2:30 PM - 2:45 PM
[MTT37-04] Estimation of a coseismic slip distribution model considering self-similarity of the 2024 Noto Peninsula earthquake using very dense geodetic observation
Keywords:von Karman regularization, Slip distribution estimation, The 2024 Noto Peninsula earthquake, Bayesian optimization
The Mw 7.5 Noto Peninsula earthquake on January 1, 2024, caused significant damage to the Noto Peninsula and surrounding regions due to strong ground motion, large-scale crustal deformation, and tsunami inundation. Yamada et al. (EPS, 2025) estimated the coseismic slip distribution using a dense geodetic observation network that integrated GEONET data, GNSS observations from SoftBank Corporation and temporary stations installed by Kyoto University and Kanazawa University to capture pre-mainshock swarm activity. They also utilized line-of-sight (LOS) displacement data from ALOS-2/PALSAR-2 via the Pixel Offset method, applying a Laplacian smoothing constraint common in slip estimations. The unprecedented density of this network offers the potential for resolving a more detailed slip distribution. Based on this background, this study applies an alternative regularization approach based on the self-similarity of slips to analyze this event and explore the complexity of the slip distribution.
The six fault planes were assumed with reference to the NT2–10 fault model from the Japan Sea Project. We adopted slip amounts and rake angles for three offshore faults estimated by Fujii and Satake (2024) from surface displacement and tsunami data. The remaining three faults near the peninsula were subdivided into approximately 2 km × 2 km subfaults, and both slip components were evaluated.
We employed Bayesian optimization to estimate the slip distribution. As a prior, we used a regularization based on the slip self-similarity, implemented via a von Karman-type autocorrelation function. The correlation length and the regularization strength were simultaneously estimated with the slip amounts. These parameters were assumed to be independent for each of the three faults and for both slip components. We used the No-U-Turn Sampler to enhance sampling efficiency for this strongly nonlinear, high-dimensional problem.
The results revealed a slip distribution with two peaks (∼8 m) exhibiting significant reverse fault slips located near the eastern and western parts of the northern coast of the peninsula. A right-lateral slip component with short-wavelength variation in the strike direction was estimated on the fault containing the maximum slip. In contrast, the estimated correlation lengths were approximately 5 km and 11 km along the strike and dip directions, respectively, indicating substantial anisotropy consistent with a model exhibiting high roughness along the strike.
Furthermore, the estimation using only GEONET data showed that the two peaks were unresolved; the slip appeared smoothed along the strike and concentrated at relatively shallow depths. The estimated correlation lengths for the strike-slip were approximately 17 km and 6 km along the strike and dip directions, respectively, aligning with a model characterized by long- and short-wavelength variations along the strike and the dip direction, respectively. These results suggest that the correlation lengths estimated alongside the slip distribution reflect both the properties of the coseismic slip and the slip resolution imposed by the observational network. Hence, evaluating the correlation lengths simultaneously is preferable rather than assuming them a priori.
We will quantitatively evaluate the assumed slip on the offshore faults by incorporating additional SoftBank observation points not used in the current inversion. Moreover, kinematic analysis will allow a more detailed discussion of the coseismic slip.
Acknowledgments: The SoftBank's GNSS observation data used in this study was provided by SoftBank Corp. and ALES Corp. through the framework of the "Consortium to utilize the SoftBank original reference sites for Earth and Space Science". We used open ALOS-2 data through JAXA EORC web site. The RINC software, which was developed by Dr. Taku Ozawa and provided through PIXEL, was used for ALOS-2 data processing.
The six fault planes were assumed with reference to the NT2–10 fault model from the Japan Sea Project. We adopted slip amounts and rake angles for three offshore faults estimated by Fujii and Satake (2024) from surface displacement and tsunami data. The remaining three faults near the peninsula were subdivided into approximately 2 km × 2 km subfaults, and both slip components were evaluated.
We employed Bayesian optimization to estimate the slip distribution. As a prior, we used a regularization based on the slip self-similarity, implemented via a von Karman-type autocorrelation function. The correlation length and the regularization strength were simultaneously estimated with the slip amounts. These parameters were assumed to be independent for each of the three faults and for both slip components. We used the No-U-Turn Sampler to enhance sampling efficiency for this strongly nonlinear, high-dimensional problem.
The results revealed a slip distribution with two peaks (∼8 m) exhibiting significant reverse fault slips located near the eastern and western parts of the northern coast of the peninsula. A right-lateral slip component with short-wavelength variation in the strike direction was estimated on the fault containing the maximum slip. In contrast, the estimated correlation lengths were approximately 5 km and 11 km along the strike and dip directions, respectively, indicating substantial anisotropy consistent with a model exhibiting high roughness along the strike.
Furthermore, the estimation using only GEONET data showed that the two peaks were unresolved; the slip appeared smoothed along the strike and concentrated at relatively shallow depths. The estimated correlation lengths for the strike-slip were approximately 17 km and 6 km along the strike and dip directions, respectively, aligning with a model characterized by long- and short-wavelength variations along the strike and the dip direction, respectively. These results suggest that the correlation lengths estimated alongside the slip distribution reflect both the properties of the coseismic slip and the slip resolution imposed by the observational network. Hence, evaluating the correlation lengths simultaneously is preferable rather than assuming them a priori.
We will quantitatively evaluate the assumed slip on the offshore faults by incorporating additional SoftBank observation points not used in the current inversion. Moreover, kinematic analysis will allow a more detailed discussion of the coseismic slip.
Acknowledgments: The SoftBank's GNSS observation data used in this study was provided by SoftBank Corp. and ALES Corp. through the framework of the "Consortium to utilize the SoftBank original reference sites for Earth and Space Science". We used open ALOS-2 data through JAXA EORC web site. The RINC software, which was developed by Dr. Taku Ozawa and provided through PIXEL, was used for ALOS-2 data processing.