In this study, we utilise GNSS data, tidal observation data, and aftershocks data to investigate the fault responsible for the 2019 Banten earthquake. The information of this data and the methodology used in this study are described below.
GNSS
The GNSS data used by this study was obtained from the Indonesia Continuously Operating Reference Station (Ina-CORS) (Gunawan et al. 2019; Gunawan and Widiyantoro 2019), which was available during the 2019 Banten earthquake event. The GNSS antennae of the Ina-CORS station are located embedded on top of a concrete structure. The GNSS data, which is recorded with a 30-s sampling interval, was processed using GipsyX software (Bertiger et al. 2020) and GAMIT software (Herring et al. 2010), described below.
During daily solution estimation using GipsyX, we conducted static solutions in the precise point positioning mode. We employed fiducial-free with five iterations and JPL’s reanalysis final set of the International GNSS Service 2014 (IGS14) orbit and clock product. Ocean loading parameters were obtained from the Onsala Space Observatory (http://holt.oso.chalmers.se/loading/) using the GOT4.8 model. In addition, we also set an elevation angle cut-off with 15°.
Meanwhile, our daily solution estimation used GAMIT incorporating processing steps as used by Gunawan et al. (2021). First, the daily position was estimated with atmospherically used, loose-constraint, prior GNSS phase observations; the orbit and earth-orientation parameters were fixed. Second, these positions and their covariance with global GNSS solutions, computed as part of Massachusetts Institute of Technology’s processing for the International GNSS Service (IGS), were combined. Third, the clarified position time series were estimated. In both the second and third steps, the loosely constrained solution was mapped onto a well-constrained reference frame by minimising the position and velocity differences of selected stations with respect to a priori values defined by the IGS14 realisation of the International Terrestrial Reference Frame (ITRF) 2014 reference frame.
Using the processed daily solutions GNSS data, we extracted the coseismic displacements by subtracting the velocity data for five days after the earthquake to five days before the earthquake. For each coseismic displacement analysis process obtained from GipsyX and GAMIT, we took the average value and used it as final coseismic displacements of the 2019 Banten earthquake.
Coseismic slip inversion
The coseismic slip inversion was calculated using the final coseismic displacements obtained from the processed GNSS data. In our search for the fault model of the 2019 Banten earthquake, we modelled using two possible fault strikes based on the earthquake focal mechanism as reported by the USGS. The first fault strike model, named Model 1, dips to the south of an ENE-WSW fault direction with a strike of 69°. The second fault strike model, named Model 2, dips to the west of a NEN-SWS fault direction with a strike of 201°. The dip angle for Model 1 is 54°, while the dip angle for Model 2 is 49°. In both models, the length of the fault is 40 km, which is estimated using an earthquake scaling relationship for dip-slip faulting system (Gunawan 2021). We also divided the main fault into sub-faults with a length and width of 5 km.
In our inversion, we utilised sdm2013 (Wang et al. 2011; 2013) to estimate the coseismic slip distribution. This process follows the objective function as follows: \(F\left(m\right)={\Vert Gm-d\Vert }^{2}+{\alpha }^{2}{\Vert H\tau \Vert }^{2}\). In this function, G is the Green’s functions obtained from an elastic half-space model (Okada 1992), m is a coseismic slip of each sub-fault, d is coseismic displacements, α2 is the smoothing factor, which controlled by the model roughness and data misfit, H is the finite difference approximation of the Laplacian operator, and τ is the shear stress drop. For every fault model, we investigated using a shallow top fault depth of 1 km, hereinafter referred to as Model 1A and Model 2A, and a deep top fault depth of 25 km, hereinafter referred to as Model 1B and Model 2B. All of the four models were constructed with a bottom depth of 50 km.
Tsunami
In this study, we modelled the tsunami waveform using the estimated coseismic slip of those four models (Models 1A, 1B, 2A and 2B) and compared it with the tide gauge data available along the coast of southern Sumatra and western Java. The tidal observation data was obtained from the national tide gauge stations network operated by BIG. Eight tide gauge stations located off the coasts of western Java and southern Sumatra were used to understand the possible tsunami waveform as recorded by the tide gauge. The location of these tide gauge stations is shown in Fig. 1.
To extract the probable tsunami waveform recorded by the tide gauge, first, we conducted a de-tiding process to separate the data from its tidal components. In this process, we used a bandpass filter FFT (Fast Fourier Transformation) with a period of 3 to 30 min (Rabinovich 1997; Heidarzadeh and Satake 2013). Then, although the tide component was removed, the recorded data may suffer interference from noise. To eliminate this type of noise, the tide gauge data was then filtered using the moving averages process. In this study, a moving average was carried out for 15 pieces of data, or for tide gauge data with a 1-min sampling rate of 15 min.
The tsunami modelling was performed with the use of TSUNAMI-N3. The model data and setup were as follows. First, the numerical domain for this study was in the boundary of geographic coordinates in longitude between 103°E and 107°E, and latitude between 5°S and 8°S. The geometric data used GEBCO Compilation Group (2020) combined with a navy chart provided by the BNPB. The grid data used in this study was set to be 450 m, and the simulation time for this model was 7200 s or 2 h.
Aftershocks
Aftershocks obtained from the Indonesian Agency for Meteorology, Climatology and Geophysics (BMKG) network were used to understand the possibility that the mainshock raised stress in the surrounding region and triggered these aftershocks. We relocated aftershocks using a Double-Difference (DD) method (Waldhauser and Ellsworth 2000). In this method, when the hypocentre distribution distance between two earthquakes is very small compared to the distance between the source station, then the ray-path and waveform of the two earthquakes can be considered to be approximately the same. With this assumption, the difference in travel time between the two earthquakes recorded at the same station can be considered only as a function of the distance between the two hypocentres.
Pesicek et al. (2010) developed the DD method for teleseismic cases by adapting the P wave beam tracking method for the case of spherical earth (Koketsu and Sekine 1998). We used the DD method for the teleseismic distance, named teletomoDD, which uses a nested regional-global 3D velocity model (Widiyantoro and van der Hilst 1997). For regional models, the 3D velocity model is used and for the global model, the AK135 velocity model is used (Kennett et al. 1995).