Tropical cyclones are tremendous natural hazards that threaten coastal populations worldwide, including in the Indian Ocean, the third most active basin. The purpose of this study is to perform data impact studies with the ALADIN Réunion Limited Area Model, which is the largest and only tropical implementation of all of the versions of the ALADIN consortium. It allows special focus on the Indian Ocean area and a "tropicalized" 3D-Var data assimilation. Studies are performed for several storms of the 2006/2007 cyclonic season of the Southwest Indian Ocean (SWIO) basin.
This last season proved to be very active with 10 named storms, 4 of which attained the "major hurricane" wind threshold of 50m/s. Satellite data has proven most invaluable when trying to initialize NWP models since the oceanic zones over which the cyclones develop are, by nature, data sparse. Yet, the occurrence of clouds or rain proves to be a challenge when trying to assimilate satellite data: non linear processes predominate and the use of refined, costly numerical methods might be required. These computational costs are usually found to be prohibitive and cloudy/rainy data assimilation usually is a missing component in most operational centers. This proves to be of critical importance when dealing with tropical cyclones the data-sparse core, where observations are systematically rejected.
Of the few centers that do not suffer from this crucial observational lack, the NCEP and the JMA have been assimilating rain rates while the European Center for Medium Range Weather Forecasting (ECMWF) has implemented a 1D-Var inversion for cloudy/rainy areas which uses complex moist physical schemes to retrieve a Total Column Water Vapor (TCWV)-equivalent from the rainy radiances, which is then used as pseudo-observation in the 4D-Var assimilation. In collaboration with the ECMWF and to bypass the costly 1D-Var inversion, we investigated a statistical multi-linear regression that fits TCWV with the brightness temperatures of the SSM/I instrument, relying on the ECMWF analyses. The algorithm is then applied to assimilate cloudy/rainy TCWV in the 3D-Var of ALADIN Réunion. Impacts of the 3D wind bogus is also investigated. We further used our reference simulation to downscale to 4km using the latest non-hydrostatic French model, AROME, over a SWIO-centered domain, using dynamical adaptation.
Finally, we have studied the flow-dependence of background error standard deviations of the day in the Aladin-Reunion area, and in particular their space and time variations as a function of cyclonic location and intensity. For this purpose, results from a 6-member global assimilation ensemble have been used. The results reflect the larger uncertainty in the vicinity of cyclones, implying that observations should be given a larger weight in these regions.