New open-source system can identify ice deposits, craters and volcanic terrain with greater accuracy, potentially helping shape the next generation of human exploration beyond Earth

NASA and IBM have unveiled a new artificial intelligence model capable of analysing decades of lunar data, giving scientists a more powerful way to identify water ice, map craters and examine geological features as space agencies prepare for a renewed era of human exploration on the Moon.

Released on September 10, the NASA-IBM Lunar Foundation Model is an open-source AI system trained on more than 30 layers of scientific data collected by nine instruments aboard four NASA missions, including the Lunar Reconnaissance Orbiter.

The technology represents a significant expansion of the growing use of foundation models in scientific research.

Unlike conventional artificial-intelligence systems built to perform a single narrowly defined task, foundation models are trained on large and varied datasets and can subsequently be adapted to multiple applications. NASA and IBM have previously used similar technology for Earth observation, weather forecasting and geospatial analysis.

The new system applies that approach to the lunar surface.

Its developers say the model can help researchers locate areas that may contain frozen water, identify and classify craters, study volcanic formations and analyse vast quantities of imagery and scientific measurements that would otherwise require substantial human effort.

In benchmark testing, NASA and IBM said the model identified important lunar features with accuracy improvements of as much as 23% compared with widely used existing techniques.

That performance could have consequences extending far beyond academic planetary science.

The Search for Water

Among the model’s most important applications is the search for lunar ice.

Scientists have long been interested in permanently shadowed regions near the Moon’s poles, where extremely low temperatures may have preserved water ice for billions of years.

For future astronauts, such deposits could become one of the Moon’s most strategically valuable resources.

Water brought from Earth is enormously expensive to transport into space. If sufficiently large and accessible deposits can instead be extracted from the lunar surface, they could potentially provide drinking water and oxygen for astronauts.

Water can also be separated into hydrogen and oxygen, both of which can be used as components of rocket propellant.

That raises the possibility that the Moon could eventually function not simply as a destination, but as a staging point for deeper exploration of the Solar System.

Identifying where usable ice exists — and determining how concentrated or accessible it is — has therefore become a major objective for lunar science.

NASA and IBM say their AI system could accelerate that process by identifying patterns across large quantities of observational data that would be difficult for scientists to examine manually.

Finding Safer Places to Land

Craters present another major challenge.

The lunar surface is covered with impact features ranging from enormous basins to relatively small depressions that may still pose dangers to spacecraft.

Precise terrain maps are therefore essential for future robotic and crewed missions.

An AI model capable of identifying craters and analysing surrounding terrain could help mission planners evaluate potential landing zones more efficiently.

This becomes especially important as lunar exploration moves toward regions near the south pole, where lighting conditions, steep terrain and extensive shadowing make navigation considerably more complicated than in many of the areas visited during the Apollo era.

NASA is placing increasing emphasis on the lunar south pole as part of its plans for long-duration exploration and infrastructure development.

Earlier this week the agency sought proposals from industry for technologies including lunar power generation, oxygen extraction from regolith, advanced manufacturing and systems intended to support a future Moon base in the south polar region.

The new AI model fits directly into that strategy.

Before a permanent or semi-permanent human presence can be established, mission planners need increasingly detailed knowledge of the terrain, resources and environmental conditions astronauts will encounter.

Artificial intelligence could become one of the tools used to build that knowledge.

Turning Decades of Data Into Usable Intelligence

NASA has accumulated an extraordinary archive of lunar observations.

The challenge is no longer simply collecting information.

It is extracting useful knowledge from the enormous volume of information already available.

Modern spacecraft carry cameras, spectrometers, radar systems and other instruments capable of generating enormous datasets. As the number of missions increases, scientists face a growing problem: there is simply too much information for researchers to analyse manually at the same speed at which it is being collected.

Foundation models offer one possible solution.

Instead of examining each image or dataset independently, AI systems can search for relationships across multiple forms of information simultaneously.

That can reveal geological structures or patterns that may not be immediately obvious when individual datasets are studied separately.

NASA and IBM have already demonstrated the same principle closer to Earth.

Their Prithvi family of open foundation models has been developed for applications including environmental monitoring, flood analysis, land-use assessment and other forms of geospatial research. One version of the Prithvi model was successfully deployed aboard orbital platforms earlier in 2026, demonstrating that sophisticated geospatial AI can operate directly in space.

The lunar system represents a natural extension of that work.

Instead of teaching AI to interpret forests, coastlines and cities, researchers are training it to understand craters, lava formations and permanently shadowed landscapes hundreds of thousands of kilometres from Earth.

Why Open Source Matters

The decision to make the model publicly available may prove almost as important as its technical performance.

Open-source scientific AI allows universities, research institutes and other organisations to experiment with the technology without having to develop equivalent models from scratch.

Researchers can potentially adapt the system for specialised questions, combine it with new datasets or use it to test competing theories about lunar geology.

That approach also spreads advanced computational capability beyond the relatively small number of organisations able to afford the enormous computing resources required to train large foundation models independently.

NASA and IBM have deliberately followed this strategy with earlier Prithvi models.

NASA has described the programme as a way of giving scientists reusable AI systems that can be adapted to environmental and planetary problems rather than repeatedly constructing entirely new models.

The philosophy mirrors a broader shift taking place across science.

Artificial intelligence is increasingly moving from being primarily a consumer technology into becoming part of the infrastructure of research itself.

AI models are now being used to predict protein structures, analyse genetic mutations, study climate systems, search astronomical observations and simulate complex physical processes.

Lunar science is becoming another frontier.

AI Becomes a Scientific Instrument

The emergence of specialised scientific foundation models could ultimately change the relationship between researchers and data.

Traditional scientific instruments observe the physical world.

Telescopes collect light. Spectrometers measure chemical signatures. Radar probes structures beneath surfaces.

AI increasingly acts as an additional analytical layer sitting above those instruments.

It does not replace the observations themselves. Instead, it helps scientists determine which patterns inside enormous datasets deserve closer investigation.

In that sense, AI is beginning to function almost like a new category of scientific instrument.

Its purpose is not merely to calculate faster, but to recognise relationships across datasets that have become too large and complicated for humans to examine exhaustively.

The potential is particularly significant in planetary science.

Future lunar missions will generate increasingly detailed information about surface composition, temperature, radiation, mineral deposits and underground structures.

Eventually similar models could analyse Mars, asteroids or the icy moons of the outer Solar System.

The same underlying technology could potentially be retrained repeatedly as humanity moves farther from Earth.

The Moon as a Testing Ground for Mars

NASA’s renewed lunar programme is ultimately about more than the Moon.

The agency sees sustained lunar exploration as a testing ground for technologies and operational methods that could eventually support human missions to Mars.

NASA’s Artemis programme currently plans to return astronauts to the lunar surface in 2028, with later missions intended to test technologies required for longer stays away from Earth.

The Moon offers an ideal intermediate environment.

It is distant enough to expose astronauts to many of the challenges of deep-space exploration, yet close enough that emergency communication and resupply remain far more manageable than they would be on Mars.

Technologies developed there could include autonomous construction, surface power generation, extraction of local resources and increasingly sophisticated robotic systems.

Artificial intelligence could become critical to all of them.

Future astronauts may operate in environments where communications with Earth are delayed, disrupted or unavailable.

Robotic vehicles and scientific systems will therefore need to perform more analysis independently.

AI capable of recognising terrain, identifying resources or interpreting scientific observations locally could dramatically reduce dependence on mission controllers thousands or millions of kilometres away.

A New Race for Lunar Intelligence

The technological development also arrives amid accelerating international competition around the Moon.

The United States, China, Europe, Japan, India and an expanding group of private companies are investing in lunar missions and technologies.

What was once primarily a scientific competition is increasingly becoming economic and strategic as governments examine the potential value of lunar resources, infrastructure and permanent operations.

In that environment, information itself becomes a strategic asset.

The nations and organisations able to build the most detailed understanding of lunar terrain and resources may gain significant advantages in selecting landing sites, planning infrastructure and designing exploration missions.

AI could dramatically accelerate that process.

Instead of waiting years for scientists to manually analyse enormous planetary datasets, automated systems may eventually process new observations almost as soon as spacecraft collect them.

That could transform mission planning.

A newly identified ice deposit, safe landing zone or geological formation could potentially move from raw observation to operational decision much faster than before.

From Artificial Intelligence to Exploration Intelligence

The NASA-IBM Lunar Foundation Model is still fundamentally a research tool.

Its performance will need to be evaluated across different datasets and scientific applications, and AI-generated results will still require validation by planetary scientists.

Foundation models can detect patterns, but they can also produce errors or misleading interpretations when confronted with unfamiliar information.

Scientific oversight therefore remains essential.

But the direction of travel is clear.

As humanity prepares to return to the Moon and potentially establish a sustained presence there, exploration will increasingly depend not simply on more powerful rockets or better spacecraft, but on the ability to understand enormous amounts of information.

Artificial intelligence may become one of the technologies that makes that possible.

The first great lunar explorers relied on maps created from telescopes and reconnaissance spacecraft.

The next generation may rely on maps interpreted by machines capable of examining almost every crater, shadow and geological structure simultaneously.

If the technology works as intended, AI will not merely help scientists study the Moon.

It could help humanity decide where to go once it gets there.