Temperature and salinity profiles observed by gliders in the Monterey Bay in August 2003 are assimilated into NCOM model in the framework of a 3dVar scheme with a hybrid background error covariance (BEC) representation. The model performance is validated against independent mooring observations for the assimilation runs with 1-hour analysis cycle. In the first experiment the background error statistics was estimated using the ensemble of model states spanning the entire observation period, whereas in the second experiment the BEC information was acquired by averaging over the 3day floating temporal window (FTW) centered at the analysis time. It is found that the FTW scheme provides lower discrepancy between the values of temperature, salinity and velocity predicted by the model and observed at the moorings. The improvement becomes more clearly visible during the upwelling and relaxations events, associated with intermittent wind forcing. During these periods the FTW scheme provides a significantly (2–3 times) better fit to the mooring data.