Tropical cyclones simulations with ALADIN

(more details yves.bouteloup++at++meteo.fr )

In this short presentation some tropical cyclone simulations made with ALADIN are presented jointly with a quick bibliographical study. The main result is the ability of ALADIN to make realistic simulations of tropical cyclones. The problem of the improvement of the analysis in order to make better track forecasts is not resolved.

It is well-known that numerical models are not able to do skilful forecasts of tropical cyclones track. An important factor that contribute to this result is the lack of observations over the ocean, another one is the inadequate models resolution. As shown by Serrano (1997) ECMWF Re-analysis has skilful performance in tracking tropical cyclones. Indeed more than 80% of them are captured by the analysis, but the positional mean error is 140 km in the Northen Hemisphere and 176 km in the Southern Hemisphere. The forecasts suffer from this inadequate analysis position. An operational use of ECMWF model (IFS) at the DIRNC (Météo-France Regional Division in Nouvelle Calédonie) reinforce these conclusions.

With satellite imagery and Dvorak method it is possible to have a very good estimation of cyclone position and intensity. Most operational track prediction models include `synthetic' or `bogus' observations near the storm to help overcome the problem of the initial position (Mathur 1991). The aim of this kind of method is to produce an initial vortex that was well adapted to the forecast model and was much more realistic in size and intensity than the analysis storm structure. In the method proposed by Kurihara et al (1993) a crudely resolved tropical cyclone in the large-scale analysis is replaced by a vortex that is properly specified for use in a high resolution prediction model. This method is used by Bender et al (1993) on four real-data cases. They concluded that the average 48-h forecast error is reduced by more than 50% when compared to the integrations run with the analyzed vortex. Serrano and Unden (1994) with a less sophisticated method, found a very unfavorable impact of predicted cyclone tracks for the cases investigated. They concluded that a bogus method must give an asymetric component of the flow consistent with recent tropical cyclone motion. As shown by Fiorino and Elsberry (1989) the asymetric flow determines the initial motion of the storm. DeMaria et al (1990) have even shown that in barotropic model forecasts, an error in the estimate of the initial motion vector of a storm has a greater impact on a track forecast than an initial position error. To overcome this problem DeMaria and Jones (1993) used a 4DVAR assimilation of synthetic data in a barotropic hurricane track model. They found that the average track errors are smaller than the errors in the control simulations out to 72 h. Nevertheless, as shown by Flatau et al (1994) baroclinic processes contribute to tropical cyclone propagation.

Some simulations of cyclones and tropical storms has been made at the DIRNC with a 20 km mesh version of ALADIN. Boundary conditions came from IFS forecasts for these first experiments. In a first set of experiments initial conditions were not pertubated. A realistic structure of the cyclone was rapidly reconstructed by the model. After a 24-hour-simulation wind force and mean sea level pressure were closed to the values estimated by the Dvorak method (Fig. 1). In case of tropical storm YALI, the IFS forecast trajectory was very accurate despite a 6-hour chronological error. ALADIN made a dynamical adaptation of the fields to the orography, with a very good strengthening of the wind in the havana canal at the south of the main island (Fig. 2). However, when the global track forecasts were not good, ALADIN did not improve them. A 20 km mesh limited area hydrostatic model is a very good tool to study the dynamical structure of tropical cyclones, but not to make track predictions. An improvement of initial conditions is really necessary.

In a second set of experiments a simple bogussing method was applied to the initial conditions. As in Serrano and Unden (1994) we found that rapidly track forecasts were not accurate. No surprise, our bogussing method did not improve the surrounding circulation. One way to reconstruct the asymetric flow is to use a 4DVAR assimilation of synthetic data, like DeMaria and Jones (1993) but with a baroclinic model.

Figure 1 : Mean sea level pressure and 10 meters wind from ALADIN simulation of tropical cyclone DRENA. Only wind larger than 40 knots (20ms-1) are plotted. Intensity of wind and mean sea level pressure are closed to their estimated values. The positional error is approximately 100 km west of the island.

Figure 2 : Same as figure 1 but for tropical storm YALI. Forecast track is very good, despite a 6-hour chronological error. One can remark the strengthning of the wind in the havana canal between the main island and the isle of pine. Observations by automatic weather stations confirm this forecast.

References




Home