PROJECT
1. Project overview
This project aims to construct a "fault-slip monitoring system" to understand the current state of fault slip, predict the short-term fault slip evolution on a given fault, and evaluate the potential of this fault slip evolution triggering a future major earthquake, as these destructive events are known to occur frequently in subduction zones. We aim to realize this system by integrating geodetic data that capture daily ground motion (i.e., crustal deformation) with physics-based numerical simulations that calculate the temporal evolution of fault slip via data assimilation, a computational technique using information science. Data assimilation techniques are already routinely used for weather forecasting; therefore, we envision that the resultant fault-slip monitoring system from this project will be like a fault-slip version of weather forecasting. However, the geodetic data that provide the input for this system contain observation noise with various factors. Understanding the noise characteristics and the crustal deformation associated with fault slip on daily to weekly time scales when the signal level is comparable to the observation noise level is not necessarily sufficient. Therefore, we will incorporate information science and seismological expertise to develop innovative methods of geodetic data analysis based on statistics and machine learning. These innovative methods will improve our current ability to detect crustal deformation by fully utilizing the available data and establishing a method for understanding fault slip that takes into account the characteristics of the observation noise. Ultimately, we aim to establish a fault-slip monitoring system by incorporating these innovative geodetic data analysis methods with a data assimilation approach. This monitoring system will elucidate the current state of slip along a given fault, and predict the short-term evolution of this fault slip in real time via an automated geodetic data analysis and data assimilation system, thereby leading to the establishment of a short-term earthquake probability evaluation method based on geodetic data.
2. Aim of this project
The following three research items have been established for this project to realize aforementioned goals.
In research item (A), "Improvement of Crustal Deformation Detection by Statistics and Machine Learning," we will develop statistical and machine learning methods for accurate detection of crustal deformation associated with fault slip, such as short-term slow slip events that produce a crustal deformation signal that is comparable to observation noise. Specifically, we aim to dramatically increase the number of detected crustal deformation phenomena through automatic separation of the deformation signal and noise via component decomposition, objective crustal deformation detection using sparse estimation, and deep learning.
In research item (B), "Improvement of state space model considering the characteristics of observation noise," we will examine the spatio-temporal characteristics of the observation noise contained in real geodetic data and implement them in a state-space model to refine the noise model. We will also employ machine learning to examine and compare multiple methods to achieve more accurate fault slip estimations. Research items (A) and (B) are related, and the mutual feedback on noise characteristics obtained from these two research items will deepen our understanding of noise characteristics, assist in developing crustal movement detection methods that make use of these characteristics, and improve fault-slip estimation methods.
In research item (C), "Establishment of a data assimilation fault-slip monitoring system," we will integrate the results of research items (A) and (B) to introduce a data assimilation method that takes into account the physical laws of friction on the fault, and investigate short-term prediction methods for fault slip on daily to weekly time scales. The performance of the monitoring system will be verified using existing geodetic data for the entire Nankai Trough region.
We hope that the geodetic data analysis methods developed in this project will be widely applied in studies using geodetic data, and we intend to promote this project as a means of advancing our understanding of earthquake- and volcano-related phenomena through geodetic data analysis.
