Estimating site effects is important for seismic hazard assessment but can be difficult due to the scarcity of data in time and space, especially in low-to-moderate seismicity regions. Numerical simulations are helpful for understanding the physical parameters controlling site effects and for modelling their spatial variability, but simulations need detailed seismic models of the subsurface in input, as well as empirical measurements of amplification for calibrating their outputs. In this study, we perform 3D numerical simulations of seismic wave propagation and amplification in the Tricastin basin (Rhône valley, France) based on a range of models built with different sources of information. A first type of models consists of layers separated by interfaces interpreted from active seismic profiles and interpolated over the area of interest. A second type of models is obtained by Ambient Noise Surface-Wave Tomography (ANSWT) applied on a dense array of 400 3C sensors and resulting in a 3D model of shear-wave velocities. Array analysis of the seismic noise recorded by this network also provides us with an estimation of attenuation parameters (Qs) within the basin. We compare our numerical results with empirical amplification measurements based on earthquake and ambient noise recordings. Our results highlight the characteristics of the input models in terms of seismic amplification. Layered models generate significant 3D wave propagation effects consistent with the observations but, if over-simplified (e.g. sediments vs. bedrock) do not fully explain the measurements in complex areas of the basin. On the other hand, preliminary models based on purely data-driven ANSWT suffer from the resolution limits of the tomographic process, especially in terms of lateral variations, and do not reproduce expected 3D wave propagation effects, but seem to provide a satisfying estimation of the subsurface velocity structure in complex areas of the basin. This study sheds light into how to acquire and combine information to design and calibrate numerical simulations for site effect estimation.