Abstract: In this paper, the feasibility of identifying nonlinear temporal changes of sites by moving time window deconvolution method is analyzed based on the simulation records and observation records from the 2011 M_W9.0 Tohoku earthquake, and the results are compared with those by the moving time window spectral ratio method. The results show that the moving time window deconvolution method can reveal the nonlinear change process of the site with ground motion level. Based on this method, the threshold of nonlinearity, the degree of nonlinearity change and the recovery degree after strong ground motion process can be identified. Compared with the moving time window spectrum ratio method, the moving time window deconvolution method can obtain more stable nonlinear temporal changes process of soil, but for the shallow surface soil with strong impedance ratio, the moving time window spectrum ratio method can obtain more accurate nonlinear degree results. The nonlinear temporal changes analysis of eight KiK-net stations during the 2011 M_W9.0 Tohoku earthquake shows that the nonlinear thresholds identified by the two methods are close to each other, ranging from 40 cm/s² to 100 cm/s², and there is no obvious correlation with site v_S30. At IBRH20 station with low PGA, the decrease of wave velocity caused by nonlinearity was small （3%） and the recovery was almost complete after the earthquake. At the other seven stations with high PGA （range of 386—822 cm/s²）, the site equivalent shear wave velocity decreased by 13%—37%, resulting in significant site nonlinearity and not fully recovered after the earthquake. As PGA is larger than 380 cm/s², there is no significant correlation between PGA and the decrease/recovery of wave velocity caused by nonlinearity.