Adaptive Observations: A Feasibility Study

Abstract

The feasibility of the recently proposed concept of adaptive observations is tested on a typical case of poorly forecast North Atlantic cyclogenesis. Only numerical tools are employed, no special observations. Although based on simulated data, this study addresses both theoretical and practical problems of adaptive observations. In the first stage of this study, the role of the data assimilation processes is neutralized; the correction is done by forcing correct continuous fields within the target area. These experiments prove that it is necessary to correct the projection of the initial errors on the first unstable plane (the first two leading singular vectors) in order to significantly improve the forecast. These results also clearly demonstrate that the quality of the initial conditions on a limited, but quite large, area could be a major factor influencing the forecast quality. In a second stage, the focus is on operational aspects. The correction is done through the assimilation of a discrete set of simulated profiles using a 3DVAR analysis system. This leads to studying the impact of the assimilation scheme and to testing different sampling strategies. These experiments suggest that the concept of adaptive observations shows great promise in situations comparable to the one studied here. But the current assimilation systems, such as 3DVAR, require that all the structure of the target has to be well sampled to have a significant beneficial effect; sampling only the extremum does not suffice.