Recent waning snowpack in the Alps is unprecedented in the last six centuries

The study investigates how recent changes in Alpine snowpack duration compare with natural variability over centuries. Snow cover is crucial for the surface energy balance, hydrology, glaciers, permafrost, ecosystems and socio-economic sectors such as water supply and winter tourism. Instrumental snow observations in the Alps show marked declines in snow depth and snow cover duration since the 1970s, but long continuous records of snowpack duration are lacking, limiting contextualization of recent trends. The research aims to reconstruct a multi-century record of snow cover duration for the Alps using a biological proxy that sensitively reflects the timing of snow cover, thereby assessing whether the current waning snowpack is unprecedented in the last six centuries.

Alpine climate records include long instrumental series of pressure, temperature and precipitation extending to the mid-18th century, facilitating numerous tree-ring reconstructions focused mainly on summer temperature due to growth limitations in heat-limited environments. However, few studies address snowpack, and long-term continuous records of snow cover duration for the Alps have been absent. Observations show an 8.4% decline per decade in seasonal (Nov–May) mean snow depth and a 5.6% decline per decade in snow cover duration between 1971 and 2019 in the Alps. Elsewhere, tree rings have served as proxies for snow hydrology in moisture- or energy-limited systems, but high-elevation Alpine environments are generally not moisture-limited, complicating use of conventional tree-ring proxies for snow. Reconstructions of glacier length changes over the last five centuries show patterns potentially linked to snow and temperature variability. Documentary evidence and climate reconstructions of temperature and precipitation since 1500 CE provide contextual references for extreme seasons, enabling cross-validation of reconstructed snow duration anomalies.

Proxy development: 572 ring-width series were collected from prostrate Juniperus communis shrubs in Val Ventina, central Italian Alps (46°18′ N, 9°46′ E), between 2,100–2,400 m a.s.l., using basal discs from main stems. Multiple radii (2–4) per disc were measured; samples were prepared via sanding and, when necessary, microtomy (12–15 μm, safranin-stained). Crossdating followed a multi-step procedure (visual matching within and between samples, COFECHA checks). Ring-width series were standardized and combined using ARSTAN; to preserve low-frequency variability, Regional Curve Standardization (RCS) was selected after testing alternative detrending methods (Hugershoff, age-dependent spline) and stratifications (with/without missing rings, pith, living/relic). The final chronology used biweight robust means. Chronology signal strength and stability were assessed with EPS and Rbar over 30-year windows. Snow model and target series: Because long snow observations are sparse, the authors built a modelled snow cover duration series from daily temperature and precipitation. Solid precipitation (SP) was estimated using daily Tmax/Tmin with a 2 °C phase threshold, allocating fractions for mixed-phase events. Snowmelt (M) was computed with a degree-day approach using mean daily temperature and DDth = 0 °C. The snow degree-day factor (SDDF, mm °C⁻¹ d⁻¹) was derived empirically: for ~500 HS (snow depth) stations (subset of 312 quality-filtered; validation on 194 stations), SP and degree-day sums over periods with continuous snow cover (HS > 0) were used to estimate seasonal SDDF; SDDF was then interpolated to target sites using a weighted linear relationship with elevation and physiographic similarity (distance, slope, aspect, steepness, distance from sea). With interpolated SDDF and interpolated daily temperature/precipitation, daily SWE evolution was modelled and integrated to obtain snow cover duration (Oct–Sep). Validation: Leave-one-out cross-validation against HS stations yielded SDDF validation statistics r = 0.80, bias = 0.01 mm °C⁻¹ d⁻¹ (0.8%), MAE = 0.35 mm °C⁻¹ d⁻¹ (20.7%), RMSE = 0.48 mm °C⁻¹ d⁻¹ (28.4%). For snow cover duration, correlations between observed and independently reconstructed series across 135 stations (≥15 years of data) had mean r = 0.75 (median 0.79; IQR 0.69–0.85). Calibration and reconstruction: The juniper ring-width chronology was calibrated against modelled annual (Oct–Sep) snow cover duration at the study site (1834–2018). Bootstrap calibration/verification correlations had medians of 0.688 and 0.687, respectively; full-period calibration r = 0.687. The reconstruction (SALP) spans 1400–2018 CE. Mean and variance of the proxy were adjusted to the instrumental target to mitigate regression attenuation; confidence intervals were defined as ±2 RMSE from calibration. Skill metrics included explained variance (R²), reduction of error, coefficient of efficiency, and a bootstrapped transfer function stability test for slope, intercept and variance across subperiods (100,000 iterations). An extreme value capture test assessed ability to reproduce upper/lower decile events. Signal attribution: To separate frequency domains, the authors applied FFT-based high- and low-pass filters (cut-off 10 years) to ring-width indices and local Oct–May precipitation and Oct–Sep temperature, assessing correlations of high- and low-frequency components. Temporal structure was analyzed with 50-year smoothing and piecewise linear regression to identify trend changes. Spatial representativeness: The model was extended over a 30-arcsec DEM on the southern Alps to compare juniper chronology against HS station snow duration series.

– The reconstructed Alpine snow cover duration (SALP) spans 1400–2018 CE and exhibits high year-to-year variability, capturing both exceptionally long-lying (e.g., 1431, 1541, 1705) and short-duration (e.g., 1532, 1875, 2012) snow seasons. The long-term mean snow cover duration is 251 days. – A marked decline in snow cover duration begins around the late 19th century (piecewise regression), culminating in the first two decades of the 21st century (2000–2019) with a mean of 215 days, i.e., 36 days shorter than the long-term mean, unprecedented in the last six centuries. – Over the last five decades, the reconstructed trend is −5.4 ± 2.5 days per decade, consistent with observational estimates for the region when considering the full snowy period (−6.67 days per decade). – Distinct multidecadal episodes are identified: prolonged long-lying phases (e.g., 1440–1460, 1780–1800) and low-snow decades (e.g., 1940–1960). – The proxy signal blends high-frequency variability largely tied to cool-season precipitation with low-frequency variability more closely associated with annual temperature; low-frequency ring-width components are negatively correlated with temperature (r ≈ −0.59, p < 0.001) and high-frequency components show positive association with precipitation. – Calibration/verification against modelled snow cover duration is strong and stable (median r ≈ 0.688/0.687), with robust spatial validation of the snow model (station mean r ≈ 0.75 for duration). – SALP low-frequency variability aligns with reconstructed glacier cumulative length changes over the past five centuries, including a long stable phase from the 16th to mid-19th centuries followed by rapid retreat and a brief recovery around 1960–early 1980s. – Extreme event checks show consistency with documentary records and seasonal reconstructions: many of the most dry–warm and wet–cold composite years show correctly signed and often large (>35-day) anomalies in SALP; 1917 is identified as the most persistent snow cover year of the 20th century (+67 days vs mean).

The prostrate growth form of common juniper at high elevation causes growth to be constrained under snowpack, making ring width primarily sensitive to the timing of snow emergence rather than snow depth or SWE. Consequently, the juniper-based chronology effectively encodes snow cover duration, reflecting an integrated response to temperature (ablation) and solid precipitation (accumulation). The reconstruction demonstrates that recent shrinkage of Alpine snow cover duration is exceptional in the context of the last 600 years, providing a crucial baseline to interpret ongoing and future climate change impacts on hydrology, glaciers, permafrost, ecosystems and winter-dependent economies. The frequency-domain analysis clarifies that high-frequency ring-width variability tracks cool-season precipitation, while low-frequency variability tracks broader temperature changes, consistent with the observed coupling between SALP and glacier length changes. While some famous cold winters (e.g., 1740, 1985) are not strongly expressed in SALP, this underscores the nuanced interplay of precipitation timing, storm characteristics, and ablation processes—cold extremes or heavy snowfall do not always yield longer snow cover duration if subsequent melt or precipitation phase modulate persistence. The alignment with independent instrumental and documentary sources, along with robust validation of the snow model and stable proxy–target relationships, supports the reliability of the reconstruction. The results highlight the importance of integrating long-term proxies with modeling to capture interannual to centennial variability in snow dynamics, a key component of cryosphere–climate interactions and water resources.

This work provides the first annually resolved, 600-year reconstruction of Alpine snow cover duration, demonstrating that the recent waning of snowpack is unprecedented over the last six centuries. The approach leverages the growth sensitivity of prostrate juniper to snow cover duration and a validated temperature–precipitation-driven snow model to generate a long target series for calibration. The reconstruction captures both high- and low-frequency climate signals and aligns with observed trends and glacier dynamics. These findings underscore urgent needs for adaptation in sectors dependent on winter conditions and for refined representation of snow processes in land-surface and climate models. Given the wide distribution and longevity of common juniper and similar prostrate taxa, this methodology could be extended to other snow-prone regions to build spatially distributed reconstructions of snow duration. Future efforts should expand sampling across the Alps and other mountains, integrate additional proxies, refine snow modelling (e.g., energy balance components, wind redistribution), and couple reconstructions with climate model evaluations to assess model fidelity across time scales.

– The snow model uses a simplified degree-day approach, relying on temperature-driven melt and empirically interpolated SDDF; it does not explicitly resolve full surface energy balance terms, wind redistribution, or local microtopographic effects. – SDDF is inferred from HS station data using only the start/end dates of continuous snow cover and interpolated using elevation and physiographic similarity; uncertainties arise from station density, representativeness, and potential gauge undercatch. – The proxy site represents the southern Alps and one valley; while spatial comparisons indicate broader relevance, regional heterogeneity in snow climate may limit generalizability. – Juniper ring width primarily reflects duration rather than snow depth or SWE, and species-specific growth peculiarities (e.g., missing/wedging rings, eccentric growth) add noise, though tested and mitigated via RCS and robust averaging. – Some known extreme cold winters are weakly expressed in SALP, highlighting that single-season synoptic anomalies may not translate directly into longer snow duration if accumulation/ablation timing differs. – Reconstruction uncertainties are conveyed via ±2 RMSE; as with regression-based reconstructions, amplitude damping and non-stationarity are potential concerns, although stability tests and scaling were applied.