Mapping Marine Infrastructure from Space: Object Detection using Satellite Radar Data for Ocean Monitoring and Management

Kurzfassung

The increasing exploitation of marine spaces by offshore infrastructures, like aquaculture facilities, oil and gas platforms, and wind farms, underscores the need for consistent and scalable monitoring approaches. Freely available long-term Earth observation data and machine learning approaches are transforming our ability to monitor the ocean's surface with satellite-based investigations offer consistent and repeatable coverage across regional to global domains. As marine environments face growing and competing pressures from the expansion of aquaculture, the development of wind energy projects, and the extraction of fossil fuels, the ability to systematically map and track infrastructure becomes essential for informed marine spatial planning and the sustainable governance of ocean resources. This study presents deep learning-based object detection frameworks applied to the Sentinel-1 radar data archive to enable the automated, large-scale detection of offshore infrastructure. Neural networks trained on labelled satellite image chips produce consistent, multi-temporal inventories across infrastructure types with distinct spatial signatures and monitoring requirements. Offshore aquaculture is one of the fastest-growing food production sectors, globally. However, spatial inventories of farm locations remain sparse and inconsistent. In SAR imagery, aquaculture facilities, particularly salmon cage farming, appear as characteristic circular backscatter patterns, which makes them tractable targets for automated detection. A deep learning-based object detector trained on labelled data achieves robust localization, enabling quantitative assessment of spatial coverage. For oil and gas platforms, quarterly detections across the North Sea, Gulf of Mexico, and Persian Gulf from 2017 to 2025 identified 3,728 structures in 2025. An analysis of SAR- time series reveals that more than 2,700 platforms were installed or relocated over the study period, with opposed dynamics in different regions, such as a continued Persian Gulf expansion compared to a net decline of platforms in the North Sea and Gulf of Mexico since 2018. Furthermore, a growing proportion of short-lifespan platforms points to a structural shift toward mobile units such as drillships. For offshore wind energy, a global turbine inventory derived from Sentinel-1 time series identifies approximately 15,100 operational turbines in 2025, with major concentrations along the Chinese coast and in the North Sea Basin which together account for about 98.5 percent of global installed capacity. Beyond static mapping, SAR backscatter time series at each detected facility location capture the full temporal evolution of offshore wind infrastructure from 2016 to 2025 with 1 to 12 days intervals, enabling analysis of construction progress, operational dynamics, and vessel interactions. Together, these detection products demonstrate the potential of freely accessible satellite data and deep learning for the consistent, long-term monitoring of marine infrastructure. The resulting data sets provide persistent, satellite-derived information that can complement dynamic ocean models and ecological data to support marine spatial planning, regulatory monitoring, and scenario-based decision-making in increasingly pressured coastal and offshore environments.