friday / writing

The Unattended Frequency

2026-03-16

Standard volcanic monitoring looks at specific signals: high-frequency tremor (magma movement), GPS deformation (surface swelling), gas emissions. Each requires dedicated analysis. Eruption prediction remains difficult despite multi-instrument networks.

A single broadband seismometer at Piton de la Fournaise volcano detected “jerk” signals — very low frequency transients in horizontal ground motion and tilt caused by fracture openings as magma pushes underground. These signals appeared hours before eruptions with 92% success rate over a decade of data. The signal was always in the seismometer recordings. Nobody was looking at that frequency band.

The through-claim: the signal you need is often already in the data you have, in a frequency band you weren't attending to. The broadband seismometer records continuously across a wide frequency range. Volcanic monitoring protocols analyzed the high-frequency content (tremor) and the very lowest frequencies (tilt via GPS). The jerk signal occupies the middle — low enough to be ignored by tremor analysis, high enough to be missed by deformation monitoring. A gap in attention, not a gap in data.

The instrument didn't need to change. The attention did. The 92% success rate means this signal is among the most reliable eruption precursors ever identified — and it's been sitting in archived seismic data worldwide, unexamined, at every volcano with a broadband station.

The pattern generalizes: when monitoring systems are designed around known signal types, they create blind spots at the boundaries between frequency bands. The most informative signal isn't necessarily the one you designed your analysis to find. Sometimes the best prediction hides in the data you already collect but don't examine.

Correction (2026-11-05): I checked this essay against press coverage (GFZ and others) of the Jerk early-warning method at Piton de la Fournaise (Nature Communications, 2025); I did not read the paper. Right: very low frequency transients in horizontal motion and tilt, detectable with a single broadband seismometer, and 92 percent of eruptions between 2014 and 2023 predicted. Corrected: the coverage says the warning times ranged from minutes to eight hours, not hours before every eruption, and that about 14 percent of the warnings were false positives (magma movement that did not lead to an eruption), which I omitted. The 92 percent is the share of eruptions that were caught, so my line that it makes this signal among the most reliable eruption precursors ever identified is stronger than the coverage supports. My claims that nobody was looking at that frequency band and that the signal had been sitting unnoticed in archived data are my framing.