17:15 〜 19:15
[HDS10-P11] Tsunami wave height prediction using data assimilation of submarine tsunami magnetic field data
キーワード：津波、津波磁場、データ同化、4DEnVAR
French Polynesia, including the Society, Marquesas, and Tuamotu Islands, faces significant tsunami risks from trans-Pacific tsunamis generated along the Pacific Ring of Fire. The 2010 Chilean tsunami demonstrated this risk. While Tahiti recorded a relatively small 28 cm wave in Papeete Harbor, the Marquesas Islands, 1,500 km northeast, experienced waves up to ~4 meters (Reymond et al., 2013). However, accurate tsunami simulation and forecasting remain challenging due to the region’s complex bathymetry, which results in significant nonlinear effects on the tsunami field, along with dispersion effects accumulating from distant tsunami sources.
This study aims to accurately model the tsunami field in French Polynesia during the 2010 Chilean tsunami using seafloor magnetic data and a novel data assimilation approach. During the event, nine ocean bottom electro-magnetometers (OBEMs) and a differential pressure gauge (DPG) recorded the tsunami's passage on the southeastern side of the Society Islands as part of the TIARES campaign (Suetsugu et al., 2012). OBEMs captured three-component magnetic field data at SOC1–9, while the DPG at SOC8 recorded pressure data. Tsunami-generated magnetic (TGM) fields arise from conductive seawater moving through Earth’s magnetic field (Tyler, 2005) and were detected at SOC1–9 (Sugioka et al., 2014). However, previous numerical simulations struggle to match observed tsunami heights and arrival times (Lin et al., 2021), and OBEM/DPG data have not been used to refine tsunami models for this event.
This study applies 4D ensemble variational (4DEnVar) data assimilation to predict tsunami heights at Papeete (PPT). Data assimilation integrates models with sparse observational data to improve predictions. The 4DEnVar method, formulated by Liu et al (2008), has advantages such as easy implementation without ajoint models and forward model linealizaiton, as well as accurate error estimation of initial conditions based on ensembles. While previous studies used optimal interpolation method with linear long wave equations (Maeda et al., 2015), no study has incorporated nonlinear effects or seafloor TGM data in assimilation studies.
Our approach follows two steps. In the first step, we build a new fault model with 12 subfaults for the 2010 Chilean event. The target region includes the TIARES array and PPT (Area 2). To reduce numerical complexity, we first conduct 12 tsunami simulations with unit slip on each sub-fault in a large domain (Area 1), which includes both the tsunami source and the TIARES area (Area 1). These simulations provide dynamic boundary conditions at the east and south edges of Area 2 as the Green’s functions, representing incoming tsunami waves. The simulations are performed assuming linear dispersive waves using the JAGURS code (Baba et al., 2017).
In the second step, we generate an ensemble of 20 fault models by adding Gaussian noise to the source model of Yoshimoto et al. (2016), which is converted to the ensemble of dynamic boundary conditions of Area2. With the boundary conditions, we then simulate nonlinear tsunami behavior and and TGM fields in Area 2 for the ensemble using JAGURS and the TMTGEM code (Minami et al., 2017), respectively. The 4DEnVar method is applied to the ensemble to obtain optimal weights of the Green functions based on the TGM magnetic data at SOC1-9. This step is iterated until the residuals of TGM data reduces enough.
The combination of step 1 and 2 balances computational efficiency by incorporating dispersion effects in Area 1 while allowing data assimilation including the nonlinear effects in Area 2. By integrating TGM data with nonlinear modeling, we aim to enhance tsunami prediction accuracy in French Polynesia. In the presentation, we plan to describe our methodology and evaluate the performance of our method by comparing the simulated tsunami with pressure gauge record at PPT, located near the northwestern corner of Area 2.
This study aims to accurately model the tsunami field in French Polynesia during the 2010 Chilean tsunami using seafloor magnetic data and a novel data assimilation approach. During the event, nine ocean bottom electro-magnetometers (OBEMs) and a differential pressure gauge (DPG) recorded the tsunami's passage on the southeastern side of the Society Islands as part of the TIARES campaign (Suetsugu et al., 2012). OBEMs captured three-component magnetic field data at SOC1–9, while the DPG at SOC8 recorded pressure data. Tsunami-generated magnetic (TGM) fields arise from conductive seawater moving through Earth’s magnetic field (Tyler, 2005) and were detected at SOC1–9 (Sugioka et al., 2014). However, previous numerical simulations struggle to match observed tsunami heights and arrival times (Lin et al., 2021), and OBEM/DPG data have not been used to refine tsunami models for this event.
This study applies 4D ensemble variational (4DEnVar) data assimilation to predict tsunami heights at Papeete (PPT). Data assimilation integrates models with sparse observational data to improve predictions. The 4DEnVar method, formulated by Liu et al (2008), has advantages such as easy implementation without ajoint models and forward model linealizaiton, as well as accurate error estimation of initial conditions based on ensembles. While previous studies used optimal interpolation method with linear long wave equations (Maeda et al., 2015), no study has incorporated nonlinear effects or seafloor TGM data in assimilation studies.
Our approach follows two steps. In the first step, we build a new fault model with 12 subfaults for the 2010 Chilean event. The target region includes the TIARES array and PPT (Area 2). To reduce numerical complexity, we first conduct 12 tsunami simulations with unit slip on each sub-fault in a large domain (Area 1), which includes both the tsunami source and the TIARES area (Area 1). These simulations provide dynamic boundary conditions at the east and south edges of Area 2 as the Green’s functions, representing incoming tsunami waves. The simulations are performed assuming linear dispersive waves using the JAGURS code (Baba et al., 2017).
In the second step, we generate an ensemble of 20 fault models by adding Gaussian noise to the source model of Yoshimoto et al. (2016), which is converted to the ensemble of dynamic boundary conditions of Area2. With the boundary conditions, we then simulate nonlinear tsunami behavior and and TGM fields in Area 2 for the ensemble using JAGURS and the TMTGEM code (Minami et al., 2017), respectively. The 4DEnVar method is applied to the ensemble to obtain optimal weights of the Green functions based on the TGM magnetic data at SOC1-9. This step is iterated until the residuals of TGM data reduces enough.
The combination of step 1 and 2 balances computational efficiency by incorporating dispersion effects in Area 1 while allowing data assimilation including the nonlinear effects in Area 2. By integrating TGM data with nonlinear modeling, we aim to enhance tsunami prediction accuracy in French Polynesia. In the presentation, we plan to describe our methodology and evaluate the performance of our method by comparing the simulated tsunami with pressure gauge record at PPT, located near the northwestern corner of Area 2.