Classification Performance and NDSI Threshold Selection
The rule-based classification algorithm performed successfully in the task of snow cover delineation. The AUC of the model using Sentinel-2 NDSI reached an excellent value of 0.98 for the period 2018–2025, which compares favorably to results reported in the literature for optical methods in alpine terrain. Poussin et al. (2023) showed that delineating Landsat-derived NDSI using a spatiotemporal threshold is highly efficient in complex topography. Tong et al. (2020) trained a similar model for snow cover delineation in Austria using MODIS data. The performance of the model trained on Sentinel-2 data was better than that of models trained on Landsat and MODIS data because of the higher spatial resolution of 10 m, compared to 30 m for Landsat and 500 m for MODIS. A threshold value of 0.4 was used to delineate the snow cover, based on established literature (Ma and Zhang, 2022; Tong et al., 2020).
Multi-sensor Fusion and SAR-Based Snow State Discrimination
The addition of Sentinel-1 C-band SAR backscatter as an input feature enabled distinguishing between Dry Snow and Wet Snow classes, which is impossible for any optical sensor. The criteria used (VV < −11 dB for absolute backscatter and ΔVV < −2.0 dB for relative backscatter) were similar to those reported by Nagler et al. (2016) and Marin et al. (2020). The addition of liquid water into the snowpack increases the dielectric permittivity of snow, which leads to higher microwave attenuation and therefore lower backscatter values at C-band (Tsai et al., 2019; Lund et al., 2022). The fusion algorithm, based on PCA of Sentinel-2 and Sentinel-1 observations, exploited the information contained in both optical and microwave observations. The high value of Pearson's correlation coefficient between the first PCA component and NDSI indicated that SAR backscatter information had added value to snow state delineation beyond that provided by the optical sensor.
Observed Snow Cover Trends in the Context of Alpine Climate Change
The tendency of the data on the transition from dominance of Dry Snow to Wet Snow and an increase in the non-snow area, as well as the lifting of the snow line, is consistent with the trends of regional warming reported in the Austrian and Alpine literature. Schöner et al. (2009) find a reduction in snow depth at the Sonnblick Observatory (3106 m a.s.l.) during the XX century. Olefs et al. (2020) confirm recent negative trends in snow depth and snow cover duration for Austria for the period 1961–2020, which were most noticeable for the lowest altitudes — matching the patterns found in this study.