IBM Launches AI System That Monitors the Earth & Tracks Climate Change
TerraMind outperforms 12 leading Earth observation models by at least 8% across all metrics in the PANGEA benchmark
IBM has teamed up with the European Space Agency (ESA) to unveil TerraMind, a powerful artificial intelligence model designed to monitor climate change and global challenges like water scarcity in real time using space-based data.
TerraMind is built on TerraMesh—the largest publicly available geospatial dataset ever compiled. The open-source AI is a multimodal model capable of processing pixel, token, and sequence-based inputs simultaneously.
The model fuses insights from nine distinct Earth observation data types, offering a powerful and intuitive understanding of our planet’s systems.
TerraMind features a unique symmetric transformer-based encoder-decoder architecture capable of processing pixel-based, token-based, and sequence-based inputs while learning cross-modal correlations.
Although trained on 500 billion tokens, it remains lightweight—requiring 10x less compute than typical models per modality. This efficiency allows for large-scale deployment at reduced costs and significantly lowers energy usage during inference.
“To me, what sets TerraMind apart is its ability to go beyond simply processing earth observations with computer vision algorithms. It instead has an intuitive understanding of geospatial data and our planet. At present, TerraMind is the best performing AI foundation model for Earth observation according to well-established community-benchmarks,” Juan Bernabé-Moreno, director of IBM Research UK and Ireland, and IBM's Accelerated Discovery lead for climate and sustainability, said in a blog post.
This efficiency doesn’t come at the cost of performance. ESA-led tests show TerraMind outperforms 12 leading Earth observation models by at least 8% across all metrics in the PANGEA benchmark, covering tasks like land cover classification, environmental monitoring, and multi-temporal analysis.
TerraMind’s edge lies in its ability to integrate data across nine modalities, such as satellite imagery, vegetation, climate, and topography—allowing it to offer unified insights on complex phenomena like water scarcity, which traditionally required multiple AI systems.
TerraMind is now available on Hugging Face and IBM’s Geospatial Studio. Fine-tuned versions tailored for disaster response and other high-impact uses are expected to roll out in the coming weeks.
