Data assimilation is the process by which observations are combined with a prognostic model to provide the best estimate (analysis) of the current state of a physical system.   Observational information, which tends to be irregular in space and time, can thus be made available in a regular or gridded form required for many applications.  The process is far more complex than a simple interpolation of information from one point to another, as it depends on aspects such as the errors of the various observation types, imperfections in the prognostic model and the physical relationships between different atmospheric variables; e.g., pressure and wind speed.  

Current atmospheric data assimilation systems used in weather and climate prediction incorporate information from millions of observations on a regular basis—typically every 6 or 12 hours—to provide the best possible estimate of the current state of the atmosphere.  The resulting atmospheric analyses are used to initialize forecasts, produce historical climate data sets and provide a dynamically consistent framework for studying specific atmospheric processes and phenomena.  This talk provides an overview of the atmospheric data assimilation problem in general terms, examines some common uses and limitations of current atmospheric analyses, and describes emerging techniques to help measure and understand the impacts of currently assimilated observations and plan for the development of future observing systems.