Programming computers to find Antarctic meteorites.

Science magazine explains how that big southerly continent has become the best place to find meteorites, and how those ancient fallen rocks offer priceless clues to the history of the solar system. The trouble is finding them in all that ice. That’s where machine learning comes in, locating the best places to consistently find the most meteorites … in “blue ice areas” (BIAs):

These meteorites are then transported along with the ice that flows under gravitational forces toward the margins of the continent. Although most of the englacially transported meteorites end up in the ocean, a small fraction is brought back to the surface of the ice sheet in some of the continent’s blue ice areas (BIAs)….If the ice within a BIA contains meteorites, these meteorites eventually become exposed through the removal of the ice by ablative processes (sublimation). Moreover, the absence of snow accumulation in a BIA implies that meteorites falling directly on a BIA can remain exposed at the surface. Thus, if the flow of the ice and specific geographical and climatological settings combine favorably, a BIA can act as a meteorite stranding zone (MSZ)…. In MSZs, meteorites are concentrated at the surface, where they can be easily recovered during field missions, as, thanks to their color, they contrast with the underlying blue ice. These MSZs make Antarctica the most productive region for collecting meteorites on Earth; to date, about 62% of all meteorites recovered on Earth originate from Antarctica.

Many of today’s known MSZs were discovered coincidentally, and to date, the identification of new MSZs remains a very labor-intensive process that strongly relies on chance and past experience. Potential MSZs are typically identified through visual examination of remote sensing data of BIAs and their vicinity, after which candidate MSZs are visited by snowmobile or helicopter, to investigate whether a meteorite concentration is present (15). The discovery of meteorite concentrations thus partly depends on the expertise and experience of the persons examining maps and imagery, and largely on costly field reconnaissance visits. Because of this big human factor in the reconnaissance approach, it is most likely that major MSZs are still to be discovered.

To overcome these limitations, we introduce a data-driven method, where multiple datasets with quantifiable uncertainties are used to predict the probability to find meteorites given an observation anywhere in Antarctica.

The where-to-go index, used to rank MSZs, is calculated by considering three parameters that are important for a successful field mission. For each of the three parameters, the rank of the MSZ is calculated and subsequently summed up to obtain the where-to-go index. Hence, MSZs with low where-to-go indices correspond to MSZs with a high potential, and vice versa.
The three considered parameters are as follows: (i) the distance to the nearest research station … (ii) The median of the a posteriori probability of all grid cells within the (potential) MSZ. The a posteriori probability relates to the probability to find meteorites at a given location … (iii) A parameter representing the presence of temporary snow layers. This parameter corresponds to the number of days during the meteorite collecting season (November to March) for which at least 50% of the MSZ is snow free. For large MSZs (>20 km2), we use an alternative criterion: here, at least 10 km2 should be snow free.