An attempt of transient tectonic signals detection from DONET seafloor pressure record using principal component analysis

The separation of the tectonic and oceanographic signals is one of the largest problems for seafloor geodetic observation using ocean bottom pressure-gauge (OBP) data. Otsuka et al. (2022, OSM) indicated that Principal Component Analysis (PCA) is helpful to identify oceanographic signals from continuous pressure data obtained by a seafloor network. They applied PCA to the DONET OBP data and oceanographic variations of different spatio-temporal patterns can be extracted as three major principal components (PCs). In this study, we evaluate whether PCA can distinguish tectonic and oceanographic signals, by applying PCA to the DONET OBP data in which synthetic tectonic signals are embedded.

We used the 40 OBP time series from 2016 to 2019, after removing the tidal components as well as instrumental drift, using a low-pass filter and exponential and linear approximate, respectively. A synthetic signal was generated assuming slow slip event (SSE), modeled as thrust motion along the slab surface with a ramp function. Duration and amplitude were defined by the SSE sizes, which also define the fault sizes. After adding the synthetic signals corresponding to SSEs of various sizes and locations, we applied PCA to the time series.

Applying PCA to the test data, the synthetic signals were successfully extracted as another PC, in addition to the oceanographic components, which had been identified from the original DONET data. The order of the PC corresponding to the synthetic signal increases as the assumed event size decreases. Therefore, monitoring the time variation of the constituents of major PCs can help to detect the occurrence of the SSE. In order to identify the change in the PC constituents, we monitor the time variation of the eigenvector corresponding to a specific order of PC. It is expected that the eigenvector changes significantly because the reorder of major PCs happens due to the change in PC constituents. We quantify the change by calculating a scalar product of a vector calculated from windowed OBP data and a reference vector, representative of the steady state, expecting that the product decreases when the transient signal is contained. The results of numerical experiments focused on the second PC demonstrated that signals of SSEs Mw5.9 or larger could be detected by the proposed monitoring method.