NASA-IBM Lunar Foundation Model Goes Open Source with Breakthrough Performance

NASA and IBM have released the NASA-IBM Lunar Foundation Model as open source, marking one of the first publicly available foundation models for scientific study of the Moon.

The model is built on a version of TerraMind, an Earth-observation model developed by IBM with the European Space Agency, which handles mixed data types and resolutions. It uses low-rank adapters to fine-tune for each lunar task while keeping 90% of the base weights frozen, making adaptation cost low.

The training corpus is a unified, spatially aligned dataset of more than 2 million image tiles from nine instruments across four missions, including imagery from NASA's Lunar Reconnaissance Orbiter and gravity-field maps from the GRAIL mission. This dataset has never existed before, according to NASA.

In benchmarking against SwinV2-B, the model cut root-mean-square error in flagging likely ice deposits by up to 22%, while matching or outperforming it on other tasks such as crater detection and volcanic-feature mapping. The model's performance is seen as a major breakthrough for lunar research.

The open-source release of the NASA-IBM Lunar Foundation Model allows scientists to explore the Moon at scale, connecting observations across instruments and revealing patterns that are difficult to see in isolation.