RV Polarstern expedition PS 106 (June–July 2017) in the early melting season gave the opportunity to systematically study the occurrence of cryogenic gypsum release and the morphological properties of gypsum crystals in the area north of Svalbard and on the Barents Sea shelf (Fig. 1a; Table 1).

Cryogenic gypsum was sampled from the upper 10 m of the under-ice water at four stations distributed throughout the expedition area (Fig. 1a; Table 1). The first part of the expedition (PS106/1) consisted of a drift study north of Svalbard, during which the vessel was anchored to an ice floe (station 32). This ice floe was revisited 6 weeks later at the end of the expedition (PS106/2) (station 80). During the second part of the expedition (PS106/2), cryogenic gypsum was collected over the western Barents Sea (station 45) and in the Nansen Basin to the north-east of Svalbard (station 66).

Gypsum crystals were sampled with a plankton net mounted on a remotely operated vehicle (ROVnet, Fig. S1). The ROVnet consists of a polycarbonate frame with an opening of 40 cm by 60 cm, to which a zooplankton net with a mesh size of 500 µm was attached (Flores et al., 2018). For gypsum sampling, a handmade nylon net with an opening of 10 cm by 15 cm and a mesh size of 30 µm was mounted in the zooplankton net opening. The concentrated particulate material of the small nylon net was collected in a 2 L polyethylene bottle attached to the cod end of the net. A gauze-covered window in the cod-end bottle allowed seawater to drain out. Both nets were mounted on the aft end of a M500 (Ocean Modules, Sweden) observation class ROV carrying an extensive sensor suite described in Katlein et al. (2017). After each ROVnet deployment, the nets were rinsed with ambient seawater to concentrate the sample in the cod end of the net. The ROVnet sampled horizontal profiles in the water directly below the sea ice. Standard ROVnet profiles were conducted at the ice–water interface at 5 m and 10 m depths. The distance covered by each profile ranged between 300 and 600 m. At station 32, the 10 m profile was aborted due to technical failure. At station 80 no 5 m profile was sampled due to time constraints, and the sub-surface sample was discarded due to handling failure (Table 1).

The concentrated particulate material collected in the cod-end bottle of the gypsum sampling net was mixed with a sample-equivalent volume of 98 % ethanol and stored at 4 ∘C until further analyses (Wollenburg et al., 2018a).

At ROVnet sampling stations, ice thickness was estimated through thickness drill holes with a tape measure. To characterize the properties of the ice floes sampled on the floe-wide scale, ice thickness surveys were conducted at each sampling station with a GEM-2 (Geophex) electromagnetic induction ice-thickness sensor (Katlein et al., 2018).

In the home laboratory the samples were rinsed onto a 32 µm mesh using fresh water. The samples were then oven-dried at 50 ∘C for 20 h. The remaining crystals were transferred into pre-weighed micropalaeontological slides, and their weight was determined with a high-precision Sartorius SE2 ultra-microbalance. Under a Zeiss Axio Zoom V16 microscope, pictures were taken with an Axiocam 506 colour camera. We made both overview images of the whole sample and detailed images of individual crystals. From all samples and crystal morphologies, individual crystals were analysed using Raman microscopy, which confirmed that the crystals were gypsum (Wollenburg et al., 2018a). As in some samples both very large and very small crystals (Figs. S3–S4) were observed; the > 32 µm samples were dry-sieved over a 63 µm analysis sieve. The length and width of the cryogenic gypsum crystals in the size fractions >32<63 and > 63 µm were determined with the software application ImageJ for 50 crystals in each sample and size fraction (Schneider et al., 2012) (Table 2).

At all ice stations, sea ice cores for archive purposes and for further measurement of bottom communities were drilled with a 9 cm diameter ice corer (Kovacs Enterprise) and stored at -20 ∘C (Peeken et al., 2018b). One ice core from station 80 and four bottom slices (10 cm) of ice cores from station 45 were studied to investigate the gypsum crystal morphologies within sea ice. Each section was transferred into a measuring jug with lukewarm tap water for approximately 2 s, and then the jug was emptied over a 32 µm analysis sieve and repeatedly refilled. This process was continued until all ice was melted. With the aid of a hand shower and a wash bottle the residue on the sieve was rinsed and transferred into a 30 µm mesh-covered funnel, dried and transferred into a micropalaeontological picking tray for inspection and documentation. For storage, the residue was transferred onto pre-weighed, labelled micropalaeontological slides.

The aim of our dissolution experiments was to investigate the persistence of gypsum crystals against dissolution in the Arctic water column (water mass trials) and under common biological sample treatment (formaldehyde trial).

Dissolution experiments were carried out on individual gypsum crystals collected from ROVnet samples. Hereby, five cryogenic gypsum crystals with different crystal morphologies and from both size fractions were used in each reaction chamber. Before the start and after the termination of each experiment, pictures of the cryogenic gypsum crystals used were taken with an Axiocam 506 colour camera under a Zeiss Axio Zoom V16 microscope. The weight of the crystals before and after each treatment was determined with a high-precision Sartorius SE2 ultra-microbalance after they had been transferred into a pre-weighed silver boat. The experimental running time of each experiment was 24 h.

The size-specific sinking velocity of cryogenic gypsum was measured in a settling cylinder (Ploug et al., 2008). The cylinder (30 cm high and 5 cm in diameter) was filled with filtered seawater (salinity 32) and surrounded by a water jacket for thermal stabilization at 2 ∘C. The settling cylinder was closed at both ends, only allowing the insertion of a wide-bore pipette at the top. Immediately before measurement, the gypsum was submerged into seawater with a salinity of 32 and a temperature of 2 ∘C, and then transferred to the settling cylinder with a wide-bore pipette. The gypsum crystals were allowed to sink out of the wide-bore pipette, which was centred in the cylinder. The descent of the crystals was recorded by a Basler 4 MP Ethernet camera equipped with a 25 mm fixed focal lens (Edmund Optics). The settling column was illuminated from the sides by a custom-made LED light source. The camera recorded seven images per second as the gypsum crystals sank through the settling column. The measurements were only done with one camera, so a two-dimensional view. We measured over a distance of ∼ 5 cm after the crystals had reached terminal settling velocity and at stable and constant temperature and salinity. The technical uncertainties of the setup were smaller than the uncertainties between two similar-sized gypsum crystals, which had up to 1000 m d-1 uncertainties (see Fig. 6, with equivalent spherical diameters of ∼ 1 mm). The setup was calibrated by recording a length scale before sinking velocity measurements. The size and settling of the individual gypsum crystals was determined with the image analysis software ImageJ. This was done by using the projected area of the crystals to calculate the equivalent spherical diameter and the distance travelled between the subsequent images to determine the sinking velocity of the individual crystals (Iversen et al., 2010)

To determine sea ice drift trajectories of sampled sea ice, we used a Lagrangian approach (ICETrack) that traces sea ice backward or forward in time using a combination of satellite-derived, low-resolution drift products. So far, ICETrack has been used in a number of publications to examine sea ice sources, pathways, thickness changes and atmospheric processes acting on the ice cover (Damm et al., 2018; Krumpen et al., 2016; Peeken et al., 2018a). A detailed description is provided in Krumpen et al. (2019).

Sea ice motion information was provided by different institutions and obtained from different sensors and for different time intervals. In this study we applied a combination of three different products: (i) motion estimates based on a combination of scatterometer and radiometer data provided by the Center for Satellite Exploitation and Research (CERSAT; Girard-Ardhuin and Ezraty, 2012); (ii) the OSI-405-c motion product from the Ocean and Sea Ice Satellite Application Facility (OSI SAF; Lavergne, 2016); and (iii) polar pathfinder daily motion vectors from the National Snow and Ice Data Center (NSIDC; Tschudi et al., 2016).

The tracking approach works as follows: an ice parcel is traced backward or forward in time on a daily basis. Tracking is stopped if (a) ice hits the coastline or fast ice edge or (b) ice concentration at a specific location drops below 50 %, at which point we assume the ice to be formed or melted. The applied sea ice concentration product was provided by CERSAT and was based on 85 GHz SSM/I brightness temperatures, using the ARTIST Sea Ice (ASI) algorithm.

Based on backtracking (Krumpen, 2018) and sea ice observations, the sampled ice floes had an age of 1 to 3 years (Fig. 1b), were originating in the Siberian Sea (station 32/80) and the Laptev Sea (station 45), and were more locally grown in the Nansen Basin (station 66). Whereas the mean sea ice thickness at the ROV survey stations ranged between 94 and 156 cm, the mean sea ice thickness of the investigated ice floes, estimated by ice-thickness sensor surveys (Katlein et al., 2018), was 1.90 m for station 32, 1.00 m for station 45 and 1.80 m for stations 66 and 80 (Fig. 1a, Table 1). Despite the different origins and thicknesses of sea ice, cryogenic gypsum crystals were found at all stations and at all depth layers sampled with the ROVnet (Fig. 1a, b, Table 1). At all stations and sampling depths, the samples were dominated by cryogenic gypsum with a proportional dry weight of > 96.5 % in the 5 m sample at station 32 and of > 99 % in all other samples (Figs. 2, S2–S5). Other lithogenic particles, as are often found in sea ice (Nürnberg et al., 1994), were essentially absent.

The samples collected at station 32 were dominated by solid, rounded, matt cryogenic gypsum crystals with a mean length–width ratio of 1.40–1.76 (Tables 2, S2). The proportional mass contribution of the smaller-sized crystals of the >30<63 µm size fraction increased with depth and outweighed the contribution of the > 63 µm size fraction with 56.30 % and 66.28 % for the 0 and 5 m water depth samples, respectively (Fig. 3). At 0 m, the mean length of the crystals was 68.46 µm in the > 63 µm size fraction and 50.64 µm in the >30<63 µm fraction. At 5 m depth, crystal dimensions were similar with mean crystal lengths ranging from 63.28 µm in the >63 µm fraction to 49.91 µm in the >30<63 µm size fraction.

At station 45, the crystals were mostly solid and for the most part hyaline rather than matt crystals as at station 32 (Figs. 2c, d, 6, S3). With decreasing weight proportion, the > 63 µm size clearly dominated the 0, 5 and 10 m samples with 79.90 %, 73.39 % and 66.14 %, respectively. In the 0 m layer samples, mean crystal lengths were 114.18 µm in the > 63 µm size fraction and 58.74 µm in the >30<63 µm size fraction (Table 2). At 5 m depth, we observed mean crystal lengths of 111 µm in the > 63 µm size fraction and 56.73 µm in the >30<63 µm fraction. The mean crystal lengths in the 10 m samples were 92.83 and 50.32 µm for the > 63 and >30<63 µm size fractions, respectively. At station 45 the crystal length–width ratio varied between 1.37 and 1.98, measured in the >30<63 µm size fraction of the surface sample and in the > 63 µm size fraction of the 10 m sample. The cryogenic gypsum crystals retrieved from the melted ice core drilled at this station were solid and hyaline. In size and shape they resembled the crystals of the 10 m layer at this station with a mean crystal length of 114.2 µm, mean width of 57.2 µm and a length–width ratio of 2 (Fig. 4).

At station 66, the crystals from 0 m water depth were dominated by large, solid, pencil-like, hyaline crystals with a mean crystal length of 1355 µm and mean width of 415 µm in the dominating > 63 µm fraction (99.25 % mass) (Figs. 2b, S4, Table 2). These crystals with an average length–width ratio of 3.27 were found as isolated crystals but very often also as intergrown crystal rosettes with 2 to more than 10 individual crystals involved (Fig. S4, Table 2). The >30<63 µm size fraction (0.75 % mass) was dominated by rounded, whitish, matt gypsum particles and tiny gypsum needles with a mean crystal length of 56.67 µm (Fig. S4, Table 2). As at the other stations, the weight proportion of the > 63 µm size fraction significantly decreased from 99.25 in the 0 m, to 75.23 at 5 m and to 61.18 % in the 10 m sample (Fig. 2). The size of cryogenic gypsum crystals collected from the 5 and 10 m layers was significantly smaller and predominantly composed of isolated small hyaline and euhedral gypsum needles. The length–width ratio ranged between 5.60 (5 m) and 4.37 (10 m) (Figs. 2a, S4, Table 2). In the 5 m layer sample, the mean crystal length was 411.42 µm in the > 63 µm size fraction and 62.03 µm in the >30<63 µm size fraction. The 10 m samples showed a mean crystal length of 101.40 µm in the > 63 and 30.71 µm in the >30<63 µm size fraction (Table 2).

In the 10 m layer sample of station 80, large tabular gypsum crystals measuring up to 1 cm in length (mean length: 3078 µm, mean width: 1830 µm) dominated the > 63 µm size fraction. Their average length–width ratio was 1.7. This size fraction contributed 89.1 % of the gypsum mass (Figs. 5, S5, Table 2). The >30<63 µm size fraction was composed of fragments of these large crystals and a few small gypsum needles. These often intergrown columnar crystals looked bladed and for the most part also dented and with numerous cracks. Their mean length was 71.8 µm. The ice core retrieved from this station was very porous and broke into pieces of 9 to 11 cm. Cryogenic gypsum was retrieved from all these ice core sections and revealed a dominance of extraordinary large crystals (Figs. 5, S5), which resembled the ROVnet samples from this station. The largest cryogenic gypsum crystals > 6000 µm (mean crystal length: 2821 µm, mean width: 1689 µm) were retrieved from the topmost 8 cm of the ice core section, whereas the maximum crystal size gradually decreased downcore (Fig. S5). The crystals themselves lacked sharp corners, and the large crystals had cavities inside, indicating an advanced stage of dissolution (Figs. 5c, d, S5).

The sinking velocity (SV) of the gypsum crystals increased with crystal size (Fig. 7). Small crystals with an equivalent spherical diameter (ESD) of 200 µm sank with 300 m d-1 velocities, while large gypsum crystals with ESDs of 2000 to 2500 µm sank with velocities of 5000 to 7000 m d-1. The size to settling relationship was best described by a power function (SV = 4239.9 ESD0.839, R2=0.84).

This study shows for the first time the widespread presence of cryogenic gypsum under melting Arctic sea ice of different origins. At all stations, cryogenic gypsum dominated the sample fraction of particles > 30 µm in Eurasian Basin surface waters, indicating a continuous cryogenic gypsum flux from melting sea ice over a period of 6 weeks.

When designing the ROVnet for cryogenic gypsum sampling, we opted for the coarser > 30 µm mesh to prohibit an overflow of the sampling container when running into a phytoplankton bloom. However, as Geilfus et al. (2013) have observed gypsum crystals as small as 10 µm, we probably lost an unknown proportion of smaller gypsum crystals by the chosen sampling strategy. The gypsum crystals described from sea ice so far retrieved from only 3-day-old experimental and 30 cm thick natural sea ice off Greenland were small (crystal length max. 100 µm). They are planar idiomorphic gypsum crystals often intergrown at large angles or occurring as rosettes (Geilfus et al., 2013). Similar but larger (crystal length up to 1 mm) gypsum crystals were observed within Phaeocystis aggregates collected in the region of the present study (Wollenburg et al., 2018a). However, here we show that gypsum crystals exhibit a strong variability in size and morphology. Particularly large crystals were characterized by more complex shapes (Figs. 2, 5, S3–4) and increased surface roughness (Fig. S8c, d), compared to the small planar euhedral (Fig. 2a) and more spherical crystals (Fig. S8a, b). Euhedral crystal needles larger but otherwise similar to those described by Geilfus et al. (2013) and Wollenburg et al. (2018a) dominated the > 63 µm fraction collected at 5 and 10 m depths at station 66, and smaller crystals contributed especially to the >30<63 µm size fraction of the station's sub-surface samples.

As cryogenic gypsum forms in sea ice brine pockets or channels, the size and morphology especially of large crystals is likely determined by sea ice texture and porosity during gypsum precipitation. Pursuing this hypothesis, the large and intergrown crystals collected from the 0 m layer at station 66 and the 10 m layer and ice core at station 80 formed in highly branched granular sea ice (Lieb-Lappen et al., 2017; Weissenberger et al., 1992). In contrast, the small cryogenic gypsum needles reported by Geilfus et al. (2013) and Wollenburg et al. (2018a) may have preferentially formed in columnar sea ice. Even sampling the same ice floe (station 32 and 80), the appearance of the crystals changed. Possibly, a widening of the brine channels during the elapsed time (6 weeks) allowed a release of larger crystals at station 80 when compared to station 32. However, crystal growth during this elapsed period or lateral advection of large crystals cannot be excluded. Thus, detailed texture analyses of sea ice cores prior to sampling are needed to validate or reject hypotheses on a link between sea ice porosity and cryogenic gypsum crystal size and morphology, which should be considered in future studies.

The sea ice microstructure dictating the formation of gypsum crystals in the brine matrix likely varied among ice floes due to different ages, origins and drift trajectories (Fig. 1b). For example, station 66 was the only station where the sea ice likely formed over the central Nansen Basin only months before our study (Fig. 1b). The surface sample of station 66 had large star-shaped intergrown hyaline gypsum crystals that were observed at no other station. They also showed a considerably higher length–width ratio than crystals from second-year ice of stations 32/80 and 45 (Figs. 1b and 2). Accordingly, a close relationship between local sea ice properties and gypsum crystal morphology in the underlying water was evident from the comparison of gypsum crystals collected with the ROVnet with those retrieved from ice cores collected at two stations. The ice core samples revealed cryogenic gypsum crystals that basically resembled the crystal morphologies collected from the water column at the same stations, indicating that the gypsum morphologies observed in the water column likely reflect the gypsum precipitation conditions and brine-channel structure of local ice floes. The current understanding of mineral precipitation in supersaturated brine relies on ice core analyses, sea ice brine, experimental studies and the mathematical modelling of the temperature window in which each mineral is likely to form (Butler et al., 2017; Marion et al., 2010). There are still many uncertainties regarding the precipitation and dissolution of gypsum within natural sea ice and during ice core storage. Although the FREZCHEM model and Gitterman pathway predict gypsum precipitation under defined conditions, only Geilfus et al. (2013) and Butler et al. (2017) succeeded in retrieving gypsum under such conditions, whereas others failed (Butler and Kennedy, 2015). According to the FREZCHEM model, cryogenic gypsum precipitates at temperatures of -6.2 to -8.5 ∘C and at temperatures <-18 ∘C (Geilfus et al., 2013; Wollenburg et al., 2018a). Accordingly, a storage temperature of -20 ∘C would allow the post-coring precipitation of gypsum from contained brine. However, in field and experimental studies, cryogenic gypsum was so far only observed to precipitate in the -6.2 to -8.5 ∘C temperature window, even when treatments were conducted below -20 ∘C (Butler et al., 2017; Geilfus et al., 2013). Furthermore, the observed signs of dissolution on the large cryogenic gypsum crystals from the ice core when compared to the sharp-edged crystals retrieved from the water column at station 80 indicate that significant new precipitation of gypsum during storage did not occur but rather the opposite.

Apart from the growing conditions of gypsum crystals within sea ice, the size spectrum of crystals retrieved from different depths in the water column was likely essentially altered by the size-dependent sinking velocity of the crystals. Because the sinking velocity of large cryogenic gypsum crystals is high, the chance to catch large crystals with horizontal transects directly under the ice should be lower compared to small crystals (Fig. 7a). Accordingly, significant amounts of large cryogenic gypsum crystals were mainly sampled from the 0 m layer where they could be scraped off the underside of the ice (see station 66, Table 2). In contrast, smaller cryogenic gypsum crystals sink at lower velocities (Fig. 7a). Hence, the large quantity of small-sized crystals retrieved in the deeper layers of station 66 and all layers of station 32 and 45 were likely influenced by the accumulated gypsum release in this size fraction, whereas the rarer large crystals indicated the momentary release at these stations. The extremely large crystals sampled at station 80 at 10 m depth probably indicated an ongoing flux event during rapid melting. According to our dissolution experiments, gypsum dissolution within Arctic surface waters should only have a minor impact on the size distribution of cryogenic gypsum crystals within the surface water. Besides vertical flux, the advection of gypsum crystals with surface currents may also have influenced the size distribution of gypsum crystals sampled in the water column.

The small temperature range of the -6.2 to -8.5 ∘C window, which is also the only gypsum precipitation temperature spectrum applicable in the Arctic Ocean, has been considered one reason why gypsum was not detected in other studies (Butler and Kennedy, 2015; Wollenburg et al., 2018a). Furthermore, the kinetics of gypsum precipitation was considered too slow for detection during experimental studies, and the amount of gypsum was considered hard to verify versus other sea ice precipitates that are quantitatively much more abundant, leading the focus towards other sea ice precipitates (Butler and Kennedy, 2015; Geilfus et al., 2013). Although cryogenic mirabilite and hydrohalite are 3 and 22 times more abundant than gypsum, respectively (Butler and Kennedy, 2015), gypsum is the only sea ice precipitate that survives for one to several days within the Arctic water column. Cryogenic gypsum dissolution increases with increasing hydrostatic pressure and increasing temperatures (Fig. 6). However, well-preserved cryogenic gypsum crystals were retrieved from algae aggregates collected from 2146 m water depth, suggesting that either the transport from the surface to this depth was very rapid or that dissolution was decreased and/or prevented once gypsum crystals were included within the matrix of organosulfur compound-rich aggregates (Wollenburg et al., 2018a). Yet, as seawater is usually undersaturated with respect to gypsum (Briskin and Schreiber, 1978) and as shown by our dissolution experiments, the disaggregation of organic aggregates would expose the gypsum to the seawater and dissolve any crystals, suggesting that the gypsum crystals sank rapidly to the seafloor within the organic aggregates. The same dissolution would occur within the sampling cups of sediment traps, explaining why gypsum has not been observed in those types of samples.

Our dissolution experiments showed that cryogenic gypsum can persist long enough in the cold polar surface water to be collected in measurable concentrations. The missing evidence of gypsum from past studies was likely due to the quick dissolution of gypsum crystals at higher temperatures and the pressure dependence of dissolution kinetics, impeding the discovery of gypsum in sediment trap samples and on the seafloor. In addition, formaldehyde preservation leads to the immediate dissolution of gypsum, destroying any evidence of cryogenic gypsum in all biological samples that are fixed with formaldehyde, including water column and net samples.

Based on our experience with the PS106 expedition samples and the experiments presented here, we propose a standardized procedure for gypsum sampling in the field. This procedure is part of the standard operating protocol for gypsum sampling on the MOSAiC expedition (Fig. S9).

We found less than 6 % dissolution of individual crystals in polar surface water per day. Thus, at depths immediately below the fluorescence maximum where a significant part of organic aggregates are formed (Iversen et al., 2010), the gypsum scavenging and ballasting of aggregates (Turner, 2015) is little affected by gypsum dissolution (Olli et al., 2007) (Fig. 6, Table 3). The incorporation of dense minerals into settling organic aggregates will increase their density and, therefore, the size-specific sinking velocities of the aggregates (Iversen and Ploug, 2010; Iversen and Robert, 2015; van der Jagt et al., 2018). The high sinking velocity of large gypsum crystals > 1 mm (5000–7000 m d-1; Fig. 7a) could create strong hydrodynamic shear that might cause disaggregation of fragile algae aggregates (Olli et al., 2007). However, smaller gypsum crystals have been observed inside Phaeocystis aggregates collected at depths below 2000 m (Wollenburg et al., 2018a). This shows that cryogenic gypsum is incorporated into organic aggregates and supports the hypothesis that gypsum can be an important ballast mineral of organic aggregates.

As chlorophyll concentrations in the surface water were mostly low (< 1 mg m-3; Hauke Flores, unpublished data), a massive gypsum-mediated export of phytoplankton was unlikely during expedition PS106. However, especially at the ice floe of station 32/80, we observed a high coverage of the ice underside by the filamentous algae Melosira arctica, and gypsum crystals were found in M. arctica filaments collected nearby (Fig. 8), as well as at station 45 (Fig. 2d). This indicates a potential for rapid M. arctica sedimentation mediated by cryogenic gypsum, as soon as the algal filaments were released from the melting sea ice. Hence, ballasting by cryogenic gypsum may also have contributed to the mass export of Melosira arctica aggregates observed in 2012 (Boetius et al., 2013).