Group B, Poster #010, Seismology

Tsunami source imaging using adjoint-state inversion of data from ocean bottom pressure gauges

SAEED Y. MOHANNA, Yuqing Xie, & Lingsen Meng
Poster Image: 

Poster Presentation

2022 SCEC Annual Meeting, Poster #010, SCEC Contribution #12021 VIEW PDF
Current frameworks for issuing rapid tsunami warnings rely on point-source models that do not take into account slip distribution or the rupture area and thus can be inaccurate in the immediate aftermath of an earthquake, leading to inaccurate wave height and wave arrival time predictions. In response to the underestimation of tsunami wave height during the 2011 Tohoku tsunami, the Japanese Meteorological Agency constructed the Seafloor Observation Network for Earthquakes and Tsunamis along the Japan Trench, a dense data network of ocean bottom pressure gauges (OBPG) to avoid reliance on accurate earthquake models in order to issue a tsunami warning. Our group has taken advantage of this by developing a method that inverts OBPG data using the adjoint-state method to yield the tsunami source directly, thus eliminating the reliance on an accurate earthquake model and reducing the computation time that would otherwise be spent calculating Green’s functions in a densely gridded source region. We tested the adjoint-state method on synthetic waveforms and OBPG data of the 2011 Mw 9.0 Tohoku and 2016 Mw 6.9 Off-Fukushima earthquakes, respectively. The results from these tests show that this method can yield wave-height predictions with an average accuracy of 93% and 78% using the first 5 and 25 minutes of the synthetic and observed OBPG data, respectively. We also sought to investigate whether our method can invert an earthquake’s rupture speed and consider its effect on our ability to predict tsunami waves. Most tsunami simulations assume a time-independent source model, where the sea-floor deforms all at once. However, this can be inaccurate for large earthquakes with slow rupture speed leading to propagation of the deformation over time, creating a time-dependent source. Consequently, we first tested our method using synthetic data of a tsunami sourced near the region of the 2011 Tohoku event, in which we mimic a scenario where a rupture is triggered 200 s after the origin time, as suggested by previous studies. Our results when using 5-10 mins of synthetic OBPG data yielded wave-height predictions that were similar to that of the synthetic data, with variance reductions converging to 0.95 after 7 iterations of the inversion in most cases. We plan to conduct optimization analysis of our method on the Peru-Chile trench in order to obtain the best hypothetical configuration and number of stations that yield accurate tsunami predictions.