A radar-based hurricane initialization scheme has been developed for high resolution numerical models. The newly developed initialization scheme utilizes the bounded derivative initialization (BDI) in conjunction with the vorticity method to accurately determine the hurricane vortex dynamic structure based on the single Doppler radar wind fields retrieved from the ground-based velocity track display (GBVTD) technique. The GBVTD technique retrieves the reliable rotational wind and vorticity of a tropical cyclone (TC) from single Doppler radar data. Based on the momentum conservation, the vorticity method derives the hurricane divergent wind/vertical velocity from the high temporal and spatial vorticity variations retrieved by GBVTD. The divergent wind inferred dynamically and the rotational wind observed from single Doppler radar data form the total wind field, including the vertical velocity in a hurricane vortex. BDI is used to smoothly insert the derived hurricane vortex into environmental flows defined by forecast fields from larger scale models. The MM5 four-dimensional data assimilation system (FDDA) is used to nudge to the derived smooth kinematic wind fields in order to retrieve unobserved fields such as temperature and moisture fields. The balanced hurricane vortex with thermodynamic and moisture fields is used as the initial condition for the simulation of hurricane Danny with high resolution MM5/WRF models. This hurricane initialization scheme has been tested by assimilating the wind fields obtained from GBVTD analysis of Hurricane Danny (1997) into 1.5 km MM5 model simulations. Numerical experiments show substantial improvements on track and intensity forecasts, in particular, wind fields, can be obtained by the initialization scheme based on radar observations. For examples, after the initialization,the simulated hourly track closely follows that of observed for a period of time. The intensity forecast of winds improves by more than 10 m/s compared with those without initialization.