AbstractThe Cretaceous and Cenozoic evolution of the North German Basin is shaped by complex processes involving basin inversion, uplift and erosion, extension and several periods of Quaternary glaciations. Based on a densely spaced long-offset 2D seismic profile network covering the Bays of Kiel and Mecklenburg, we employ a Machine Learning algorithm to pick refracted first-arrival travel-times. These travel-times are used in a travel-time tomography to derive velocity models for the approximately upper 800 m depth of the subsurface. Investigating velocity-depth relations within the Upper Cretaceous strata and analyzing lateral velocity anomalies within shallow depths provide new insights into the magnitude of the Cenozoic basin exhumation and the locations of glacial tunnel valleys. Our findings suggest that previously observed bent-up structures in seismic images are caused by heterogeneous velocities in the overburden and do not represent actual reflectors. We provide strong indications that these misinterpretations of imaging artifacts are related to tunnel valleys even though these valleys might not always be resolvable in seismic reflection or sediment sub-bottom images. Comparing Upper Cretaceous velocity-depth trends to reference trends reveals significantly higher velocities in our study area. We interpret these differences as overcompaction and estimate the apparent Cenozoic exhumation in the Bay of Mecklenburg to be about 475 m. Within the Bay of Kiel, we observe an increase of the apparent exhumation from about 385 m (south) to about 480 m (north). Our study demonstrates the importance of near surface velocity analysis for the investigation of geological processes in shallow marine settings.