Spatiotemporal variations in trawling intensity in the German Baltic Sea Basins based on bathymetric data

Abstract

Bottom trawling is the largest source of anthropogenic seafloor disturbance globally, yet spatiotemporal variations are poorly constrained in most regions due to non-reporting or low-resolution fishing effort data. To overcome this gap in knowledge, we use a U-Net convolutional neural network to segment trawl marks across approximately 1,069 km 2 of seafloor in the German sector of the Western Baltic Sea. The model was trained and applied to open-source bathymetric data collected between 2016 and 2025 with a resolution of 1 m. Trawling intensities vary substantially in space and over time, as evidenced by time-lapse bathymetric data for the Fehmarn Belt area. Here, repeated surveys of a Marine Protected Area (MPA) in 2024 and 2025 confirm seafloor regeneration timescales of approximately one year. Within the MPA, trawl mark densities and AIS-derived fishing effort both increased from 2024 to 2025, indicating continued trawling activity despite a bottom trawling ban that entered into force in 2025. Circular untrawled zones in Mecklenburg Bay, often with diameters of more than 500 m, coincide with seafloor pockmarks indicating localized fluid seepage. We attribute these untrawled zones to fishermen avoiding the pockmark depressions which pose a risk to bottom-contact gear. Trawl marks are morphologically healed in areas with intense commercial ship traffic. This study demonstrates the potential of open-source bathymetric data to monitor