MIT uses AI to document warming Gulf of Maine ecosystems

MIT researchers have launched LOBSTgER, an innovative project that combines artificial intelligence with underwater photography to document marine life in the rapidly warming Gulf of Maine. The initiative uses generative AI models trained exclusively on curated underwater photographs to create scientifically accurate imagery that could revolutionize how environmental stories are told and help raise awareness about vulnerable ocean ecosystems.

What you should know: LOBSTgER stands for Learning Oceanic Bioecological Systems Through Generative Representations and represents a collaboration between underwater photographer Keith Ellenbogen and MIT mechanical engineering PhD student Andreas Mentzelopoulos.

• The project focuses on the Gulf of Maine, one of the world’s most biologically diverse marine ecosystems that is warming faster than 99 percent of the world’s oceans.
• Custom diffusion models are trained exclusively on Ellenbogen’s original underwater photographs, ensuring both visual integrity and ecological accuracy.
• The AI models can generate new synthetic images unconditionally or enhance real photographs by recovering detail in turbid water and adjusting lighting.

How it works: The generative models learn from thousands of real underwater images to internalize fine-grained details like natural lighting gradients, species-specific coloration, and atmospheric textures created by suspended particles and refracted sunlight.

• Ellenbogen’s comprehensive dataset comes from his “Space to Sea: Visualizing New England’s Ocean Wilderness” project, capturing diverse angles, lighting conditions, and animal behaviors.
• Mentzelopoulos developed original code to train the latent diffusion models from scratch, requiring hundreds of hours of computation and meticulous hyperparameter tuning.
• The models can simulate scenes that would be nearly impossible to capture in the field, helping construct more complete visual narratives of underwater life.

In plain English: Think of it like teaching an AI to paint underwater scenes by showing it thousands of real ocean photos first. The AI learns the subtle details—how light filters through water, what different fish look like, how particles float in the current—then uses that knowledge to create new, scientifically accurate images or improve existing blurry photos.

The big picture: LOBSTgER operates at the intersection of art, science, and technology, challenging traditional boundaries between photography and AI-generated imagery.

• The project treats photography and generative AI as complementary rather than competing technologies, with photography capturing “what is” and AI extending vision toward “what could be understood, inferred, or imagined.”
• Underwater photography in New England’s coastal waters faces constant challenges from limited visibility, swirling sediment, bubbles, and unpredictable marine life movement.

What they’re saying: The researchers emphasize that AI should enhance rather than replace traditional documentation methods.

• “Getting a high-quality dataset is not easy,” Ellenbogen says. “It requires multiple dives, missed opportunities, and unpredictable conditions. But these challenges are part of what makes underwater documentation both difficult and rewarding.”
• “The goal isn’t to replace photography,” Mentzelopoulos explains. “It’s to build on and complement it — making the invisible visible, and helping people see environmental complexity in a way that resonates both emotionally and intellectually.”

Why this matters: The project addresses the urgent need to document and raise awareness about marine ecosystems undergoing rapid environmental change.

• The Gulf of Maine’s accelerated warming threatens hundreds of species including whales, sharks, jellyfish, herring, and plankton.
• In regions where ecosystems are changing rapidly, visualization becomes more than documentation—it becomes a tool for awareness, engagement, and conservation.
• The approach could benefit other underwater photographers and expand to marine ecosystems worldwide.

Key details: The project has already captured high-resolution images of lion’s mane jellyfish, blue sharks, American lobsters, and ocean sunfish while free diving in coastal waters.

• LOBSTgER’s models are built using custom code to protect the process and outputs from potential biases from external data or models.
• The team’s long-term goal is developing a comprehensive model that can visualize a wide range of Gulf of Maine species and eventually apply similar methods globally.