Development of a snow reanalysis pipeline using downscaled ERA5 data: application to Mediterranean mountains

<p>The Snow Water Equivalent (SWE) is a key variable to characterize water resource availability in mountain catchments. Despite its hydrological significance, the snow cover is poorly monitored in many regions due to a lack of in situ measurements.&#160;</p><p>Global climate reanalysis products provide increasingly accurate data but are too coarse to be used directly in mountain regions to reconstruct snow related variables. However these reanalyses have been successfully used to generate high resolution meteorological forcing and run a snowpack model in the central Andes and the High Atlas mountain ranges (Mernild et al. 2017; Baba et al. 2018).&#160;</p><p>The method is based on the MicroMet/SnowModel package (Liston and Elder 2006a; 2006b). MicroMet performs spatial interpolation of meteorological variables using the digital elevation model (downscaling) and the other routines of SnowModel computes the snowpack energy and mass balance. We have implemented a tool to improve the automation and scalability of this method to simulate the snow cover distribution in other regions using ERA5 or ERA5-Land. Our snow simulation tool only requires a digital elevation model as input. The land cover is extracted from the Copernicus global land cover map and the meteorological data are retrieved from either ERA5 or ERA5-Land over the period of interest.&#160;&#160;</p><p>We used three catchments under the influence of Mediterranean climate to evaluate the performance of this tool: Tuolumne (USA), Bassies (France) and Yeso (Chile). For each catchment either the modeled SWE depth or snow depth are compared with the validation data, over periods going from 3 to 8 years.&#160; In the Tuolumne basin, where the dataset is the most accurate with several SWE maps per year, we find a very good agreement at the basin scale (RMSE 40 mm w.e.). However, the mean RMSE in the highest elevation band (3500-4000 m asl) can exceed 500 mm w.e., which we attribute to the&#160; lack of gravitational transport in SnowModel and errors in the spatial distribution of precipitation. To reduce these errors in particular, we are implementing a non-deterministic representation of the precipitation input data to eventually allow the assimilation of globally available remote sensing data.&#160;</p><p>This tool will allow us to compute snow reanalyses in key mountain ranges around the Mediterranean sea over the past two decades (Pyrenees, Atlas and Mount Lebanon) and study the influence of topography and climate on the snow cover variability.</p>