This study aims to assess the sensitivity of river level estimations to the stream–aquifer exchanges within a hydrogeological model of the Upper Rhine alluvial aquifer (France/Germany), characterized as a large shallow aquifer with numerous hydropower dams. Two specific points are addressed: errors associated with digital elevation models (DEMs) and errors associated with the estimation of river level. The fine-resolution raw Shuttle Radar Topographic Mission dataset is used to assess the impact of the DEM uncertainties. Specific corrections are used to overcome these uncertainties: a simple moving average is applied to the topography along the rivers and additional data are used along the Rhine River to account for the numerous dams. Then, the impact of the river-level temporal variations is assessed through two different methods based on observed rating curves and on the Manning formula. Results are evaluated against observation data from 37 river-level points located over the aquifer, 190 piezometers, and a spatial database of wetlands. DEM uncertainties affect the spatial variability of the stream–aquifer exchanges by inducing strong noise and unrealistic peaks. The corrected DEM reduces the biases between observations and simulations by 22 and 51% for the river levels and the river discharges, respectively. It also improves the agreement between simulated groundwater overflows and observed wetlands. Introducing river-level time variability increases the stream–aquifer exchange range and reduces the piezometric head variability. These results confirm the need to better assess river levels in regional hydrogeological modeling, especially for applications in which stream–aquifer exchanges are important.