A newly developed machine learning tool could help scientists search for signs of life on Mars and other alien worlds.
With the ability to collect samples from other planets severely limited, scientists currently have to rely on remote sensing methods to look for signs of extraterrestrial life. This means that any method that could help guide or refine this search would be incredibly useful.
With this in mind, a multidisciplinary team of scientists led by Kim Warren-Rhodes of the SETI (Search for Extraterrestrial Intelligence) Institute in California has mapped the sparse life forms living in salt domes, rocks and crystals in the Salar de Pajonales, a salt flat the border of Chile’s Atacama Desert and Altiplano or Plateau.
Related: The Search for Extraterrestrial Life (Reference)
Biosignature probability maps from convolutional neural network models and statistical ecology data. The colors in a) indicate the probability of detecting biosignatures. In b) is a visible image of a geological feature of a gypsum dome (left) with biosignature probability maps for different microhabitats (e.g. sand versus alabaster) inside. (Photo credits: M. Phillips, F. Kalaitzis, K. Warren-Rhodes.)
Warren-Rhodes then teamed up with Johns Hopkins University’s Applied Physics Laboratory’s Michael Phillips and University of Oxford researcher Freddie Kalaitzis to train a machine learning model to recognize the patterns and rules associated with the distribution of life connected in the rough region. Such training taught the model to recognize the same patterns and rules for a variety of landscapes – including those that may be on other planets.
The team discovered that by combining statistical ecology with AI, their system could locate and detect biosignatures up to 87.5% of the time. This compares to no more than a 10% success rate achieved through random searches. In addition, the program could reduce the area required for a search by up to 97%, helping scientists significantly refine their search for potential chemical life traces or biosignatures.
“Our framework allows us to combine the power of statistical ecology with machine learning to discover and predict the patterns and rules by which nature survives and disperses across Earth’s harshest landscapes,” Warren-Rhodes said in one Explanation (opens in new tab ). “We hope that other astrobiological teams will adapt our approach to map other habitable environments and biosignatures.”
Such machine learning tools, the researchers say, could be applied to robotic planetary missions like that of NASA’s Perseverance rover, which is currently searching for signs of life on the floor of Jezero Crater on Mars.
“With these models, we can design custom road maps and algorithms to guide rovers to locations most likely to harbor past or present life — no matter how hidden or rare,” Warren-Rhodes explained.
Choosing an analogue for Mars on Earth
The team chose the Salar de Pajonales as a test phase from their machine learning model because it is a suitable analog for the arid and arid landscape of modern-day Mars. The region is a high-altitude dry salt lake bed blasted with high levels of ultraviolet radiation. Although the Salar de Pajonales is considered extremely inhospitable to life, it is nevertheless home to some critters.
The team collected nearly 8,000 images and over 1,000 samples from the Salar de Pajonales to discover photosynthetic microbes living in the region’s salt domes, rocks and alabaster crystals. The pigments secreted by these microbes represent a possible biosignature on NASA’s “ladder of life detection” (opens in new tab), designed to guide scientists to search for life beyond Earth within the practical constraints of robotic space missions.
The team also studied Salar de Pajonales using drone images that are analogous to images of the Martian terrain captured by the High-Resolution Imaging Experiment (HIRISE) camera aboard NASA’s Mars Reconnaissance Orbiter. Using this data, they were able to determine that the microbial life in the Salar de Pajonales is not randomly distributed but is concentrated in biological hotspots that are strongly linked to water availability.
Warren-Rhodes’ team then trained convolutional neural networks (CNNs) to detect and predict large geological features at the Salar de Pajonales. Some of these features, such as Patterned soil or polygonal networks are also found on Mars. The CNN has also been trained to recognize and predict smaller microhabitats that are most likely to contain biosignatures.
For now, the researchers will continue to train their AI at the Salar de Pajonales to next test the CNN’s ability to predict the location and distribution of ancient stromatolite fossils and salt-tolerant microbiomes. This should help him figure out if the rules he’s using in this search might also apply to hunting for biosignatures in other similar natural systems.
After that, the team plans to start mapping hot springs, frozen permafrost-covered soils, and the rocks in dry valleys, and hopefully teach the AI to study potential habitats in other extreme environments here on Earth before potentially exploring those of other planets .
The team’s research was published this month in the journal Nature Astronomy (opens in new tab). (opens in new tab)
Follow us on Twitter @Spacedotcom (opens in new tab) and on Facebook (opens in new tab).
