Scale patterns of the Sentinel-1 SAR-based snow depth product compared to station measurements and airborne LiDAR observations

Abstract. Water storage in snowpacks in mountain areas is critical for hydropower production, hydrological forecasting, and freshwater availability. Spaceborne synthetic aperture radar (SAR) is a powerful tool for quantitatively measuring snow mass because of its high spatial resolution and the sensitivity of signals to snow depth (SD). In particular, the first SAR SD product (C-snow) based on Sentinel-1 satellites displays high sensitivity to depolarization signals for dynamic SD monitoring in mountainous areas. Moreover, upscaled C-snow retrievals (e.g., 10 and 25 km) have been used to provide reference data to train machine learning models, improve passive microwave-based retrieval, and calibrate many hydrological models. However, a systematic assessment of C-snow products at various scales has not been conducted, until now. In this study, the performance of C-snow products at three scales (1, 10 and 25 km) is comparatively assessed via station-based measurements and airborne LiDAR observations, and the scale patterns associated with the heterogeneity of the geographic environment and the representativeness of so-called truth data are analyzed. The results indicate that the scale patterns of the C-snow products across various resolutions differ from those of station- and airborne-based reference data. As the spatial scale increases from 1 to 25 km, the error of C-snow retrieval in reference to station measurements tends to increase (e.g., ubRMSE from 68.18 to 77.47 cm, bias from -9.81 to 10.68 cm), whereas it tends to decrease compared with airborne snow observatory (ASO) data, with ubRMSE values ranging from 104.3 to 83.29 cm, and the bias values from -91.31 to -52.73 cm. We also found that land cover types, e.g., tree cover and permanent ice, affect the C-snow product at various scales. Especially an overestimation tends to occur in coarse pixels covered with even a small amount of permanent ice. It is concluded that C-snow retrieval at three scales is characterized by high uncertainty. Researchers should focus on developing a robust SD retrieval algorithm by combining SAR backscattering signals and polarimetric and interferometric information.