Potential short-term earthquake forecasting by farm animal monitoring
Summary
WIVVI summary: Researchers used biologging sensors to monitor the collective activity of cows, sheep, and dogs in an earthquake-prone region of central Italy. The reported animal activity patterns were compared with nearby earthquake occurrence, distance, and magnitude. The authors reported unusual collective activity before some earthquakes, particularly while the animals were housed in a stable, and proposed that continuous animal monitoring might contribute to short-term seismic research. These observational findings do not establish a reliable earthquake-prediction method.
Source
Ethology, Volume 126, Issue 9, pages 931–941
Keywords
animal behaviour, earthquake precursors, seismic activity, farm animals, cows, sheep, dogs, biologging, accelerometers, collective activity, earthquake forecasting
Source Type
research article, peer reviewed, observational study
Key Findings
The authors reported that:
Collective animal activity increased before some nearby earthquakes while the animals were housed in a stable.
Comparable anticipatory patterns were not detected while the animals were grazing freely on pasture.
Reported warning times ranged from approximately 1 to 20 hours and tended to increase as the animals’ distance from the earthquake hypocentre decreased.
Combining observations across cows, sheep, and dogs appeared more informative than examining individual animals alone.
These are the authors’ reported results, not independently established forecasting capability.
Limitations
This was an observational study conducted in one region with a limited group of animals; it does not establish causation or generalizability.
Activity can be affected by feeding, handling, weather, predators, noise, human presence, housing conditions, and other uncontrolled influences.
The reported pattern appeared in the stable but not on pasture, showing that environmental context substantially affected the result.
A published critique argued that the analysis did not adequately distinguish the reported patterns from chance or fully assess false alarms and missed earthquakes.
In their subsequent response, the authors acknowledged that earthquake prediction is not possible from this dataset and should not be attempted.
Independent, preregistered replication with prospective alert thresholds and out-of-sample testing is required before forecasting value can be assessed.
Methods Summary
Researchers attached biologging sensors to cows, sheep, and dogs on a farm near the epicentre of the 2016 magnitude-6.6 Norcia earthquake in central Italy. Animal activity was monitored continuously during parts of the 2016–2017 earthquake sequence, including periods in a stable and on pasture. The researchers aggregated activity across the animal group and compared it with the timing, magnitude, hypocentral distance, and estimated arrival of thousands of recorded earthquakes: 5,304 events in 2016 and 12,948 in 2017, spanning magnitudes 0.4–6.6. Statistical thresholds were used to identify unusually elevated collective activity and examine whether it occurred before seismic events.