On the initialization of Tropical Cyclones

Recent increases in the IFS model resolution have improved its ability to produce realistic forecasts of tropical cyclones but they also present new challenges to the data assimilation system. This is because standard assumptions made in 4D-Var are called into question when the data assimilation system has to deal with observations sampling the extreme conditions, both actual and forecasted, around a tropical cyclone centre. We present here examples of these problems as encountered in the operational ECMWF analysis and the solutions adopted to correct them. These include introducing an adaptive quality control algorithm for dropsonde observations, the reduction of the resolution of the EDA-derived background errors used in 4D-Var together with a non-homogeneous noise filtering technique. These investigations have also highlighted a number of promising avenues of further research, which have the potential to improve both the initialisation of tropical cyclones and the global accuracy of the ECMWF analyses and forecasts.