In This Article:
The European Space Agency (ESA) and IBM Research Europe have unveiled TerraMind, a next-generation AI model to help transform Earth observation.
This self-supervised learning tool is designed to process vast data sets, providing precise insights into climate and environmental issues.
TerraMind interprets Earth observation images with a comprehensive understanding of geospatial context, unlike other AI models that may confuse similar-looking objects, the ESA claims.
The multimodal capabilities of TerraMind allow it to analyse various data types, including topography and satellite imagery, to deliver accurate environmental assessments.
TerraMind has been developed using an extensive training data set that consists of more than nine million samples from around the world, spanning eight varied data types.
This data set includes radar information from the Copernicus Sentinel-1 satellite and detailed optical imagery from Sentinel-2.
Furthermore, TerraMind can create artificial data to fill gaps when certain inputs are unavailable, a frequent challenge in the field of remote sensing.
Employing a method known as Thinking-in-Modalities, the AI model can engage in a logical sequence of problem-solving steps, simultaneously producing new data to aid in its analysis.
TerraMind also emerged as the “best-performing” AI foundation model for Earth observation tasks during benchmark testing.
Additionally, the AI model requires up to ten-times less power compared with the use of individual models for different data types.
IBM Research (UK and Ireland) director and IBM's Accelerated Discovery lead for climate and sustainability Juan Bernabé-Moreno said: “Developed by IBM Research and ESA, TerraMind is an AI model trained to understand our planet through satellite data, landforms and other key surface features.
“Rather than just analysing images, it builds a deeper, more intuitive understanding of Earth’s systems.”
Trained using the infrastructure and expertise of the Jülich Supercomputing Centre in Germany, the TerraMind model has been developed within the Fostering Advancements in Foundation Models via Unsupervised and Self-supervised Learning for Downstream Tasks in Earth Observation (FAST-EO).
The FAST-EO initiative is led by a consortium comprising Forschungszentrum Jülich, DLR, IBM Research Europe and KP Labs, with ESA Φ-lab's support and funding.
ESA Φ-lab earth observation data scientist and TerraMind technical officer Nicolas Longépé said: “TerraMind can understand satellite images in a smart, powerful way, making it great for tracking large-scale events on Earth. It is trained on lots of data and built with novel technology that helps it to see the big picture.”