3:30 PM - 3:45 PM
[SSS10-13] Study on validation process of simulation data to supplement the strong-motion database
Keywords:strong-motion database, ground-motion simulation
We have been developing a strong-motion observation record database as an infrastructural database utilized for seismic hazard assessment (Morikawa et al. JpGU 2022), from which data-driven ground-motion models (GMMs) are to be constructed. GMM is incapable of predicting phenomena that do not exist in the database. This is a critical issue because the time period range considered in seismic hazard assessment is far longer than that covered in the observation record database. To overcome this problem, we have started an attempt to construct a simulation dataset to supplement the observation database (Iwaki et al. JpGU 2021).
The performance of the observation database is investigated in discretized magnitude-distance space (M-X space) (Tomozawa et al. JpGU 2022). In order to supplement the observation database by a simulation dataset while maintaining the characteristics of the observation database, we set two steps to deal with. The first step is to validate the simulation data. The second step is to evaluate the performance of the unified database supplemented by the simulation dataset.
For the first step, we analyzed the frequency distribution of the simulation dataset in M-X space to investigate how to validate the amount and quality of the dataset. We have made a simulation dataset using active fault models in the Kanto region from the National Seismic Hazard Map (NSHM; HERP, 2020). Simulation method follows Iwaki et al. (2016) in which short-period (< 1s) component is computed using site-specific empirical characteristics of acceleration envelopes and long-period component is computed by 3D FDM.
In order to validate the frequency distribution of the simulation data in M-X space, we first evaluated the robustness of the dataset by bootstrap sampling, then compared the distribution with that of the observation.
We also compared the dataset with the original NSHM simulation data (PGV on the ground surface) to investigate the difference of the simulation methods. There are two types of NSHM simulation data, hybrid simulation (3D FDM and stochastic Green’s function) and GMM. Since our simulation data reflects the empirical ground-motion characteristics at each site, within-event variability was generally the largest compared with the other two methods.
Our future plan is to increase the amount of the simulation data for different types of earthquakes and proceed with the validation process.
Acknowledgement: This study is supported by JSPS KAKENHI 20H00292.
The performance of the observation database is investigated in discretized magnitude-distance space (M-X space) (Tomozawa et al. JpGU 2022). In order to supplement the observation database by a simulation dataset while maintaining the characteristics of the observation database, we set two steps to deal with. The first step is to validate the simulation data. The second step is to evaluate the performance of the unified database supplemented by the simulation dataset.
For the first step, we analyzed the frequency distribution of the simulation dataset in M-X space to investigate how to validate the amount and quality of the dataset. We have made a simulation dataset using active fault models in the Kanto region from the National Seismic Hazard Map (NSHM; HERP, 2020). Simulation method follows Iwaki et al. (2016) in which short-period (< 1s) component is computed using site-specific empirical characteristics of acceleration envelopes and long-period component is computed by 3D FDM.
In order to validate the frequency distribution of the simulation data in M-X space, we first evaluated the robustness of the dataset by bootstrap sampling, then compared the distribution with that of the observation.
We also compared the dataset with the original NSHM simulation data (PGV on the ground surface) to investigate the difference of the simulation methods. There are two types of NSHM simulation data, hybrid simulation (3D FDM and stochastic Green’s function) and GMM. Since our simulation data reflects the empirical ground-motion characteristics at each site, within-event variability was generally the largest compared with the other two methods.
Our future plan is to increase the amount of the simulation data for different types of earthquakes and proceed with the validation process.
Acknowledgement: This study is supported by JSPS KAKENHI 20H00292.