The problem of data assimilation that is concerned with the complete and accurate specification of the atmospheric state based upon observations and other types of information can be approached either by variational or sequential algorithms. While variational techniques proceed by the global fitting of an assimilating model to the available information, sequential assimilation involves a statistical minimum mean-square-error estimation approach. In this paper both algorithms are compared in a systematic manner with regard to assimilation/forecast accuracy, computational efficiency, and storage requirements based on a limited series of observing-systems simulation experiments. The barotropic vorticity equation on a rotating sphere is used as the assimilating model. The results indicate that under a variety of conditions the variational algorithm performs at least as well as the sequential algorithm. The variational algorithm is also found to be more successful than the sequential algorithm in the reconstruction of the physical fields in data-void regions. Limitations and possible extensions, as well as operational implications of this work, are briefly discussed.
