The number of seismological observations available in the vicinity of faults is still too limited to fully catch the complexity of strong motion and properly calibrate Ground Motion Models (GMMs). This problem is exacerbated in areas of moderate seismicity, where earthquakes often occur on unknown faults and are only exceptionally recorded in damage areas. Physics-based simulation methods are a very promising approach but they require a very good understanding of the physical processes controlling the strong motion (variability of the rupture process, radiation of high frequency seismic energy, effects of shallow geological structures, etc.), which also requires more observations. Here we propose an approach to constrain near-fault ground motion predictions based on measures of grave slab sliding displacements, e.g., as observed during the Le Teil earthquake in 2019 (Mw 4.9, France) and in Petrinja in 2020 (Mw 6.4, Croatia). In a Bayesian framework, the approach combines a priori information on the ground motion distribution obtained using physics-based simulations and a likelihood function representing the probability of the observed slab displacement for a given ground motion intensity measure, to produce a posterior distribution of ground-motions.