https://orcid.org/0000-0001-8746-8615
SUMMARY
On 2015 November 4, an Mw 6.5 earthquake struck the east region of Alor Island, eastern Indonesia. Here, we use Sentinel-1 and ALOS-2 data to explore the coseismic surface displacement of this earthquake. Based on the ALOS-2 coseismic interferograms, the first fault model and coseismic slip distribution of the 2015 Alor earthquake are presented in this study. The preferred slip model links to a blind south-southeast striking, south-southwest dipping strike-slip fault with amount of normal slip component and a peak slip of 2.09 m at 2.34 km of depth. Considering the fact that the ascending and descending InSAR prediction of the distributed slip model does not fit the InSAR observations at the northwest and southeast tips with the simple planar fault model. We tried to construct a strike-variable fault model to further improve data fit. The results of the calculated Coulomb stress change imply that the regions with positive CFS changes mainly located at the northwest and southeast extremities of the rupture of the Alor earthquake, and in the lobes north and south of the rupture. The most striking discovery from the InSAR observations of the Alor earthquake is that most of the displacements occurred on a fault whose existence was unknown before the earthquake.
1 INTRODUCTION
On 2015 November 4 (UTC Time 03:44:15), a magnitude 6.5 earthquake struck the east region of Alor Island, eastern Indonesia. This earthquake caused at least three injuries and hundreds of houses damaged. More than 40 landslides triggered by the earthquake were identified from pre- and post-earthquake Sentinel-2 and Landsat-8 optical images (Supplementary Text S1, Figs S1–S2). Focal mechanism solutions from the United States Geological Survey (USGS) indicate that the rupture occurred on either a right-lateral oblique-normal slip rupture dipping to the southwest or a left-lateral oblique-normal slip fault dipping to the east. Both nodal plane results demonstrate that the earthquake rupture had a normal slip component. The USGS W-phase solution shows that the 2015 Alor earthquake released the seismic moment of 7.43 × 1018 N m, which nucleated at a centroid depth of 11.5 km. The aftershocks from 2015 November 4 to 2016 May 4 included 21 events with magnitudes greater than 4.0, and the largest aftershock was the Mw 5.3 event on 2015 November 4. The 2015 Alor earthquake is the first Mw ≥ 6.5 oblique-slip earthquake to have occurred within the ashore of Alor Island in 40 yr of instrumentally recorded history, providing a rare opportunity to study crustal deformation in this area.
Alor Island, the site of the earthquake, is located in a complex tectonic setting in eastern Indonesia. This setting was formed by the subduction of the Australian Plate beneath the Eurasian Plate (Fig. 1), and arc-continent collision is one of the most prominent tectonic processes occurring in this area (Katili 1975; Nishimura & Suparka 1986; Breen et al. 1989; Hantoro et al. 1994; Tsuji et al. 1995). North–South tectonic convergence, driven by the subduction of the Timor Trough to the north and backarc thrusts to the south, is regionally accommodated by the slip on the Wetar thrust, a major backarc thrust structure located offshore to the north of Alor Island (Nishimura & Suparka 1986; Breen et al. 1989; McCaffrey 1996). The 350-km long, East–West trending Wetar backarc thrust is a product of the collision between Australia and the Indonesian island arc, which is responsible for most of the large earthquakes near and below Alor Island (Breen et al. 1989). The Wetar Zone, one of the three distinct zones of the Banda arc, is an inactive zone bounded by two active segments, the Flores zone in the west and the Damar Zone in the east (Jones et al. 2014). This collision between the Australian plate and the Sunda plate has resulted in a lack of volcanic activity in the Wetar Zone (Katili 1975) and a moderately deep seismicity gap, which is in contrast to continuous deep seismicity below all three zones of the arc (Jones et al. 2014). Although the Wetar zone in the Banda arc is relatively inactive, some motion of Alor Island relative to the Australian plate is likely mainly accommodated by thrusting along the western section of the Wetar thrust fault, north of the Alor Island. Seismicity is related to both northward subduction of the Australian plate and southward thrusting of the islands of Wetar, but the rarity of thrust earthquakes in the forearc compared to those in the backarc suggests that subduction in the forearc is occurring aseismically, or that the backarc thrust zones are more actively slipping than the Timor trough (Jones et al. 2014). In the past four decades, there has been little seismicity larger than magnitude 4.0 inland of Alor Island, while the seismicity mainly concentrates offshore near the Flores and Wetar Thrust fault. This raises the question of what kind of strain build-up is taking place on the island, as it is located in such a complex deformation zone.
Tectonic background of the 2015 November 4 Alor earthquake. The white and red rectangles outline the footprint of the Sentinel-1 and ALOS-2 SAR data, respectively. S1-T39A and S1-T141A denote Sentinel-1 ascending track 39 and 141, respectively. The beach balls denote historical earthquakes from the Global Centroid Moment Tensor (GCMT) catalogue (Dziewonski et al. 1981), which are coloured by depth. The red star denotes the location of the 2015 Alor earthquake from GCMT. The blue arrows represent the interseismic GPS velocity (Koulali et al. 2016). The fault data (Flores Thrust, Wetar Thrust and Timor Trough) is from Koulali et al. (2016). The magenta box in the inset map outlines the area in the main figure.
In this study, we use coseismic InSAR observations to investigate the 2015 Alor earthquake from the following aspects: (1) Mapping the coseismic deformation of the 2015 Alor earthquake with C-band and L-band SAR interferometry (InSAR); (2) Decomposing the ascending and descending LOS data into coseismic horizontal and vertical displacement; (3) Determining the optimal fault parameters and coseismic slip distribution of the earthquake by using InSAR observations; (4) Assessing the landslide caused by this earthquake. Finally, some discussions, including the resolution test of the InSAR data, possible strike-variable fault model and assessment of future seismic risk in this area, as well as the understanding of the strike-slip faulting in a convergent boundary, are provided.
2 ESTIMATION OF STRAIN RATES AND RIGID ROTATION COMPONENT IN EPICENTRAL AREA
To evaluate the deformation mode and magnitude near the Alor earthquake area, we employed the regional GPS velocity field of east Indonesia (Koulali et al. 2016; Fig. 1) to calculate the crustal deformation strain rates in the Indonesian island arc region. Using the stations from the Sunda Block-fixed reference frame solutions, we constructed a Delaunay triangulation and estimated the horizontal strain rate components for each triangle via a linear least-squares inversion, with the horizontal velocities as the input (Savage et al. 2001). The results indicate that the area with highest values for the second invariant and max shear strain rate lies in the northern part of the island arc, directly facing the Flores and Wetar thrust faults (Fig. 2). These areas are also known to be the locations of historical earthquakes (Fig. 1). The Flores thrust fault is characterized by compressional rates, while the Wetar thrust has extension rates in its west segment and compressional rates in the east. Alor Island lies in the western section of the Wetar thrust fault, where E–W-oriented extensional processes dominate, reaching 98 nanostrain yr−1. The compression orientation at Alor Island is perpendicular to the coast line. Historically, the majority of earthquakes that have occurred offshore of Alor Island, along the western section of the Wetar thrust fault, have had dip-slip thrust mechanisms. Earthquakes on Alor Island, however, are quite rare.
Strain rate field of the east Indonesia deduced from GPS velocities. (a) Principal axes of strain rates and the second invariant of strain rate tensor (the colour in the background); (b) maximum shear strain rate. The maximum shear strain rate is dextral relative to the fault.
Relative to Sunda Block, the GPS velocity field of the east Indonesia includes a rigid rotational component and an interior deformation component (e.g. Gan et al. 2007). In order to take out the rigid rotation so as to highlight the block rotation which could be the reason to load interseismic strain energy on the strike-slip fault within a convergent zone, we solved for the Euler vector that minimized the RMS velocity of the 28 GPS stations in Supplementary Fig. S3. The resulting Euler pole has a location and magnitude of (114.55°±0.94° E, -6.44°±1.67° N, 2.68°±0.2951° Ma−1). Supplementary Fig. S3 shows the estimated rigid rotation velocities and the observed velocities in the Sunda Block-fixed reference frame. The results show that the rigid rotation velocity field (Supplementary Fig. S3b) is a first-order approximation to the GPS velocity field, except for the north of the Flores and Wetar thrust fault where slow anticlockwise movement dominates the crustal deformation. For the south of the Flores and Wetar thrust fault, crustal motion can be well modelled near the Alor island, which suggest that the block rotation could be the reason to load interseismic strain energy in this area.
3 COSEISMIC DEFORMATION AND MODELLING
3.1 Sentinel-1 and ALOS-2 SAR data
For the 2015 Alor earthquake, there are two Sentinel-1 and ALOS-2 tracks flying over the epicentral area, of which ascending track 141 of Sentinel-1, ascending track 124 and descending track 22 of ALOS-2 cover the entire earthquake deformation area. We collected SAR images encompassing the 2015 Alor earthquake from the Sentinel-1 Interferometric Wide Swath (IWS) and ALOS-2 Fine mode. Due to the inadequate Sentinel-1 IWS data acquisitions over Alor Island, there were no descending track data obtained prior to the earthquake. Two ascending track Sentinel-1 interferograms (track 039 and 141) were acquired to show the coseismic deformation, but both interferograms were noisy (Supplementary Fig. S4), which could be the result of temporal decorrelation and the dense vegetation cover on Alor Island. Nevertheless, we were still able to observe a clear coseismic deformation signal in the Sentinel-1 interferograms. Compared to C-band InSAR, the ALOS-2 interferograms showed excellent coherence, even in heavily vegetated areas (Funning & Garcia 2019; Morishita 2019). Thus, for this study, we collected ascending track 124 and descending track 22 ALOS-2 data, in addition to the two ascending Sentinel-1 images, to map the coseismic surface displacement of the Alor earthquake. The detailed information of the SAR data used can be found in Table 1.
Details of SAR images used in this study.
| Satellite | Track | Pass | Reference | Secondary | Perp. B (m) | Inc. Angle | Azi. Angle |
|---|---|---|---|---|---|---|---|
| Sentinel-1 | 141 | Ascending | 2015/10/22 | 2016/11/27 | −55.3 | 33.6 | −12 |
| 39 | Ascending | 2015/10/15 | 2015/12/02 | −2.6 | 43.7 | −12 | |
| ALOS-2 | 124 | Ascending | 2015/9/22 | 2016/2/9 | 57.4 | 30.0 | −12 |
| 22 | Descending | 2015/3/4 | 2016/3/2 | −18.5 | 30.5 | −168 |
| Satellite | Track | Pass | Reference | Secondary | Perp. B (m) | Inc. Angle | Azi. Angle |
|---|---|---|---|---|---|---|---|
| Sentinel-1 | 141 | Ascending | 2015/10/22 | 2016/11/27 | −55.3 | 33.6 | −12 |
| 39 | Ascending | 2015/10/15 | 2015/12/02 | −2.6 | 43.7 | −12 | |
| ALOS-2 | 124 | Ascending | 2015/9/22 | 2016/2/9 | 57.4 | 30.0 | −12 |
| 22 | Descending | 2015/3/4 | 2016/3/2 | −18.5 | 30.5 | −168 |
Details of SAR images used in this study.
| Satellite | Track | Pass | Reference | Secondary | Perp. B (m) | Inc. Angle | Azi. Angle |
|---|---|---|---|---|---|---|---|
| Sentinel-1 | 141 | Ascending | 2015/10/22 | 2016/11/27 | −55.3 | 33.6 | −12 |
| 39 | Ascending | 2015/10/15 | 2015/12/02 | −2.6 | 43.7 | −12 | |
| ALOS-2 | 124 | Ascending | 2015/9/22 | 2016/2/9 | 57.4 | 30.0 | −12 |
| 22 | Descending | 2015/3/4 | 2016/3/2 | −18.5 | 30.5 | −168 |
| Satellite | Track | Pass | Reference | Secondary | Perp. B (m) | Inc. Angle | Azi. Angle |
|---|---|---|---|---|---|---|---|
| Sentinel-1 | 141 | Ascending | 2015/10/22 | 2016/11/27 | −55.3 | 33.6 | −12 |
| 39 | Ascending | 2015/10/15 | 2015/12/02 | −2.6 | 43.7 | −12 | |
| ALOS-2 | 124 | Ascending | 2015/9/22 | 2016/2/9 | 57.4 | 30.0 | −12 |
| 22 | Descending | 2015/3/4 | 2016/3/2 | −18.5 | 30.5 | −168 |
3.2 InSAR data processing
The data processing was conducted using the GAMMA InSAR processing software (Werner et al. 2000). All of the interferograms were generated from the Single Look Complex (SLC) products, with the multilook ratios between the range and azimuth directions being 10:2 for the Sentinel-1 and 8:8 for ALOS-2 data. ALOS-2 Fine mode interferometry is similar to the traditional interferometry. For the TOPS interferometry of Sentinel-1, co-registration of several thousandths of a pixel accuracy in the azimuth direction was necessary to avoid phase jumps between subsequent bursts (González et al. 2015). Thus, a method for taking into account the effect of the scene topography and a spectral diversity method (Scheiber & Moreira 2000) considering the interferometric phase of the burst overlap region were used to ensure very high co-registration accuracy. After obtaining the high-quality co-registration between the TOPS SLC data, the topography effects were removed from the interferograms using precise orbits together with the 30 m resolution Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) (Farr et al. 2007). However, slight height-related signals still remained in the interferograms. To reduce the effect of phase noise, we then applied a power spectrum filter (Goldstein & Werner 1998) to the interferograms, followed by an unwrapping process executed with the minimum cost flow method (Chen & Zebker 2001). Finally, the interferograms were geocoded to the WGS84 geographic coordinates with 30 m resolution. A linear function among the location (x, y), elevation (h), and error phase was estimated with observations away from the deformed areas to eliminate both the residual orbital errors and topography-correlated atmospheric delays (e.g. Feng et al. 2019; Wen et al. 2021). The decorrelated regions (white areas in the line-of-sight (LOS) displacement maps) extended approximately 1–2 km off the rupture, likely indicating a zone of high damage and ground disruption around the fault.
Sentinel-1 interferograms show significant noise levels due to the effects of temporal decorrelation, while ALOS-2 images present a high level of interferometric correlation and a clear coseismic deformation signal due to the longer wavelength of the L-band SAR. Two prominent features of the coseismic deformation can be seen in both ascending and descending ALOS-2 interferograms, one is a steep fringe gradient towards the southwest edge of the interferograms, which indicates the dip direction of the seismogenic fault; another is similar deformation pattern, suggestion that a large amount of vertical surface displacements occur. The interferograms imply that the earthquake source of the 2015 Alor earthquake may be located near the dense high-fringe gradient areas at the eastern end of Alor island, which is further east of the USGS epicentre (Funning & Garcia 2019; Morishita 2019). A detailed description of InSAR coseismic deformation can be found in Supplementary Text S2.
3.3 Quasi-horizontal and vertical deformation field
To gain further insight into the surface deformation caused by the rupture of the Alor earthquake, we attempted to retrieve the quasi-horizontal and quasi-vertical deformation field by decomposing ascending and descending ALOS-2 InSAR data. This method takes advantage of the fact that InSAR LOS data are most sensitive to vertical displacements, moderately sensitive to East–West displacements, and weakly sensitive to North–South displacements (Fialko & Simons 2001). By assuming the North–South displacements were equal to zero, we were able to use InSAR data from two different look directions to invert the quasi-horizontal and quasi-vertical deformation (Fujiwara et al. 2000; Wen et al. 2016). The decomposing results highlighted the significant horizontal and vertical deformation caused by the coseismic rupture (Fig. 3), with ranges of the quasi East–West and vertical displacements of −75.4 cm ∼ 51.4 cm and −63.4 cm ∼ 30.7 cm, respectively. The deformation pattern of the quasi East–West displacement resembled the shape of a butterfly, with the left part of the surface moving to the west and the right part to the east, which is consistent with the surface deformation caused by the rupture of a right-lateral strike-slip fault. The vertical deformation was primarily characterized by surface subsidence, forming a long and narrow deformation area. This deformation pattern is consistent with the type of surface deformation caused by a normal fault. The horizontal and vertical displacements suggest that a large amount of strike-slip and normal slip may have been present in the earthquake rupture.
Coseismic displacement of (a) quasi East–West and (b) quasi upward; (c, d) represent deformation along the two profiles (AA’ and BB’) shown in (a) and (b). The black dashed line in (c) and (d) denote the location of seismogenic fault.
3.4 Modelling
In order to gain a better understanding of the seismogenic structure of the 2015 Alor earthquake, a classical two-step inversion method was used to invert for the location and geometry of the rupturing fault and retrieve the detailed slip distribution (Wright et al. 2004; Funning et al. 2005; Atzori et al. 2009; Feng et al. 2014; Xu 2017; Wen et al. 2019). Since Sentinel-1 InSAR data had a low signal-noise ratio (SNR), only the ALOS-2 data were used for modelling. To improve the computational efficiency of the inversion process, a resolution-based downsampling algorithm (Lohman & Simons 2005) was employed to reduce the large number of InSAR observations to 989 points for the ascending track and 1220 points for the descending track.
For a first-order description of the Alor earthquake, we model the downsampled InSAR data with a rectangular dislocation in a homogeneous elastic half-space (Okada 1985), with a uniform slip fault patch used to model the coseismic deformation. To search for the optimal parameters of the fault model, a Bayesian inversion algorithm was used, which is available in the open source package, Geodetic Bayesian Inversion Software (GBIS) (Bagnardi & Hooper 2018). This algorithm, which is based on the Bayesian framework, integrates a Markov chain Monte Carlo algorithm with the Metropolis-Hastings algorithm (Hastings 1970) to effectively sample the posterior probability distribution of each unknown parameter (Bagnardi & Hooper 2018). The focal mechanism reported by USGS was used as a guide to set a search region for the optimum parameters. The seismic solution provided two possible nodal planes: a right-lateral strike-slip fault with abundant normal-slip component dipping west, and a left-lateral strike-slip fault with some normal-slip component dipping east. Since the LOS displacement field was more aligned with the rupture of a strike-slip fault dipping west, the nodal plane dipping east was excluded. The search boundaries were then set based on the nodal plane dipping west of the seismic solution.
Based on the determined fault geometry, we extended the fault length to 25 km along the strike direction and fault width to 16 km along the dip direction, subsequently dividing the entire fault plane into 400 patches with a size of 1 km × 1 km. We employed a constrained least-squares algorithm to retrieve the detailed slip distribution on the fault plane. To prevent the occurrence of unphysical oscillating in the slip distribution, Laplacian smoothing was applied in the linear inversion. We use the covariance matrix estimated through the 1-D covariance function to weigh the InSAR data (e.g. Funning et al. 2005). As two tracks of ALOS-2 InSAR data were utilized in the inversion, we introduce the concept of relative weight to reconcile the different contributions of the two tracks. To solve the parameters of the smooth factor employed in the Laplacian smoothing operation and the relative weight, we incorporated a Helmert Variance Component estimating method (HVCE) (Xu et al. 2010) in the linear inversion procedure. Finally, we resort to a Monte Carlo technique to estimate uncertainties of the slip distribution (e.g. Wright et al. 1999; Funning et al.2005).
4 RESULTS
The best-fitting uniform slip model suggests that the earthquake occurred on a southwest dipping fault with a strike of 117.01° and dip of 69.35° (Table 2). The uniform slip model produces root-mean-squares-errors (RMS) of approximately 6.2 and 7.0 cm for ALOS-2 ascending and descending InSAR data, respectively (Supplementary Fig. S5). The optimal distributed slip model suggests that the maximum slip of 2.09 m occurred at a depth of 2.34 km (Fig. 4), with the major slip area concentrated between 1 and 6 km. The released moment from the distributed slip model is 4.73 × 1018 N m, approximately equivalent to a moment magnitude of Mw 6.4, which is slightly lower than the GCMT and USGS results. The best-fitting distributed slip model shows good agreement with the InSAR data, with RMS of 4.2 and 6.1 cm for ALOS-2 ascending and descending data, respectively (Fig. 5). However, we also noted that there is one residual fringe near the southeast edge of fault plane, which may result in a lower moment magnitude determined by the distributed slip model.
Distributed coseismic slip model. (a) coseismic slip model projected on the surface. The orange dots denote the aftershocks from USGS. The red beach ball denotes the focal mechanism of the 2015 Alor earthquake from GCMT. The red and yellow stars represent the location of the main shock from GCMT and USGS, respectively. The red solid line outlines the surface projection of the fault trace. (2) Perspective view of slip distribution from the southwest. The arrows show the magnitude and direction of fault motion.
InSAR observations and predictions of distributed coseismic slip model. (a) ascending and (d) descending ALOS-2 data. (b) and (e) are predicted InSAR data from the optimal distributed coseismic slip model. (c) and (f) are residuals. (g–i) LOS displacements along the three profiles (AA’, BB’ and CC’) shown in (a) and (d). The black rectangle denotes the surface projection of the fault. The dashed red line denotes the projection of the fault trace. The orange dots denote the aftershocks from USGS. The red beach ball denotes the focal mechanism of the 2015 Alor earthquake from GCMT. The red and yellow stars represent the location of the main shock from GCMT and USGS, respectively.
Fault parameters of the 2015 Alor earthquake.
| Source | Longitude (°) | Latitude (°) | Depth (km) | Strike (°) | Dip (°) | Rake (°) | Length (km) | Width (km) | Slip (m) | Seismic moment (1018 Nm) | Mw |
|---|---|---|---|---|---|---|---|---|---|---|---|
| USGS | 124.875 | −8.338 | 20 | 115/11 | 61/67 | −154/−32 | − | − | − | 7.44 | 6.5 |
| GCMT | 124.95 | −8.20 | 12 | 113/11 | 56/73 | −159/−36 | − | − | − | 8.15 | 6.5 |
| Uniform slip model | 124.973 [-0.01/0.02 (km)] | −8.230 [-0.01/0.01(km)] | 0.45 [-0.01/0.01] | 117.01 [-0.08/0.10] | 69.35 [-0.53/0.45] | −149 [-0.87/+1.60] | 7.00 [-0.04/0.04] | 10.13 [-0.23/0.29] | 1.62 | 3.45 | 6.3 |
| Source | Longitude (°) | Latitude (°) | Depth (km) | Strike (°) | Dip (°) | Rake (°) | Length (km) | Width (km) | Slip (m) | Seismic moment (1018 Nm) | Mw |
|---|---|---|---|---|---|---|---|---|---|---|---|
| USGS | 124.875 | −8.338 | 20 | 115/11 | 61/67 | −154/−32 | − | − | − | 7.44 | 6.5 |
| GCMT | 124.95 | −8.20 | 12 | 113/11 | 56/73 | −159/−36 | − | − | − | 8.15 | 6.5 |
| Uniform slip model | 124.973 [-0.01/0.02 (km)] | −8.230 [-0.01/0.01(km)] | 0.45 [-0.01/0.01] | 117.01 [-0.08/0.10] | 69.35 [-0.53/0.45] | −149 [-0.87/+1.60] | 7.00 [-0.04/0.04] | 10.13 [-0.23/0.29] | 1.62 | 3.45 | 6.3 |
Note. The uniform slip model shows a maximum posteriori probability solutions of fault parameters and 2.5 and 97.5 percentiles of posterior PDFs of model parameters (in squared brackets). The location (longitude and latitude) of the uniform slip model derived in this study represents the midpoint of the upper boundary of the seismogenic fault.
Fault parameters of the 2015 Alor earthquake.
| Source | Longitude (°) | Latitude (°) | Depth (km) | Strike (°) | Dip (°) | Rake (°) | Length (km) | Width (km) | Slip (m) | Seismic moment (1018 Nm) | Mw |
|---|---|---|---|---|---|---|---|---|---|---|---|
| USGS | 124.875 | −8.338 | 20 | 115/11 | 61/67 | −154/−32 | − | − | − | 7.44 | 6.5 |
| GCMT | 124.95 | −8.20 | 12 | 113/11 | 56/73 | −159/−36 | − | − | − | 8.15 | 6.5 |
| Uniform slip model | 124.973 [-0.01/0.02 (km)] | −8.230 [-0.01/0.01(km)] | 0.45 [-0.01/0.01] | 117.01 [-0.08/0.10] | 69.35 [-0.53/0.45] | −149 [-0.87/+1.60] | 7.00 [-0.04/0.04] | 10.13 [-0.23/0.29] | 1.62 | 3.45 | 6.3 |
| Source | Longitude (°) | Latitude (°) | Depth (km) | Strike (°) | Dip (°) | Rake (°) | Length (km) | Width (km) | Slip (m) | Seismic moment (1018 Nm) | Mw |
|---|---|---|---|---|---|---|---|---|---|---|---|
| USGS | 124.875 | −8.338 | 20 | 115/11 | 61/67 | −154/−32 | − | − | − | 7.44 | 6.5 |
| GCMT | 124.95 | −8.20 | 12 | 113/11 | 56/73 | −159/−36 | − | − | − | 8.15 | 6.5 |
| Uniform slip model | 124.973 [-0.01/0.02 (km)] | −8.230 [-0.01/0.01(km)] | 0.45 [-0.01/0.01] | 117.01 [-0.08/0.10] | 69.35 [-0.53/0.45] | −149 [-0.87/+1.60] | 7.00 [-0.04/0.04] | 10.13 [-0.23/0.29] | 1.62 | 3.45 | 6.3 |
Note. The uniform slip model shows a maximum posteriori probability solutions of fault parameters and 2.5 and 97.5 percentiles of posterior PDFs of model parameters (in squared brackets). The location (longitude and latitude) of the uniform slip model derived in this study represents the midpoint of the upper boundary of the seismogenic fault.
5 DISCUSSION
5.1 Checkerboard tests
Inversion stability and data resolution are commonly examined using checkerboard tests. The resolution analysis comprises of forward calculations to simulate synthetic surface displacements caused by a given slip distribution on a given fault, and a subsequent inversion to recover the input slip model. Therefore, we build an input slip model in which the fault plane uses the optimal fault geometry. The slip patch has 2 × 2 subfaults. Each subfault has a slip of 1 m with a pure dextral strike-slip movement (Fig. 6). We then compute the synthetic LOS displacements at the InSAR sites for the input slip model, and add Gaussian noise with the same standard deviations as the data errors. Some regularization is applied to the checkerboard inversions to stabilize the results, similar to what we apply to actual data. The inversion results show that the slip distribution inferred from InSAR data has a better resolution at shallow depth (0–2 km) (Fig. 6). As the depth increases, the resolution of InSAR data decreases gradually. Slip at the right corner of the shallowest row is underestimated. Between 3 and 5 km depth, slip tends to be smeared, and the along-strike patchiness is not well resolved. Further down, along-strike change of slip is almost indistinguishable.
Checkerboard test of the inversion resolution. The input pattern has alternating patches of 2 m of slip and no slip. (a) Input model; (b) The results of inverting synthetic InSAR data. The purple arrows indicate slip vectors.
5.2 Branching of fault geometry at the north-western and south-eastern end of the seismogenic fault
We noted that the ascending and descending prediction of the distributed slip model does not fit the observations well at the northwest and southeast tips with the simple planar fault model (Fig. 5). Therefore, we suspect that weather the fault geometry used in the previous modelling needs further adjustment. Here, we try to discuss about the detailed fault geometry at the north-western and south-eastern end of the seismogenic fault. From the modelling result of distributed slip model (Fig. 5g), we can see that both ascending and descending prediction are overestimated across the fault along profile AA’, and the projected fault trace of the distributed slip cut across the outmost fringe in both ascending and descending InSAR observations at the northeast end (Figs 5a and d). Considering the fact that, the modelling fit well along the profile BB’, we suspect that the fault geometry at the north-western end should be modified. At the south-eastern end of the seismogenic fault, the descending prediction overestimated at the southwest part along the profile CC’, and the ascending prediction show opposite trend compared with ascending InSAR observation at the distance of 0 km along the profile CC’ (Fig. 5i). We also can see a residual fringe at the south-eastern end of the seismogenic fault (Figs 5c and f). Therefore, we also wonder weather the level of data fit can be further improved through modifying the fault geometry at the south-eastern end of the seismogenic fault. To further verify the possibility of branching at both ends of the fault, we carried out five sets of test scenarios (See Supplementary Text S3). The results indicate that taking the fault branching at both ends into fault modelling can further improve the data fitting in these area.
From the ALOS-2 ascending and descending InSAR data, we can see that there is a strike variation at the maximum deformation gradient from northwest to southeast, which can be seen in the observations of profiles (Fig. 7). Then, we construct a strike-variable fault of discrete rectangular patches (Mildon et al. 2016) to refine the coseismic slip model. We used the location of the steep displacement gradient to outline the strike-variable surface projection of the fault. From the surface projection of fault extracted based on the ascending and descending InSAR data, a projection direction (the mean fault strike) is assigned to subfaults (Mildon et al. 2016). The initial fault dip we adopted is based on the best-fitting uniform slip model. The strike-variable fault geometry is composed of 560 rectangle patches with a size of 1 km × 1 km. Because there are dislocation gaps and overlaps in the strike-variable fault model with rectangle patches (Jiang et al. 2013), we also construct a strike-variable fault of discrete triangle patches, which can avoid this phenomenon. We first determined the fault cover area based on the above models and then discretized its 3-D geometry into triangles of approximately 1 km sides (Fukushima et al. 2018). The two strike-variable fault models provide a similar coseismic slip pattern compared to the simple planar model, but an obvious difference appears at the location where the strike changes (Fig. 8). The two models all provided a better fit to the ascending and descending InSAR data at the northwest and southeast part, where the simple planar model does not fit well. Although the strike-variable models provide a better data fitting compared to the simple planar fault model, the data fitting along profiles AA’ and CC’ is still not satisfactory. Considering the fact that the poor data fitting along the profils mainly occurs near the area across the fault, we suspect that whether changing the fault dip can further fit the profile data. Then the dip angel is estimated with a simple 1-D grid search at the linear inversion stage of the strike-variable fault model (Supplementary Fig. S12). The model with optimal dip angle further improve the data fitting along the profiles AA’ and CC’ (Fig. 9). The RMS between the ascending and descending track InSAR data and the model prediction are 3.0 and 5.1 cm for the rectangle strike-variable slip model. From the point of view of data fitting, the rectangle strike-variable slip model seems to be preferable. The released moment from the strike-variable slip model is 5.0 × 1018 N m, approximately equivalent to a moment magnitude of Mw 6.4, which is closer to the GCMT and USGS result. We also should note that these are all based on the assumption that the InSAR data are mainly composed of the coseismic deformation of the Alor earthquake.
(a) Coseismic displacements observed by ascending and descending orbit are plotted as black and red dots along each profile. The light blue dotted line outlines the possible fault strike variations. (b) and (c) are coseismic LOS displacements of ALOS-2 ascending and descending track. The locations of eight profiles are marked with black rectangles.
Three coseismic slip models. (a) slip model with rectangle patches; (b) strike-variable slip model with rectangle patches; (c) strike-variable slip model with triangle patches. The orange dots denote the aftershocks from USGS. The red star represents the location of the main shock from GCMT. The red and black solid lines outline the surface projection of the fault trace.
Coseismic deformation and model prediction of strike-variable fault model with rectangle patches. (a) ascending and (d) descending ALOS-2 data. (b) and (e) are predicted InSAR data from the optimal strike-variable fault model. (c) and (f) are residuals. (g–i) LOS displacements along the three profiles (AA’, BB’ and CC’) shown in (a) and (d). The black rectangle denotes the fault projection on the surface. The dashed black line denotes the projection of the fault trace. Orange dots denote the aftershocks from USGS. The red beach ball denotes the focal mechanism of the 2015 Alor earthquake from GCMT. The red and yellow stars represent the location of the main shock from GCMT and USGS, respectively.
5.3 Coulomb stress change due to the 2015 Alor earthquake and the historical earthquakes
Stress changes caused by an earthquake in the crust can accelerate the failure of adjacent faults and lead to earthquake sequences in certain cases (Stein et al. 1992). To assess the seismic risk in neighbouring zones after 2015 Alor earthquake, we calculated the coseismic Coulomb failure stress (CFS) change at the seismogenic depths of 2, 4, 6 and 8 km using Coulomb 3.3 software (Toda et al. 2011). In our calculation, the moderate value of the effective friction coefficient of 0.4 was applied, and the parameters of the receiver fault refer to optimal planar fault model. The results show that the regions with positive CFS changes mainly located at the northwest and southeast extremities of the rupture of the Alor earthquake, and in the lobes north and south of the rupture (Fig. 10). As the depth increases, the CFS changes varies from negative to positive at the location corresponding to the main rupture area. Furthermore, the CFS changes induced by the main shock was mainly concentrated in a local area, and quickly attenuated with a certain distance away from the seismogenic fault.
Coseismic Coulomb stress changes induced by the 2015 Alor earthquake at different depth levels.
The regional tectonic of the Alor area is dominated by backarc thrusting in which the ocean crust of the South Banda basin underthrusts the arc southward (e.g Hantoro et al. 1994). The Wetar thrust fault as the major fault in this region is gaining attention, which accommodates the main plate convergence at 7.5 cm yr−1 (Price & Audley-Charles 1987). One of the important earthquakes that occurred in this area is the 2004 November 11, Mw 7.5 earthquake, which is just close to Wetar thrust fault. Based on the focal mechanism and location of the 2004 Alor earthquake, we suspect that the occurrence of this earthquake may be related to the Wetar thrust fault. To further assess the seismic risk on the Wetar thrust fault, we collect the focal mechanism parameters (Supplementary Table S2, Fig. 11) of the history earthquakes since 1976 with focal depths less than 35 km and distances smaller than 100 km away from the 2004 Alor earthquake (Yang et al. 2021). Then, we use the empirical magnitude-area relations to estimate appropriate fault lengths and widths for the history earthquakes (Wells & Coppersmith 1994). Finally, we calculate the evolution of CFS changes in Alor island area for all Mw ≥ 6 events since 1976 (Supplementary Table S2) and slip distribution of 2015 Alor earthquake. The parameters of receiver fault are set based on the nodal plane of the 2004 Mw 7.5 Alor earthquake. All of our calculation of CFS changes are at a certain depth of 8 km, which is the location of the maximum slip of 2004 Alor earthquake from finite fault model of USGS. The calculation results indicate that the Coulomb stress changes induced by 1981 Mw6.0, 1987 Mw6.5, 1995 Mw6.8 and 2015 Alor earthquake mainly concentrated in local areas due to their moderate magnitude (Fig. 12). Since they occurred at a distance from the Wetar thrust fault, they cannot affect it. The 1977 Mw7.0 and 2008 Mw6.1 earthquake impart positive CFS changes on the east section of the Wetar thrust fault, while the 1991 Mw6.7 earthquake have an influence on the west section. The 2004 Mw7.5 earthquake have a significant CFS changes on the Wetar thrust fault, which may increase the Coulomb stress change on the two ends of the Wetar thrust fault.
The distribution of history earthquakes (Mw ≥ 6) since 1976 within 100 km of the epicentre of the 2004 Alor earthquake. The corresponding earthquake events for the serial numbers are: (1) 1977 Mw 7.0 earthquake, (2) 1981 Mw 6.0 earthquake, (3) 1987 Mw 6.5 earthquake, (4) 1991 Mw 6.7 earthquake, (5) 1995 Mw 6.8 earthquake, (6) 2004 Mw 7.5 earthquake, (7) 2008 Mw 6.1 earthquake.
The evolution of the cumulative coseismic CFS changes resolved at 8 km depth for all Mw ≥ 6 events since 1976 and slip distribution of 2015 Alor earthquake. (a) After 1977 Mw 7.0 earthquake, (b) After 1981 Mw 6.0 earthquake, (c) After 1987 Mw 6.5 earthquake, (d) After 1991 Mw 6.7 earthquake, (e) After 1995 Mw 6.8 earthquake, (f) After 2004 Mw 7.5 earthquake, (g) After 2008 Mw 6.1 earthquake, (h) After 2015 Alor earthquake.
5.4 Understanding the strike-slip faulting in a convergent boundary
Strike-slip faults are common in obliquely convergent subduction settings where interplate strain is partitioned into arc-parallel strike-slip zones within the forearc, arc or backarc region (Beck 1983; Cunningham & Mann 2007). When plate boundaries have a relative velocity vector that is markedly oblique to the boundary normal, a significant proportion have vectors that are nearly parallel to the boundary. Accommodation of the oblique motion usually involves the strike-slip motion, back-arc extension or some combination of both behind the trench (Woodcock 1986; Yu et al. 1993). To further understand the role of strike-slip faults in subduction zones, we collected focal mechanisms for most earthquakes of Mw ≥ 6.5 along three convergent plate boundary (Sunda-Java subduction zone, Aleutian subduction zone and Peru-Chile subduction zone) from GCMT for the period 1976 to 2023, shallow earthquake (0–30 km) showing strike-slip focal mechanisms were selected. For the Sunda-Java subduction zone, we can see that the strike-slip earthquakes mainly concentrated in Sumatra and Timor area, where the predicted plate motion vectors (blue arrows in Supplementary Fig. S13) is oblique to the trench. There are no large strike-slip earthquakes in Java area where the plate motion vector nearly normal to the trench. In Aleutian subduction zone, strike-slip earthquakes are chiefly centred around the longitude of −175° and 167° (Supplementary Fig. S14), where the geometry of Aleutian trench changes rapidly. The plate motion is nearly parallel to the trench at the longitude of 167°. In Peru-Chile subduction zone, the large strike-slip earthquake is rare near the trench (Supplementary Fig. S15).
The 2015 Alor earthquake occurred on the inland of Alor island, where it is affected by both the Timor trough and backarc thrust faults. The plate motion vector is oblique to the Timor area and the Alor island is just located behind the Timor island. The 2015 Alor earthquake may accommodate a portion of the oblique motion. Meanwhile, we also noted that the potential rigid rotation component estimated from GPS data can fit well at the south of the backarc thrust faults, while the data fitting on the northern side is not ideal. This phenomenon indicates that the block rotation could be a reason to load interseismic strain energy on the strike-slip fault within a convergent zone.
6 CONCLUSION
In this study, we investigate the coseismic deformation of the 2015 Mw6.5 Alor earthquake with Sentinel-1 and ALOS-2 interferograms. The preferred coseismic fault model shows that the rupture occurred on a southwest dipping fault with a strike of 117.01° and dip of 69.35°. The major slip area is concentrated between 1 and 6 km. We find that constructing a strike-variable fault model can fit the InSAR data better, which may imply that the seismogenic fault is likely to have a geometric feature of a curved fault plane. The most striking discovery from the InSAR observations of the Alor earthquake is that most of the displacement occurred on a fault whose existence was unknown before the earthquake. The Coulomb stress change results show that the regions with positive CFS changes mainly located at the northwest and southeast extremities of the rupture of the Alor earthquake, and in the lobes north and south of the rupture.
ACKNOWLEDGEMENTS
We thank the editors, Dr Wanpeng Feng and another anonymous reviewer for their constructive comments, which significantly helped improve the manuscript. The study is funded by the National Natural Science Foundation of China (42104008, 42130101, 41974004, 42074007) and the Fundamental Research Funds for the Central Universities (2042023kf1035). ZG has been funded by the China Scholarship Council (202006270005). Figures in this study were prepared by using the Generic Mapping Tools (GMT) (Wessel & Smith 1998).
DATA AVAILABILITY
The Sentinel-1 SAR data were downloaded from the Sentinel-1 Scientific Data Hub (https://scihub.copernicus.eu). The ALOS-2 PALSAR-2 data are provided by JAXA through EO RA3 (No. ER3A2N040).
SUPPORTING INFORMATION
supplement_material.docx
Please note: Oxford University Press is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the paper.
