By pairing AI with decades of underwater video, scientists have uncovered hidden shifts in marine life, offering a powerful new tool to protect fragile ecosystems in a warming world.
Image Credit: Leonardo Gonzalez/Shutterstock.com
A recent study published in Ecology and Evolution showcases how artificial intelligence (AI) and long-term video monitoring can significantly improve the management of Marine Protected Areas (MPAs), enhancing the conservation of vulnerable species and supporting ecosystem resilience in the face of environmental change.
The Complexities of Monitoring Marine Ecosystems
Marine ecosystems are inherently complex and ever-changing, shaped by a mix of biological, chemical, and physical forces that fluctuate across time and space. To understand how these systems respond—especially to rising temperatures caused by global climate change—long-term observational data is essential. But traditional monitoring methods, such as manual benthic sampling and diver-based surveys, face significant limitations. They're costly, logistically demanding, and often infrequent, making it difficult for managers to detect gradual or subtle ecological changes that could have major implications for biodiversity and sustainable use.
Recent advancements in remote sensing, underwater imaging, and data analysis are beginning to address these challenges. In particular, deep learning—a branch of machine learning that uses neural networks to identify patterns—has unlocked new potential for analyzing the vast amounts of visual data collected from marine environments.
The Study
In this study, researchers deployed autonomous underwater video systems along Sweden’s coastline within a designated MPA. Over more than two decades, these systems recorded footage across various depths and habitats, resulting in a massive dataset comprising thousands of hours of video.
To make sense of this data, the team developed and trained convolutional neural networks (CNNs)—a type of deep learning algorithm—to automatically identify 17 species of benthic invertebrates. The process began with manually annotating a selection of video frames to create a training set. This allowed the AI models to learn key traits of each species, such as body shape, coloration, and movement.
Once trained, the models were applied to the entire dataset, automatically detecting and logging species in each frame. This approach drastically cut down the time and effort needed for manual identification, enabling continuous monitoring at high temporal resolution.
The occurrence data generated by the AI were then used to analyze species abundance and distribution over time and across depth gradients. The researchers used statistical models to account for potential biases and included environmental variables like temperature and sediment type to build a more comprehensive picture of ecological change.
Results and Discussion
AI-driven image analysis revealed notable and complex changes in benthic communities over the study period. Several species showed shifts in depth distribution, likely in response to warming waters. Larger, heat-sensitive species such as certain starfish and crustaceans were increasingly found in deeper, cooler waters—an apparent adaptation to thermal stress. In contrast, smaller or more heat-tolerant species became more abundant in shallower zones, highlighting temperature as a key factor shaping community structure.
The long-term, high-resolution monitoring also illuminated trends in ecosystem dynamics. Some sensitive species saw population increases within the protected areas, pointing to the success of conservation measures such as reduced trawling. In other cases, species were observed beyond their historically documented depth ranges in Swedish waters—suggesting ongoing ecological shifts tied to environmental change.
These insights reinforce the value of continuous, detailed observation in detecting changes that might otherwise go unnoticed, and in assessing the effectiveness of protection efforts.
Conclusion
This study highlights how integrating AI with remote sensing can significantly improve our ability to monitor and manage complex marine ecosystems. Automating species identification from long-term video footage provides high-frequency, fine-grained data that capture ecological responses in real time. Such data are essential for designing and managing MPAs that effectively protect vulnerable species and habitats in an era of rapid environmental change.
The authors advocate for wider adoption of AI-assisted video monitoring as a key element of future ecosystem management. They recommend expanding protections to deeper habitats and establishing new MPAs to help safeguard cold-water species from warming trends. Embracing these technology-supported strategies could help maintain biodiversity and ecosystem function for the long haul.
Journal Reference
Nilsson C. L., Faurby S., et al. (2025). Applying Deep Learning to Quantify Drivers of Long-Term Ecological Change. Ecology and Evolution, 25, p15-20. DOI: 10.1002/ece3.72091, https://onlinelibrary.wiley.com/doi/10.1002/ece3.72091