Producing unbiased analyses in the presence of forecast bias

Analysis methods involve many assumptions about the errors associated with observing and predicting atmospheric fields. For example, the short-term model forecasts, which serve as a first guess for the analysis, are generally assumed to be unbiased. For most atmospheric model fields, however, observational evidence shows that the systematic component of errors in forecasts and analyses is of considerable magnitude, in some cases, being of the same order of magnitude as the random component.

We have recently shown that biases are largely reduced when a simple analysis algorithm is used to estimate and correct biases in the moisture field from the forecast model of the Goddard Earth Observing System Data Assimilation System (GEOS DAS). As a continuation of this effort, we will present results of applying this algorithm to further reduce forecast biases in all of the GEOS DAS upper-air analysis fields.