GRAPHIC APPAREL SHOPDespite covering the majority of our planet, Earth's oceans remain some of the least understood environments. Harsh conditions and logistical challenges have long limited our ability to monitor marine ecosystems and their crucial role in the global climate.
Now, artificial intelligence is changing the game. IBM Research, in partnership with leading UK institutions, has introduced an open-source AI model, Granite-Geospatial-Ocean, poised to upgrade marine science without the need for constant, costly expeditions.
Granite-Geospatial-Ocean: A New Kind of Ocean Intelligence
Granite-Geospatial-Ocean builds on proven geospatial AI technology, adapting it from land to the unique environments of the sea. Developed collaboratively with Plymouth Marine Laboratory, UK STFC Hartree Centre, and University of Exeter, this model processes massive volumes of satellite images and real-world marine data. The result is detailed, color-coded maps that reveal the health of marine ecosystems and the nuanced dynamics of the ocean’s carbon cycle.
What sets this model apart is its efficiency. With just 50 million parameters, it's significantly smaller and more accessible than previous land-focused AI models, making it suitable for broader scientific and practical use.
Illuminating Marine Productivity
Granite-Geospatial-Ocean is not just theoretical, it’s paired with practical tools built on IBM’s TerraTorch framework. These applications estimate where phytoplankton thrive and measure net primary production, the process by which these microscopic organisms convert sunlight and CO2 into organic matter.
- Phytoplankton’s importance: Like forests on land, phytoplankton drive marine life, producing about half of Earth's oxygen and absorbing vast amounts of carbon dioxide.
- Climate connections: By moving carbon from the atmosphere into the deep ocean, phytoplankton play a pivotal role in climate regulation, although the scale of this impact is still being uncovered.
- AI-powered accuracy: Granite-Geospatial-Ocean has already outperformed traditional machine-learning methods in mapping ocean productivity, doing more with less data.
From Space to Science: Leveraging Satellite Imagery
Historically, ocean productivity data came from research cruises and monitoring buoys—accurate, but geographically limited. Advances in satellite technology now allow scientists to survey vast marine areas, picking up subtle color changes that signal phytoplankton activity.
The model was pre-trained on a staggering 500,000 images from the Copernicus Sentinel-3 mission and then fine-tuned with select field data. The model’s ability to accurately map productivity and phytoplankton distribution, even with minimal ground-truth measurements, marks a major leap forward for large-scale marine monitoring.
Versatile Tools for Environmental Monitoring
The model’s design supports a variety of applications crucial to ocean and climate science:
- Detecting harmful algae blooms that threaten marine life and human health
- Monitoring sediment and nutrient runoff impacting coastal water quality
- Enhancing climate models through improved estimates of ocean carbon uptake
This combination of AI and satellite data enables dynamic, up-to-date ocean maps without the high costs and risks of traditional field campaigns. As satellite data quality improves, Granite-Geospatial-Ocean’s capabilities will only grow—deepening our understanding of the ocean’s role in environmental health and climate stability.
The Future: Data-Driven Ocean Stewardship
The Granite-Geospatial-Ocean model is a milestone for open-source, scalable ocean monitoring. By making advanced AI accessible for marine science, this model empowers researchers, policymakers, and conservationists to better track and respond to changes in the world’s oceans. As we face mounting environmental challenges, data-driven insights from AI will be essential in safeguarding ocean health for future generations.
AI Foundation Models For Exploring the Depths: Ocean Research