Abstract: Extreme wind like thunderstorm has a destructive effect on structures adopted by buildings and transmission towers. In order to analyze the dynamic response of a structure, it is essential to obtain reliable wind samples. Thusly, a time-frequency analysis-based method, composed of multivariate empirical mode decomposition (MEMD), amplitude modulation/frequency modulation (AM/FM) decomposition and proper orthogonal decomposition (POD), is proposed to simulate multivariate non-stationary wind speed. Firstly, MEMD decomposes multivariate non-stationary wind speed into a set of intrinsic mode functions. Secondly, AM/FM decomposition calculates the instantaneous frequency and instantaneous amplitude of each intrinsic mode function. Thirdly, the instantaneous frequency is decoupled by POD. Finally, the instantaneous amplitude and decoupled instantaneous frequency is used to reconstruct multivariate non-stationary wind speed. The simulation results of measured thunderstorm wind datas shows that MEMD-AM/FM decomposition-POD serves as a powerful tool in the simulation of multivariate non-stationary thunderstorm wind speed.