Deep Learning Techniques for High-Resolution Coastal Modeling

In this work, we focused on the development and application of artificial intelligence methods for coastal downscaling, particularly in the context of the northern Adriatic Sea. The motivation stems from the fact that regional-scale oceanographic models, such as those provided by the Copernicus Marine Service, lack the spatial resolution needed to represent fine-scale coastal processes. River discharges, salinity gradients, and nutrient variability are often poorly captured by these models, limiting their usefulness for coastal monitoring and management. To overcome this challenge, our activities concentrated on designing, training, and validating a deep learning model capable of downscaling coarse-resolution outputs into high-resolution fields. The work has been published in Ocean Modelling 2025 [Adobbati et al. 2025].