Stewarding our blue planet

Introduction

The ocean, a vast interconnected system fundamental to planetary habitability, regulates climate by absorbing anthropogenic CO₂ and over 90% of excess heat.¹ It sustains crucial biodiversity and food webs underpinning global ecosystems and resources. However, escalating anthropogenic pressures, overexploitation, pollution, habitat degradation, and climate change impacts demand effective governance through robust sustainable and equitable management frameworks.

Sustainable ocean governance is complicated by a fragmented legal landscape and diverse international agreements and management efforts. Key measures include mitigating climate change impacts (e.g., by improving energy efficiency, promoting sustainable transport, and protecting and restoring blue carbon ecosystems), establishing marine protected areas, implementing science-based fisheries management, and tackling pollution.² However, persistent gaps in scientific research and monitoring, and inconsistent data, hinder evidence-based decision-making. Traditional monitoring methods are often resource-intensive and limited in scope, impeding accurate ecosystem assessment and adaptive policy development.³

Artificial intelligence (AI), encompassing machine learning (ML), deep learning (DL), and advanced analytics, emerges as a transformative tool to bridge these gaps. It can facilitate large-scale, real-time surveillance of marine activities, automate the identification of marine species and threats, improve predictive modeling of oceanographic changes, and advance data processing efficiency.⁴ This technological advancement can provide critical insights for informed, adaptive ocean governance, supporting the implementation of international agreements. Nevertheless, AI integration poses challenges, including data representativeness, algorithmic biases that may exacerbate existing inequalities, and ocean justice concerns.

This chapter illustrates AI applications in ocean governance, conservation, and area-based management for understanding dynamics and predicting change. Several applications that focus on optimizing commercial operations, infrastructure, and engineering efficiency (e.g., port operations and logistics, engineering, and offshore infrastructure) fall beyond this chapter’s scope.

AI for Data Collection, Monitoring, and Surveillance

Addressing the ocean’s persistent monitoring deficit requires novel, AI-driven approaches that can complement traditional resource-intensive methods. Autonomous sensor-equipped platforms can use onboard analytics power to optimize survey routes and extract features from data in real time, potentially informing real-time ecosystem management.^5,6 Such platforms have fundamentally expanded our capacity to collect and analyze data by integrating a broad variety of sensors for detecting biological and non-biological scatterers in the water, and above-water and underwater imagery including hyperspectral imaging for water quality assessment. Many of these data streams are suitable for developing automated analysis pipelines around, using ML and DL approaches for real-time feature identification:

Marine acoustics is a research area with large potential for automation and real-time feature extractions. It includes both passive acoustics (with autonomous recorder units capturing impulsive underwater sound signals) and active acoustics (devices emitting sound waves and recording the returned echoes, i.e., echo sounders). These complex acoustic datasets can be processed with various ML methods to, for example, distinguish seabed types and structures,7 identify fish schools,⁸ detect hydrothermal emissions, and monitor small whales based on their emitted contact sounds.⁹ This use of AI in analyzing marine soundscapes not only advances mapping and classification, but can also help detect ecological changes and anthropogenic impacts that might not be visible using optical techniques.

Big data is starting to play a major role in imagery and video data. Underwater imagery,¹⁰ as well as above-water imagery of marine organisms such as seabirds¹¹ and marine mammals,¹² is increasingly collected using automated systems, and analyzed using DL frameworks enabling large-scale, automated identification of marine species. Platforms such as FathomNet¹⁰ leverage DL to analyze millions of underwater images and recordings, dramatically accelerating species classification and population tracking. Advanced ML applications are also starting to expand beyond simple classification tasks toward identifying behavior, growth, and individual performance, which is often highly informative for revealing population status and threats to marine animal populations.^13,14

Such automation of data collection and feature extraction can aid biodiversity monitoring. Emerging threats to marine biodiversity, especially in remote or under-observed regions, are increasingly studied using a combination of remote sensing and sequencing data.¹⁵ Remote sensing technologies allow near real-time tracking of changes in large, inaccessible marine areas,¹⁶ while tagging and in-situ fieldwork support a robust picture of ecosystem health.¹⁷ Meanwhile, environmental DNA analyses allow ML models to detect rare or endangered species from trace genetic material in ocean samples,¹⁸ uncovering microbial diversity and revealing how different organisms contribute to nutrient cycling and ecosystem resilience. This integration of remote and autonomous sensing, acoustic monitoring, and advanced analytics is fundamentally redefining our ability to assess ecosystem changes by delivering comprehensive, near-real-time insights that support both scientific discovery and adaptive, evidence-based ocean stewardship.

AI-Assisted Predictive Modeling

Predictive modeling plays an increasingly vital role in helping scientists and policymakers anticipate oceanic responses to accelerating anthropogenic pressures. However, the complexity of marine systems, marked by feedback loops, threshold-driven events, and persistent data scarcity, often limits the capacity of traditional modeling frameworks. Advances in AI are beginning to address some of these challenges by enabling integration of multimodal data streams and supporting higher-resolution forecasts.

DL methods are advancing the detection and classification of marine pollutants. For example, U-Net neural networks trained on Sentinel-2 satellite imagery demonstrate great potential for large-scale identification of floating marine plastics, providing a scalable monitoring framework aimed toward automated global plastic tracking systems.¹⁹ Similarly, integrating hyperspectral sensors with ensemble algorithms enhances the monitoring of oil spills and other contaminants. When combined with ocean current modeling, these approaches help anticipate pollutant drift and accumulation, enabling more targeted cleanup efforts. Emerging evidence shows that AI algorithms (e.g., ML, DL), along with big data analytics, Internet of Things (IoT)-enabled sensors, and smart sensor technologies, can improve environmental monitoring, disease and feed management, production optimization, and traceability in fisheries and aquaculture. The combination of technologies can thus support sustainability, waste reduction, and increased supply chain transparency.²⁰

The application of AI in modeling global marine changes is advancing. Convolutional neural networks (CNNs) help correct systematic biases in sea ice projections—crucial indicators of climate-driven ocean changes—while physics-informed neural networks trained on long-term hydrographic and turbulence observations more accurately represent small-scale ocean processes such as vertical mixing.^21,22 These models, which integrate data-driven insights with physical constraints, not only outperform traditional parameterizations, yielding more realistic predictions of ocean temperature and heat fluxes in both stand-alone and coupled climate models, but also improve projections of phenomena like marine heatwaves and ocean acidification.

AI-driven analysis is also expanding our understanding of the interplay between human activities and marine environments. Notably, ML models trained on global datasets of daily ship movements have revealed fundamental patterns in maritime trade evolution, enabling accurate forecasts of shipping routes and trade flows.²³ These predictive models might improve our ability to anticipate the impact of maritime logistics on ecological systems and climate change, potentially supporting more informed sustainability planning. Another example is the global surveillance of fishing activities.²⁴

AI is also being used to model changes in marine biodiversity, for instance by predicting coral bleaching through analysis of thermal stress, nutrient levels, and ocean chemistry. In blue carbon ecosystems, ML algorithms like random forest and XGBoost are used to synthesize light detection and ranging (LiDAR), remote sensing-, and field data, resulting in more accurate estimates of carbon stocks across mangroves and seagrasses.²⁵ The use of cloud computing and high-performance hardware further enables the processing of these increasingly complex, multi-source datasets.

AI-Assisted Decision-Making

AI is increasingly supporting decision-making in ocean governance, particularly in fields such as marine protected area (MPA) management, maritime spatial planning (MSP), and combating illegal, unreported, and unregulated (IUU) fishing. These applications are grounded in the integration of diverse, spatially explicit ecological and operational datasets, with the goal of informing management that balances conservation needs and human activities. Several important application areas and contributions can be identified:

For MSP, coupling Bayesian networks with GIS maps enables multi-scenario assessments of cumulative impacts under various climate and management conditions. The coupling also allows the identification of key pressure-driving impacts and potential protective measures to reduce environmental vulnerability.²⁶ In the European context, a case study was recently conducted on the Italian region and emerging AI hub of Emilia-Romagna. The case study demonstrates how integrated AI-supported MSP platforms, which combine multiple geospatial decision support tools with extensive ecological and socioeconomic datasets, can enable transboundary, ecosystem-based MSP. The MSP platforms do so by providing rapid, spatially explicit outputs for scenario analysis.²⁷ Globally, AI can help identify high-priority areas for MPA establishment by analyzing integrated global datasets on species diversity, habitat heterogeneity, benthic features, productivity, and human activities such as fishing effort. AI can also support the implementation of the UN’s Biodiversity Beyond National Jurisdiction Agreement (BBNJ)—the key multilateral framework for high seas conservation.²⁸

AI contributes to the detection and analysis of IUU fishing. CNNs, applied to satellite imagery (including synthetic aperture radar, vessel monitoring systems, and automatic identification system data), have enabled more accurate and persistent monitoring of vessel activity on the open ocean. In particular, these methods can help to identify vessels operating without tracking signals (“dark vessels”), recognize atypical fishing patterns, and estimate unreported fishing effort.^29,30 Identifying such deviations from normal vessel behavior can also support enforcement efforts by authorities.31Combining fishing activity data with satellite altimetry, this approach can also reveal fine-scale fishing activity patterns linked to ecological drivers, facilitating sustainable management and the protection of marine biodiversity.³²

Limitations and Key Challenges

Oceans remain critically understudied and “under-governed,” with 64% of waters beyond national jurisdiction and deep-sea ecosystems largely unexplored.6,³³ This reality intensifies three interconnected challenges that constrain AI’s transformative potential and risk perpetuating existing inequities:

Persistent data gaps form the first challenge. AI’s performance depends on abundant, high-quality data, yet ocean observations remain sparse and fragmented. The Intergovernmental Oceanographic Commission of UNESCO (IOC-UNESCO) estimates that more than 90% of marine ecosystems lack sufficient data for effective management,³⁴ which hinders progress toward global targets such as the Convention on Biological Diversity’s goal to protect 30% of the ocean by 2030—a target aimed at conserving marine biodiversity and ensuring the sustainable use of ocean resources.³⁵

These issues stem from deep-seated inequalities in scientific capacity between the Global North and the Majority World, which manifest in disparities in funding, access to research infrastructures, thematic priorities, and influence over global ocean governance discourses.^36,37 These data coverage and scientific capacity gaps converge on issues of ocean justice. Without dedicated capacity building aimed at more inclusive, diverse, and equitable exploration and exploitation practices,³⁸ AI models risk producing inaccurate or unjust outcomes due to bias in training data. Algorithmic biases emerging from unrepresentative training data can skew conservation priorities away from vulnerable regions or communities. Making private data available to coastal states that cannot afford to buy or collect it themselves can represent a first step in that direction.³⁹

Developing adaptive, multi-scalar governance regimes will be essential to harness AI’s potential for addressing the “wicked problems” of the global ocean and the challenge of scaling up the potential of AI within ocean governance frameworks.⁴ AI makes it possible to monitor ecosystems faster than ever before, but it also brings new challenges. The timing and scale of ecological data don’t always match up with what’s practical or realistic for social, economic, or political decisions. Bridging that gap will be key if policymakers are going to make real use of AI.