Typhoon winds near its external eye wall are strongly non-stationary and disastrous, requiring a deep understanding. This paper first presents an S-transform-based method for obtaining the time-varying power spectra and coherences of a multivariate non-stationary process. The accuracy of the proposed S-transform-based method is examined through a comparison with currently-used two methods. The analytical expressions of time-varying power spectra and coherences of non-stationary typhoon winds are then proposed by introducing time-varying parameters into the stationary Von Karman wind spectra and the stationary Krenk wind coherence functions respectively. Finally, the S-transform-based method is applied to the wind data recorded by the multiple anemometers installed in the Stonecutters Bridge in Hong Kong during Typhoon Hato, and the resulting time-varying wind spectra and coherences are fitted by the analytical expressions of time-varying Von Karman wind spectra and Krenk wind coherence functions respectively. The results show that the typhoon winds recorded during Typhoon Hato are clearly non-stationary and that the time-varying Von Karman wind spectra and Krenk wind coherence functions could well fit the wind data recorded during Typhoon Hato.