Fig 1.
Schematic picture depicting some of the fundamental aspects and processes by which the cryosphere and atmosphere in snow-covered regions impact global climate, ecological, and social systems (modified from Intergovernmental Panel on Climate Change [18]; https://www.ipcc.ch/srocc/chapter/chapter-3-2).
Variations in snowfall and snow-cover, mediated by complex teleconnections and feedbacks, have significant implications for various Earth system components, including water supplies, ecosystems and global temperature dynamics [19]. Teleconnected weather patterns provide critical insights into the drivers behind changes in snowfall, both encompassing long-term trends and short-term atmospheric circulation variability [20], thereby enhancing climate prediction capabilities.
Fig 2.
Environmental setting and snow outline.
a. Swiss region setting (red square) adapted from Shaded Relief (http://www.shadedrelief.com/natural2/globes/africa.jpg); b. Pre-alpine region (yellow areas) adapted from the National Centre for Climate Services, MeteoSwiss (https://www.nccs.admin.ch/nccs/en/home/regions/grossregionen/pre-alps.html); c. Mean annual snowfall days across Switzerland (1981–2010), with the Pre-Alpine region outlined in white, adapted from the National Centre for Climate Services-MeteoSwiss (https://www.nccs.admin.ch/nccs/en/home/climate-change-and-impacts/swiss-climate-change-scenarios/facts-and-figures/climate-iindicators.html); d. Winter cycle of heavy snowfall days (black line), snow amount (histogram) and rainfall days (dashed blue line) per weeks in each month from October to May, averaged on the period 2007–2022 for the Swiss Valley, adapted from Snow-Forecast (https://www.snow-forecast.com/resorts/Swiss-Valley/history).
Fig 3.
Swiss pre-Alpine region: heavy snowfall days (NHSD) data (1884–2023).
a. Annual NHSD (blue line) with a significant linear trend (grey dashed line); b. Detrended residuals (violet line) overlaid with an 88-year cosinusoidal filter (red line), emphasising periodic trends; c. NHSD conditional variance, estimated using an 11-year moving window. The vertical scale of the band-pass filter in b is magnified threefold to improve phase visibility in relation to periodic trends.
Fig 4.
Analysis of the number of heavy snowfall days across the Swiss pre-Alpine region.
a. Autocorrelation function (ACF), and b. Partial autocorrelation function (PACF) on the standardised residuals; featuring horizontal black lines indicating the 95% confidence level limits; c. Time-series attractor in the phase-space domain, showing the trajectory of NHSD (blue) and the distinctive ellipse-like branching (red contour) for the period 1884–2023.
Fig 5.
Visual representation of the workflow illustrating the criteria for the PARMAX(TVAR)-CSD model criteria employed in predicting the NHSD for the Swiss pre-Alpine region (image of snowy landscape derived from Freepik, https://www.freepik.com/free-vector/watercolor-background-with-winter-landscape_1433112.htm#query=snowfall%20landscapes&position=1&from_view=search&track=ais&uuid=3a748666-b7de-4ee9-b1ba-1bd3383c2928).
Fig 6.
Climate drivers of heavy snowfall in Switzerland:
a. Scatter-plot illustrating the linear relationship between the Climate Forcing Index (CFI) and the annual number of heavy snowfall days (NHSD) in the pre-Alpine region (1884–2023), with 90% (dark pink) and 95% (light pink) confidence limits; b. Spatial correlation between Arctic Oscillation (AO) sea-level pressure (https://www.atmos.colostate.edu/~davet/ao/Data/ao_index.html) and snow cover (1966–2022); c. Same as b, but for the Dipole Model Index (DMI) HadSST1 sea-surface temperature dataset (Hadley Centre Sea Surface Temperature dataset version 1; https://www.metoffice.gov.uk/hadobs/hadisst). Panels b and c are based on Rutgers University Climate Lab snow-cover data (https://climate.rutgers.edu/snowcover) and were generated using Climate Explorer (http://climexp.knmi.nl; [84]). The black box in both maps highlights the Swiss region.
Fig 7.
Comparison of modelled (orange) – Eq. (2) – and observed (blue) NHSD annual time-series in the Swiss Pre-Alpine region, with change-points in 1988 (observed) and 1986 (joint modelled).
Table 1.
Performance comparison of PARMAX(TVAR)-CSD and PARMA(TVAR)-CSD models during training (1884–1993) and validation (1994–2023) periods in the Swiss Pre-Alpine region. Sim: simulated values; Obs: observed values; RMSE: root mean square error (optimal, 0; worst: +∞); MAE: mean absolute error (optimal, 0; worst: +∞); MAPE: mean absolute percentage error (optimal, 0; worst: +∞); r = Pearson’s correlation coefficient (optimum, + 1; worst: –1); KGE: Kling-Gupta Efficiency index (worst, -∞ to 1, optimum); Log Loss: logarithmic loss (optimum, 0; worst: +∞).
Fig 8.
Decadal prediction of heavy snowfall days.
Observed data (1950–2023, blue) and forecasted values (2024–2060, red) are shown for the Swiss Pre-Alpine Region alongside the long-term mean (grey line), with the 10th and 90th percentiles (indicating snowfall deficits and heavy snowfall exceedances, respectively) based on the 1970–2023 reference period. The grey shaded band represents uncertainty, calculated using a five-year moving standard deviation. The dashed grey line indicates a five-year running ensemble mean of heavier snow loads for the northeast Italian Alps under the RCP8.5 scenario (12 km resolution; [99]).