We recovered three firn cores from Eclipse Icefield between 2 and 4 June 2023 (Fig. c). Cores B501 and B502 were drilled from the surface to depths of 16.2 and 10.9 m, respectively, at 60.844° N, 139.851° W, 0.6 m apart from each other, to assess the small-scale spatial variability of near-surface melt features. Core B201 was drilled from the surface to 14.7 m depth at 60.835° N, 139.830° W (at a similar elevation ∼ 1.5 km to the southeast). Cores B501 and B502 were drilled in one night over the course of 5 h; core B201 was drilled the following night. A summary of core recovery information is given in Table .

We collected ground-penetrating radar (GPR) data at frequencies of 900 and 5 MHz using two different systems to image the near subsurface (∼ 20 m depth) and full ice thickness (up to 700 m depth), respectively. We collected very high-frequency (VHF) data on 2 June 2023 with a GSSI system using a shielded 900 MHz center frequency antenna (model 3101) and an SIR4000 control unit paired with a Garmin GPSMap78 (horizontal and vertical accuracy of ±3 m). We collected data at 2048 samples per scan and 24 scans per second with a range of 240 ns. We collected high-frequency (HF) data on 4 June 2023 with a Blue Systems Integrated (BSI) 5 MHz antenna, with 30 m separation between the transmitter and receiver. We ski-towed both GPR systems at a rate of ∼ 1 ms-1 to obtain profiles of stratigraphy between core sites B2 and B5 (Fig. c). Radar data were processed in ImpDAR using standard processing techniques, including clipping stationary periods, constant trace spacing, bandpass filtering, and normal move-out analysis to correct for antenna separation. HF data were also migrated using the SeisUnix sumigtk routine to correct for hyperbolic reflections from basal topography. Depth-variable density profiles obtained from shallow (2023 B501) and deep (2002) cores were used to calculate variable, permittivity-dependent radar-wave velocities to determine the depth of reflected layers. The LSS was semiautomatically picked across the VHF profile; a full description of this picking can be found in .

We model the firn pack evolution over a period of 20 years (2013–2033) at Eclipse using the Community Firn Model (CFM), an open-source, modular model framework coded in Python 3 and available for download on GitHub . We used the CFM to simulate firn density and temperature evolution as well as meltwater retention and refreezing. We parameterize firn density evolution using the densification scheme from and a time-invariant surface density derived from sensitivity testing (Table ). The densification equation from was calibrated for use in Greenland using firn data from sites spanning a range of climates, and it is reasonable to assume that the climate at Eclipse is similar to some locations in Greenland. Any firn densification model is possibly subject to tuning biases (e.g., Kuipers Munneke model was tuned using RACMO climate data, so biases in RACMO will affect the model tuning). We do not expect that using a different firn densification equation would change our results and conclusions. The CFM uses a bucket scheme to simulate meltwater percolation, and it uses an enthalpy-based heat transfer scheme to simulate heat diffusion in the presence of phase changes. We use a parameterization for thermal conductivity from and a Neumann boundary condition for the heat equation at the bottom of the 50 m deep firn column. We force the model with air temperature data from a weather station near the ice divide between the Kaskawulsh and Hubbard glaciers (30 km away, “Divide AWS”; Fig. b), surface melt calculated from air temperatures using a simple degree day model, and a mean annual accumulation rate of 1.4 m w.e. a-1 with monthly scalars applied (Fig. ). We calculate these scalars using an in situ snow accumulation record from Divide. We use 4 years of complete coverage (2004, 2005, 2006, 2008) to compute each month's mean fractional contribution to annual accumulation. When incorporated into the CFM, we distribute each month's portion of the annual 1.4 m w.e. evenly across the days of the month. We elevation-correct Divide AWS air temperatures to Eclipse (∼ 400 m higher) using a lapse rate of -3.98 °C km-1 following .

We spin the model up from ∼ 1983–2013 (exact spin-up time varies slightly among model runs as it is the time required to refresh the entire firn column and therefore dependent on densification rate and surface melt) using monthly averaged downscaled North American Regional Reanalysis (NARR) air temperatures for Eclipse from 1983 to 2013 . We also test the model sensitivity to three other spin-up schemes using different air temperature datasets: (1) elevation-adjusted Divide AWS data from 2013 to 2024 repeated for the duration of the spin-up, (2) synthetic climate data selected from a Gaussian distribution of temperatures based on elevation-adjusted 2013–2024 Divide AWS data, and (3) like spin-up scheme (2) but with a historical 0.024 °C a⁻¹ rate of temperature change applied for the duration of the spin-up such that the mean annual temperature at the start of the main model run is consistent with elevation-adjusted 2013 Divide AWS data. All model spin-ups are forced with a mean annual accumulation rate of 1.4 m w.e. a-1. For our spin-up sensitivity tests, we test each spin-up scheme with 15 different degree day factor (DDF) values used to estimate surface melt from air temperatures and 20 different surface density values (Table ). We select a representative pairing (DDF =6.2, ρ=450 kg m-3) from all the combinations of DDF and surface density values that produce no liquid water down to 14 m depth in the firn in both spring 2016 and spring 2023 (consistent with firn cores and GPR showing no evidence of liquid water at those times) and a firn temperature between -2 and -4 °C at 14 m depth in spring 2023 (consistent with 2023 borehole temperature measurements). We refer to this selected model as our “reference model”. Our exploration of model sensitivity to spin-up scheme, DDF, and surface density values is detailed in Appendix .

We run the model with approximately 1 d time steps. After spin-up (∼ 1983 to 2013), we apply the elevation-adjusted Divide AWS temperature data as forcing from 2013–2024 and then run the CFM under a suite of climate scenarios from 2024–2033, including (a) continuation of current climate, (b) 0.1 °C cooling, (c) 0.1 °C warming, (d) 0.2 °C warming, (e) 0.5 °C warming, and (f) 1 °C warming by 2033. We generate synthetic air temperatures for all 2024–2033 climate scenarios by computing the mean and standard deviation of the 2013–2024 Divide AWS daily mean temperature values (computed from hourly measurements) for each day of year. We then assign synthetic daily air temperatures randomly drawn from Gaussian distributions described by these means and standard deviations. Temperatures for scenario (a) are drawn from the distribution described by the mean and standard deviation of 2013–2024 data. Temperatures for scenarios (b–f) are drawn from distributions with a prescribed rate of change applied to the mean such that the specified degree of warming or cooling for each scenario is reached by 31 December 2033. For each climate scenario, we run the model 50 times and calculate the percentage of model runs that produce temperate firn at 15 m depth by 2033.

To examine the conditions associated with the production of year-round temperate firn at 15 m depth, we quantify the mean winter temperature, melt season start, melt season end, melt season length, and total PDDs each year. We also quantify the number of individual melt events and each one's magnitude. We define melt events as any period over which the snow surface is continuously melting without refreeze.

To contextualize the firn conditions at Eclipse, we compare our results with data from two neighboring sites near the ice divide between the Kaskawulsh and Hubbard glaciers: Icefield Divide Camp (60.68° N, 139.78° W; 2603 m a.s.l.), which we refer to as “Icefield Camp” to avoid confusion with reference to the broader Kaskawulsh/Hubbard Divide area, and the upper Kaskawulsh Glacier (60.78° N, 139.63° W; 2,640 m a.s.l.), which we refer to as “Kaskawulsh”(Fig. b). On 4 June 2018, VHF (400 MHz) and HF (5 MHz) ground-penetrating radar data were collected at Icefield Camp; for a full description of GPR deployment and processing, see McConnell (2019). In the same year, two firn cores were recovered from the upper northern arm of Kaskawulsh Glacier by . The cores were each 8 cm in diameter and drilled 60 cm apart from each other to depths of 36 m (Core 1) and 21 m (Core 2); liquid water was encountered at 34.5 m below the surface in Core 1 . Stratigraphy of both cores and density of Core 1 were measured in the field; densities for a subset of samples from Core 2 were measured after transporting the core from the field to nearby Kluane Lake Research Station. For a complete description of Kaskawulsh core recovery and analysis, see .

We also contextualize the Eclipse, Icefield Camp, and Kaskawulsh sites within the St. Elias Range by producing a regional hypsometric curve for the region's glacierized terrain. We derive this curve from the ArcticDEM digital elevation model at 10 m resolution and highlight areas ranging from 2600 to 3000 m a.s.l., which cover the elevation band between our sites of interest.

The St. Elias Range has been an area of glaciological study for decades. To investigate how firn at Eclipse has changed over time, we compare our 2023 data to other measurements made at Eclipse since 2002. i.

We compare 2023 GPR data to both VHF (400 MHz) and HF (10 MHz) GPR data from 2016, described in McConnell (2019).

The stratigraphy of all three 2023 cores shows ice layers, ice lenses, and melt-affected firn throughout the core; however, variability among individual melt features is high among all three cores despite the proximity (<1 m) of cores B501 and B502 (Fig. ). The first ∼ 4 m of all three cores comprises the seasonal snowpack, with the last summer surface (LSS) appearing as a crusty layer of melt-affected firn. Unless otherwise specified, all following stratigraphic depths are reported relative to the LSS, which we designate as below the LSS or “BLSS”.

In core B501, most of the top meter of firn below the LSS is coarse-grained and contains several hairline to 1.0 cm ice layers (Fig. ). Approximately 1 m BLSS, the firn transitions from coarse-grained and very melt-affected to fine-grained with fewer melt features. It remains fine-grained until 3.40 m BLSS, where there is a sharp transition from fine- to coarse-grained firn. The largest sections of fine-grained firn occur from 2.15 to 3.40 m BLSS (1.25 m thick) and from 7.54 to 8.83 m BLSS (1.29 m thick). The rest of the core contains clusters of ice layers and lenses or is otherwise melt-affected based on texture and appearance. The thickest ice layer in the core begins at 5.46 m BLSS and is 12.0 cm thick. Core B501 has a total firn ice content of 5 % by volume and 40 % of its length below the LSS (11.34 m) is visibly metamorphized and/or melt-altered (ice layers, coarse-grained, and melt-affected sections as shown in Fig. ).

In core B502, the first meter of firn below the LSS lacks the large crystals observed in core B501 but does contain numerous hairline to 1.0 cm thick ice layers (Fig. ). The largest sections of fine-grained firn occur from 1.80 to 3.03 m BLSS (1.23 m thick) and from 3.72 to 4.82 m BLSS (1.10 m thick). The thickest ice layer in core B502 begins at 1.41 m BLSS and is 10.0 cm thick. Core B502 has a total ice content of 3 % by volume, and 19 % of its length below the LSS (5.96 m) is melt-altered.

In core B201, the LSS is defined based on the first appearance of coarse-grained firn rather than a crust as in B501 and B502 (Fig. ). The top meter of firn below the LSS is coarse-grained and contains several ice lenses. At 1.18 m BLSS, the firn transitions from coarse- to fine-grained. The largest sections of fine-grained firn occur from 1.70 to 2.71 m BLSS (1.01 m thick) and from 7.10 to 8.90 m BLSS (1.80 m thick). Core B201 has three ice layers over 10 cm thick, the largest being at least 33 cm, where we stopped drilling because of mechanical difficulties. Core B201 has a total ice content of 10 % by volume, and 27 % of its length below the LSS (9.91 m) is melt-altered.

Ice content and melt alteration of cores B501 and B201 from the LSS down to 14.15 m depth (below snow surface) are summarized in Table . Although all three cores differ in their ice content and layer characteristics, we focus on the difference between cores B501 and B201 to maximize the depth of overlap for a greater sample domain.

In general, density increases with depth throughout the core. However, cyclic variations can be seen, which are likely seasonal, particularly in the top 10 m. Individual ice layers can also be identified by peaks in density (Fig. ). We report density values beginning at 1.92 m depth (below the snow surface) because of high uncertainty in our volume measurements for the surface snow. The mean density from 1.92 m to the LSS at 4.24 m depth is 612±20 kg m-3. Seasonal snow above the LSS was dry with no ice content. We focus here on the firn below the LSS, much of which shows signs of melt alteration. The mean density of the top 2 m of firn below the LSS is 638±21 kg m-3. The mean density of the top 10 m of firn below the LSS is 689±10 kg m-3. The mean density of the bottom 2 m of the core is 722±23 kg m-3. The mean overall density of core B501 is 679±9 kg m-3. Note that although measured densities for some ice layers are implausibly high, their values are physically reasonable within uncertainty.

We obtained borehole temperatures down to 14.1 m depth (below the snow surface) for B201 and to 15.5 m depth (below the snow surface) for B501 (Fig. ). Both profiles show an initial cooling for the top ∼ 3 m, which then transitions to a warming, with a temperature minimum just below -8 °C in the top 4 m of the profile. In B201, the warming continues until ∼ 12 m depth, below which the profile shows a temperature stabilization and slight cooling. In B501, the warming that begins around 3 m depth continues through the bottom of the profile domain (15.5 m). At 14 m depth, B2's temperature is -1.74±0.01 °C and B5's temperature is -3.37±0.01 °C. Although the temperature probe was continuing to equilibrate in the borehole at 1 min (30 s equilibration and 30 s measurement), our fully equilibrated spot measurements (indicated by stars in Fig. ) are consistent with measurements acquired after only 30 s equilibration. We therefore consider our full temperature profiles to be adequately representative of borehole conditions.

Clear, continuous englacial horizons are observable throughout the radargrams, which is generally indicative of a non-temperate snowpack (Fig. ). For example, our VHF (900 MHz) radargram shows a continuous reflector between 3 and 4 m depth (Fig. b), which we interpret to be the LSS. The LSS is determined to be at 3.66 m at site B2 and 3.60 m at site B5, differing from the average LSS depth observed in our firn cores because of assumptions about the radio-wave velocity used to interpret the GPR data. Similarly, our HF (5 MHz) radargram shows a continuous reflector at ∼ 150 m depth, which we interpret to be the ash layer associated with the Katmai eruption of 1912 (; Fig. c). Our HF radargram also shows the total ice thickness, which ranges from ∼400 m to greater than 700 m (Fig. c). The deepest ice is found near the center of the profile, with the bedrock sloping up toward either end. Bedrock is located 484 m below the surface at site B2 and 566 m below the surface at site B5. There is no evidence for a firn aquifer in any of our GPR data.

Assuming a continuation of current climate conditions through 2033, our reference model predicts little change in firn temperature below 10 m depth; over 50 replicate runs, our median predicted firn temperature is -1.75 °C at 10 m depth, -1.50 °C at 15 m depth, and -1.34 °C at 20 m depth in 2033. However, our range of predicted firn temperatures is 5.31 °C at 10 m depth, 4.06 °C at 15 m depth, and 3.53 °C at 20 m depth, with the reference model producing temperate firn for at least some of the 50 replicate runs at each depth (Fig. ). Eclipse therefore appears to be near a threshold for supporting temperate firn below the penetration depth (∼ 10 m) of the annual temperature wave.

Additionally, firn conditions (even below 10 m depth) at Eclipse are sensitive to changes in air temperature and exhibit threshold behavior with projected warming (Fig. ). Over 50 replicate model runs with 0.1 °C cooling by 2033, firn at 15 m depth has a 4 % chance of being temperate year-round in 2033 compared to a 6 % chance over 50 replicate runs under a continuation of current climate or with 0.1 °C warming. The probability of firn at 15 m depth becoming temperate year-round increases to 26 % with 0.2 °C warming, 72 % with 0.5 °C warming, and 98 % with 1 °C warming by 2033. The evolution of density, liquid water content, and temperature over the full firn column is shown under each modeled climate scenario in Fig.  for a model run randomly selected from the 50 replicates.

Conditions associated with the development of year-round temperate firn at 15 m depth include higher total melt season positive degree days (PDDs), more individual melt events, higher average melt event magnitude, and higher mean winter temperature (Fig. ). We find a significant (p<0.05) difference (independent sample t test) for mean winter temperatures during all years after 2025 between model runs that do produce year-round temperate firn by 2033 and those that do not (Table ).

Additionally, for model runs that do not produce temperate firn, the median of the total melt season PDDs (over all model runs for any given year) never exceeds 35 (Fig. ). In contrast, the median total melt season PDDs over all model runs that do produce temperate firn by 2033 range from 31.58 in 2025 to 52.88 in 2033. Finally, we find a significant (p<0.05) difference (Wilcoxon rank sum test) for the total melt season PDDs, the number of individual melt events, and the median melt event magnitude (mm) between model runs that produce temperate firn at 15 m depth and those that do not during most years from 2024–2033 (Table ). We do not find any significant (p≥0.05) difference in median melt season start, end, or length (Table ).

Over 80 % of the St. Elias Range's glacier cover lies below both Eclipse and the Kaskawulsh/Hubbard Divide in elevation (Fig. ). Although it is only about 400 m higher in elevation than Divide, the firn pack at Eclipse to date remains drier than that at both the Icefield Camp and Kaskawulsh study sites. Comparisons between VHF (400 MHz) and HF (5–10 MHz) GPR data from Eclipse and Icefield Camp in 2016 and 2018, respectively, showed stratigraphic differences indicating a wetter firn pack at Icefield Camp relative to Eclipse, including a bright reflector in the HF data at ∼ 25 m depth suggesting the presence of a liquid water table in the firn at Icefield Camp but not at Eclipse . VHF data from Icefield Camp show greater signal attenuation than those from Eclipse, consistent with a wetter firn pack at Icefield Camp . GPR data from Eclipse in 2023 show the same results: clear stratigraphy with low signal attenuation and a lack of a bright reflector that would indicate a water table.

At the Kaskawulsh site, two firn cores were drilled in 2018 that allow comparison of firn properties between Kaskawulsh and Eclipse. The Kaskawulsh cores showed extensive evidence of meltwater percolation and freezing events, including the saturation of firn with liquid water below 34.5 m depth . The saturated firn, or firn aquifer, likely developed since 2013 based on model results, but its age cannot be confirmed . The seasonal snow layers at both Eclipse in 2023 and on Kaskawulsh in 2018 were dry since drilling at both sites occurred in the very early stages of the melt season. Ice content in Kaskawulsh 2018 Core 1 is similar to that in Eclipse 2023 core B201 and more than Eclipse 2023 cores B501 and B502 (Table ). However, the density of the upper 10 m of firn (below the LSS) was higher in Eclipse core B501 (688±10 kg m-3) than in the Kaskawulsh 2018 cores (588±8 and 572±7 kg m-3; ). Much like cores B2 and B5 at Eclipse, Cores 1 and 2 on Kaskawulsh showed some similarities in general regions of melt alteration but distinct individual stratigraphic layers.

A borehole temperature of -3.37±0.01 °C was recorded at 14 m depth in May 2023 at site B2, approximately the same location (within ∼ 50 m) where borehole temperatures of -5.04±0.5 and -5.50±0.5 °C were recorded at 14 and 20 m depth, respectively, in May 2016 (Fig. ). The 1.67 °C increase in temperature at 14 m depth over those 7 years supports the notion of a warming regional firn pack suggested by changes in borehole temperatures and stratigraphy at the Kaskawulsh/Hubbard Divide between 1965 and 2018, where an increase in melt and refreeze was indicated by an increase in both the quantity and thickness of ice layers and lenses observed in the firn . The warming firn pack and increased melt are likely due to atmospheric warming over that period, which has been shown to be amplified at high elevations; downscaled NARR gridded surface air temperatures in St. Elias show a 1979–2016 warming rate of 0.028 °C a-1 between 5500 and 6000 m a.s.l., approximately 1.6 times larger than the global-average warming rate from 1979–2015 .

Additionally, Eclipse density profiles from 2002, 2016, and 2023 show an apparent densification of the top 20 m of firn at Eclipse over the 21-year period (Fig. ). We interpret this apparent densification with caution because of the unrealistically high measured densities of ice layers in the 2023 core (>917 kg m-3); however, both the location of peaks and transitions from periods of high to low measured density are consistent with our stratigraphic observations, coinciding with ice layers and seasonal changes in firn grain size. Results suggest an increase in melt within the snow and firn, which leads to densification first by rounding snow grains, allowing them to pack more closely, and eventually by filling in pore space and refreezing .

We see no evidence of seasonal melt onset in the top ∼ 4 m of snow. However, below this depth all three firn cores display a variety of features associated with snowmelt, including ice layers, ice lenses, and bubbly and coarse-grained firn (Fig. ). The presence of such features indicates that snow at Eclipse experiences melt and percolation during the summer months, leading to the formation of ice lenses and layers that freeze either in subsurface firn below 0 °C or as ambient surface temperatures drop below freezing.

In the early stages of summer melt, capillary action causes the small amount of liquid water produced to stay in place between the crystals ; if refreezing occurs at this stage, the refrozen melt only partially fills the boundaries between individual grains. It therefore does not produce a solid blue ice layer, instead appearing bubbly or composed of very coarse but well-sintered grains, which we note in our observations as “coarse-grained”. The strength of these layers differentiates them from depth hoar and is consistent with formation from meltwater, as repeated melt–freeze cycles increase both the size of snow crystals and the bond strength between them . If the temperature remains above freezing for an extended period, ponding and wet grain growth can occur . Although not necessary for ponding or grain growth, blue ice layers, which we observe in all three firn cores, may form under these conditions .

Ice lenses (ice layers that do not extend across the entire core diameter) are observed in all three firn cores, indicating the development of preferential pathways for meltwater movement. Percolation is a notoriously spatially variable phenomenon, often occurring via vertical pipes that develop as surface melt progresses and can leave surrounding firn unaltered . Preferential flow through vertical pipes has been observed extensively in both seasonal snowpacks and glacier and ice sheet snow cover . We interpret the appearance of ice lenses as evidence of vertical piping in the snow and firn at Eclipse, although no pipes were sampled directly.

We interpret thin (≤2 mm) ice layers in all three cores as buried sun crusts. Sun crusts form when meltwater in the surface snow refreezes due to radiative cooling; the forming crust reduces shortwave absorption, allowing additional water vapor to condense below the initial glaze . Buried sun crusts account for 31 % and 42 % of observed ice layers in cores B501 and B502, respectively, but only 16 % of observed ice layers in core B201. Sun crusts are surface formations that are later buried; they do not indicate movement of liquid melt through the snow and firn pack.

Thicker ice layers can form at the surface or deeper in the snow and firn, either through prolonged or intense individual melt events or through the cumulative effect of multiple melt events once an impermeable barrier to deeper percolation is established . Multimeter-thick ice slabs, such as those observed in Greenland, can develop with multiyear meltwater production . Alternatively, rapid freeze–thaw cycles can produce melt but inhibit its percolation, resulting in melt complexes that comprise many thin melt layers in close proximity, rather than thick consolidated slabs . Our results are more consistent with the latter scenario, with the thickest ice layers observed in cores B501 and B502 being 12.0 and 10.0 cm, respectively. However, the possibility of an ice slab at the bottom of core B201 cannot be dismissed, as we drilled through 33.0 cm of ice before stopping due to mechanical issues; the total thickness of that layer therefore remains unknown.

Neither our VHF (900 MHz) nor HF (5 MHz) radar is high-frequency enough to identify thin and discontinuous ice layers in the near surface, which would indicate subsurface meltwater movement. However, both systems do provide a sense of the overall wetness of the firn pack. Liquid water is very effective at attenuating a radar signal, so the transition from a dry to wet firn pack can be inferred from the disappearance of stratigraphy in GPR data (e.g., ). The presence of clear stratigraphy in both our VHF and HF radar profiles supports previous observations of an overall dry firn pack at Eclipse . This is in contrast to a bright reflector atop washed-out deeper stratigraphy seen in a 2018 radar profile from nearby Icefield Camp (∼ 400 m lower than Eclipse), indicative of a liquid water table in the firn .

GPR results, core stratigraphy, and borehole temperatures together indicate that while spatially consistent annual melt–freeze cycles can be observed at Eclipse at the kilometer scale, the effects of meltwater production and percolation across this distance can vary substantially. For example, annual cycles as approximated by transitions between ice-rich and unaltered sections of firn are consistent between cores B501 and B501 and concur with those suggested by cyclic variations in B501 density. However, the location, thickness, and type (e.g., clustered ice layers vs. coarse-grained firn) of melt features differ between the cores due to small spatial variations in the amount of snow deposited at each site and to the irregular nature of ice lens, finger, and layer formation. Furthermore, core B201, drilled ∼ 1.5 km from the B5 cores and at a similar elevation, has a higher firn ice content (10 % by volume) than either B501 (5 % by volume) or B502 (3 % by volume), evidence of either greater surface melt production or greater retention of liquid water in the near surface. Additionally, B201's ice content tends to comprise thicker ice layers despite B201 having lower borehole temperatures than B501 below ∼10 m depth (Table ).

This difference in borehole temperatures is likely due to a difference in the surface energy balance and/or meltwater input at the two sites; the relative magnitude of surface energy balance terms can vary substantially based on prevailing weather, topographic shading, and seasonal effects, even between closely located sites with similar mean annual temperatures . One explanation for observed differences between sites B2 and B5 is that B5 on average receives slightly more solar radiation due to its southeast aspect, especially in the winter. Eclipse is high enough in latitude that the sun rises considerably further south in the winter than summer. This being a predominantly winter phenomenon could explain why B5 has fewer melt features despite being slightly warmer; however, we consider it unlikely that the wintertime southerly migration of sun position would outweigh the influence of summertime melt production at the two sites. Moreover, in areas with surface melt and percolation into the subsurface, the role of conduction in downward heat transport is comparatively minor relative to that of latent heat associated with the refreezing of meltwater . Another explanation is that meltwater percolation is responsible for warming the firn at B5 but occurs adjacent to our core sites and is therefore not recorded in our core stratigraphy. For example, the lateral transport of liquid water along a subsurface layer boundary from the southeast-facing areas upslope of site B5 may account for the higher subsurface heat content of the site relative to B2. However, subsurface meltwater flow would be limited by the large cold content of below-freezing firn. Additionally, site B2 could also experience liquid water transport from upslope areas, predominantly with a western aspect; though based on surface debris, the areas upslope of B2 experience greater avalanche disruption than those upslope of B5, making meltwater transport along consistent subsurface pathways less likely.

The present difference in firn water content between Eclipse, Icefield Camp, and Kaskawulsh is likely because Eclipse sits ∼ 400 m higher in elevation than the other sites. The lower elevation but generally similar environment of Icefield Camp and Kaskawulsh relative to Eclipse make them useful case studies for predicting the future firn evolution at Eclipse with continued warming in the St. Elias Range.

Unlike the Kaskawulsh/Hubbard Divide area, Eclipse has yet to develop a firn aquifer, as demonstrated by GPR data from both 2016 and 2023. HF (5–10 MHz) data from both years show bedrock that slopes to a trough in the middle of the ice field, a bright reflector at ∼ 150 m depth interpreted as the ash layer from the 1912 Katmai eruption, and continuous stratigraphy with the exception of surface zones of avalanche debris. Neither 2016 nor 2023 HF data show a bright reflector indicative of a liquid water table. VHF (400–900 MHz) data from both years also look similar, with clear and continuous stratigraphy down to 20 m depth indicative of a dry firn pack. Despite remaining dry to date, borehole temperatures indicate that at least part of Eclipse Icefield is <2 °C from the melting point at depth. Moreover, model results indicate that Eclipse is close to the threshold for developing temperate firn, with a 2 % chance of year-round temperate firn at 15 m depth by 2033 without continued atmospheric warming, a 51 % chance with 0.5 °C warming, and a 98 % chance with 1 °C warming. Model behavior also suggests an increase in liquid water content at depth, with the potential for firn aquifer development. During many of our model runs, the CFM failed to produce any outputs below ∼ 25 m depth starting in the late 2020s. Because the CFM is limited in its ability to handle large amounts of liquid water, we associate this pattern of model failure with high amounts of persistent liquid water, consistent with the possibility that Eclipse follows a similar trajectory to Kaskawulsh and Icefield Camp, developing a firn aquifer in the next 5–10 years.

We suggest that more and higher-magnitude melt events during the height of summer combined with warmer wintertime temperatures promote the development of year-round temperate firn in St. Elias. Model results for Eclipse show the development of year-round temperate firn at 15 m depth associated with an increase in total PDDs throughout the melt season, as well as with a greater number of individual melt events, higher average melt event magnitude (mm melt produced), and warmer winter temperatures, rather than an earlier or prolonged melt season (Fig. ; Tables , ). In Greenland, “extreme melt events” have been related to firn's multiyear response to surface melt via the formation of thick ice slabs and ice layer complexes, which cause a near-surface barrier to downward percolation . In St. Elias, however, an increase in the number of melt events and an increase in average individual melt event magnitude are more likely to result in sustained heat transport to depth because of the insulating effect of the region's high annual accumulation (1.4 m w.e. a-1 at Eclipse) relative to accumulation rates in Greenland (0.3–1.2 m w.e. a-1; ; ; ). Also consistent with observations in Greenland , our results show that the development of year-round temperate firn at Eclipse is associated with an increase in winter temperatures. Rather than directly relating to melt production, wintertime warming affects firn properties by reducing the regeneration of cold content that occurs between melt seasons, effectively enabling the warming effects of summertime melt to compound from one melt season to the next.

If Eclipse does indeed develop a firn aquifer in the next decade, over 90 % (>22000 km²) of the region from 59.59 to 61.58° N and 138.02 to 142.23° W could support liquid water in the firn based on its elevation (Fig. ). Moreover, areas above Eclipse in elevation are largely steep mountain peaks and represent a far smaller portion of the St. Elias hydrological reservoir than the broad ice fields below, which retain more snow and ice. Deep (>10 m depth) temperate firn up to 3000 m a.s.l. would therefore represent widespread meltwater percolation and constitute a wholesale change in the region's hydrological system. In particular, the capacity of the ice fields to buffer runoff would be reduced by meltwater's direct occupation of pore space and by intensified compaction associated with its rounding and lubrication of grains . Such processes associated with warming have contributed to a 5 % reduction in firn pore space in Greenland since 1980 . Additionally, firn loses pore space in response to warming more readily than it gains pore space in response to cooling; observed densification of the firn to date therefore has long-term consequences for runoff buffering .