Sampling took place in four different spots within the 3×1 m accumulation of CCMs in July 2019. Four samples (KC19CCC-1, KC19CCC-2, KC19CCC-3, and KC19CCC-4) were picked up with a knife, wrapped in aluminium foil, and stored in plastic containers. There were no obvious visual differences between the samples during collection. During sampling, disturbance to the overall appearance of the CCM accumulation and Cove Cave was kept to a minimum.

In the lab, the samples were homogenised and transferred to glass vials in a laminar flow hood. For all subsequent analyses, macroscopic contaminants (e.g. pieces of bedrock or insect remains) were removed under a binocular microscope. To aid the calculation of the DCF, four artificial mixtures ranging from 100 wt % CCM fraction to 100 wt % non-CCM fraction were produced from KC19CCC-4 by splitting the sample into the different mineral fractions and mixing the relative amounts. Overall, analysis options and the reproducibility of analyses were limited by the available sample amount.

A Bruker D8 DISCOVER X-ray diffractometer (XRD), using Bragg–Brentano geometry equipped with a Cu target and a LYNXEYE detector, was used to analyse the mineralogical composition of the samples (Cu–Kα radiation 1.5406 Å; 2θ range 8–45∘). Furthermore, a HORIBA Jobin Yvon LabRam HR800 spectrometer, excited by a frequency-doubled Nd:YAG laser (100 mW, 532 nm), was used for the in situ determination of the mineralogy of crystals and crystal aggregates at a resolution of ca. 5 µm.

A KEYENCE VHX-6000 digital microscope was used to examine the morphology of the CCMs. To study the fine fraction, a field-emission scanning electron microscope (SEM) operating at 10 kV accelerating voltage was utilised (DSM 982 GEMINI, ZEISS Group).

In total, 14 aliquots of 0.15 to 0.7 mg of the carbonate fraction were taken from the four samples, and their carbon and oxygen stable isotopic composition was analysed using a Thermo Fisher DELTA V Plus isotope ratio mass spectrometer coupled with a GasBench II (Spötl and Vennemann, 2003), yielding a long-term precision of ±0.08 ‰ for δ18O and ±0.06 ‰ (1σ) for δ13C (Spötl, 2011). For comparison, carbon and oxygen isotope data from inactive common speleothems in the study area (a few square kilometres) were included. The common speleothems were collected in several caves, including Cove Cave. All results were calibrated against international standards and reported relative to the Vienna Pee Dee Belemnite (VPDB) standard.

All four samples show a complex mineralogy and a range of particle morphologies that consist of crystal aggregates and single crystals of variable size. The aggregates range from 50 to 400 µm in diameter, rarely exceeding 500 µm, while individual crystals range from 1 µm to a few tens of micrometres. Brownish submicrometric crystals dominate sample KC19CCC-1, which contains larger translucent and opaque white and brownish crystals, as well as crystal aggregates (Fig. 3a). Sample KC19CCC-2 (Fig. 3b) is homogeneously white, with opaque and translucent crystals/crystal aggregates. The other two samples (KC19CCC-3 and KC19CCC-4; Fig. 3c) are made up of larger white and brownish crystals/crystal aggregates and brownish submicrometric crystals, resulting in a speckled appearance.

XRD results (see Appendix A) show that the samples are mostly mixtures of calcite and the sulfate minerals gypsum (CaSO4⚫2H2O), eugsterite (Na4Ca(SO4)3⚫2H2O), löweite (Na12Mg7(SO4)13⚫15H2O), and mirabilite (Na2SO4⚫10H2O). Sample KC19CCC-1 contains additional phases such as quartz, dolomite, and potassium feldspar, while quartz is also present in samples KC19CCC-3 and KC19CCC-4.

The CCMs contain a variety of crystal morphologies, comparable to the morphologies of CCCcoarse (Žák et al., 2018), although having much smaller sizes. Predominant morphologies are spherulitic and acicular and also aggregates of elongated prismatic crystals. Furthermore, rhombic and dumbbell-shaped morphologies were observed. Optical microscopy combined with micro-Raman spectroscopy showed that calcite is mostly represented as translucent and opaque spherulitic crystals and crystal aggregates, such as chains of different colours, whereas sulfate minerals mostly consist of white fibrous, acicular, and prismatic crystals. Raman spectroscopy and SEM imagery also showed that microcrystalline calcite and sulfate minerals are often intermixed (Fig. 3d).

The stable isotopic composition of the 14 aliquots ranges from -24.0 ‰ to -16.0 ‰ for δ18O and from 7.0 ‰ to 11.4 ‰ for δ13C (Fig. 4). The δ13C values overlap with those of CCCfine from mid-latitude caves (Žák et al., 2018). The δ18O values, however, are significantly lower than values from mid-latitude CCCfine and fall at the lower end of δ18O values of mid-latitude CCCcoarse (Žák et al., 2018).

Of the four 230Th / U analyses, three yielded results (KC19CCC-2, KC19CCC-3, and KC19CCC-4), while KC19CCC-1 contained too much detrital material to be analysed. Table 1 shows that the 238U content is high (1113±2 to 1278±2 ng g-1), as is the concentration of detrital thorium (indicated by the low 230Th/232Th activity ratio between 1.30±0.04 and 3.89±0.22). Uncorrected ages show a large range from 143±11 to 7335±48 a BP. However, the appearance of the sampling site and the presence of the same types of minerals in all samples suggest that they likely formed near-synchronously. The results yielded a highly correlated isochron (R2=0.998), yet the maximum likelihood regression is heavily controlled by the sample with the highest analytical precision, which is the cleanest sample (KC19CCC-2). The high mean squared weighted deviation (MSWD) of 69 indicates that the ages are overdispersed compared to the stated analytical uncertainties and that an isochron accounting for this overdispersion is needed (Vermeesch, 2018). The resulting age of 65±17 a BP agrees with the ages of two of the three samples (KC19CCC-2 is 64±12 a BP; KC19CCC-3 is 172±153 a BP) when corrected for detrital Th, using the isochron-derived initial 230Th/232Th activity of 1.39±0.006 (Fig. 5). The isochron-corrected age of KC19CCC-4 (353±74 a BP) is, however, not in agreement with the other two samples within dating uncertainty, thus suggesting multiple sources of initial 230Th (e.g. Dorale et al., 2004). Nonetheless, this disagreement could also result from an underestimation of the uncertainty by the maximum likelihood regression. The isochron-corrected δ234Uinitial is high for all three samples (1872±3 to 2193±5).

Conventional radiocarbon ages for the four untreated and four mixed aliquots range from 140±27 to 2261±74 a BP, with corresponding F14C values of 0.98±0.003 to 0.06±0.0006 (Table 2). When applying the radiocarbon method to speleothems, the DCF needs to be determined (Genty and Massault, 1997) in order to correct for 14C-free material from bedrock and/or soil (Bajo et al., 2017; Hajdas et al., 2021; Hua et al., 2012). The DCFs of all analysed aliquots range from 1.24±0.50 % to 93.67±0.06 % (1σ uncertainty). For the untreated aliquots, 96.2 % of variance in radiocarbon age can be explained by the DCF (R2=0.96). When artificial mixtures are included, this value is lower (R2=0.88), yet it indicates that the artificial mixing was reasonably accurate. The radiocarbon age extrapolated to 0 % dead carbon by linear approximation is 2±28 BP. Calibration using IntCal20 (Reimer et al., 2020) yielded three distinct peaks of calibrated ages, namely 36–73, 112–139, and 226–255 cal BP, with respective probabilities of 37.8 %, 28.3 %, and 29.4 % (Fig. 6). The peak with the highest probability (36–73 BP) is in agreement with the 230Th / U isochron age of 65±17 a BP and provides the most likely timing for CCM formation. We attribute the other two peaks to a plateau of the calibration curve in this time interval.

The dating results show that the CCMs formed during the late Little Ice Age, a period of relatively cold climate conditions and glacier advances, which in Greenland lasted between about 700 and 50 a BP (i.e. 1250–1900 CE; Kjær et al., 2022).

In carbonate-hosted caves, CCCfine typically show a simple mineralogical composition, mostly consisting of calcite or, less commonly, other carbonate minerals, such as aragonite, or mixtures thereof (Žák et al., 2018). In some caves, the occurrence of non-carbonate minerals of cryogenic origin has been reported (Dublyansky et al., 2017; Žák et al., 2018). Cryogenic gypsum and, locally, other sulfate minerals are mostly known from caves hosted in gypsum rock (Kadebskaya and Tschaikovskiy, 2015; Žák et al., 2018) but also from some limestone caves, where the sulfate is derived from pyrite oxidation (Bartolomé et al., 2023; Dublyansky et al., 2017). Mirabilite is known from several caves, where its existence has been tied to evaporative conditions (e.g. Audra and Nobécourt, 2013; Bieniok et al., 2011). Although it has been reported from cold caves (Harmon et al., 1983), and its formation has been tied to preceding cryochemical processes, mirabilite was not regarded as primarily cryogenic in origin (Žák et al., 2018), and we are unaware of reports about cryogenically formed mirabilite in caves. Eugsterite is a rare sulfate mineral which has only been reported from two caves, namely Mammoth Cave, USA (White, 2017), and Chamois Cave, France (Audra and Nobécourt, 2013), where its formation is not related to cryogenic processes. Löweite has only rarely been reported from volcanic caves (Hill and Forti, 1997).

In Cove Cave, the most likely source of dissolved sulfate, necessary for the precipitation of sulfate minerals, is finely disseminated pyrite in the dark-coloured Silurian limestones in which this cave developed (Smith and Rasmussen, 2020), whereas magnesium for löweite formation was likely sourced from dolostones in the area. In contrast, the presence of sodium, which is needed for the minerals mirabilite, löweite, and eugsterite, is more challenging to explain. One possibility is that, given the proximity to the coast, the Na is sourced from aerosols from the ocean and/or the sea ice surface, which has been recognised as a dominant source of Na in coastal Arctic ice cores (Rhodes et al., 2018). The most probable source for the traces of dolomite is the host rock, while traces of quartz and potassium feldspar are, most likely, detrital material that entered the cave via water influx or aeolian transport.

The mode of formation (during freezing or sublimation) of the hydrated sulfate minerals must also be considered, especially with respect to whether their formation was synchronous with the precipitation of the carbonate fraction. While careful examination of SEM images yielded no conclusive evidence of coeval crystal growth, the high solubility of eugsterite, mirabilite, löweite, and to a lesser extent gypsum can be used to provide insights. For instance, it is clear that the four sulfate minerals did not form prior to the influx of water and subsequent freezing; otherwise pre-existing soluble minerals would have been dissolved and/or washed away. Likewise, the high solubility of these sulfate minerals suggests that the ice in which the cryogenic carbonates were embedded did not melt but rather sublimated; otherwise, these delicate crystals would not have been preserved. Furthermore, the cave section hosting the sampling site must have remained dry and ice-free since the formation of the CCMs. This also implies that the cave ventilation pattern at the time of formation was already similar to the present mode in this sag-type cave.

Based on the observation that the calcite and sulfate minerals were found intermixed, it seems likely that the formation of the sulfate minerals was associated with the formation of the carbonate fraction, in the sense that both their formations were triggered by the same event and therefore took place near-synchronously. Whether this formation was cryogenic, or occurred during the subsequent sublimation of ice, cannot be determined with certainty. A cryogenic formation of gypsum is likely, similar to the observed formation in other ice caves (see above; e.g. Žák et al., 2018), but the other sulfate minerals may have formed during the subsequent sublimation of ice after gypsum and calcite had already formed.

The relatively high δ13C (7.0 ‰–11.4 ‰) values of the Greenland CCMs are comparable to those of CCCfine from the Canadian Arctic Circle (Clark and Lauriol, 1992); however, the δ13C of the Canadian samples reach higher values (up to 17 ‰). Nevertheless, these highly positive values reflect kinetic isotope fractionation as a result of rapid freezing and the associated degassing of carbon dioxide (Lacelle et al., 2009).

The Cove Cave CCMs show highly depleted δ18O (-24.0 ‰ to -16.0 ‰) values that are much lower than the δ18O values of mid-latitude CCCfine and overlap with the lower range of δ18O values of CCCcoarse from mid-latitude caves (Fig. 4). While CCCcoarse δ18O reflects closed-system freezing in small drip water pools in ice, the low values of the Greenland samples are largely related to the isotopically light meteoric precipitation in this high-Arctic setting. We are interpreting the low δ18O values, which can be observed in Cove Cave CCMs, Cove Cave common speleothems (-17 ‰ to -13 ‰), and Canadian CCCfine (although at a smaller magnitude; Clark and Lauriol, 1992; Fig. 4), to be the result of the high-latitude setting in which both sampling sites are located. While existing CCC data are often biased towards central European/mid-latitude sites, with the isotopic compositions of CCCcoarse and CCCfine plotting in distinct O and C ranges, our data demonstrate that such borders are not universally applicable, especially at higher latitudes.

Previous studies have shown that the δ18O values of CCCfine and common speleothems often overlap (Fig. 4; e.g. Luetscher et al., 2013; Žák et al., 2018); however, the δ18O values of Cove Cave CCMs are more depleted than those of common speleothems from the same cave. While our data set is not sufficient to draw definite conclusions, we hypothesise that the isotopic composition of the source water, from which the CCMs and common speleothems precipitated, differed. While the CCMs precipitated recently from melted snow/ice (winter signal), the common speleothems formed from rainwater (annual or summer signal) during an earlier period within the Quaternary under different climatic boundary conditions.

In this study, fine-grained CCMs were successfully dated using both a 230Th/U isochron approach and 14C dating with DCF correction. The isochron indicated that the initial 230Th/232Th activity ratio (1.39±0.006) was elevated in comparison to the bulk-Earth-derived value (i.e. 0.8; Wedepohl, 1995), which had therefore resulted in an undercorrection for detrital Th (Table 1). For dating CCMs in this study, it is sufficient to solely use 230Th / U methods, but we also show that successful and reliable dating of fine-grained CCMs is possible using 14C, provided that the DCF can be constrained and corrected. The independent age information, necessary for the DCF calculation, however, does not have to be provided by 230Th / U dating; it could, for instance, come from 14C dating of stratigraphically coeval organic matter. Where 230Th / U dating of fine-grained CCMs is impracticable (e.g. due to low 238U concentrations and/or high detrital Th contamination), 14C dating could provide a viable alternative. In this study, the 14C dating results complement the 230Th / U isochron age well (Fig. 6). However, because the DCF-corrected 14C age hits a plateau on the calibration curve (Reimer et al., 2020), the 230Th / U method is considered a better approach in this study. On the other hand, 14C dating of fine-grained CCMs could be as reliable as, or better than, 230Th / U dating for periods of time characterised by a steeply sloped calibration curve. Ultimately, the choice of dating will be study-dependent, and regardless of the method used, the dating of fine-grained CCMs could provide additional chronological control on cave ice bodies, which is an emerging (but rapidly disappearing) palaeoclimate archive (e.g. Kern and Perşoiu, 2013; Racine et al., 2022), in particular in those settings where organic inclusions are lacking.

High δ234Uinitial values suggest that the karst system of the cave was hydrologically inactive prior to the event that caused CCM formation. During arid climate periods and/or under permafrost influence, 234U accumulates in the crystal lattice of bedrock minerals through alpha recoil and is easily mobilised when water becomes available (Fleischer, 1982). Speleothems, including CCMs, may record this first mobilisation of 234U after a period of accumulation by elevated δ234Uinitial values (e.g. Wendt et al., 2020). The high δ234Uinitial values of the Cove Cave CCMs (1872±3 to 2193±5) contrast with the comparatively low δ234Uinitial values of common speleothems from the same cave, which represent a hydrologically active karst system (<150). The high δ234Uinitial values of the Cove Cave CCM therefore indicate a prolonged period of permafrost presence prior to their formation.

Combining the results of mineralogical analyses and dating, it can be concluded that the Cove Cave CCMs formed as a result of a singular event at 65±17 a BP (1885±17 CE) and therefore during the cold climate of the late Little Ice Age (ca. 700–50 a BP; Kjær et al., 2022). A reconstruction of surface temperature, based on Greenland ice core data since 1840 CE, indicates that the mid- to late 1880s CE (65–60 a BP) were some of the coolest years on record (Box et al., 2009). It can therefore be assumed that the climate of northeastern Greenland during the time of CCM formation was colder and permafrost was more abundant than today. In order for water to enter the cave through the frozen rock and not immediately freeze and sublimate on the walls, it must carry enough latent heat. Any event that led to these conditions in and around Cove Cave must have been anomalous, which is supported by the lack of CCMs observed in other caves in the area (based on the findings of two expeditions that were part of this project).

In order to explain the formation of fine-grained CCMs in Cove Cave, we consider the following scenarios: (i) an extreme rainfall event providing enough water and thus latent heat to enter Cove Cave, (ii) migration of the hoar frost boundary and melt pond deeper into the cave due to the influx of warm and moist air, (iii) an increase in the active layer thickness of (wet) permafrost due to an anomalously warm summer, year, or interval, (iv) higher cave air temperatures and water availability due to temperate ice covering the cave, and (v) enhanced melting of the local ice cap.

There is no evidence of (i) an extreme rainfall event in the high-resolution Greenland ice cores during the time of CCM formation, although it cannot be excluded that such an event occurred locally or regionally. Though a tectonic fracture exists in the roof of Cove Cave, which could promote water infiltration, the lack of CCM formation in other caves in the area renders this scenario unlikely. Furthermore, in this arid region, with ca. 200 mm yr-1 of precipitation (Schuster et al., 2021), extreme rainfall events with high volumes of water do not occur. Scenario (ii) seems unlikely, as the hoar frost boundary is located at the high point in Cove Cave (Fig. 2). Such boundaries are only observed at the highest points in the Greenland Caves (Barton et al., 2020); hence, shifting this boundary deeper into the cave is unlikely to occur, based on the sag-type geometry of the cave, which determines the location of the cold pool interface. An unusually warm summer, year, or interval that could potentially lead to (iii) an increase in the active layer thickness would have been recorded by Greenland ice cores and observational records along the coast of southern and western Greenland. Based on a merged southwestern Greenland temperature record (Vinther et al., 2006), there is no summer, year, or interval that stands out as being particularly warm in the period of 1868–1902 CE (i.e. 65±17 a BP). Another argument against scenario (iii) is, again, the lack of CCMs in other caves in the area. While it is very likely that an ice cap covered the plateau above Cove Cave in the time interval of interest (Fig. 1), scenario (iv) fails to explain why there was only one generation of CCMs found in Cove Cave, since higher cave air temperatures and water availability would probably have lasted for a prolonged interval. The ice cap does, however, appear to be a key factor together with a tectonic fracture in the roof of the cave, as they enable a mechanism in which sufficient water can pass through the permafrost and enter the underground. Specifically, enhanced melting of the local ice cap (v) could have been triggered by anomalously high temperatures and concurrent lowering of the albedo due to black carbon deposition. Both of these factors have been recognised as the cause for ice-melting conditions all over the Greenland ice sheet, including the dry snow zone, in 1889 CE (Clausen et al., 1988; Fischer et al., 1998; Keegan et al., 2014; Neff et al., 2014). This event is often referred to as the “summer melt episode” of 1889 CE, which, according to Neff et al. (2014), lasted for a few days. The summer melt episode occurred synchronously within the dating uncertainty related to the timing of the CCM formation (1885 CE; 65±17 a BP). Closer to our study site, an ice core from the Flade Isblink ice cap recorded a warming between 1920 and 1930 CE (30–20 a BP) as increased melt percentage, but it did not record enhanced melt during 1889 CE, which might simply be the result of a non-definitive timescale that does not allow for the investigation of short-term events (Lemark, 2010). The ice core from Flade Isblink did, however, record a high concentration of black carbon in 1889 CE (Eckhardt et al., 2023). We therefore infer that the unusually high air temperatures and black carbon deposition associated with this event affected our study area as well, leading to enhanced melting of the local ice cap, with more water entering moulins, reaching the base of the ice cap, and then finding its way through conduits, possibly through tectonic fractures, into the cave. This water turned to ice in the heavily undercooled cave, resulting in the precipitation of fine-grained cryogenic cave carbonates and potentially also cryogenic gypsum during a rather rapid freezing process. Subsequently, the ice sublimated in the cold and dry microclimate of the cave, releasing enclosed cryogenic calcite and gypsum particles which accumulated on the cave floor. As discussed above, the origin of the other hydrous sulfate minerals is less clear; they were, however, found intermixed with calcite and gypsum in the same spot, hinting towards a near-synchronous formation.

The presence of CCMs in Cove Cave indicates that the conditions leading to the summer melt episode of 1889 CE also reached further northeast and thus affected the area surrounding the cave. Linking the CCM formation to this short-term extreme event further provides hints about the rate of formation of fine-grained CCMs (which is a highly understudied field), while demonstrating why fine-grained CCMs cannot be used as a permafrost proxy as the presence or absence of permafrost is generally not influenced by short-term weather conditions.