<p>Although the temporal decay of earthquake influence on post-seismic landslides is widely recognized, the duration of seismic legacy effects remains debated. This uncertainty partly arises from differences in the types of post-seismic landslides considered across studies, resulting in poorly constrained spatiotemporal controls of post-seismic new landslides. Here we focus on landslides newly triggered in areas that remained stable during the mainshocks of the 2008 M_w 7.9 Wenchuan, China, 2015 M_w 7.8 Gorkha, Nepal, and 2018 M_w 7.5 Palu, Indonesia earthquakes. We propose a nested random forest framework designed to address modeling uncertainty, and apply it to analyze nine landslide inventories forming a coherent post-seismic temporal sequence. Results indicate that the influence of peak ground acceleration (PGA) on new landslides declines rapidly during the early post-seismic period and becomes a minor contributor after approximately 3 years. Over the same post-seismic period, individual topographic proxies exhibit temporal variability, whereas their aggregated contribution remains remarkably stable and consistently exceeds 50%. In contrast, lithology type and distance to faults consistently show low relative importance. These results delineate distinct temporal behaviors between time-dependent variables, such as seismic forcing and extreme rainfall, and more stable terrain-related controls on post-seismic new landslides. These findings provide a more constrained basis for developing spatiotemporal susceptibility models of post-seismic new landslides.</p>