Machine Learning Can Predict the Timing and Size of Analog Earthquakes

Despite the growing spatiotemporal density of geophysical observations at subduction zones, predicting the timing and size of future earthquakes remains a challenge. Here we simulate multiple seismic cycles in a laboratory-scale subduction zone. The model creates both partial and full margin ruptures, simulating magnitude M w 6.2-8.3 earthquakes with a coefficient of variation in recurrence intervals of 0.5, similar to real subduction zones. We show that the common procedure of estimating the next earthquake size from slip-deficit is unreliable. On the contrary, machine learning predicts well the timing and size of laboratory earthquakes by reconstructing and properly interpreting the spatiotemporally complex loading history of the system. These results promise substantial progress in real earthquake forecasting, as they suggest that the complex motion recorded by geodesists at subduction zones might be diagnostic of earthquake imminence. Plain Language Summary Large and devastating subduction earthquakes, such as the 2011 magnitude 9.0 Tohoku-oki earthquake (Japan), are currently considered unpredictable. Scientists lack a long enough seismic catalog that is necessary for drawing statistical insights and developing predictions. For this reason, we simulate tens of earthquakes using a small-scale experimental replica of a subduction zone. We show that machine learning (a group of algorithms that make predictions based on the "information" acquired in past "experience") can predict when, where, and how big the next experimental earthquake will be. The "information" in our study is provided by the slow deformation accumulating in the analog tectonic plates during the periods in between earthquakes. Since such slow deformation is also measured by means of space geodesy along real subduction zones, there is the possibility that, in the future, variations of this machine learning approach can predict the timing and size of natural subduction earthquakes.

[1]  J. Avouac,et al.  From Geodetic Imaging of Seismic and Aseismic Fault Slip to Dynamic Modeling of the Seismic Cycle , 2015 .

[2]  Kenji Satake,et al.  Long-Term Perspectives on Giant Earthquakes and Tsunamis at Subduction , 2007 .

[3]  M. Rosenau,et al.  Analogue earthquakes and seismic cycles: experimental modelling across timescales , 2016 .

[4]  J. Friedman.  Greedy function approximation: A gradient boosting machine. , 2001 .

[5]  Mian Liu,et al.  Comparison of Seismicity Rates in the New Madrid and Wabash Valley Seismic Zones , 2010 .

[6]  Hiroo Kanamori,et al.  The physics of earthquakes , 2001 .

[7]  R. Briggs,et al.  Coseismic slip and early afterslip of the 2015 Illapel, Chile, earthquake: Implications for frictional heterogeneity and coastal uplift , 2016 .

[8]  H. Kanamori,et al.  An Asperity Model of Large Earthquake Sequences , 2013 .

[9]  M. Métois,et al.  Interseismic Coupling, Megathrust Earthquakes and Seismic Swarms Along the Chilean Subduction Zone (38°–18°S) , 2016, Pure and Applied Geophysics.

[10]  Paul A. Johnson,et al.  Breaking Cascadia's Silence: Machine Learning Reveals the Constant Chatter of the Megathrust , 2018 .

[11]  M. Moreno,et al.  Impact of megathrust geometry on inversion of coseismic slip from geodetic data: Application to the 1960 Chile earthquake , 2009 .

[12]  K. Sieh,et al.  A paleogeodetic record of variable interseismic rates and megathrust coupling at Simeulue Island, Sumatra , 2015 .

[13]  B. Meade,et al.  Two decades of spatiotemporal variations in subduction zone coupling offshore Japan , 2015 .

[14]  K. Heki,et al.  Accelerated Pacific Plate Subduction Following Interplate Thrust Earthquakes at the Japan Trench , 2013 .

[15]  Julian J. Bommer,et al.  Scaling of the Source Dimensions of Interface and Intraslab Subduction-zone Earthquakes with Moment Magnitude , 2010 .

[16]  B. Bookhagen,et al.  Using uplifted Holocene beach berms for paleoseismic analysis on the Santa María Island, south‐central Chile , 2006 .

[17]  R. Yeats,et al.  Superquakes and Supercycles , 2013 .

[18]  C. Goldfinger,et al.  HOLOCENE EARTHQUAKE RECORDS FROM THE CASCADIA SUBDUCTION ZONE AND NORTHERN SAN ANDREAS FAULT BASED ON PRECISE DATING OF OFFSHORE TURBIDITES , 2003 .

[19]  J. K. Sveen.  An introduction to MatPIV v. 1.6.1 , 2004 .

[20]  M. Moreno,et al.  2010 Maule earthquake slip correlates with pre-seismic locking of Andean subduction zone , 2010, Nature.

[21]  Rongjiang Wang,et al.  Gradual unlocking of plate boundary controlled initiation of the 2014 Iquique earthquake , 2014, Nature.

[22]  J. Avouac,et al.  Megathrust friction determined from mechanical analysis of the forearc in the Maule earthquake area , 2013 .

[23]  Paul A. Johnson,et al.  Estimating Fault Friction From Seismic Signals in the Laboratory , 2017, 1710.04172.

[24]  Mark D. Petersen,et al.  Cascadia Subduction Zone , 2008 .

[25]  Naoshi Hirata,et al.  Propagation of Slow Slip Leading Up to the 2011 Mw 9.0 Tohoku-Oki Earthquake , 2012, Science.

[26]  R. Kulkarni,et al.  Statistical Analyses of Great Earthquake Recurrence along the Cascadia Subduction Zone , 2013 .

[27]  Yehuda Bock,et al.  Plate-boundary deformation associated with the great Sumatra–Andaman earthquake , 2006, Nature.

[28]  M. Moreno,et al.  Estimating coseismic coastal uplift with an intertidal mussel: calibration for the 2010 Maule Chile earthquake (Mw = 8.8) , 2012 .

[29]  P. Segall,et al.  A decadal‐scale deformation transient prior to the 2011 Mw 9.0 Tohoku‐oki earthquake , 2014 .

[30]  M. Ando.  Source mechanisms and tectonic significance of historical earthquakes along the nankai trough, Japan , 1975 .

[31]  Control of asperities size and spacing on seismic behavior of subduction megathrusts , 2017 .

[32]  M. Bevis,et al.  Heterogeneous plate locking in the South–Central Chile subduction zone: Building up the next great earthquake , 2011 .

[33]  Chen Ji,et al.  Partial rupture of a locked patch of the Sumatra megathrust during the 2007 earthquake sequence , 2008, Nature.

[34]  Jim Mori,et al.  Are asperity patterns persistent? Implication from large earthquakes in Papua New Guinea , 2007 .

[35]  P. Mai,et al.  The seismic cycle at subduction thrusts: 1. Insights from laboratory models , 2013 .

[36]  M. Peyret,et al.  Roughness Characteristics of Oceanic Seafloor Prior to Subduction in Relation to the Seismogenic Potential of Subduction Zones , 2018, Geochemistry, Geophysics, Geosystems.

[37]  Mario Pardo,et al.  Reappraisal of great historical earthquakes in the northern Chile and southern Peru seismic gaps , 1991 .

[38]  Hai Cheng,et al.  Earthquake Supercycles Inferred from Sea-Level Changes Recorded in the Corals of West Sumatra , 2008, Science.

[39]  Brendan J. Meade,et al.  Spatial correlation of interseismic coupling and coseismic rupture extent of the 2011 MW = 9.0 Tohoku‐oki earthquake , 2011 .

[40]  Takashi Nakata,et al.  Time‐predictable recurrence model for large earthquakes , 1980 .

[41]  C. Humphreys,et al.  Machine Learning Predicts Laboratory Earthquakes , 2017, Geophysical Research Letters.

[42]  John McCloskey,et al.  Limited overlap between the seismic gap and coseismic slip of the great 2010 Chile earthquake , 2011 .

[43]  Nadia Lapusta,et al.  Towards inferring earthquake patterns from geodetic observations of interseismic coupling , 2010 .

[44]  F. Viégas,et al.  Deep learning of aftershock patterns following large earthquakes , 2018, Nature.