3:45 PM - 4:00 PM
[SSS02-08] Real-time estimation of displacement, velocity, and tsunami height with ocean-bottom pressure data using a Kalman filter and an Ensemble Empirical Mode Decomposition
Keywords:Tsunami, real-time estimation
Dense and wide-span ocean-bottom observation networks such as DONET and S-net enable us to observe offshore earthquakes and tsunamis inside a coseismic area. At these networks, tsunamis and ground motions are designed to be recorded by collocated ocean-bottom pressure (OBP) gauges and strong-motion accelerometers. During an earthquake, however, tsunami components of OBP records are contaminated by non-tsunami components such as ocean-bottom displacements (e.g., Saito and Tsushima, 2016), and a simple integration of acceleration records leads to unreasonable velocity and displacement waveforms by the presence of unphysical drifts (e.g., Iwan et al., 1985).
In this study, we propose a new method to estimate coseismic displacement, velocity, and tsunami height on the real-time basis. This method is based on the use of a Kalman filter and an ensemble empirical mode decomposition (EEMD). We combine the independent records of OBP gauges and accelerometers with the Kalman filter to estimate velocity and displacement waveforms. This is analogous to studies on on-land sites by the combination of GNSS and accelerometer records (e.g., Bock et al., 2011). Tsunami height estimation with the EEMD is based on Wang et al. (2020) who used raw OBP data with 1 min resampling. On the other hand, our method uses 1-sec data which extracted tsunami and displacement components with the extraction method of Mizutani et al. (2020).
We applied our method to the DONET records of the 2016 Off-Mie earthquake (Mw 6.0) and the S-net records of the 2016 Off-Fukushima earthquake (Mw 7.4). In both events, even with very different sizes, we successfully estimated displacement, velocity and tsunami height waveforms (Figure 1).
To confirm the validity of our results, we made three comparisons (Figure 2). Figure 2a compares the residual displacements estimated by the 60-85 sec average of the resulted waveforms and by the OBP offset. Figure 2b compares of the absolute maximum tsunami height estimated from the EEMD and a band-pass filter of 100-500 sec. We also estimated the velocity magnitudes defined by Katsumata (2007) from our estimated velocity waveforms (Figure 2c). All the comparisons show that the present method will be useful for earthquake and tsunami early warnings because we need records of less than 2 min after origin times.
In this study, we propose a new method to estimate coseismic displacement, velocity, and tsunami height on the real-time basis. This method is based on the use of a Kalman filter and an ensemble empirical mode decomposition (EEMD). We combine the independent records of OBP gauges and accelerometers with the Kalman filter to estimate velocity and displacement waveforms. This is analogous to studies on on-land sites by the combination of GNSS and accelerometer records (e.g., Bock et al., 2011). Tsunami height estimation with the EEMD is based on Wang et al. (2020) who used raw OBP data with 1 min resampling. On the other hand, our method uses 1-sec data which extracted tsunami and displacement components with the extraction method of Mizutani et al. (2020).
We applied our method to the DONET records of the 2016 Off-Mie earthquake (Mw 6.0) and the S-net records of the 2016 Off-Fukushima earthquake (Mw 7.4). In both events, even with very different sizes, we successfully estimated displacement, velocity and tsunami height waveforms (Figure 1).
To confirm the validity of our results, we made three comparisons (Figure 2). Figure 2a compares the residual displacements estimated by the 60-85 sec average of the resulted waveforms and by the OBP offset. Figure 2b compares of the absolute maximum tsunami height estimated from the EEMD and a band-pass filter of 100-500 sec. We also estimated the velocity magnitudes defined by Katsumata (2007) from our estimated velocity waveforms (Figure 2c). All the comparisons show that the present method will be useful for earthquake and tsunami early warnings because we need records of less than 2 min after origin times.