Abstract
Snow Water Equivalent (SWE) is a key variable for several applications (e.g., hydrology, ecology, agriculture, or hydropower production). The typical high spatial variability of SWE linked to snow distribution processes requires its monitoring at a proper spatial and temporal scale. However, spatialized high-resolution (HR) time-series (TS) of SWE maps are rarely available. We explore the use of multisource remote sensing data to derive information about snow depletion curves (SDC), i.e., the relationship with the snow cover area (SCA) and SWE. We propose a fusion approach that merges different sensors, if available, as optical HR and low-resolution (LR) data, Synthetic Aperture Radar (SAR), and in-situ temperature and snow depth observations to reconstruct SWE TS. The method does not require precipitation data as input, which could be a relevant advantage in poorly monitored mountain regions. The approach allows to carry out a historical reconstruction obtaining spatial reanalyses SWE TS, with daily sampling time and HR spatial detail (i.e., few dozens of m). The approach, designed for a hydrological catchment, assumes that similar snow patterns repeat interannually due to the climate and geomorphology of the area. Moreover, the catchment is characterized by a state: it is either subjected to an accumulation (SWE increase) or ablation (SWE decrease). The state is identified by mean of in-situ snow depth observations (if available) and/or SAR data and is used to regularize the SCA as well as to decide whether SWE is added or removed to the reconstruction. The potential melting is calculated through a degree day model. The output SWE TS present an unprecedented spatial and temporal detail. The proposed approach has been tested in reanalysis in few mountainous catchments with different climate and when evaluated against a reference product (i.e., Airborne Snow Observatory), shows a bias of -40 mm and a RMSE of 216 mm for a catchment of 970 km2 in Sierra Nevada (CA). The power of the method is represented by a good precision to capture the snow persistence on ground, that in turn is linked to the SWE amount of the pixel. This is achieved using HR spatialized data as input, that allow to properly sample the phenomenon at the correct spatial scale resulting in a good detection of the typical SWE spatial patterns. The availability of a long TS of accurate spatial reanalysis products opens the door to possible near real time predictions that exploit SDC.