The history of ground ice formation and intra-permafrost fluid flow as documented by Ra and Th isotopes

While permafrost is considered a permanently frozen soil, it often demonstrates evidence for internal processes, including fluid migration. Here, we present data on the chemistry, Ra, Th, and Ac isotopes of saline permafrost from three closely retrieved cores drilled at Adventdalen, a fjord Valley in central Svalbard, which provides evidence for a fingering-style intra-permafrost recent fluid flow. Ground ice of the different cores differs markedly in its salinity and composition. In one core, which has a relatively high salinity (up to > 20 000 mgCl L⁻¹), and a composition similar to seawater, the long to short-lived isotope activity ratios (AR), e.g., (²²⁶Ra $/$ ²²³Ra)_AR and (²²⁸Ra $/$ ²²³Ra)_AR, are relatively low, being similar to parent isotope activity ratios (²³⁰Th $/$ ²²⁷Ac and ²³²Th $/$ ²²⁷Ac, respectively) on grain surfaces (exchangeable fraction). Ground ice of another core, which is less saline and has $\mathrm{Na}/\mathrm{Cl}$ and SO₄ $/$ Cl ratios higher than seawater, demonstrates much higher Ra isotope ratios, closer to parent ratios in the bulk sediment. Ground ice in a third core, with chemical composition similar to the latter, shows high (²²⁶Ra $/$ ²²³Ra)_AR, albeit low (²²⁸Ra $/$ ²²³Ra)_AR. It is suggested that the different isotope ratios are due to different residence times, and that the parameter controlling the longer-lived (²²⁶Ra and ²²⁸Ra) activities (hence, long to short isotope ratios) is radium diffusion from inside the grains via partly liquidized nano-pores. While ground ice in the less saline cores could have been formed during permafrost formation (10–9 ka), ground ice in the more saline core went recently (years to decades) through reset of its Ra clock, which did not allow a significant diffusion of the long-lived ²²⁶Ra and ²²⁸Ra from inside the grains. This reset is probably the result of a Late Holocene intrusion of saline fluids, which arrived from a low-Th or high water : rock ratio basement aquifer, and mixed with the original, less saline ground ice. Recent salinization is also supported by the presence of high salinity all the way up into the syngenetic permafrost, which has been deposited during the Late Holocene. The above highlights the internal dynamics of saline permafrost, which may affect its resilience to the ongoing global warming.

1 Introduction

Permafrost is a soil or rock that has remained at or below 0 °C for at least two consecutive years. While permafrost covers more than 20 % of the northern hemisphere land area, it is evident that its distribution has dramatically declined during the last several decades due to the ongoing global warming (Brown et al., 2002; Obu et al., 2019; Li et al., 2022), with continuous permafrost areas changing into discontinuous and sporadic permafrost zones (Kwong and Gan, 1994; Anisimov and Nelson, 1997; Osterkamp and Romanovsky, 1998; James et al., 2013). While some of the permafrost is very old and thick, in particular in inland settings, such as Siberia, Canada, Alaska, and Antarctica (Gilichinsky and Wagener, 1995; Dobinski, 2011; Abramov et al., 2021), other areas are dominated by young permafrost (Late Pleistocene to Holocene), related to the last deglaciation, which exposed vast areas to the atmosphere and opened the way for cooling and permafrost formation (e.g., Jin et al., 2007; French, 2007). Following exposure, permafrost aggradation may go two ways. The first, applied to the already deposited sediments and bedrock, is the top-down, epigenetic permafrost formation, while the other is the bottom-up, syngenetic cooling and freezing of the newly deposited sediments.

Whereas most permafrost is significantly colder than 0 °C, it is not necessarily strictly frozen (French, 2007; Dobinski, 2011; Keating et al., 2018), which is due to both the lowering of the freezing point in saline permafrost (Ahonen, 2001), as well as due to capillary or surface adsorption effects in the sediment pore space (e.g., Sheshukov and Nieber, 2011; Wang et al., 2020), which in turn could be related to water content and grain size (e.g., Zhang et al., 2019; Wang et al., 2021). The presence of fluids may boost intra-permafrost processes, allowing the re-distribution and migration of fluids in the cryotic pore space (Fisher et al., 2019; Lacelle et al., 2022).

Intra-permafrost processes are often associated with supra-permafrost processes, such as infiltration from the active layer (Mackay, 1983; Marsh and Woo, 1993; Boike et al., 1998), runoff on slopes to form foothill mounds (Åkerman and Malmström, 1986), intrusion from rivers (Alekseyev, 2015), and thaw-freeze crack-filling, which produce the commonly-observed ice wedges and polygon structures (e.g., Harry and Gozdzik, 1988; Mackay, 1989; Christiansen et al., 2005; Opel et al., 2018). On the other hand, intra-permafrost ice segregation is also common (Mackay, 1983; Mackay and Dallimore, 1992; Solomatin and Xu, 1994; Fu et al., 2022; Kipp et al., 2025). While this is more difficult to trace, Weinstein et al. (2019) showed by Ra and Th isotopes, combined with ³H, that a segregated ice layer at Adventdalen, central Svalbard (Fig. 1), had been formed very recently (less than one year).

Naturally, due to its lower freezing temperature, saline permafrost (i.e., soils with significant content of soluble salts, such that results in a freezing point depression, e.g., Banin and Anderson, 1974; Marion, 1995) is more prone to fluid migration and intra-permafrost processes. Near-surface saline permafrost is widely distributed in the northern circumpolar region, especially in the continuous permafrost zone, covering up to 35 % of that area (Brouchkov, 2002). The salinity source is commonly related to seawater (Hivon and Sego, 1993; Brigham-Grette and Hopkins, 1995; Forman et al., 2004; Lønne and Nemec, 2004; Jones et al., 2023), although it can also be strictly terrestrial, related to evaporation processes in inland basins and to water-rock interaction (e.g., Dickinson and Rosen, 2003; Henkemans, 2016; Gao et al., 2017). Salinity and cryotic state (i.e., freezing degree of the pore fluid) are highly heterogeneous in the saline permafrost environment (e.g., Brouchkov, 2002; Ahonen, 2001; Dafflon et al., 2016). This heterogeneity is sometimes manifested by the presence of lenses of unfrozen ground, or overcooled brines (“cryopegs”), which usually form due to freezing and salt expulsion processes (e.g., Yershov, 1998; Iwahana et al., 2021).

In Svalbard, near-surface saline permafrost is commonly observed in fjord valleys, where seawater ingression occurred during the early Holocene, following glacial retreat and sea level rise (e.g., Cable et al., 2018; Gilbert et al., 2019; Rotem et al., 2023). Permafrost formation had been inhibited until after seawater regression and the exposure of the recently deposited sediments to the atmosphere (Gilbert et al., 2018). The actual formation of the permafrost depends on the combination of climatic and soil physical attributes. Accordingly, while Hornum et al. (2020) argued that freezing at Adventdalen (Fig. 1) did not occur until 4500 years BP due to the relatively warm mid-Holocene, Rotem et al. (2023) suggested that permafrost started forming immediately after exposure (ca. 9 ka), which was assisted by the thermal conductivity difference between solid ice and liquid water. In this paper, we show by chemistry and radioisotopes that the permafrost in eastern Adventdalen demonstrates a complex and multi-stage history of ground ice formation.

Figure 1 Location and detailed map of the ADE study site. The dashed line represents sea ingression limit at the early Holocene (Lønne and Nemec, 2004, drawn after Hodson et al., 2020). The map is provided with courtesy of the © Norwegian Polar Institute. All rights reserved.

Radium isotopes

In this work, we report the activities (disintegrations per time) of radium isotopes in ground ice, as well as their radioactive parents (thorium and actinium) on the permafrost sediments. Radium has four naturally occurring isotopes: ²²³Ra, ²²⁴Ra, ²²⁶Ra, and ²²⁸Ra. All isotopes are produced by decay chains, stemming from ²³⁸U, ²³⁵U, and ²³²Th (Fig. 2). Specifically, all Ra isotopes are produced by the α decay of different Th (thorium) isotopes (²²⁷Th, ²²⁸Th,²³⁰Th, and ²³²Th, respectively). In the case of ²²³Ra, we usually relate to its grandparent ²²⁷Ac, with a half-life of 21.7 years, due to the short life of its direct parent ²²⁷Th (half-life: 18.7 d). An important aspect is that thorium (parent isotope) is highly particle-reactive and tends to strongly adsorb onto mineral surfaces, whereas radium is more mobile and is more common in the dissolved phase. Also, the alpha-recoil effect during radioactive decay provides the kinetic energy necessary to eject radium from the mineral lattice into the pore space. The ²²⁶Ra is the longest-lived radium isotope, with a half-life of 1601 years; ²²⁸Ra has a half-life of 5.75 years, while ²²⁴Ra and ²²³Ra are short-lived, with half-lives of 3.66 and 11.4 d, respectively. The wide range of half-lives of the radium isotopes allows their application to various processes with very different time scales. For example, for short-term groundwater processes, ²²⁸Ra, ²²³Ra, or ²²⁴Ra, and their activity ratios are good tracers (e.g., Moore, 2003; Hsieh et al., 2013). However, for a distinction between thawed ground ice and active layer water or between young and old sub-permafrost groundwater, ²²⁶Ra would be a better tracer, since its growth toward equilibrium with its parent nuclide, ²³⁰Th, takes 5–6 half-lives (8–10 000 years; e.g., Weinstein et al., 2019; Rotem et al., 2024).

Radium isotope research in the polar regions has been mostly focused on the land-ocean interface, like shelf-deep basin exchange processes (Rutgers van der Loeff et al., 1994; Kipp et al., 2019, 2023), river discharge as a carrier of carbon and nutrients to the oceans (Rutgers van der Loeff et al., 2002; Bullock et al., 2022) or lakes (Dabrowski, 2020) and direct groundwater discharge to the sea (e.g., Dimova et al., 2015; Charkin et al., 2020). While also studied in relation to soil or river contamination (e.g., Chevychelov and Sobakin, 2017), as tracers of weathering processes (Linhoff et al., 2020), or as part of geological and pedological surveys (e.g., Wojtasik et al., 2017), radium isotopes have just recently been studied as a potential tracer of permafrost degradation-segregation or as a proxy for residence time determination (Weinstein et al., 2019; Kipp et al., 2025). In fact, the only radioisotopes that have previously been used for the determination of ground ice residence time in permafrost soil were uranium isotopes (Ewing et al., 2015) and ³H (for young ice, e.g., Burn and Michel, 1988).

Figure 2 The three U-Th decay series. Radium isotopes are designated orange, and Thorium is labelled green. Note that all Ra isotopes are produced by the decay of Th isotopes, and that two of the four Ra isotopes are along the ²³²Th decay chain.

2 Study site

The data presented in this study are from cores drilled into the permafrost in the Adventdalen Valley (hereafter Adventdalen), Svalbard (Fig. 1), during March 2017 and March 2022. The Adventdalen is a U-shaped fjord valley, carved by advancing and retreating glaciers. During the last glacial cycle, it was eroded to its basement (Elverhøi et al., 1994), which was followed by Early Holocene glacier retreat and seawater ingression up valley (Lønne and Nemec, 2004), resulting in the deposition of deltaic sediments (Gilbert et al., 2018). A relatively fast elastic rebound (Forman et al., 2004) resulted in the exposure of the sediments to the atmosphere, starting at the eastern part of the valley ca. 9.5 ka and arriving at the current coastline location no later than 4 ka (Gilbert et al., 2018). The exposed sediments froze epigenetically within a short time after exposure to the atmosphere (100 s of years; Rotem et al., 2023). The marine sediments were then covered by several meters of mid to Late Holocene fluviatile sediments, which were overlain by 2–3 m of aeolian deposits, and both went through syngenetic freezing (Gilbert et al., 2018). At present, the thickness of the permafrost in the valley is estimated at less than 100 m next to the coast (Humlum et al., 2003) and up to 250 m up-valley (Isaksen et al., 2001), most of it being epigenetic and the shallow 4–5 m syngenetic, while active layer depth ranges between 0.5–1 m (Christiansen et al., 2005; Weinstein et al., 2019). Permafrost in Svalbard is considered continuous (Obu et al., 2019), although non-frozen and partially frozen permafrost were also reported (Keating et al., 2018; Weinstein et al., 2019), and recent works about sub-permafrost hydrology also question the continuity of the permafrost in this area (e.g., Rotem et al., 2024).

Four boreholes were drilled at the ADE site, a river terrace, 23 m a.s.l., 9.8 km up-valley from the Adventfjorden (Fig. 1). Two closely spaced (0.5 m apart) boreholes, centered at 78.17220° N 16.06130° E, were drilled during spring 2017, and they are treated here as one core, ADE-17. The other two cores, ADE-1 and ADE-2, were drilled in spring 2022. ADE-1 is located within a few meters from ADE-17 (78.17216° N 16.0610° E), and ADE-2 was drilled ca. 30 m away (78.17243170° N 16.06114840° E). The valley-fill section at the site (14–20 m, Gilbert et al., 2018, and this study) consists of 1.5 m of fine-grained aeolian deposits underlain by fluviatile gravel down to 5.5 m, which in turn is underlain by deltaic sediments (ca. 12 m). The Holocene section covers glacial sand deposits, 2.5 m thick (Gilbert et al., 2018), which are underlain by basement rocks (Lower Cretaceous, shales) (Grundvåg et al., 2019).

3 Methods

The drilling campaigns were executed using the University Center in Svalbard (UNIS) permafrost drill rig (Gilbert et al., 2018), using core barrels of 43 mm diameter (ID) in 2017 and 58 mm in 2022. The two boreholes drilled in 2017 reached depths of 13 and 9 m, while the 2022 ones (ADE-1 and -2) reached a depth of 16 m. In the latter, the bottom 1–2 m was composed of bedrock, which was excluded from sample processing. The upper 5 m (syngenetic part) of ADE-2 was drilled with a jackhammer, and the extracted cuttings were analyzed only for chemistry (no Ra isotopes). Core length, borehole depth, core freezing status, and gravel content were recorded in the field. Retrieved core sections were sealed in plastic bags and stored frozen (−18 °C) at UNIS. The cores were subsequently sectioned in a cold room (−5 °C) into 0.5–1 m depth intervals. Subsamples were scraped and crushed into small fragments, which were transferred to 250 mL centrifuge tubes. For samples with low water (ice) content, up to 40 mL of Ra-free water was added to facilitate porewater extraction. The samples were then thawed in a microwave oven at 600 W for 2 min and centrifuged for 8 min at 11 000 rpm (high gravity) to separate the meltwater from the soil matrix. The extracted water was sequentially filtered through 3 and 0.45 µm. Net extracted water volumes varied strongly (in accordance with the high variability of ice content) and summed to 42–1391 mL per sample. It is important to note that in this paper, we consider all extracted water as thawed ground ice, although some of the core segments were not fully frozen, thus liquid water could also be present to a certain amount, especially in the more saline samples (see Keating et al., 2018; Weinstein et al., 2019). We also note that centrifuging, as well as the addition of Ra-free water to samples with low ice content, could result in some desorption from grain surfaces, as well as in salt dissolution when it exists. Yet, both salts and adsorbed ions are assumed part of the pore space chemistry (see a similar conception by Ewing et al., 2015, regarding U isotopes in ground ice). Nevertheless, the discussion in this study focuses on Ra isotope ratios, which are less affected by possible desorption. Please also see the discussion of (²²⁴Ra $/$ ²²⁸Ra)_AR in Sect. 5.4.

Most of the extracted water was used for Ra isotope analysis, while 30–60 mL was used for elemental chemistry analyses. For Ra isotopes, the solution was run at least three times through columns, filled with 20 g manganese-coated fibers (flow rate of 0.5 L min⁻¹). Samples with a pH lower than 6 were pre-treated by adding a low-concentration NaOH solution (Ra-free). Sample water was also run through a secondary Mn-coated fiber to account for Ra adsorption efficiency. In six samples (DR-AD-121-126), the adsorption efficiency was determined by measuring and comparing ²²⁶Ra on both fibers and water samples. Efficiencies were very variable, from 90 % down to < 50 %, with the lower end being mainly due to the relatively low pH and/or reducing conditions. Measured activities of all isotopes were corrected accordingly, assuming no mass-dependent fractionation. Nevertheless, due to the uncertainties about both the extraction protocol and the adsorption efficiency on the fibers, in this paper, we mainly focus on isotope ratios.

The short-lived ²²⁴Ra and ²²³Ra were measured by the RaDeCC system (Moore and Arnold, 1996) within 2–3 d after water-soil separation, while ²²³Ra was also measured after ca. 10 d, as to check for 220–219 cross-talk interferences (e.g., Diego-Feliu et al., 2020). We note that we found no evidence for cross-talk (i.e., time-corrected ²²³Ra activities showed no higher activities during the first compared with the later measurement), which is probably due to the low activities of both isotopes (²²⁴Ra was mostly < 0.5 cpm). Both ²²⁴Ra and ²²³Ra were corrected for chance coincidence, following Moore and Arnold (1996). ²²⁴Ra was re-measured after 3–4 weeks, after reaching equilibrium with its radioactive parent ²²⁸Th. Fibers were measured for ²²⁶Ra by an emanation system and Lucas Cells (Mathieu et al., 1988) after a 3-week incubation, to allow for ²²²Rn equilibration with ²²⁶Ra. ²²⁸Ra was measured by a low-background well-type HPGe gamma spectrometer (Canberra), using the 911 keV peak of ²²⁸Ac. All results are presented as dpm (disintegrations per minute) per liter of thawed ground ice, which can change strongly due to the ice content of the permafrost. Analytical errors on ²²⁶Ra and ²²⁸Ra were ≤ 7 %, while on ²²⁴Ra they were ≤ 13 %, and those on the low activity ²²³Ra averaged 28 %, and in two cases exceeded 40 % (errors determined, following the protocol of Garcia-Solsona et al., 2008).

The Radium isotopes accumulate in the frozen pore space following recoil from their parent nuclides Th and Ac, which are mostly located either on or within the sediment grains. Accordingly, thorium isotopes and ²²⁷Ac were measured both in the bulk grains and in the exchangeable fraction of the permafrost sediment, after thawing and completing the ground ice-soil separation. Several grams of soil samples were disaggregated and homogenized in an agate mortar for the bulk sediment analysis. For the bulk sediment measurements, 0.3 g of each sample was placed in a Teflon beaker and was treated with 1 mL HF and 5 mL HNO₃(both concentrated, ultrapure grade). A droplet (ca. 0.035 mL) of a ²²⁹Th spike (86 ppt) was added to the solution, and the beakers were placed on a hot plate (200 °C) to let evaporate. The residue was then treated with 5 mL of HCl. A few drops of H₂O₂ were added to treat the organic matter content. The solution was evaporated, and then dissolved by 6 mL 7N HNO₃. Two mL (out of the 6), dedicated for ²³²Th and ²³⁰Th analyses, were run through Bio-Rad AG 1X8 200–400 mesh resin, following the protocol for thorium ion chromatography separation (Grant et al., 2012). The remaining fraction of 4 mL was kept for ²²⁸Th and ²²⁷Ac analyses. The exchangeable fraction of thorium and actinium on grain surface was determined by the CEC (Cation Exchange Capacity) ammonium acetate leaching protocol (e.g., Sumner and Miller, 1996). Five grams of sediment was shaken with 25 mL of 1N ammonium acetate overnight. The sample was then centrifuged, and the liquid phase was evaporated, dissolved in 2 mL 7N HNO_3,and run through the Bio-Rad resin (above) for ²³²Th and ²³⁰Th analyses (after adding a ²²⁹Th spike, as in the above). A second set of 5 g sediments was prepared for ²²⁸Th and ²²⁷Ac, using the same protocol (albeit no chromatography). ²³⁰Th and ²³²Th (radioactive parents of ²²⁶Ra and ²²⁸Ra, respectively) were determined by the Neptune Plus Multicollector ICPMS at the Institute of Earth Science, The Hebrew University of Jerusalem. All samples for ²²⁸Th and ²²⁷Ac were run through 20 g manganese-coated fibers (the bulk sediment solutions were first buffered with NaOH to pH ∼ 7) and measured by the RaDeCC system (Moore and Arnold, 1996), assuming equilibrium has been achieved with their short-lived radium isotope daughters. Results are presented as dpm g⁻¹ sediment. Analytical errors for the MC-ICPMS thorium measurements were below 2 % and 0.5 % for the bulk and exchangeable fractions, respectively. In the RaDeCC measurements, while errors for ²²⁸Th were < 17 % (average of 8 %), those of the ²²⁷Ac averaged 23 % for the exchangeable fraction and 34 % for the bulk sediment which is due to the low counts.

Major elements were analyzed at the Geological Survey of Israel (GSI). Cations and SO${}_{\mathrm{4}}^{-\mathrm{2}}$ were measured by inductively coupled plasma atomic emission spectroscopy (ICP-OES; Optima 5300). The Cl⁻ concentrations were determined according to their concentration in the solutions; above 50 ppm via potentiometer titration (Metrohm 702 SM Titrino titrator connected to a chlorine electrode), while below 50 ppm by Ion Chromatograph, which was also used for the Br⁻ concentrations. Bicarbonate was analyzed as alkalinity via titration (analyzed only in 60 % of the samples). The analytical error for all majors is less than 5 %. We note that the chemistry of ground ice in ADE-17 was already presented in Rotem et al. (2023). Ice content was determined by overnight drying of sediment samples at 55 °C, and is defined as gravimetric moisture content $(W-D)/D\times \mathrm{100}$, where W and D refer to the wet and dry sample mass (Gilbert et al., 2018).

4 Results

4.1 Ground ice chemistry

Ice content (frozen water mass divided by dry mass) of the samples varied strongly, from 2 % to > 180 % (180 % ice content translates into 64 % water content), and was, in general, higher in the shallow, syngenetic part (Table A1, Appendix A).

Ground ice (i.e., water extracted from thawed permafrost) chemistry differs strongly between the cores and between the shallow, syngenetic, and the deeper, epigenetic parts of the permafrost profile. In cores ADE-17 and ADE-1, chemistry of the syngenetic ground ice is basically that of fresh water with typical TDS concentrations of < 400 mg L⁻¹ (Table A1), while ionic ratios of major elements to Cl⁻ are ≫ 1 (e.g., $\mathrm{Na}/\mathrm{Cl}$ of 1.01–3.58, Figs. 3 and 4). On the other hand, ground ice in the epigenetic section, deeper than 5–5.5 m, is mostly brackish to saline, with 450–3230 mg Cl⁻ L⁻¹ (TDS of 900–6500 mg L⁻¹; note that HCO${}_{\mathrm{3}}^{-}$ and Mg²⁺ were not measured in some of the samples). In ADE-2, salinity is significantly higher (up to 7960 mg Cl⁻ L⁻¹ and TDS of 17 190 mg L⁻¹), with one outlier (DR-AD-125) reaching > 23 000 mg Cl⁻ L⁻¹ and TDS of > 40 000 mg L⁻¹ (Table A1). While the higher salinities were observed at epigenetic depths (> 5.5 m), salinities in the syngenetic section of ADE-2 were also relatively high (TDS ≫ 1000 mg L⁻¹), which reached TDS concentrations > 6000 mg L⁻¹ at the base of this section (Table A1).

In ADE-17, Cl⁻ and other major elements show a mixing pattern with depth, with concentrations increasing from 470 mg Cl  L⁻¹ at 6 m to 1900–3200 mg Cl L⁻¹ at 9–12 m (Fig. 3a–e). This is not observed in ADE-2, where high salinities are observed up to the top of the epigenetic section. Unlike Cl⁻, the Ca⁺² does not show any pattern with depth also in ADE-17, and concentrations in the epigenetic section are similar to those in the syngenetic one (Fig. 3f).

Ion ratios in the ground ice of ADE-17 and ADE-1 vary strongly, with $\mathrm{Na}/\mathrm{Cl}$ and SO₄ $/$ Cl showing up to 3.60 and 0.77, respectively, in the syngenetic section (Figs. 4 and 5a–b), and lower ratios in the epigenetic part (≤ 2 and ≤ 0.45, respectively, Fig. 5a–b). On the other hand, in ADE-2 ratios are more uniform, with $\mathrm{Na}/\mathrm{Cl}$ equivalent ratios similar to that of seawater (0.84–1.02) both in epigenetic and the syngenetic section (except for the shallowest-measured sample, at 3.5 m, with $\mathrm{Na}/\mathrm{Cl}=\mathrm{0.4}$), and SO₄ $/$ Cl (in the epigenetic section) mostly lower than that of seawater (≤ 0.1, Figs. 4 and 5a–b). $\mathrm{Ca}/\mathrm{Cl}$, $\mathrm{Mg}/\mathrm{Cl}$, and $\mathrm{K}/\mathrm{Cl}$ generally show similar patterns, with very high $\mathrm{K}/\mathrm{Cl}$ and $\mathrm{Ca}/\mathrm{Cl}$ in the syngenetic part, while close to seawater ratios in the epigenetic part of the profile, this time in all cores (Fig. 5c–d; one outlier, with very high K and low Ca and Mg at 11m, ADE-2, which could be analytical).

Figure 3 Concentration of selected major elements along the profile of the three ADE cores (ADE-17, ADE-1, and ADE-2). The dashed line shows the boundary between syngenetic (syn) and epigenetic (epi) permafrost at the site (Gilbert et al., 2018). The insets in the Cl⁻ and K⁺ diagrams are for sample DR-AD-125 and DR-AD-150, respectively.

Figure 4 (a) Sodium and (b) sulphate against chloride in the ground ice of the ADE cores. Salt expulsion and WRI trends are indicated, with WRI refers to Water-Rock Interaction.

Figure 5 Ground ice equivalent ratios in the ADE cores. The black full diamond and vertical dashed line is for seawater composition (de Baar et al., 2017, Salinity of 35 ppt). The horizontal dashed line as in Fig. 3.

4.2 Radium isotopes in ground ice

Radium isotope activities in ground ice are presented in Table A2. Activities of ²²⁶Ra, ²²⁸Ra, and ²²⁴Ra in ground ice are quite similar, varying between < 1 to 60 dpm L⁻¹ (averages of 10–13 dpm L⁻¹), while ²²³Ra presents much lower activities, with an average of 0.7 dpm L⁻¹ (Fig. 6a–d). Activities of ²²⁴Ra and ²²³Ra are fairly uniform and low in ADE-17 and ADE-1 (slightly higher in the epigenetic section), while significantly higher in the more saline ground ice of ADE-2 (epigenetic section; Figs. 6a–b, and 7). ²²⁶Ra and ²²⁸Ra also show higher activities in the epigenetic, compared with the syngenetic section, with no difference between the cores, and with the highest activities observed at the top of the section (6–10 m depth, Fig. 6c–d).

Figure 6 Radium isotope activities in ground ice of the three ADE cores. The dashed line as in Fig. 3.

Ratios of the long to short-lived radium isotopes, (²²⁶Ra $/$ ²²³Ra)_AR, demonstrate a large range (4 to 72; Fig. 8a, with three outliers > 100) with marked differences between the cores. While in ADE-17 and ADE-1, (²²⁶Ra $/$ ²²³Ra)_AR both in syngenetic and the epigenetic section are close to or higher than the secular equilibrium activity ratio (21.7, which is the natural ratio of parent nuclides ²³⁸U $/$ ²³⁵U, e.g., Kiro et al., 2015; Weinstein et al., 2021), ratios in ADE-2 (epigenetic section) are relatively low, ranging between 4–14 (Figs. 7c and 8a).

Figure 7 Radium isotope activities and ratios against chloride concentrations. (a) ²²⁶Ra, (b) ²²³Ra, (c) ²²⁶Ra $/$ ²²³Ra.

(²²⁶Ra $/$ ²²⁴Ra)_AR shows a similar pattern, where ratios in ground ice from cores ADE-17 and ADE-1 are higher than those in ADE-2 samples (> 1 and 0.3–1, respectively, Fig. 8b). Ratios of ²²⁶Ra to the medium-lived (half-life of 5.75 years) ²²⁸Ra are mostly ≤ 1, with no difference between the cores (Table A2), although some of the ADE-17 samples show ratios ≫ 1. Despite of the much shorter life of ²²⁸Ra, (²²⁸Ra $/$ ²²³Ra)_AR shows a pattern very similar to that of and (²²⁶Ra $/$ ²²³Ra)_AR (Fig. 8c). Last, the daughter-parent ratio (²²⁴Ra $/$ ²²⁸Ra)_AR in ADE-17 ranges between 0.13–0.68, while in ADE-1 and ADE-2, ratios in the epigenetic part are mostly higher (0.55–1.58, Fig. 9).

Figure 8 Radium isotope activity ratios in ground ice and their parent activity ratios in bulk soil grains and on grain surfaces (exchangeable fraction). (a) (²²⁶Ra $/$ ²²³Ra)_AR and (²³⁰Th $/$ ²²⁷Ac)_AR; (b) (²²⁶Ra $/$ ²²⁴Ra)_AR and (²³⁰Th $/$ ²²⁸Th)_AR. (c) (²²⁸Ra $/$ ²²³Ra)_AR and (²³²Th $/$ ²²⁷Ac)_AR. The vertical dotted line in (a) indicates the natural ratio of (²³⁸U $/$ ²³⁵U), parents of ²²⁶Ra and ²²³Ra, respectively. The dashed line as in Fig. 3.

Figure 9 (²²⁴Ra $/$ ²²⁸Ra)_AR in the ADE cores. The dashed line as in Fig. 3.

4.3 Thorium isotopes and ²²⁷Ac in permafrost soil

Thorium isotope activities in soil samples (bulk and exchangeable fraction) from core ADE-17 are presented in Table A3. ²³²Th and ²³⁰Th activities in bulk samples are two to three orders of magnitude higher than in the exchangeable fraction (4.3–8.4 and 0.01–0.17 dpm g⁻¹, respectively), reflecting their main residence inside the grains (mineral lattice). On the other hand, ²²⁸Th activities in the exchangeable fraction are similar to those in the bulk (both are in the range of 0.1–1.2 dpm g⁻¹, Table A3 and Fig. 10), implying that most of the ²²⁸Th is adsorbed onto grain surfaces. Importantly, although being progeny of ²³²Th, ²²⁸Th activities are, on average, twice the ²³²Th activities in the exchangeable fraction. This is due to the fact that ²²⁸Th is also produced from ²²⁸Ra that recoiled via the decay of ²³²Th close to grain rims. The above is also true for ²²⁷Ac (Table A3), where exchangeable fraction activities are close (though lower) to bulk activities. This is especially true for the epigenetic section, while in the syngenetic part, bulk sediment ²²⁸Th is up to 12 times that in the exchangeable fraction (i.e., grain surface ²²⁸Th is not dominant, Fig. 10), although still lower than bulk/exchangeable ratios of ²³²Th and ²³⁰Th (10's to 100's, Fig. 10).

(²³⁰Th $/$ ²³²Th)_AR in the bulk sediment is quite uniform along the whole permafrost section, ranging between 0.7–1.0 (Table A3; note that these are activity ratios, while concentration ratios average $\mathrm{5}\times {\mathrm{10}}^{-\mathrm{6}}$). Ratios in the exchangeable fraction change from < 1 in the syngenetic permafrost to > 1 in the epigenetic section (5–12 m depth). (²³⁰Th $/$ ²²⁷Ac)_AR in bulk sediment are significantly higher than the secular equilibrium activity ratio (21.7, Fig. 8a), while all exchangeable fraction samples show lower than equilibrium activity ratios (2–12, Table A3, and Fig. 8a). The latter is apparently due to the high concentration of ²²⁷Ac on grain surfaces. Notably, daughter ratios, (²²⁶Ra $/$ ²²³Ra)_AR in ground ice of ADE-2 are similar (slightly higher) to the (²³⁰Th $/$ ²²⁷Ac)_AR in the exchangeable fraction, while in ADE-17 ratios are much higher, “mid-way”, between the exchangeable fraction and bulk sediment ratios (Fig. 8a). (²³²Th $/$ ²²⁷Ac)_AR and (²³⁰Th $/$ ²²⁸Th)_AR show a similar pattern, with exchangeable fraction ratios being much lower than the bulk sediment ratios, and with the respective daughter (²²⁸Ra $/$ ²²³Ra)_AR and (²²⁶Ra $/$ ²²⁴Ra)_AR in the ground ice of ADE-2 being similar to the exchangeable fraction parent ratios and those of ADE-17 positioned mostly between exchangeable fraction and the bulk sediment ratios (Fig. 8b–c).

5 Discussion

5.1 The heterogeneous nature of ground ice in Adventdalen

Our study site is located within the reach of the marine early Holocene transgression (Fig. 1, Gilbert et al., 2018), suggesting a-priori that the saline water observed in the permafrost epigenetic section is of an estuary-marine origin. Nevertheless, ground ice retrieved from the same depths in the three cores, located ≤ 30 m apart (Fig. 1), differ markedly in their salinity and chemical composition, as well as in their radium isotope ratios (Figs. 3–8). The salinity of ground ice from borehole ADE-2 is relatively high, with up to more than seawater salinity and no distinct pattern with depth (Fig. 3). In ADE-17, salinity does not exceed 20 % that of seawater, and it increases with depth in the epigenetic section. Finally, in ADE-1, salinity is less than 6 %–7 % that of seawater, and no distinct pattern is observed. The difference between the cores is not just about the extent of dilution with fresh water/ground ice, but also about the contribution of water : rock interaction, as demonstrated by the ionic ratios. While in ADE-2, ionic ratios are relatively similar to seawater (Figs. 4 and 5; although $\mathrm{Na}/\mathrm{Cl}$ of 1 could indicate halite dissolution in some of the samples and high SO₄ $/$ Cl in two of the samples could be due to sulfide oxidation, possibly related to the coal layers, common in the nearby slopes), ground ice from ADE-17 and ADE-1 shows $\mathrm{Na}/\mathrm{Cl}$ and SO₄ $/$ Cl significantly higher than in seawater (Figs. 4 and 5a–b). The similarity to seawater chemistry in ADE-2 ground ice favors a marine source for its saline component. On the other hand, the > 1 $\mathrm{Na}/\mathrm{Cl}$ in ADE-17 and ADE-1 (compared with 0.86 in seawater) suggests that, although it probably also contains a seawater component, porewater in these two cores went through significant water-rock interaction (WRI, e.g., dissolution or ion exchange). Accordingly, this water is the result of both mixing between fresh, meteoric water and Early Holocene seawater and of WRI. It should be noted that seawater freezing and saline fluid expulsion could not be a significant player here, since in this case fractionation would result in $\mathrm{Na}/\mathrm{Cl}$ ratios lower than that of seawater (e.g., Herut et al., 1990), which is hardly observed (see Fig. 4a).

Figure 10 Bulk sediment $/$ exchangeable (B $/$ ex) activity ratios of thorium isotopes in sediments from ADE-17. The dashed line as in Fig. 3. Note the logarithmic scale.

The observation of a high salinity profile in ADE-2 all the way to the top of the epigenetic, marine section could be the result of an immediate freezing following exposure to the atmosphere (e.g., Rotem et al., 2023), which limited mixing with fresh water, as well as reduced the effect of WRI. However, the fact that this profile is very different from the nearby boreholes, including its salinity, composition and pattern with depth, favors a later introduction of saline water into the permafrost pore space at this location, which could occur either via intrusion or by intra-permafrost segregation processes (e.g., Ahonen, 2001; French, 2007; Iwahana et al., 2021). Late Holocene intrusion of a saline fluid is firmly supported by the fact that the high salinity in ADE-2 continues into the syngenetic section (Table A1 and Fig. 3), which has been deposited during 4–2 ka (Gilbert et al., 2018). This will be further discussed in light of the radium and thorium isotope data, as follows.

5.2 Radionuclides insights into ground ice timing

The differences in chemistry between the cores are accompanied by differences in the Ra isotope ratios, where long to short-lived isotope activity ratios (²²⁶Ra $/$ ²²⁴Ra, ²²⁶Ra $/$ ²²³Ra and ²²⁸Ra $/$ ²²³Ra) are significantly lower in the ground ice of ADE-2, compared with those in ADE-17 and ADE-1 (Figs. 8a-c). In particular, while (²²⁶Ra $/$ ²²³Ra)_AR in ADE-17 and ADE-1 are ≥ 20, in ADE-2 these ratios are ≤ 14 (Fig. 8a). The low ratios in ADE-2 are apparently because activities of the short-lived isotopes are significantly higher than in the two other cores (Fig. 6b), while ²²⁶Ra activities are quite similar (Fig. 6a). However, the significantly higher salinity in ADE-2 should result in higher activities of all radium isotopes in this core due to the reduced adsorption (e.g., Kraemer and Reid, 1984; Gonneea et al., 2008; Kiro et al., 2012; Vinson et al., 2013), and assuming a non-complete frozen nature of the permafrost (e.g., Gilbert et al., 2019; Keating et al., 2018; Weinstein et al., 2019). The observation that this is the case only with the short-lived isotopes (²²⁴Ra and ²²³Ra), while not with the longer-lived ²²⁶Ra and²²⁸Ra, raises the possibility of a timing control on the radium isotopes of ground ice in this core (e.g., Weinstein et al., 2019, 2021), namely that the relatively low ²²⁶Ra and ²²⁸Ra (compared with ²²³Ra) is the result of a short residence time of the ADE-2 ground ice. Time constraints will be further discussed below.

Another important observation is that in ADE-2, (²²⁶Ra $/$ ²²³Ra)_AR, (²²⁶Ra $/$ ²²⁴Ra)_AR and (²²⁸Ra $/$ ²²³Ra)_AR in ground ice are similar to (i.e., close to equilibrium with) their parent nuclide activity ratios (²³⁰Th $/$ ²²⁷Ac, ²³⁰Th $/$ ²²⁸Th and ²³²Th $/$ ²²⁷Ac, respectively) in the exchangeable fraction (i.e., surface of grains). On the other hand, in ADE-17 radium ratios are consistently higher and are closer to parent ratios in the bulk sediment (Figs. 8a–b). We note that while Th and Ac isotope activities were measured only in the ADE-17 sediment, the very close proximity of the cores (within 30 m) and the relatively uniform Th isotope ratios in the exchangeable fraction along the sediment profile (Table A3) suggest that the measured sediment nuclide ratios, in particular the differences between ratios in bulk and the exchangeable fractions, should not be very different between the ADE core locations.

In fact, ²²⁶Ra activities in ground ice (1–60 dpm L⁻¹) of all three cores are significantly lower than the ²³⁰Th activities in the exchangeable fraction (0.02–0.17 dpm g⁻¹, Table A3), which considering a dry solid density of 2.5 g cc⁻¹, a relatively high porosity of 50 % (100 % ice content, Table A1) and recoil release fraction of 50 %, is equivalent to 25–200 dpm L⁻¹ (Fig. 11; a more conservative porosity of 25 % would triple these activities). The significantly lower ²²⁶Ra activities in the ground ice are probably the result of most radium (both ²²⁶Ra and other isotopes) being kept on grain surfaces (Fleischer and Raabe, 1978; Suksi and Rasilainen, 1996), either physically attached or adsorbed, if liquid water is present. Notably, ground ice of DR-AD-125, which is characterized by an exceptionally high salinity (Cl⁻ higher than in seawater, Fig. 4, Table C1), showed unusually high radium activities of all isotopes (Table A2), probably due to lower radium adsorption onto grains.

5.2.1 Ground ice residence times

The similarity between (²²⁶Ra $/$ ^slRa)_AR (sl = “short-lived”) and (²²⁸Ra $/$ ^slRa)_AR in ADE-2 and the parent nuclide activity ratios in the exchangeable fraction (Figs. 8a–c) suggests that pore space radium isotopes are in equilibrium with parent nuclides on grain surfaces. Equilibrium activity ratios of radium isotopes are supposed to be achieved within 5–6 half-lives of the longer-lived isotope (e.g., 8–10 kyr for ²²⁶Ra; 28–35 years for ²²⁸Ra), or shorter if adsorption is involved (Kiro et al., 2013; Weinstein et al., 2021). The concentration of a certain Ra isotope in pore space (water or ground ice) at time t is described by Eq. (1), assuming steady production and negligible dissolution and diffusion (e.g., Kiro et al., 2015; Weinstein et al., 2021), as is probably the case in the ground ice.

$$\begin{array}{}\text{(1)}& A={\displaystyle \frac{P}{\mathit{\lambda }\left(K+\mathrm{1}\right)}}\left(\mathrm{1}-{\exp }^{-\mathit{\lambda }\left(K+\mathrm{1}\right)t}\right)+A_{\mathrm{0}}{\exp }^{-\mathit{\lambda }\left(K+\mathrm{1}\right)t}\end{array}$$

where A is the activity [dpm L⁻¹] of the isotope, A₀ is the initial activity [dpm L⁻¹] (when the fluid/ice was emplaced in pore space), P is the production by recoil from the sediment [dpm L⁻¹ yr⁻¹], λ is the decay constant of the isotope [yr⁻¹], K ($=R+\mathrm{1}$; Krishnaswami et al., 1982) is the unitless adsorption coefficient and t is the time since water (or ice) emplacement in the pore space [yr]. The actual activities of ²²⁶Ra and ²²⁸Ra in our samples are controlled by adsorption, as well as by the efficiency of the (thawed) ground ice extraction. Accordingly, we prefer to normalize these longer-lived isotopes to ²²³Ra activities. The short-lived isotopes ²²⁴Ra and ²²³Ra reach steady state activities within weeks, which for the discussed time scales practically means a uniform activity. Therefore, ratios of long to short-lived isotopes solely depend on the initial ratio and on the buildup of the long-lived isotope.

In Fig. 11, we show the buildup of (²²⁶Ra $/$ ²²³Ra)_AR in the ground ice with time. For ²²³Ra, we make use of the average isotope activity in ADE-2, e.g., 1.4 dpm L⁻¹ (Table A2; SD is 1.1 dpm L⁻¹, which is mainly due to one sample of 3.9 dpm L⁻¹). While C₀ of ²²⁶Ra is not well-constrained for the shallow (syngenetic) layers, we assume that fluid in the deeper, epigenetic layers had initial activities similar to that of typical shelf seawater (e.g., 0.1 dpm L⁻¹, Moore, 1996). This is a minimum value, because fluid could reside in pore space much before ground ice formation, but will not change the pattern of buildup to steady-state ratios, as discussed below.

Assuming that ground ice radium in ADE-2 is produced only from the exchangeable fraction, the time that will take (²²⁶Ra $/$ ²²³Ra)_AR to reach exchangeable (²³⁰Th $/$ ²²⁷Ac)_AR is dependent on the adsorption. With K=0 (i.e. R=1), this will take several thousand years. On the other hand, if some liquid is present in pore space (e.g., Keating et al., 2018; Weinstein et al., 2019), this will result in adsorption joining the process, and calculated residence times will significantly shorten. For instance, with K=10, as is the case for seawater salinities (e.g., Weinstein et al., 2021; Garcia-Orellana et al., 2021; Diego-Feliu et al., 2021), the time to reach equilibrium will be on the order of 100's years, with K=100 times will reduce to 10's years and even less with higher K's. Nevertheless, while timing is not constrained, the relatively low (²²⁶Ra $/$ ²²³Ra)_AR and (²²⁸Ra $/$ ²²³Ra)_AR in ADE-2 evidently indicates that there is no significant contribution from inside the grains, where parent activity ratios (²³⁰Th $/$ ²²⁷Ac and ²³²Th $/$ ²²⁷Ac, respectively) are much higher (Fig. 8a, c). Below, we use this observation and the circumstantial evidence of the high salinity in the syngenetic section of ADE-2 to infer about the mechanism that is responsible for the higher ratios in the other cores.

As noted, and unlike ADE-2, (²²⁶Ra $/$ ^slRa)_AR and (²²⁸Ra $/$ ²²³Ra)_AR in ground ice of ADE-17 and ADE-1 are significantly higher than parent nuclide ratios in the exchangeable fraction (e.g., (²²⁶Ra $/$ ²²³Ra)_AR is mostly 20–70), being much closer to (although not in equilibrium with) bulk sediment ²³⁰Th $/$ ²²⁷Ac parent ratios (Fig. 8a). This suggests that on top of contribution from the exchangeable fraction, there was enough time in ADE-17 for some ²²⁶Ra and ²²⁸Ra to arrive from a different source, e.g., from inside the grains. While contribution from the grains could be accomplished by direct recoil from grain rims, there is no reason for this not to happen within the same time scale as that of the exchangeable fraction, which is not the case in ADE-2. Alternatively, we suggest that this contribution is via ²²⁶Ra and ²²⁸Ra diffusion out of the mineral grains, which is time-dependent, and which is not relevant for the short-lived ²²³Ra and ²²⁴Ra. Diffusion is further discussed in Sect. 5.2.2.

The maximum time allowed for ground ice formation in the study site is constrained by the deposition age of the valley-fill marine sediments at the site, which is Early Holocene age (≤ 10 500 years, e.g., Lønne and Nemec, 2004), and furthermore by the actual permafrost formation (i.e., freezing), which could not occur until elastic rebound allowed exposure to the atmosphere, at 9–9.5 ka (e.g., Gilbert et al., 2018; Rotem et al., 2023). This means that ground ice in ADE-17 could be as old as Early Holocene (9–9.5 ka; note that some authors, e.g. Hornum et al., 2020, argue that permafrost formation has been postponed until about 4 ka). On the other hand, the salinization event in ADE-2 should be much younger, as is evident by the high salinity, documented also in the 4–2 ka syngenetic permafrost (Figs. 3 and 4). Below (Sect. 5.2.2), we show that the salinization (Ra clock reset) event probably happened much more recently.

Figure 11 Simulations of the buildup of (²²⁶Ra $/$ ²²³Ra)_AR in permafrost pore space with different adsorption coefficients (K=0 means no adsorption). See text for further explanation.

Figure 12 Conceptual model of ground ice versatility and residence times at the ADE site. See text for explanation. Note that locations of ADE-17 and ADE-2 are not to scale. In fact, they are located next to each other (within 30 m).

5.2.2 Diffusion from grains in the permafrost

Assuming that diffusion from inside the soil grains controls (²²⁶Ra $/$ ^slRa)_AR and (²²⁸Ra $/$ ^slRa)_AR in ground ice, it is the life time of ²²⁶Ra and ²²⁸Ra that could constrain the residence times involved. While for ²²⁶Ra, this could span up to thousands of years, it is the shorter-lived ²²⁸Ra that is of interest here. The observation that the ²²⁸Ra managed to make its way to pore space in ADE-17 (²²⁸Ra $/$ ²²³Ra closer to equilibrium with bulk sediment parent activity ratios) implies that the effective diffusion time scales are within the time life of this isotope (with half-life of 5.75 years, this practically means up to 20–30 years). Assuming similar sediment composition and texture (same site), this practically means that in ADE-2 there was no time enough even for the ²²⁸Ra to diffuse out of the grains, which suggests a very recent “age” for the ground ice salinization in this core.

Extrapolation from higher temperature data (e.g., for Ca) predicts that D_Ra (diffusion coefficients) in sub-zero solids should be very low (≪ 10⁻²⁴ m² s⁻¹, e.g., Brady and Cherniak, 2010; Cherniak, 2010). This should result in diffusion time scales of at least thousands of years for the 1 µm scale, which practically means that no ²²⁸Ra could make its way out of the grains.

Rama and Moore (1984) showed experimentally that radon can diffuse out from sand grains within days, which was attributed to diffusion via nano-pores. On the other hand, they argued that other, non-gaseous nuclides of the U-Th chains, including radium, could not make it to the intergranular space, which they claimed is due to adsorption in the nano-pores. We believe that while radium diffusion could indeed be negligible in the short-term (days to weeks), it is not necessarily the case for the longer term (e.g., years), in particular when silt and clay are considered (unlike the sand and coarser grain size used in Rama and Moore's paper). Accordingly, we suggest that the actual diffusion follows the nanopore pathway. While diffusion through the ice crystals is probably negligible (D approaches zero), it is commonly believed that the actual diffusion of ions mainly occurs through liquidized (partially melted) ice along grain boundaries (e.g. Jellinek and Chatterjee, 1971; Dash et al. 2006). Hence, we suggest that ²²⁶Ra and ²²⁸Ra diffuse out of the grains via liquid films, where occur, along nano-pores walls.

In Fig. 13, we used effective diffusion coefficients of 10⁻¹³ m² s⁻¹ and distances relevant to silt (20 µm). Simulations were performed using the following equation: Ra_p $/$ Ra${}_{\mathrm{g}}=\mathrm{erfc}\mathit{\left\{}\frac{X}{\mathrm{2}\cdot \sqrt{\mathrm{De}\cdot \frac{t}{p\cdot R}}}\mathit{\right\}}$ (Crank, 1975; Bear, 1972), where Ra_p $/$ Ra_g= activity in pore space relative to that in the grain, x= distance [m], De = effective diffusion coefficient [m² s⁻¹], t= time [s], p= porosity [unitless], R=retardation factor [$=K+\mathrm{1}$, unitless, Krishnaswami et al., 1982].

Molecular diffusion for Ra is assumed similar to that of Ba ($\mathrm{0.85}\times {\mathrm{10}}^{-\mathrm{9}}$ m² s⁻¹ at 25 °C; Vany'sek, 2017), although Ra²⁺ probably has the largest self-diffusion coefficient in water among the alkaline earth ions, owing to its smallest ionic potential (e.g., Yamaguchi et al., 2022). This should reduce by about an order of magnitude at sub-zero temperature due to the significant increase in water viscosity (Stokes-Einstein relation). Since liquid water is probably highly uncommon in the nano-pores, owing to its low water : rock ratio, we used a much lower intrinsic effective diffusion of 10⁻¹³ m² s⁻¹. Diffusion along grain boundaries is considered as diffusion in veins, where porosity =1. We assumed that the liquid in grains is fresh water, therfore the high R's (up to 10⁶, e.g., Kumar et al., 2020) used in Fig. 13.

The results show that radium could escape silt-size grains within years, assuming the partly (slightly) liquidized nano-pore concept. We suggest that this is the way that both ²²⁶Ra and ²²⁸Ra are enriched and produce the high long : short-lived ratios in ADE-17. The observation that high ratios are not observed in ADE-2 (note: long-lived activities are similar to ADE-17, but this is apparently due to the ground ice's higher salinity, therefore enhanceed desorption in ADE-2) is probably due to its short residence time (years). Namely, the salinization had to occur quite recently (the actual timing is practically dependent on the effective diffusion coefficient).

Interestingly, ²²⁸Ra $/$ ²²³Ra and ²²⁶Ra $/$ ²²³Ra show a similar pattern in ADE-17 (compare Fig. 8a and c), despite of the significant difference in their half-lifves. This suggests that in this core, it is the diffusion pathways (i.e. the liquidized fraction of nano-pores), rather than time, that practically limits Ra diffusion out of the grains. As a matter of fact, only a small fraction of the in-grain produced ²²⁶Ra had to diffuse out of the grains in order to produce the high (²²⁶Ra $/$ ^slRa)_AR. With a bulk ²³⁰Th of 4–8 dpm g⁻¹ (Table A3), a dry density of 2.5 g cc⁻¹, porosity of 25 %–50 % (i.e., 50 %–100 % ice content, Table A1) and 50 % emanation (release to pore space), ²²⁶Ra in ground ice should be as high as 5000–30 000 dpm L⁻¹, which is 2–3 orders higher than that observed in the ground ice (Table A2). Similar calculation could be shown for ²²⁸Ra. This probably means that diffusion is quite limited also in ADE-17, probably due to the frozen nature of most nano-pores.

Diffusion time constraints are further illustrated by the ground ice in ADE-1. While (²²⁶Ra $/$ ^slRa)_AR in this core are similar to those in ADE-17, its (²²⁸Ra $/$ ^slRa)_AR are actually very similar to those of ADE-2 (Fig. 8). This probably means that in this location (within 10's meters), diffusion has been slower, therefore ²²⁸Ra could not make it to the pore space. Slower diffusion at that site could be due to slightly different cryogenic conditions (e.g., less liquid water).

Figure 13 Simulations of diffusion from sediment grains into the pore space. $(\mathrm{Ra}{)}_{\mathrm{p}}/(\mathrm{Ra}{)}_{\mathrm{g}}=$ Ra activity in pore space normalized to that in grain. Diffusion assumingly occurs through occasional melted ice along in-grain frozen nano-pores. D_e (effective diffusion) assumed 10⁻¹³ (see discussion in text). Retardation values are 10⁴–10⁶ to cover for fresh water at low T.

5.3 Sources of ground ice and permafrost dynamics

The co-existence of pore fluids (ground ice) with different chemistry and supposedly distinct history (residence time) at the same site and depth (from the surface), and the comprehension that ground ice formation (or regeneration) along the profile of one of the cores (ADE-2) is very young, calls for dynamic visualization of the permafrost at Adventdalen. Fluid intrusion and segregation in permafrost has been commonly documented (e.g., French, 2007). This often involves brine intrusion and the formation of cryopegs (e.g., Colangelo-Lillis et al., 2016; Stephani et al., 2020). Fluid migration could be old (millenia and more, e.g., Gilichinsky et al., 2003; Iwahana et al., 2021), but sometimes is apparently young and even very recent (e.g. < 1 year, Weinstein et al., 2019).

Rotem et al. (2023) suggested that the saline water (and the frozen fresh-saline water interface) found at the top of the epigenetic section in ADE-17, at > 10 m a.s.l., has been preserved due to the immediate formation of permafrost following seawater regression from this site, ca. 9 ka. It is also apparent that some of the water went through water : rock interaction while in the sediment, probably prior to freezing (e.g., high $\mathrm{Na}/\mathrm{Cl}$ and SO₄ $/$ Cl, Fig. 5), possibly when still covered with seawater.

Accordingly, the similar to seawater composition of the ground ice at ADE-2 and its implied short residence time is quite enigmatic. A direct active connection to the nearby fjord, the Adventfjorden (Fig. 1) via sub-permafrost groundwater flow (i.e., seawater circulation) is unlikely, considering the distance from the fjord (∼ 10 km). With basement geology of shales underlain by sandstones (e.g., Ogata et al., 2014) and a relatively low hydraulic gradient (ca. 2 ‰, assuming it is parallel to the topographic gradient), the flow rate should not be much more than 1 m yr⁻¹ (e.g., Hornum et al., 2020), which would result in residence times of several thousand years at the study site. Thus, it seems that salinization and the low (²²⁶Ra $/$ ^slRa)_AR and (²²⁸Ra $/$ ²²³Ra)_AR at ADE-2 are the result of local, within aquifer fluid migration and a recent Ra clock reset (“apparent” residence time), rather than the “real” residence time of intruding water in the aquifer.

The buildup of radium isotope activity in the pore space, whether hosting water or ground ice, depends on several factors, including (1) the mineralogy of aquifer solids, (2) grain size, and (3) pore space : solid ratio. The migration of water from Th-poor to a Th-rich lithology, as well as from high water : rock to a low water : rock ratio environment, should result in an increased supply of radium, therefore possibly in the reset (or a partial reset) of the Ra isotope-ratio clock. Impact on radium release and resultant reset will be even stronger, if the intruding water is more saline than the older one (assuming a non-completely frozen environment), therefore resulting in an enhanced desorption. Accordingly, it is suggested that the ground ice in ADE-2 recently experienced a local intrusion of saline water, which previously resided in a low-Th or a high water : rock ratio aquifer. Such an aquifer could be the Early Cretaceous sandstone Helvetiafjellet Formation, which is located > 70 m beneath the surface at the site (ca. 60 m below the Adventdalen sediment fill, e.g., Ogata et al., 2014), although the possibility of a deeper source (e.g., the Triassic to Early Jurassic fractured sandstone Geerdalen Formation; Mulrooney et al., 2018) could not be excluded. Sandstone, in particular fractured, usually has a high water : rock ratio, as well as a low thorium content, compared with the silt-dominated sediments that comprise the valley infill. Sandstone could also explain the limited water-rock interaction, which is demonstrated by the similarity to seawater composition in the ADE-2 ground ice (Fig. 4). Logs taken at deep boreholes in Adventdalen, albeit closer (5 km) to the sea, indicated that groundwater salinity in the Helvetiafjellet Formation (and deeper) is ≥ 20 mS cm⁻¹ (∼ 40 % seawater conductivity; Braathen et al., 2012), which could also be the case in the basement rocks beneath the ADE site.

The ascent of water from the sub-permafrost zone is demonstrated in Svalbard by artesian discharge, associated with the common pingo structures in the vicinity of our Adventdalen site (e.g., Demidov et al., 2019; Hodson et al., 2020; Rotem et al., 2024). This discharge is controlled either by hydraulic heads from the surrounding mountains or by basal permafrost aggradation (e.g., Hornum et al., 2020). Intra-permafrost intrusion or segregation events were also documented in various environments (e.g., French, 2007; Weinstein et al., 2019), sometimes related to non-frozen cryopeg lenses (e.g., Stephani et al., 2020; Iwahana et al., 2021). Although not a massive ice body, like in the common segregated and intrusive ice (e.g., Humlum et al., 2003; Demidov et al., 2021), it is proposed that the very young (modern) ground ice of ADE-2 was formed by saline fluids that arrived from the underlying sandstone reservoir (Fig. 12), possibly forced by the same hydraulic pressure that drives artesian discharge in the nearby pingos (e.g., Hornum et al., 2020). These fluids have been emplaced (intruded) and mixed with ground ice in the shallower permafrost zone in a fingering style, which resulted in the local heterogeneity at the ADE site.

We note that since ground ice analysis has been performed on large core segments (length > 0.5 m) and considering the very variable salinity (e.g., 2000 to > 20 000 mg Cl L⁻¹ in the epigenetic zone, Fig. 3a), we do not have a good resolution of the actual distribution of the saline ground ice. By this we mean that the high salinity observed could in fact be due to small pockets of even higher salinity, where the fingering intrusive fluids managed to intrude and salinize pore space, as well as to reset its Ra clock. We also have no portrayal of the intrusion pathways, which should be further studied.

5.4 The ²²⁴Ra $/$ ²²⁸Ra enigma

The relatively low daughter-parent (i.e. (²²⁴Ra $/$ ²²⁸Ra)_AR≪ 1) in permafrost ground ice has been noticed in several cases (Weinstein et al., 2019, in Svalbard; Kipp et al., 2025, in Arctic Canada; and Zhang et al., 2025, worldwide). This is remarkably different from the common steady-state ratios of 1–2, mentioned in most groundwater publications (e.g., Krishnaswami et al., 1982; Davidson and Dickson, 1986; Kiro et al., 2015). Interestingly, low ratios also appear in deep, old brines from coastal Israel (Weinstein et al., 2021). In both cases, this was attributed to the adsorption of ²²⁸Th (direct ²²⁴Ra parent, ²²⁸Ra daughter) onto grain surfaces or aquifer rocks, thereby retaining some of the produced ²²⁴Ra on the grains and preventing it from reaching the pore space fluid/ice. This, in turn, required the presence of some non-frozen fluid in the pore space, probably as thin films around the grains, in order to allow for the highly particle-reactive thorium to adsorb onto grain surfaces. This is in agreement with the observed partially frozen nature of the permafrost in Svalbard and elsewhere (e.g., Keating et al., 2018; Weinstein et al., 2019).

Nevertheless, the results of this study (Fig. 9) suggest that this is not an across-the-board observation, and that some of the samples show ratios within the “expected” range of 1–2, although this could be affected by enhanced desorption of high (²²⁴Ra $/$ ²²⁸Ra)_AR during sample processing in the lab. High ratios are particularly observed in the deep, epigenetic samples of ADE-2, which could be related to the higher salinity (hence, less adsorption or enhanced desorption) in these samples. This calls for further and systematic research of ²²⁴Ra $/$ ²²⁸Ra in old brines and permafrost.

6 Summary and Conclusions

In this paper, we used Ra and Th isotopes, together with general chemistry, to get an insight into the ground ice history and fluid migration in the continuous permafrost of a Svalbard valley. While the dominant control on the contribution of short-lived Ra isotopes into the pore space is the recoil of Th-parent nuclides residing on the soil grain surfaces, the supply of the long-lived ²²⁶Ra and ²²⁸Ra is also controlled by diffusion out of the soil grains, probably mainly through slightly liquidized nano-pores. The actual activities of all isotopes are also controlled by adsorption, which in turn is dependent on the existence of liquid water in the pore space under the discussed cryotic conditions. The relatively low (²²⁶Ra $/$ ^slRa)_AR and (²²⁸Ra $/$ ²²³Ra)_AR in the ground ice of one core (ADE-2) and its equilibrium with grain surface ratios of thorium isotope parents, while the much higher Ra isotope ratios in ADE-17, suggest that ground ice in the first is younger, or more precisely – has been recently regenerated, probably as a result of fluid intrusion and Ra isotope clock reset. This fluid was significantly more saline than the original one, as reflected in the higher salinity of this core's ground ice, compared with the other cores. Salinization goes almost all the way to the surface, affecting Late Holocene sediments, also demonstrating its recent history. The similarity to seawater composition of this young ground ice suggests that the intruding fluids were less subjected to WRI, which is probably due to the characteristic higher water : rock ratio of its source aquifer, most likely an underlying sandstone formation. On the other hand, the different from seawater composition in ADE-17 suggests that the water had a significantly longer history in the valley-fill sediments prior to freezing, which is in accordance with their much longer residence time.

This study demonstrates the heterogeneity of the saline permafrost and highlights its complex history and liability to fluid migration, which is probably also the case in other saline permafrost terrains. While the suggested intrusion is probably the result of basement-originated fluid intrusion, the salinization process could also be affected by the current cryotic conditions of the shallow permafrost. This further highlights saline permafrost's vulnerability to the ongoing climate change and warming. As Arctic temperatures continue to rise, the degradation of saline permafrost may trigger or allow subsurface fluid mobilization, posing significant risks to infrastructure and accelerating the release of stored carbon to the atmosphere.

Appendix A

Table A1 Chemistry and ice content^* of ground ice in the ADE cores, East Adventdalen.

^* Ice content = ice/dry weight. NA: Not analyzed.

Table A2 Radium isotope activities and ratios in ground ice, ADE cores.

NA = Not Analyzed, LE = Low Ra Efficiency on fibers (< 50 %), and BDL = Below Detection Limit.

Table A3 Thorium isotope activities and ²²⁷Ac in sediments from ADE-17.

PF = permafrost, AL = Active Layer, Syn and Epi = Syngenetic and Epigenetic permafrost respectively, exch = exchangeable fraction. ^* Errors of all exchangeable fraction of ²³²Th and ²³⁰Th are ≪ 0.001.