Abstract: Based on the filtered white noise model, a time-frequency domain hybrid dimension-reduction model of fully non-stationary ground motion random field is proposed to overcome the bottleneck that the traditional filtered white noise model can only simulate the ground motion process but not the field. First, to solve the problem that the impulse response function in the filtered white noise model is difficult to directly reflect the spatial coherence of the ground motion random field, the frequency domain representation of the white noise random vector process with spatial coherence is realized by introducing the proper orthogonal decomposition-based dimension-reduction simulation method. Then, the impulse response function and modulation function at different action points of the ground motion random field are used to filter and modulate the corresponding white noise component process, so as to realize the efficient time-frequency domain dimension-reduction expression of the fully non-stationary ground motion random field. Finally, numerical examples are presented to verify the accuracy and engineering applicability of the proposed hybrid time-frequency domain dimension-reduction model of fully non-stationary ground motion random field through the comparisons of mean value, standard deviation, auto-/cross- correlation function with the corresponding targets, and response spectrum, coherence function with strong motion records.