An Ensemble Transform Kalman Filter (ET KF) is used to predict the error variance reducing effects of different deployments of atmospheric observations. Using real and synthetic observations together with an ensemble of 12-36h old NCEP and ECMWF members (from which background fields and error covariances are produced), an analysis increment is calculated via a linear transformation of ensemble perturbations. Applying the same transformation to the ensemble members valid at later times, a ``pre-emptive'' forecast can then be produced almost instantaneously. The skill of such forecasts is assessed.

We use this technique to analyze (1) the effects of current and potential future routine observational networks on analyses and forecasts, and (2) the forecast error reducing effect of adaptive observations on high-impact weather events (such as winter storms), in the framework of a future data assimilation scheme that utilizes flow-dependent error covariance information. Implications for the design of future observational networks will be discussed.