AI early-warning systems fuse many seismometers: the moment P-waves (fast, low-amplitude) arrive at several stations, a network estimates magnitude and epicentre, then predicts shaking and tsunami risk before the slower, destructive S-waves and surface waves arrive:
r_P(t) = v_P·t, r_S(t) = v_S·t (v_P ≈ 6.0, v_S ≈ 3.5 km/s, scaled)
risk(d) = σ( M − 5.0 − k·d ), σ(x) = 1 / (1 + e^−x)
tsunami_ETA = d_coast / v_wave (v_wave ≈ 0.7 km/s, shallow-water scaled)
- Magnitude — sets the earthquake's energy; higher M widens the radius at which the AI model still flags meaningful risk.
- Network density — the grid of instanced sensor nodes; a denser array reaches the detection threshold sooner and pinpoints the epicentre with less error, mirroring how AI early-warning improves with more stations.
- Detection threshold — the sigmoid cutoff a node's local risk score must cross to switch from grey "listening" to red "alert", visualising a classifier's decision boundary sweeping outward with the wavefront.
- Tsunami wave — once magnitude crosses a coastal-risk threshold, a ring propagates across the ocean plane toward the shoreline at a much slower group velocity than the seismic waves, the same timing gap tsunami-warning AI exploits to issue alerts.