AbstractIn research and development of photo-electrochemical (PEC) cells for water splitting, most results up to now are based on simulations and experiments on laboratory scales. However, to make PEC cells attractive for application, scaling up their size and energy efficiency is necessary. Therefore, we investigate the effects of stepwise upscaling of PEC cells. On the experimental level, cells with metal oxide electrodes of different sizes and shapes as well as cell tank geometries are characterized with respect to their surface reactions and photo-current output. In order to predict their behavior on different scales, a computer-aided reference model is developed simultaneously. This is benchmarked by testing various cell sizes and shapes, enabled by fast and cost-efficient fabrication via 3D printing. Machine learning via Bayesian optimization was employed to optimize the PEC cell simulation model input parameters, resulting in very good agreement within a few percent of computed and measured current–voltage curves with Pt-electrodes. Transferring these input parameters to the same cell geometry but with a semiconductor anode, deviations of less than 25% were observed. Here, we present experimental results of the PEC cells, as well as the first drafts of the simulation model and the optimization approach.