Snow cover remote sensing and spectral indices
Optical satellite remote sensing has been the dominant approach for large-scale snow cover mapping since the early 1990s. The normalized difference snow index (NDSI) is calculated using surface reflectance in the visible and shortwave-infrared spectral bands. The high reflectance of snow in the green band and its strong absorption at 1.6 µm wavelength allow distinguishing snow from most other land covers. The NDSI value ≥ 0.4 is generally accepted as a threshold for pixel-specific binary snow classification (Poussin et al., 2023). For heterogeneous mountainous areas, an improved algorithm of automated threshold selection was suggested by Ma & Zhang (2022). Tong et al. (2020) demonstrated that the assimilation of daily MODIS Collection 6 snow cover products significantly improved the simulation of streamflow dynamics.
SAR backscatter and snow detection
The use of Synthetic Aperture Radar (SAR) imaging is partially complementary to the information derived from optical sensors since, unlike optical data, SAR is independent of solar irradiance conditions. The underlying physical mechanism that enables wet snow detection with SAR is related to the change in dielectric properties of snowpack with a transition from solid to liquid state. Liquid water content significantly increases snowpack's dielectric constant, thereby decreasing C-band backscatter by several dB. Nagler et al. (2016) applied a change detection algorithm to VV-polarisation of Sentinel-1 SAR data enabling automated delineation of the snowmelt front. This method was later found to be applicable to a variety of terrain and vegetation (Tsai et al., 2019). Marin et al. (2020) used a temporal stack of Sentinel-1 data combined with derivative topographic information for the Italian Alps. Brangers et al. (2024) analysed C-band radar backscatter from a tall tower to characterise SAR sensitivity to liquid water content, and Carletti et al. (2025) combined multitemporal Sentinel-1 acquisitions with radiative transfer modelling.
Dry and wet snow classification
The ability to discriminate between dry and wet snow greatly improves the utility of snow monitoring products for water management and avalanche forecasting applications. Varade et al. (2019) showed that combining dual-polarisation SAR observations with multispectral indices allows separation of dry and wet snow classes in high-mountainous regions. Karbou et al. (2021) highlight the potential of using Sentinel-1 backscatter time-series together with numerical weather prediction data. Lund et al. (2022) employed Sentinel-1 observations in combination with the physical snow model SnowModel to relate backscatter variations to snowpack evolution during melt, freeze, and ripening periods.
Multi-sensor data fusion (SAR + optical)
The combined use of SAR and optical remote sensing has been identified as the most operationally efficient approach to snow cover monitoring through cloud cover. Sentinel-1 and Sentinel-2 offer the best potential for combined snow cover mapping as both sensors have a comparable revisit capacity and spatial resolution. The present thesis applies a Principal Component Analysis (PCA) using Sentinel-1 backscatter coefficients (VV, VH), the spectral bands B2, B3, B4, B8 and B11 of Sentinel-2, and the Normalized Difference Vegetation Index as input features. In contrast to earlier approaches, NDSI is not used as an input feature or a classification threshold.
Snow cover dynamics in the Austrian Alps
A significant tendency towards reduction of snow depth and snow season was identified for the high-altitude weather station Sonnblick, which is representative for the whole area (Schöner et al., 2009). Recently available high-resolution gridded data sets of daily temperature and precipitation over Austria (1 km resolution, SPARTACUS: 1961–present) enabled analysis of trends in detail (Hiebl & Frei, 2018). Olefs et al. (2020) assessed changes in snow depth, snow cover duration, and snow-making potential for Austria in the period 1961–2020, and Hanzer et al. (2018) estimated future snow-pack trends under various climate scenarios. Hohe Tauern National Park is located in the highest and most glaciated part of the Alps in Austria, characterized by a large altitudinal gradient and complex topography.