Abstract: The spatial distribution pattern of faults significantly influences the spatial distribution of ground motions generated by fault rupture. Taking the Dingri M_S6.8 earthquake as an example, this study proposes a source model construction method that inversely maps the slip distribution generated by the Gaussian process (GP) method onto a three-dimensional fault, based on a 3D fault model, the GP source model, and the principal component analysis (PCA) algorithm. By comparatively analyzing the waveforms, peak values of velocity time histories, and seismic intensity distributions between those simulated by the spectral element method and the recorded ones, the rationality and applicability of the proposed method are demonstrated. The characteristics of ground motion attenuation are analyzed. Furthermore, by analyzing the near-field characteristics of the simulated ground motions, the influence of the 3D fault spatial distribution on the spatial distribution of ground motions is revealed. The results indicate that the simulated ground motions satisfactorily reproduce the near-field ground motions of the Dingri earthquake. The attenuation of peak ground velocity (PGV) for the two horizontal components with distance is generally consistent with the PGV attenuation characteristics reflected by ground motion prediction equations. Large PGV values are primarily distributed in the northern segment of the fault. Influenced by the curved fault geometry, these large values also extend towards the southwest, forming an extreme value zone near the fault bend. The spatial distribution of the direction-independent Sv_RotD50 shows that the fault's spatial geometry has a more pronounced effect on long-period Sv_RotD50. The directivity effect of PGV along the main rupture direction is relatively obvious, more significant in the near-fault region, while also exhibiting characteristics of asymmetric bilateral rupture.