The potential use of high-resolution SWE estimates from remote sensing imagery to predict snow melt rates

Snow water equivalent is a key variable in hydrology. An accurate SWE estimation is crucial for runoff prediction, especially for catchments with strong nival regimes. Direct observations are unfortunately rare and are available only at a point scale. Accurate spatialized estimates of SWE are thus difficult to be obtained. Physically based models often suffer from the inaccuracies of input data and uncertainty of model parametrization. In this sense, the integration of traditional techniques with remote sensing observation is valuable. Although current satellite missions do not provide direct SWE observation, they allow us to extract important proxy information that is crucial for SWE reconstruction. In this sense, we propose to exploit optical and radar sensors to retrieve accurate information on the persistence of snow on the ground. In fact, the longer the persistence, the deeper the snowpack. To achieve enough spatial and temporal detail, we merged multi-scale information from MODIS, Sentinel-2, and Landsat missions. The key idea is to exploit the snow pattern persistence that we can observe with good spatial detail from Landsat and Sentinel-2 missions to reconstruct the scene when a low-resolution image (MODIS) is acquired. Furthermore, information on the duration of the melting phase can also be retrieved by exploiting the synthetic aperture radar (SAR) mounted on board of Sentinel-1. Hence, we can estimate the number of days of melting. In-situ data, when available, are also exploited in the reconstruction. In detail, air temperature is used to estimate the potential melting and the snow depth increases to determine the number of days in accumulation. The reconstruction approach is then simple: by knowing the days in melting, the total amount of melted SWE is determined. Assuming that the melted SWE is equal to the accumulated SWE, we can redistribute SWE throughout the season using a simple approach as the degree day. The final output is a daily time-series with a spatial resolution of few dozens of m. One of the major advantage of the proposed approach, compared to more traditional SWE estimation techniques, is that it does not depend from precipitation observation, often highly uncertain in high-elevation catchments. When evaluated against a reference product (i.e., Airborne Snow Observatory), the method shows a bias of -22 mm and an RMSE of 212 mm for a catchment of 970 km2 in Sierra Nevada (CA). In this work, we investigate the relationship between the melted SWE and the measured riverine discharge for a number of catchments in South Tyrol (Italy). The results may be of great interest, especially for poorly monitored basins with highly variable snow accumulation that are exploited for hydroelectric energy production. In detail, we propose a long-term analysis on SWE time-series to understand if there are evident trends that might improve hydroelectric power management.