IBM and NASA Release Open-Source AI Model for Lunar Mapping

IBM and NASA have released an open-source artificial intelligence model built to map the moon's surface, bringing machine learning into planetary science at a practical scale. The system, called the NASA-IBM Lunar Foundation Model, identifies craters, volcanic formations, and potential ice deposits by analyzing decades of archived observation data IBM. According to IBM, it outperforms widely used methods by up to 23% in detecting these geographic features.

The model launched on Thursday and is now available on Hugging Face CNET. It was trained on lunar observation data spanning multiple decades, curated jointly by IBM and NASA researchers. Campbell Watson, a senior research manager at IBM Research, said the pair aim to give the global scientific community a shared foundation that can be adapted to new questions about the moon.

The unified dataset

Alongside the model, the teams published what they describe as the first open-source lunar dataset of its kind, combining tens of thousands of maps and images from nine instruments across four space missions IBM. The breadth of that compilation matters for how scientists work. Previously, researchers had to manually examine maps and images collected by different instruments over many decades, or rely on low-resolution machine learning models designed for a single task CNET.

A foundation model approach changes that by learning cross-instrument patterns simultaneously, letting one system handle diverse analytical jobs. The unified dataset effectively serves as an artificial intelligence index of lunar observations, consolidating fragmented archives into a resource that individual research teams could not easily reconstruct on their own.

Historical context and limitations

IBM and NASA have collaborated for over 60 years, tracing back to the Apollo era CNET. The Lunar Foundation Model builds on that long partnership and on decades of publicly released lunar data, continuing a tradition of broad scientific access. Its deployment under NASA's Artemis program signals that foundation models are becoming practical tools for space exploration rather than laboratory experiments.

Practitioners should note, however, that the 23% improvement figure comes from IBM's own benchmarks, and independent verification has not yet been published. The model's performance on edge cases, such as permanently shadowed craters where ice detection is hardest, remains unspecified in the current documentation. Researchers probing those limitations should consult the dataset directly on Hugging Face.

What comes next

Whether the Lunar Foundation Model becomes a standard reference for lunar science will depend on how the broader community adopts and benchmarks it. Accurate surface mapping is essential for the Artemis program's goals of identifying safe landing sites and resource deposits. If independent teams confirm the reported gains, this release could set a template for how space agencies apply artificial intelligence to decades of archival mission data.

FAQ

What is the NASA-IBM Lunar Foundation Model?
It is an open-source artificial intelligence model released by IBM and NASA to analyze lunar observation data and identify geographic features including craters, volcanic formations, and potential ice deposits.

How accurate is the model?
According to IBM, it outperforms existing methods by up to 23% in detecting key lunar surface characteristics. Independent benchmarks have not yet been published.

Where can I access the model and dataset?
Both are available on Hugging Face, alongside the unified dataset combining data from nine instruments across four space missions.

Why does this matter for lunar exploration?
The model and dataset provide a shared, open foundation for lunar analysis, replacing manual workflows and single-task models under NASA's Artemis program.