This report provides scientific rationale and justification for a five-year research plan for the assimilation of meteorological observations into numerical weather prediction models. In the broadest sense, meteorological data assimilation is partly the incorporation of atmospheric measurements into computer models that predict atmospheric behavior and partly the accommodation of such models to a set of observations. The goal of data assimilation is to produce a regular, physically consistent, four-dimensional representation of the atmosphere from a heterogeneous array of in situ and remote instruments which sample imperfectly and irregularly in space and time. Models and observations are inextricably linked in building this representation. Though data assimilation seldom receives public notice, it is true that today's computer forecasts would be impossible without it. Moreover, improvements in data assimilation are just as important as improvements to the model itself for accurate forecasts. For example, it is estimated that at least half the error in a two-day forecast of wind, temperature, or pressure is due to errors in the initial analysis. Meteorological data assimilation addresses a number of problems that will be introduced here and treated more thoroughly in Sections 2 and 3.
