Abstract: Accurate three-dimensional (3D) velocity models are essential for fitting high-frequency seismic waveform records. This process usually requires regional-scale 3D numerical simulations that are computationally expensive, especially with sparse seismic networks. Because of the significance of source domain modeling, we propose a hybrid waveform simulation approach that combines the 3D spectral-element method (SEM) with the displacement representation theorem. By separating near-source wavefield excitation from long-distance wave propagation to stations, only the source domain wavefield needs to be recomputed when the local velocity and source models change. We apply the method to the 2019 M_w 5.0 Changning shallow earthquake to verify its flexibility and effectiveness. We compare high-frequency waveforms computed with different regional velocity models against observations. Results show that the hybrid method achieves accuracy comparable to full SEM 3D simulations while reducing computation costs by more than two orders of magnitude when the structure of the source region updates. Our results further indicate that high-frequency waveforms are highly sensitive to shallow structures. Introducing low-velocity shallow layers into the source region improves near-field waveform fits, indicating pronounced low-velocity sediments in the Changning area. Large surface-wave time delays suggest that shallow velocities within the Sichuan Basin are lower than those in existing published models. In addition, an Interferometric Synthetic Aperture Radar (InSAR)-derived finite-fault model outperforms the point-source model in near-field waveform fitting and better reproduces rupture directivity. The proposed method is practical for high-frequency waveform modeling in areas with complex subsurface structures and rupture processes.