European offshore wind energy (OWE) production is growing at an accelerated rate in the blue economy sector. However, existing knowledge about wind farm interactions with the marine environment needs improvement to make OWE expansion sustainable. DTO4OWE – Digital Twin of the Ocean for Offshore Wind Energy, a 3-year European research project, aims to extensively study long-term OWE impacts on physical and biological systems in the North Sea and Baltic. Numerical modeling, combined with data-assimilation, is the gold standard in ocean science, but with OWE the increased need for fidelity makes studies on climate timescales impractical. Here there is an obvious need for complementary machine learning (ML) methods that downscale (upsample) ocean model output. Broadly speaking the task is to learn a time-dependent residual by training on both coarse and high-fidelity experiments. This residual then constitutes a subgrid parametrization to augment coarse climate simulations. This project is a proof-of-concept/pilot study into ML downscaling methods to complement classical numerical modeling in a limited domain. We will test both purely image/snapshot-based methods (SRResNet) and lightweight time-dependent ocean emulators (CAE-ESNc, OceanNet). The aim is to use a downscaling approach to include OWE impacts in ocean currents, wave fields, mixed-layer depth, etc. on long (climate) timescales. To this end we will not only train using point-wise losses (MSE) but also test adding derived quantities such as energy, enstrophy and spectral properties in the objective functions.