About This Project

Lava tubes are underground conduits formed by previous or currently flowing lava. Several have been explored on Earth, but space missions are revealing that they may also be present on the Moon and Mars. Lava tubes could be used as habitats for astronauts and targeted for astrobiological research. We aim to use machine learning to find candidate lava tubes on the Moon (and possibly Mars) by training a model to detect where their roofs have collapsed in satellite imagery.

Ask the Scientists

What is the context of this research?

Lava tubes are sub-surface conduits formed by the flowing of lava, which can either still have lava present, or be completely evacuated. The latter leaves an entire underground cave system which can be explored by entering through collapses in the tube's roof (called pits). Lava tubes have been explored extensively on Earth in regions with high volcanic activity. Without the ability to easily explore these features on the Moon and Mars, the presence of lava tubes have up to now been predicted by looking at space satellite data containing sinuous chains of pits. Only around 280 and 1,000 pits have been catalogued on the Moon and Mars, respectively thus far. Therefore, automated techniques for detecting sinuous pit chains will be critical for searching for potential lava tubes.

What is the significance of this project?

Underground cavities are likely to play a huge role in future space exploration. On bodies without thick atmospheres (such as on the Moon and Mars) the surface will be exposed to harmful solar and cosmic radiation, limiting how long humans can spend on the surface without sufficient protective shielding. Evacuated lava tubes would provide natural shelter from this radiation, as well as micro-meteorite impacts, without the need for building and transporting shielding. Extra-terrestrial caves are also of astrobiological significance for their propensity to harbour water ice, particularly on Mars. A bolstered catalogue of possible Lunar lava tubes would aid planetary geologists and space agencies in knowing which would be the best to further study – and eventually explore.

What are the goals of the project?

We will train at least two machine learning models to detect Lunar pits. The first model will be trained on Lunar Reconnaissance Orbiter (LRO) imagery containing Lunar pits. The second will be trained with a combination of this Moon data, but also with Mars Reconnaissance Orbiter imagery containing relevant Martian pits to determine if accuracies improve. These models will then be used to detect new pits in the Lunar poles using the LRO controlled polar mosaics. Using the newly detected collapses, we will infer where there are possible intact lava tubes – and then estimate the surface areas, depths and volumes of the pits and lava tubes, respectively.