9676-6b93893f66cf. We have obtained permission from the rights-holder for its use in this publication, and it is now licensed under a CC BY 4.0 license https://creativecommons.org/licenses/by/4.0/. The original blue background and original annotations have been removed, and all annotations on this figure are modifications. Ice thickness equivalent (ITE, mm) is the thickness of sublimation of the sea-ice surface over 21 days. The pie charts show the approximate ratio of 16O (red) to 18O (blue) of solid ice (block color) and vapor (dashed). The red arrows represent vapor fluxes. The lower figure shows a cross-section of sea ice, where hs = snow height, hi = ice thickness, not drawn to scale. b One point represents one winter snow sample. The solid and dotted lines are the average profile with one standard deviation ( ± 1σ). 0 cm = snow-ice interface. Samples  > 30 cm are in ridged areas. The pie charts show the snow-ice interface 18O enrichment.

We can learn from isotopic snow studies in tundra environments to understand the upward contribution of vapor from the snow’s substrate4, despite a substrate’s physical, chemical, and biological characteristics influencing its thermal properties, composition, and interaction with the above snowpack.27 found stable water isotopic enrichment in the basal snowpack directly above the soil substrate, implying an upward mass flux from the soil substrate. A schematic diagram of this isotopic enrichment is shown in Fig. 1a using pie charts to represent the ratio of heavy 18O to light 16O. A pie chart with an equal amount of 16O and 18O represents δ18O = 0‰. It is important to note that phase transitions within the snow lead to fractionation processes in the snowpack’s isotopic composition28,29. Accordingly, TGM can enrich the basal snow in δ18O in the range of 2‰30 to 5‰31, with an equivalent deposition and depletion in δ18O in the layer above31, which then, in turn, goes on to be enriched by further TGM. This effect reduces the isotope variability over time29,32. However, if an isotopically different vapor source (like one from a sea-ice substrate) is diffused through a snowpack, the snow’s isotopic composition is modified4,33,34.

In summary, three factors suggest that sea-ice substrate contributes mass to the overlying snow: (i) the typically large temperature gradients in snow on sea ice, exceeding those in tundra environments, due to a relatively shallow snow depth, (ii) the abundant moisture supply from the ice substrate, and (iii) evidence from tundra studies indicating an upward moisture flux from the underlying soil substrate. However, until now, no mass quantification of sea-ice recrystallization, referred to in this study as recrystallized sea-ice snow, has been attempted for the central Arctic. In this study, we examined snow measurements on sea ice from November 2019 to May 2020 above 83 ∘N during the MOSAiC (Multidisciplinary Drifting Observatory for the Study of Arctic Climate) expedition to quantify the mass contribution of recrystallized sea ice to the Arctic snowpack. To understand and quantify the metamorphic processes of snow on sea ice, it is essential to combine our density and isotope measurements to make a case for this sea-ice-sourced snow. In particular, since the snow heterogeneity complicates the mass flux quantification from density changes, we also attempt to quantify how much the sea ice sublimates up into the overlying pack using an isotopic mixing model.

In this work, we use co-located in-situ measurements of snow temperatures, stable water isotopes, and snow density at specific areas, ~1m2, obtained from excavated snow areas above the sea ice surface. We measured snow temperatures in snowpits using a needle thermometer inserted horizontally into the snow at 3 cm increments (including Ts and Ti). We calculated ΔT using the equation ΔT = (Ti–Ts)/hs. Isotopic snow samples were collected within the snowpits, in vertical profiles of 3 cm vertical resolution, and transported to the research vessel (RV) frozen in sealed pots. In total, 515 snow isotope samples were collected and analyzed. Two sampling methods were used, named SLF and CiASOM, throughout this manuscript. Details can be found in the isotope and salinity sample collection methods section. An overview of all isotopic profiles is shown in Fig. 1b. In addition to the stable water isotopes, the same samples were measured for conductivity to obtain salinity. The density of the snowpack was measured using primarily density cutter measurements, which obtained density by weighing a fixed volume of snow (100 cm3) on an in situ scale with a 6% uncertainty, adapted from ref. 35. In addition, density was also measured using a desktop cone-beam microCT90 (microCT) installed in a laboratory onboard the RV Polarstern. This instrument allowed us to measure additional microstructural properties of the snow, e.g., anisotropy and specific surface area, which are not used in the main body of this study but are outlined in the supplementary information. The microCT samples and density cutter measurements were co-located with isotopic samples through cross-comparison of the height of the measurements. More details can be found in the methods section.

Results

Snow temperature gradients drive mass redistribution

During the MOSAiC expedition, the regional atmospheric winter conditions were normal relative to the preceding four decades, with a few anomalous synoptic events outlined in detail in ref. 36. In summary, an event in December had temperatures at the edge of the 95th percentile, and ranked as the fifth warmest period in the climatology, and a handful of events between late fall and early spring transported warm, moist air into the Arctic36. On the contrary, March had anomalous northerly winds bringing cold polar air into the region. Snow depth measurements revealed a shallower snowpack in this region, relative to other areas in the Arctic Basin1,19. These two conditions influence ΔT and the amount of TGM within the internal snowpack.

The ΔT values for each snowpit are shown in Fig. 2a against the measurement date. The average ΔT after calculating it for each snowpit was 61 K m−1 (standard deviation, σ = 52 K m−1). The snow-ice interface temperature gradient is predominantly determined by the snow’s thickness and thermal properties, dampening atmospheric temperature fluctuations. The snow generally has a stronger impact on the snow-ice interface temperature gradient than the sea ice thickness and the relatively stable ocean temperature. We therefore identify that the significant influence on the sea ice surface sublimation is the snow depth, and it is less significantly impacted by the sea ice thickness. Nonetheless, we agree that thicker sea ice would reduce the temperature gradient across the entire ice profile; however, as we are only interested in the interface, we continue to analyze our data relative to the snow depth. The mean snow depth (hs) was 192 mm (σ = 158 mm) (within the 100–300 mm range obtained from the magnaprobe dataset19), and the average temperature was –17 ∘C (σ = 6 ∘C) from the snowpit temperature dataset. The large standard deviation of ΔT is due to varying hs (seen on the x-axis in Fig. 2b) and late-winter atmospheric warm air intrusions from the mid-latitudes into the high Arctic, causing ΔT < 037. The transition between equilibrium fractionation and kinetic growth in the snow for different vapor pressure gradient thresholds (mbar/m) is visualized in Fig. 2c as outlined in ref. 38. Sixty-seven percent of the measured temperature gradients are in excess of the 5 mbar/m threshold required for kinetic-growth metamorphism to occur.

Fig. 2: Snow temperature gradients (ΔT) measured in snowpits during the MOSAiC expedition between November 2019 and May 2020.
Fig. 2: Snow temperature gradients (ΔT) measured in snowpits during the MOSAiC expedition between November 2019 and May 2020.
Full size image

ΔT is the surface temperature minus the snow-ice interface temperature, so ΔT  > 0 when the sea-ice interface (Ti)  > snow surface temperature (Ts). This figure shows ΔT plotted a against the measurement date, showing that the strongest gradients were at the base of the snowpack and atmospheric temperature at 2 m (T2m) over time (red), b relative to snowpit height, and c relative to the average snowpit temperature as a boxplot with data outliers shown as markers. The dotted lines represent vapor pressure gradient thresholds adapted from ref. 38, showing the transition between equilibrium fractionation and kinetic growth for any measurements above the 5 mbar/m plot.

Quantification of mass redistribution

The upward vapor flux during snowpack TGM typically reduces the density of the lower snowpack and forms depth-hoar structures as a result of upward vapor transport along vapor pressure gradients9. One main driver of diffusion is the divergence of moisture flux caused by either a change in temperature gradient between the sea ice and the snow or a change in porosity between the sea ice and the snow13. identified a strong sublimation layer attached to the sea ice-snow interface with the largest negative change −66 kg m−3.

We assume that all vapor from the underlying sea ice is taken up into the snowpack, and it is unlikely that warmer vapor will pass through the snow without deposition. Especially due to surface energy processes (e.g., longwave cooling) as the vapor reaching the snow surface from below will condense at or just below the cooler surface39. Nonetheless, it is important to note that there is also significant sublimation of the upper snowpack due to the moisture gradient with the atmosphere (dependent on air temperature, specific humidity of the air, surface temperature) and turbulent fluxes (i.e., higher with increased wind or higher roughness length). However, we focus on the mass lost from the snow-ice interface and have not attempted to estimate the mass sublimation from the snow surface.

Figure 3a shows the heterogeneity of the snow density cutter dataset, highlighting the complexity of the snowpack density and its limitations as a straightforward basis for estimating the mass contribution of recrystallized sea-ice snow. We can see by looking horizontally along the bottom of Fig. 3a that there was a systematic density reduction in the lower half of the snowpack after January 2020 (see samples at and below 0.4 on the y axis in Fig. 3a), implying upward vapor flux and mass re-distribution. In addition, mid- and upper-pack densities increased, in part due to wind slab development but perhaps also due to upwards mass deposition from below.40 proposed that these low-permeable wind slabs could prevent convection, and ref. 13 confirmed that convection could occur in low-density snow (i.e., depth hoar), but once a dense layer of snow forms on top (i.e., wind compacted rounded grains), the convection process is halted. Due to the spatial heterogeneity, the analysis of these snow cutter density measurements was limited. Therefore, we compare the empirical relationship from the controlled laboratory study from ref. 24 given in Equation (1).

Fig. 3: A figure linking density changes and isotopic enrichment of snow measured during the MOSAiC expedition.
Fig. 3: A figure linking density changes and isotopic enrichment of snow measured during the MOSAiC expedition.
Full size image

a A heatmap of height-normalized monthly density measured using the density cutter. The values within each cell show the number of measurements. The average monthly density values for all profiles are shown as a red curve, plotted on the secondary/right y axis. b A heatmap of height-normalized monthly averaged isotopic samples of δ18O. The values within each cell show the number of snow samples. The average monthly isotopic values are shown in the overlying gray, also plotted on the right y axis.

The laboratory snow density (201 kg m−3) used by Wiese24 was marginally lower than the average snow density in our study during the MOSAiC expedition (290 ± 34 kg m−3). The 100 K m−1 temperature gradient and average snow temperature from ref. 24 is plotted in Fig. 2c as a horizontal black line, showing the high metamorphism rate of this experiment spanning the 15–20 mbar/m thresholds. This shows that the laboratory conditions are relative to the upper quartile of our measured conditions (93 K m−1). Using the empirical relationship in Equation (1) from ref. 24, and multiplying ITE by the daily vapor pressure gradient ratios, as shown on the y-axis in Fig. 4, we can obtain an approximation for the sea ice sublimation. The ratios are calculated using the August-Roche-Magnus formula41 to obtain the saturated vapor pressure over ice (see Equation 2), where es is in hPa, and T is in degrees Celsius. Calculations of es for each snowpit measurement are then used to obtain the vapor pressure difference at the snow surface and the snow-ice interface individually. The vapor pressure difference is then divided by the distance between the two measurements (i.e., the snow height) to estimate the vapor pressure gradient. The vapor pressure gradient ratio refers to the values relative to the laboratory experiment in ref. 24. We can use this method to estimate the sublimation of the ice surface between 25th October 2019 and 10th May 2020 (198 days). When multiple temperature gradients were measured on one day, an average was taken. If no measurements were taken, the seasonal average was used, indicated using a black cross in Fig. 4. 95 days had measurements, and the daily values were summed to give the cumulative ITE. The remaining 103 days used the seasonal average. This resulted in a cumulative ITE of 13 mm, which can be converted to snow depth equivalent (SDE) by taking the average density of snow and ice as ρs = 290 ± 34 kg m−3 (see methods for details), and ρi = 900 ± 36 kg m−3, respectively. Applying Equation (3)42, SDE = 39 ± 7 mm. The error is obtained from ref. 24, however, we acknowledge the large uncertainty in this method, which we discuss again in the discussion and implications section.

$${e}_{s}(T)=0.61094\exp \left(\frac{17.625T}{243.04+T}\right)$$
(2)
$${{\rm{SDE}}}={{\rm{ITE}}}\times \frac{{\rho }_{i}}{{\rho }_{s}}$$
(3)
Fig. 4: A figure showing the range of vapor pressure gradient ratios of each measured snowpit.
Fig. 4: A figure showing the range of vapor pressure gradient ratios of each measured snowpit.
Full size image

The vapor pressure gradient ratios are calculated relative to the conditions found in Wiese et al. 24, plotted against temperature gradient and average snowpit temperature. These ratios were used to scale the estimated sublimated ice thickness equivalent for our study period. For days with no snowpit measurements, the seasonal average was used, indicated with a black cross.

Stable water isotope gradients in the lower snowpack

In all but four snow isotopic profiles, we found distinguishable signals of a) fresh snow precipitation in the upper layers of the snowpack and b) signals at the bottom of the snowpack that indicate it had undergone TGM. In general, measurements showed the average snow sample at the snow-ice interface was substantially more enriched in δ18O (–11 ‰, σ = 6 ‰) than a surface snow sample in contact with the atmosphere (–33 ‰, σ = 7 ‰) and was closer to the δ18O values of the sea-ice substrate (–4 ‰, σ = 3 ‰ for this study period). In Fig. 1b, a negative covariance (–6 ‰-cm) of the average δ18O snow profile is shown.

We propose that this late-winter signal mainly originates from the sublimation of sea ice and vapor deposition into the snowpack above. The fractionation between 18O and 16O is represented by simple pie charts in Fig. 1a, showing phase transitions where heavy isotopes are less likely to make it into the vapor phase than the light isotopes, and the light isotopes are less likely to transfer into the solid phase. Since the sea ice is isotopically much heavier than snow43, the empirical evidence suggests basal enrichment.

The lower snowpack was stable and unlikely to become completely eroded by wind. However, wind mixed and redistributed the upper snowpack, which caused an averaging effect on the isotopic composition of the surface snow, especially apparent in ridged areas44. The temporal evolution of the isotopic profiles (see Fig. 3b) and a few sampling sites with repeat measurements (see Supplementary Figure 1 from ref. 43) show that despite the snow’s high spatial variability, this vertical gradient is apparent throughout winter. Although no observations of flooding or snow-ice were recorded in the MOSAiC central observatory45, identified flooded areas in local depressions in close proximity to ridged areas, these processes would enrich the lower snowpack whilst simultaneously increasing the density to near-ice values. See the supplementary information on excluding flooding processes for further details.

In the literature, a second-order isotope parameter, deuterium-excess (d-excess = δ2H − 8 × δ18O)25, provides insights into i) the meteoric input defined by the primary moisture source conditions and ii) post-depositional processes within the snowpack46. For example, non-equilibrium fractionation processes during sublimation and re-sublimation can result in d-excess changes47. Our d-excess calculations revealed unexpectedly negative values close to the snow-ice interface, seen in Supplementary Figure 7. Between snow height 0–5 cm, the average d-excess equaled –1 ‰ (σ = 11 ‰), but single values as low as –27 ‰ were also measured. Snow at the surface had an average d-excess of 13.4 ‰ (σ = 5.4 ‰), and the sea-ice substrate had an average d-excess value of 0.8 ‰ (σ = 1.9 ‰43).

We use d-excess alongside conductivity salinity measurements to eliminate the possibility that the enriched isotope values were due to saline snow from processes such as sea spray deposition, brine-wicking (the movement of brine by capillary action), or frost flower inclusion processes. Each marker in Fig. 5b shows a snow sample’s salinity (measured through conductivity) and isotopic d-excess. For this analysis, we separated our snow samples into the substrate types upon which they were collected using the categories: snow on level first-year ice (FYI) and second-year ice (SYI), snow on re-frozen leads, and snow in ridged areas. The distribution of the d-excess values for each ice type category can be seen in Fig. 5a plotted against δ18O. Figure 5b shows how saline snow is typically found on refrozen leads and level FYI and SYI snow areas. However, the difference between these samples can be seen with the isotope d-excess values. Excluding a few outliers, we see that younger, less-metamorphosed snow on refrozen leads typically has positive d-excess values, whereas more metamorphosed snow on level ice has the majority of negative d-excess values. The presence of salty snow with positive d-excess values shows that the above-mentioned processes, which lead to saline snow, typically do not result in decreasing d-excess values43. We hypothesize that there is a greater proportion of recrystallized sea-ice snow on level ice in contrast to the meteoric snow on re-frozen leads, and the negative d-excess values are likely to stem from complex metamorphic recrystallization processes at the snow-ice interface, leading to a disproportionate fractionation of δ2H and δ18O.

Fig. 5: Second-order stable water isotope parameter, deuterium-excess, grouped by ice-type substrate, plotted for all winter snow samples.
Fig. 5: Second-order stable water isotope parameter, deuterium-excess, grouped by ice-type substrate, plotted for all winter snow samples.
Full size image

Deuterium-excess against a δ18O, and b salinity values. The colors in both plots indicate snow collected on level first-year (FYI) and second-year ice (SYI), refrozen leads, and ridged ice areas. Each marker in this figure is one snow sample collected between November 2019 and May 2020. b shows how saline samples on level ice are mainly associated with a negative d-excess signal, in contrast to snow on refrozen leads, where the saline samples have a positive d-excess signal. This indicates that d-excess could be used as a potential proxy for the amount of metamorphic recrystallization in saline snow on sea ice.

Supplementary Figure 5 shows the deviation from the global meteoric water line in the lower samples (shown in red), causing this negative d-excess. In general, this is not a strong deviation and an argument that, in the first approximation, the isotopes can be used for two-component mixing calculations (see section below). Individual samples with very negative d-excess values indicate non-equilibrium kinetic fractionation beyond a two-component mixing model, where care is needed.

Quantifying isotopic enrichment with two-source endmembers

We used a two-source endmember approach of our isotopic measurements to quantify the contribution of recrystallized sea-ice snow. In the sea-ice environment, there is a clear differentiation of snow that has precipitated from the atmosphere (meteoric snow, δ18O between –15 ‰ to –35 ‰48,49,50) and sea ice frozen from the ocean water (with second-year ice (SYI) δ18O = –4.5 ‰ and first-year ice δ18O = –0.7 ‰43). These two origins of snow (the atmosphere and sublimation from the sea ice) could be treated as two source endmembers51. Following the methods outlined in ref. 52, we used a mixing model and the end-member δ18O signatures53,54,55,56 to calculate the fraction of recrystallized sea-ice snow by mass, fs, for each snowpit (Equation 4,53). The distribution of fs values is visualized in Fig. 6, where δi is the average δ18O value for sea ice (–4 ‰, σ = 3 ‰, n = 79 for this study period), δx is each snow profile’s average δ18O, δs is the δ18O value for meteorically precipitated snow, of which we used the surface snowpit snow sample. The two sample sets (CiASOM and SLF) were comparably taken, and we treat them both separately and combined, as shown in the three histograms in Fig. 6. This gives an average fs of 0.33 (33 ± 9%, σ = 14 %) of recrystallized sea-ice snow. The error for CiASOM samples  ± 0.1 ‰ and for the SLF samples  ±0.5 ‰ gives an absolute error of 25% for each calculation of fs; details of this calculation are given in the isotopic overview supplementary information.

$${f}_{s}=\frac{{\delta }_{x}-{\delta }_{s}}{{\delta }_{i}-{\delta }_{s}}$$
(4)

An average snow height (hs) of 192 ± 10 mm (σ = 158 mm, see the methods for details) corresponds to a recrystallized sea-ice SDE of 63 ± 21 mm (σ = 60 mm) (SDE = hsfs) and a SWE of 18 ± 7 mm (σ = 19 mm) (SWE = SDEρs, when ρs = 290 ± 17 kg m−3, σ = 34 kg m−3, using the 6% uncertainty given by Proksch et al.35, and SDE in m). The large range of values for fs in this calculation, and seen in Fig. 6, is due to the large σ of our isotope profiles (77 %), which can be attributed to snow heterogeneity through processes such as wind re-distribution. Values of fs below 0 are where the surface snow is more enriched than the interface snow, typically occurring in thin snowpits reflecting the high variability of snowpack in this region. In summary, isotopes suggest the sea ice contributed to the equivalent of a snow layer ~6 cm thick at the bottom of the snow.

Fig. 6: Histogram of the contribution of recrystallized sea-ice snow in the central Arctic winter.
Fig. 6: Histogram of the contribution of recrystallized sea-ice snow in the central Arctic winter.
Full size image

Quantifying the contribution of recrystallized sea-ice snow to each individual snowpit was calculated using the isotopic two-source endmember approach. fs is the fraction of recrystallized sea-ice snow by mass for each isotopic profile collected in a single snowpit. The dataset has been split to show the two different sets of isotope samples (SLF and CiASOM), and the average recrystallized sea-ice snow contribution for winter is shown in the horizontal dotted lines.

Discussion

Our estimated contributions of recrystallized sea-ice-sourced snow should be used as a starting point for future discussion and improved modeling efforts in this domain.

Based on mass redistribution, a cumulative SDE of 39 ± 7 mm was obtained for the study period. The main uncertainties in this method come from the use of the August-Roche-Magnus approximation for the saturated vapor pressure and the 103 days of unknown temperature gradients in the snowpack, where we use the average winter temperature gradient instead. Future work implementing a physical model to investigate vapor fluxes through the snow under the conditions measured in this region is needed to compare with our estimations and reduce these quantification uncertainties. However, snowpack modeling is beyond the scope of this work.

The second method determines average isotopic enrichment from the underlying sea ice to be on average 63 ± 21 mm using an isotope two-source endmember approach. This second method neglects snow fractionation processes during metamorphism, which enriches the snow in δ18O (in the range of 5 ‰27), causing an overestimation. However, wind homogenization would mix the isotopically depleted meteoric snow, causing an underestimation of the recrystallized sea-ice contribution of an unknown amount.

At the beginning of the season, the snow’s isotopic enrichment at the interface is less apparent and this is likely due to a number of processes (1) the wind redistribution and mixing of the snowpack before it has sintered, this prevents the preservation of atmospheric precipitation events in the snowpack stratigraphy, and (2) spatial and temperature influences in the precipitation’s isotopic signal. This latter can be addressed by building on the work of ref. 37, who investigates the atmospheric water vapor isotopic signal throughout the expedition. We assume precipitated snow follows the isotopic trends of the atmospheric vapor. This atmospheric water vapor was also depleted at the start of the season (likely because it is already quite cold and far from the ice edge) and remains reasonably stable between November 2019 and April 2020, excluding the warm air intrusion events, which bring enriched isotopic vapor. However, these intrusions may cause transient enrichment but are not frequent or sustained enough to explain a systematic enrichment at the snowpack base57 showed that these warm air intrusions do not always align with enriched snowfall. Unfortunately, we did not explicitly collect fresh snow samples for isotopic analysis, and due to large amounts of snow mixing, the fresh snow isotopic signature was hard to disentangle. Finally, the negative d-excess values would not be obtained simply from the accumulation of early-season δ18O-enriched snow, implying that another process must come into play. Persistent negative d-excess values indicate post-depositional modification (e.g., sublimation), which preferentially removes heavy isotopes and further depletes the remaining snow.

The isotopic samples grouped by normalized height in Fig. 3 need to be interpreted with care, as we also have deepening snow over the season. This could be why we do not see a continuously enriching snowpack. Over time, the snowpack becomes more stable and compact, resulting in less mixing of the entire snowpack. As a result, the late-season upper snowpack is likely to be less influenced by the recrystallized sea ice and more by the atmospheric deposition, creating the apparent upper and lower snowpit differentiation of their isotopic signatures. We also see this with the snow stratigraphy, as we have a lower depth hoar layer (likely with more sea-ice sourced snow) and an upper wind slab layer forming over the season (with more atmospheric snow). While it is important to note that the largest uncertainty for snow accumulation on Arctic sea ice is wind redistribution, we show that a high proportion of the snow depth may originate from recrystallized sea-ice snow. This general understanding has implications for snow mass balance estimates, gas flux, light transmission, and heat flux estimates in the Arctic.

Stable water isotopes have proven valuable in showing that there is likely to be substantial recrystallization of the sea ice contributing to the snowpack enrichment; however, calculations of the contribution are uncertain using this method. For this reason, future work using a combined physical and isotopic model is needed to compare with our estimations, similar to the work done by Touzeau et al.32, bridging the vapor diffusion with isotope modeling. However, this is beyond the scope of this work.

To test our calculations, we compare our study to snow TGM at tundra sites. The work by Sturm20 suggests 2–5 mm SWE enter the base of the snow from the soil in 150 days. Less vapor is likely to be provided by a tundra substrate compared to a sea-ice substrate, as vapor from ice is not trapped in between the soil structure. Additionally, as the soil sources moisture, the surface soil desiccates, leading to the ice source being deeper in the ground, which further inhibits upward mass transfer. This supports our larger estimates of recrystallized sea ice snow in the sea-ice environment. This also agrees with Wiese’s24 approximation of ice mass flux being approximately four times larger than Sturm’s20, likely because of the larger ice surface area for sublimation. These study values are provided in the introduction. To conclude, we expect greater availability of vapor from an ice substrate compared to a porous soil substrate.

Our measurement in a shallower snowpack (relative to other areas in the Arctic Basin1,19) with higher ΔT have experienced more metamorphism, increased isotope enrichment in heavy isotopes, and as a result, we are likely to overestimate the SWE contribution. Nonetheless, sea-ice recrystallization is a general process of the sea-ice domain, and any snow on sea ice that is exposed to a positive temperature gradient has the potential to incorporate mass from the sea-ice substrate. Snow on sea ice in Antarctica is expected to undergo the same process under comparable snow and sea-ice conditions; however, the thicker the snow cover, the smaller the SWE contribution. Alternative processes also come into play in Antarctica, for example, in some areas, a persistent snowpack from the previous years is present on the sea ice. Additionally, our study excludes flooding processes (See the supplementary information on excluding flooding processes), however in Antarctica, flooding processes need more consideration due to increased precipitation.

This study does not explore impurity origins in the snowpack, but notes that recrystallized sea ice contributes mass to the snowpack and, as a result, the impurity concentration in the snow may progressively dilute58,59 throughout the winter. These impurities refer to black carbon59, soluble ions (e.g., ammonium;60), and natural minerals (e.g., salts and dust)61. Alternatively, impurities (e.g., salt and halogens) could be made more available to atmospheric transport, as ions with high ice solubility can be incorporated into the ice interior and relocated during recrystallization60,61. TGM and sea-ice-sourced snow will influence snow depth, snow density, specific surface area, grain size and potentially sea-salt distributions in the snowpack; understanding this is critical to improving our understanding of remote seeing backscatter signals with altimetry applications62. Finally, the vertical distribution of snowpack impurities will become increasingly critical in Arctic warming during wind-redistribution and blowing snow events63, where the snowpack acts as a source of atmospheric aerosols64, trace gases65, and other chemical compounds like bromine and molecular iodine66, which contribute to Arctic ozone destruction and gaseous mercury67,68.

Central Arctic climatic predictions point towards substantial regional differences in precipitation69. Regions with thicker snowpacks will have smaller ΔT, resulting in less metamorphism and recrystallized sea-ice snow. This has the potential to reduce the dilution of impurities. Other regions with reductions in snow accumulation would have increased gas fluxes through the snow to the atmosphere, alongside changes in light transmission to the ocean63 and thermal conductivity of the snowpack. This process could have implications for climate modeling, and future work improving the representation of this process could benefit Earth system models. In summary, this improved understanding of the mass contribution from recrystallized sea-ice snow could help disentangle the intricate snow-sea ice system and its trajectories.

Methods

This manuscript studies snow on sea ice in the winter months from November 2019 to May 2020 (therefore excluding melt influences) in the high Arctic between 83.4°N and 88.6°N. During the MOSAiC expedition, the RV Polarstern drifted within the Arctic pack ice, allowing repeat measurements of a single ice floe. In total, three ice floes were studied throughout the whole expedition; however, this study only investigates the first, named Central Observatory 1 (CO1), between October 2019 and May 2020. This ice floe had a diameter of ~1 km. Five aspects of the system were investigated, including the atmosphere, ocean, sea ice (and snow), ecosystem, and biogeochemistry. To describe the evolution of the snowpack conditions over time, snowpit sites were set up and marked with flags on the ice floe, which were designated locations assigned to repeat in situ snow measurements. A time series of the snowpack’s physical (e.g., structural), (isotope) geochemical, and salinity development was constructed by repeatedly visiting the same measurement locations and following the same measurement protocols. The snowpit sites were returned to weekly or bi-weekly unless ice dynamics meant they were inaccessible. The snowpit sites were chosen randomly. However, the measurer ensured snowpit sites included a selection of ice substrate conditions, including level first-year ice and second-year ice areas, leads, and ridges. A snowpit event was one trip to the ice and included a library of different snow measurements measured simultaneously; the events are labeled in the format PS122/"Leg#”_"Week#”-"UniqueID”, e.g., PS122/2_19-9270. The measurements conducted during one snowpit event depended on the time and ice conditions on the given day. In all, 128 snowpits are included in this analysis. More details on the data collection methods, data storage, and format can be found in the following sections, the published datasets70 and the online data paper71.

Density and SSA as a function of height

Snow microstructural measurements were taken within the snowpits alongside the snow height72, isotope and salinity profiles. The density cutter was used to obtain the data used for Fig. 3a. This measurement measures the weight of 100 cm3 of snow in situ on calibrated weighing scales. The density cutter was used to take vertical profiles of snow samples every 3 cm.

To obtain more detailed microstructural measurements, we also used a microCT73. This could measure density at high vertical resolution, specific surface area, and provide details on snowpack anisotropy. We used the microCT data to obtain data for Supplementary Figure 8. Analysis with the microCT requires snow samples to be collected in the field, transported to the Polarstern, and measured using the microCT. The microCT has a resolution of 22 μm. MicroCT profiles are publicly accessible on Pangaea73. The ice and air volumes were separated to obtain profiles of density. SSA was calculated using the triangular-based total surface area of snow per unit mass (with units of m2 kg−1) of sub-volumes within the snow samples. Geometric anisotropy is defined in this manuscript by the relationship outlined in Equation (5), in terms of the correlation lengths ξ in different coordinate directions x, y, z. The correlation lengths were obtained by fitting the decay of the two-point correlation function in different directions to an exponential74. These parameters are used for Supplementary Figure 8.

$${A}_{g}=\frac{2{\xi }_{z}}{{\xi }_{x}+{\xi }_{y}}$$
(5)

Samples of isotope and salinity as a function of height

Isotopic snow samples were collected within the snowpits, in vertical profiles in the field, and transported to the RV Polarstern in sealed pots. In total, 515 snow isotope samples were collected and analyzed later in two laboratories (The Swiss Federal Research Institute WSL and the ISOLAB Facility at the Alfred Wegener Institute Helmholtz Center for Polar and Marine Research in Potsdam, Germany) for the study period between November 2019 and May 2020. Both datasets contribute to this manuscript. The first dataset75, named SLF, took vertical profiles of snow samples (each 100 cm3). The vertical resolution of these samples was every 3 cm. After the samples were transported to RV Polarstern, the samples were melted for salinity measurements (using the YSI 30 Salinity, Conductivity and Temperature sensor,76). These samples were sealed into 20 ml glass vials onboard Polarstern. Each sample was then transported to Switzerland and measured for stable water isotopic composition at WSL, Zurich, using a Los Gatos Research (LGR) Isotopic Water Analyzer (model IWA-45EP) instrument. Each sample was measured five times in this dataset,75 with a resulting measurement accuracy of δ18O ± 0.5 ‰ and δ2H ± 1 ‰. The second dataset, named CiASOM, was collected in the snowpit at three layers: top, middle, and bottom. A plastic shovel was used to sample the snow in these three layers directly into plastic bags. These samples were transported to RV Polarstern and remained frozen onboard as they were stored at −4 ∘C. After the expedition, this second dataset was sent to AWI, Germany, thawed at room temperature, poured into 20 ml glass vials, sealed with parafilm tape, and stored at 4 ∘C. This dataset was measured for stable water isotopes using mass spectrometers (DELTA-S Finnigan MAT, USA) using equilibration techniques to an accuracy of better than  ± 0.1 ‰ for δ18O and  ± 0.8 ‰ for δ2H. Every sample is equilibrated and then transferred to the Isotope Ratio Mass Spectrometry, where it is measured 10 times against a standard of known isotope composition. The statistics of one single datapoint are based on these ten measurements. Both datasets are published with open access in Pangaea, where more details can be found.

Sea ice samples were collected during ice coring events using a Kovacs Mark II 9 cm diameter corer from ice of differing ages. The snow on top of the sea ice was brushed off the top of the cores to minimize snow affecting the ice surface. More details on the sampling and measurement methods are outlined in Section 2.2.2. in ref. 43.

A comparison was conducted between the two laboratories by measuring fifty samples twice to show continuity between the datasets. The fifty selected samples primarily originated from snow measurements on level ice with a range of isotopic signatures (see Supplementary Figure 2), but included a selection of samples with negative d-excess values. This negative d-excess was consistently observed in both datasets. It was found that the WSL-measured samples needed a correction due to evaporative fractionation during sample storage. The correction was done by calculating the mean of this dataset relative to the mean of the parallel dataset from AWI and correcting for the difference. This was possible as the histograms for both datasets were of a similar shape but appeared to have a shift, see Supplementary Figure 3. As a result, the δ18O was corrected by –6.4 ‰, the δ2H was corrected by –36.4 ‰, visualized in Supplementary Figure 4. Despite this correction factor, the enrichment signal analyzed in this study was apparent in both datasets, however, as mentioned in the main text, d-excess calculations need interpreting with care for the SLF dataset, as done in this study.