Artificial intelligence is becoming an increasingly useful tool for scientific research, and the Moon is now becoming part of that trend.
NASA and IBM Research have developed an open-source Lunar Foundation Model, an artificial intelligence system designed to help researchers study the Moon's surface and geological features. The model was trained using around 17 years of lunar observations collected by NASA's Lunar Reconnaissance Orbiter and data from several other space missions.
Instead of relying on a single type of image, the system combines information from different instruments. This gives researchers a broader view of the lunar surface and could make it easier to identify important features such as craters and possible deposits of water ice.
A Huge Collection of Lunar Data
Training an AI system to understand the Moon requires an enormous amount of data.
For this project, researchers created a dataset called SomBench. It contains nearly two million groups of lunar observations covering multiple types of scientific information.
A significant portion of the data comes from NASA's Lunar Reconnaissance Orbiter, which has been observing the Moon for many years. The researchers also incorporated information from missions including GRAIL, Lunar Prospector and Japan's Kaguya spacecraft.
The combined dataset contains more than 30 aligned data layers collected from nine instruments and four different missions.
Some of the images provide extremely detailed views of the lunar surface, while others cover much larger areas. By bringing these different perspectives together, the AI can learn about both small surface features and larger geological patterns.
Why Studying the Moon With AI Matters
Scientists already have access to an enormous amount of lunar data, but analyzing all of it manually can be difficult and time-consuming.
An AI foundation model can potentially help researchers examine large datasets more efficiently. Once trained, the model can be adapted for different scientific tasks instead of building a completely new AI system for every individual problem.
NASA and IBM tested the model on several applications, including crater detection, identifying potential ice deposits and analyzing unusual geological formations known as Irregular Mare Patches.
One of the most interesting results involved possible water-ice deposits near the Moon's poles.
AI Shows Promise in Finding Lunar Ice
Water ice on the Moon is particularly important because permanently shadowed areas near the lunar poles can remain extremely cold.
If future missions confirm and successfully use these resources, lunar water could have several potential applications. Water could potentially be used for drinking and life-support systems, while its components could also be useful in producing oxygen and rocket propellant.
According to IBM's reported testing, the Lunar Foundation Model reduced prediction error for potential ice deposits by as much as 22 percent compared with one of the baseline systems used in the evaluation.
The model also produced strong results when detecting craters, including situations where researchers used less training data.
These results don't mean the AI has discovered previously unknown lunar resources by itself. Rather, the system can help scientists analyze existing observations and identify areas that may deserve closer investigation.
Teaching AI About Lunar Lighting
The Moon presents a unique challenge for computer vision.
The way the lunar surface appears in an image can change dramatically depending on where the Sun is positioned. Shadows and differences in illumination can make the same type of terrain look very different.
To deal with this problem, researchers provided the model with information about the imaging conditions. This includes details such as the position of the Sun and the illumination angles associated with each observation.
Giving the AI this additional information helps it understand that some differences in an image may be caused by lighting rather than actual changes in the Moon's surface.
The model can also work with imagery captured at different levels of detail, allowing it to consider both small features and wider geological surroundings.
The Model Has Limitations
Despite its promising results, the Lunar Foundation Model is not designed to replace scientific instruments or direct measurements.
Researchers found that the system should not be treated as a tool for determining precise geographic coordinates or absolute elevation measurements.
In other words, the AI can be useful for recognizing patterns and supporting analysis, but scientists still need traditional instruments and measurements to establish accurate physical properties of the lunar surface.
The researchers also note that additional testing is necessary. Some of the evaluation datasets are relatively small, and more experiments are needed to determine exactly which parts of the system are responsible for its improvements.
An Open-Source Tool for Researchers
One of the notable aspects of the project is that the Lunar Foundation Model has been released as an open-source resource.
Researchers can access the model, associated datasets and code to explore new applications. It has also been incorporated into the open-source TerraTorch framework.
This could make the technology useful beyond the original experiments. Scientists may be able to adapt the model for other lunar research questions as more data becomes available.
AI Could Change How We Explore the Moon
The development of the Lunar Foundation Model is part of a larger movement toward using artificial intelligence in scientific research.
Space missions generate enormous amounts of information, and future missions are expected to produce even more. AI systems could help researchers organize this information, identify patterns and prioritize areas that need closer examination.
For lunar science, combining observations from different missions could be particularly valuable. Each spacecraft and instrument provides a different perspective, and bringing those observations together can provide a more complete picture of the Moon.
The NASA-IBM project does not mean AI is replacing planetary scientists. Instead, it provides researchers with another tool for working with the growing volume of information collected from the lunar surface.
As humanity prepares for more lunar missions, technologies like this could help scientists make better use of the data already collected and prepare for the discoveries that future spacecraft may bring.
0 Comments