About $1.36M to Woods Hole Oceanographic Institution — machine learning over sediment metal and fish fossil data to predict where deep-sea rare earths concentrate

1Fish fossils are among the inputs

Two materials go into the machine learning here: data on metals in the sediment, and data on fish fossils. Why fish fossils, for predicting where rare earths lie.

The abstract does not set out the mechanism in detail, saying that combining the two will reveal when, where and how rare earths become enriched in sediment. As records of what settled on the sea floor, metals and fossils lie in the same layers. The reading is that what neither shows alone can appear when they are set side by side.

2Prediction fills in where sampling cannot reach

Studying deep-sea sediment means putting a ship out and taking samples. The abstract notes that traditional methods are inefficient for mapping spatial variability. The ocean is wide and the points that can be sampled are few.

That is where prediction comes in: rather than sampling everywhere, infer the unsampled places from the points already taken. Deep-sea mining is also a field where argument over environmental impact continues, and that argument struggles to advance while what lies where remains unknown.

3The recipient is an independent oceanographic institution

Among the 120 NSF awards this site holds as of 2026-09-01, recipients break down as 104 universities and the like, 13 companies and three independent research institutes. This award is one of those three.

Those 120 awards are spread across 93 institutions, and only 19 hold more than one. Most institutions hold a single award. The obligated amount of about $1,356,482 matches the estimated total; among the 120 awards this site holds as of 2026-09-01, 86 match in that way and this is one of them. The period runs three years, from 2026-09-01 to 2029-08-31.

4Sample it, or infer it

Studying deep-sea sediment means putting a ship to sea and collecting it. The abstract notes that conventional approaches are inefficient for mapping spatial variability. The ocean is wide and the points that can be sampled are few.

That the machine learning takes two inputs bears on the same design: metal data in the sediment, and fish fossil data. As records of what settled to the sea floor, metals and fossils remain in the same layers. What one alone cannot say may be readable when the two are set side by side — the mechanism goes undescribed in the abstract, but that intention is stated plainly.