09:45 〜 10:00
[SCG40-24] Characterizing regular and slow earthquake seismogenic zones by frictional locking and kinematic coupling
キーワード：Geodetic data inversion、Frictional fault modeling、Coupling-locking semantics
Slow earthquakes in subduction zones often trace the rim of the focal zones of past large earthquakes (Obara & Kato, 2016). Source areas of regular earthquakes are characterized by high slip deficits, while those of slow earthquakes are characterized by moderate slip deficits (Baba et al., 2020; Sherill et al., 2022). Those facts are analogous to the following observations in physics-based earthquake cycle simulations: a locked zone brakes the surrounding stably creeping zone, and the incurred slip deficits in creeping zones are compensated for by intermittent aseismic slips (Tse & Rice, 1986). To test whether this relationship holds between actual slow-to-fast seismogenic zones and frictional locking, we study frictionally locked zones in the Nankai subduction zone, Japan.
Coupling (slip deficit) is a conventional proxy of locking, but in essence, locking is defined by friction (Wang & Dixon, 2004). Thus, a series of modern geodetic inversions distinguish locking from coupling and estimate locked zones in the sense of Amonton-Coulomb friction (Burgmann, 2007; Johnson & Fukuda, 2010; Herman & Govers, 2020). This locking inversion defines locking as the static-friction (stick) phase with zero slip rate and unlocking as the dynamic-friction (slip) phase with zero stressing rate. While the conventional slip deficit inversion estimates a slip-deficit field, the locking inversion estimates a field of stick-slip binaries (locking parameters) by imposing the physical constraint that expresses the slip-deficit field as a functional of the locking-parameter field.
We conduct this locking inversion. Of course, it is nontrivial how the Amonton-Coulomb definition of locking would be corrected for other friction laws. Therefore, we first examine the above definition of locking and confirm its validity for long periods over which the slip rate is almost constant (i.e., interseismic periods) for general frictional models that obey the yielding law. Besides, grid-based locking inversions easily overfit (Herman & Govers, 2020), so we adopt a transdimensional scheme, where the number of locked segments varies and each segment is approximated into a circular patch. The radii and positions of respective segments are model parameters. The number of segments is a hyperparameter. Locked segments correspond to asperities in fault mechanics, so our multi-patch model may be regarded as a reduced-order model for extracting robust information about the locked zone.
We analyze the data of onshore GNSS of GEONET and offshore GNSS-Acoustic (Yokota et al., 2016). We use a homogeneous half-space Green's function, and the fault geometry model follows the Japan Integrated Velocity Structure Model Version 1 (Koketsu et al., 2012). Green's function errors are accounted for by the method of Yagi & Fukahata (2011). We also check that using Green's function for a realistic 3D velocity model (Hori et al., 2021) does not much change the results of the homogeneous half-space model at least for the conventional slip deficit inversion. The number of asperities, the hyperparameter, is determined by the marginal likelihood maximization using the Bayesian information criterion (Schwarz, 1978).
Our optimal estimate has detected five asperities. Their along-strike sections, surprisingly, closely resemble the five basins of the Nankai subduction zone: Hyuga, Tosa, Muroto, Kumano, and Enshu, the envisioned five focal sections of the Nankai megathrust earthquake (Furumura et al., 2011; Saito & Noda, 2022). Our estimate also concludes the unlocking at the segmentation boundary of the 1944 Tonankai and 1946 Nankai earthquakes. Fittingly, this segmentation boundary is the neighborhood of the initiation zones of those earthquakes, which are supposed to be near the locked zone edges hosting the nuclei of those earthquakes (Hori et al., 2006). Observation points exist above that boundary, so its unlocking is likely. The estimated locked zones coincide with the previous focal zones of the same basins (Obara & Kato, 2016) in all basins but the Hyuga. The Hyuga locked zone includes the slip zones of the 1968 Hyuga-nada earthquake (Yagi et al., 1998) and the Bungo-channel long-term slow-slip events (Obara & Kato, 2016), though documented deep low-frequency tremors are all outside the locked zone estimate. Our results suggest that, on geodetic scales, the seismogenic zone of regular earthquakes is locked, whereas the slip zone of slow earthquakes at depth is coupled but unlocked. The only exception is the slip zone of the Bungo-channel slow-slip events, which may be the nucleation zone that often fails to slip faster, sometimes succeeding as supposedly in 1707 (Furumura et al., 2011).
Coupling (slip deficit) is a conventional proxy of locking, but in essence, locking is defined by friction (Wang & Dixon, 2004). Thus, a series of modern geodetic inversions distinguish locking from coupling and estimate locked zones in the sense of Amonton-Coulomb friction (Burgmann, 2007; Johnson & Fukuda, 2010; Herman & Govers, 2020). This locking inversion defines locking as the static-friction (stick) phase with zero slip rate and unlocking as the dynamic-friction (slip) phase with zero stressing rate. While the conventional slip deficit inversion estimates a slip-deficit field, the locking inversion estimates a field of stick-slip binaries (locking parameters) by imposing the physical constraint that expresses the slip-deficit field as a functional of the locking-parameter field.
We conduct this locking inversion. Of course, it is nontrivial how the Amonton-Coulomb definition of locking would be corrected for other friction laws. Therefore, we first examine the above definition of locking and confirm its validity for long periods over which the slip rate is almost constant (i.e., interseismic periods) for general frictional models that obey the yielding law. Besides, grid-based locking inversions easily overfit (Herman & Govers, 2020), so we adopt a transdimensional scheme, where the number of locked segments varies and each segment is approximated into a circular patch. The radii and positions of respective segments are model parameters. The number of segments is a hyperparameter. Locked segments correspond to asperities in fault mechanics, so our multi-patch model may be regarded as a reduced-order model for extracting robust information about the locked zone.
We analyze the data of onshore GNSS of GEONET and offshore GNSS-Acoustic (Yokota et al., 2016). We use a homogeneous half-space Green's function, and the fault geometry model follows the Japan Integrated Velocity Structure Model Version 1 (Koketsu et al., 2012). Green's function errors are accounted for by the method of Yagi & Fukahata (2011). We also check that using Green's function for a realistic 3D velocity model (Hori et al., 2021) does not much change the results of the homogeneous half-space model at least for the conventional slip deficit inversion. The number of asperities, the hyperparameter, is determined by the marginal likelihood maximization using the Bayesian information criterion (Schwarz, 1978).
Our optimal estimate has detected five asperities. Their along-strike sections, surprisingly, closely resemble the five basins of the Nankai subduction zone: Hyuga, Tosa, Muroto, Kumano, and Enshu, the envisioned five focal sections of the Nankai megathrust earthquake (Furumura et al., 2011; Saito & Noda, 2022). Our estimate also concludes the unlocking at the segmentation boundary of the 1944 Tonankai and 1946 Nankai earthquakes. Fittingly, this segmentation boundary is the neighborhood of the initiation zones of those earthquakes, which are supposed to be near the locked zone edges hosting the nuclei of those earthquakes (Hori et al., 2006). Observation points exist above that boundary, so its unlocking is likely. The estimated locked zones coincide with the previous focal zones of the same basins (Obara & Kato, 2016) in all basins but the Hyuga. The Hyuga locked zone includes the slip zones of the 1968 Hyuga-nada earthquake (Yagi et al., 1998) and the Bungo-channel long-term slow-slip events (Obara & Kato, 2016), though documented deep low-frequency tremors are all outside the locked zone estimate. Our results suggest that, on geodetic scales, the seismogenic zone of regular earthquakes is locked, whereas the slip zone of slow earthquakes at depth is coupled but unlocked. The only exception is the slip zone of the Bungo-channel slow-slip events, which may be the nucleation zone that often fails to slip faster, sometimes succeeding as supposedly in 1707 (Furumura et al., 2011).