We enhanced our 2024 measurements with two CMP 2025 measurements at each location (CMP3 and CMP4) (Fig. ), approximately 50–100 m apart, and an additional new CMP gather at the lower part of the Ewigschneefeld (CMP2 in Fig. ). We also extended the GPR transect to trace the spatial firn-to-ice transition and the variation in firn thickness with elevation along a flow line. All three CMP measurements were collected using a 200 and 600 MHz dual-frequency, mono-static IDS GPR system. Table  illustrates the GPR system settings used for all CMP gathers and GPR transects during both 2024 and 2025 field campaigns. The common midpoint (CMP) method is a geophysical approach that assumes subsurface reflectors (IRHs in radargrams) are horizontal, homogeneous layers. The CMP gather follows a similar acquisition procedure as that described by . During CMP acquisition, the transmitter (Tx) and receiver (Rx) are moved so that they remain at equal distances from a selected fixed midpoint. With this approach, subsurface targets appear as diffraction hyperbolas in the radargram (Fig. a). The CMP setup used 20 cm offsets from the fixed midpoint, perpendicular to the GPR transect direction (i.e., a 10 cm shift of the antennas on either side of the midpoint), covering a total CMP measurement length of 20 m. The details of CMP acquisition with schematic representation are given in the supplement along with a representative CMP radargram as in Fig. S1 in the Supplement.

The GPR profile (same instrument as for the CMP) with a length of 4.4 km extends below the equilibrium line altitude (ELA), covering an elevation range from 3514 to 3202 m a.s.l (Table , black GPR transect in Fig. ) and extends beyond the 1.8 km long GPR track from 2024 (blue GPR transect in Fig. ). The GPR was towed manually using external GPS location with a time trigger. The overview of measurement locations from both winter campaigns (2024 and 2025) is shown in Fig. .

We acquired an approximately 18.3 m deep firn core from the base of the 1.12 m shallow snow-pit, using an electromechanical drill (FELICS small, ). The corer retrieves 0.9 m core segments (58/73 mm core/borehole diameter) at the upper part of Ewigschneefeld near the CMP3 location (FC3 in Fig. ). This enables direct observation of the firn density profile to a depth of nearly 19.4 m. Density within the snow pit was estimated by measuring the weight of the collected snow sample in a plastic bag using an analogue scale and the volume of a 14 cm-long cylindrical cone with a diameter of 4.75 cm. At the same time, the firn core sections were split into approximately 20 cm intervals for density measurements. The initial 20 cm of the first firn core was soft, and we assumed some disturbance due to drilling during density estimation at a shallow depth of around 2–3 m. We used plastic bags and analogue weighing springs to measure a sample weight. Further, density was determined from the known corer volume and weight of the measured core sample. The 20 cm firn core sections were stored in plastic bags, melted at the research station, and the meltwater was transferred into sampling tubes. The collected samples were analysed in the laboratory of the Geozentrum Nordbayern at Friedrich-Alexander-Universität Erlangen-Nürnberg for stable isotopes (expressed as per-mil deviation from a defined standard: VSMOW Vienna Standard. Mean Ocean Water) and chemical impurities (major ions). The details of the water isotope and chemical impurities analysis are provided in the Supplement.

An additional shallow firn core (FC2) of approximately 7 m depth from the bottom of a 1.12 m snow-pit was gathered on 30 March 2025, using a “Mark II Ice Coring System” from Kovacs Ice Drilling Equipment with a 9 cm diameter at the lower part of the Ewigschneefeld near the CMP2 location (orange FC2 in Fig ). When translating from snow-pit to firn core drilling, we expect a loss of around 15–20 cm in core length due to the crumbling of fresh snow, and the data gap is filled by assuming a fresh-snow density measured from the snow-pit. A snow pit (orange SP4 in Fig. ) approximately 2.8 m deep was dug near the CMP4 location to measure near-surface density and to investigate melt-induced ice lenses at the Jungfraufirn. The density of all snow pits was estimated using weight of collected sample and cone volume (14 cm-long, 4.75 cm-diameter).

The GPR transect was processed using ReflexW software following the conventional processing steps as described in previous studies . A sequence of processing steps, such as filtering and gain, removes noise and improves the visibility of the radargrams. The application of de-wow filtering, stacking, and a bandpass Butterworth filter increases the signal-to-noise ratio. Other processing steps included moving the start time to adjust the first arrival at the surface, background removal, and static correction to reduce system-induced irregularities. We also applied interpolation technique for equidistant traces (Fig. a) similar to that of the study. The visible IRHs on GPR transect radargrams were traced/picked using the built-in semi-automated phase follower tool within ReflexW software. The basic GPR processing steps used for GPR transects and CMP gathers are detailed in Table . A flowchart and detailed explanation on GPR transects and CMP data processing flow is provided in the Supplement (Fig. S2).

Studies such as , , and illustrate that the subsurface properties of the medium can be derived from the radar wave propagation velocity, which is necessary to accurately estimate internal reflection horizon (IRH) depths from the measured two-way travel time (TWT). The radar propagation velocity as Vrms in the medium above the target can be estimated by matching the reflections of a synthetic propagation model to the curvature of the observed reflections . The CMP data were processed in a manner comparable to our 2024 workflow and, as outlined in Sect. 4.1, the only difference is that each Tx–Rx offset was acquired individually and then merged into a continuous profile within ReflexW. The parameters and parameter values used for the semblance analysis are provided in Table .

The semblance analysis (Fig. ) aids in picking Vrms and TWT based on the peak energy coherence between waveforms that are centred on hyperbolic trajectories (Fig. b) within the analysis windows . The picked Vrms velocities and corresponding TWT of each reflector from the semblance analysis can be used to determine the interval velocities (Fig. c) using the Dix equation (Eq. ), which is a measure of subsurface dielectric properties within each firn layer (between two IRH boundaries): 1Vint=vrmsi2twti-vrmsi-12twti-1twti-twti-1 where Vint is the interval velocity through layer i, twt_i, twt_i−1, vrmsi and vrmsi-1 are two-way travel times and root-mean square (vrms) velocities through layer ith, and i−1th interfaces, respectively. Further, the density within each firn layer can be obtained by using the relation between the interval velocity (Vint) and the density by modifying the critical refractivity index method CRIM as 2ρ=VairVfirn-1VairVice-1ρice Here, Vair and Vice are radar wave velocities in air and ice (0.3 and 0.167 m ns⁻¹, respectively), whereas the velocity of the radar wave propagation within the firn layer (Vfirn) is estimated from the Dix equation (Eq. ) as Vint using picked Vrms from CMP semblance analysis and density of ice is assumed as 920 kg m⁻³. Further details of the semblance analysis and processing technique as a flowchart (Fig. S2) can be found in the Supplement with added screenshots of CMP-1D velocity analysis from the ReflexW software (Fig. S3).

illustrates the identification of internal reflection horizons (IRHs) as annual layers through direct comparison with firn core observations. presents an iterative chronological method to identify internal reflection horizons (IRHs) as annual layers by estimating snow water equivalent (SWE) from GPR-derived CMP-based radar velocity and comparing it with point mass-balance observations. We used this approach to identify observed internal reflectors in CMP acquisition as annual layers from the CMP (CMP2 and CMP4) acquisition. This is achieved firstly by determining each reflector's depth from the measured TWT, and estimating the snow water equivalent (SWE) of the respective layer by multiplying the radar interval velocity (Eq. ) with the estimated density within each layer (Eq. ): 3SWEi=0.5×TWTi-TWTi+1×Vi+1-Vi×ρi with 0.5 × TWT_i × Vi being the thickness of the respective layer and ρi is the density of the firn layer.

The estimated SWE within each reflector was compared chronologically with the long-term point mass balance observations (Stake in Fig. ). The matched SWE between the CMP-derived and long-term point mass balance measurements from the identified IRH is treated as an annual layer; otherwise, the next reflector (IRH) is evaluated iteratively until a SWE agreement indicates the correct annual layer boundary for that particular year. This process was applied to CMP-derived reflectors at CMP2 and CMP4 locations, where no deep firn-core data is available (Fig. ), assuming constant relative accumulation, which is likely to be constant over a limited area and conservation of mass within each identified annual layer. Figure  illustrates the flowchart of the iterative method for identifying internal reflectors as annual layers, applied only to the CMP2 and CMP4 datasets.

The study by demonstrates a method for tracing spatial firn density and accumulation history using a CMP-derived firn density-depth profile and an extrapolated profile from a shallow firn core. We improved this method by using a deep firn core (FC3) and a CMP-derived (CMP2) firn density-depth profile approximately 2 km apart (Fig. ). We estimated spatial firn density variation and accumulation distribution, following a similar approach to our previous work , using a CMP-derived density-depth profile (CMP2) for the lower part and a firn core (FC3) for the upper part of the Ewigschneefeld. We identified annual layers (Sect. 4.3 and “Annual layer dating using firn core”) and estimated corresponding densities within firn layers (Eq. ). Further, we obtained the spatial firn density distribution by interpolating the density gradient along the GPR profile (black line in Fig. ) between the FC3 and CMP2 locations. The accumulation distribution can be estimated in terms of SWE by multiplying the firn layer thickness by the estimated density of the respective layer (Eq. ). The IRHs identified on the radargram (Fig. b) are considered as annual layers by following the procedure in Sect. 4.3 and Fig. . We compared the results of accumulation and density variations with those from 2024 investigations and interpreted the temporal changes. We provide corresponding figures with brief interpretation in Appendix  (Figs.  and ). However, we discuss their effects on temporal changes in density and compaction rate estimates in the respective discussion sections (Sect. 6.3 and 6.4).

To derive the firn compaction rate from the CMP data, we compute the change in annual-layer thickness between two time points , in our case between the 2024 and 2025 measurements. After identifying annual layers, we determine the change in TWT in nanoseconds (ns) and, subsequently, the depth for each layer from the interval velocity. This change in layer thickness with time provides the compaction rate. 4Compaction rate=Layer thicknessi2025-Layer thicknessi2024time The equation used here (Eq. ) results in negative values for thinning layers between repeated measurements.

Figure  shows snow and firn density-depth profiles at three locations of the Grosser Aletschgletscher accumulation area using firn cores and snow-pits from winter 2025 (orange coloured FC2, FC3, and SP4 in Fig. ). The 19.4 m deep firn core (FC3) display firn density fluctuation from surface snow density of 200 to ∼ 830 kg m⁻³ at 18 m depth. The typical firn density of 550 kg m⁻³ is reached at around 5 m, and the depth of near pore close-off density (830 kg m⁻³) is deeper than the drilling depth despite some local maxima at ∼ 13 and 18 m, which can be attributed to increased local density due to ice lenses. The shallow firn core obtained at the lower part of the Ewigschneefeld (FC2 in Fig. ) demonstrates a similar density-depth variation compared to that of the deep firn core (FC3), but with generally higher densities. The depth to reach firn density (550 kg m⁻³) is at 3–3.25 m. The density offset between FC2 and FC3 locations is pronounced at depths greater than 2 m. The estimated bulk SWE up to the common depth of around 8 m at the FC3 location is around 3950 mm w.e. in comparison with 4140 mm w.e. at FC2.

Ice lenses were found in the firn core (FC2) at shallower depths (around 3 m) and at the bottom of the firn core (7–7.5 m), depicting the corresponding density maxima at these depths. Within the deep firn core (FC3), we could not identify any near-surface ice lenses up to a depth of 8.5 m. Several ice lenses were identified at greater depths, corresponding to local density maxima. The visible Sahara dust layers (orange bars) can also be seen near 9 and 11.5 m in the deep firn core (FC3) and at 6.5 m (brown bar) in the shallow firn core (FC2). The density trend from the snow-pit (SP4) shows similar variations as that of FC2, except for the higher surface snow density and a sharp kink at 2.5 m depth with a surface density offset of around 80 kg m⁻³ when compared with the FC3 measurement results.

We compared repeated glaciological measurements collected at the lower part of Ewigschneefeld on 29 February 2024 and 30 March 2025, approximately 50 m apart. Figure  shows the firn cores, the snow pit-derived density-depth profiles, corresponding identified ice lenses and the visible Sahara dust layer during two field measurements approximately 400 d apart. The comparison of the 2024 and 2025 results show depth to the last summer horizon at this elevation varies between the two years. In 2024, the measured winter accumulation up to the measurement day was approximately 1985 mm w.e. (∼ 5 m depth), whereas on 30 March 2025, the estimated accumulation was 1380 mm w.e. (∼ 3 m depth). Because of the lack of isotope analysis from the shallow firn core in 2025, we can not provide further supporting evidence for the summer-to-winter transition, as compared to the firn core in winter 2024, where the result was supported by isotope analysis (see Fig. 9 in ). The detailed comparison reveals that the variation in density with depth remains similar up to the depth of the last summer horizon (3 m). However, winter snow density at the surface in 2024 was nearly 100 kg m⁻³ lower than that of 2025 measurements (depth < 1 m). The winter 2025 firn core results show significant fluctuations with a maximum density offset of 200 kg m⁻³ between 4 and 5 m depth, which is not observed in winter 2024 (FC2 and SC2-2024). This can be attributed to the observed ice lenses at that depth. Furthermore, the depth to firn density (550 kg m⁻³) is much shallower (depth < 4 m) in FC2-2025 than in FC2-2024 where it occurred at around 5 m. The maximum density of around 650 kg m⁻³ occurs between 5.5 and 6.5 m in 2025, whereas in 2024 it occurs at 7 m depth.

The firn density-depth profiles of all three CMP acquisitions show increasing density with depth. Figure  illustrates the variable densification across three locations of the Grosser Aletschgletscher. At the lower part of Ewigschneefeld, the identified first reflector depth is approximately 3 m, and density above the first reflector is around 450 kg m⁻³ (CMP2), which is similar to the results from the CMP4 profile with an initial density of 460 kg m⁻³. In contrast, the CMP3 located at the higher elevation has a somewhat lower density (420 kg m⁻³) from the surface to the first reflector depth (around 3.5 m). The depth to the firn density (550 kg m⁻³) also varies across three locations, ranging from 3–3.5 m at CMP2 and CMP4 to approximately 5 m at CMP3 (higher elevation). Similarly, the depth to pore close-off density (830 kg m⁻³) is significantly shallower, around 18–19 m at the CMP2 and CMP4 locations, compared to 27 m at the CMP3 location.

The comparison of the density-depth profiles between the firn core (FC3) and CMP3 measurements (Fig. b) illustrates similar density variations with depth. It is noticeable that certain sharp spikes in FC3 and CMP3 density measurements are located at approximately the same depths, and both show a similar depth for the snow-firn transition. Comparison of density-depth derived from FC3 and CMP3 gives a mean bias of -5.73 kg m⁻³ and RMSE of ∼ 57 kg m⁻³ (Fig. ). Whereas the comparison of the shallow firn core (FC2) and the CMP2 density profile shows a strong spike at 5 m depth in the firn core, and a similar spike is visible in CMP2 near 4.5 m depth (Fig. a). The radar-derived depth to the deepest discernible IRH at all locations also varies from 23 m (CMP4) to 27 m (CMP2) to 32 m (CMP3). The uncertainty of radar wave velocity determination from the semblance analysis increases with depth, as seen on all three CMP density-depth profiles in Fig.  (shaded area).

The radargram obtained from the 4.4 km-long GPR transect collected on 30 March 2025 is shown in Fig. a. Although it is treated as a secondary result, it provides essential spatial context that improves the firn-density and accumulation estimates discussed in Sect. 6.2 and Appendix . The depicted radargram spans ∼ 300 m surface elevation range (Table ), starting at the upper accumulation area (left side of Fig. a) and reaching into the ablation area of the Ewigschneefeld. Between 0 and 3500 m, the visible IRHs show decreasing thickness with increasing distance along the profile. We identified the upper white region of the radargram, preceding the first prominent IRHs, as accumulated winter snow from the 2024/2025 winter; below, IRHs are interpreted as firn layers. At a depth below 25 m (∼ 220 ns), a prominent undulated reflector is visible (black arrows in Fig. a), which spans approximately between 200 and 2500 m in distance within the accumulation area. It is evident that the total firn thickness reduces with the elevation, and almost no firn exists from 3500 m onwards (∼ 3200 m a.s.l.) (rightmost part in Fig. a) reaching an ablation area. There, no distinct IRHs are visible except the 2025 winter accumulation and glacier ice boundary. This can be interpret as a spatial firn-to-ice transition and beginning of an ablation area (elevation ∼ 3200 m a.s.l) as discussed in Sect. 6.2.

Figure b and c illustrate the comparison of the processed GPR long profile radargrams acquired on 30 March 2025 and 29 February 2024. Here, we use only the upper 2.2 km of the 2025 profile, which also covers the extent of the 2024 profile (Fig. b). The first reflector between the surface and 60 ns (4.5 m) depth in Fig. c (black line) depicts the 2024 winter accumulation, which is overlaid by the 2025 winter accumulation of approximately 3 m depth (40 ns) in Fig. b (red line). Both radargrams show various distinct IRHs, with a prominent reflector at approximately 240 ns (24 m) in 2025 (Fig. b, black arrows) and 260 ns (26 m) in 2024 (Fig. c, red arrows), along the GPR transect. We used the 2025 GPR transect (Fig. b) along with FC3 and CMP2 to trace spatial firn density and accumulation. The corresponding results and discussion are presented in Appendix A, Figs.  and .

The temporal changes in firn density at the accumulation area of the Grosser Aletschgletscher from the CMP measurements are shown in Fig. . We identified more than 30 IRHs in each CMP3 and CMP4 measurement in both years (2024 and 2025). The use of a 600 MHz GPR system in 2025 provides higher resolution but less penetration depth than the 500 MHz GPR system used in 2024. Therefore, we present CMP results at a common depth from all CMP measurements. It is also evident from Fig. a and b that the radar wave attenuates faster in the presence of denser firn (CMP4), providing the deepest discernible IRH approximately at 23 m depth. In contrast, at the CMP3 location, the deepest IRH is at nearly 33 m.

The comparison of temporal changes in CMP-estimated firn density at the CMP3 location (Fig. ) shows an increase in firn density with depth, with some fluctuations along the profiles occurring at similar depths. It is observable that the firn density profile from 2025 (black line in Fig. a) shifted towards higher values compared with the 2024 CMP measurements. The depth to the firn density (550 kg m⁻³) is around 7 m in 2025 and 8 m in 2024, and the pore close-off density (830 kg m⁻³) is reached at approximately 27 m in the 2025 and 2024 measurements. Similarly, increased firn density with depth is seen in Fig. b, estimated from CMP4 measurements near the Stake (Fig. ), which is located at a lower elevation than the CMP3 measurements. Comparison of firn density-depth profiles shows a distinct, higher density shift in 2025. The depth to firn density is at around 3.5 m in 2025 and 8 m in 2024, whereas the pore close-off density is observed at approximately 19 m in both years.

Figure  represents the temporal changes within the identified annual firn layers using repeated 2024 and 2025 CMP measurements at two locations of the accumulation area of the Grosser Aletschgletscher. As explained in the method section (Sect. 4.3), we identified 10 and 11 annual layers from more than 30 IRHs at CMP3 (Fig. a) and CMP4 (Fig. b) measurements. At the CMP3 locations, firn density within the annual layers is higher than in the 2024 measurements. The 2025 winter accumulation shows a lower snow density (430 kg m⁻³) than the 2024 winter accumulation, which is close to 500 kg m⁻³ (orange line in Fig. a). We estimated an increase in density offset across all identified layers from 2024 to 2025, with the change more pronounced in layers 3 and 4 (years 2022 and 2021). The results indicate that the effect of densification is low in deeper or older layers (layers 9 and 10 correspond to 2016 and 2015, respectively), and the estimated density within the identified oldest layer (2015) is approximately 750 kg m⁻³. The identifiable 2022 firn layer at the CMP3 location densified from 540 kg m⁻³ in 2024 to 610 kg m⁻³ in 2025.

Similarly, the CMP4 results show pronounced firn densification in the near-surface annual layers (2023 and 2021), with complete ablation of layer 3 (2022) at this location. The maximum densification is around 140 and 100 kg m⁻³ yr⁻¹ in the 2023 and 2021 firn layers between 2024 and 2025. The estimated density change (offset) within older layers (2017 to 2014) is small. The density difference between the winter 2024 and 2025 accumulation (orange line in Fig. b) is not as significant as recorded at the CMP3 location (near 40 kg m⁻³). The identified oldest layer (2014) has a density near the pore close-off density (830 kg m⁻³) in the 2024 and 2025 measurements. Whereas the 2024 winter-accumulated snow densifies further, from 450 to 620 kg m⁻³ in 2025. Figure  also depicts a form of visualisation of the temporal changes in firn density within identified annual layers at two locations.

The CMP-derived compaction rate of commonly identified annual layers at both CMP3 and CMP4 locations is shown in Fig. a, and the corresponding temporal change in firn density is depicted in Fig. b. We estimated that commonly identified annual layers, particularly those near the surface (younger layers), were compacted by nearly 0.3 m yr⁻¹ at the CMP4 location (red line) at 7–8 m depth during 2023 (Fig. ). In contrast, the compaction rate of the same layers at the CMP3 location was approximately 0.2 m yr⁻¹ at the same depth (7–8 m) in 2023 (Fig. ). The compaction rate of the deepest layers at the CMP3 and CMP4 locations is nearly zero. The estimated bulk compaction from the reference surface (layer 0 as 2023) to the common firn layer (layer 8 as 2015) at CMP3 is 0.5 m, and at the CMP4 location is 1.18 m between the 2024 and 2025 measurement periods. The mean compaction rate of commonly identified annual layers from 0–8 (2023 to 2015) at the CMP4 location is approximately -0.14 m yr⁻¹, whereas at the CMP3 location it is approximately -0.06 m yr⁻¹. It is noticeable that the higher compaction of younger layers and lower compaction of older (deeper) layers correlate well with the observed change in firn layer density over a year (Fig. b).

Cumulative depth-age Fig. a and b illustrate the depth of commonly identified annual layers from the referenced layer (2023 as layer 0) to the last identified layer 8 (2015) in both measurement years at CMP4 and CMP3 locations. At the CMP3 location, the cumulative depth of 9 identified annual layers is approximately 11.52 m in 2024, compacted to 11 m in 2025. The compaction is most pronounced in the near-surface layers (Fig. a), as reflected by the close spacing of layers 0–3 compared to the deeper layers (4–8), which are nearly parallel. At the CMP4 location, the total thickness of the 8 identified annual layers is around 13.4 m in the 2024 measurements and 12.2 m in the 2025 measurements. Similar to the CMP3 location, the effect of compaction is pronounced in near-surface layers (0–4). As mentioned previously, the 2022 extreme summer melt completely ablated layer 1 at the CMP4 location, which is identifiable at the CMP3 location, resulting in a thicker firn body of around 8 m up to layer 6 in 2025 compared to 6 m at the CMP4 location. The overall bulk thickness from 0–9 layers at the CMP4 location is almost similar to that of the CMP3 location, which we attribute to the uncertainty (Figs.  and ) in annual layer dating using the firn core (FC3, method section “Annual layer dating using firn core”) as discussed in Sect. 6.5.

Direct observations at three accumulation-area sites reveal both common firn features and site-specific stratigraphy. The deep (FC3) and shallow (FC2) cores from the upper and lower Ewigschneefeld (Fig. ) display similar density–depth profiles with increasing density with depth. Surface snow density is slightly lower at FC3 (3450 m a.s.l.) than at FC2 (3360 m a.s.l.), which we attribute to the temperature lapse rate, resulting in denser snow at lower elevation. This is supported by the SP4 (3340 m a.s.l.) profile, which shows a higher surface density of 80 kg m⁻³ (Fig. ) than FC3, consistent with the other observations. Temperature records indicate mean summer temperatures of about 2.3 and 3.2 °C in July and August 2024 at 3570 m a.s.l. , implying enhanced melt at lower elevations (FC2 and SP4 in Fig. ). We further hypothesise that wind-driven snow and spatial variability in accumulation (Fig. ) contribute to higher surface densities at lower elevations (FC2 and SP4), but quantifying this is beyond the scope of this study. Our results indicate that the depth to the last summer horizon is about 3.5 m, based on FC3 and FC2, supported by visible ice lenses and a local density increase from 420 to 550 kg m⁻³ at FC2. In contrast, we did not detect distinct ice lenses at 3–4 m depth in FC3, which we attribute to reduced surface melt at higher elevations (> 3440 m a.s.l.) that would otherwise form such lenses. Overall, the data indicate a well-defined summer layer beneath the winter snow in 2025 (Fig. b and c). We interpret the ice lenses as evidence of melt–refreeze events, which reflect temperature variations during their formation.

The isotope results show clear variation at shallower depths (< 4 m), indicating changes in winter precipitation. The δ18O range of -24 ‰ to -12 ‰ matches expected values for this altitude , implying that both winter and summer accumulation are preserved in the upper 0–4 m. Below 4 m, the strongly smoothed signal with little seasonal contrast suggests the effect of the 2024 summer surface melt. Attenuation of δ18O between 8 and 14 m, together with numerous ice lenses, likely reflects extreme melt in summer 2022, which diluted isotopic signals and marks the refreezing depth of that melt (Fig. b). demonstrated that ice cores from the Swiss Alps, including those from the Jungfraujoch Saddle (1987), illustrate how wind and surface melt smooth δ18O. Likewise, a sharp electrical conductivity peak (∼ 80 µScm-1 at ∼ 3.5 m; Fig. c) provides independent validation for identifying the position of the last summer horizon, complementing the other results (Fig. b, c). We attribute the smoother conductivity signal below 4 m to signal loss due to surface meltwater washout and runoff of soluble ions. The increase in conductivity between 8 and 15 m likely records melt-driven impurity concentration during the extreme 2022 summer (Fig. c).

The analysed chemical impurities (Na, Ca, Ti, and Fe; Fig. ) in the firn core reflect atmospheric pollution history from anthropogenic and natural sources (e.g., ). Surface melt can significantly disturb these records by altering signal preservation (e.g., ), as shown by the loss of signal strength with depth (Fig. ). Annual layer dating at CMP3 is based on Fe concentration spikes, supported by Ti and visible Sahara dust layers (Fig. g), because Ti and Fe are less affected by meltwater infiltration and are better preserved. In contrast, Na and Ca are more soluble and diluted by meltwater percolation 10.1029/2019JF005026. Figure  compares FC2 2025 and 2024 and shows temporal changes in firn density. Winter 2024 had intense snowfall of about 3000 mm w.e., while winter 2025 had about 1900 mm w.e. , indicating changes in seasonal accumulation. This is further evident in our two firn core results, indicating varying fresh-snow density and depth to the last summer surface (3.25–3.5 m, 1700 mm w.e. in 2025, vs. 5 m in 2024), which we attribute to differences in winter accumulation. Surface-melt ice lenses formed in summer 2024 appear in the FC2 2025 core below 3 m, raising density to nearly 600 kg m⁻³ at 4 m depth; the same density occurred at 6.5 m in 2024. We interpret the cluster of ice lenses at 6.25–6.5 m in 2024 as the same feature observed at 6.8–7.1 m in 2025 (Fig. ). This 0.3–0.75 m downward shift over 1 year suggests additional 2025 winter accumulation. Because FC2-2025 was drilled on 30 March 2025 and winter 2024 accumulation was high, the precise position of the 2023 annual layer is uncertain. We tentatively assign the ice lenses near 7 m depth to the 2023 layer (Fig. ), assuming 3–4 m of 2025 winter accumulation (about 2000 mm w.e.) and a ∼ 3.5 m thick 2024 firn layer.

The GPR-based CMP measurements provide an indirect firn density–depth profile, which we compared with densities from direct observations, FC3 (Fig. ), near CMP3 (Fig. b). Both profiles show similar structures, with sharp density increases at comparable depths (∼ 5, 12, 14, 16, and 17 m in Fig. b), indicating ice lenses at these depths (Fig. ). Both observations show good agreement in the upper part, supporting the use of CMP measurements for density estimation. All three CMP results also reveal firn thickness in the accumulation area of the Grosser Aletschgletscher, highest at the Ewigschneefeld (CMP3) and lowest at the Jungfraufirn (CMP4 near the Stake in Fig. ). Earlier studies and reported similar firn thickness, with a firn–ice transition depth of 32 m at the Grosser Aletschgletscher.

Radar wave penetration depth varies across the glacier (∼ 32 m at CMP3, ∼ 28 m at CMP2, and ∼ 23 m at CMP4 in Fig. ), which we attribute to attenuation caused by increasing firn density with depth, altering the dielectric property of the medium . The estimated depths to firn and pore close-off density at all CMP locations indicate spatial variation in firn densification, mainly driven by melt and refreezing, which enhance densification at CMP4 relative to the other two CMP locations. We interpret these melt–refreezing processes as elevation-dependent and minimal at CMP3.

In the GPR transect, we interpret the first strong reflector as the 2025 winter accumulation depth (Fig. a). Snow layer thickness decreases from ∼ 30 to ∼ 20 ns over the first 2000 m, then thins from ∼ 20 to 15 ns between ∼ 2000 and 4400 m, indicating spatially variable winter accumulation. From ∼ 3500 m onward (Fig. a), IRHs below the 2025 winter layer vanish, marking the transition to the ablation area and the approximate ELA, consistent with . We compared the upper 2.2 km of this profile (Fig. b) with a 1.8 km profile from 29 February 2024 (Fig. c). In the lower part (distance > 1200 m), strong compaction of the full winter accumulation (3000 mm w.e., thickness > 6 m) and melt–refreeze processes dominate. Accordingly, in both radargrams, near-surface IRHs (Fig. b, c) and all identified IRHs (Fig. a) thin with increasing distance and decreasing elevation, consistent with melt-driven ablation controlled by lapse rate and variable accumulation.

A strong reflector is visible at the deepest part of the radargrams (Fig. a and b) across the accumulation area (0–2400 m). In the 2024 GPR transect, this reflector was interpreted as an annual layer (Fig. c) by , but the 2025 radargrams show that its geometry is uneven and becomes increasingly distorted toward the saddle. This behaviour is consistent with enhanced scattering and stratigraphic disturbance. We hypothesise that the observed variation in stratigraphy and the prominent reflector may result from crevasses and strong accumulation gradients near the ridge separating Ewigschneefeld from Mönchsjoch (Fig.). Similar undulated reflectors have been reported in for the same region. We need further research and additional field observations to support our hypothesis.

Studies such as suggest that melt and refreezing can produce high-density layers, or ice lenses, that can be misinterpreted as annual layers. Therefore, not all IRHs necessarily represent annual layers, which is also clear from the comparison of density-depth profiles between the firn core (FC3) and CMP3 (Fig. b). The method explained in Sect. 4.3, which is introduced in , was used to identify IRHs as annual layers (Fig. a). Figure a demonstrate the annual layer identification by chronologically comparing CMP-derived SWE with point mass balance estimations (Sect. 4.3), whereas Fig. b shows the layer dating using CMP3-estimated SWE and FC3 (section “Annual layer dating using firn core”). Thus, the identified annual layers from multi-year CMP measurements at two locations (CMP3 and CMP4 in Fig. ) are shown in Fig. , illustrating temporal changes in firn density within these layers. It is noticeable that the firn density has increased over the year at both CMP locations. The density-depth profiles at CMP3 show that the vertical structure of the density variations is preserved, while the absolute values increase. The changes we observe therefore reflect differences in the magnitude of densification within individual annual layers rather than variations in the densification pattern itself. We attribute these layer-specific density increases primarily to compaction of accumulated snow, as surface melt is generally negligible at this site. The change in density within identified annual layers over the year increases and is highest (∼ 70 kg m⁻³) in layers 3 and 4 (2022 and 2021) and is relatively low within older layers (∼ 40 kg m⁻³ in 6–10 layers). The year-to-year density change of layer 5 remains minimal (∼ 30 kg m⁻³), despite the absolute density in layer 5 being higher. We attribute the higher absolute density to lower seasonal accumulation, summer melt and refreezing (year 2020 in Fig. ). Several studies have shown that the firn layers with initial lower density densify faster than the firn layers with initial higher density (e.g., ). This aligns with our results from the CMP3 location, where shallower layers (3–4) densify faster than older layers (7–10) in measurements from 2024 to 2025. This suggests that processes such as pressure sintering dominate densification in denser layers with densities above 550 kg m⁻³ .

According to Sorge's law, the density of snow does not change with time at a given depth below the surface, provided no melting occurs in summer . This is not valid at the CMP3 location. Although repeated CMP-derived density profiles show similar density fluctuations in the upper firn column, they show significant density offset below the 20 m depth. This difference in density profiles reflects the influence of variable seasonal accumulation, summer melt (Fig. ), and ice-lens formation (Fig. ), all of which disrupt the steady-state conditions required for validation of Sorge's law. Furthermore, we also observe the depth to firn density at around 7–8 m and pore close-off density at around 25–26 m in 2024 and 2025. A similar study by , but using direct observations (firn cores), addresses firn density evolution over a temporal scale (2016–2022) at Wolverine Glacier, Alaska, and demonstrates the importance of firn properties for glacier mass-balance estimates. The study illustrates that firn densification accelerated due to increased melt and refreezing, leading to reduction in pore space and enhanced firn compaction. The study also demonstrates that the substantial variability in firn and snow air content over 5 years has direct implications for geodetic mass-balance calculations. This further suggests the importance of investigations of mountain glacier firn to improve glacier hydrology and reduce uncertainties in geodetic mass-balance calculations . Here, we show that temporal firn variations can be detected and quantified with CMP methods, which is important for characterising the climatic impact on the firn body. Our CMP surveys resolve density–depth structure and IRH displacement across multiple locations, enabling spatially distributed estimates of firn evolution that complement the point-scale precision of firn cores. Although firn-core data remain necessary, CMP offers a faster approach with greater penetration depth than traditional, time and labour-intensive firn core methods.

The changes in firn density at the CMP4 location (Fig. ) are mainly driven by summer surface melt and refreezing in addition to compaction, because it is located at a lower elevation than CMP3 (Table ). The CMP-derived density profile indicates the higher densification at CMP4 (750 kg m⁻³ between 0 and 10 m depth) compared to CMP3 (600 kg m⁻³ between 0 and 10 m depth). This is further supported by the SP4 density profile depicting higher surface density. Absence of melt layers between 0 and 8 m depth from FC3 also indicates spatial variability in melt and refreezing processes between CMP3 and CMP4. These melt-related processes explain significantly higher densification rates, i.e., layer 2 densified from 520 to 680 kg m⁻³ and layer 4 from 600 to 700 kg m⁻³ (Fig. ). Studies such as and demonstrate firn density as a power-law dependency on time while the densification rate increases exponentially with pressure, which we also attribute to our results at this location. Further, a shift in depth to firn density over a year at the CMP4 location (Fig. b), which is approximately 7.5 m in 2024 and 3.5 m in 2025, indicates that the firn is densifying faster at shallower depths. This is explained by additional compaction between two measurement periods (Table ), predominantly due to intense 2024-winter accumulation, the refreezing of surface melt in 2024-summer, and an additional accumulation layer in 2025-winter. However, this is not the case at the CMP3 location (Fig. a), where surface melt is likely much lower, and percolation plays a minor role; thus, densification is significantly lower in layer 2 (520–550 kg m⁻³). This implies that the firn densifies faster at low-lying locations compared to firn located at higher elevations, demonstrating different climate settings at the two locations of the glacier accumulation area.

Our secondary results on the spatial changes in firn density (Fig. ) and accumulation (Fig. ) were estimated by interpolating density gradient and estimating the SWE (Sect. 4.4) between two CMP gathers at the upper and lower parts of the Ewigschneefeld (Fig. ) using picked IRHs from the long GPR transect acquired in 2025 (Fig. b). Similar to the 2024 results presented in , our new results show that spatial firn density increases with increasing radar profile distance and decreasing elevation. This further illustrates that the same firn layers at lower elevations are denser than those at higher elevations, whereas the thickness of annual layers reflects differences in spatial accumulation variation.

Our attempt to estimate the compaction rate in an Alpine glacier using CMP measurements will contribute new information to this poorly resolved issue. Figure a and b illustrate the estimated compaction rate and the corresponding change in density at two locations in the Grosser Aletschgletscher accumulation area, respectively. The change in density at the CMP4 location is higher in the near-surface annual layers 0 and 1, which represent 2023 and 2021, with the 2022 layer melted away. This results in higher compaction (0.3 m yr⁻¹) compared to older layers, likely due to reduced firn air content caused by overburden stress from the 2025 winter accumulation, summer melt (Fig. ), and refreezing of the record-high 2024 accumulation (∼ 3000 mm w.e. as in Fig.). The recorded point mass balance measurement near the CMP4 location indicates an annual mass balance (year 2024) at the end of the summer seasonal measurement is around 2000 mm w.e. Additionally, the meteorological data suggest that the mean surface temperature at 3570 m a.s.l. in 2024 July and August months was around 2.3 and 3.2 °C and in 2023 was around 1.4 and 1.0 °C , therefore, we expect a more pronounced surface melt at 3340 m a.s.l. near the CMP4 location. Assuming partial meltwater percolated down and refroze at a certain depth within the 2024 winter accumulated snow, and compacting it further, which increases the density from 450 to 620 kg m⁻³ (Fig. b). The process of refreezing in the 2024 summer and the accumulation of 2025 winter snow further compacted the 2023 layer from 530 to 680 kg m⁻³ and the 2021 layer from 600 to 700 kg m⁻³. This change in density corresponds to compaction rates of 0.3 and 0.26 m yr⁻¹ in the 2023 and 2021 layers, which we interpret as reasonable estimates. Compaction in the deeper layers is primarily driven by overburden stress from the denser 2023 and 2021 firn layers, yielding minimal compaction rates. This implies that the compaction rate decreases gradually with age; we estimated almost zero compaction within deeper layers (layers 4–6 in Fig. a), which are already denser than near-surface layers and require more time to compact further. Thus, we highlight the importance of surface meltwater percolation in resolving compaction rates within the firn pack. The same is highlighted by , suggesting the importance of considering the effect of meltwater percolation on compaction rate to improve firn modelling in mountain glaciers.

However, at the upper part of the Ewigschneefeld, where summer melt is relatively low, we expect minimal refreezing of the 2024 surface melt without producing ice lenses, which is also evident from our firn core results (no ice lenses, FC3 in Fig. ). This is observed from the density change (30 kg m⁻³) in the near surface annual layer 0, represents 2023 (Figs. a and b) as well as the compaction rate of around 0.19 m yr⁻¹ (Fig. 10a), which we ascribe to the overburden stress from accumulated snow. The presence of the firn layer 1 (2022 in Fig. ) at this elevation experienced an increased change in density (nearly 70 kg m⁻³) with an estimated compaction rate of 0.2 m yr⁻¹, predominantly due to a ∼ 9 m thick accumulation on top of this layer. We attribute the increase in change in density within layer 1 (2022) to the fact that firn density increases with pressure . Even though the change in density is decreasing in layers 2, 3, 4, and 5, whereas the compaction rate in certain layers (3 and 4) is increasing, this is inconsistent with our results from the CMP4 location. This discrepancy likely arises from uncertainty in picking denser but relatively thin layers (2020, 2019, and 2018 in Fig. ). We discuss these uncertainties in the subsequent section.

illustrate that firn compaction rates are primarily driven by climate and expected to vary spatially with surface climate conditions. Similarly, the study by emphasises region-specific compaction rates that depend on mean annual temperature, lower latitude, higher elevation, and higher annual accumulation. Therefore, direct comparison of Alpine compaction rates with polar studies is not meaningful. However, for the context, report compaction rate of ∼ 0.33 m yr⁻¹ in West Antarctica, while find ∼ 0.05–0.250.19 m yr⁻¹ near Ross Island, Antarctica. Further, we did not consider the effect of strain rate on our compaction rate estimate because finely resolved vertical strain rates within each identified annual layer are unavailable. However, we presented a quantitative estimate of the strain contribution to IRH displacement in the supplement by assuming a vertical strain rate, using available ice velocity data from and .

Figure  shows the depth-age (depth-firn layers) from the reference year 2023 (layer 0) to the commonly identified annual layers at the CMP3 and CMP4 locations (layer 8, 2015). The observed changes in depth to annual layers in 2024 and 2025 (Fig. a) are closely related to the estimated compaction rates at the CMP3 location (Fig. a). Annual layers appear closer to each other when compacted more (layers 0–4), and layers with minimum compaction are almost parallel to each other. This observation also holds at the CMP4 location, where the near-surface layers (0–4) are non-parallel and exhibit a higher compaction rate than the older layers (4–8). The comparison of the CMP3 and CMP4 depth-age plots shows that the cumulative depth of 7 layers at the CMP3 location is approximately 8 m, whereas at the CMP4 location, it is 6 m in 2025. This is mainly attributed to the ablation of the 2022 layer at this location. However, the total depth to layer 8 at both locations is nearly the same (11 m), indicating that layers 6–8 are thicker at CMP4 and relatively thinner at CMP3. We expect that the identified similar layer (layer 8) should appear at a greater depth at the CMP3 location than at the CMP4 location, due to better preservation of firn layers where surface melt is relatively less. The presence of the 2022 layer should also result in a higher cumulative thickness for all identified layers at the CMP3 location than at CMP4. We attribute this discrepancy to the uncertainty associated with the layer dating using Fe spikes (Fig. g) at the CMP3 location (section “Annual layer dating using firn core”) by comparing the CMP-derived SWE with SWE above the Fe peaks rather than considering the point mass balance observation (Fig. ). We discuss all uncertainties mentioned in this study in the following section.

Identification of CMP-derived IRHs as annual layers by chronologically comparing the CMP4-estimated SWE with point mass balance-derived SWE (Sect. 4.3) and firn core-based Fe peaks as summer horizon, and matching CMP3-derived SWE layers to FC3-estimated SWE (section “Annual layer dating using firn core” and Fig. g) requires the assumption of conservation of mass. This means that between 2024 and 2025, no mass loss was observed, implying surface meltwater at the CMP4 location refroze within the same layer. The same is also valid at the CMP3 location, where the surface melt is relatively low. This is supported by the estimated SWE in the respective layers in 2024 and 2025, which match well (R2 > 0.9; Fig. ) at both locations. The comparison of shallow firn cores from 2024 and 2025 (Fig. ) indicates that the conservation-of-mass assumption is valid, as identified ice lenses at ∼ 6.25–6.5 m depth in 2024 are still present in 2025 at ∼ 7–7.5 m depth. This indicates that despite the 2024 summer surface melt at the CMP2 location (at a similar elevation to CMP4 in Fig. ), the meltwater refreezed within the 2024 winter layer, with no evidence of internal accumulation in subsequent layers. Further, ice core results from the Jungfraujoch drilled in 2002 illustrate that the subzero borehole temperatures (-4 and -6 °C) in the upper part of the firn zone, and high annual net accumulation (1–2 m w.e.), suggest the negligible percolation of melt through underlying annual layers . These limited direct observations support our mass-conservation assumption to some extent. However, our measurements do not account for mass conservation during extreme melt seasons, such as the 2022 summer, due to the lack of data during those periods.

Previous studies suggest that refreezing within the seasonal snowpack occurs when preferential flow occurs through vertical piping . We hypothesise that higher surface meltwater during extreme summers (e.g., summer 2022) percolates deeper and refreezes within the subsequent firn layer, thereby altering the mass of the deeper layer through internal accumulation. These processes challenge our assumption of mass conservation when meltwater drains from the glacier system. However, tracking meltwater percolation, preferential flow, and refreezing complicates the understanding of firn, even though we can, to some extent, derive this information from deep firn cores and CMP measurements. For this reason, we propose continuous monitoring of the Alpine firn using strain meters , and radar techniques that are implemented to study polar firn, such as Autonomous Phase Sensitive Radar (ApRES), as stated in and .

exemplifies uncertainties in Vrms picking from Semblance analysis, which increases with depth, and also illustrates challenges in picking IRHs that generate high permittivity contrast, such as ice lenses and visible Sahara dust layers. As explained by , we assigned a sensitivity of 0.005 m ns⁻¹ for the Vrms picking from the semblance analysis, which resulted in a density fluctuation of 60–100 kg m⁻³. We observe that these uncertainties affect our density-depth profiles (Figs. , , and ). As depth increases, the uncertainty in Vrms picking increases, and so does the density estimation. Whereas uncertainty in compaction rate estimation also depends on the quality of the Vrms pick. Figure a shows the error bars corresponding to the 0.005 m ns⁻¹ for the Vrms picking, illustrating uncertainty in compaction rate related to the magnitude of the compaction rate. This means layers with a high compaction rate have higher uncertainty than those with a lower compaction rate. Other uncertainty lies with identifying the same IRH across multiple or repeated surveys, and any misidentification can amplify errors in compaction rates and annual layer depth estimates . The accuracy of picking IRHs from CMP gathers and identifying annual layers can introduce significant uncertainties in temporal changes in firn density and compaction rate, as well as in spatial accumulation estimates.

Furthermore, the thickness of identified layers is less sensitive to the Vrms pick than to the TWT pick. For instance, uncertainty in 1 ns picking changes the neighbouring layer thickness by 10 cm. This uncertainty amplifies when the layers are thin and close to each other, impacting the compaction rate estimations (Fig. a). Additionally, this is reflected in the deeper layers, as shown in Fig. , where the error bars are more pronounced at greater depth (layers 5–8). Similar uncertainties are discussed in , suggesting that bias in density measurements alters the TWT-to-depth estimations and emphasises density measurements as a potential source of error when deriving radar-wave velocities from firn-core-derived density estimations. In our study, we assign potential sources of error to radar velocity and TWT picking, which govern the identification of IRHs as annual layers.

Firn-core studies help attain high-accuracy, direct measurements of density, stratigraphy, and melt features, but are spatially sparse, time-intensive, and logistically demanding . In contrast, CMP-based GPR surveys yield spatially distributed density–depth profiles and IRH geometries to depths of ∼ 30–40 m , depending on the antenna frequency and requirement of vertical resolution. CMP method further allows us to map firn structure and compaction across the accumulation area . Our repeat CMP approach therefore complements firn cores by extending point-scale observations into a continuous framework. Our multi-year CMP observations help quantify temporal changes in firn density and firn compaction rates within the identified annual layers. The application of the CMP method, given its non-invasive, less laborious nature and ease of data collection , is well suited to challenging conditions such as high-mountain glaciers. Whereas Phase-sensitive radar systems (such as ApRES/pRES) provide precise, time-resolved measurements of vertical movement (strain rates) and firn compaction at single sites , however they require independent density data. We can address density data by applying the CMP method wherever the ApRES data are available, to further improve the temporal and quantitative resolution of the density profile to attain better firn compaction and strain rates. Considering the CMP ability to resolve radar velocity with depth to derive the density profile, we propose that a combination of CMP, firn cores, and ApRES as a complementary set of tools: firn cores offer ground truth and detailed process data, CMP provides radar velocity-based spatial and temporal measurements of density and compaction, and ApRES delivers high-precision temporal changes at key reference locations.