6.4
Sensitivity to observations and targeted observations
Alex Doerenbecher, Meteo-France, Toulouse, France; and T. Bergot and F. Bouttier
The forecast of rapid cyclogenesis with NWP models depends on a severe predictability problem, even in the short range (1-2 days). Such a topic was among the scientific objectives of FASTEX (Front and Atlantic Storm Track Experiment). This field context allowed the first test of real time targeting of observations. Targeted observations are extra observations deployed over a sensitive region, numerically identified with adjoint sensitivity or singular vectors (unstable plane), among other techniques.
First evaluations of targeting during FASTEX, as well as simulated studies, suggest that the data assimilation scheme used at that time (3D-VAR) is not efficient enough to use in an optimal way the information provided by such additionnal observations. During FASTEX, the location of adaptive observations was only based on the information given by the definition of the sensitive area and no quantitative information about unstable structures is used after these zone have been determined.
The challenge is then to sample efficiently this sensitive region. Further progress on targeting depends on a consistent global approach, gathering data assimilation together with targeting observations and dynamical properties of the current trajectories (predictability informations provided by the sensitive structures). The aim of targeting becomes then to determine in advance how many extra observations are needed to maintain a control of the unstable part of the analysis error, knowing the assimilation scheme (3D or 4D-VAR).
The conceptually straightforward but technically non trivial way to insure this is to include the assimilation scheme in the algorithm used to determine the sensitive area where to sample. An extension of the adjoint sensitivity technique in order to take the analysis into account provides an estimate of areas where extra observations will have the most impact on some forecast aspect through assimilation.
The theoretical approach needed will be outlined and some of the difficult or open points that it contains will be documented by diagnostic calculation performed on some FASTEX or simulated cases.
Session 6, Adaptive observations
Wednesday, 12 January 2000, 9:30 AM-10:45 AM
Previous paper