Applications of deep learning super-resolution methods for coastal ocean modelling

Deep learning algorithms are gaining popularity in oceanography due to their lower computational cost compared to classical numerical methods and to the increasing availability of ocean data. In particular, super-resolution techniques can help recover small-scale features, which are especially important in the analysis of coastal regions. These methods have a wide range of applications, including the generation of high-resolution forecast data, of long-term projections, and the reconstruction of past conditions through the integration of model outputs and observational data. In a recent work, we developed a deep learning method for super-resolution in coastal areas by adapting techniques originally designed for super-resolution of digital images. We applied this method to the northern Adriatic Sea, a sub-basin of the Mediterranean Sea. This allowed us to identify the conditions under which the neural network performs best. We are currently working to integrate the neural network with our existing tools, to combine the advantages of both deep learning and numerical approaches. The aim of this contribution is to present our ongoing work on super-resolution for the northern Adriatic Sea, summarizing the results achieved so far and outlining the current challenges in practical application scenarios.