Aftershock study of the 2019 Ambon earthquake using moment tensor inversion: identification of fault reactivation in northern Banda, Indonesia

On September 26, 2019, an Mw 6.5 earthquake occurred 23 km northeast of Ambon City, Indonesia, followed by numerous aftershock series related to a complex fault network reactivation in the Ambon and Seram region. Using moment tensor inversion, we identify the kinematics of fault reactivation based on the focal mechanism solution of 20 aftershocks with Mw > 3.2 and analyze the earthquake sequence from both focal mechanism solutions and spatiotemporal seismicity. The MTs solution of aftershocks revealed three different characteristics of fault reactivation: (i) a 35 km long N-S oriented main fault characterized by dextral strike-slip (ii) a NE-SW reverse fault segment with a ~ 55° northeastward dip located in southwest Seram, and (iii) two strike-slip segments (NNW-SSE and NNE-SSW trends) and an E-W normal fault in Ambon Island. Analysis of spatiotemporal seismicity with the MTs solution suggests that the Mw 6.5 Ambon aftershock sequences can be described as follows: (i) an Mw 6.5 mainshock rupture that was primarily made up of a major strike-slip component and an insignificant minor normal fault; (ii) first aftershock cluster propagate along the main N-S ruptures, followed by the strike-slip and normal cluster in Ambon Island (iii) The reverse fault events cluster appeared next in Southwest Seram. The presence of complex strike-slip segments in Ambon agrees with the regional structure trends in Halmahera, located in the north of the study area, while the E-W oriented normal fault might be related to the eastward velocity increase in Banda Arc, which causes extensional deformation. Given that the fault reactivation identified in Ambon and Seram is located close to the densely populated urban regions of Ambon City and Kairatu, the analysis of future seismic hazards related to this fault reactivation should consider the risks in a region with complex fault settings.


Introduction
Indonesia is located in a complex tectonic region with frequent occurrences of earthquakes.In eastern Indonesia (insert map in Fig. 1a), the major plates responsible for the high seismicity are the Australian plate which moves to the north-northeast relative to Eurasia, and the Pacific plate which moves to the west relative to Australia (DeMets et al. 2010).The westward movement of the Pacific plate is the primary driving force for a series of left-lateral displacements in the northern part of eastern Indonesia, including the Yapen Fault Zone (YFZ), Sorong Fault Zone (SFZ), the Kawa Shear Zone (KSZ), as well as the Matano and Palu-Koro Faults (in Sulawesi, exclude from the insert map) (Stevens et al. 2002;Bock 2003;Watkinson and Hall 2017).The Banda Arc, which curves 180° around the Banda Sea, forms a prominent geological feature in eastern Indonesia.Its collision with the Australian continental margin contributes most to seismicity in the Banda region.The arc developed by the eastward subduction rollback as ancient cold oceanic Proto-Banda Sea lithosphere which formed a pre-existing embayment in the Australian continental margin was subducted from the Miocene onwards (Spakman and Hall 2010).This old lithosphere has been almost completely subducted under the Banda Sea.The study area is located in the northern part of the outer Banda Arc, part of the collision zone between the Australian continental margin and the Banda Arc (Audley-Charles et al. 1979).This arc, which represents the deformation front of the foldand-thrust belt, is mostly comprised of imbricated thrust faults and is bounded by the Timor, Tanimbar, and Seram Troughs (Audley-Charles et al. 1979;Hall R et al. 2017;Baillie et al. 2020).The northern outer Banda Arc now converges with the Bird's Head at a rate of 30-50 mm/ year in an ENE direction, oblique to the Seram Trough (Bock et al. 2003).Convergence in the Seram region is accompanied by strong seismicity dominated by shallow thrusts and strike-slip faults (Mccaffrey 1989;Engdahl et al. 1998).
Several major destructive earthquakes occurred in the northern Banda Arc (black stars in Fig. 1a), such as the 1674 and 1708 earthquakes which triggered a tsunami in Ambon, Haruku, Buru, and the surrounding regions (Harris and Major 2016); an Mw 7.8 earthquake in 1899 that struck Tehoru, Seram (Brune et al. 2010); an Mw 7.6 on 8 October 1950 close to the south coast of Ambon (Båth and Duda 1979); and an Mw 8.2 earthquake in 1965 located in northern Buru (Løvholt et al. 2012;Patria et al. 2021).Ambon City, located on Ambon Island, is the capital of the Maluku province with a population of more than 368,987 inhabitants in 2021 (BPS: Statistics Indonesia 2021).Although this region is tectonically active and Fig. 1 a Active fault map of the northern Banda region adapted from Patria et al. (2021) shown by red lines.The epicenter of the historical earthquakes from NOAA (National Oceanic and Atmospheric Administration) from 1600 to before 1900 are presented as black stars.Centroid moment tensor solutions (navy beach balls) of events with depth < 20 km and a magnitude more than Mw 5 in the period of 1976-2018 are derived from Global CMT.b Map of the aftershock distributions (indicated by blue circles) in the study area derived from Sahara et al. (2021a;b) between October 18th to December 15th, 2019 recorded by local and BMKG regional networks.The epicenter of the mainshock is shown by a magenta star (Global CMT) and a red star (BMKG).Focal mechanism solutions from both agencies are inserted by red beach balls.Fault segments from Patria et al. (2021) are plotted in black lines.The area colored in orange represents the urban or populous region experienced several historical destructive earthquakes, not all the source faults of the past earthquakes were fully understood due to the lack of marine geophysical surveys and insufficient seismic data.Thus, seismic hazards related to the active faults in this densely populated area should be analyzed further.
Recently, fault observations in this area have become more intensive.Patria et al. (2021) integrated their geomorphic feature studies with the existing dataset from geological structure, seismic, and geodetic observation to map the complex active faults in the northern Banda Arc, including Seram and Ambon (the identified faults are shown by red lines in Fig. 1a as well as black lines in Fig. 1b).The study observed that the fault system in Seram is generally divided into two different tectonic regimes.North Seram is mostly characterized by the activity of the Seram fold-and-thrust belt.Thus, features such as back-tilted marine terraces, fault scarps, and uplifted ridges are well observed there.The focal mechanism solution from Global CMT (navy beach balls) in this area added an understanding of the distribution of the thrust fault in this area.Meanwhile, Central, and east Seram, are dominated by left-lateral strike-slip faulting, which is identified by the presence of river offsets, decapitated rivers, shutter ridges, and linear valleys (Patria et al. 2021).
The ~ 137 km-long Kawa Shear Zone (Pownall et al. 2013), or Kawa Fault (Patria et al. 2021), with observed offsets of about ~ 2 km is the main significant structure in Central Seram, separated from the Bobot Fault in the east by Teluti Bay.A left-lateral strike-slip fault is also recorded in the offshore area east of Seram (Teas 2009;Patria and Hall 2017).Watkinson and Hall (2017) suggested this offshore strike-slip fault, the Bobot Fault and the Kawa Fault are part of a KSZ.Patria et al. (2021) also identified several dip-slip faults with the east to northeast strike direction in Amahai, Ambon, and surrounding islands that cut marine terraces.The Ambon Fault extends from Ambon Bay for about 34 km to the northeast coast of Ambon.Patria et al. (2021) suggested that the Ambon Fault is relatively young since it deforms the lower marine terrace surfaces of the early to middle Pleistocene age.
The most recent significant earthquake with an Mw 6.5 struck the northern Banda region on September 26, 2019, with a hypocenter located 23 km northeast of Ambon City.Sahara et al. (2021a;b) have produced an accurate seismic catalog of the aftershock events due to the seismic monitoring for 59 days using 11 local seismic networks and four regional seismic networks from Agency of Meteorology, Climatology, and Geophysics of Indonesia (BMKG).Figure 1b shows the location of the mainshock and aftershock distribution.Based on the analysis of aftershock distribution and finite fault modeling, Sahara et al. (2021a) suggested that the Mw 6.5 Ambon earthquake occurred on an N-S-oriented fault plane 35 km in length that extends from Kairatu down to the Haruku strait, in which both south (label I in Fig. 2a) and north (label II in Fig. 2a) ends have more frequent aftershocks.In addition, there was seismicity trending NE-SW in Seram Island (label III in Fig. 2a), and seismicity trending NE-SW in Ambon Island (label IV in Fig. 2a).The widespread aftershock distribution separates into several distinct clusters raising the issue of possible reactivation of preexisting faults in this region.Interestingly, most of the aftershock sequences occurred on faults that are still unmapped.The series of mainshocks and aftershocks severely impacted the surrounding region including central Maluku and West Seram (BNPB: National Agency for Disaster Management 2019; BPS: Statistics Indonesia 2021).This data (which provides statistics about damage to houses) support that the aftershock distribution agrees well with these two most impacted regions.
Although the aftershock trends had been resolved well, there is not enough active fault data to be associated with this region's mainshock and aftershock sequences.Previous structure geological studies have shown that northeast-trending dip-slip fault scarps face southeast or northwest and cut marine terraces on Ambon and in the surrounding region and it seems too far from the seismicity clusters (see Fig. 1b) (Patria et al. 2021;Watkinson and Hall 2017).However, Ambon and Seram are in a tropical zone between latitudes 3 and 4°S, with heavy rainfall and high daily temperatures in which surficial processes may rapidly erode indications of active deformation.As a result, Ambon and Seram might have a more complex geological structure than one could map.Thus, revealing and interpreting the hidden faults is essential for assessing the potential zone of future seismic hazard.Additional information from source mechanism inversion can provide important insights to reveal the unmapped fault (in this case: fault reactivation) which could not be resolved by geological observation.
Using the aftershock sequences of the 2019 Mw 6.5 Ambon earthquake recorded by well-distributed temporary local and regional BMKG seismic networks, we attempt to identify the kinematic of possible fault reactivation.We estimated the focal mechanism of significant aftershock events by implementing moment tensor (MT) inversion using ISOLA, the ISOLated asperities software package (Sokos andZahradnik 2008, 2013).This program is widely used in retrieving focal mechanism solutions, including when dealing with low-to moderate-magnitude earthquakes (e.g., Abdel-aal and Yagi 2017;Ataei and Rezapour 2016;Paul et al. 2018;Serpetsidaki et al. 2010).The distribution of aftershock sequences is used to identify the geometry of the fault plane and a study of the focal mechanism of an individual seismic event provides kinematic information on the fault movement (e.g., Hallo et al. 2019Hallo et al. , 2017;;Herman et al. 2017;Umino et al. 2003;Zheng et al. 2010).We analyzed the major and minor MT decomposition from the mainshock's existing MT solution to observe the possible complexity of the source mechanism and examine the possible relationship between the mainshock and the aftershock.

Data
The aftershock monitoring was conducted using a total of 15 stations, consisting of 11 three components temporary broadband stations and 4 regional BMKG stations.The distribution of the seismic networks is shown in Fig. 3.The aftershock monitoring was conducted from October 18th, 2019, 21 days after the mainshock, to December 15th, 2019.As a result of the seismic monitoring, Sahara et al. (2021a; b) produced a highly accurate catalog consisting of 1152 double-difference relocated event locations.Figure 2a shows the distribution of the relocated events and the event grouping selection for the moment tensor processing.Here, we divided our investigation into four regions (denoted by the green dashed rectangles).The corresponding labels in each rectangle represent the region number.Regions I and II are located on the south and north of the N-S trending fault which is previously defined as the main rupture by Sahara et al. (2021a).Region III represents a separate cluster in the southwest area of Seram Island with a NE-SW orientation.Meanwhile, Region IV represents aftershocks located on Ambon Island which follow a NE-SW orientation.
We initially selected aftershock events with an Mw magnitude greater than 4.0 in the catalog of Sahara et al. (2021b).In the case of high magnitude, the moment tensor results are stable over various frequency bands ranging.When implemented in a lower magnitude, using high frequency would result in a reduction of solution quality, particularly when using a simple 1-D velocity model (Cesca et al. 2006;Petersen et al. 2021).However, we could not find enough significant events with Mw > 4 in regions II and III.Therefore, in these two regions, we included the events with a magnitude of Mw 3.2-3.8which have a high signal-to-noise ratio (SNR) and were recorded by at least 5 stations.Based on these criteria, 20 aftershocks were selected for moment tensor inversion in all four regions.Figure 2b shows the location of the selected events, while detailed information about these events is presented in Table 1.

Moment tensor inversion
The moment tensor inversion is implemented to the aftershock data using the ISOLA code developed by (Sokos andZahradnik 2008, 2013).In this approach, the source is assumed to be made up of a sequence of point sources of various focal mechanism.They proposed that the moment tensor at a given position and time centroid can be considered a linear combination of the six components of the moment tensor (MTs).Calculations of MTs are performed by minimizing the difference between the observed and synthetic displacements.The match between the observed and synthetic data is characterized by variance reduction (VR), i.e., greater VR indicates better inversion matching and vice versa.In our case, we inverted the deviatoric moment tensor, in which we more concerned about the best-fitting DC source.In terms of the fault-related case, we expect the result to be a shear faulting with significant DC.As a preprocessing procedure, we initially selected the waveform based on SNR by visually inspecting in all components for each station per event.We set the time window at 80-100 s length and then the instrument response was deconvolved to obtain displacement seismograms that were bandpass filtered between 0.04 and 0.12 Hz.This range is the minimum and maximum frequency bands utilized in the MT inversion to reduce the influence of low and high-frequency noise.Considering the range of aftershock depth based on the catalog (Sahara et al. 2021b), the trial source was calculated at various points below the epicenter from 2 to 20 km with an interval of 2 km.At each source depth, Green's function calculation is performed using the frequency wavenumber method (Bouchon 1981).This step was carried out to test the correlation between the observed waveform and the synthetic waveform of both vertical and horizontal components.For each source depth, the source time shift is also defined to form a 2-D correlation contour (see Fig. 6b).The inversion process is carried out at each depth and time shift where the optimal solution with the highest correlation will be determined as the best fit through an analytic grid search.We manually inspected and selected well-correlated waveforms and removed the poor ones in each iteration of the MT inversion.In the end, around 5 to 7 stations were used for the final MT inversion for each event.

Accounting for the quality parameter and uncertainty of the MT
We used VR as the main evaluation of the inverted moment tensor quality as it represents how good the fitting between the real waveform and the synthetic waveform.In addition, we also used Focal Mechanism Variability Index (FMVAR), Condition Number (CN), Space-Time variability index (STVAR), and DC percentage to assess the reliability and robustness of the MT solution.CN represents the quality related to the stationevent configuration, STVAR and FMVAR are related to the robustness of the solution in the space-time dimension, while DC measures the percentage of the shear slip on a planar fault (detailed explanations of the quality parameters, including the calculation and visualization, are served in the additional files).We took events that satisfy at least four out of five criteria based on Sokos and Zahradník (2013) as follows: (i) VR > 50%, (ii) CN < 10, (iii) FMVAR < 30°, (iv) STVAR < 0.3, (v) DC > 70%.
We also assessed the MTs solution uncertainties for all events to add the control quality factors.In this study, we only calculated relative uncertainty assessments, in which we investigated the resolvability of the DC component of moment tensor solution without using actual waveforms (Sokos and Zahradník 2013).In this case, errors of the model parameters are related to data errors through green's functions matrix, G .As a result, we may analytically examine a given source-station configuration, quantify the uncertainty of a certain focal mechanism, and prescribe the data variance σ 2 d and missfit Δχ2 (Zahradník and Custódio 2012).The uncertainty of MT solutions was calculated by averaging the 5D error ellipsoids (for deviatoric MT) on each position and source time.The ellipsoids are constructed numerically by regularly sampling the parameter space around and the assumed moment tensor, in which the shape and orientation are determined by the source-station configuration, frequency ranges, and the crustal model used.From the error ellipsoid, a statistical set of the strike, dip, and rake is obtained whose histograms can be constructed (the detailed procedure and example of the uncertainty calculation is presented in the additional file).

Major and minor moment tensor decomposition
Unlike the MT analysis on the aftershock data, in which we just used deviatoric MT in the processing, for the mainshock, we would further carry out a deviatoric MT decomposition into major and minor DC MTs (Kanamori and Given 1982).Both decomposed MTs may be physically interpreted by the complex geometry of the rupture.However, Jost and Herrmann (1989) pointed out that these typical decompositions may be suitable in some specific cases.This means that the physical interpretation of the major and minor DC MTs should be confirmed by other data derived from the rupture (e.g., local tectonic settings, source mechanisms of other events, or slip inversion on a segmented fault) (Hallo et al. 2017(Hallo et al. , 2019)).In the case of 2019 Ambon earthquake, the coseismic slip model from Meilano et al. (2021) has clarified and supported the interpretation of the rupture.As additional support from the perspective of MT analysis, we would like to conduct this MT decomposition to examine the possible complex rupture.Furthermore, comparing these decomposition results with the distribution of MT solution of aftershock could give an understanding of the relationship between the mainshock and the aftershock.
The major DC MT is the best approximation of a nonshear (non-DC) MT by a shear (DC) source.As in (Wallace 1985), assume absolute values of deviatoric MT eigenvalues and v 3 , the major and minor DC MTs with pre- served dominant axis can be formulated as follows: The deviatoric condition 1 + 2 + 3 = 0 applies.The dominant eigenvector v 1 refers to the direction of either P-or T-axis, based on the sign of the dominant eigenvalue 1 (Hallo M et al. 2017).The python code from Hallo et al. (2017Hallo et al. ( , 2019) is used to perform major and minor MT decomposition.
As the MTs solution of the mainshock, due to the limited local network provided (beyond our local monitoring), we considered the existing data from both Global CMT (Ekström et al. 2012) and GFZ Potsdam (Saul et al. 2011).For both two MTs solutions, we take advantage of the Full MT concept and decompose it into major and minor DC MTs assuming preserved P-axis (Wallace 1985;Hallo et al. 2017Hallo et al. , 2019)).The more reliable result was analyzed further by considering the aftershock's MT solution.

1D crustal model
We tested several crustal models to select the most suitable one for representing the study area's subsurface structure which is reliable for Green Function calculation.The four models were selected, namely the PREM (Preliminary Reference Earth Model) (Dziewonski and Anderson 1981), a crustal model for the central part of Java (Koulakov et al. 2007), a crustal model for eastern Indonesia (Sabtaji and Nugraha 2015), and a new regional crustal model around Ambon derived from Sahara et al. (2021a).The regional model from Sahara et al. ( 2021a) is an updated 1D velocity model utilized by implementing VELEST program (Kissling et al. 1995) which used AK135 as initial crustal model to better fit the simple geological model in the Ambon region.Figure 4 describes the value of P and S wave velocity and the layer depth of all models. (1) (2) The Green Function is calculated using four crustal models, followed by an MT inversion.We performed MT inversion tests for the Mw 5.1 aftershock event that occurred on November 12th, 2019 at 10:10:40 UTC. Figure 5 depicts the MT results in the form of correlation graphs using four distinct crustal models.For each model, the graphic shows the focal mechanism solution derived from ten trial source numbers, and the correlation values between observed and synthetic waveforms.

MTs solution of aftershock events
An example of the inverted MTs obtained in this study is presented in Fig. 6.We plot the MTs inversion results for an aftershock event that occurred on November 12th, 2019 at 10:10:40 UTC with a magnitude of 5.1 Mw (event ID 11).It was the greatest aftershock recorded by our network.The frequency range used for this earthquake is 0.04-0.10Hz. Figure 6a shows the correlation between the observed (black) and synthetic (red) waveforms for all the components of 7 stations.VR values for each station are indicated in blue numbers, in which the higher the absolute value, the better the correlation is.We deactivate the E-W and vertical component of MA01 (ID MA1) stations due to the poor correlation (depicted by the grey waveforms).The elimination is also applied to the N-S component of stations KRAI (ID KRA) for the same reason.The average VR value of all stations is 71.9% (see Table 2).A plot between the source time shift and trial source numbers in correlation contours is plotted in Fig. 6b.The background color shows the correlation value a scale of 0-1, while the beachball color represents the DC percentage.Here, the maximum correlation is obtained for a source time shift of ~ 1.5 s which corresponds to the increased size of the resultant beachball shown in red.The graph of correlation vs depth is also shown (Fig. 6c).We can see that the maximum correlation is obtained for a focal depth of 12 km with 94.5% DC of the beach balls.Figure 6d shows the map view of the event epicenter (a black star) and the stations used in the MT inversion (reverse red triangle).
Figure 7a shows the distribution of the focal mechanism solution.Our MTs inversion results suggest that strike-slip mechanisms predominated in the 2019 Ambon aftershock sequences, followed by minor normal and reverse faults.The label number in the corresponding beachball indicates the event ID as shown in Table 2.The ternary diagram adapted from Frohlich (1992) and Hallo et al. (2017Hallo et al. ( , 2019) ) is presented in Fig. 7b to indicate the style of faulting of each MTs solution.
Table 2 shows the MT solutions of 20 events, which consist of nodal plane parameters (strike, dip, and rake), the frequency band used, magnitude, DC percentage, the four quality parameters (VR, CN, FMVAR, and STVAR), and nodal uncertainties.Table 2 shows that almost all solutions fit the criteria, except the bold and underlined values.The DC percentages are significant (in the range between 70 and 95%), with 80.9% on average.The VR is consistent in the range of 51-72% with 59.7% on average.For STVAR values, 5 of 20 events are out of the criteria (event ID 6, 7, 10, 16, and 20), while event ID 4 and 15 does not fit the CN criteria.However, the values of both events in those two parameters still have small differences from the threshold criteria with a maximum gap of 4.9.For FMVAR, all events are in the criteria range, showing that the solution is stable around the optimum correlation value.
The uncertainty values for nodal parameters (strike, dip, and rake) for 20 events is shown in Additional file 1: Fig. A22 and Additional file 1: Table A1 in the additional files.The average value of uncertainties for the dip, strike, and rake is 8.80°, 4.64°, and 10.29°, respectively.The visualization of the nodal uncertainty (Table A2 column 8) shows that the uncertainty values are relatively small, to represent the coherence solutions of all the sampling nodal planes.into major strike-slip and minor normal fault.For the global CMT solution, the non-DC MT is decomposed into a combination of strike-slip and the normal fault with 84% and 16% scalar seismic moment ratios, respectively.Meanwhile, the solution from GFZ which has a large non-DC component is decomposed into a combination of strike-slip and the normal fault with 68.1% and 31.9%,respectively.

Discussion
We observed heterogeneity of MT solutions corresponding to different clusters.The strike-slip events were primarily associated with Regions I and II (the main N-S rupture), as well as a portion of Ambon Island (Region IV), while events with normal fault mechanisms were observed in the northern part of Ambon Island.Interestingly, we also identified a NE-SW reverse fault in Region III, southwest of Seram, where the aftershock's trends and dip direction corresponded with the focal mechanism solutions.We discuss the mechanism of events in each region.In addition, we also analyze the MTs decomposition result of the mainshock, and spatiotemporal seismicity of the aftershock sequences.

Analysis of region I and II (MTs solution along the main N-S rupture)
The Mw 6.5 2019 Ambon mainshock occurred on an N-S trending strike-slip fault with a rupture length of about ~ 35 km from Kairatu down to the Haruku strait (Sahara et al. 2021a).We identified that most of the strike-slip faults of the aftershock sequence was located along this mainshock rupture.We chose the N-S nodal plane as the fault plane in these regions since it corresponds to the rupture length as well as the aftershock trends.
Figure 9a shows the map view of focal mechanism distribution along the N-S main rupture (region I and II) with two-line sections both in the north and south including the hypocenters in a radius of 6 km (inside the blue dashed rectangles).We plotted cross-sections map to provide a better geometric illustration of the aftershocks distribution and their focal mechanism solutions (Fig. 9b, c).The red dashed lines on the figure depict the average fault dip angles inferred from both the aftershock trends and focal mechanisms.We observed that all strike-slip mechanisms in the south (Fig. 9b) have large dips angles which are mostly westward with strike orientations N-S.The average westward dip of events at the southern end of the main fault is 77.3 °.In the north (Fig. 9c), four focal mechanism solutions consistently show a strike-slip mechanism with a lower dip angle compared to the south with an average of 69 ° eastward.
The change in dip direction along the N-S main fault from westward in the south to eastward in the north that is revealed in this study might refer to the transition zone of the possible bend fault geometry in the north, where the NE-SW oriented reverse fault (discussed in the next section) appears roughly 5 km east of the northern strike-slip.A bent fault system with different dips also occurred in the 2016 Kumamoto, Japan, earthquakes.It propagates on a near-vertical dip on one side of the bend and a northwestward dip on the other side.In this case, the inflection point represented the boundary of the two different ruptures (i.e., a rupture from the foreshock and the mainshock) and formed a ~ 10 km-long seismic gap (Asano and Iwata 2016;Fukahata and Hashimoto 2016;Lozos 2021).The presence of a less seismic zone of the 2019 Ambon earthquake is found in the central part of the main N-S rupture or between region I (in the south) and region II (in the north) (denoted as the red dashed circle shadow in Fig. 9a).A further study of bathymetry structure is required to assess the variation of the main fault dip direction.
We also found that differences in dip angles in the north and south affect the direction and length of the rupture.The transition from vertical to nonvertical dip along a strike-slip fault can stop rupture from propagating through the entire fault due to the rupture deviation (Lozos 2021).This might affect the rupture propagating longer to the south direction, characterized by a high dip angle, than continuing rupture to the north, which has a shallower dip angle.

Analysis of region III (MTs solution in the southwest of Seram)
In the north, the aftershock distribution forms two distinct subclusters.Beside the north end cluster with strike-slip mechanisms (Region II), another cluster around 5 km further east shows a reverse fault mechanism striking NE-SW.This reverse fault is parallel with a surface fault mapped by Watkinson and Hall (2017) that was identified from a NE-SW lineament interpreted as an unknown fault from the Seram digital elevation model (SRTM).However, the fault is not identified in the most current geological study by Patria et al. (2021).Figure 10a The presence of a reverse fault, as well as faults with strike-slip mechanisms, might be linked to the Coulomb stress change (ΔCFS) calculated by Sahara et al. (2021a), who observed positive stress changes in the north, hypothesizing that high CFS concentrations at the northern tip of the main rupture could reactivate minor faults.Based on the Coulomb stress transfer (ΔCFF) calculation imparted by the mainshock to the existing fault, the NE-SW oriented fault in Seram also shows a stress increase value of more than 1 bar.This indicates that some of the reverse fault aligned NE-SW in Seram Island which experienced high-stress transfer had been reactivated.

Analysis of region IV (MTs solution in Ambon Island)
The onshore fault mapping by Patria et al. (2021) observed the presence of several active faults with dipslip mechanisms distributed along the regions located south of the Northern Banda Arc (i.e., Amahai (Amahai fault), Saparua (Saparua Fault), Ambon (Ambon Fault), and Buru (Rana Fault) (Fig. 11a).The major trends of these faults are in E-W, N-S, and NE-SW.We also inserted the CMT solution from Global CMT which is associated with inland fault from events with depth < 20 km and a magnitude more than Mw 5 in the period of 1976-2018.The strike orientation of the normal fault solution agrees with several faults (i.e., E-W oriented Amahai Fault and N-S oriented Saparua Fault).(1992) and Hallo et al. (2019Hallo et al. ( , 2017)).The number represented by black labels above the beachball corresponds to the event ID as shown in Table 2.In the current study, we observed three fault subclusters in Ambon Island (Fig. 11b, c): (1) the oriented NNW-SSE strike-slip segment with an average strike around 175° and with nearly vertical dip westward direction (SF1 fault); (2) the oriented NNE-SSW strike-slip segment with average strike around 189° with high dip angle (SF2 fault); (3) an oriented E-W normal fault segment (NF fault).
Based on those findings, our preferred interpretation is that the presence of two strike-slip segments with NNW-SSE and NNE-SSW trends might correlate with the regional structure trends in Halmahera, north of the study area (see Fig. 1a).The onshore volcano-chain segment in Halmahera is overlain by a graben-like median zone of an anticline and is bounded to the east by a sequence of steep NNW-SSE trending faults (Verstappen 1964;Passarelli et al. 2018).Hall (1987) also discovered a series of steep faults trending NE-SW and active since the Pliocene with dip-slip and strike-slip motion in Kau Bay.An identical strike-slip mechanism with either NNW-SSE or NEE-WSW is also observed on Buru Island from the CMT solution.Regarding the E-W normal fault segments in Ambon Island, we correlate this with the finding of displacement velocity by Patria et al (2021) in which they showed that the GPS velocity of Ambon and Banda Arc increases toward the east.This region's normal faults are reasonably related to this eastward velocity increase, which causes extensional deformation in the E-W strikes along the Amahai, Ambon, and Buru.In agreement with our interpretation, Spakman and Hall (2010) hypothesized a Banda Subduction rollback mechanism to explain extensional deformation in a converging zone.

Spatiotemporal seismicity of the aftershock sequences
Figure 12a-d shows the spatiotemporal seismicity of the 2019 Ambon aftershock.We compiled our aftershock data (18 October to 15 December 2019) with the earlier aftershock data (25 to 30 September 2019) from Sianipar et al. (2022), added by aftershock data (1 to 17 October 2019) from Meilano et al. (2021).The focal mechanism solutions for the first 20 days after the mainshock were obtained from Global CMT, while the solution for the next three periods was obtained from the current study.Figure 12a shows the aftershock distribution in the first 20 days after the mainshock.The distribution of aftershocks is mostly dominated by the north-south trend from Kairatu to the south on the Haruku Strait and the low intensity of the event appearance on Ambon Island.Based on the slip inversion results by Sahara et al. (2021a) and the coseismic slip model by Meilano et al. (2021), the mainshock unilaterally ruptured the South and this pattern is possibly making the N-S trends in the early period.The low intensity, as well as the sparse distribution of aftershocks in this period, could not give a clear lineament rupture in Ambon Island.However, during this period, we could observe that there were two significant aftershock events with magnitude > Mw 5.0 with two different mechanisms, i.e., normal fault around 3 km east of the N-S trend (labeled as Ag1) and strike-slip fault on Ambon Island (labeled as Ag 2).
In 21-40 days after the mainshock (Fig. 12b), in which 11 temporary local broadband seismometer was included in the monitoring, the aftershock distribution is wellconstraint in three rupture areas.In this period, we can observe the first strong lineament of the NNE-SSW of the Ambon cluster (shown by a dashed circle and defined by SF1 fault in Sect.5.3).There is also an indication of the first appearance of NE-SW aftershock trends in the southwest of Seram (black dashed circle) associated with the reverse fault mechanism.
For the next term, Fig. 12c shows the seismicity in the 41-60 days after the mainshock.The most significant feature that occurred in this term is the extension of rupture propagation in Ambon Island to the northeast direction.The existence of the normal fault cluster in Ambon Island was also observed in this sequence.In more detail, from Fig. 12b, c we could interpret that aftershock sequence in Ambon Island began with a rupture of the NNW-SSE strike-slip (SF1 in Fig. 11b) represented by the focal solution of A2.The rupture continued to the east and northeast just after the occurrence of the largest aftershock with Mw 5.1 on November 12th, 2019 at 10:10:40 (focal  11b).The E-W normal faults (NF in Fig. 11b) appear simultaneously with the SF2 strike-slip segment, characterized by both Mw 4.9 (focal solution A13) and Mw 4.4 (focal solution A14).In the main N-S cluster, the increase of aftershocks occurred intensively in the south end.Meanwhile, in the north, the seismicity is not so significant but it is still clear enough to describe the event propagation of both strike-slip and reverse fault events in the southwest of Seram.During the last period (61-80 days) after the mainshock (Fig. 12d), low-intensity of seismicity was observed in all three clusters.
Figure 13 below shows the temporal evolution of the aftershock in term of the cumulative number of events in which we associated it with the occurrence of the significant aftershock.We observed the most significant increasing event occurred in the first 5 days after the mainshock and was followed by a gradual decay until the break of the Mw 5.0 (11/10/2019 04:39:43 UTC) in Ambon Island.Besides this Mw 5.0 aftershock, two other significant aftershocks influenced most of the increased seismicity, i.e., Mw 4.7 (18/10/2019 19:44:24 UTC) and Mw 5.1 (12/11/2019 10:10:40 UTC), depicted by the steep increase of the cumulative lines (magenta circles).All these significant aftershocks occurred on Ambon Island.

Analysis of the mainshock MTs decomposition and the relation with the aftershock sequence
It is worth addressing how the ruptures occurred during the mainshock and aftershock and their relationship (i.e., joint movement or a series of different fault reactivation).In the 2019 Ambon Earthquake case, we found no indication of joint movements.From our results on the MTs decomposition of the mainshock (see Fig. 8), it might be revealed that the source of the mainshock consists of a major strike-slip and minor normal fault, which indicate that the N-S main fault cluster might observe the presence of a small portion of normal fault aside of strike-slip mechanism.However, all MTs solutions in this cluster consistently show strike-slip mechanisms which means normal faulting resulting from MTs decomposition is insignificant.This result also shows that the mainshock MTs solution from global CMT (84% of major strike-slip and 16% of minor normal fault in the scalar moment) is more reliable and reasonable compared with the GFZ that reflects the significant percentage of minor normal fault (31.9%).
On the other hand, according to Meilano et al. (2021), the earthquake ruptured two faults that are practically orthogonal to each other: the first fault going roughly north-south, and the second fault trending roughly east-west.This data raise the possibility that the Ambon earthquake was caused by conjugate faults that were active at the same time during the aftershock.However, in our monitoring (which has better coverage with local temporary networks), the presence of the E-W lineament was not detected or was not sufficient to be considered as a conjugate fault.We just found a small portion of the E-W lineament in Ambon Island.Our analysis of the MTs mainshock decomposition and the aftershock MTs solutions above also confirms that the possibility of the conjugate faults occurring in this 2019 Ambon case is unlikely.Although there are differences in interpretation, both studies consistently revealed the main N-S trends.Therefore, instead of joint movements of different faults, the complexity of the 2019 Ambon aftershock sequence may appear in the form of ruptured locations which clustered in three separated regions, each with its distinctive source mechanism.Thus, we prefer the hypothesis that the Ambon aftershock sequence occurred as a series of separate fault reactivations in a critically stressed area.

Conclusions
The focal mechanism solutions provide a better understanding of fault reactivation in a tropical region where significant surficial processes may erode the trace of active deformation.Using combined local and regional station data, we infer the MT solution for 20 aftershock events of the 2019 Ambon earthquake sequences.This was achieved by carefully selecting the stations used in the inversion, the quality of the waveform, the epicentral distance, and the frequency band utilized for inversion, to satisfy four out of five criteria based on Sokos and Zahradník (2013), including VR, CN, FMVAR, STVAR, and DC parameters.
Our MTs solution could reveal characteristic of fault reactivation in Ambon and Seram region.The N-S main rupture has a 35 km long orientation with the dextral strike-slip mechanism with the dipping changing from westward in the south to eastward in the north separated by a less seismic zone in the middle.Furthermore, we analyzed the fault mechanism of the two reactivated fault clusters from Ambon Island and southwest Seram.The presence of two strike-slip segments with NNW-SSE and NNE-SSW strike orientation in Ambon agree with the regional structure trends in Halmahera, located in the north of the study area, while the E-W oriented normal fault might be related to the eastward velocity increase in Ambon and Banda Arc, which causes extensional deformation.Moreover, the presence of a reverse fault about 5 km east of the N-S main fault in west Seram indicates reactivation of a previously interpreted lineament aligned NE-SW probably due to high-stress transfer imparted by the mainshock.
Analysis of spatiotemporal seismicity with the MTs solution suggests that the Mw 6.5 Ambon mainshock might consist of a geometrically complex rupture.The 2019 Ambon mainshock and its aftershock sequences can be explained as follows: (i) a Mw 6.5 mainshock rupture, characterized by the major strike-slip component Fig. 13 Temporal evolution of the 2019 Ambon aftershock, visualized by the cumulative number of events within all regions (magenta line) and the occurrence of the daily event ordered based on magnitude (gray circles).The red star shows the mainshock with a magnitude Mw 6.5.The day of the mainshock and several significant aftershock occurrences (which impact the seismicity most) is plotted as red lines and minor normal fault; (ii) Rupture propagated to the south and making up the first aftershock cluster along the main N-S ruptures, followed by the second cluster in Ambon; (iii) The reverse fault events cluster appeared in the southwest of Seram, in addition with increased seismic activity in the other two clusters.
The triggering of significant fault reactivations due to a relatively moderate Mw 6.5 2019 Ambon earthquake indicates that the faulting regime in the Ambon and Southwest Seram is critically stressed.Moreover, the fault reactivations observed in this study are located nearby the highly populated urban areas of Ambon City and Kairatu.We suggest the stress transfer generated along the fault segments passing through Ambon City in the south of the island should be modeled to estimate the seismic hazard.The reactivation of the small portion of reverse faults in the northern part of the Mw 6.5 main fault indicates that the stability of major NE-SW faults in Seram Island needs to be assessed further.Furthermore, this study provides important insight into active fault identification in Northern Banda Arc.

Fig. 2 a
Fig. 2 a Hypocenters of aftershock events based on a catalog produced by Sahara et al. (2021a; b) are plotted as orange circles (magnitude ≥ 4.0), blue circles (magnitude between 3.0 and 4.0), and grey circles (magnitude < 3).The area of investigation is divided into four regions denoted by green dashed rectangles.The corresponding labels in each rectangle represent the cluster number.The epicenter of the mainshock form BMKG (red star) and GCMT (magenta star) is also inserted.Fault segments from Patria et al. (2021) are plotted in red lines.b Final selected events for the MT inversion process, depicted by the blue rectangles

Fig. 3
Fig. 3 Distribution of seismic stations in the study area.The reverse yellow triangles indicate the ITB local stations, the reverse blue triangles indicate the BMKG regional stations, while the reverse yellow-blue one shows the local-regional stations located close to each other.The aftershock epicenters from Sahara et al. (2021a; b) are plotted as black circles, while the magenta and red star represent the mainshock from GCMT and BMKG (relocated), respectively The centroid moment tensor (CMT) solution from Global CMT (navy beach ball) is also present for a more accurate comparison.The results indicate that the crustal model proposed bySahara et al. (2021a) has the highest correlation value and provides a more stable solution which is relevant to the Global CMT results.Although theSabtaji and Nugraha (2015) model has a fairly high correlation, the focal mechanism solution is not consistent if compared to the solutions presented by the crustal model fromSahara et al. (2021a).The correlation values for the PREM and Koulakov models are relatively low (below 0.5).Furthermore, the MT solutions using koulakov model have a different mechanism from the global CMT reference solutions, which could lead to misinterpretation.Finally, we used the crustal model from Sahara et al. (2021a) to calculate Green Function calculation for all events during the inversion process.

Figure 8
Figure 8 below shows the result of the MT decomposition from two MT solutions derived from the global CMT and GFZ.In general, both MTs solutions were decomposed

Fig. 4
Fig. 4 Velocity models used in the MT inversion test, consist of Vp (left) and Vs values (right) in km/s for each layer depth in km.The four models consist of: (i) Sahara et al. (2021a), shown by green lines; (ii) Koulakov et al. (2007), shown by magenta lines, (iii) Sabtaji and Nugraha (2015), shown by red lines, and (iv) PREM model (Dziewonski and Anderson 1981), shown by blue lines

Fig. 5
Fig. 5 MT solutions in a correlation diagram by implementing different crustal models.The graphic shows the focal mechanism solution derived from ten different trial source numbers, as well as the correlation values between observed and synthetic waveforms.The crustal model from Sahara et al. (2021a), Koulakov et al. (2007), Sabtaji and Nugraha (2015), and PREM model are, respectively, shown by green, magenta, red, and blue beach ball solutions.The MT solution from Global CMT (navy beach ball) is also inserted as a reliable comparison

Fig. 6
Fig. 6 Inversion results for the Mw 5.1 aftershock earthquakes (corresponding to Event ID 11 Table 2) using ISOLA.a The correlation between the observed (black) and synthetic (red) waveforms for all the components of all the stations are shown.VR values for each station are indicated in blue.The frequency range used for this earthquake is 0.04-0.10Hz and is shown at the top of the figure.b A plot between the source time shift and trial source number in correlation contours.The color scale of the correlation values is shown on the right.The maximum correlation is obtained for a source time shift of ∼1.5 s.The resultant beachball is increased in size and shown in red.c The plot of correlation with depth is shown.Maximum correlation is obtained for a focal depth of 12 km with 94.5% DC of the beach ball.d Map of stations used in the MT inversion, showed by a red reverse triangle.The event epicenter is shown as a black star Fig.7a Map of the focal mechanism solutions derived in this study.The MTs solutions are plotted with colored beachballs adapted fromFrohlich (1992) andHallo et al. (2019Hallo et al. ( , 2017)).The number represented by black labels above the beachball corresponds to the event ID as shown in Table2.Hypocenter locations fromSahara et al. (2021a; b)  are plotted in grey circles based on magnitude size.The blue dashed rectangle shows the region of investigation.The land fault data were derived fromPatria et al. (2021), depicted by red lines.b Distribution of all fault plane solutions of aftershock events in a ternary diagram, adapted fromFrohlich (1992) andHallo et al. (2017Hallo et al. ( , 2019) ) (See figure on next page.)

Fig. 8
Fig. 8 Major and minor DC MT decomposition of the non-DC MT solution for the Mw 6.5 Ambon mainshock shown by the trinity of beach-ball: a GFZ Potsdam, b Global CMT solution.The size of the beach ball is proportional to their respective scalar seismic moment

Fig. 9 a
Fig. 9 a Map view of focal mechanism distribution along N-S main rupture.The blue dashed rectangles show the area of the beachball and the aftershock hypocenter covered by lines A-A' and B-B' (depicted by blue lines) in a radius of 6 km.The red straight lines show the fault segments mapped by Patria et al. (2021).The red dashed circle depicts the less seismic zone in the central part of region I. b Vertical cross-sections of focal mechanism solutions and hypocenter distributions along lines A-A' .c Vertical cross-sections of focal mechanism solutions and hypocenter distributions along lines B-B' .Focal mechanisms obtained in this study are plotted as colored beachballs based on Frohlich (1992) and Hallo et al. (2017, 2019) with numeric labels to indicate the event ID.The size of the circles represents event magnitudes in Mw.The red dashed lines are the interpreted fault planes derived in this study.The magenta circle plotted in (a-c) is the event hypocenter of the corresponding beach ball solution

Fig. 10 aFig. 11 a
Fig. 10 a Map view of focal mechanism distribution in the northern part of the study area.The blue dashed rectangles show the area of the beachball and the aftershock hypocenter covered by lines C-C' (depicted by a blue line) in a radius of 6 km.The red straight lines show the fault segments mapped by Patria et al. (2021).b Vertical cross-sections of focal mechanism solutions and hypocenter distributions along lines C-C' .Focal mechanisms obtained in this study are plotted as colored beachballs based on Frohlich (1992) and Hallo et al. (2017, 2019, ) with numeric labels to indicate the event ID.The size of the circles represents event magnitudes in Mw.The red dashed lines are the interpreted fault planes derived in this study.The magenta circle plotted in (a) and (b) is the event hypocenter of the corresponding beach ball solution

Fig. 12
Fig. 12 Spatiotemporal seismicity of the 2019 Ambon aftershock sequences with the focal mechanism solution between 25 September and 15 December 2019 (0-80 days after mainshock): a day 0-20 after the mainshock, the mainshock epicenter from BMKG (relocated) is shown by a red star.b day 21-40 after the mainshock.c day 41-60 after the mainshock.d day 61-80 after the mainshock.The blue circles show the hypocenter distribution in the active period and its size represents the magnitude value in Mw.The hypocenter catalog for Fig. 13a is compiled from Meilano et al. (2021) and Sianipar et al. (2022).The focal mechanism solutions for Fig. 13a are derived from Global CMT (labeled as the letter 'Ag'), while Fig. c-d used our MTs solution (labeled as letter 'A' followed by the event ID).The red straight lines show the fault segments mapped by Patria et al.(2021)

Table 1
List of processed earthquakes from the Ambon

Table 2
Parameters of MTs solutions for 20 selected aftershock events with the quality parametersThe bold and underlined numbers indicate that the quality parameter values are out of the criteria a Here, VR represents the average VR of all station's components used in the MT inversion

Event ID strike1/dip1/rake1 ( o ) strike2/dip2/rake2 ( o ) frequency band (Hz) Magnitude
Fig. 7 a Map of the focal mechanism solutions derived in this study.The MTs solutions are plotted with colored beachballs adapted from Frohlich