Abstract:Due to the lack of observational data over the tropical oceans, TC (tropical cyclone) initialization for numerical models is one of major difficulties in TC track forecasting. Since the National Meteorological Centre of China Meteorological Administration (NMC/CMA) put the global model T639 into operation in 2009, an effective TC initialization scheme has been used. When TC occurs at first cycle time, a bogus vortex is inserted into the background fields, which is called vortex formation. In the second, third, …, cycle time, the evolutive vortex generated with 6 h output of global model prediction at the previous time in the background fields is initialized by relocation (moving to observed position) and modification (adjusted to observed intensity) techniques. It is obvious that the vortex formation at the first cycle time is so important that it can affect the structure of TC and the forecast results of the subsequent cycle time. But the initial bogus vortex used in the first cycle time is not consistent with the dynamics and physics of the global model. Recently, a new vortex formation scheme based on T639 global analysis forecast cycle in NMC/CMA has been developed successfully. Compared with the operation scheme, the initial vortex of the new scheme is mostly formed by assimilating TC bogus data into the variational assimilation system. It is significant that the initial vortex structure is analyzed with constraint of variational method, not largely affected by the empirical and statistical manual factors. Meanwhile, the analyzed vortex is consistent with the dynamics and physics of the global model and also compatible with environmental flow in the boundary. To evaluate the impact of the new scheme on TC track predictions, more than 672 cases from 27 different TCs during 2011-2012 seasons are examined. The initial study result shows that, compared with operational scheme, the new scheme can generate reasonable and realistic vortex in the initial fields and produce TC track forecast more accurately. Statistical analysis shows a decrease in the average track error of 3%-15% during the 48-120 h time period.