Professional Documents
Culture Documents
Coseismic Slip in The 2016 M 7.8 Ecuador Earthquake Imaged From Sentinel-1A Radar Interferometry
Coseismic Slip in The 2016 M 7.8 Ecuador Earthquake Imaged From Sentinel-1A Radar Interferometry
○
E
doi: 10.1785/0220160151 Seismological Research Letters Volume 88, Number 2A March/April 2017 1
SRL Early Edition
▴ Figure 1. Tectonic setting of the 2016 M w 7.8 Ecuador earthquake. Blue rectangles depict the footprints of the Sentinel-1A images used
in this study, and black lines depict active faults (Alvarado et al., 2014). Focal mechanism plots depict the focal mechanism solutions from
the Global Centroid Moment Tensor (Global CMT; Ekström et al., 2012) and U.S. Geological Survey (USGS). Yellow circles are two months
of aftershocks taken from the International Seismological Centre (ISC) catalog. Stars denote the epicenters of 1906–1998 M w > 7:0 earth-
quakes (Chlieh et al., 2014). Gray-shaded ellipses are the inferred high-slip region of the 1942, 1958, 1979, and 1998 earthquakes (Beck and
Ruff, 1984; Swenson and Beck, 1996; Segovia, 2001). Black dashed ellipse depicts the region of the megathrust inferred to have slipped in
the great 1906 earthquake, and dashed contours show the aftershocks zones of the 1942, 1958, and 1979 events (Mendoza and Dewey,
1984). The inset map shows the larger tectonic setting.
Table 1
Interferometry Pairs Used in This Study (A or D Denotes Ascending or Descending, Respectively)
Master Slave Inclined Azimuth
Track (A/D) (yyyy/mm/dd) (yyyy/mm/dd) Perp. B (m) Angle (°) Angle (°) σ (mm) α (km)
018 (A) 2016/03/29 2016/04/22 10 30–44 −168 8.2 15.8
040 (D) 2016/04/12 2016/04/24 18 30–44 −12 3.7 8.9
“Perp. B” is the perpendicular baseline of the pair; σ is the standard deviation calculated with all points in the nondeforming
area; and α is the e-folding correlation length scale of 1D covariance function (SAR data; see Data and Resources).
As demonstrated by numerous studies on large earth- as the 2015 M w 7.9 Gorkha, 2015 M w 8.3 Illapel, and 2015
quakes, spatially dense, near-field geodetic measurements of M w 6.4 Pishan earthquakes (Grandin et al., 2016; Solaro et al.,
coseismic surface deformation yields more insights into both 2016; Wen et al., 2016).
coseismic slip distribution and fault geometry than teleseismic We generate interferograms for both the ascending and
data (e.g., Lindsey et al., 2015; Zhang et al., 2015), although descending tracks (T018A and T040D) using the traditional
onshore geodetic data do have resolution issues when infer- two-pass differential InSAR method (e.g., Hanssen, 2001). All
ring offshore slip in megathrust earthquakes (e.g., Simons three subswaths within a satellite pass are combined to achieve
et al., 2011; Evans and Meade, 2012; Zhang et al., 2016). an efficient coverage of the coseismic deformation. We use the
Although there is a continuously operating Global Position- commercial processing software GAMMA (Werner et al.,
ing System (cGPS) network of 70 permanent stations in 2000), with geometric alignment based on precise orbit deter-
Ecuador (Mothes et al., 2013), the cGPS network is perhaps mination (SAR orbit data; see Data and Resources). Because
too sparse to characterize the full pattern of surface displace- straightforward interpolation of the slave image results in
ments associated with this earthquake. As a new generation banding with each burst, deramping the single look complex
satellite-based Synthetic Aperture Radar (SAR) system, the (SLC)-format images is done prior to interpolation to over-
European Space Agency’s Sentinel-1A has proven to be suc- come limitations of large phase mismatches (Xu et al., 2016).
cessful in determining ground deformation due to recent ma- We adopt the 90-m resolution Shuttle Radar Topography Mis-
jor earthquakes (e.g., Lindsey et al., 2015; Zhang et al., 2015; sion digital elevation model to remove topographic phases
Solaro et al., 2016). In this study, we process SAR images ac- (Farr et al., 2007). Finally, the interferograms are filtered using
quired by the new generation, C-band, Sentinel-1A SAR system a power spectrum filter algorithm (Goldstein and Werner,
and use these data to constrain both rupture geometry and slip 1998) to reduce the effects of phase noise, and they are un-
distribution of the 2016 earthquake. We also explore the reso- wrapped using the SNAPHU algorithm (Chen and Zebker,
lution of the Interferometric Synthetic Aperture Radar (InSAR) 2000). The unwrapped and geocoded interferograms have 90-m
data and discuss the earthquake within the context of the pre- resolution and represent range changes in the direction of line-
vious megathrust earthquakes offshore of Ecuador. of-sight (LOS) of the surface toward the satellite (Fig. 2a,b and
Ⓔ Fig. S1, available in the electronic supplement to this article).
INSAR DATA Because of short perpendicular baselines for the pairs of images,
there are no visible phase ramps in the resulting interferograms.
Onshore coseismic deformation in the 2016 Ecuador earth- The interferograms are shown in Figure 2, and they resolve
quake is imaged quite well by the Sentinel-1A satellite on both one incomplete lobe of coseismic deformation onshore. Even
ascending and descending orbits (Table 1). The Sentinel-1A though the coseismic deformation offshore is invisible by In-
radar operates in four modes with different resolution and cov- SAR, the onshore surface displacements are sufficient to con-
erage: strip-map mode (SM), interferometric wide swath mode strain the coseismic slip in this earthquake, as we show later.
(IW), extrawide swath mode (EW), and wave mode (WM). Compared to the ascending interferogram, the descending in-
Both IW and EW mode employ the novel terrain observation terferogram shows a clearer pattern of surface displacements
with progressive scans (TOPS) acquisition for interferometry because the descending image pairs are separated by a shorter
(Grandin et al., 2016), which achieves an extended swath by time span, resulting in a higher degree of coherence. The stan-
steering the beam in range, from backward to forward in the dard deviations of the LOS measurements in the region of the
azimuth direction for each burst (De Zan and Guarnieri, interferograms away from the prominent coseismic deforma-
2006). However, the TOPS success was realized at the cost tion (e.g., about 1° S and 1° N; Ⓔ Fig. S1) are 8.2 and
of a much higher accuracy of pixel co-registration at the level 3.7 mm for the ascending and descending interferograms, re-
of <0.001 pixel than is required to eliminate the azimuth phase spectively. We take these standard deviations as lower bounds
deramping (De Zan and Guarnieri, 2006). Currently, a tech- on the uncertainties of the interferometric data and set the
nique termed as enhanced spectral diversity performs well for a uncertainty of the LOS measurements to be 18 and 6 mm
single-frame data case (Xu et al., 2016). Sentinel-1A SAR data for the ascending and descending interferograms, respectively.
have been used for the investigation of recent earthquakes, such We note that range direction in the descending interferogram
▴ Figure 2. Coseismic interferograms of the (a) descending Sentinel-1A track T040D and the (b) ascending track T018A. Synthetic in-
terferograms on the (c) descending and (d) ascending tracks of the best-fit uniform slip-fault-plane solution; red barbed line represents
the surface trace of the modeled fault plane. Residual interferograms in (e) descending and (f) ascending geometries. Black line is the
subduction trench. Each contour on the interferograms indicates 0.05 m increment in displacement.
is approximately normal to the strike of the fault, which yields homogeneous elastic half-space on the inferred slip distribution
its maximum LOS offset of 73 cm, nearly double the maximum in the Discussion section.
offset (33 cm) in the ascending interferograms (Fig. 2a,b). We first determine the fault geometry through the
To reduce the computational burden when jointly inverting method of Clarke et al. (1997). We assume uniform slip on a
the two interferograms for coseismic slip, we subsample both rectangular fault plane, solving for fault strike, dip, length, and
interferograms using the QuadTree algorithm (Yan et al., 2013), width, as well as slip magnitude and rake. The method of
resulting in two data sets of 552 and 568 samples from the as- Clarke et al. (1997) uses a simplex search algorithm, starting
cending and descending interferograms, respectively. from an initial solution. We construct 100 initial models with
strike, dip, and rake randomly sampled from Gaussian proba-
COSEISMIC SLIP MODEL bility density functions (PDFs) with means corresponding to
the strike, dip, and rake of the nodal plane corresponding to
We determined the coseismic slip from the InSAR data in two the megathrust in the USGS focal mechanism and standard
steps. First, we use a nonlinear inversion to constrain the fault deviations of 15°. We randomly choose the fault length, width,
geometry, assuming uniform slip on a rectangular fault. Second, and slip in the initial models from Gaussian PDFs with means
we invert the InSAR data for the distribution of coseismic slip, given by dimensions consistent with the USGS magnitude
assuming either the fault plane determined in the first step or (80 km × 35 km × 2:5 m) and standard deviations of 1 m in
using a strike from the focal mechanism solutions of this earth- slip, 20 km in length, and 15 km in depth. After 50 restarts in
quake. We assume that the fault is embedded in a Poisson half- the simplex search, each initial model results in a unique sol-
space with shear modulus 32 GPa (Okada, 1992). We discuss ution, and we take the collection of final models obtained to
the resolution of these data and the ramifications of using a indicate the dependence on the initial starting models in the
Table 2
Source Parameters of the 2016 Ecuador Earthquake
Longitude Latitude Strike Dip Rake Depth Length Width Slip Moment Rms
Model (°) (°) (°) (°) (°) (km) (km) (km) (m) (×1020 N·m) M w (cm)
USGS-BW −79.926 0.352 26 16 113 20.6 – – – 7.054 7.8 –
Global −80.35 −0.16 26 23 123 24.1 – – – 5.530 7.8 –
CMT-SW
USM −80.97 0.03 8.7 17.1 110.7 – 80.8 28.3 2.54 6.32 7.83 5.8
±0.04 ±0.01 ±1.2 ±0.7 ±1.9 – ±0.5 ±1.6 ±0.06
Reference −80.97* 0.03* 8.7* 17.1* 106.7 – 180 140 2.48† 7.86 7.90 5.1
model
Preferred −80.97* 0.03* 26.0 17.1* 123.1 180 140 2.56† 7.15 7.87 4.1
model
Source Parameters of the 2016 Ecuador Earthquake Determined from the U.S. Geological Survey Body-Wave (USGS-BW) Focal
Mechanisms, the Global Centroid Moment Tensor (Global CMT) Surface-Wave Focal Mechanisms (Global CMT-SW), the Nonlinear
Inversion Assuming Uniform Slip Model (USM, Best-Fitting Parameters Are Given, along with the Standard Deviations of the Parameters
from the 100 Nonlinear Inversions Started with Unique Initial Models), a Linear Inversion Solving for the Distribution of Slip on the Best-
Fitting Fault Geometry in the Uniform Slip Model (Reference Model; Ⓔ Fig. S5), and a Linear Inversion Using the Focal Mechanism Strike
(Preferred Model; Fig. 3). Rms is root mean square between observation and simulation with model.
*The value is taken from the best-fitting USM model
†Maximum slip.
nonlinear inversion (Clarke et al., 1997). Most surprisingly, our grams. From this, we conclude that the InSAR alone, even
best-fit uniform slip model has a more north–south strike than with both ascending and descending orbits, cannot uniquely
either the USGS or Global CMT focal mechanisms (Table 2; resolve the strike of the fault. There is a trade-off between the
Ⓔ Fig. S1), and the 100 solutions from each chain of the non- strike and inferred slip rake in these two models, with mean
linear inversion all have strikes at least 10° more northerly from coseismic rake weighted by the slip magnitude of 123° and
the focal mechanism solutions (Ⓔ Fig. S2). 107° (90° corresponding to pure thrust motion of hanging
We next invert for the distribution of coseismic slip as- wall) using a strike of 26° or 8.7°, respectively, both of which
suming a planar fault geometry. As the best-fitting strike are consistent with the footwall moving about N83°E. That
found in the above nonlinear inversion is quite different from this footwall direction is coincident with the plate conver-
the strike of the trench, we consider two fault geometries. In gence and is independent of assumed fault strike indicates
both cases, the faults dip 17.1° to the east, which is the best that there is minimal, if any, slip partitioning of the subduc-
dip obtained in the nonlinear inversions above, and we as- tion convergence in this earthquake.
sume fault lengths and widths of 180 and 140 km, respec- We take the coseismic slip model with strike of 26° as our
tively, and divide the faults into 10 × 10 km fault patches preferred coseismic slip model (Figs. 3a and 4; we show the
(for a total of 252 patches). We separately consider two fault coseismic slip model and data fits using an 8.7° strike in Ⓔ
strikes, either a strike of 8.7°, the best fault strike obtained in Figs. S5 and S6). The preferred model contains a relatively
the nonlinear inversions above, or a strike of 26°, which is the compact region of oblique-thrust slip, with maximum slip of
strike of the nodal plane corresponding to the trench in the about 2.5 m at a depth of just under 20 km (corresponding to
USGS and Global CMT focal mechanism solutions (Table 2). about 65 km along the dip direction along the fault plane),
Because the primary mechanism of the earthquake was thrust with slip extending down to about 35 km depth (∼120 km
(revealed by both the moment tensor solution and the slip along dip). We infer a decrease in slip toward the trench, with
rake in our best-fit uniform slip model), we constrain the slip under 2 m of slip at the trench; however, the resolution near
rake within 85°–135°. We linearly invert for slip on each of the trench is the poorest with these onshore data, a point we
the two fault planes, assuming a second-order smoothing to address in the Discussion section. Our preferred coseismic slip
avoid unphysical oscillation of inferred slip, determining the model yields a geodetic moment of 7:15 × 1020 N·m, equiva-
penalty parameter on the regularization through a standard lent to M w 7.87. The moment of our preferred model is
L-curve analysis (Ⓔ Fig. S3; e.g., Aster et al., 2013). We find slightly higher than the USGS or Global CMT solutions,
that the solution using a strike of 26° is slightly better than although we note that the ascending and descending interfero-
using the strike found in the nonlinear inversions above (root grams contain 6 and 8 days, respectively, of the postseismic
mean square error of 4.1 cm vs. 5.1 cm), with no clear period and thus may contain a minor amount of ground
differences in the residuals of the downsampled interfero- deformation due to postseismic deformation.
▴ Figure 3. (a) Coseismic slip model determined from the interferograms in Figure 2, assuming a fault location and dip determined in the
nonlinear inversion and using the fault strike of 26° taken from the USGS and Global CMT focal mechanism solutions. Arrows show the
sense of motion of the hanging wall, and contours are labeled in meters. (b) Standard deviation of 100 coseismic slip models, each inferred
by adding random noise to the original interferograms prior to inversion.
▴ Figure 4. Modeled interferograms for the (a) descending track T040D and (b) ascending track T018A predicted by the slip model in
Figure 3a and the associated residuals of the observed interferograms in Figure 2c,d. The red line is the trace of the model fault plane, and
the white dashed lines represent the surface projections of the modeled fault. Each contour indicates a 0.05-m increment in displacement.
These interferograms are also shown wrapped in Ⓔ Figure S4 (available in the electronic supplement to this article).
the earthquake and an assumed rupture velocity of 2:0 km=s. would be required to have ruptured to match the moment. A
Using this rupture size, however, the average slip would need 50-km circle approximating the 1942 earthquake roughly cor-
to be greater than 8 m to match the moment (Swenson and responds to the region in which we infer more than 1.0 m of
Beck, 1996). If on the other hand, we assume uniform slip of slip in our preferred model (Fig. 5). The largest MMI (IX)
2.0 m, consistent with the slip we infer for this earthquake, in the 1942 earthquake was reported to be just to the south
then a region of the megathrust with radius about 50 km of the epicenter, with MMI ≥ VIII over the coastal region