The study area is located on Mt. Ortles (3905 m a.s.l.; 46.508° N, 10.541° E), the highest peak in the Eastern Alps, within the Ortles-Cevedale Mountain Group (Fig. 1). The Alto dell'Ortles glacier lies on the north-western slope of the mount, extending from 3018–3905 m a.s.l. and covering an area of 1.19 km2 (2017). Ice thickness reaches up to ∼75m (Gabrielli et al., 2012), with the basal ice layers preserving a paleoclimatic record of the past ∼7kyr (Gabrielli et al., 2016, 2025). The glacier exhibits a polythermal structure, with temperate firn overlying colder ice below the firn-ice transition at ∼30m depth. The firn was risen to temperate conditions due to meltwater percolation, which was particularly high in some recent ablation seasons, for example during the 2003 heatwave (Beniston, 2004; Gabrielli et al., 2012).

Since 2008, this glacier has been the focus of the “Ortles Project” (https://ortles.org, last access: 26 August 2026), an international research program aimed at recovering deep ice cores for paleoclimate reconstruction and collecting data for multiple cryospheric components.

Mt. Ortles is located near the inner dry zone of the Alps and is subject to a continental climate. Long-term observations from Solda (valley floor, Fig. 1) indicate a mean annual precipitation of 800–950 mm between 1981 and 2010 (Adler, 2015). On the top of Mt. Ortles, annual precipitation is estimated between 1300 and 1400 mm, based on mass balance observation conducted from 2009–2016 (Carturan et al., 2023). This precipitation estimate may vary significantly across space due to the influence of the wind on snow accumulation and redistribution, and to the influence of spatial variations in precipitation.

An automatic weather station (AWS) operated in the upper accumulation zone of the Alto dell'Ortles Glacier from September 2011–June 2015. The solar-powered AWS was installed at 3830 m a.s.l., approximately 200 m downslope of the drilling site, and on a site with western exposure. A Campbell Scientific CR-1000 data logger stored 15 min values coming from sensors of air temperature and relative humidity (Vaisala HMP155A), wind speed and direction (R. M. Young 05103), incoming and outgoing shortwave and longwave radiation (Delta Ohm LP Pyra 05 and LP PIRG 01), and snow depth (Campbell Scientific SR50A). Some sensors have been duplicated to provide a backup in the event of a malfunction. Details on sensors characteristics have been reported in Table 1 whereas details and pictures of the AWS can be found in Carturan et al. (2023).

The AWS was mounted on an aluminium tower, composed of two-meter segments and anchored two meters deep into the snow, supported by wooden platforms at the bottom. Once installed, the tower rose 4 m above the snow surface. Sensors and photovoltaic panels were placed on the top of the structure, while the data loggers and batteries were housed in a fiberglass box just above the snow surface. Because of the snow accumulation, the height of the sensors relative to the glacier surface varied seasonally. To ensure continuous operation and avoid sensor burial by net accumulation of snow, the tower was extended each year by adding a two-meter segment each time.

Precipitation could not be measured with reasonable accuracy at the Ortles AWS, because it would have required a heated rain gauge with too high-power demand. In addition, due to the high elevation and wind exposure, any data collected would have been highly underestimated and difficult to correct, due to the well-known issues (high wind speed, high frequency of snowfall, rain gauge perturbations of the wind field) affecting precipitation measurements in similar environments (Rasmussen et al., 2012; Kochendorfer et al., 2017). Consequently, a rain gauge was not installed at the AWS, using and scaling instead precipitation data measured at the nearest weather station in Solda (1907 m a.s.l.), operated by the local meteorological agency (Ufficio Idrografico, Provincia Autonoma di Bolzano, https://meteo.provincia.bz.it, last access: 26 August 2026).

The raw meteorological data from Mt. Ortles and Solda were quality-checked and validated against observations from neighbouring weather stations (details are reported in Carturan et al., 2023), specifically Madriccio (2825 m) and Cima Beltovo (3328 m) for air temperature and wind speed, and Fontana Bianca (1900 m) for precipitation. Precipitation data from Solda were also corrected following the procedure described in Carturan et al. (2012,) to account for the rain gauge underestimation. Since the SNOWPACK model requires a continuous meteorological input in the investigated period, missing or unreliable data were reconstructed using backup sensors at the AWS when available, and data from Madriccio and Cima Beltovo weather stations in case backup sensors were unavailable or unreliable. In this case, the missing data at the Ortles AWS were calculated by means of linear regression from Madriccio and Beltovo weather stations. For wind speed and direction, data gaps (7 % of the total period of observation, Table B1 in the Appendix) were filled with seasonal-mean values. Gap-filling was carried out using the MeteoIO/SNOWPACK toolchain (Bavay and Egger, 2014). The obtained final dataset is shown in Fig. 2.

At 3850 m a.s.l. the mean annual air temperature is approximately -9°C. Snow accumulation occurs mainly from October to May. The melt season typically spans from June to September, though summer snowfalls are frequent on the glacier summit. The intensity and duration of ablation exhibit substantial differences from year to year and are strongly influenced by heatwaves. Rainfall is rare at these elevations, but isolated rain events have been observed on the summit over the past 15 years (Carturan et al., 2023).

In the period of direct observations, between September 2011 and June 2015, the mean monthly air temperature ranged between -17.24 and -1.63°C, peaking in August (Fig. 3a). The precipitation showed a distinct maximum in summer and a minimum in winter, which is typical of the local climate (Fig. 3b). The wind speed was highest in winter and lowest in spring and in September. The dominant wind direction was from W–SW (Fig. 3c), with significant variations considering all hours of observations (Fig. 4a), hours with precipitation (Fig. 4b), hours of erosion during precipitation (Fig. 4c), and hours of erosion after precipitation (Fig. 4d).

Seasonal and annual glacier mass balance measurements were carried out between June 2009 and September 2014 at the ice core drilling site and at the AWS location. Winter balance surveys were performed in June or early July before the start of the ablation season. Summer and annual balance measurements took place in late August or early September, at the end of the melt season. Field observations included snow depth probing around both sites, combined with snow and firn density measurements in snow pits dug down to the previous summer surface. Snow stratigraphic profiles were carried out along shaded pit walls, including observations of snow and firn temperature, hardness, grain type and size, and the presence of ice lenses and dust layers (Gabrielli et al., 2010). These stratigraphic observations were helpful in identifying the summer surface, both within snow pits and while probing snow depth. Density measurements were used to convert snow depth into water equivalent.

We used the snow cover model SNOWPACK (Lehning et al., 2002a, b) to calculate the energy and mass balance at the Ortles AWS site and to obtain a continuous snow water equivalent data series at the study site in the period from September 2011–June 2015. This data series was required for this study, but could not be derived directly from snow depth data because snow density (and its variations) were unknown. Originally developed for avalanche warning, SNOWPACK is widely used to study the alpine cryosphere, including snow and glacier mass balance and mountain permafrost (Meirold-Mautner and Lehning, 2004; Obleitner and Lehning, 2004; Rasmus et al., 2004; Luetschg et al., 2008; Bavay et al., 2009, 2013). SNOWPACK is a physics-based, one-dimensional, multi-layer numerical model designed to simulate the evolution of the snow cover using meteorological data from automatic weather stations. The model simulates the stratigraphy and the microstructure of the snow by modelling energy and mass fluxes. It employs a Lagrangian finite-element method to solve the heat transfer equations. A key feature of SNOWPACK is its detailed representation of snow metamorphism, which is closely linked to mechanical properties such as the thermal conductivity and viscosity. The model also accounts for the interactions between the snow surface and the atmosphere, including radiative transfer processes such as the penetration of shortwave radiation. The surface energy exchange in SNOWPACK is detailed in Lehning et al. (2002a). A fixed roughness length of 0.002 m has been assumed together with the Holtslag stability correction in a bulk aerodynamic Monin–Obukhov formulation for the turbulent fluxes. The standard SNOWPACK parameterization for albedo has been employed.

The computational framework is based on a set of key state variables that serve as the foundation for simulating the snow microstructure and its metamorphic changes. These variables are used to derive essential bulk properties needed to solve the model core conservation equations. A detailed description of the mathematical formulations and governing equations implemented in SNOWPACK can be found in Bartelt and Lehning (2002), and in Lehning et al. (2002a, b).

SNOWPACK explicitly accounts for wind-driven snow erosion through its snowdrift module (Lehning et al., 2000). The scheme first determines a threshold friction velocity (ms-1) at the snow surface following Schmidt (1980), based on its erodibility calculated as a function of snow grain shape, size and bond strength (Lehning et al., 2000). When the actual friction velocity, derived from the logarithmic wind profile including stability corrections, exceeds the threshold friction velocity, snow erosion initiates. The horizontal drifting snow mass flux is then calculated following Lehning and Fierz (2008). To translate the horizontal mass flux into a vertical surface erosion mass flux, a fetch length of 70 m is assumed, which is an arbitrary length scale under the assumption that the mass that is transported (e.g. over a mountain ridge) is then distributed along the fetch distance on the lee side (Lehning and Fierz, 2008; Wever et al., 2023). At each time step, the erodibility of the surface layer is updated, allowing SNOWPACK to consistently represent erosion also during snowfall.

To run the model the following meteorological variables were used as inputs: air temperature, relative humidity, wind speed, incoming and outgoing shortwave radiation, incoming and outgoing longwave radiation and precipitation. All input data came from the Mt. Ortles AWS, with the only exception of precipitation that comes from the Solda village, as detailed in Sect. 2.2. The model framework relies on the MeteoIO library (Bavay and Egger, 2014) for meteorological data pre-processing. This module acts as a middleware layer between raw meteorological data and the model, pre-processing input data. Gap filling methods are mentioned in Sect. 2.2; in addition, the MeteoIO module adjusts input data for the varying sensor's height above the surface, based on readings from the snow depth sensor.

A point simulation of the snowpack was conducted at the AWS site using a 15 min time step interval. The simulation began on 7 September 2011 and extended until 14 of June 2015. SNOWPACK was initialized using snow pit observations collected at the AWS site a few days before the start of the simulation. These observations extended from the glacier surface down to 400 cm and included snow and firn stratigraphy (grain size and shape), density, and the temperature profile. The snow accumulation was simulated using the Solda precipitation, corrected for gauge undercatch as explained in Sect. 2.2, and then adjusted using a 28 %km-1 vertical precipitation lapse rate, obtained from 40 years of meteorological and snow water equivalent data in a nearby study site (Val de La Mare, 10 km south-east of Mt. Ortles, Carturan et al., 2019). In this way, the vertical precipitation lapse rate was derived from independent meteorological observations and was not used as a calibration parameter. A fixed lapse rate was preferred to a month-specific lapse rate (as employed by Carturan et al., 2019) because the seasonality of lapse rates is expected to vary significantly across the region, in particular during summer when thunderstorms prevail, but also considering the orographic characteristics. As we have no information on how monthly lapse rate might change in the Ortles area, compared to Val de La Mare, we consider it more appropriate to use a fixed lapse rate.

The wind speed was also adjusted by means of a multiplicative factor, to account for the fact that on top of Mt. Ortles the wind speed is not expected to follow strictly a vertical logarithmic profile as in horizontal, unobstructed terrain (Mott et al., 2010; Stiperski and Rotach, 2016). The multiplicative factor for wind speed was adjusted iteratively to minimize the Root Mean Square Error (RMSE) between measured and modelled mass balance at the simulation site.

The snow erosion process was analysed in terms of snow water equivalent (SWE), rather than in terms of snow height, allowing changes due to wind redistribution to be isolated from those caused by other processes, like compaction. We excluded from analyses the periods with interpolated wind speed data (7 % of the total, Table B1), to avoid introducing possible external noise to the data.

The analyses regarded hourly snow erosion events, identified using the SNOWPACK output variable MS_Wind (mmh-1), which describes the mass loss during erosion. An hourly snow erosion event was identified when MS_wind exceeded 0.1 mmh-1. Each hourly snow erosion event was characterized using descriptors that quantify the meteorological conditions during erosion, and the average conditions during the “life” (abbreviation used in this study to indicate `lifetime') of each snow layer that was removed by wind erosion (Table 2). The beginning of each layer's life was set when snow accumulation exceeded 0.1 mmh-1. The duration of each new layer's life was up to the end of the simulation period, unless the layer was removed by melt, erosion, or sublimation/evaporation.

Because the mechanical response of snow to wind differs under wet and dry conditions, erosion events were classified accordingly. “Wetted” snow was defined as snow exposed to positive air temperatures since layer formation, representing a cohesive wet regime, even if wetting was followed by refreezing of liquid water (because in that case the snow surface keeps high mechanical strength). “Dry” snow, in contrast, was defined as snow that never experienced positive air temperatures. Based on this criterion, 1731 dry snow erosion events and 155 wetted snow erosion events were identified.

The snow erosion process was analysed using correlation and frequency distribution analyses. Two correlation matrices were calculated for dry snow and other two for wetted snow: (i) the first dry/wetted snow matrix including all hourly erosion events, and (ii) the second dry/wetted snow matrix that includes the threshold wind speed for snow erosion (WS_Th) and the temperature threshold for snow erosion (T_Th) variables, which could be calculated only when erosion starts (i.e. they could not be calculated over consecutive hourly erosion events).

To assess the sensitivity of wind-driven snow erosion to changes in air temperature, we performed additional SNOWPACK simulations applying a fixed offset to the input temperature data. Six scenarios were simulated, shifting the observed air temperature data by -3,-2,-1,+1,+2, and +3°C, relative to the baseline (i.e. the observation period from 2011–2015). With the word “scenario” we mean a mere increase or decrease in air temperature, which is used to assess the model's sensitivity exclusively to air temperature perturbations. This approach, even if simplified, aims to investigate the response to both colder conditions (representative of past centuries) and plausible near-future warming at high elevation in the Alps (IPCC, 2022).

All other meteorological data were kept unchanged considering the lack of significant trends in long-term observation series and high uncertainty in future projections (Brunetti et al., 2009; De Blasi, 2018; IPCC, 2022), as discussed in Sect. 5.1. In this framework, the word “scenario” does not refer to past or future shifts in climate conditions, which would have implied modifications in all meteorological variables.

Each experiment covered the same period as the reference run (7 September 2011–14 June 2015).

For each scenario, we analysed the model output in terms of snow water equivalent, the change in the relative contribution of erosion, melt and net accumulation to the total mass balance, and the change in the monthly regime of snowfall erosion by wind. We included also sublimation and evaporation fluxes to assess their relevance in comparison to the other mass fluxes.

The RMSE between measured and modelled mass balance at the simulation site was optimized using a multiplicative factor for wind speed of 0.70. Considering all mass balance measurement sub-periods, the RMSE between measured and modelled SWE above the previous year's summer surface is 0.115 m w.e. The agreement between measured and modelled mass balance was higher in the accumulation season (RMSE=78mmw.e.), compared to the ablation season (RMSE=209mmw.e.). An average overestimation of 182 mm w.e. was detected in summer balance, which occurred mostly in 2013. The winter balance was underestimated by 52 mm w.e. on average (Fig. 5).

According to SNOWPACK calculations, 4.401 m w.e. accumulated at the study site between 17 September 2011 and 23 September 2014. The annual accumulation rate was lower in the 2011/2012 balance year and higher in the 2012/2013 and 2013/2014 balance years, with similar rates in the latter two years. The largest part of snow accumulation occurred during the cold season (from October–March), while the warm season (from April–September) still contributed, with summer gains ranging from 0.04 m w.e. (2012) to 0.49 m w.e. (2014), whereas 2013 experienced a net summer ablation of 0.24 m w.e.

The build up of the snowpack modelled at the Mt. Ortles AWS site from 7 September 2011–14 of June 2015, and the associated snow layering, is displayed in Fig. 6a. Figure 6b shows a close-up period between 13 and 17 December 2011, when several snow erosion events took place in a sub-freezing temperature regime. Two major events can be recognized: the first occurred on 13 December, after a snowfall the day before that was accompanied and followed by low wind speed. Wind erosion exclusively regarded fresh snow and was triggered when the hourly-mean wind speed rose above 10 ms-1, ceasing when it fell below 12 ms-1. The second major erosion event, removing 35 mm SWE (7 times higher erosion compared to the first event) was triggered when the wind speed rose above 15 m s-1 on 16 December, and was characterized by a much higher wind speed, averaging 22 ms-1 with hourly-mean peaks above 27 ms-1. This second event eroded much older snow, including 42 d-old snow which was also wetted (even if with a very small positive degree hours equal to 1.6 °C).

In the period between 2011 and 2015 (baseline conditions), the mass balance was dominated by snow accumulation and wind erosion, which removed 20.9 % of the total accumulated snow. Melt was comparatively less important and removed only 2.9 % of total snow accumulation (Table 3). Evaporation and sublimation removed 326 mm (5.5 % of total snow accumulation), whereas condensation and inverse sublimation added 43 mm, giving a net flux of -283mm (-4.7% of total snow accumulation).

The sensitivity of SNOWPACK simulations to the simulated temperature changes was remarkably higher for warming scenarios, compared to cooling scenarios. (Fig. 12). The modelled mass balance was little affected by warming up to 1 °C above baseline conditions, which caused a decrease of 11.5 % in the final SWE. However, a larger warming had a much stronger impact, causing a 38.5 % smaller final SWE for 2 °C warming, and a 71.8 % smaller final SWE for 3 °C warming. In contrast, cooling scenarios produced a minor increase of 0.4 % in the final SWE for a 1 °C temperature decrease, whereas the final SWE decreased under larger cooling (-2.2% for 2 °C cooling and -3.8% for 3 °C cooling).

Up to 1 °C warming, wind erosion (represented by black vertical bars in Fig. 12) remained dominant in removing accumulated snow (18.7 %) over melt (12.5 %, represented by red vertical bars in Fig. 12). Larger warming implied strong melt increase and a significant decrease in the relevance of snow erosion (16.5 % and 14.9 % of the total accumulation for 2 °C and 3 °C warming, respectively). On the other hand, melt decreased almost to zero with 1 °C cooling, whereas the share of total snow accumulation removed by wind erosion increased about 2 % for each degree of cooling, reaching 27 % for the -3°C scenario. Fluxes related to water vapor showed a lower sensitivity to changing temperature scenarios. Evaporation and sublimation losses increased by 1.2 % under the largest cooling scenario, and decreased of the same amount under the largest warming scenario. Changes in condensation and inverse sublimation were negligible and reached a few tenths of a percent.

For baseline conditions, the seasonal regime of snow erosion (Fig. 13) was characterised by a prominent maximum in winter months, when 48 % of monthly snowfall was eroded on average, and a minimum in spring, when 11 % of monthly snowfall was eroded on average. In summer and fall, the wind erosion averaged 16 % and 17 % of the monthly snowfall (Fig. 13 and Table A1). The highest sensitivity of wind snow erosion to warming temperature was found in spring and fall, with a maximum decrease of about 61 % in March for the +3°C scenario. High sensitivity was calculated in mid-summer, whereas lower sensitivity is visible in the coldest months. On the other hand, colder temperatures affected mainly the months of May, August, September and November. In particular, for August a nearly doubling of wind erosion was calculated for the -3°C scenario. The sensitivity was smaller during winter months and in April and June.

In this work, we used the model SNOWPACK to investigate the snow erosion on top of the Alto dell'Ortles glacier, based on mass balance and nivo-meteorological observations collected between 2011 and 2015. It was not possible to analyse the snow erosion directly from the data collected by the snow-depth sensors, because several other processes are involved in snow depth variations, such as sublimation, melt, compaction, and because the density of the surface layers is unknown. This made impossible to quantify erosion mass fluxes just from automatically measured snow depth and required the application of a physics-based model, constrained by field measurements, to disentangle snow erosion from other processes.

The SNOWPACK model reproduced well the measured mass balance variability across four years at the study location, once the wind speed had been adjusted to account for deviations from smooth, logarithmic profiles over complex topography (Panofsky and Zhao, 1983), and the precipitation measured at the valley floor had been corrected for rain gauge errors (Carturan et al., 2012) and for the vertical lapse rate.

The requirement for a precipitation vertical lapse rate is justified by the orographic enhancement of precipitation due to the forced orographic uplifting of air masses on windward slopes of mountain ranges, which enhances the condensation of water vapor and the precipitation (e.g., Roe, 2005; Houze, 2012; Avanzi et al., 2021).

Similarly, the adoption of a multiplicative factor of 0.7 (reduction factor) for the measured wind speed is physically consistent with knowledge on wind speed-up effects over mountain tops and ridges (Jackson and Hunt, 1975; Raderschall et al., 2008). On Mt. Ortles, the wind sensor was exposed to a higher relative speed-up effect than what occurred close to the snow surface. Literature reports speed-up ranging between 30 %–100 % in these settings compared to the logarithmic wind profile, with a maximum speed-up effect occurring within the first few meters above the ground, which is where the wind sensor was located at the study site (Taylor et al., 1987; Pellegrini and Bodstein, 2004).

In absence of higher time resolution snow water equivalent data from the field, we adopted constant values for the precipitation laspe rate and for the wind multiplier. Consequently, despite the overall good performance the model, it is possible that the model was affected by the time invariance of the two parameters for single events or periods of simulation.

For example, there is very good correspondence between modelled and measured seasonal balances (Fig. 5), except for summer 2013, when the model overestimated the mass balance by 0.33 m w.e. In this case, SNOWPACK calculated too much snowfall on Mt. Ortles, which increased surface albedo and inhibited melt. This was due to the seasonal variability of the precipitation lapse rate and type of precipitation. In summer, the study site is mainly affected by local convective precipitation (thunderstorms) that are scattered by nature, making precipitation variability more randomly distributed and less correlated with elevation when compared to frontal precipitation, which is more sensitive to orographic uplift (Adler, 2015). Using a constant precipitation correction factor led to significant overestimation of summer snowfall and mass balance in 2013, however possible errors in 2013 mass balance measurements in the field cannot be ruled out as well.

Low accuracy affecting mass balance measurements on Mt. Ortles can be mostly related to the difficult detection of the previous year's summer surface during snow depth soundings. The related deviation from the true value can be potentially large (up to 0.5 m w.e.) but was significantly reduced by regularly digging pits into snow and firn, which enabled detection of the previous year's summer surface with good confidence based on observations of grain shape and size, dust layers and melt/refreeze crusts. This check excluded large uncertainties in the evaluation of the annual snow depth. However, it could not completely remove inaccuracies linked to the spatial variability of snowpack thickness and density. Overall, we estimate a random uncertainty of 0.1 m w.e. per mass balance measurement, which lies within the range reported in literature for measurements under similar conditions (Zemp et al., 2013).

Inaccuracies in nivo-meteorological variables measured by the AWS may also have contributed to the total uncertainty of the model's output. Considering these sources of possible uncertainty (Sect. 3.1) and the extreme environmental conditions at the study site, the RMSE between measured and modelled seasonal and cumulated balance (0.115 m w.e.) looks acceptable. However, we preferred to remove from the presented model results periods with data gap filling to minimise external and spurious sources of variability in the modelled SWE series and snow erosion events analysed in this work, while still preserving a sample size that was sufficient for attaining statistical significance.

Using simple scenarios of changed air temperature is a simplification of future warmer climates (e.g., Dumont et al., 2025) and is mainly intended to assess the model's temperature sensitivity, as outlined in Sect. 3.3. However, a change in air temperature represents the main effect as climate models predict small changes in other meteorological variables such as precipitation or radiation (Kotlarski et al., 2023). The main additional effect of a warmer atmosphere in the surface energy balance is through increased longwave incoming radiation, which is captured in our simulations as this forcing is parameterised in SNOWPACK (Schmucki et al., 2015). Relative humidity can be assumed unchanged as suggested by secular meteorological observations at high elevation in the Alps showing the absence of clear trends (Brunetti et al., 2009). In addition, a minor (2 %–4 %) decrease in relative humidity is indicated by future projections at the end of the 21th century in the European Alps (Gobiet et al., 2014), but they are highly uncertain due to the relationship between the relative humidity and precipitation, and to the high uncertainty in future trends of precipitation in the study area (Kotlarski et al., 2023; Pepin et al., 2025).

A possible limitation of the study is the relatively short observation period. This may affect, for example, the representativeness of threshold wind speed values for wetted snow erosion (due to the smaller sample size compared to dry snow), or the robustness of monthly erosion regimes (which fluctuate due to the existence of single months with unusual meteorological behaviour). Nevertheless, in view of how rare this high-elevation four-year dataset is, we consider it suitable to characterize the key processes affecting snow erosion and its temperature sensitivity at the study site.

At the Mt. Ortles AWS site and during the investigated period, snow erosion played a relevant role in regulating the surface mass balance, removing 21 % of the total accumulated snowfall. Snow erosion (associated with sublimation of drifted snow, Table 3) represented the dominant ablation process at this site, whereas melt only removed 2.9 % of the accumulated snow during the modelled period. The western exposure enhanced snow erosion, as it frequently resulted in the AWS site lying on the windward side of the mountain (Figs. 1, 3c and 4a). Several other studies documented the importance of wind exposure of mountain slopes compared to dominant regional wind fields, as a direct control on local snow accumulation and location of glacier accumulation areas (e.g. Humlum, 1987).

Snow erosion was also promoted by high elevation, which resulted in air temperature below the freezing level most of the time (Fig. 2a). Air temperature had a dominant effect on the susceptibility to snow erosion at the glacier surface, because it was directly related to snow metamorphism and, in particular, to the formation of bonds between snow crystals and of melt-and-refreeze crusts. Dry snow was consequently more frequently eroded when compared to wetted snow, regardless of the wind speed causing drifting (Fig. 9). Above 0 °C snow erosion becomes negligible (Fig. 10b).

The threshold wind speed for the initiation of erosion was highly dependent on the snow thermal regime. A wind speed of about 8–9 ms-1 was required to achieve a frequency of dry snow erosion significantly above zero, while this nearly doubles for wetted snow (Fig. 10a). These values lie towards the upper end of the range reported in literature, which is between 4–11 ms-1 for dry snow and 7–14 ms-1 for wet snow (Li and Pomeroy, 1997b). The high threshold wind speed on Mt. Ortles may depend on the high wind speed during precipitation (6.7 ms-1 on average), which increased the density and cohesion of fresh snow during deposition (Liston et al., 2018).

Precipitation events represent a source of ice particles that can create initial saltation and consequently a lower threshold for wind transport (Schmidt, 1980). This process was likely responsible of the high amount of snow erosion calculated during precipitation at the Mt. Ortles AWS. Indeed, more than half of the snow erosion occurred during precipitation (53 %), or shortly afterwards (Fig. 8). This was somehow unexpected because during precipitation the atmospheric wind flow in this region is mostly from South or South–East (Fig. 4b), creating lee-side conditions at the AWS site. In fact, most wind erosion was expected to occur during northerly windstorms that frequently follow atmospheric disturbances in this region, especially during winter. In contrast, a prevailing wind erosion by south-easterly or south-westerly winds occurred, during or shortly after weather disturbances (Fig. 4c and d) and well before the wind rotation to north (generally associated with the end of the precipitation events).

Snow erosion on Mt. Ortles was mostly a winter phenomenon, due to the combination of high wind speed and low temperature. On average, 48 % of the accumulated snow was removed by erosion in winter, whereas this amount ranged between 10 % (in March and May) and 21 % (in October) during the other parts of the year. The monthly regime of precipitation, with a maximum in the warm season (Fig. 3b), was more favourable to snow preservation than it could have been if precipitation were equally distributed over the year, or with a winter maximum. Under these conditions, the AWS site would have experienced a significantly lower annual snow accumulation.

The sensitivity of snow erosion to temperature fluctuations is expected to be highest on climates characterized by a summer maximum in precipitation, like on Mt. Ortles, because of the proximity of air temperature to the melting point. Our calculated monthly estimates confirm the highest sensitivity of snow erosion during the warm and transition seasons, and low sensitivity during winter when air temperature is far from the melting point, even under the +3 °C scenario (Fig. 13).

Based on the reported 2±0.3 °C warming experienced by the European Alps since 1850–1900 (Dumont et al., 2025, and references therein), the -2°C scenario should represent thermal conditions in the second half of the 19th century. Compared to that colder period, current conditions lead to a calculated reduction of snow erosion by 22 % (29 % for a thermal depression of 3 °C) (Fig. 12 and Table 3). At the same time melt increased only from 0.2 %–0.3 % to 2.9 % of the total accumulated snowfall. Therefore, based on these calculations, the reduced snow erosion with increasing temperatures acted as negative feedback, increasing the annual accumulation rate at this site. Our calculations thus confirm the negative feedback deriving from a moderate temperature increase, which could increase the annual accumulation rate at high elevation in the Alps, as suggested for example by Haeberli and Alean, (1985).

A further small increase in air temperature compared to baseline conditions is expected to strongly overcome this negative feedback in the near future (IPCC, 2022). With just 1 °C additional warming (already partially occurred since 2015, Jacquemart et al., 2024), there would be more than a fourfold increase in melt and only a 2 % decrease in snow erosion. When combined with a 2.5 % decrease in the fraction of solid precipitation, this would result in a significant decrease in the annual accumulation rate (Table 3 and Fig. 12). Therefore, a “tipping point” for the wind erosion feedback is approaching at the AWS site. It is also worth noting that areas at lower elevation with similar or higher exposure to wind erosion probably already experienced this critical temperature increase in the last decades and quickly transitioned from an erosion-dominated to a melt-dominated mass balance regime.

This study highlights and quantifies the high sensitivity of glacier mass balance to snow erosion, and of snow erosion to air temperature fluctuations, at this specific study site. These interactions may lead to a non-linear response of glacier mass balance to climate change, and to negative feedbacks that should be taken into account when projecting the future behaviour of similar, high elevation and wind exposed glacierized areas. Comparable interactions are likely to occur also in non-glacierized areas, affecting for example the thermal regime of the ground and the behaviour of the underlying permafrost.

We conclude that snow erosion processes and feedbacks should also be considered when paleoclimatic data recorded in the alpine cryosphere are used for reconstructing past environmental conditions such in the case of ice cores retrieved from high elevation glaciated areas with characteristics similar to the one analysed in this study. Like melt, wind erosion is potentially effective in removing a large fraction of the seasonal accumulated snow. On Mt. Ortles, this can occur especially during the coldest months, which are also the driest and windiest. In this case, for instance, a significant fraction of the isotopic winter signal might be missing from the Ortles ice core records. In addition, it is likely that during colder climatic phases, such as the Little Ice Age, an even lower winter “imprint” is preserved in ice cores, because wind erosion increases at lower temperature.

For these reasons, we argue that proxy system models aimed at improving dating and interpretation of ice cores, could benefit from the inclusion of simple parameterizations of snow erosion, in addition to melt (e.g. Carturan et al., 2025). This refinement looks essential for a realistic modelling of the formation and preservation of paleoclimatic proxies such as the isotopic signal and its changes through time.