Satellite mapping reveals extensive industrial activity at sea

The study addresses the challenge that major ocean industries—fishing, transport, and offshore energy—are poorly quantified at global scale due to limitations in public tracking and restricted data access. Despite the ocean underpinning food security, global trade, and energy production within a rapidly growing blue economy, industrialization has contributed to environmental decline, overfished stocks, and habitat loss. Existing monitoring approaches are incomplete: many vessels do not use or are not required to use AIS, some manipulate signals, and coastal reception can be poor; VMS data are often proprietary; and information on fixed offshore infrastructure can be restricted or outdated. Consequently, human expansion into the ocean is poorly documented. The purpose of this study is to create a detailed, global map of industrial activities at sea by integrating satellite imagery, AIS, and deep learning, thereby revealing the scale, distribution, and temporal dynamics of fishing and other vessel activities, as well as offshore oil and wind infrastructure, from 2017–2021.

Prior work has mapped fisheries using AIS and regional VMS, but these approaches are limited by regulatory variability, proprietary data, poor satellite reception, and AIS manipulation or deactivation, especially among illicit fleets. Large blind spots exist along many coastlines, and public datasets on offshore infrastructure are fragmented or restricted. AIS-based analyses have suggested regional activity patterns (e.g., comparable fishing activity between Europe and Asia), but these may misrepresent true distributions because many vessels are not publicly tracked. Studies have also documented reductions in marine traffic during COVID-19 and highlighted the role of large transnational corporations in the ocean economy. Overall, the literature underscores the need for transparent, global, high-resolution monitoring that includes vessels not publicly tracked and comprehensive mapping of offshore infrastructure.

The authors analyzed approximately 2 petabytes of satellite imagery covering coastal waters worldwide from 2017 to 2021, focusing on more than 15% of the ocean where over 75% of industrial activity occurs. They processed more than 67 million image tiles using deep convolutional neural networks. Input imagery included dual-polarization synthetic-aperture radar (SAR) from Sentinel-1 and optical (RGB and NIR) imagery from Sentinel-2. Three deep-learning models were developed: (1) object detection and length estimation for vessels and structures (>97% accuracy; length estimation R^2 ≈ 0.84); (2) classification of offshore infrastructure into oil, wind, and other structures (>98% accuracy); and (3) classification of vessels as fishing or non-fishing (>90% accuracy). The SAR resolution enabled detection of most objects larger than 15 m, with a detection rate >70% for 25-m vessels and >90% for vessels 50 m and larger. The authors also analyzed 53 billion AIS GPS positions, matching AIS tracks to satellite detections to determine whether vessels were publicly tracked. Spatial analyses summarized detections across 0.1° grid cells (~11 km), depth gradients, continents, and exclusive economic zones (EEZs). Temporal analyses constructed time series from average detections per satellite overpass to evaluate annual cycles and pandemic-related changes, stratifying patterns inside and outside China.

• Scale of untracked activity: An estimated 72–76% of industrial fishing vessels were not visible in public AIS data, whereas 21–30% of transport and energy vessel activity was not publicly tracked. - Global distribution: At any given moment during 2017–2021, around 63,300 vessel occurrences were detected, with 42–49% being fishing vessels (23.1 million vessel detections total). Vessels were detected at least once in 84% of imaged 0.1° cells, but half of all activity occurred in less than 3% of cells, indicating strong spatial concentration. - Depth and regional patterns: Most activity occurred in shallow waters: 86% of fishing and 75% of non-fishing detections were in waters <200 m deep (which represent only ~7% of the ocean). By continent, 67% of all vessel activity was in Asia, followed by Europe (12%), North America (7%), Africa (7%), South America (4%), and Australia (2%). - Fishing dominance in Asia: Contrary to AIS-based impressions, Asia dominated industrial fishing, accounting for 70% of fishing vessel detections; nearly 30% of all mapped fishing vessels were in China’s EEZ. The Mediterranean showed more balanced fishing detections between its northern (European) and southern (African) coasts than AIS alone suggests. - Hotspots of untracked activity and possible illegality: Extensive not publicly tracked fishing occurred around Indonesia, South and Southeast Asia, and along the northern and western African coasts. Near the Korean Peninsula, the western side exhibited the world’s highest density of fishing vessels (about 40 vessels per 1,000 km²) during 2017–2019; activity peaked in May (coinciding with China’s fishing moratorium) and then fell by ~85% during the COVID-19 pandemic as North Korea closed borders. - Activity in protected areas: Not publicly tracked vessels were detected inside notable MPAs, including averages of >5 per week in the Galápagos Marine Reserve and >20 per week in the Great Barrier Reef Marine Park. - Temporal trends and COVID-19: Global industrial fishing activity decreased by 12 ± 1% at the onset of COVID-19 and had not fully recovered by 2021. The decline was greater outside China than within China. In contrast, transport and energy vessel activities remained stable or slightly increased over 2017–2021, with greater growth in China than elsewhere. - Offshore infrastructure scale and growth: By end of 2021, ~28,000 offshore structures were mapped. Of these, 48% were wind turbines and 38% oil structures, with the remainder other structures (e.g., piers, bridges, power lines, aquaculture). The Gulf of Mexico had the largest concentration of offshore oil infrastructure in open ocean; the USA accounted for >2,200 oil structures, Saudi Arabia >770, and Indonesia >670. - Offshore wind expansion: Offshore wind development was concentrated in northern Europe (52%) and China (45%). The number of offshore oil structures grew ~16% since 2017 (with a decline in the USA offset by increases elsewhere), whereas wind turbines more than doubled since 2017, likely surpassing the number of oil structures by the end of 2020. China led with a ~900% increase in turbines from 2017 to 2021 (~950 turbines added per year); the UK and Germany increased offshore wind capacity by 49% and 28%, respectively, since 2017. - Vessel interactions with infrastructure: Trawlers avoided fishing within ~1 km of oil structures (likely to avoid net entanglement), whereas other gear types were attracted to structures, possibly due to fish aggregation. Wind-related effects on industrial fishing appear smaller currently because wind farms are concentrated and generally farther from shore. Oil-related vessel traffic had a much larger footprint than wind-related traffic in 2021, accounting for ~4,140,000 hours of vessel activity versus ~792,500 hours near wind turbines.

The integration of SAR and optical satellite imagery with AIS and deep learning overcomes key limitations of AIS-only approaches, revealing the true extent and distribution of industrial activities at sea. The findings demonstrate that AIS substantially underrepresents fishing, particularly in Asia and in coastal regions with poor AIS reception or where regulations do not require AIS. The more accurate global distribution of fishing—dominated by Asia and including significant untracked activity in MPAs—highlights regions where management, enforcement, and transparency need strengthening. The observed 12% global decline in fishing during COVID-19, contrasted with the resilience or growth in transport and energy activity, underscores sector-specific sensitivities to global disruptions. Offshore energy is undergoing a transition: offshore wind is rapidly expanding and has likely surpassed oil structures in count by 2020, though oil-related vessel traffic still dwarfs wind-related traffic. The observed behavioral responses of fisheries to infrastructure (e.g., trawl exclusion zones near oil platforms) suggest that expanding offshore infrastructure will reshape fishing patterns and potential habitat effects. Overall, the results refine global assessments of human pressure on marine ecosystems and provide actionable intelligence for fisheries management, maritime governance, and planning for offshore energy expansion.

This study delivers a high-resolution, global map of industrial vessel activity and offshore infrastructure by fusing satellite imagery, AIS, and deep learning, revealing that most industrial fishing is not publicly tracked and that offshore wind has grown rapidly to rival or exceed oil infrastructure in count. The openly available dataset enables nations and stakeholders to detect potential illegal activities, monitor encroachment into artisanal fishing grounds and MPAs, and more accurately assess fishing effort and vessel traffic for environmental and climate policy. The COVID-19-driven reduction in fishing may reflect a longer-term plateau or decline in global fishing activity, while transport and energy sectors continue to expand. Future research should: integrate additional sensors and higher revisit rates to improve detection of smaller vessels; expand coverage beyond current coastal focus; link vessel activity to environmental and socio-economic outcomes; refine classification of gear types and behaviors; and evaluate cumulative ecological impacts of growing offshore infrastructure and associated traffic.

• Coverage and detectability: Analyses covered more than 15% of the ocean (primarily coastal waters) where most activity is concentrated, so truly pelagic and remote regions are less represented. SAR detection is most effective for objects larger than ~15 m; smaller vessels may be under-detected, biasing results toward larger industrial vessels. - AIS data constraints: Matching relies on publicly accessible AIS, which is affected by regulatory gaps, deliberate disabling or spoofing, and poor coastal satellite reception; thus, “not publicly tracked” includes both truly non-broadcasting vessels and those not captured by public AIS services. - Classification uncertainties: Although model accuracies are high (>90–98%), misclassification of vessel type (fishing vs non-fishing) or infrastructure category (oil vs wind vs other) can occur, and length estimates have uncertainty (R^2 ~0.84). - Temporal inference: Pandemic-related comparisons use time series derived from average detections per satellite overpass and may be influenced by regional variations in image acquisition frequency. - Infrastructure attribution: Counts of oil and wind structures include confidence bounds (with possible inclusion of lower-confidence detections outside major development areas), and changes in national totals may reflect both real development and detection uncertainties.