Monitoring of glacier outflow and the development of naledi was done by setting up time-lapse cameras (LTL-6210MC Plus Scouting Cameras) across the glacier forefield and monitored between 2022 and 2024 (Fig. 2 and Table S1 in the Supplement). Cameras were set up at six different locations and orientations to capture changes in the main outflow channel of Isunnguata Sermia and to review active upwellings at the time of installation. Photographs were taken at 11:00 am, 12:00 pm, 01:00 pm and 02:00 pm (local time) each day. Full descriptions of camera installations and observations can be found in Table S1.

Four field campaigns were conducted to collect a representative sample set of ice and water samples across the Isunnguata Sermia forefield over multiple years both pre- and post-melt season (Fig. 2). We focused on three different sample types (i) naled ice (dome shaped structures, formed over winter likely during low discharge), (ii) naled waters upwelling in the outwash plain (capturing active water sources feeding naled ice over the winter months) and (iii) accreted ice (sediment-laden ice structures associated with active upwellings in the summer months) (Fig. 2).

Over the course of four field campaigns (October 2022, April 2023, September 2023 and March–April 2024), we sampled 12 naledi and 4 accreted ice structures (Figs. 2a and S1, S2 in the Supplement). Ice cores were extracted with a motorised SIPRE Auger to a maximum depth of 2 m. Cores were logged immediately, then sectioned by changes in ice facies and transported in individual sterile Whirl-Pak bags (Fig. 3). Core sections were then melted in a gently heated water bath in the field (less than 4 h after sampling) and filtered using Whatman 0.45 µm cellulose membrane filters into centrifuge tubes for major ion analysis. All samples were refrigerated and stored at 3–4 °C until analysis. Four of the naled ice cores terminated in water sources. Cores at the other seven sites did not reach water and were limited to depths between 8 and 139 cm, often due to thick gravel deposits which rendered further drilling impossible.

Of the cores that reached active water sources, naled I8 was located at the margin of the main meltwater channel and the three others (I2, I3 and I10) in an adjacent lateral meltwater stream (Fig. 2). In each case, samples of the naled upwelling waters were collected in sterile 50 mL centrifuge tubes and filtered using Whatman 0.45 µm nylon membrane filters. Accreted ice cores were sampled around the glacier front and within the main active frozen meltwater channel area in October 2022 (n=4, Figs. 2a, 4 and S3 in the Supplement). Unlike the naled ice core sites, none of the accreted ice cores reached free-flowing waters.

Contextual samples of glacier ice (n=2) and supraglacial meltwater (n=7) were also collected to represent end-member water types. Glacier ice was sampled from the upper ice margin of Isunnguata Sermia in April 2024 using a chisel and hammer, with the upper 10 cm layer of ice discarded to remove the outer layer of weathered ice (Fig. 2a). Supraglacial water was sampled from the ice margin and atop the Isunnguata Sermia glacier snout in 50 mL centrifuge tubes.

Concentrations of major anions in water and melted ice samples were analysed in triplicate on a Dionex Integrion high-pressure ion chromatography (IC) system at Northumbria University using 5 mL of sample. Standard solutions of fluoride (F-), chloride (Cl-), sulphate (SO42-), and nitrate (NO3-) were prepared using 18.2 MΩ ultrapure water to obtain standard curves. Concentrations of major cations were analysed in triplicate on an Optima 8000 inductively coupled plasma optical emission spectrometer (ICP-OES) at Northumbria University using 7 mL of sample. Standard solutions of calcium (Ca), potassium (K), magnesium (Mg), and sodium (Na) were prepared using ultrapure water to obtain standard curves. Limits of detection (LOD) for both instruments was determined by standard regression on the intercept of the calibration curve and measured 0.02 mgL-1. Analytical precision was assessed as the relative standard deviation (RSD%) of triplicate measurements for each sample. RSD was below 3 % for all analytes and across all run days for the ICP-OES and the IC, indicating good analytical reproducibility. HCO3- was then calculated using the charge balance error (CBE) method, where ϵ+ and ϵ- are total cation and anion equivalents. Uncertainty in HCO3- concentrations is derived from uncertainties of all measured cation and anion concentrations.

Aliquots of samples were analysed for their hydrogen and oxygen isotopic composition using a Picarro L2130-i cavity ringdown spectrometer at Northumbria University. Four calibration standards were measured at the beginning of each run (including in-house laboratory standards supplied and calibrated by the University of Bergen (“EVAP-2”, “FINSE”, “GLW”) and international standard SLAP2 (Standard Light Antarctic Precipitation 2). Separate aliquots of reference water “Medium Natural Water” or “Greenland Water” (purchased from Elemental Microanalysis Ltd.) were analysed every 10 to 15 samples to monitor and correct instrument drift. Results were corrected for memory and drift and normalised to the VSMOW-SLAP (Vienna Standard Mean Ocean Water – Standard Light Antarctic Precipitation) scale, using the FARLAB Liquid Water Isotope Measurement Processor (FLIIMP) software package (Sodemann et al., 2023). No humidity correction was applied (following Weng et al., 2020). The drift standards (as well as one aliquot of VSMOW2 per run) were used as data quality monitors. Analytical error is equal to or better than 0.4 ‰ for δD and 0.07 ‰ for δ18O (SE); overall reproducibility based on monitor standards is equal to or better than 1.0 ‰ for δD, and 0.38 ‰ for δ18O (SD) (Table S2 in the Supplement). Individual “freezing packages”, which are discrete isotopic units within the naled ice body representing a temporally distinct episode of water upwelling and subsequent freezing at the ice surface, were also identified by calculating freezing curves for each ice core. In the context of refreezing Greenland Ice Sheet waters, a δD:δ18O slope of around 6 would be expected. The composition of the source water can be calculated from the two-isotope slope (Souchez and Jouzel, 1984), allowing us to determine the isotopic composition of the source water where a freezing slope is identified.

A SuperSting R8 eight-channel earth resistivity metre (Advanced Geosciences, Inc.) was used to investigate subsurface conditions and potential water transport routes through the forefield of Isunnguata Sermia (Fig. 2) in April 2024. The frozen ground had to be prepared for electrode stake deployment by thawing small areas to a depth of ∼ 30 cm using boiled water, heated from locally sourced snowmelt. A mixture of locally sourced glacial clay and salt or isotonic powder (Gatorade™) was poured over the thawed ground to secure each electrode stake and ensure sufficient ground coupling once the electrodes froze in. Fifty-six electrodes were installed in a forward and reverse dipole-dipole configuration along profile A with the same configuration along profile B (Fig. 2) for 47 electrodes. The maximum source current was set to 50 mA for both profiles, and electrode contact resistance was monitored throughout. All data were processed and inverted using ResIPy v3.6.3 (Blanchy et al., 2020) (see Supplement).

Automatic cameras captured active naled and accreted ice formation as well as changes in winter outflow over the foreland of Isunnguata Sermia from 2022 to 2024. In 2023, Camera 5, orientated on the proglacial outwash plain facing the ice margin, captured the formation of winter-accumulating naled ice and an outflow of yellow-coloured water on the 22 October when the rest of the meltwater channel was frozen (Fig. 5). The source of flow from this yellow water originate from a location off-camera. The distinct colouration of this water can be seen freezing over the channel and naled structure over 3 d, before being covered by fresh snow on 28 October. These accumulation patterns suggest that similar ice layering may be expected in the naled icings.

A glacial outburst flood from Isunnguata Sermia was also detected between September and early October 2023 (see Livingstone et al. 2026). By analysing footage from two separate automatic cameras and orientations (see Supplement), a gradual increase in water levels of the main meltwater channel was observed before a vigorous outburst flood event on 2 October (Fig. 4a and b). We also observed the sudden build-up of sediment-laden accreted ice around the upwelling during the outburst event (Fig. 6). Much of the resulting accreted ice structure persisted up to 6 months later, preserved by the cold winter temperatures, though it appeared to have eroded, suggesting some melt wastage had occurred.

Ice cores revealed distinct ice facies, which we group into four classifications: light ice, opaque ice, yellow-coloured ice, and sediment-rich ice (Fig. 3). Some ice has either horizontal or vertical stripes – which are more visible in the “light ice” and “opaque ice”. Light ice and opaque ice packages do not have sediment inclusions or discolouration, with the light ice mostly translucent and the opaque white or “clouded”, and were combined for geochemical analysis as “clear ice” for simplicity. Yellow-coloured ice was distinctly yellow in colour and appeared as discrete layers within the naled cores. Five ice cores recorded these yellow-coloured ice layers, which ranged in thickness from 10–25 cm, at various depths (Figs. 7b, e and 8a, c, and e). Yellow-coloured ice was not found in any of the accreted ice cores. Sediment-rich ice had clear sediment inclusions that accounted for > 50 % of the core section. Sediment layers were present in nearly all core samples (except I2, I7 and I8), with more pronounced sediment-rich layers (between 10–40 cm thick) recorded in I1, I3, I5 and I10 (Figs. 7 and 8). These sediment-rich layers only appeared as discrete layers within the naled cores, but they account for almost the full core length in the accreted ice cores (Figs. 3 and 4). Facies logs for each ice core and their resultant geochemical profiles are provided in Figs. 7–9.

Accreted ice cores were typically multi-metre in length and consisted of heavily sediment-laden ice (Fig. 4), which appeared to be a mix of frazil ice and silt to fine sand sediments. Cores ranged in length from ∼ 25 to 240 cm (Fig. 9). The longest core, A4, had no visible layers, but the full core was sediment-laden with clay to fine sand-sized particles (Fig. 4d). Bands of opaque ice, less than 10 cm thick, were recorded at various intervals throughout cores A1, A2 and A3.

Proportions of major ions and total dissolved solids (TDS) are shown for each core in Figs. 7–9. The naled ice cores and their sub-naled waters display a range of TDS concentrations, broadly spanning from very dilute (less than 1 mgL-1) to highly concentrated (up to 3000 mgL-1) depending on the core and depth. Across most cores, HCO3- is the dominant anion, while Ca2+ is consistently the most abundant cation. In general, we observe that ice layers with low TDS, such as naled I1 and lower I7 (Figs. 7a and 8b), are typically SO42- poor (though sometimes NO3- enriched). Whereas very concentrated TDS such as the naled waters in I2 and I3 (Fig. 7b and c) are more often SO42- enriched and have higher TDS concentrations than the overlying ice.

Naled cores I1, I7, I8, I11 and I12 were all located in the main proglacial river channel. Naled I1 was located closest to the glacier and extracted from beneath a layer of mud, whereas all other naled icings were collected from clean ice surfaces. All cores sampled from the main channel lack yellow ice layers apart from I8, which had a distinct yellow layer at the very surface and yellow water at the base of the core (Fig. 8). These main channel cores generally exhibit moderate TDS concentrations, except I8, which reaches up to 132 mgL-1 at its base and 3098 mgL-1 in the water. HCO3- dominance is consistent across all main channel cores, SO42- contributions are relatively low compared to some side channel cores. NO3- is slightly more elevated in I8 than in the other main channel cores. I12 was sampled as a newly emerging naled with only a thin layer of domed ice extracted after it began forming in the main channel overnight in March 2024. Here, the very surface of the naled ice was sampled and showed moderate TDS concentrations of 24–40 mgL-1 with an almost even proportion of HCO3-and SO42-.

Side channel cores, I2, I3, I4, I5, I6, I9 and I10 show greater variability in both TDS concentrations (2–500 mgL-1in ice and 2100 mgL-1in naledwaters) and ion proportions. While HCO3- and Ca2+ dominate, SO42- contributes more proportionally across several cores, most notably the naled water in I2, I6, and I10. Core I10 is particularly distinct, with SO42- comprising over 40 % of total ions throughout and TDS values reaching up to 548 mgL-1 in the yellow ice layers. Naled waters sampled beneath I2 and I3 also show higher TDS concentrations of 2030 and 1032 mgL-1, respectively.

Accreted ice cores (A1–A4) are more uniform in their geochemical signature both within single cores and across cores (Fig. 9c–f). All cores have TDS concentrations between 13 and 39 mgL-1 (Fig. 9c–f). HCO3- and Ca2+ consistently dominate, comprising 50 %–80 % of anion and cation proportions, respectively. The most variability is seen in accreted core A4, where SO4- dominates the anions in some of the core sections. Accreted cores also have marginally higher proportions of K+ compared to the naled ice cores.

Isotopic compositions (δ18O and δD) reveal accreted ice samples have slightly higher δ18O and δD values (Fig. 10a; -21 ± 1.7 ‰) compared to naled ice (Fig. 10a; -24 ± 2.3 ‰). Surface glacier ice plots separately from all naled samples with the lowest δ18O value (-33.5 ‰), indicating the naled ice is distinctly different in water source and/or formation mechanism than surface glacier ice (Fig. 10a). Deuterium excess (D-excess = δ2H-8⋅δ18O) values range from approximately 3 ‰ to 32 ‰ with a general trend of decreasing deuterium excess with increasing δ18O values (Fig. 11).

Changed in isotopic composition were also observed down-cores, revealing potential freezing mechanisms. Within naled core I8 (Fig. 10b) δ18O values generally decrease downcore (the lowest values detected at 132–146 cm of -29.5 ‰) whereas deuterium excess increases from 9.8 ‰ at the top of the core to 25 ‰ at the base due to increasing fractionation processes downcore (Fig. 10b). However, a distinct but slight increase in δ18O and decrease in deuterium excess was noted at 104–108 cm. When plotted on a freezing curve, naled ice samples I5, I8, I10 and I11 all display R2 values of > 0.9, suggesting sequential freezing downcore (Fig. 11). However, some core sections (in I2, I3, I4, I6, I7 and I9) do plot away from the freezing line with R2 values of 0.07–0.93 (Fig. S4 in the Supplement). These sections represent discrete intervals of an increase in δ18O and decrease in deuterium excess (also seen in Fig. 10b), suggesting the start of a new freezing package.

Electrical Resistivity Tomography (ERT) surveys were conducted < 400 m from the terminus of Isunnguata Sermia, in an area where gravel-rich sediments dominate (Fig. 2). The resulting ERT images (Fig. 12) reveal considerable spatial heterogeneity in subsurface resistivity, spanning from log10ρ ∼ 0.83 to > 8.1. In both ERT profiles, the uppermost layer (∼ 1–5 m below the surface) is dominated by high resistivity values (log10ρ > 10.5), often indicative of dry and (at least partially) frozen material (Kneisel et al., 2014). As the surface is dominated by unsaturated gravel-rich glacial outwash material and compacted till, we expect this uppermost subsurface layer is similar in character.

Beneath the high resistivity layer, an intermediate zone (approximately 5–15 m below ground level) exhibits markedly lower resistivities (log10ρ ∼ 2–4), varying across both profiles (yellow/white/blue colours in Fig. 12a). Bound by higher resistivity above and below, this zone likely represents a saturated sediment layer composed of sands/gravel and/or finer material or fractured bedrock (Palacky, 1987; Hauck and Kneisel, 2008; Reynolds 2011). Each of these materials can host groundwater, ice or subglacial meltwater. Localised zones of even lower resistivity (log10ρ ∼ 1–2) are observed within this intermediate zone in both profiles (Fig. 12a, insets). Their geometry is suggestive of discrete pockets or conduits of moisture, potentially reflecting melting permafrost or glacial outflow pathways (e.g., Hauck and Kneisel, 2008). In both profiles, resistivity values abruptly return to log10ρ ∼ 10.5 and then log10ρ ∼ 10.6 Ωm beneath the intermediate zone. We interpret this to be drier sediments and/or frozen ground/ice (Kneisel et al., 2014), which would be likely in the forefield of Isunnguata Sermia in winter.

Time-lapse camera footage, along with geochemical analyses of naledi and geophysical investigations of the glacial foreland at Isunnguata Sermia have allowed us to capture naledi formation mechanisms in unprecedented detail. Our data show icings forming under both confined and unconfined water input conditions.

In the confined condition, ice layers build up into dome-like structures, fed by the upwelling of waters from below (Figs. 5 and 13). The waters likely originate from subglacial sources and are transported through subsurface conduits and/or saturated till (Fig. 12). In April 2024, naled ice domes were observed to have started forming overnight suggesting that they may form quickly (Fig. 5e). Following formation of an ice layer that restricts water flow, resultant build-up of water pressure beneath the naled can fracture the ice surface (e.g. Fig. 5c). The method of formation we describe here is akin to the formation mechanism described by Wadham et al. (2000) and Stachnik et al. (2016) (outlined in Fig. 1d–f).

Not all of the naledi studied were exclusively fed from waters below, and our results show that more than one formation mechanism exists with the incorporation of unconfined waters from, for example, englacial flow or precipitation. Automatic camera footage in Fig. 5 demonstrates naledi being built from these unconfined waters, as ice layers accumulate from overland water flow such as supra- and englacial melt and from local precipitation such as the snow event captured in the automatic cameras (Fig. 13). The cameras also captured the incorporation of yellow coloured waters originating off-camera. Due to the colour similarity with the waters collected below naled structures, we believe this overland flow originated from a large upwelling of overwinter waters at a nearby naled location. These camera-derived observations, combined with geochemical analysis, help to explain some of the unique ice facies we observed in the naled ice cores.

The waters with relatively high concentrations of TDS at the base of the naled cores are interpreted to reflect prolonged water storage in the subglacial system (at least over the winter months) - where chemical equilibration and mineral dissolution result in the highest TDS concentrations across our study site (e.g., Gokhman et al., 1987; Hodgkins et al., 2004; Stachnik et al., 2016; Yde et al., 2012). Layers of similar colours and geochemical signatures are also found within the naled cores such as I2, I8 and I10, suggestive of an eruptive process injecting these TDS-rich yellow coloured waters into the ice structure. Our geophysical analysis of the substrate supports these ideas, as low resistivity TMS suggests a saturated sediment layer could be fed by subglacial meltwaters and/or groundwater. The geophysical survey also identifies a potential conduit for water flow (∼ 5 m in diameter), which could efficiently transport subsurface waters over wintertime, contributing to naled ice structures when air temperatures limit overland flow.

The clear naled ice with low solute concentrations suggests waters have had limited mineral surface contact for reactions to occur (such as supra- or englacial melt/storage and/or overwinter precipitation). An alternative explanation is that these layers (most often found at the top of the naled cores) are partially frozen (or partially wasted) ice/slush, which just retains some of the solutes available – as the rest of the solutes have either run off completely (if unconfined), or moved downwards, where they have become concentrated in the bottom layer of the freezing package. For example, in I8 we see very low TDS values (0.8–1.2 mgL-1) in the middle of the core, followed by giher concentrations of 15–141 mgL-1 further down the core, within the most isotopically enriched layers.

The downcore chemical and isotopic variation in naled ice can be explained, at least in part, by freezing processes. During freezing, isotopically enriched water freezes first, with a stronger effect on 2H than 18O. Remaining, freezing water will fractionate on a freezing slope unlike that of meteoric water, controlled by the initial composition of the water and the fractionation factors of freezing (Jouzel and Souchez, 1982; Souchez and Jouzel, 1984). In a package of ice that froze in a single outburst event, we expect the heaviest isotopes on the top and lighter isotopes downcore. We also expect the concentration of dissolved solutes to significantly increase at the bottom of a freezing package, from the effects of progressive freeze concentration of dissolved solutes in the source water (Halde, 1980).

The stable water isotopes reveal sequential freezing downcore with discrete freezing packages resulting from multiple upwelling events. These upwellings freeze on at the base of existing ice, pushing up into the ice, and creating domed structures over wintertime. However, we also observe that in the top sections of some samples of longer cores (I2 and I7) and the shorter cores below 50 cm (I3, I4, I6 and I9) do not plot on this freezing curve. This indicates a different freezing sequence, potentially forming from a top-down process.

As water discharge is reduced over the winter months, and less glacial sediment is transported, the resulting naled ice is less sediment-laden than summer accreted ice, which forms under faster water flow conditions (Fig. 13). We therefore conclude that the naledi on the forefield of Isunnguata Sermia form through a combination of unconfined and confined water sources, where subsurface waters emerge through the substrate overlain by ice freeze-on below, whereas overland flows and precipitation accumulate from above (Fig. 13).

The formation of accreted ice was captured on automatic cameras in the late melt season, September–October 2023, after a large upwelling event (Figs. 4 and 6). Based on the speed of their formation and supported by our geochemical analyses, we propose that they form where supercooled meltwater is evacuated into the glacier forefield and quickly freezes creating structures composed of a mix of frazil ice and silt to fine sand sediments (Fig. 14).

The accreted ice samples exhibit a moderately uniform geochemical signature throughout, where the cores are dominated by sediment-rich ice layers. This uniformity between sites and within each core suggests consistent water sources, potentially derived from single or time-constrained outflow events. These cores are also geochemically similar to previously characterised summer outflow waters in other studies (i.e., Graly et al., 2014, 2017). Preserved by cold air temperatures, the ice accretions formed by the late melt season persist through the winter months, although gradual wastage is evidenced by deflation and erosion of the ice surface towards the end of the winter (due to discrete melt events), where melting leaves behind large sediment accumulations. Due to our focus on overwinter ice accumulations, we were not present in the summer months to observe the complete wastage of accreted ice; however, our repeat field campaigns allow us to note that accreted ice structures form in almost the exact same location each year. This observation indicates that the subsurface hydrological system must be stable over interannual timescales.

The subglacial hydrological system in this region drives fast-moving, highly pressurised subglacial waters into the proglacial field, generating large upwellings several metres in height during summer months (Graly et al., 2017). This water is presumably routed through the deep subglacial trough beneath Isunnguata Sermia (Lindbäck et al., 2015), where water will accumulate and interact with sediments and bedrock, before being driven up the reverse bed slope by high water pressure (when the subglacial hydrological system reaches capacity) (Cook et al., 2006; Lindbäck et al., 2014). The subsurface here is likely to be very high-porosity sand/gravel deposited during the Holocene ice margin retreat, which favours “groundwater” flow and not channelised flow. These highly pressurised supercooled waters then freeze into large icing structures close to minimal meltwater output zones at the glacier terminus (Figs. 6 and 14). Currently, the growth and wastage rates of this accreted ice formation are unknown, though we observed that sporadic hours- to days-long upwelling events contribute to the majority of the ice formation (Fig. 6a–c). We hypothesise, based on the results presented here that the formation timescale is likely fast with sudden upwellings of high velocity, highly pressurised waters erupt into the glacial forefield (Fig. 14). Where these structures are formed from single outburst events in the late summer season, they may also incorporate waters from sub-, supra- and ice-marginal lakes where their outbursts are the primary source of the upwelling (e.g., Livingstone et al., 2026).

Previous theories of naled formation have adapted either a top-down or bottom-up approach (Fig. 1). One approach details growth from waters below icing structures, with liquid water conduits within naled ice providing transport pathways insulated from the cold surface (Fig. 1f) (Stachnik et al., 2016; Wadham et al., 2000). The other approach describes accumulations from above where the cold, winter air temperature acts as a heat sink, freezing any waters or precipitation that accumulate from above (Fig. 1a–c) (Schohl and Ettema, 1986; Yde et al., 2012). Our observations and results support both of these existing theories. Winter-forming naledi at Isunnguata Sermia form through a combination of top-down and bottom-up freezing processes, whilst summer accreted ice structures form by freeze-on of higher velocity, supercooled waters erupting onto the glacier foreland.

These systems offer insights into hydrological systems in Greenland that are difficult to observe by other means. They also show how water routing can vary spatially and temporally through the melt season. Geochemical differences in ice samples also raise important questions for solute and nutrient release from the Greenland Ice Sheet. With pronounced peaks in TDS detected in the overwinter naled waters, year-round analyses of glacial waters are critical for capturing the full extent of foreland water flow and the impact of these waters on downstream ecosystems, particularly at the release of these solute-rich waters during the initial onset of the melt season. The seasonal changes in geochemistry we observe suggest that solute export in winter may be under-represented in existing literature, which should be considered for a thorough understanding of biogeochemical processes and wider ecosystem effects downstream (Hawkings et al., 2026).