지진 탐지 기술의 미래

Where Earthquake Detection Is Heading

Earthquake science has made remarkable progress since the first seismographs were installed in the late nineteenth century, but fundamental challenges remain unsolved: we cannot reliably predict earthquakes before they occur, our Seismic NetworkA coordinated group of seismograph stations that continuously monitor earthquake activity. The Global Seismographic Network (GSN) includes 150+ stations providing worldwide coverage. coverage of ocean floors and developing-world land areas remains sparse, and early warning warning times are limited by the speed of light relative to seismic waves. Next-generation detection technologies address these gaps through four converging approaches: fiber optic distributed sensing, satellite-based geodetic monitoring, artificial intelligence applied to seismic data streams, and quantum sensing.

Distributed Acoustic Sensing: Fiber Optic Seismology

Distributed Acoustic Sensing (DAS) technology transforms existing fiber optic cables into continuous SeismographAn instrument that detects and records ground motion caused by seismic waves. Modern digital seismographs can detect movements smaller than a nanometer. arrays with station spacing of meters rather than kilometers. A DAS interrogator unit sends laser pulses down the fiber and measures the tiny backward scattering variations caused by acoustic vibrations along the fiber's length. A single cable tens of kilometers long effectively becomes a seismic array with thousands of virtual sensors. Telecom cables installed under city streets, in boreholes, and across the ocean floor become seismic networks at negligible incremental cost, since the fiber infrastructure already exists.

DAS has already demonstrated detection of earthquakes, microseismicity associated with Induced SeismicityEarthquakes triggered by human activities such as hydraulic fracturing (fracking), wastewater injection, mining, or reservoir impoundment. Most are small (M<4) but some have exceeded M5.5. from fluid injection, and even traffic and environmental noise patterns that contaminate traditional network data. In submarine deployments, DAS using trans-oceanic telecom cables provides the first dense seismic coverage of the ocean floor — historically the most sensor-sparse region on Earth. The 2019 DAS experiment on the MARS cable offshore Monterey Bay demonstrated detection of an M 3.5 earthquake with comparable quality to land-based instruments.

InSAR (Interferometric SAR)A satellite radar technique that measures ground deformation with centimeter accuracy by comparing radar images taken before and after an earthquake. Reveals fault slip patterns. and Next-Generation Radar Satellites

InSAR (Interferometric SAR)A satellite radar technique that measures ground deformation with centimeter accuracy by comparing radar images taken before and after an earthquake. Reveals fault slip patterns. provides spatially dense surface deformation maps by comparing radar phase between repeat satellite passes, but current radar satellites (Sentinel-1, ALOS-2) have 6–24 day revisit times. This temporal sampling is sufficient for measuring slow interseismic deformation but misses the rapid post-seismic deformation immediately after large earthquakes. Next-generation SAR constellation concepts — including NASA-ISRO NISAR (scheduled 2024) and planned commercial SAR fleets — will achieve 1–3 day global revisit times, capturing the complete temporal evolution of post-seismic deformation from the first day onward.

Continuous InSAR monitoring at 1–3 day cadence will enable near-real-time tracking of volcanic inflation, fault creep episodes, and the days-scale strain transients that sometimes precede major earthquakes. When combined with continuous GPS GeodesyThe use of Global Positioning System receivers to measure tectonic plate motion and crustal deformation with millimeter precision. Reveals how strain accumulates on faults between earthquakes. and the emerging DAS seismic networks, this multi-sensor fusion will provide unprecedented spatial and temporal resolution of crustal deformation.

Artificial Intelligence in Seismic Phase Detection

Traditional seismic phase picking — identifying P-wave and S-wave arrival times on seismograms — was performed manually by trained analysts or using simple threshold algorithms. Deep learning models trained on millions of labeled seismogram examples now outperform both human analysts and classical algorithms for phase detection and arrival time measurement, particularly for small events near the noise floor. PhaseNet, EQTransformer, and GPD (Generalized Phase Detection) networks achieve sub-sample precision picking on continuous data streams at thousands of stations simultaneously.

AI-based catalogs produced from Southern California data have revealed two to ten times more events than conventional catalogs at the same detection threshold, by identifying events previously masked within the coda of larger events or within continuous noise. This expanded catalog density improves b-valueThe slope of the Gutenberg-Richter frequency-magnitude relationship. A b-value near 1.0 is typical; higher values indicate more small earthquakes relative to large ones. Changes may signal stress changes. estimation, reveals previously invisible Earthquake SwarmA sequence of earthquakes occurring in a localized area over days to months with no clearly dominant mainshock. Often associated with volcanic activity or fluid injection. sequences, and provides better constraints on fault geometry. The dense catalogs also enable improved Omori's LawAn empirical law describing the decay rate of aftershock frequency over time: the rate of aftershocks decreases roughly as the inverse of time since the mainshock. parameter estimation for operational aftershock forecasting.

Machine Learning for Ground Motion Prediction

Beyond phase picking, machine learning is transforming ground motion prediction equations (GMPEs). Traditional GMPEs parameterize ground motion as a function of magnitude, distance, fault type, and Vs30 site parameter using regression on hundreds to thousands of recordings. Neural network GMPEs trained on the NGA-West2 dataset (21,000+ recordings) capture nonlinear source-path-site interactions that parametric models cannot represent, reducing residual scatter and improving prediction accuracy for complex geological settings. Better GMPEs directly improve Probabilistic Seismic Hazard Analysis (PSHA)A method for quantifying earthquake hazard that considers all possible earthquake sources, magnitudes, and ground motion levels, expressing results as probability of exceeding specific shaking levels. accuracy and the design ground motions that flow from it.

Quantum Sensing for Earthquake Detection

Quantum gravimeters and seismometers represent the most frontier frontier of detection technology. Cold-atom interferometers measure gravity gradients with sensitivity exceeding conventional spring-based gravimeters by orders of magnitude. Because the prompt gravity signal from an earthquake's mass redistribution travels at the speed of light (rather than at seismic wave velocities), quantum gravimeters could in principle detect large earthquakes and estimate their magnitude before any seismic wave arrives. Detection of the prompt elastogravity signal from the 2011 Tohoku earthquake was demonstrated in 2017 using the existing gravimeter network — a proof-of-concept that dedicated quantum sensors could extend to smaller events.

Quantum seismometers based on atom interferometry also promise thermal-noise-limited sensitivity far below current MEMS and broadband seismometer technology. At this sensitivity, global monitoring of the Earth's free oscillations after large earthquakes, and possibly direct detection of gravitational waves from seismic sources, become feasible research targets.

Autonomous Ocean Floor Observatories

The global Seismic NetworkA coordinated group of seismograph stations that continuously monitor earthquake activity. The Global Seismographic Network (GSN) includes 150+ stations providing worldwide coverage. has a critical data gap: the ocean floor covers 70% of Earth's surface but hosts fewer than 1% of seismograph stations. Autonomous ocean bottom seismometers (OBS) deployed from research vessels record for months to years before being recovered, but this sampling is episodic. Permanent broadband observatories connected to shore by fiber optic cables provide continuous real-time data but are expensive to deploy and maintain. Proposed innovations include autonomous underwater vehicles that service ocean floor seismometers, reducing recovery costs, and networks of pressure-sensor-equipped floats (deep Argo-style buoys) that provide low-frequency seismic monitoring at global scale.

The Convergence: Fused Real-Time Monitoring

The future of earthquake detection lies in the fusion of all these sensor modalities into integrated real-time monitoring systems. Seismic NetworkA coordinated group of seismograph stations that continuously monitor earthquake activity. The Global Seismographic Network (GSN) includes 150+ stations providing worldwide coverage. data will be supplemented by DAS arrays, continuous GPS GeodesyThe use of Global Positioning System receivers to measure tectonic plate motion and crustal deformation with millimeter precision. Reveals how strain accumulates on faults between earthquakes., InSAR (Interferometric SAR)A satellite radar technique that measures ground deformation with centimeter accuracy by comparing radar images taken before and after an earthquake. Reveals fault slip patterns. satellite passes, ocean bottom observatories, and smartphone crowdsourced sensors. AI systems will continuously process all data streams simultaneously, detecting events, characterizing sources, issuing Earthquake Early Warning (EEW)A system that detects an earthquake and sends alerts to people and systems before strong shaking arrives. Can provide seconds to tens of seconds of warning, enough to take protective action. alerts, and updating hazard state estimates in real time. This convergence will not eliminate earthquakes or make exact prediction possible, but it will dramatically reduce the information gap between an earthquake's occurrence and an emergency manager's situational awareness — saving lives through faster, better-informed response.

Summary

The future of earthquake detection integrates distributed fiber optic sensing, satellite InSAR (Interferometric SAR)A satellite radar technique that measures ground deformation with centimeter accuracy by comparing radar images taken before and after an earthquake. Reveals fault slip patterns. at daily cadence, AI-based seismic phase detection, quantum gravimetry, and ocean floor observatory networks. Each technology addresses a specific current limitation of the Seismic NetworkA coordinated group of seismograph stations that continuously monitor earthquake activity. The Global Seismographic Network (GSN) includes 150+ stations providing worldwide coverage. — sparse coverage, temporal gaps, analyst bottlenecks, or fundamental sensitivity limits. Together they promise a monitoring capability that will reveal seismic phenomena invisible to current instruments and enable the Earthquake Early Warning (EEW)A system that detects an earthquake and sends alerts to people and systems before strong shaking arrives. Can provide seconds to tens of seconds of warning, enough to take protective action. systems of the next decade to be faster, more accurate, and more globally available than anything operating today.