Multiscale ensemble filtering in snow data assimilation [INVITED]

A data assimilation system that merges model predictions of snow awater equivalent (SWE) and remotely sensed observations of snow cover extent (from MODIS) and passive microwave brightness temperatures (from AMSR-E), is evaluated over one winter seasons in the Upper Snake River basin. The potential value of assimilating these observations was examined in comparison with an open loop simulation in which a macroscale hydrology model (Variable Infiltration Capacity, VIC) was forced only with globally available data (ERA-40), hence the results are applicable to data sparse parts of the world. A model truth, or baseline run was formed by forcing the VIC model with gridded in situ data. Two data assimilation techniques, the Ensemble Kalman filter (EnKF) and the Ensemble Multiscale Kalman filter (EnMKF) are tested, where the VIC multilayer snow hydrology model is forced by ERA-40 reanalysis. A screening procedure for the AMSR-E observations was used to address the signal saturation problem and the effects of different land cover types, such as forest cover. Both the EnKF and EnMKF produced modest improvements relative to the baseline simulation as compared with the open-loop simulation. The differences from the baseline simulation were smaller (performance better) for the filters in higher elevation areas, while the use of the 18.7 GHz frequency channel partly alleviated some of the negative effects of forest cover, leading to different reductions in forested areas ($\sim$10\%). Comparisons with in-situ SWE measurements showed an overall improvement when assimilating remotely sensed observations in at least 24 of 32 stations, however there were a number of stations where correlations were higher without assimilation.