Based on the records of six KiK- net stations in the 2021 Fukushima earthquake (M_JMA7.3), the nonlinear time- varying process of the site was studied using the moving- time- window deconvolution method, moving- time- window spectral ratio method, and moving- time- window autocorrelation function method. Then, the threshold and degree of nonlinearity were analyzed. Results show that the ability of the three methods to identify the nonlinear time- varying characteristics of the site differs. The moving- time- window deconvolution method can easily obtain a more stable nonlinear time- varying response for soil, but the degree of nonlinearity is low. The method reflects the change of average wave velocity from the surface to the underground. For shallow soils with a strong wave impedance ratio, the nonlinear degree of identification results derived from the moving- time- window spectral ratio method and moving- time- window autocorrelation function method are strong. These two methods cannot provide stable nonlinear time- varying results of shallow soils with a low wave impedance ratio. The strongest nonlinear moments identified by the three methods are basically the same and are located near the peak acceleration of the whole record. The nonlinear threshold values of the six stations identified by the three methods are about 40-140 gal, and the shear wave velocity drops between 16% and 53%, indicating that the degree of site nonlinearity is high.