Variational assimilation of time sequences of surface observations with serially correlated errors

Assimilation of observations from frequently reporting surface stations with a four-dimensional variational assimilation system (4D-Var) is described. A model for the serial observation error correlation is applied to observed time sequences of surface pressure observations, whereby the relative weight of the mean information over the temporal variations is decreased in the assimilation. Variational quality control is performed jointly for each time sequence of observations so as to either keep or reject all observations belonging to a time sequence. The operational practice at ECMWF has previously been to use just one pressure datum from each station within each 6-h assimilation time window. The increase of observational information used in these assimilation experiments results in a small but systematic increase in the short-range forecast accuracy. The r.m.s. of the analysis increments is decreased in the experiments, whichmeans there is an improved consistency between the background and the observations. A study of a rapidly developing small-scale synoptic system (the Irish Christmas Storm in 1997) showed that both the background and the analysis became more accurate when more frequent observations were assimilated. Single-observation experiments showed that a surface pressure timesequence of data from a single surface station can intensify the analysis of a mid-latitude baroclinic system, that was underestimated in the background, when used in a 6-h 4D-Var. The method to assimilate time sequences presented in this paper has been implemented into the ECMWF operational 4D-Var assimilation system.