First Pockmark susceptibility map of the Italian continental margins

Scientific advancements often emerge from pivotal discoveries and technological breakthroughs, expanding the frontiers of exploration. In geoscience, natural hazard studies have predominantly focused on terrestrial environments, while submarine settings remain relatively unexplored due to the scarcity of high-resolution data, particularly in deep-sea regions. In recent years, explainable machine and deep learning methodologies have shown significant promise in geohazard prediction, enhancing both predictive reliability and process understanding. A key submarine geohazard is linked to fluid flow processes, which influences seabed morphology and geological processes. Pockmarks (seafloor depressions) formed by fluid flow are widespread across diverse geodynamic settings but remain enigmatic in terms of formation mechanisms. This study examines 5,932 pockmarks mapped along the Italian continental margins, primarily on gently sloping, muddy sand seafloor with Plio-Quaternary sediment layers up to 400 meters thick. Stylistically, pockmark distribution appears to be closely linked to fault systems. By integrating field observations with machine and deep learning techniques, we developed a neural network-based pockmark susceptibility model, the first of its kind for the Italian continental margins. Susceptibility maps, widely used in geohazard assessments, differentiate between high- and low-risk areas based on past occurrences and predictive modelling. This tool has significant applications in the planning of submarine and floating infrastructure, navigation safety, and environmental studies related to fluid seepage, climate change, and marine biodiversity. Our findings contribute to a better understanding of submarine geohazards and demonstrate the potential of artificial intelligence in improving geoscientific assessments.