IBM Research Europe has partnered with the European Space Agency to develop TerraMind, a new Earth observation model accessible on IBM’s open-source library Hugging Face. Through pretraining on TerraMesh geospatial data, the artificial intelligence-based model combines insights from nine classes of Earth observation data, providing an intuitive knowledge of the planet Earth, IBM said Tuesday.
In a separate statement, ESA said TerraMind can accurately answer questions about climate and nature and has practical applications, such as detecting methane gas leaks or tracking forest and land use changes.
Earth Observation
“At present, TerraMind is the best-performing AI foundation model for Earth observation according to well-established community benchmarks,” said Juan Bernabé-Moreno, director of IBM Research UK and Ireland.
Simonetta Cheli, director of ESA Earth observation programs, explained that TerraMind’s output accuracy is thanks to several modalities of training data that the model combines. “The ability to intuitively bring in contextual information and generate unseen scenarios is a critical step in unlocking the value of ESA data. Compared to competitive models, [TerraMind] can uncover a deeper understanding of the Earth for researchers and businesses alike,” she stressed.
International Partnership
Through international collaboration in Earth observation science and technology, TerraMind is fully harnessing space-based data for the protection of the planet, said Nicolas Longepe, Earth observation data scientist at ESA. “This project is a perfect example where the scientific community, big tech companies and experts have collaborated to leverage this technology for the benefit of Earth sciences,” he added.
The Jülich Supercomputing Center and the German Space Agency are the other contributors to the project.
Although with an architecture trained across 500 billion tokens, TerraMind is a small, lightweight model running at 10 times less compute than standard models, enabling its deployment at lower costs and less overall energy consumption.