Velocity Changes Associated to the 2019 Ambon 6.5 Mw Earthquake Using Ambient Seismic Noise

Seismic velocity changes which occurred before, during and after large magnitude earthquake carry information about damage fault surrounding the area. Research focused on Ambon 6.5 Mw earthquake on September 26, 2019. We use ambient noise cross-correlation technique to compute empirical Green’s Function between station pairs from January to December 2019. We use doublet method to determine perturbation calculated in frequency domain by using Moving Windows Cross Spectrum in frequency between 0.1 to 1.0 Hz. Our results show no obvious significant velocity change before the mainshock. Velocity fluctuation average about ±0.1% caused by non-isotropic of noise source around the area. Sudden velocity decrease occurred and reached average value of -0.35% in 7 days after mainshock. After reaching its lowest value, velocity gradually increased about +0.4% in 14 days. Velocity fluctuations about ±0.2% also occurred for three months after the earthquake which may caused by the aftershocks around the earthquake epicentre. Following the large magnitude mainshock, velocity reduction may be induced by cracks opening and fractures due to deformation. While the velocity increase may caused by post-seismic relaxation due to elastic behaviour and crack healing of the fault throughout the surrounding region.


Introduction
Monitoring of crustal properties in earthquake-prone region is very useful to obtain the information of properties and seismic activity.Seismic surface wave velocities always change based on elastic behavior of earth's crust and stress activity around the area.Ambient seismic noise can be used to evaluate the crustal velocity changes which associated to stress change underneath.Velocity changes which happened after mainshock can be governed by the post seismic stress relaxation in crustal area around the fault zone [3].Ambon 6.5 Mw earthquake on September 2019 cause deformation and strong shaking in crustal and surface area around the fault zone.These strong shaking and deformation in seconds cause changes in seismic velocity because the changes in elastic properties.In this research, we investigate the temporal changes of seismic velocity in the epicentral region of mainshock by using vertical component from 4 seismometers around the area (Figure 1a).To determine the velocity changes, we use ambient noise cross correlation method to obtain the empirical Green's Function that mimics the impulse response between two stations [6].To calculate the velocity changes, we use Moving Window Cross Spectrum (MWCS) analysis, using earthquake multiplets [10].
IOP Publishing doi:10.1088/1755-1315/1245/1/012022 2 Ambon earthquake on 6.5 Mw located between Seram island and Ambon Island detected by BMKG station network.Ambon earthquake triggered by north-south fault plane orientation [11] and cause the displacement around ~4 cm horizontal and `10 in vertical [8].In this study, the station pair spacing are around 45 km for the nearest station AAI -KRAI while the furthest station pair is MSAI -NLAI which around 210 km.We need to note that all stations are located on different island except for KRAI and  MSAI because the source of ambient noise mainly from ocean microseism which may cause non isotropic source distribution [5].This study aims to obtain seismic velocity changes in pre-seismic, coseismic and post-seismic phase which associated to the 26 September Ambon 6.5 Mw mainshock.

Data and Method
This study uses waveform data acquired from continuous measurement on vertical component from 4 seismometers around Ambon: AAI KRAI, MSAI and NLAI recorded by BMKG from January to December 2019.The missing and incomplete daily waveform are detected in manual checking especially after the mainshock on NLAI station.To reduce the effect of large phases caused by earthquakes, we used filter and adjustments applied to waveform each station.The filter and analysis of waveform generally follows the common procedure [2] to obtain reliable Green's Function using the MSNoise package in Python [7].
First, we demeaned the waveform to extinguish the trend effect of waveform reading.Then we applied bandpass filter to between 0.1 -1.0 Hz to obtain the ambient noise frequency [1].For temporal normalization, we used one bit normalization to as it is most robust to remove amplitude caused by earthquake aftershocks [2].Meanwhile for spectral whitening which aimed to suppress frequency, we used several experimental frequency ranges adjusted in the bandpass and whitening filter to evaluate process in MWCS (Table 1) and analyze which frequency shows the clearest velocity change.Then after the signal pre-processed, the daily waveform data cross correlated between two stations (Figure 2a).For daily pair of stations, signals were cross correlated in 30 mins windows resulting in positive and negative time lags and then stacked in linear method.9 months of daily GF processed for reference stacking from January 1 to September 1, 2019, or before the mainshock occurred to obtain stable Green's Function.Meanwhile for current stacking, we used 10 days to 20 days of moving windows.The result of cross correlation and current stacking also can be used to analyze the direct and scattered arrival wave to determine the window length for MWCS process.

Table 1. Filter used in filtering and MWCS processing.
We use the MWCS method to obtain the relative velocity variation measurement.MWCS applied to time-series which computed by cross correlating noise between two stations.The reference stacking which has longer stacking duration as representative of background value when the medium is relatively stable (Figure 2b).Meanwhile, current stacking has shorter duration than reference stacking represents the situation at a given time period of crust (Figure 2c).The time delay between two Green's Function (reference and stacking) determined in the unwrapped phase of cross spectrum, proportional to frequency [7].The delay for each window calculated by using weighted least square within the frequency needed.Meanwhile in c) when in unstable condition, the cross correlation will produce time lags.

Cross Correlation Result
After the cross correlation and stacking process, we need to analyse the amplitude or energy recovered in positive and negative time lags to determine the window length used for MWCS process.
The cross correlated energy in Green's Function criteria must be enough to process in MWCS and obtain the recovered energy in coda wave.To enhance the reliability, we also do the SNR analysis to obtain the coherent signal energy in time domain for different frequencies (Figure 4g).Clear direct arrivals are observed in all station pairs in zero-time lag in red arrow (Figure 4a -f).On the other hand, the scattered arrival which shown in yellow arrows cannot be observed in station pair AAI-NLAI, KRAI-NLAI and MSAI-NLAI, while observed clearly in other station pairs.From the crosscorrelation and SNR analysis, we decided to choose the positive and negative time lag between 40 s and 90 s to avoid the direct and scattered arrival (Figure 4 a -f).

Velocity Variation
To obtain the velocity change around the epicentral region which caused by nonlinear strong ground motion, we observed the highest velocity reduction and correlated with the event timeline.While the high fluctuation after the mainshock may prompted by the aftershock around the area.Several research about temporal velocity variation using ambient noise cross correlation have shown that temporal changes were mainly constrained around hypocentre and ruptured area [4].Seismic velocity in cracked medium also dependent on stress.When crack density decrease, the velocity increased caused by an increase of pressure without any fluid participation [9].We focused on velocity variation of co-seismic phase.In overall velocity variation, the fluctuations before the mainshock vary between ± 0.25% but shows very high fluctuation in filter range 0.1 -0.2 Hz, 0.1 -0.4 Hz and 0.1 -1.0 Hz which may provoked by high non-isotropic ambient noise source around the area, so we decided to exclude them for analysis.While frequency range 0.1 -0.6 Hz and 0.1 -0.8 Hz show more clear fluctuations and velocity changes when mainshock occurred.Then the mainshock 6.5 Mw followed by hundreds of aftershocks around the epicentral region.
The velocity variation in co-seismic phase should only associated with nonlinear strong ground motion and deformation around the epicentral area.In this case, we focused on investigating the velocity variation with small obscuration of noise in frequency range between 0.1 -0.6 and 0.1 -0.8 Hz on 10 days and 12 days moving windows current stacking which have relatively low fluctuation.Before the mainshock occurred, the average temporal velocity relatively fluctuates between ±0.1% (Figure 6a and  6b).Ambient noise source mostly interfered by ocean microseism [1] while in this research also interfered by hollow waveform data few days after the earthquake because the velocity calculation composed by stacked correlation.Significant temporal changes of velocity occurred when mainshock occurred with average of velocity reduction -0.25% between all station pairs.Seismic velocity dramatically decreased in all frequency bands and observed clearly in 10 days and 12 days current stack resolution (Figure 6a and 6b).The velocity decreased after the mainshock keep declining gradually for 7 days.This declining pattern observed on all frequency range.The lowest value clearly observed -0.35% in frequency range 0.1 -0.6 Hz in current stack 10 days (Figure 6c).After reached the lowest value, the velocity recovered in average of +0.4% in 14 days and bounced back to 100% to the value before the mainshock occurred.This pattern observed clearly in all frequency range.After the significant decreased and increased velocity change, the pattern of velocity become highly fluctuated around average of ±0.3%.This pattern also observed in all frequency range.The strong ground motion and deformation are the main cause of velocity decrease of Ambon.The velocity decreases mainly dominated in the crust area around the hypocentre depth and rupture area.The deformation process causes significant decrease and gradually decreased because of the deformation process.While the velocity increase after the mainshock shows the process of post-seismic relaxation.The recovery need 14 days to recover to 100% as before the mainshock occurred.And following the mainshock, the hundreds of aftershock events are responsible for the high fluctuation after the recovery (±0.2%).These mechanisms are often located deep around the hypocentre depth and are caused by cracks opening and closing.

Conclusion
The main objection of this study is to analyse the velocity change which associated to Ambon 6.5 Mw earthquake.We choose frequency band and current stack parameter which produced minimum fluctuation to make sure minimize the non-isotropic noise interference.We decided to use frequency band 0.1 -0.6 Hz and current stack 10 days as the clearest velocity variation graph.We investigate there are no significant velocity change before the mainshock (January -August).The average temporal velocity relatively fluctuates around ±0.1% before the mainshock.When the mainshock occurred on September 26, the velocity decreased significantly in 7 days to -0.35%.This process mainly caused by deformation and rupture process around the epicentral region.After reach the lowest value, the velocity increased +0.4% or recovered 100% to the value before the mainshock occurred in 14 days.We interpret

Figure 1 .
Figure 1.a) Daily seismicity frequency detected around the area of Ambon in map c).Red star denotes the mainshock.b) Map of epicenter of earthquake 6.5 Mw inside the red rectangle area c).White circles show the earthquake events January -Desember 2019.Earthquake events catalog, location and focal mechanism obtained from BMKG catalog.

Figure 2
Figure 2Adapted from[12], shows illustration of how velocity changes measured from cross correlation when a) direct and scattered arrival recorded on two stations.Cross correlation in stable medium produces cross correlation in figure b) shows in positive and negative lags have no delay.Meanwhile in c) when in unstable condition, the cross correlation will produce time lags.

Figure 3
Figure 3 Workflow data processing

Figure 4 .
Figure 4. (a -f) Daily cross correlation result from each station pairs to analyse the direct and scattered arrival wave in positive and negative time lags.Red arrows denote the direct arrival wave and yellow arrows shows the clear scattered arrival around the time lags.And SNR analysis also applied to all cross correlation in different frequencies, as shown in figure g) for AAI -KRAI pair.

Figure 5 .Figure 6 .
Figure 5. Average of velocity variation for different frequency ranges (a-e) in whole time period from all station pairs in this study.Blue and orange line denote the mean and median of velocity from all station pairs.Red line denotes when the mainshock on September 26, 2019.