9:30 AM - 9:45 AM
[SCG61-15] Crustal Strain Rate Distribution Derived from an Ultra-Dense GNSS Observation Network in Japan
Keywords:GNSS, Dense network, SoftBank Corp., Strain Rate, Crustal Deformation
The crustal strain field estimated using coordinate time series from the Geospatial Information Authority of Japan (GSI) GEONET has been reported in previous studies, such as Sagiya et al. (2000) and Okazaki et al. (2021). These studies have discussed the existence and distribution of inland strain concentration zones. However, the sensitivity of GEONET, with an average inter-station distance of approximately 20 km, to strain caused by individual faults and localized deformation is not sufficiently high. In contrast, the independent GNSS network operated by SoftBank (hereinafter referred to as SB) numbers over more than 3,300 nationwide, nearly three times of GEONET stations. This higher station density is expected to improve the spatial resolution in crustal deformation analysis.
The GNSS observation data provided within the framework of the "Consortium to utilize the SoftBank original reference sites for Earth and Space Science (CSESS)" (https://csess.jp/) are currently subjected to routine estimation of daily coordinate results by Tohoku University. At Tohoku University, daily coordinate values are estimated by applying the precise point positioning with ambiguity resolution (PPP-AR) method using GipsyX Ver. 2.2 from SoftBank's independent reference sites obtained through the CSESS framework and GEONET sites. Although the RINEX data included multi-GNSS observables, only GPS data was used for daily coordinate estimation. After removing offsets due to maintenance and earthquakes, the displacement was calculated as the difference between the median daily coordinate values from May 2–11, 2021, and those from April 24–May 2, 2022. Hierarchical clustering was performed separately for GEONET and SB to exclude stations exhibiting anomalous displacement distinct from surrounding areas, grouping stations into approximately 40 stations each. Outlier detection was then conducted within each cluster using the interquartile range method for east-west and north-south components.
The strain field was calculated using the Strain_2D software (Materna and Maurer, 2023), specifically employing the Velocity Interpolation for Strain Rate (VISR) algorithm (Shen et al., 2015). Compared to the conventional method by Shen et al. (1996), the VISR algorithm still requires the subjective determination of the hyperparameter Wt, but it improves upon previous methods by adjusting the distance-decaying constant (D) based on station density. The VISR algorithm allows for different weighting schemes, among which (i) Voronoi × Gaussian and (ii) Azimuth × Gaussian were used in this study.
Initially, strain rate calculations were conducted separately for GEONET and SB data using Wt = 20 and the (i) Voronoi × Gaussian weighting scheme. The distribution of the estimated D values exceeded 50 km in regions with sparse GEONET stations, such as Hokkaido and the edges of the observation network. In contrast, in relatively dense regions like Shizuoka, D was approximately 20 km, indicating a highly heterogeneous distribution. In contrast, for SB data alone, the D values were more uniformly distributed nationwide at approximately 20 km, reflecting a denser and more uniform observation network than GEONET. This difference in D distribution indicates that the denser the observation network, the smaller the D value, meaning that the range over which local uniformity is assumed becomes narrower. This confirms that the VISR algorithm is more appropriate when applied to denser station distributions.
Based on these findings, we proceeded with strain rate field estimation using two datasets, which provide a denser observation network than previous studies: (1) SB data alone and (2) a combined dataset of GEONET and SB data. The L-curve method was used to determine an appropriate Wt value for each dataset to avoid subjective selection of the hyperparameter Wt. The strain rate fields obtained using these optimized Wt values successfully captured shorter wavelength strain concentration compared to previous studies. In the presentation, we will further compare these results with information on microseismicity, volcanoes, and active faults to facilitate further discussions.
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".
The GNSS observation data provided within the framework of the "Consortium to utilize the SoftBank original reference sites for Earth and Space Science (CSESS)" (https://csess.jp/) are currently subjected to routine estimation of daily coordinate results by Tohoku University. At Tohoku University, daily coordinate values are estimated by applying the precise point positioning with ambiguity resolution (PPP-AR) method using GipsyX Ver. 2.2 from SoftBank's independent reference sites obtained through the CSESS framework and GEONET sites. Although the RINEX data included multi-GNSS observables, only GPS data was used for daily coordinate estimation. After removing offsets due to maintenance and earthquakes, the displacement was calculated as the difference between the median daily coordinate values from May 2–11, 2021, and those from April 24–May 2, 2022. Hierarchical clustering was performed separately for GEONET and SB to exclude stations exhibiting anomalous displacement distinct from surrounding areas, grouping stations into approximately 40 stations each. Outlier detection was then conducted within each cluster using the interquartile range method for east-west and north-south components.
The strain field was calculated using the Strain_2D software (Materna and Maurer, 2023), specifically employing the Velocity Interpolation for Strain Rate (VISR) algorithm (Shen et al., 2015). Compared to the conventional method by Shen et al. (1996), the VISR algorithm still requires the subjective determination of the hyperparameter Wt, but it improves upon previous methods by adjusting the distance-decaying constant (D) based on station density. The VISR algorithm allows for different weighting schemes, among which (i) Voronoi × Gaussian and (ii) Azimuth × Gaussian were used in this study.
Initially, strain rate calculations were conducted separately for GEONET and SB data using Wt = 20 and the (i) Voronoi × Gaussian weighting scheme. The distribution of the estimated D values exceeded 50 km in regions with sparse GEONET stations, such as Hokkaido and the edges of the observation network. In contrast, in relatively dense regions like Shizuoka, D was approximately 20 km, indicating a highly heterogeneous distribution. In contrast, for SB data alone, the D values were more uniformly distributed nationwide at approximately 20 km, reflecting a denser and more uniform observation network than GEONET. This difference in D distribution indicates that the denser the observation network, the smaller the D value, meaning that the range over which local uniformity is assumed becomes narrower. This confirms that the VISR algorithm is more appropriate when applied to denser station distributions.
Based on these findings, we proceeded with strain rate field estimation using two datasets, which provide a denser observation network than previous studies: (1) SB data alone and (2) a combined dataset of GEONET and SB data. The L-curve method was used to determine an appropriate Wt value for each dataset to avoid subjective selection of the hyperparameter Wt. The strain rate fields obtained using these optimized Wt values successfully captured shorter wavelength strain concentration compared to previous studies. In the presentation, we will further compare these results with information on microseismicity, volcanoes, and active faults to facilitate further discussions.
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".