A variational assimilation for a barotropic model of tropical cyclone track forecast.
Y Gregoris, S Greard, A Lasserre-Bigorry, I Soulan,
Meteo-France, DIRRE/CRC, La Reunion
The purpose of this paper is to improve our objective forecast methods of tropical cyclones. Considering the very sparse meteorological observations available on South West Indian Ocean, we will not try to modelize the thermodynamics of tropical cyclones and will just interest us on the dynamics. For this study we will use a shallow-water model. It is a global shallow-water model with variable mesh, it conciliates some advantages of global models to those of limited area models. It didn't have disturbance generated at (lie boundary which can propagate throughout the domain) and it can have a fine resolution on the specific area without the need of powerful computer.
One can justify this choice because barotropic models are most representative of the deep trade wind regions of the tropics. Moreover, De Maria et al. (1992) have shown that barotropic predictions of tropical cyclone tracks have as much skiff as predicions with baroclinic models for forecasts up to about 48 hours. Numerous studies have focused on the motion dynamic of tropical cyclone. Through numerical integration of a nondivergent barotropic model, Chan and Williams (1987) first found that the linear beta-effect set up an east-west asymmetry in the vortex that then leads to a north-westward (northern hemisphere) movement of the vortex, when non-linear effects are included.
The major problem for barotropic simulation is that this important asymmetric part of the wind is much weaker than the symmetric part. Its analyse require then a fine resolution data network like the one used in some field experiments (ex. : TCM 90, SPECTRUM, ...). These resolutions are indeed not available for operational forecasts, especially in south-west Indian Ocean where most of the time only polar satellite data are available. Consequently, in the Saint-Denis RSMC the forecasters generate only synthetic observations for the symmetric part of the flow from Dvorak (1984) method. The absence of the asymmetric part of the flow introduces an initial bias in the analyse which grows with time and debases the forecast skill of barotropic simulation. The first goal of this study is to improve this analyse, starting from only basic observations of the storm. One way to realise that is to use data assimilation techniques. De Maria and Jones (1993) and Bennett et al. (1993) have shown that these methods could improve the forecast skill out to about 48 hours. They have indeed worked on situations where numerous observations are available. In the present case, we will use only the synthetic observations provided by the forecasters plus the background field given by the baroclinic global model of Meteo-France (ARPEGE). We will demonstrate with some school cases that variational assimilation over the past fixes may determine an asymmetric circulation that has a realistic structure (i.e. one can find a realistic solution of this inverse problem). Then we will apply the method on real cases of tropical cyclones