Methods for dealing with spurious covariances arising from small samples in ensemble data assimilation

Jeff Whitaker
ERSL, Physical Sciences Division - NOAA

Ensemble Kalman filters are increasing used in numerical weather prediction to initialize forecast models. They utilize a sample covariance estimate from a ensemble of short-term forecasts to estimate the state vector of the model given new observations. Since the ensemble size is typically many orders of magnitude smaller than the size of the forecast model state vector, it is crucial to deal with the spurious covariances that can arise due to sampling error. I will discuss methods currently used for dealing with these spurious covariances, as well as some promising new techniques that have yet to be tested in large models.

Back to agenda