Every volcano is supposed to be unique. Each has its own plumbing, its own magma composition, its own structural weaknesses, its own history. Eruption forecasting has traditionally been volcano-specific: build a model of this volcano from this volcano's past behavior, calibrate to this volcano's baseline, interpret this volcano's anomalies in light of this volcano's precedents.
A study across 24 volcanoes worldwide finds that eruption precursors follow a universal pattern. Seismicity, deformation, and gas emission all accelerate before eruptions in a statistically similar way regardless of the volcano. A machine learning classifier trained on precursor data from one set of volcanoes predicts eruptions at volcanoes it has never seen with an area under the curve of approximately 0.8.
The universality is in the scaling, not the specifics. Each volcano has different absolute levels of seismicity, different background deformation rates, different gas fluxes. But the relative change — the acceleration from baseline to eruption — follows similar curves. The ratio of current activity to recent baseline is the informative variable, not the absolute value. A volcano at its personal 90th percentile of seismicity is approximately as likely to erupt as any other volcano at its personal 90th percentile, regardless of whether that means 50 events per day or 500.
This means eruption forecasting can work at volcanoes with no prior eruption history — the majority of potentially dangerous volcanoes worldwide. A model trained on well-monitored volcanoes can transfer to poorly studied ones because the diagnostic pattern is relative, not absolute.
The uniqueness of each volcano is real but confined to calibration. The physics of failure — pressurization, fracturing, venting — produces similar acceleration patterns in different containers because the failure process is more universal than the container.