Hillslope-scale daily snow cover reconstruction in complex mountains: a novel framework integrating long-term snow persistence

Spatially continuous daily snow cover mapping at the hillslope scale (<100 m) is essential for resolving terrain-induced heterogeneity in hydrological processes and supporting mountain infrastructure management. Yet accurately characterizing daily snow cover heterogeneity at this scale remains challenging due to observational limitations and the difficulty of resolving terrain-controlled snow accumulation and ablation dynamics. To address this challenge, we developed a hillslope-scale daily snow cover reconstruction (HDSR) framework that explicitly incorporates long-term snow persistence as a structural spatial prior. The framework reconciles coarse-resolution daily Normalized Difference Snow Index (NDSI) observations with a 30 m climatological snow cover frequency (SCOF) dataset, which characterizes terrain-mediated long-term snow persistence and constrains the spatial allocation of daily snow cover at the sub-pixel scale. HDSR is implemented through three sequential stages: spatiotemporal continuity restoration (STR), SCOF-constrained spatial refinement (SSR), and spatiotemporal snow-cover inference (SSI). A high-quality 30 m reference dataset was first constructed using the Snow Mapping Algorithm Considering Different Lighting Conditions (SALC), which was identified as the optimal approach through a systematic evaluation of seven candidate snow-mapping methods. The HDSR framework was subsequently applied to the Qilian Mountains and adjacent regions (QMAR), generating a daily 30 m binary snow cover product over an extensive mountainous region (536,018 km2) from September 2013 to August 2025. Validation against independent Landsat 8 observations yielded an F1-score (FS) of 0.873 and Cohen’s kappa (CK) of 0.859, while Bias values close to unity indicated negligible systematic bias in snow-cover representation. Consistent results obtained from independent Landsat 9 observations and time-lapse photography further supported the validity of the reconstruction, although site-specific evaluation revealed reduced capability for reconstructing short-lived snow events. The reconstructed dataset markedly outperforms the corresponding coarse-resolution baseline while providing enhanced spatial details. Compared with machine-learning models that exclude SCOF-related predictors, the proposed framework yields notable improvements, increasing FS and CK by 4.1% and 4.6%, respectively. Long-term snow persistence provides unique information unavailable from daily MODIS observations alone and is essential for resolving hillslope-scale snow heterogeneity. This study advances hillslope-scale daily snow cover mapping and potentially contributes to improved hydrological modeling and snow hazard assessment in complex mountainous environments.