The non-linear quality control of all data types within the National Center for Environmental Prediction's Regional Data Assimilation

This paper discusses the incorporation of quality control within National Center for Environmental Prediction s (NCEP s) 3-dimension variational (3dvar) Regional data assimilation, which provides the initial fields for the Eta model. In earlier formulations, the quality control was a separate, previous step. The usual 3dvar formulation is modified to incorporate the fact that each datum may have a combination of random error and possibly also a large non-random error component (gross error). This leads to a modification of the penalty function which effectively modifies the weight given to each datum, depending upon its perceived quality. This weight changes as the solution for the analysis is made; thus a datum which at first may receive little weight, if the analysis adjusts toward the datum, the weight given to it will also increase.

Examples are given of the sensitivity of analyses to this new quality control. In general the influence is small, but with bad data having even less influence than before. A record is kept of those data that are given small weight by the analysis.