For decades, seismologists classified continental mantle earthquakes as rare anomalies — isolated events in a region that should be too hot and ductile for brittle failure. The mantle deforms plastically. It flows. Earthquakes require snapping, and the mantle should not snap. Individual mantle earthquakes had been identified, debated, and filed as exceptions. No one had assembled a global catalog because no one had a reliable way to tell, from a seismogram recorded hundreds of kilometers away, whether an earthquake nucleated in the crust or in the mantle below it.
Wang and Klemperer at Stanford solved this not by improving location accuracy — the standard approach, which requires dense station networks and produces unreliable depth estimates for moderate-depth events — but by listening to the signal differently. They measured the ratio of two seismic wave types that propagate through fundamentally different structures.
Sn waves travel along the uppermost mantle lid — the high-velocity layer immediately below the Mohorovičić discontinuity. They are guided by the mantle lithosphere, efficiently propagated when the source is in or near the mantle. Lg waves are high-frequency surface waves that reverberate within the crustal waveguide — they bounce through the crust and are the dominant phase for crustal earthquakes at regional distances. An earthquake in the mantle generates strong Sn and weak Lg, because the energy originates in the Sn waveguide and must couple upward into the crust to generate Lg. An earthquake in the crust generates strong Lg and weak Sn, because the energy originates in the crustal waveguide and must couple downward to generate Sn.
The ratio is the diagnostic. Not the absolute amplitude of either phase — that depends on distance, attenuation, station conditions. The ratio depends primarily on origin depth relative to the Moho. High Sn/Lg means mantle. Low Sn/Lg means crust. The measurement requires no depth modeling, no velocity structure assumptions beyond the basic crust-mantle layering, no dense network. A single well-placed station recording both phases at regional distance can make the determination.
Applied to over 46,000 earthquakes since 1990, the method identified 459 continental mantle earthquakes — the first global catalog. They are not rare anomalies scattered randomly. They cluster in the Himalayas, the Tibetan Plateau, and the Bering Strait — regions of active continental convergence with thickened lithosphere. Some originate more than 80 kilometers below the Moho, at total depths that may exceed 160 kilometers beneath the Tibetan surface.
The catalog settles a long-running debate about lithospheric rheology. The “jelly sandwich” model posits a strong upper crust, a weak lower crust, and a second strong layer in the uppermost mantle. The “crème brûlée” model posits that all mechanical strength resides in the upper crust, with everything below too weak for brittle failure. Four hundred and fifty-nine continental mantle earthquakes across multiple tectonic settings are strong evidence for jelly sandwich. The mantle lithosphere is, at least locally, strong enough to store and release elastic strain energy seismically.
But the structural insight is in the method, not the result. The conventional approach to depth determination — triangulating the source from arrival times at multiple stations — treats the earthquake as a point to be located in space. The Sn/Lg approach treats the earthquake as a signal that couples differently to different propagation channels depending on where it was generated. You do not find the earthquake. You observe which channel carries it, and the channel tells you where it came from. The location is determined by the coupling, not by the triangulation.
This is a general principle with applications beyond seismology. Any system with multiple propagation channels — waveguides, pathways, modes — carries information about the source in the relative amplitude of the channels. The source's position in the structure determines which channels it feeds efficiently and which it feeds poorly. The ratios are the tell. Measuring them requires understanding the structure well enough to know what each channel implies, but it does not require locating the source directly. The structure itself is the instrument.