Landslides are dramatic and complex surface processes that can result in extensive casualties and property damage. The broadband seismic signals generated by landslides provide datasets essential for understanding time‐dependent sliding processes. However, traditional methods for computing Green’s functions based on wavenumber integration converge very slowly for surface sources, especially at high frequencies. Usually, long‐period synthetic waves with a cutoff ‐integral for an approximated near‐surface source are adopted for landslide studies, which may lead to artifacts. Thus, the development of efficient methods for computing the broadband Green’s functions of surface sources is important. The generalized reflection and transmission method with the peak‐trough averaging technique can overcome the difficulties in wavenumber integration for surface sources, quickly converging even for high‐frequency calculations. We use this improved method to compute Green’s functions for surface single‐force sources and invert the force histories of the 2017 devastating Xinmo landslide in different frequency bands. The results indicate that the complex sliding process of this drastic event can be revealed by broadband signals (0.02–0.5 Hz), and that the initiation stage of this event shows a dominant frequency up to 0.2 Hz.