Peering Into the Moon’s Shadows: AI Provides Sharper Images of Lunar Craters That Contain Water Ice

Craters and Depressions Moon South Pole

The 17 newly studied craters and depressions are located near the South Pole. While the smallest of these regions (region 11) has a size of only 0.18 square kilometers, the largest (region 9) measures 54 square kilometers. Region 9 is not located in the section of the south polar region shown here, but a bit further to the North, in Schrödinger Basin. The representations of the lunar surface shown here are based on altimeter data from the Lunar Reconnaissance Orbiter. Credit: © MPS/University of Oxford/NASA Ames Research Center/FDL/SETI Institute

Permanently shadowed lunar craters contain water ice, but are difficult to image. A machine learning algorithm now provides sharper images.

The Moon’s polar regions are home to craters and other depressions that never receive sunlight. Today, a group of researchers led by the Max Planck Institute for Solar System Research (MPS) in Germany present the highest-resolution images to date covering 17 such craters. Craters of this type could contain frozen water, making them attractive targets for future lunar missions, and the researchers focused further on relatively small and accessible craters surrounded by gentle slopes. In fact, three of the craters have turned out to lie within the just-announced mission area of

Unnamed Crater South Polar Region of Moon

An as-yet unnamed crater in the south polar region of the Moon. It is located on the Leibnitz plateau, in direct proximity to the targeted mission area of NASA’s Volatiles Investigating Polar Exploration Rover (VIPER). The left image shows a picture taken by the Lunar Reconnaissance Orbiter. The interior of the crater is almost not visible. The right image shows the same image after it was processed with the HORUS machine learning algorithm. Credit: © Left: NASA/LROC/GSFC/ASU; Right: MPS/University of Oxford/NASA Ames Research Center/FDL/SETI Institute

To address this problem, the researchers have developed a machine learning algorithm called HORUS (Hyper-effective nOise Removal U-net Software) that “cleans up” such noisy images. It uses more than 70,000 LRO calibration images taken on the dark side of the Moon as well as information about camera temperature and the spacecraft’s trajectory to distinguish which structures in the image are artifacts and which are real. This way, the researchers can achieve a resolution of about 1-2 meters per pixel, which is five to ten times higher than the resolution of all previously available images.

Using this method, the researchers have now re-evaluated images of 17 shadowed regions from the lunar south pole region which measure between 0.18 and 54 square kilometers in size. In the resulting images, small geological structures only a few meters across can be discerned much more clearly than before. These structures include boulders or very small craters, which can be found everywhere on the lunar surface. Since the Moon has no atmosphere, very small meteorites repeatedly fall onto its surface and create such mini-craters.

“With the help of the new HORUS images, it is now possible to understand the geology of lunar shadowed regions much better than before,” explains Moseley. For example, the number and shape of the small craters provide information about the age and composition of the surface. It also makes it easier to identify potential obstacles and hazards for rovers or astronauts. In one of the studied craters, located on the Leibnitz Plateau, the researchers discovered a strikingly bright mini-crater. “Its comparatively bright color may indicate that this crater is relatively young,” says Bickel. Because such a fresh scar provides fairly unhindered insight into deeper layers, this site could be an interesting target for future missions, the researchers suggest. 

The new images do not provide evidence of frozen water on the surface, such as bright patches. “Some of the regions we’ve targeted might be slightly too warm,” Bickel speculates. It is likely that lunar water does not exist as a clearly visible deposit on the surface at all – instead, it could be intermixed with the regolith and dust, or may be hidden underground.

To address this and other questions, the researchers’ next step is to use HORUS to study as many shadowed regions as possible. “In the current publication, we wanted to show what our algorithm can do. Now we want to apply it as comprehensively as possible,” says Bickel.

Reference: “Peering into lunar permanently shadowed regions with deep learning” by V. T. Bickel, B. Moseley, I. Lopez-Francos and M. Shirley, 23 September 2021, Nature Communications.

DOI: 10.1038/s41467-021-25882-z