Replication Data for : Weakly supervised learning for snow cover segmentation in mountainous areas from Sentinel-1 SAR images using interpolated NDSI time series

This dataset is designed for deep learning-based snow cover detection from SAR imagery. It contains Sentinel-1 imagery (VV and VH backscatter in dB) on three different orbits (descending 139, descending 66 and ascending 88) over the Guil basin in the French Alps, during 2018-2019 hydrological year, paired with MODIS-derived binary snow cover labels obtained through NDSI thresholding at 0.4, projected on SAR geometry. Additional data includes orbit-specific geographic projection maps, reference images for each acquisition geometry and normalization statistics for model input preprocessing. The dataset is organized into training, validation and test sets, with the test set subdivided into three evaluation scenarios: no transfer, spatial transfer (Gyronde basin) and temporal transfer (Guil basin during 2019-2020 hydrological year). Training set labels come time-interpolated using NDSI values in three variants: non-interpolated, with closest neighbour interpolation (CNI) and Kalman smoother (KS). The dataset folder tree is described in the readme.