AI
NASA Opened 17 Years of Moon Data to AI, and the Ice Maps Are the Point
NASA and IBM just released one of the first open source AI models built specifically for lunar science. The Lunar Foundation Model is trained on 17 years of data from the Lunar Reconnaissance Orbiter, hosted publicly on Hugging Face, with the complete codebase on GitHub for testing and experimentation. Anyone can use it, fine tune it, and build on it.
The training set is staggering: roughly 2 million image tiles, more than 1 million high resolution camera images at 1 meter resolution, nearly 964,000 multispectral images at 100 meter resolution, plus terrain data from GRAIL, Lunar Prospector, and Japan's SELENE mission. The LRO data alone is larger than all other NASA planetary missions combined, an almost seamless high resolution mosaic of the entire Moon.
What the model does: map craters at scale, spot young volcanic features called irregular mare patches, and estimate where ice stays stable near the lunar poles. Dark permanently shadowed regions preserve ice for up to billions of years, and that ice is the most valuable resource in cislunar space: drinking water, air, and rocket fuel waiting in the cold.
Kevin Murphy, NASA's chief science data officer, framed the mission plainly: collecting data is only part of the job, and AI turns petabytes of records into new discoveries. The team also released machine learning ready datasets and benchmark collections through the open TerraTorch toolkit, with a companion paper on Hugging Face, so scientists worldwide can compare and refine models on the same foundation.
This joins NASA's growing AI for science portfolio with IBM, alongside the Prithvi models for Earth observation and the Surya model for space weather prediction. The lunar model matched or beat strong baselines on crater mapping and volcanic feature detection, and showed a clear advantage in estimating polar ice stability.
The open release strategy is the quiet revolution here. Alongside the model, the team published machine learning ready pre training datasets and benchmark collections, integrated into the open source TerraTorch toolkit, with a companion paper on Hugging Face. That combination lets any lab reproduce the results, challenge them, and build better versions, which is how science compounds. The Moon's data just became a commons.
For you, the Moon just got an open source copilot. The next generation of lunar missions will navigate by maps AI helped draw, and every crater count and ice estimate feeds directly into where humans land next. Seventeen years of moonlight, now readable by machines and free for everyone.
Quick answers
What is this story about?
NASA and IBM just released one of the first open source AI models built specifically for lunar science. The Lunar Foundation Model is trained on 17 years of data from the Lunar Reconnaissance Orbiter, hosted publicly on Hugging Face, with the complete codebase on GitHub for testing and experimentation. Anyone can use it, fine tune it, and build on it.
Why does this story matter?
For you, the Moon just got an open source copilot. The next generation of lunar missions will navigate by maps AI helped draw, and every crater count and ice estimate feeds directly into where humans land next. Seventeen years of moonlight, now readable by machines and free for everyone.
Sources
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.