Snow plays an important role in the global energy and water budgets. But it is difficult to forecast snow accurately in atmospheric and hydrologic models, since the scale of its variability is usually smaller than that resolved by most models, and the forcing provided by models usually contain errors. Direct insertion of observed snow into a model may induce artificial moisture transfer. A data assimilation scheme that optimally merges snow observations with model forecasts is needed.Here we present a data assimilation study for North America with the catchment-based land surface model (LSM) used by NASA Seasonal-to-interannual Prediction Project (NSIPP), which includes a three-layer snow model. We use one-dimensional Kalman filter for each catchment of the LSM to sequentially update model states and model error covariances. This scheme takes into account of snow melting as a result of bias in the LSM temperature. Results from identical-twin experiments are presented, in which the data and true states are generate by the same LSM model. The synthetic data are in the same format as those from satellite observations. Special attention is paid to the updating of land surface temperature through snow data, so that erroneous snowmelt may be prevented. In a follow-up study, we will assimilate SMMR and SSM/I satellite-measured snow data into the model.