Co-leads
- Robert Shorten (Imperial College, UK)
- Geeth de Mel (IBM Research, UK)
Projects
GEO-Bench
A transformative benchmarking suite for Earth Observation foundation models. GEO-Bench tackles the lack of reproducibility and systematic evaluation of geospatial foundation models by enhancing the initial version (GEO-Bench V1) with an expanded set of tasks, user-friendly features, and robust community engagement. It also includes a dedicated track for vision-language models (GEO-Bench-VLM), supported through a collaboration with NASA and ESA. With a leaderboard on Hugging Face and integration with IBM TerraTorch, GEO-Bench sets the standard for benchmarking geospatial foundation models. Planned release for 1H 2025.
Alliance Partners & Steering Committee: ServiceNow, IBM Research, MBZUAI, NASA, ESA, TU Munich, Arizona State University, Clark University
EarthDial
A groundbreaking vision-language model (VLM) designed to transform how we interpret Earth observation data. With capabilities extending from text to multiresolution, multispectral, and time-series datasets, EarthDial bridges complex geospatial analytics and accessibility. EarthDial generates both textual and pixel-level insights using a compounded strategy for domain-specific alignment, enabling non-experts to interactively explore geospatial data. EarthDial already surpasses state-of-the-art VLM benchmarks for a range of geospatial AI tasks and is currently being enhanced with Granite 3.0 integration. Planned release for 1H 2025.
Alliance Partners: IBM Research, MBZUAI
Agenda
Join our network to co-develop innovative AI solutions for climate and sustainability.
We invite volunteers to help advance AI for through:
- Geospatial data fusion and model validation.
- Creating sustainable AI tools for Earth observation.
- Organizing community-driven events and hackathons.
Other resources
IBM-NASA Geospatial Foundation Models: IBM and NASA have teamed up to build a family of geospatial foundation models named Prithvi. You can access these via Hugging Face.
The IBM Granite Hugging Face page also provide enterprise-ready fine-tuned versions of the Prithvi models, including models for biomass estimation, urban heat island detection, predicting the canopy height, and downscaling climate projections.
Frequently Asked Questions (FAQ)