Assimilation of AMSR-E snow water equivalent data in a spatially-lumped snow model

Abstract

Summary Accurately initializing snow model states in hydrologic prediction models is important for estimating future snowmelt, water supplies, and flooding potential. While ground-based snow observations give the most reliable information about snowpack conditions, they are spatially limited. In the north-central USA, there are no continual observations of hydrologically critical snow variables. Satellites offer the most likely source of spatial snow data, such as the snow water equivalent (SWE), for this region. In this study, we test the impact of assimilating SWE data from the Advanced Microwave Scanning Radiometer – Earth Observing System (AMSR-E) instrument into the US National Weather Service (NWS) SNOW17 model for seven watersheds in the Upper Mississippi River basin. The SNOW17 is coupled with the NWS Sacramento Soil Moisture Accounting (SACSMA) model, and both simulated SWE and discharge are evaluated. The ensemble Kalman filter (EnKF) assimilation framework is applied and updating occurs on a daily cycle for water years 2006–2011. Prior to assimilation, AMSR-E data is bias corrected using data from the National Operational Hydrologic Remote Sensing Center (NOHRSC) airborne snow survey program. An average AMSR-E SWE bias of −17.91 mm was found for the study basins. SNOW17 and SAC-SMA model parameters from the North Central River Forecast Center (NCRFC) are used. Compared to a baseline run without assimilation, the SWE assimilation improved discharge for five of the seven study sites, in particular for high discharge magnitudes associated with snow melt runoff. SWE and discharge simulations suggest that the SNOW17 is underestimating SWE and snowmelt rates in the study basins. Deep snow conditions and periods of snowmelt may have introduced error into the assimilation due to difficulty obtaining accurate brightness temperatures under these conditions. Overall results indicate that the AMSR-E data and EnKF are viable and effective solutions for improving simulations of the operational forecast model.