11:00 AM - 1:00 PM
[SSS13-P03] Estimation of clock time error of seismometer based on extended Kalman filter using ambient noise cross-correlation function: an application to Kirishima volcanoes
Keywords:ambient noise, seismic interferometry, clock time correction
Monitoring the edifice of volcanoes is crucial to understanding the functioning of an active magmatic system. Noise-based passive seismic monitoring is a powerful tool to detect slight temporal changes in the mechanical properties of a volcano. However, seismic velocity variations have sensitive to large earthquakes and environmental changes such as precipitation, which could mask the seismic velocity variations related to volcanic activities. The clock time errors of seismometers also affect the temporal changes in ambient noise cross-correlations. The instrumental time error has been estimated by using ambient noise cross-correlations (e.g., Stehly et al., 2007; Sens-Schönfelder, 2008). This work proposes a method to detect time differences caused by instrumental time error based on an extended Kalman filter using ambient noise cross-correlation functions following Nishida et al. (2020).
We determine the time shifts caused by clock time error from the daily stacked ambient noise cross-correlations with an extended Kalman filter based on a state-space model. In the Kalman filter processing, we minimized the squared differences between the reference and observed noise cross-correlations. A model function of the observed correlations is expressed by changing the amplitude, stretching factor, and time shifts due to clock time error. The time shift is equal for all lapse times. The reference correlation function at each station is computed by averaging for the observation duration. We estimate the state vector, which represents the amplitude, the stretching factor, and the time shift through the extended Kalman filter, changing an initial stretching factor and time shift and a prior model covariance as hyperparameters. These hyperparameters are estimated by a maximum likelihood method.
We apply this method to three-component seismic records at Kirishima volcanoes. We use the seismic stations deployed by the Earthquake Research Institute (ERI), the University of Tokyo, and the National Research Institute for Earth Science and Disaster prevention (NIED). We first calculate the ambient noise cross-correlations from May 2010 to September 2021. We then estimate time differences between two stations caused by clock time errors of seismometers. For the pair of stations containing SMW, a step change of about 0.1 seconds was seen near January 2014, and for the pair of stations containing SMN, a change of about 0.05 seconds was seen in the summer of 2015. For the station pair containing KVO, a slight change was observed near January 2011. The timing of clock time error at SMN and KVO seems to be consistent with the date when the seismometers were modified.
We thank the Volcano Research Center of ERI and NIED for providing continuous seismic records.
We determine the time shifts caused by clock time error from the daily stacked ambient noise cross-correlations with an extended Kalman filter based on a state-space model. In the Kalman filter processing, we minimized the squared differences between the reference and observed noise cross-correlations. A model function of the observed correlations is expressed by changing the amplitude, stretching factor, and time shifts due to clock time error. The time shift is equal for all lapse times. The reference correlation function at each station is computed by averaging for the observation duration. We estimate the state vector, which represents the amplitude, the stretching factor, and the time shift through the extended Kalman filter, changing an initial stretching factor and time shift and a prior model covariance as hyperparameters. These hyperparameters are estimated by a maximum likelihood method.
We apply this method to three-component seismic records at Kirishima volcanoes. We use the seismic stations deployed by the Earthquake Research Institute (ERI), the University of Tokyo, and the National Research Institute for Earth Science and Disaster prevention (NIED). We first calculate the ambient noise cross-correlations from May 2010 to September 2021. We then estimate time differences between two stations caused by clock time errors of seismometers. For the pair of stations containing SMW, a step change of about 0.1 seconds was seen near January 2014, and for the pair of stations containing SMN, a change of about 0.05 seconds was seen in the summer of 2015. For the station pair containing KVO, a slight change was observed near January 2011. The timing of clock time error at SMN and KVO seems to be consistent with the date when the seismometers were modified.
We thank the Volcano Research Center of ERI and NIED for providing continuous seismic records.