Improving dissolved oxygen simulation is challenging as it depends on uncertain physical and biogeochemical processes and their interactions. We present a machine learning-based emulator, O₂EMU, capable of reducing model biases and inter-model spread by replacing biogeochemical parameterizations with learned relationships between dissolved oxygen and physical variables. The emulator first learns from the historical shipboard and autonomous observations of dissolved oxygen and temperature and salinity from ocean reanalyses, and then applies the learned relationships to the temperature and salinity data in areas with limited or no abservatons or output from models, making it ideal for digital twin implementations. O₂EMU offers a computationally efficient and scalable alternative to standard biogeochemical models, and provides a stepping stone towards hybrid biogeochemical projections that blend mechanistic models with observationally constrained tracer distributions.