The relatively low damage in the Kathmandu Valley caused by the 2015 7.8 Gorkha earthquake has attracted much attention. To gain a deeper understanding of this phenomenon, we conduct broadband ground‐motion simulations for both the mainshock and the 7.2 Dolakha aftershock through a hybrid method that combines deterministic 3D synthetics at relatively low frequencies () and semistochastic synthetics at higher frequencies (). Because they are summarized in a companion paper (Wei et al., 2018), the 3D deterministic synthetics were generated by embedding a finite‐fault rupture model in a 3D velocity model that is characterized by a simplified basin structure for the Kathmandu Valley. We tested different weighting schemes using a finite slip model and backprojection results to weight the high‐frequency sources. Our simulations were guided by fitting the observations from five strong‐motion stations in Kathmandu Valley and the intensity and mortality distributions. Site effects were handled by amplitude spectra ratio derived from the vertical component of a hard‐rock station (KTP). Our broadband ground‐motion simulations show that (1) the stress parameter (3.8 MPa) of the mainshock was much lower in comparison to the 7.2 aftershock (23 MPa) that suggests the rupture process of the mainshock was relatively deficient in radiating high‐frequency energy and different fault friction property between the mainshock and the aftershock; (2) the soft deposits in the Kathmandu Valley experienced a pervasive nonlinear site response during the mainshock and the 7.2 aftershock, which also contributed to the reduction of high‐frequency motions; and (3) the high‐frequency ground motions during the mainshock were primarily radiated from the down‐dip rupture. Hence, we suggest considering the difference in the distribution of high‐frequency radiation and fault slip in the broadband ground‐motion simulations for scenario and historical earthquakes.