5:15 PM - 7:15 PM
[HDS10-P08] Calculation of the Tsunami Caused by the 2010 Chile Earthquake Based on Tide Gauge Records Along the Japanese Coast
Keywords:Chile earthquake, Tsunami
A massive tsunami was generated by the M8.8 Chile earthquake that occurred on February 27, 2010, and it traveled across the Pacific Ocean, reaching the coast of Japan. In Japan, a tsunami warning was issued, urging coastal residents to evacuate. The actual arrival time of the tsunami was approximately 20 to 30 minutes later than predicted. Subsequent research indicated that this discrepancy could be resolved by considering the Earth's elastic deformation. This was confirmed by comparing the observed tsunami recorded by deep-sea bottom pressure gauges with the computed tsunami. However, there has not been sufficient investigation into the reproducibility of tsunamis in coastal areas, which is crucial for mitigating tsunami damage.
This study examined the reproducibility of tsunami waveforms observed at Japanese tide gauge stations for the tsunami generated by the 2010 Chile earthquake. Specifically, we compared two numerical simulation methods and evaluated them regarding arrival time, maximum tsunami height, and computational cost.
One method applied a phase correction by Watada et al. (2014) to waveforms obtained using nonlinear long-wave equations, while the other method employed a numerical model (JAGURS, Baba et al., 2017) that accounts for wave dispersion and the Earth's elastic deformation. The simulation domain covered the entire Pacific Ocean, and the fault model proposed by Yoshimoto et al. (2016) was adopted. The tsunami was simulated for 40 hours following the earthquake.
Observational data from six tide gauge stations in Japan (Hanasaki, Ofunato, Ayukawa, Mera, Okada, and Awayuki) were used for validation. Bathymetric data resolutions of 81 arc-seconds, 27 arc-seconds, 9 arc-seconds, 3 arc-seconds, and 1 arc-second were employed, and four different nesting models were examined: two-layer nesting with a minimum grid of 27 arc-seconds, three-layer nesting with a minimum grid of 9 arc-seconds, four-layer nesting with a minimum grid of 3 arc-seconds, and five-layer nesting with a minimum grid of 1 arc-second.
Computations were performed using the Fujitsu Supercomputer PRIMEHPC FX1000 and the Fujitsu Server PRIMERGY GX2570 (Wisteria/BDEC-01). The two-layer to four-layer nesting models were executed using 16 nodes, while the five-layer nesting model was run using 32 nodes (with two threads per node).
The analysis results showed that, for both methods, considering the Earth's elastic deformation nearly eliminated the discrepancy in tsunami arrival times. At four of the observation points, excluding Mera and Awayuki, the maximum tsunami height was reproduced within ±0.2 m. At Mera and Awayuki, the maximum tsunami height was underestimated compared to the observed waveforms.
Increasing the topographic resolution improved the accuracy of initial waveform reproduction by better-capturing reflections within harbors and enhancing the precision of observation point locations. However, it did not result in significant changes to the maximum tsunami height or arrival time. Furthermore, in the five-layer nesting model with a minimum grid of 1 arc-second, the computation did not finish before the tsunami arrival time.
In conclusion, no significant differences were observed between the two methods in terms of tsunami arrival time and maximum tsunami height. The quantitative reproduction accuracy showed an arrival time discrepancy of approximately 3–4 minutes, with a geometric mean (K) of 1.25 and a geometric standard deviation (κ) of 0.18 for maximum tsunami height. However, in terms of computational cost, the phase correction method was more effective for real-time prediction, as it only required precomputed waveforms without considering dispersion and Earth's elastic deformation, making it more practical than the JAGURS-based method.
Future work will focus on developing methods to improve the accuracy of tsunami warnings.
This study examined the reproducibility of tsunami waveforms observed at Japanese tide gauge stations for the tsunami generated by the 2010 Chile earthquake. Specifically, we compared two numerical simulation methods and evaluated them regarding arrival time, maximum tsunami height, and computational cost.
One method applied a phase correction by Watada et al. (2014) to waveforms obtained using nonlinear long-wave equations, while the other method employed a numerical model (JAGURS, Baba et al., 2017) that accounts for wave dispersion and the Earth's elastic deformation. The simulation domain covered the entire Pacific Ocean, and the fault model proposed by Yoshimoto et al. (2016) was adopted. The tsunami was simulated for 40 hours following the earthquake.
Observational data from six tide gauge stations in Japan (Hanasaki, Ofunato, Ayukawa, Mera, Okada, and Awayuki) were used for validation. Bathymetric data resolutions of 81 arc-seconds, 27 arc-seconds, 9 arc-seconds, 3 arc-seconds, and 1 arc-second were employed, and four different nesting models were examined: two-layer nesting with a minimum grid of 27 arc-seconds, three-layer nesting with a minimum grid of 9 arc-seconds, four-layer nesting with a minimum grid of 3 arc-seconds, and five-layer nesting with a minimum grid of 1 arc-second.
Computations were performed using the Fujitsu Supercomputer PRIMEHPC FX1000 and the Fujitsu Server PRIMERGY GX2570 (Wisteria/BDEC-01). The two-layer to four-layer nesting models were executed using 16 nodes, while the five-layer nesting model was run using 32 nodes (with two threads per node).
The analysis results showed that, for both methods, considering the Earth's elastic deformation nearly eliminated the discrepancy in tsunami arrival times. At four of the observation points, excluding Mera and Awayuki, the maximum tsunami height was reproduced within ±0.2 m. At Mera and Awayuki, the maximum tsunami height was underestimated compared to the observed waveforms.
Increasing the topographic resolution improved the accuracy of initial waveform reproduction by better-capturing reflections within harbors and enhancing the precision of observation point locations. However, it did not result in significant changes to the maximum tsunami height or arrival time. Furthermore, in the five-layer nesting model with a minimum grid of 1 arc-second, the computation did not finish before the tsunami arrival time.
In conclusion, no significant differences were observed between the two methods in terms of tsunami arrival time and maximum tsunami height. The quantitative reproduction accuracy showed an arrival time discrepancy of approximately 3–4 minutes, with a geometric mean (K) of 1.25 and a geometric standard deviation (κ) of 0.18 for maximum tsunami height. However, in terms of computational cost, the phase correction method was more effective for real-time prediction, as it only required precomputed waveforms without considering dispersion and Earth's elastic deformation, making it more practical than the JAGURS-based method.
Future work will focus on developing methods to improve the accuracy of tsunami warnings.