11:00 AM - 1:00 PM
[SGD01-P04] Estimation of slip distribution of slow slip event by joint inversion of several diffrent types of data
Keywords:Slow slip events, Joint inversion of several diffrent types of data, Detection of data changes
Several types of slow earthquakes commonly occur along the Nankai Trough subduction zone of the Philippine Sea plate southwest of Japan. Among these, short-term slow slip events (SSEs) with durations of several days have been detected from records of borehole strainmeters (Kobayashi et al. 2006) and tiltmeters (Obara et al. 2004; Hirose and Obara 2005). These observational data have been used in rectangular fault models with uniform slip to analyze SSEs. Alternatively, Tsuyuki et al. (2021) developed a joint inversion of strain and tilt data using ABIC (Akaike 1980) and estimated slip distributions on the plate boundary of short-term SSEs. This method can be used not only for strain and tilt data but also for crustal movement data such as GNSS. Tsuyuki et al. (JpGU 2020) estimated the slip distribution of long-term SSE in Bungo channel during 2018 - 2019 by using of joint inversion of strain and GNSS data and showed that the change in strain data in this period is considered to be due to long-term SSE.
In this presentation, the following two points were examined in order to improve the method of joint inversion of several types of crustal deformation data.
1. How to arrange the sub-faults used as the basis function of the slip distribution when estimating the slip distribution on the plate interface.
2. A method for detecting changes due to slow slip events in observation data, especially data like strain data with high resolution but with large variance for a long period of time.
Regarding #1, sub-faults used for estimating slip distribution are rectangular faults and they are arranged so as to be consistent with the plate contour data. If a triangular fault is used instead of a rectangular fault, It will be possible to be closer to the shape of the plate interface. The effect of using a triangular fault was examined in this study. Green's function matrix due to a triangular sub-fault slip is computed by the method of calculating surface deformation in a homogeneous elastic half-space due to a triangular dislocation (e.g. Meade 2007; Nikkhoo and R. Walter 2015), which was compared with the results using rectangular sub-faults. Also, when calculating the prior probability density function of the slip distribution, slip distribution is supposed to be a smooth distribution, which is formulated using discrete Laplacian. For the method of formulation when a triangular sub-fault were used as a sub-fault, Maerten et al. (2005) was examined.
Regarding #2, in order to detect changes due to slow slip events in observation data, change in trend is generally detected. Strain data, which the presenter mainly deals with, is sensitive to seasonal factors or long-term effects of precipitation, and it is difficult to objectively estimate what the trend will be in the steady state (the period when no crustal movement due to slow slip is occured on the plate interface). I examined a detection method using a state-space model (JpGU 2019), but did not go so far as to objectively estimate trends. In machine learning, the k-nearest neighbor method is used as a method for detecting anomalies in time-series data. However, since crustal movement data contains many local changes, such anomaly detection is not sufficient to detect the change due to slip on the plate interface. On the other hand, stacking method of strain data (Miyaoka and Yokota 2012) stacks the data assuming a slip on a certain subfault, so it is possible to detect the start of the chnage and the location of the slip at the same time (Tsuyuki et al. 2017). When dealing with crustal movement data, it is considered to be appropriate to use the detection method that estimates from the data of the entire observation network. Therefore, I examined whether the network inversion filter method (Segall and Matthews 1997) could be applied to strain data.
In this presentation, the following two points were examined in order to improve the method of joint inversion of several types of crustal deformation data.
1. How to arrange the sub-faults used as the basis function of the slip distribution when estimating the slip distribution on the plate interface.
2. A method for detecting changes due to slow slip events in observation data, especially data like strain data with high resolution but with large variance for a long period of time.
Regarding #1, sub-faults used for estimating slip distribution are rectangular faults and they are arranged so as to be consistent with the plate contour data. If a triangular fault is used instead of a rectangular fault, It will be possible to be closer to the shape of the plate interface. The effect of using a triangular fault was examined in this study. Green's function matrix due to a triangular sub-fault slip is computed by the method of calculating surface deformation in a homogeneous elastic half-space due to a triangular dislocation (e.g. Meade 2007; Nikkhoo and R. Walter 2015), which was compared with the results using rectangular sub-faults. Also, when calculating the prior probability density function of the slip distribution, slip distribution is supposed to be a smooth distribution, which is formulated using discrete Laplacian. For the method of formulation when a triangular sub-fault were used as a sub-fault, Maerten et al. (2005) was examined.
Regarding #2, in order to detect changes due to slow slip events in observation data, change in trend is generally detected. Strain data, which the presenter mainly deals with, is sensitive to seasonal factors or long-term effects of precipitation, and it is difficult to objectively estimate what the trend will be in the steady state (the period when no crustal movement due to slow slip is occured on the plate interface). I examined a detection method using a state-space model (JpGU 2019), but did not go so far as to objectively estimate trends. In machine learning, the k-nearest neighbor method is used as a method for detecting anomalies in time-series data. However, since crustal movement data contains many local changes, such anomaly detection is not sufficient to detect the change due to slip on the plate interface. On the other hand, stacking method of strain data (Miyaoka and Yokota 2012) stacks the data assuming a slip on a certain subfault, so it is possible to detect the start of the chnage and the location of the slip at the same time (Tsuyuki et al. 2017). When dealing with crustal movement data, it is considered to be appropriate to use the detection method that estimates from the data of the entire observation network. Therefore, I examined whether the network inversion filter method (Segall and Matthews 1997) could be applied to strain data.