Abstract: The spatial variabilities of S- wave velocity (V_s) structures have a significant influence on the results of seismic site response analyses and liquefaction discrimination. Based on the borehole data in Bohai Bay, the variations of means and standard deviations of seabed V_s with depth h are developed by using the point estimation method when the power function is used to describe the nonlinear trend of the seabed V_s with h, the Monte Carlo simulation is utilized to verify the accuracy of simulated results through the point estimation method. Combined with measured V_s -profiles at the boreholes, the random seabed V_s- structures are simulated by using the non-stationary conditional random field method considering the trend nonlinearity. A series of possible distributions of the random seabed V_s -structures are obtained. This method can significantly reduce the standard deviations of the simulated random seabed V_s-structures compared with the random field method which neglects the constraint of the measured V_s-profiles at the boreholes and the nonlinear trend of V_s changing with h. A series of possible V_s- structures obtained from random simulations are employed to evaluate the probability of soil liquefaction in the study area, and it is found that ignoring the spatial variability of V_s- structures may underestimate the potential of soil liquefaction. The accuracy of the method proposed in predicting seabed V_s -structures and liquefaction potential is verified by comparing the simulation results with the measured V_s values and liquefaction results of the soil at the reserved boreholes which are not used in the simulations. The relevant methods can provide reasonable shear wave velocity reference values for the seismic response analysis and liquefaction probability discrimination of marine engineering sites.