Deepwater marine seismic measurements undergo statics from seawater velocity variations that corrupt seismic imaging, and subsequent time-lapse analysis. The conventional method to correct for statics consists of a two-step workflow, of deriving and applying static corrections to seismic measurements, which has inherent complications. Thus, our research goal is to determine if the conventional workflow can be bypassed by a deep learning algorithm that corrects statics by performing offset and traveltime dependent time-shifts in traces of shot gather data. We use SymAE, an autoencoder-based learning algorithm, to disentangle information arising from subsurface geology coherence and seawater (acoustic) velocity variation from instances of simulated shot gather data. With the learned-SymAE, subsurface instances are redatumed with a reference instance containing the known Hood’s velocity profile, which homogenizes the effects of water velocity across instances. Our results exhibit a strong convergence in arrival times of redatumed instances to those of reference instances. Thus, we conclude SymAE is a learning algorithm capable of correcting for water velocity induced time-shifts. This confirmation paves the way for further investigation into using SymAE in real deepwater acquisition environments, which add further complexities to seismic measurements.