A new method for selecting adaptive observation locations is demonstrated. This method is based on optimal estimation (Kalman filter) theory; it determines the observation location which will maximize the expected improvement, which can be measured in terms of analysis or forecast error variance. This technique presupposes a large ensemble of forecasts is available for generating an accurate model for background error statistics which vary both in space and in time.

This technique is demonstrated using in a quasigeostrophic channel model under perfect-model assumptions. Three data assimilation schemes are tested, two variants of the standard ensemble Kalman filter and a third perturbed observation (3D-Var) ensemble. The technique is shown to find large differences in the expected variance reduction depending on observation location and the flow of the day. The perturbed observation ensemble was not particularly useful for selecting observation locations and assimilating the targeted data, but the two variants on the ensemble Kalman filter defined consistently similar targets to each other, and assimilation of the targeted observation typically reduced analysis errors significantly. It was also found that the spread in background forecasts in the ensemble Kalman filter provided similar target locations to those generated with the full algorithm.