11:00 AM - 11:15 AM
[HDS10-02] Impact of dense offshore tsunami measurements on tsunami source and propagation modeling: case study of 2016 Mw 6.9 off Fukushima earthquake, Japan
Keywords:Tsunami, Tsunami source, Tsunami propagation, Tsunami decay
Background and purpose
A dense ocean-bottom observation network along the Japan Trench, called S-net (Aoi et al., 2020, EPS), has been in operation since 2016. On November 22, 2016, an Mw 6.9 intraplate earthquake occurred off Fukushima Prefecture, Japan, and the resultant tsunami was observed by many ocean-bottom pressure gauges (OBPGs) of S-net. Many studies have investigated the source and propagation characteristics of this tsunami (e.g., Gusman et al., 2017, Pageoph). Kubota et al. (2021, JGR) showed that the S-net pressure waveforms can constrain the earthquake fault model better than the coastal tsunami waveforms, and the model can reproduce the nearshore tsunami waveforms. In this study, we focus on the following two points: (1) How well can S-net data constrain initial tsunami height distribution, compared to coastal data? (2) How well can the source inverted from S-net data reproduce tsunami later phases and decay in the near-field coast?
Data
The pressure waveforms recorded at 36 S-net OBPGs and sea-surface height waveforms at a GPS buoy were used as offshore tsunami data. The sea-level waveforms at 9 wave gauges and 11 tide gauges in the near-field coast were used as the coastal tsunami data. Ocean tide and short-period components were removed to extract tsunami signals. According to Kubota et al. (2021, JGR), the coseismic pressure data of the S-net contain steps and drifts which are not related to the crustal deformation and tsunami. To reduce the noise, we estimate linear trend from the 1-hour pressure waveform after the earthquake and then subtract it from the detided pressure records, because the noises seem to be approximated as linear trend in the time window of our source inversion.
Method
We conducted linear inverse analysis in a Bayesian framework (Yabuki and Matsu'ura, 1992, GJI) to estimate initial tsunami height distribution from the observed waveforms of the first tsunamis. Then, the resultant tsunamis were calculated by numerical simulation based on nonlinear long wave theory. To evaluate the uncertainty of the calculated waveforms at the coast, we adopted ensemble approach proposed by Takagawa and Tomita (2015, JJSCE). For the numerical modeling of tsunami propagation, we used JAGURS (Baba et al., 2015, Pageoph). High-resolution bathymetric data (the finest resolution is 50 m) was used to reproduce tsunami behavior around the coastal tide gauges.
Results and Discussion
We compared the tsunami source model inverted from the offshore data and that from the coastal data, and found the insights that the dense offshore data is effective for the source estimation in terms of the model covariance and resolution, and the validity of the linear approximation in tsunami propagation. Here, we will show the validity of the linear approximation. The two source models have commonly the significant subsidence area whose extent is elongated in the northeast-southwest direction, but the source based on the coastal data has the additional subsidence area to the south and northwest. As a result of our synthetic test assuming this earthquake, we found that the additional subsidence area is probably the source artifacts due to the strong nonlinearity in the waveforms of the first tsunamis at the coastal tide gauges.
The results of nonlinear tsunami calculations using the source based on the offshore data showed that the waveforms of the first tsunamis at the coast were well reproduced even though they were not used in the inversion. The maximum waves that were observed within ~4 hours after the earthquake were well reproduced at most coastal stations, but at some stations the later phases, which are the maximum waves in some cases, could not be reproduced. As for the decay process within ~15 hours after the earthquake, it is well reproduced at some stations, but not at some stations. Considering that the source is well constrained by dense offshore observations and that the difference between observed waveforms and calculated ones exceeds the uncertainty due to the source estimation error, there will be room for improvement in modeling of tsunami propagation such as bathymetry and bottom friction.
Acknowledgments: This work was partially supported by JSPS KAKENHI Grant Number JP19H02409.
A dense ocean-bottom observation network along the Japan Trench, called S-net (Aoi et al., 2020, EPS), has been in operation since 2016. On November 22, 2016, an Mw 6.9 intraplate earthquake occurred off Fukushima Prefecture, Japan, and the resultant tsunami was observed by many ocean-bottom pressure gauges (OBPGs) of S-net. Many studies have investigated the source and propagation characteristics of this tsunami (e.g., Gusman et al., 2017, Pageoph). Kubota et al. (2021, JGR) showed that the S-net pressure waveforms can constrain the earthquake fault model better than the coastal tsunami waveforms, and the model can reproduce the nearshore tsunami waveforms. In this study, we focus on the following two points: (1) How well can S-net data constrain initial tsunami height distribution, compared to coastal data? (2) How well can the source inverted from S-net data reproduce tsunami later phases and decay in the near-field coast?
Data
The pressure waveforms recorded at 36 S-net OBPGs and sea-surface height waveforms at a GPS buoy were used as offshore tsunami data. The sea-level waveforms at 9 wave gauges and 11 tide gauges in the near-field coast were used as the coastal tsunami data. Ocean tide and short-period components were removed to extract tsunami signals. According to Kubota et al. (2021, JGR), the coseismic pressure data of the S-net contain steps and drifts which are not related to the crustal deformation and tsunami. To reduce the noise, we estimate linear trend from the 1-hour pressure waveform after the earthquake and then subtract it from the detided pressure records, because the noises seem to be approximated as linear trend in the time window of our source inversion.
Method
We conducted linear inverse analysis in a Bayesian framework (Yabuki and Matsu'ura, 1992, GJI) to estimate initial tsunami height distribution from the observed waveforms of the first tsunamis. Then, the resultant tsunamis were calculated by numerical simulation based on nonlinear long wave theory. To evaluate the uncertainty of the calculated waveforms at the coast, we adopted ensemble approach proposed by Takagawa and Tomita (2015, JJSCE). For the numerical modeling of tsunami propagation, we used JAGURS (Baba et al., 2015, Pageoph). High-resolution bathymetric data (the finest resolution is 50 m) was used to reproduce tsunami behavior around the coastal tide gauges.
Results and Discussion
We compared the tsunami source model inverted from the offshore data and that from the coastal data, and found the insights that the dense offshore data is effective for the source estimation in terms of the model covariance and resolution, and the validity of the linear approximation in tsunami propagation. Here, we will show the validity of the linear approximation. The two source models have commonly the significant subsidence area whose extent is elongated in the northeast-southwest direction, but the source based on the coastal data has the additional subsidence area to the south and northwest. As a result of our synthetic test assuming this earthquake, we found that the additional subsidence area is probably the source artifacts due to the strong nonlinearity in the waveforms of the first tsunamis at the coastal tide gauges.
The results of nonlinear tsunami calculations using the source based on the offshore data showed that the waveforms of the first tsunamis at the coast were well reproduced even though they were not used in the inversion. The maximum waves that were observed within ~4 hours after the earthquake were well reproduced at most coastal stations, but at some stations the later phases, which are the maximum waves in some cases, could not be reproduced. As for the decay process within ~15 hours after the earthquake, it is well reproduced at some stations, but not at some stations. Considering that the source is well constrained by dense offshore observations and that the difference between observed waveforms and calculated ones exceeds the uncertainty due to the source estimation error, there will be room for improvement in modeling of tsunami propagation such as bathymetry and bottom friction.
Acknowledgments: This work was partially supported by JSPS KAKENHI Grant Number JP19H02409.