What Do Animals Sense Before Earthquakes? Pre-specified Tests of Magnetic, Infrasound and Seismic Precursors with Open Data

Animals are often said to sense earthquakes minutes or hours in advance. To find outwhich physical stimulus an animal could be noticing, we measured four candidates withsensors and open data. Every test followed a plan written before any computation, with asingle primary test, positive controls, placebo catalogues and dated changes.Magnetic field. Over 82 nights before M ≥ 4.5 earthquakes within 100 km of fourINTERMAGNET observatories (74 at Kakioka, Japan), residual power at 0.001–0.01 Hzchanged by +3% (95% CI −10% to +19%; p = 0.77). Local signals larger than about 0.02nT are ruled out. At 0.01–0.1 Hz, and with the induction coils at Parkfield and Hollister(California), no reference removed geomagnetic storms, so those tests were notevaluated.Infrasound. With USArray microbarometers within 50 km of M ≥ 3.5 earthquakes, thestrongest minute at 1–10 Hz was, in median, 7 times more powerful in the 10 minutesafter the earthquake than on days without one (p = 0.0001). In the previous hour it didnot change (mean z +0.14; p = 0.18; 236 pairs): an added signal does not exceed about0.1 Pa (74 dB) on average.Foreshocks. With the same rule in four regions, a small earthquake within 10 kmpreceded, in the previous hour, 19.0% of mainshocks in Japan, 10.7% in the IberianPeninsula, 28.1% in Italy and 20.4% in California, against less than 1% expected in quiethours at the same epicentre. In Italy and California the fraction stays far above chancewith radii of 5–20 km and windows of 30–120 minutes. Between 38% and 50% of thosecases had their last foreshock in the final 5 minutes, somewhat fewer than the almost60% of last-hour animal reports; without the cases in the final minute, which may bepart of the rupture itself, the share falls to 17–26%.Animal activity. In published data from a farm near Norcia (Italy), with 15-minute meanactivity, not even earthquakes felt by people produced a detectable rise afterwards.With that measure, anticipation could not be evaluated.Foreshocks are a known, measurable stimulus, present at the right time, that can explainpart of the reports of animals reacting minutes before an earthquake; that they exist doesnot show that animals react to them. The magnetic and infrasound signals did not appearat the levels measured, in the bands, distances and windows tested. Between 72% and89% of mainshocks had no foreshock in the previous hour, and the reports linked to themremain unexplained. The plans, code and log of steps are published with this work Supplementary file: code, pre-specified plans (in Spanish), dated log and SHA-256 fingerprints