Towards the assimilation of satellite reflectance into semi-distributed ensemble snowpack simulations

Abstract

Uncertainties of snowpack models and of their meteorological forcings limit their use by avalanche hazard forecasters, or for glaciological and hydrological studies. The spatialized simulations currently available for avalanche hazard forecasting are only assimilating sparse meteorological observations. As suggested by recent studies, their forecasting skills could be significantly improved by assimilating satellite data such as snow reflectances from satellites in the visible and the near-infrared spectra. Indeed, these data can help constrain the microstructural properties of surface snow and light absorbing impurities content, which in turn affect the surface energy and mass budgets. This paper investigates the prerequisites of satellite data assimilation into a detailed snowpack model. An ensemble version ofMétéo-France operational snowpack forecasting system (named S2M) was built for this study. This operational system runs on topographic classes instead of grid points, so-called 'semi-distributed' approach. Each class corresponds to one of the 23 mountain massifs of the French Alps (about 1000km 2 each), an altitudinal range (by step of 300m) and aspect (by step of 45 o ). We assess the feasability of satellite data assimilation in such a semi-distributed geometry. Ensemble simulations are compared with satellite observations from MODIS and Sentinel-2, and with in-situ reflectance observations. The study focuses on the 2013-2014 and 2016-2017 winters in the Grandes-Rousses massif. Substantial Pearson R 2 correlations (0.75-0.90) of MODIS observations with simulations are found over the domain. This suggests that assimilating it could have an impact on the spatialized snowpack forecasting system. However, observations contain significant biases (0.1-0.2 in reflectance) which prevent their direct assimilation. MODIS spectral band ratios seem to be much less biased. This may open the way to an operational assimilation of MODIS reflectances into the Météo-France snowpack modelling system. Highlights -Ensemble simulations of the snowpack are compared with satellite reflectances -Spatial aggregation into the semi-distributed geometry filters the observation noises -Satellite reflectances carry useful information worth to assimilate -MODIS reflectances can not be directly assimilated because they are biased -Ratios of MODIS reflectances show no evidence of bias and could be assimilated

CrocO_v1.0 : a Particle Filter to assimilate snowpack observations in a spatialised framework

“…In the following, only a short description of the system and its elements is provided. More details on the ensemble modelling setup are available in Cluzet et al (2020).…”

“…It enables to optimally combine the spatial and temporal coverage of snowpack models with the available information from observations. Assimilation of optical reflectance could reduce modelled SWE errors by up to a factor of two (Charrois et al, 2016), and preliminary studies showed its potential for spatialised assimilation (Cluzet et al, 2020). Assimilation of HS is very efficient in reducing modelled SWE errors (Margulis et al, 2019).…”

“…For instance, snowpack VIS-NIR reflectances from moderate resolution (250-500 m) satellites such as MODIS or Sentinel-3 can help constraining the snowpack surface properties (Dozier et al, 2009). However, in the areas covered by clouds, forests, or concerned by high sub-pixel variability (ridges, roughness, fractional snow cover) and shadows, satellite retrievals are less accurate (Masson et al, 2018;Lamare et al, in prep, 2020), and data should be filtered out (Cluzet et al, 2020).…”

“…CrocO uses an ensemble of stochastic perturbations of SAFRAN meteorological analysis (Durand et al, 1993;Charrois et al, 2016) to force ESCROC (Ensemble System CROCus, Lafaysse et al (2017)), the multi-physical version of Crocus snowpack model (Vionnet et al, 2012). The ensemble setup accounts for the major sources of uncertainties in snowpack modelling (Raleigh et al, 2015) and was formerly described and evaluated in the semi-distributed geometry by Cluzet et al (2020).…”

Toward Snow Cover Estimation in Mountainous Areas Using Modern Data Assimilation Methods: A Review

“…In addition, the assimilation of reflectance (spectral albedo) has been shown to further improve the simulated snowcover physical properties (e.g., Charrois et al, 2016). The uncertainties related to the conversion of raw optical satellite imagery into snow products remain the main limitation of surface reflectance and albedo assimilation (Zaitchik and Rodell, 2009;Hall et al, 2010;De Lannoy et al, 2012;Cluzet et al, 2020). The calculation of the observation errors is thus critical for the efficiency of the system.…”