AI could give planetary archaeology a much larger search area

Roads, terraces and settlements leave patterns that can remain visible from orbit long after their builders are gone. Ana Henrique, Sofia Sheikh and Benjamin Fields propose applying that archaeological perspective beyond Earth: use machine learning to search spacecraft imagery for possible traces of technological activity on rocky bodies. Their October paper lays out a search programme using existing archives; it reports no extraterrestrial discovery.[1]

The attraction is the longevity of a physical trace. Surface changes can outlast the society that made them and need no continuing radio transmission to be detectable. On geologically inactive bodies, the authors argue that some traces could persist for hundreds of millions of years. Human spacecraft have already left recognisable marks on bodies that were never habitable, illustrating why a surface search need not be restricted to places where life could have originated.[1]

Teach the search what to look for—and let it find surprises

The first task is image selection. Software could flag an unusual spatial pattern and direct an investigator to the relevant region, then follow-up observations could test its origin. Faults, impacts, lighting and camera artefacts can all create striking geometry. The useful output is therefore a ranked set of features to investigate, with geological explanations examined alongside artificial ones.[1]

The authors advocate two complementary approaches. A supervised model can learn from labelled examples of human structures and landscape engineering. An unsupervised or self-supervised detector can identify images that differ from the surrounding terrain without needing to know what an unfamiliar technology should look like. The first approach offers a defined target; the second widens the search beyond recognisably human forms.[1]

There is already a concrete precedent in the work cited by the proposal: a self-supervised search of Lunar Reconnaissance Orbiter imagery returned the known Apollo 17 lander as well as recent impacts and volcanic pits. That mix is useful. A pipeline that can recover known human artefacts and identify unusual geology has testable performance before it is asked to evaluate an extraordinary candidate.[1]

Build evidence across observations

The paper proposes using optical images together with complementary observations, including reflectance spectroscopy, radar and LiDAR where available. These methods probe different properties: shape, material composition and possible buried structure. A feature seen repeatedly under different conditions, with supporting information about its material and context, offers more to investigate than a single suggestive outline.[1]

A well-defined survey could also produce a quantitative upper limit. Researchers would specify the area searched, image resolution and kinds of features the method can recover. A search finding no qualifying candidates could then constrain how common detectable examples are within that scope. Features smaller than the resolution, deeply buried or unlike the search target would remain outside the conclusion.[1]

Archived imagery makes the programme relatively economical: the spacecraft have already collected the data. The work lies in organising it, validating detection methods and examining the candidates. Links between terrestrial archaeology and planetary science give the programme both training examples and expertise in distinguishing intentional patterns from natural processes.[1]

The proposed search can therefore produce useful maps even when its most ambitious target is absent. Each unusual volcanic pit or impact feature is a planetary-science finding worth inspecting. That broader return gives researchers a practical reason to build and test the machinery for looking.[1]

Report an error or suggest a correction