What the surface reveals about what lies beneath

A chemical fingerprint in stream sediment

Exploration geologists face a search problem: rare-earth elements matter, but surveying every catchment is slow. In a preprint, the authors test whether the age and type of the rock underneath can narrow the search, using stream sediments across the western United States. m:…3723

The signal is strong. Mesoproterozoic intrusive terrain is associated with a median acid-leachable yttrium concentration 2.03 times higher than other geological periods. Yttrium itself is useful because it serves as a proxy for heavy rare earth elements, with a Spearman correlation of 0.900 against independent whole-rock data. m:…3723

The authors also check that the pattern is geological rather than an artifact of age. The observed yttrium enrichment is magmatic in origin, as evidenced by the lack of enrichment in the Belt Supergroup control group, which shares the time period but not the intrusive rock type. m:…3723

For exploration teams, the value of the result is prioritization rather than proof. A terrain type that carries more than double the median yttrium of other periods gives surveyors a place to start, and a proxy validated at 0.900 against whole-rock data means stream sediment samples can stand in for far more expensive rock sampling in the first pass. The preprint does not claim that any single catchment is economic,, so the next step for a user is still field verification. m:…3723

A sixth-century manual, tested against wells

A second preprint applies the same discipline to a very different archive: the groundwater guidance in Varahamihira's sixth-century Brhatsamhita, tested against India's modern well network. The question is which of the manual's plant indicators actually predict shallow water. m:…1907

Part of the manual survives the test. Near-water tree species identified in the text show a statistically significant correlation with shallow water tables, at p=0.020. The manual's depth predictions for wet and dry regions, 3 to 17 metres, are also consistent with modern dug-well depths in India. m:…1907

The rest does not. The broader set of tree indicators in the manual is not significantly associated with shallow water compared with common Indian plant species, at p=0.43. The honest reading is selective: some indicators carry real signal, the full list does not. m:…1907

Cameras as seismometers

The third preprint turns an everyday sensor into a scientific one. By integrating security camera footage with seismic and field measurements, the authors show that security camera footage can be used to retrieve displacement time series that match seismic records for near-fault ground motion. m:…1773

The test case is the 2026 Kumamoto earthquake, which produced two velocity pulses of about 1 metre per second amplitude and about 1 second duration, attributed to subsurface and surface fault rupture. Two pulses, two rupture stages, read off cameras that were never installed to watch the ground move. m:…1773

The three preprints share a method as much as a theme: take a surface record — sediment chemistry, a tree list, a camera feed — and ask what it proves about what lies below. In each case the answer is partial and specific, which is what makes it usable: a terrain type worth prioritizing, a subset of indicators worth keeping, a displacement record worth trusting. m:…3723 m:…1907 m:…1773

That partial quality is the point. A proxy that works at 0.900 correlation still leaves room for misses, a manual whose broad indicator list fails still contains a subset that passes, and a camera record that matches seismic data for one earthquake still needs testing elsewhere. Surface evidence earns its use case by case, and these three preprints show what that earning looks like. m:…3723 m:…1907 m:…1773