Detection and Characterization of Systematic Errors In Atmospheric Models

Abstract

Modern numerical weather prediction models typically use 3- or 4-dimensional variational techniques (3D- or 4DVAR) for data assimilation and model initialization. In most cases, these methods assume that the distribution of observation and model error is Gaussian and unbiased. Owing to the importance of water vapor in weather forecasting and climate modeling/prediction, both scientific areas benefit by identifying and correcting systematic moisture errors in observations and models. Recent work at ESRL's Global Systems Division has detected clear evidence of systematic errors in the analysis and prediction of total atmospheric column precipitable water vapor (TPW) in operational NWP models over the continental U.S. This paper describes how these errors were detected and how they appear to propagate with time. Because the cause(s) of these errors are not well understood, climate and operational forecasting interests will mutually benefit by collaborating to discover these answers.