A methodology for snow data assimilation in a land surface model

[1] Snow cover has a large influence on heat fluxes between the land and atmosphere because of its high albedo and insulating thermal properties. Hence accurate snow representation in coupled land-ocean-atmosphere global climate models has the potential to greatly increase prediction accuracy. To this end, a one-dimensional extended Kalman filter analysis scheme has been developed to assimilate observed snow water equivalent into the NASA Seasonal-to-Interannual Prediction Project (NSIPP) catchment-based land surface model. This study presents the results from a set of data assimilation twin'' experiments using an uncoupled version of the land surface model. First, true'' snow states are generated by spinning-up the land surface model for 1987 using an observation-constrained version of the European Centre for Medium-Range Weather Forecasts (ECMWF) 15-year Re-Analysis (ERA-15) data set for atmospheric forcing. A degraded 1987 simulation was then made by initializing the model with no snow on 1 January 1987. A third simulation assimilated the synthetically generated snow water equivalent observations'' from the true simulation into the degraded simulation once a day. This study illustrates that by assimilating snow water equivalent observations, which are readily available from remote sensing satellites, other state variables (i.e., snow depth and temperature) can be retrieved and effects of poor initial conditions removed. Runoff and atmospheric flux predictions are also improved.