The 20102011 Canterbury, New Zealand ... - Wiley Online Library

6 downloads 0 Views 4MB Size Report
Aug 22, 2012 - started with a Mw 7.1 earthquake and continued with large aftershocks with dramatic consequences, particularly for the city of Christchurch.
JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117, B08305, doi:10.1029/2012JB009178, 2012

The 2010–2011 Canterbury, New Zealand, seismic sequence: Multiple source analysis from InSAR data and modeling S. Atzori,1 C. Tolomei,1 A. Antonioli,1 J. P. Merryman Boncori,1 S. Bannister,2 E. Trasatti,1 P. Pasquali,3 and S. Salvi1 Received 23 January 2012; revised 2 July 2012; accepted 5 July 2012; published 22 August 2012.

[1] The 2010–2011 Canterbury sequence is a complex system of seismic events that started with a Mw 7.1 earthquake and continued with large aftershocks with dramatic consequences, particularly for the city of Christchurch. We model the main earthquakes using InSAR data, providing displacement maps and the respective modeling for the September 4th, 2010, February 22nd, 2011 and June 13th, 2011 events. Relocated aftershocks, field and GPS surveys are used to constrain models obtained by inversion of InSAR data; the fault slip distribution is retrieved with a variable patch size approach aimed at maximizing the spatial resolution on the fault plane. For the September 2010 earthquake we estimated significant slip values below 10 km depth; the calamitous February 2011 event in Christchurch is modeled with a double fault source with slip values less than 2 m down to 7 km depth; for the second June 13th event in Christchurch we identified a NW-SE striking fault as responsible for the earthquake. Last, we introduce the use of InSAR coherence maps to quickly detect the areas subject to soil liquefaction in Christchurch, as shown for the two main events. Citation: Atzori, S., C. Tolomei, A. Antonioli, J. P. Merryman Boncori, S. Bannister, E. Trasatti, P. Pasquali, and S. Salvi (2012), The 2010–2011 Canterbury, New Zealand, seismic sequence: Multiple source analysis from InSAR data and modeling, J. Geophys. Res., 117, B08305, doi:10.1029/2012JB009178.

1. Introduction [2] From late 2010 through to January 2012, the Canterbury region in the South Island of New Zealand was struck by a complex earthquake sequence, which started on September 4th, 2010 at 4:35 A.M. (NZ standard time, adopted for the whole paper) with a Mw 7.1 main shock near the town of Darfield, about 30 km west of Christchurch. This event was followed by an aftershock sequence that included 3000 aftershocks of magnitude greater than 3 up to the end of July 2011 [Bannister et al., 2011]; two of the larger aftershocks, Mw 6.2 on February 22nd and Mw 6.0 on June 13th, 2011 heavily damaged Christchurch city (Figure 1). [3] This earthquake sequence occurred in an area with a low strain rate and moderately low historical seismicity 100 km from the Marlborough strike-slip region to the north, and the Alpine Fault strike-slip fault to the west [Cox and Sutherland, 2007]. The Alpine Fault is thought to accommodate ca.70–75% of the 35–45 mm/yr of motion between the Australian and Pacific plates, while kinematic models of GPS measurements, using an elastic block 1

Istituto Nazionale di Geofisica e Vulcanologia, Rome, Italy. GNS Science, Lower Hutt, New Zealand. SARMAP SA, Purasca, Switzerland.

2 3

Corresponding author: S. Atzori, Istituto Nazionale di Geofisica e Vulcanologia, Via di Vigna Murata 605, Rome I-00143, Italy. ([email protected]) ©2012. American Geophysical Union. All Rights Reserved. 0148-0227/12/2012JB009178

approach, suggest a total slip rate of 2.5–7 mm/yr on the faults in the foothills west of the Canterbury Plains region, higher than estimated from geological studies [Wallace et al., 2007]. Although not unexpected – earthquakes up to magnitude 7.2 are contemplated by the national seismic hazard model [Stirling et al., 2002] – the events occurred on previously unrecognized faults. [4] The three largest events took place beneath the Canterbury Plains, which are gravel-dominated alluvial plains with a 200 to 600 m thickness of alluvial gravels, underlain by varying thicknesses of late Cretaceous to Tertiary marine sedimentary layers, and, in the east, also by Late Miocene volcanics, which form the Banks Peninsula, with a maximum elevation of 910 m [Forsyth et al., 2008]. The sedimentary layers are thought to have been a factor in the strong damage from soil liquefaction and basin reverberation effects in the city of Christchurch and its suburbs [Sibson et al., 2011; Guidotti et al., 2011; Cubrinovski et al., 2011; Green et al., 2011; Fry et al., 2011], while the presence of the Banks Peninsula volcanics, with different rheology, may locally alter the regional stress field, and have led during inception to more complex fault structures in the immediate vicinity of Banks Peninsula (i.e., near Christchurch city). [5] In this work we analyze different aspects of the seismic sequence, the complexity of the seismogenic fault system for the three major events, the fault interaction and the soil liquefaction induced by the ground shaking. This research is mostly based on the analysis of the ground displacement measured through InSAR, (Differential) Interferometry from

B08305

1 of 16

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

B08305

Figure 1. New Zealand 2010–2011 seismic sequence: focal mechanisms of main earthquakes from USGS-CMT. Aftershocks from Bannister et al. [2011], extended up to July 10th 2011 in this work, are shown color coded following the main events: after September 4th (black), February 22nd (red) and June 13th (yellow). Tectonic setting in the inset is from Wallace et al. [2007].

Synthetic Aperture Radar images [Franceschetti and Lanari, 1999] and through SAR amplitude offset tracking techniques (Appendix A and references therein), providing a picture of the complex fault system underlying the seismic sequence. Modeling is conducted with a standard and a new full-resolution approach, based on a fault subdivision into patches of variable size (Appendix B and references therein). We also used the interferometric coherence maps, by-products of the InSAR processing, to determine the area affected by soil liquefaction, comparing our preliminary results with field observations from Green et al. [2011].

2. SAR Data [6] The Canterbury seismic sequence is characterized by a large availability of SAR data, already analyzed and interpreted by various authors [Beavan et al., 2010, 2011; Barnhart et al., 2011; Stramondo et al., 2011; Elliott et al., 2012], acquired from the European Space Agency ENVISAT satellite (ASAR sensor, a C-band SAR antenna with wavelength of 5.6 cm) and the Japan Aerospace Exploration Agency ALOS satellite (PALSAR sensor, an L-band antenna with wavelength of 23.6 cm). [7] Moreover, after the September 4th, 2010 event, the SiGRIS monitoring service (www.sigris.it), a joint effort of the Italian Space Agency and INGV, the Italian Institute of Geophysics and Volcanology, requested a full coverage of a large area around Darfield with the 4-satellite COSMOSkyMed constellation, with a mean revisit time of 8 days, for a total duration of at least a year. The frequent revisit time was requested to minimize the X-band decorrelation problems likely to occur in the highly vegetated Canterbury

region, and to monitor the temporal evolution of post-seismic deformations (see in Figure S1 of the auxiliary material the footprints of all the satellite tracks processed in this work).1 [8] We retrieved the Line-of-Sight (LoS), i.e., the groundsatellite direction, co-seismic ground displacement induced by the largest earthquake of the sequence using the two-pass InSAR technique [Massonnet et al., 1993; Franceschetti and Lanari, 1999], adopting the SRTM 3 arc second Digital Elevation Model to remove the topographic contribution. In addition, for the Darfield earthquake, we also calculated the displacement in the azimuth direction using a cross-correlation technique (see Appendix A). [9] Measurements used in the inversion are obtained by resampling the unwrapped interferograms on a variable size mesh, with 250 m of maximum resolution in the fault proximity. This value is a compromise between the calculation load (more than 28,000 points for the Darfield event, see Figure S2 in the auxiliary material) and the resolution achievable over the fault plane. Although resolution is qualitatively intended as the level of detail achievable on the fault from observed data, in text and figures we refer to “resolution value” as the diagonal values of the Model Resolution Matrix (see Appendix B and references therein for a more complete description).

3. Data Processing and Modeling [10] Geodetic data were processed to infer the seismic source for three 6+ magnitude events between September 1 Auxiliary materials are available in the HTML. doi:10.1029/ 2012JB009178.

2 of 16

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

B08305

Table 1. Darfield Earthquake Data Sets RMSa

Baselines Satellite

Path

Pair

Look Angle (deg)

Spatial (m)

Temporal (d)

Orbit

Points

Fixed (m)

Full Resolution (m)

ALOS

337

38.9

357

46

Asc

9951

0.12

0.13

ALOS

336

38.9

157

920

Asc

2752

0.13

0.12

ALOS

631

38.9

2650

782

Dsc

7517

0.14

0.14

ENVISAT

51

23.3

113

105

Asc

2510

0.08

0.09

ENVISAT

323

13/08/2010 28/09/2010 03/05/2008 11/09/2010 22/07/2008 12/09/2010 04/06/2010 17/09/2010 01/09/2010 06/10/2010 13/08/2010 28/09/2010

23.3

300

35

Asc

2669

0.06

0.07

357

46

Asc

2625

0.21

0.23

ALOS

337

b

0

a

Rms of residuals between observed and modeled data with the fixed patch size and the full-resolution approaches. Azimuth displacement in the satellite flight direction, horizontal, 12 counter-clockwise from North.

b

2010 and June 2011. Data modeling was carried out with the typical two-step approach, where fault geometries are retrieved via nonlinear inversion and the slip distribution is then obtained with a linear inversion, estimating at the same time possible residual ramps in the InSAR data [Atzori et al., 2009]. In both steps, the underlying model is a rectangular dislocation in a homogeneous and elastic half-space [Okada, 1985], only slightly improved to account for the gentle topography of the area [Lungarini et al., 2005]. As previously mentioned, we discuss the outcome of two alternative approaches, where fault planes are subdivided into patches of equal and variable size respectively. We believe that the full-resolution approach is particularly suitable in fault configurations where high dip angles favor the rapid decrease of resolution with geodetic data: this condition might result in slip distributions mostly driven at depth by a priori constraints, such as smoothing or boundary conditions. At the same time, understanding where and if slip propagated at depth can provide useful information from a tectonic point of view.

4. September 4th, 2010: Darfield Earthquake [11] The Mw 7.1 Darfield earthquake occurred at 4:35 A. M. on September 4th on a previously unknown fault. The rupture is thought to have started on a steeply dipping reverse fault, then propagating on the main Greendale fault characterized by a transcurrent, predominantly right-lateral, mechanism [Gledhill et al., 2010; Beavan et al., 2010; Quigley et al., 2010]. Several authors have already presented complex fault systems to describe the segmented strike-slip Greendale plane, which released most of the seismic moment, and surrounding faults. Beavan et al. [2010] proposed 3 transcurrent segments and 3 reverse faults on the basis of InSAR and GPS data; this solution was then adopted by Stramondo et al. [2011]; 2 transcurrent segments and 2 reverse faults were modeled by Holden et al. [2011] from strong motion recordings; 4 segments accommodating only the Greendale fault were proposed by Barnhart et al. [2011] from InSAR data; 5 transcurrent segments for Greendale, 2 reverse faults and an additional NNW–SSW transcurrent fault were proposed by Elliott et al. [2012] from InSAR and teleseismic data.

[12] In this work we analyze the Darfield seismic source with a geodetic data set consisting of three ALOS-PALSAR interferograms, two from ascending (path 336 and 337) and one from descending (path 631) orbits, two ENVISATASAR interferograms from ascending orbits (track 51 and 323) and one azimuth offset map from ALOS path 337 (Table 1). Every PALSAR interferogram is composed of two consecutive frames (all the processed interferograms are shown in Figure S3 in the auxiliary material). This large set of highly coherent interferograms, integrated with precise 3D locations of aftershocks from Bannister et al. [2011] extended up to July 10th 2011 in this work, allowed us to identify the displacement pattern for 5 distinct segments of the main fault, here referred to as Greendale 1 to Greendale 5 from east to west (G1–G5 in all the figures), and three more sources consisting of two reverse faults and a transcurrent fault. Such a configuration is highly consistent with that presented by Elliott et al. [2012], which is in fact partially based on the same geodetic data. Four of the Greendale segments (Greendale 1 to 4) are clearly visible in the ascending interferograms, while the displacement induced by the fifth segment (Greendale 5) has a larger component in the descending line-of-sight. Three further fringe patterns reveal the presence of several faults, which we identify as Hororata Fault (HF in this paper, not to be confused with the Hororata fault in Forsyth et al. [2008]), Charing Cross South (CCS) and Charing Cross North (CCN) faults. [13] Such a high number of sources complicates the simultaneous retrieval of all parameters through a nonlinear inversion; however, fringe patterns, high resolution aftershock patterns [Bannister et al., 2011] and ground surface faulting [Quigley et al., 2011] were used to constrain the initial value of some parameters, primarily fault locations and strike angles. Almost all were then varied in turn for a further refinement (Table 2). After the definition of the fault geometries, we calculated the slip distribution through a linear inversion of the six data sets of Table 1, with the standard and the full-resolution algorithms introduced in section 3 and described in Appendix B. [14] Maps of observed, modeled and residual displacement obtained from the linear inversion are shown in Figures 2 and 3. Modeled and residual displacement maps

3 of 16

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

B08305

Table 2. Darfield Earthquake Fault Parameters Linear Inversion Nonlinear Inversion Fault Segment Greendale 1 Greendale 2 Greendale 3 Greendale 4 Greendale 5 Charing Cross North Charing Cross South (near) Hororata

Fixed Patches a

a

Strike (deg) Dip (deg) Rake (deg) Longitude (deg) Latitude (deg) 307 269 261 91 108 338 27 210

86 77 81 80 60 90b 60 40

164 173 172 169 153 7 64 84

43.572 43.598 43.591 43.576 43.574 43.532b 43.571 43.575 TOTAL

172.023 172.135 172.246 172.338 172.455 172.189b 172.118 171.952

M0 (Nm) 0.5  10 1.1  1019 1.2  1019 1.0  1019 0.2  1019 0.3  1019 0.7  1019 0.3  1019 5.5  1019b (M 7.1) 19

Full Resolution

Number

M0 (Nm)

Number

120 140 112 96 140 130 48 100 886

0.6  10 1.2  1019 1.4  1019 1.5  1019 0.2  1019 0.7  1019 0.7  1019 0.4  1019 6.7  1019 (M 7.2c)

24 16 11 15 22 13 12 31 144

19

a

Coordinates of the fault trace center. Parameter fixed a priori. Rounded from M 7.15.

b c

refer only to the linear inversion with the full-resolution algorithm, since those for the fixed patch-size model are virtually identical (cf. also the r.m.s. values of the residuals in Table 1). The outcome of both models is shown in the 3D

views of Figures 4 and 5, where the same perspective is adopted to show the slip distribution and the resolution values (a second 3D view, from the North, is available in Figures S4 and S5 in the auxiliary material). The model

Figure 2. (left) Observed, (middle) modeled, and (right) residual displacement maps of three unwrapped ALOS-Palsar interferograms relating to the September 4th 2010, Mw 7.1, Darfield earthquake. Color legends are scaled according to the minimum/maximum value for observed and modeled maps, while the same color scheme is adopted for the three residuals maps. Interferograms can be found in Figure S3 in the auxiliary material. Locations of the modeled fault traces, image acquisition dates and satellite paths are also provided, together with the line-of-sight directions. Data set details are supplied in Table 1. 4 of 16

B08305

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

Figure 3. (left) Observed, (middle) modeled, and (right) residual displacement maps of two unwrapped ENVISAT-ASAR interferograms and one ALOS-Palsar azimuth offset map relating to the September 4th, Darfield earthquake. Fault traces, image acquisition dates, satellite paths and line-of-sight directions as for Figure 2. Data set details are supplied in Table 1. ENVISAT interferograms are shown in Figure S3 in the auxiliary material. uncertainty, available only for the full-resolution subdivision is given in Figure S6 of the auxiliary material.

5. Discussion of the Darfield Modeling Results [15] A quantitative comparison with already published models [Beavan et al., 2010; Holden et al., 2011; Stramondo et al., 2011; Barnhart et al., 2011; Elliott et al., 2012], is not straightforward, due to the different levels of modeling complexity and to the use of different or differently processed and sampled data. We therefore limit the discussion to macroscopic differences. There is a good agreement with previous studies concerning the position of the Greendale fault, based on field surveys and InSAR fringe patterns, while a small disagreement concerns its dip angle; we find it to be northward for G1, G2 and G3, the three segments west of the step over, while it is reported to be nearly vertical or even slightly southward in Beavan et al. [2010] and Elliott et al. [2012]. However, we consider discrepancies on the order of ten degrees highly acceptable for nearly vertical faults. A more significant disagreement is the absence, in our model, of the reverse segment located at the step over between G3 and G4 by Beavan et al. [2010], of which the InSAR displacement maps show no evidence.

[16] Major differences, however, concern the slip distribution, peak values and the related moments. Holden et al. [2011] and Barnhart et al. [2011] locate most of the slip at shallow depth (less than 3 km), while almost all the inversions based on InSAR data [Beavan et al., 2010; Stramondo et al., 2011; Elliott et al., 2012] result in significant slip down to 5–7 km; peak values of the slip range from 5 m [Beavan et al., 2010] to 8 m [Elliott et al., 2012]. These differences can be attributed, more than to different data, to the non-uniqueness of the solution of the linear inversion. In this respect the two different inversion schemes used in our study also provided significantly different results. The fullresolution method, described in Appendix B, located most of the slip between 2.5 and 5.0 km beneath the central segment G3, with a mean slip value of 6 m, in agreement with previous studies (Figure 4). However, significantly higher mean slip values were inferred at depth, compared to the standard modeling approach with patches of predefined size and to other studies: we retrieve a mean slip of about 0.5 m (s = 0.7 m) and 2.1m (s = 0.7 m) for the deepest 10 km of segments G3 and G4, respectively. An exception to this is the work of Barnhart et al. [2011], which shows nonzero values down to about 14 km, although the change in rake to pure dip-slip may have played an important role.

5 of 16

B08305

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

Figure 4. (top) Slip distribution for the September 4th, Darfield earthquake obtained with (middle) standard (fixed patch size of 1 km) and (bottom) full-resolution approaches; the latter is obtained with an automatic fault subdivision according the algorithm of Atzori and Antonioli [2011]. Resolution values for both models are shown in Figure 5. Fault positions and extent are identical for the two solutions and are shown in the top panel; G1-G5 are the Greendale fault segments; HF, CCN and CCS show the traces of (near) Hororata, Charing Cross North and Charing Cross South segments. Further fault parameters are reported in Table 2. A second 3D view from North can be found in Figure S4 of the auxiliary material. Uncertainty for the full-resolution model is shown in Figure S6 of the auxiliary material. [17] Concerning the comparison of the two inversion methods used in this study, shown in Figure 3, although these inversions fit the observed data equally well (cf. r.m.s. in Table 1), the number of patches is definitely lower in the full-resolution approach (144 versus 886), since the patch size is driven by the model resolution matrix and any further patch subdivision would not meaningfully improve the data fit. Though not perfectly resolved, as shown by the resolution values in Figure 5, this subdivision is possibly one of the best obtainable with rectangular patches for the given data sampling. It is worth noting that resolution values, especially for patches at depth, are up to 2 orders of magnitude higher than those obtained assuming 1 km patches. Slip uncertainty increases at depth to a standard deviation of 0.7 m in the big patches between 10 km and 20 km (see Figure S6 in the auxiliary material).

[18] The geodetic moment based on the full-resolution model is 6.7  1019 Nm, corresponding to M 7.15, which is not significantly higher than seismological estimates ranging from 3.5  1019 Nm (Mw 7.0) from the USGS-CMT solution to 6.2  1019 Nm (Mw 7.15) of the GeoNet moment tensor database (http://www.geonet.org.nz/canterburyquakes/significant.html); the latter is derived from moment tensor inversion using New Zealand local and regional stations [Ristau, 2008]. The differences with respect to the value derived in this study could be due to the contribution of postseismic deformation in the InSAR maps, spanning from 7 days up to 1 month after the Darfield earthquake (Table 1), as well as to the uncertainty in the slip distribution, which is expected to explain differences up to 1.1  1019 Nm (Figure S6 in the auxiliary material). The full-resolution estimate can be therefore considered fully compliant with

6 of 16

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

B08305

Figure 5. Resolution values of the two proposed models for the September 4th, Darfield earthquake. Values, ranging from 0 (completely undefined) to 1 (perfectly resolved), are derived from the diagonal of the Model Resolution Matrix, as described in Appendix B. The full-resolution subdivision is obtained with the algorithm of Atzori and Antonioli [2011], starting from a single rectangular fault; fixed patches have 1 km of size. An alternative 3D view from North can be found in Figure S7 of the auxiliary material. seismological results. We finally stress that the slip distribution derived in this study does not account for any a priori boundary condition; thus our results indicate that, even if resolution strongly decreases with depth, significant slip occurred at greater depth than suggested by previously published models and this implies a higher seismic potential even for short faults in the area.

6. February 22nd, 2011: The First Christchurch Earthquake [19] The Mw 6.2, 22nd February 2011 event was a major Darfield aftershock, which released a seismic moment ranging from 1.9  1018 Nm (USGS-CMT) to 6.2  1018 Nm (GeoNet MT). Due to the strong ground motion [Iizuka et al., 2011] and its shallow depth (5 km, from Bannister et al. [2011]), 185 people lost their life, 2000 were injured and about 105 buildings were affected. Moreover, strong liquefaction phenomena affected large areas of the city [Cubrinovski et al., 2011; Orense et al., 2011; Mucciarelli,

2011; Green et al., 2011]. To model this source we analyzed data from ascending and descending COSMO-SkyMed and ascending ALOS-PALSAR images (Table 3). Temporary GPS survey data, published by Beavan et al. [2011], were also jointly inverted. The small post-seismic displacement detected by the GPS campaign [Beavan et al., 2011] allows us to consider the static displacement measured by InSAR as a good approximation of the coseismic displacement. [20] During InSAR processing, we encountered a diffused and unexpected loss of interferometric coherence in the urban area that prevented the unwrapping of some isolated areas, as already noticed by Beavan et al. [2011] and attributed to ground failure and building damages. The coherence maps in Figures 6a–6c have been calculated for two image pairs characterized by similar spatial baselines, thus minimizing the coherence change due to differences in the acquisition geometry. In several zones, the stability of the radar phase was altered mainly by the soil liquefaction that affected large parts of Christchurch. In Figures 6a–6c

Table 3. February 2011, Christchurch Earthquake Data Sets Baselines

RMS

Satellite

Path

Pair

Look Angle (deg)

Spatial (m)

Temporal (d)

Orbit

Points

Fixed (m)

Full Resolution (m)

COSMO

n.a.

35.9

22

4

Asc

4095

0.03

0.02

COSMO

n.a.

44.5

487

16

Des

3177

0.01

0.01

ALOS

335

38.9

655

46

Asc

975

0.04

0.04

ALOS

336

38.9

1350

138

Asc

2235

0.04

0.03

GPSa

-

19/02/2011 23/02/2011 20/02/2011 08/03/2011 10/01/2011 25/02/2011 27/10/2010 14/03/2011 -

-

-

-

-

70  3

0.03

0.03

a

From Beavan et al. [2011].

7 of 16

B08305

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

Figure 6. Coherence analysis with COSMO-SkyMed data for February 22nd (top) and June 13th (bottom) 2011 Christchurch earthquakes. (a) pre-seismic coherence from images acquired before the February earthquake (pair 02/12/2010–10/01/2011, baseline 37 m); (b) co-seismic coherence map from images acquired across the February earthquake (pair 19/02/2011–23/02/2011, baseline 22 m); (c) coherence drop induced by the earthquake with threshold 0.2, overlapped by boundaries of areas subject to liquefaction from aerial and field survey [Green et al., 2011]; (d) coseismic coherence across June 13th earthquake; (e) postseismic coherence (pair 24/06/2011–10/07/2011, baseline 39 m); (f) coherence drop for the February event. Drop in Figures 6c and 6f is calculated only for areas already showing coherence above 0.4, to avoid the inclusion of inherently low coherence zones. we show a qualitative comparison between the areas affected by liquefaction, derived from aerial and field surveys [Green et al., 2011], and those that show a drastic drop of coherence. The good correspondence between these areas suggests that coherence analysis might be effective in mapping liquefaction phenomena, although a dedicated study would be required. [21] As for the Darfield earthquake, the Christchurch earthquake occurred on previously unknown faults [Beavan et al., 2011], but several factors make its source modeling more complicated. First, the rupture did not reach the surface, thus preventing the unequivocal identification of the fault trace; second, coherence loss occurred in the transition area between the hanging wall and the footwall deformation; finally, part of the deformation is offshore and is completely unmapped in the interferograms. In addition, the patterns of relocated aftershock from Bannister et al. [2011] suggest a complex, segmented rupture for this event. We discarded a single fault model because of the high residuals, especially for the COSMO-SkyMed data, and a fault orientation and mechanism hardly compatible with the aftershock distribution and the USGS-CMT and GeoNet MT solutions. Constraining strike and rake with the USGS-CMT or GeoNet MT solutions lead to deterioration of the already poor fit. On the other hand, the hypothesis of a multiple fault system is suggested by

several authors, based on aftershock distribution [Beavan et al., 2011], InSAR and strong motion data [Elliott et al., 2012; Holden, 2011]. There is a general agreement on the presence of a significant right-lateral strike-slip component, though not constrained in time [Holden, 2011] and location. In the solution from Beavan et al. [2011] its position is driven by two major aftershocks, but this configuration leaves an unmodeled displacement lobe covering the city of Christchurch. [22] In the solution proposed by Elliott et al. [2012] the inversion of body waves and InSAR data distinguishes a northern SW–NE segment, with an almost inverse mechanism, and a southern WSW–ENE segment, with a right lateral, slightly compressive mechanism. Though based only on InSAR and GPS data, our two-fault solution is in good agreement with the solution from Elliott et al. [2012]. In our inversion, the strike angle of the sources was constrained based on the relocated aftershocks and the displacement patterns in the interferograms, while the rake value was recovered by nonlinear inversion of InSAR and GPS. The northern segment shows a rake value of 100 , highly consistent with that from Elliott et al. [2012], while the southern segment shows an almost pure right-lateral mechanism with a very small normal component (Table 4). [23] As for the Darfield earthquake, we examined the slip distribution with patches of equal dimension as well as with

8 of 16

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

B08305

Table 4. February 2011 Christchurch Fault Parameters Linear Inversion Nonlinear Inversion Fault Name Christchurch north Christchurch south

Fixed Patches a

a

Strike (deg) Dip (deg) Rake (deg) Longitude (deg) Latitude (deg) b

30 70b

b

60 70b

100 173

43.556 43.532 TOTAL

172.695 172.682

M0 (Nm) 1.36  10 0.54  1018 1.9  1018b (M 6.1) 18

Full Resolution

Number

M0 (Nm)

Number

144 192

1.96  10 1.79  1018 3.7  1018 (M 6.3)

33 13

18

a

Coordinates of the fault trace center. Parameter fixed a priori.

b

the full-resolution approach. For the former, we introduced the seismic moment as a further constraint, adopting a rigidity modulus of 30 GPa, in agreement with other authors (30 GPa for Beavan et al. [2011]; 32 GPa for Elliott et al. [2012]). As for the Darfield event, the data fit of the two approaches is equally good (see Table 3). In Figures 7 and 8 the observed, modeled and residual displacements are shown for the fullresolution inversion for InSAR and GPS, respectively. [24] With the fixed patch model, 70% of the seismic moment is released by the northern segment, with a peak of 2 m, 4 km deep, with significant slip down to 6–7 km depth. This is in general agreement with the outcomes of other studies based on geodetic data [Beavan et al., 2011; Barnhart et al., 2011; Elliott et al., 2012] but the maximum slip is about half of that retrieved by Holden et al. [2011] from strong motion inversion. The contribution to the seismic moment from the southern segment is much lower; we found a maximum slip of about 1.7 m at 2 km depth. A second peak of about 1 m is located at its western edge, close to the location of two major aftershocks (ML 5.8 and 5.9) which occurred on the same day, and which have been used by Beavan et al. [2011] to constrain the second fault plane. [25] The full-resolution approach leads to a solution with a geodetic moment of 3.8  1018 Nm (corresponding to M 6.3), shared almost equally between the two segments (Table 4). As for the Darfield earthquake models, the main difference between the two solutions is the spread of the dislocation up to a depth of 7 km (Figure 9), where the patch size that can be resolved is on the order of 4 km (Figure 10), while below 7 km we estimate a null mean dislocation (with a standard deviation up to 0.4 m, see Figure S7 of the auxiliary material). The peak of the rupture and its extension at depth is also supported by the slip distribution of Barnhart et al. [2011], the only other study adopting patches of variable size. [26] We note that the ratio between the moments of the two faults differs for the two proposed solutions. Although the presence of strong decorrelation prevents the unequivocal slip attribution to the faults, significant trade-offs that characterize the model covariance matrix for poorly resolved parameters (see AppendixB) might have played an important role. Since the full-resolution approach is also associated with a strong diagonalization of the covariance matrix, we consider this slip distribution more convincing than that with fixed patch size.

7. June 13th, 2011: The Second Christchurch Earthquake [27] On June 13th, a further strong aftershock of Mw 6.0 occurred beneath the city of Christchurch, 5 km east of the hypocenter of the February event. Peak ground accelerations reached 2 g at some sites, 0.4 g in the central city, with

horizontal accelerations dominant [Fry et al., 2011]. Further damage was caused to vulnerable buildings in the city, while cliff collapse and rockfalls occurred on some slopes to the east and south of the city. Liquefaction was again widespread in the southern and eastern residential suburbs. To model this event we processed two interferometric pairs, from a COSMO-SkyMed descending orbit and from an ENVISAT ascending orbit (Table 5 and Figure 11); both interferograms show a high coherence and have been subsampled to a 300 m regular grid. [28] No surface ruptures are present to constrain which of the conjugate planes of the focal mechanism (Figure 1) slipped during the earthquake. We initially adopted the E–W striking fault, in agreement with the previous events, with the interpretation of Barnhart et al. [2011] and with the fringe pattern of the ENVISAT interferogram (see Figure S3 in the auxiliary material). However, this approach led to several inconsistencies, including high residuals in Christchurch and unrealistic rake values. When left free to vary, the rake was estimated to be 97 , almost as for a pure thrust fault, far from the 160 of the USGS-CMT catalog. On the other hand, the rake fixed to a transpressive mechanism led to solutions with considerable high residuals. [29] The distribution of the aftershocks after June 13th (Figure 1), however, supported the adoption of the conjugate plane, as also noted by Sibson et al. [2011]. Aftershocks unambiguously show a NW–SE alignment, dipping westward at an angle of 70 , in almost perfect agreement with the 71 of the USGS-CMT focal mechanism; however we eventually fixed the dip angle to 80 to avoid the unrealistic intersection with the February fault planes. All the other fault parameters were left free in the nonlinear inversion, leading to the fault parameters of Table 6. [30] Good results, of about 1 cm of r.m.s. on the residuals, can be already achieved with the uniform slip solution. We then ran the linear inversion with the fixed patch size and fullresolution approaches, obtaining the slip distribution shown in Figure 9, where this model is shown together with the February faults (see also r.m.s. of the residuals in Table 6). A peak slip value of 3 m is retrieved for a 3  4 km patch at depths from 3 to 5.5 km, with a resolution value of 0.7 (see Figure 10). In this case, the full-resolution approach corresponds to M 6.2 earthquake, but estimates here can be altered by the higher uncertainty (up to 0.6 m, as visible in Figure S7 of the auxiliary material) due to the high dip angle and the absence of data coverage for the offshore displacement field. [31] In addition to the seismic source, both interferograms also show fringe patterns related to locally induced displacements. We just address here the high fringe rate involving the Godley Head promontory, between Harris Bay and Mechanics

9 of 16

B08305

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

Figure 7. (left) Observed, (middle) modeled, and (right) residual displacement maps of COSMO-SkyMed and ALOS-Palsar interferograms relating to the February 22, Mw 6.2, Christchurch earthquake. Color legends are scaled according to the minimum/maximum value for observed and modeled maps, and the same color scheme is adopted for the three residuals maps. Interferograms are shown in Figure S3 in the auxiliary material. Annotations as for Figure 2 are supplied: fault traces, image acquisition dates, satellite path, line-of-sight directions (see Table 3 for data set details). 10 of 16

B08305

B08305

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

Figure 8. Observed (blue) and modeled (red) GPS data from Beavan et al. [2011], for the (a) horizontal and (b) vertical components. Predicted values refer to the full-resolution model shown in Figure 9 and are nearly identical to those with fixed patch size (rms 0.03 m for both approaches, see Table 3). Bay (Figures 11a–11d); here we found a SW–NW gradient of 9 fringes in the COSMO interferogram, equivalent to about 14 cm, and a NW–SE gradient of 3.5 fringes in the ascending ENVISAT interferogram, corresponding to about 10 cm. This

could be ascribed to additional local subsidence of the promontory, but the converging direction suggests the presence of a significant horizontal component, that could be identified with the availability of 3D GPS measurement from the site “B9DV.”

Figure 9. Slip distribution for the February 22nd and June 13th 2011 Christchurch earthquakes obtained with fixed patch size of 1 km and full-resolution subdivision. Two different 3D views are provided to show the spatial relationship between the February and the June faults. Resolution values are shown in Figure 10. Fault parameters are reported in Tables 4 and 6 for February and June events, respectively. The dislocation standard deviation is given in Figure S7 in the auxiliary material. 11 of 16

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

B08305

Figure 10. Resolution values of the fixed patch and full-resolution models, with the same 3D view as in Figure 9. Refer to Appendix B for a description on how resolution values are obtained from the Model Resolution Matrix. [32] Finally, we point out that this event was also accompanied by the occurrence of soil liquefaction, the discrimination of which in the field is not easy, because of its focus in the same area of the previous event. However, coherence change analysis overcomes such a limitation, since it is related to the change of soil backscattering characteristics, regardless of the presence of ejecta from a previous event. We recorded such a change in backscattering across the June event, finding the results shown in Figures 6d–6f. From Figure 6e, which shows a coherence map from a post-event pair, we see that the coherence in the Christchurch city center reverted to its expected stability after the earthquake.

8. Stress Variation Analysis [33] Using the retrieved variable patch slip distributions, we calculated the stress change induced by the September event on the two fault planes of the February 22nd event,

and the joint effect of these two earthquakes on the June 13th event. A static analysis of the Coulomb Failure Function variation (DCFF e.g., Harris [1998] for details) was carried out; positive DCFF means the effect of previous events advanced subsequent shocks toward failure, while negative DCFF represents fault relaxation and failure-time delay. Our aim is not to demonstrate that stress transfer is the mechanism to explain the main shock sequence in the region, but to investigate whether the effect of each earthquake contributed in stressing the fault planes of the following large events. Furthermore, we do not focus on other major aftershocks, such as the ML 5.8 aftershock following the February 22nd earthquake, because they mainly lay too close to the fault planes to consider the stress redistribution alone as a separate effect on their trigger. [34] We did not project the stress variation on the entire seismic sequence, due to the lack of a reliable mean fault geometry, as clearly indicated by the dramatically different

Table 5. June 2011, Christchurch Earthquake Data Sets Baselines

RMS

Satellite

Path

Pair

Look Angle (deg)

Spatial (m)

Temporal (d)

Orbit

Points

Fixed (m)

Full Resolution (m)

ENVISAT

195

41.0

144

30

Asc

7384

0.01

0.01

COSMO

913

08/06/2011 08/07/2011 21/04/2011 24/06/2011

43.3

61

64

Des

2372

0.01

0.01

12 of 16

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

B08305

B08305

Figure 11. (left) Observed, (middle) modeled, and (right) residual displacement maps of COSMO-SkyMed and ENVISAT-ASAR interferograms relating to the June 13rd 2011, Mw 6.0, Christchurch earthquake. Fault trace, image acquisition dates, satellite paths and line-of-sight directions as for Figure 2. Ancillary information for every data set is provided in Table 5. Insets in Figures 11a and 11c show the interferometric fringes due to a local deformation of the Godley Head promontory, representing 1.5 cm and 2.8 cm of displacement per fringe in the line-of-sight for COSMO-SkyMed and ENVISAT, respectively. geometries of all the inferred fault segments. We focused only on the direct possible triggering effect between the largest events. The stress variation tensors were projected on the fault planes geometries, along the rake directions of the receiver faults. Geometries and rake directions were previously retrieved from the coseismic nonlinear inversion described above and summarized in Tables 2, 4, and 6. A similar analysis has been previously performed, among others, by Barnhart et al. [2011], Stramondo et al. [2011] and Zhan et al. [2011]. However, none of these studies had included the double fault identified for the February Christchurch earthquake, nor the June event fault. [35] The effect of the Darfield earthquake on the February Christchurch event is unambiguous; both of the modeled faults show a significant increase in stress, up to 2.5 bars in the shallowest part of the southern fault (Figure 12). The

February Christchurch earthquake shows an anomalously high stress drop, but 2.5 bars still represent about 5% of the total stress drop which occurred during the February event [Barnhart et al., 2011]. Both February planes are only positively loaded, with the southern one characterized by higher values; although the hypocenter cannot be undoubtedly ascribed to either faults, it lays in a positively loaded area. The modeling suggests that it is possible that the February earthquake was advanced in its seismic cycle by the stress readjustment caused by the strain induced by the September Darfield earthquake. [36] We calculated, for the June earthquake, the stress readjustment from the September and February events. We computed the separate contribution of these events on the June fault plane; because of the geometry and the distance of the September faults, the stress induced by the latter is

Table 6. June 2011 Christchurch Fault Parameters Linear Inversion Nonlinear Inversion Fault Name Christchurch

Strike (deg) 150

Dip (deg) b

80

Rake (deg) 11

Fixed Patches

Longitudea (deg) 172.769

b

Latitudea (deg) 43.544

a

Coordinates of the fault trace center. Parameter fixed a priori.

b

13 of 16

M0 (Nm) 2.3  10

18

(M 6.2)

Full Resolution

Number 192

M0 (Nm) 2.3  10

18

(M 6.2)

Number 31

B08305

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

Figure 12. (a) Stress change based on the Coulomb Failure Function [Harris, 1998] induced by the September 2010, Darfield, earthquake on the fault plane of the February 2011, first Christchurch event (epicenter with a red star). (b) Stress change induced on the June 2011, second Christchurch earthquake (epicenter with a yellow star) by September 2010 and February 2011 events. completely masked by the effect of the February earthquake. The vicinity of the February seismogenic faults induces a strong gradient in the stress variation. The hypocenter of the June 13th earthquake, however, is clearly located in a positively loaded fault zone (Figure 12b). Dynamic stress redistribution cannot explain the triggering of the June event; the few months delay could be due to the different spatial orientation of the faults, or to more complex mechanisms such as fluid migration that has been found responsible of the loading increase with time on earthquake sequences [Noir et al., 1997; Antonioli et al., 2005].

9. Conclusion [37] The 2010–2011 Canterbury (New Zealand) seismic sequence involved a complex series of events, raising important questions on the seismic hazard of this region and on the relationships between earthquakes. Although these issues remain far from being solved, we provide a contribution based on the analysis of remotely sensed data, introducing constraints from seismological and geological data. The large availability of space-based data allowed a multidimensional picture, in space and in time, of the seismogenic crust, consisting of a complex set of active faults. This picture was used

B08305

to infer the relationships between events, using Coulomb failure analysis. We showed that static stress variations consistently favored the advance of fault failure, but on the other hand the occurrence of the main events cannot be ascribed to this variation alone; other factors (e.g., fluid migration) could justify the interval of several months between the main events. In the light of this, a further study based on the SBAS time series analysis of tens of COSMO-SkyMed images between September and February is ongoing and will address the transient between the larger events to give more insight into their interaction. [38] Another important aspect analyzed in this work concerns the way geodetic data were used to infer the seismic model. InSAR data, despite their dense coverage, always show the surface effects and their use to infer parameters at depth requires caution. We have shown that at a depth of 7– 8 km, especially for high dipping angles, we can only infer large scale average values. We therefore compared the classical approach with patches of predefined dimension with that involving patches dimensioned according to the model resolution matrix, referred to as full-resolution approach. [39] The two solutions are substantially equivalent in terms of data fit, as shown by the r.m.s. of the residuals in Table 1, 3, and 5. However, with the full-resolution approach we found that significant slip might have occurred at depth greater than those inferred from other studies and this would suggest an increase of the earthquake hazard potential even for short faults in the area. We finally addressed the suitability of coherence change to rapidly detect areas repeatedly affected by soil liquefaction.

Appendix A: Offset Tracking [40] Intensity cross-correlation, also referred to in SAR literature as speckle tracking [Gray et al., 2001], intensity offsettracking [Strozzi et al., 2002] or intensity pixel-tracking, was applied to the Aug. 13th 2010 and Sep. 28th 2010 PALSAR Fine Beam Dual image pair, using the commercial ISP software distributed by GAMMA Remote Sensing and Consulting AG, and a set of scripts and utility programs created at the Technical University of Denmark. The cross-correlation window size was 64  256 pixels in slant-range and azimuth respectively, corresponding to about 1 km  1 km on the ground. Complex data were oversampled by a factor two, prior to detection, and weighted by a Kaiser window for sidelobe reduction. Cross-correlations were computed on a 10  50 pixel grid, corresponding to 200 m  200 m on the ground and the sub-pixel shift of the cross-correlation function peak with respect to the origin, at the center of the data patch, was computed by a local quadratic least squares fit. The resulting slant-range and azimuth offset fields were smoothed using a 5  5 moving average window, and outliers discarded based on cross-correlation Signal-to-noise ratio thresholding [de Lange et al., 2007]. At a later processing stage a second outlier removal step was carried out, based on the local deviation from the median of the magnitude and direction of the displacement vectors [Pritchard et al., 2005]. [41] The offset fields were then calibrated for a single constant assumed as a zero-displacement reference, and converted to geocoded displacements in slant-range and azimuth on a 500 m  500 m posting. The error prediction framework of [Mohr and Merryman Boncori, 2008] was adapted to associate

14 of 16

B08305

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

predicted displacement error standard deviations to the measurements, accounting for the spatial correlation of atmospheric delays with the model of [Merryman Boncori and Mohr, 2008], and using a simplified model of the spatial correlation introduced by correlation window overlap and lowpass filtering [Joughin, 2002]. [42] The azimuth displacement map (Figure 3g), exhibited distinctive linear features, aligned at a mean angle of 23.5 deg. from the east, with peak-to-peak amplitude of 0.2 pixels (70 cm). This is the typical signature of azimuth image mis-registration errors, caused by the uncompensated Doppler modulation induced by km scale variations of ionospheric Total Electronic Content (TEC) [Gray and Mattar, 2000]. This signal was estimated and removed following [Raucoules and De Michele, 2010]. Areas with large slant-range displacements were masked and the azimuth displacement map was rotated clockwise by 5.5 deg (the flight direction is at about 18 deg. from north). A one-degree polynomial was fitted in the horizontal direction in the upper part of the image and a 0 degree polynomial in the lower part, where the extension of the masked-out area did not allow a reliable linear fit. The estimated correction and the corrected displacement map are shown in Figure S8 in the auxiliary material. [43] The predicted mean accuracies after the ionospheric delay corrections were in the order of 1/60th of the resolution cell, (Figure S8d in the auxiliary material), corresponding to 7 cm in azimuth, for a coherence level of 0.5. This is consistent with theoretical expectations, considering a reduction of the effective correlation window area by a factor 4, due to the Kaiser weighting [Bamler and Eineder, 2005]. The expected phase screen associated to the azimuth streaks was also computed, but was however not observable in the interferometric phase. This is probably due to the fact that its order of magnitude is less or equal to the standard deviation of tropospheric turbulent delay differences, typically in the order of 1 cm at midlatitudes [Merryman Boncori and Mohr, 2008]. This also prevented determination of the appropriate scale factor and integration constant for the quantification of the phase screen, based on the estimated azimuth offsets.

Appendix B: Full-Resolution Inversion [44] The full-resolution approach is used to find the best fault discretization able to maintain a meaningful spatial detail of the slip distribution. In this inversion, the linear relation between data and model parameters is such to provide the best configuration according to the model resolution matrix R [Menke, 1989], defined as: R ¼ Gg  G

where G is the Green’s function matrix (without a priori constraints) and Gg is the generalized inverse used to obtain the slip distribution, i.e., that with the Laplacian operator. Gg, regardless of the way it is calculated, must contains a certain amount of regularization to get reliable solutions. However, in the full-resolution approach damping affects the patch dimension and not the slip distribution. In this way we minimize spurious features only due mathematical arrangement of the slip.

B08305

[45] An iterative algorithm, described in Atzori and Antonioli [2011], starting from a rectangular fault, subdivides the plane into patches of variable size keeping the model resolution matrix R as close as possible to the identity matrix I. Observing this condition guarantees that every parameter is resolved at its best, since the diagonal of R describes the resolution of every parameter, from 0 (completely undetermined) to 1 (perfectly solved). These are the values shown in Figures 5 and 10. [46] The maximization of the patch number is a second rule driving the algorithm and is necessary since any rough subdivision in few patches is in general perfectly resolved. [47] A further advantage with the full-resolution subdivision concerns model uncertainty, defined as ½cov m ¼ Gg ½cov dGg T

where [cov m] and [cov d] are the full variance-covariance matrices of model and data, respectively. It follows that in the full-resolution approach the off-diagonal values of [cov m] are minimized and trade-offs between parameters are strongly reduced, thus leading to a more realistic assessment of the standard deviation. [48] A detailed description of the algorithm and the role played by the data distribution (and other inversion parameters) can be found in Atzori and Antonioli [2011]. [49] Acknowledgments. Russell Green for data on liquefied areas after the February earthquake; Paolo Riccardi (SARMAP) for the assistance in InSAR processing with SARscape; John Beavan (GNS) and two anonymous reviewers for their valuable comments; Simona Zoffoli (ASI) and JAXA for ALOS data. COSMO-SkyMed images were provided by the Italian Space Agency in the SiGRIS project framework (www.sigris.it).

References Antonioli, A., D. Piccinini, L. Chiaraluce, and M. Cocco (2005), Fluid flow and seismicity pattern: Evidence from the 1997 Colfiorito (central Italy) seismic sequence, Geophys. Res. Lett., 32, L10311, doi:10.1029/2004GL022256. Atzori, S., and A. Antonioli (2011), Optimal fault resolution in geodetic inversion of coseismic data, Geophys. J. Int., 185, 529–538, doi:10.1111/ j.1365-246X.2011.04955.x. Atzori, S., I. Hunstad, M. Chini, S. Salvi, C. Tolomei, C. Bignami, S. Stramondo, E. Trasatti, A. Antonioli, and E. Boschi (2009), Finite fault inversion of DInSAR coseismic displacement of the 2009 L’Aquila earthquake (central Italy), Geophys. Res. Lett., 36, L15305, doi:10.1029/2009GL039293. Bamler, R., and M. Eineder (2005), Accuracy of differential shift estimation by correlation and split-bandwidth interferometry for wideband and delta-k SAR systems, IEEE Geosci. Remote Sens. Lett., 2(2), 151–155, doi:10.1109/ LGRS.2004.843203. Bannister, S., B. Fry, M. Reyners, J. Ristau, and H. Zhang (2011), Finescale relocation of aftershocks of the 22 February Mw 6.2 Christchurch earthquake using double-difference tomography, Seismol. Res. Lett., 82(6), 839–845, doi:10.1785/gssrl.82.6.839. Barnhart, W. D., M. J. Willis, R. B. Lohman, and A. K. Melkonian (2011), InSAR and optical constraints on fault slip during the 2010–2011 New Zealand earthquake sequence, Seismol. Res. Lett., 82(6), 815–823, doi:10.1785/gssrl.82.6.815. Beavan, J., S. Samsonov, M. Motagh, L. Wallace, S. Ellis, and N. Palmer (2010), The Darfield (Canterbury) earthquake: Geodetic observations and preliminary source model, Bull. N. Z. Soc. Earthquake Eng., 43(4), 228–235. Beavan, J., E. Fielding, M. Motagh, S. Samsonov, and N. Donnelly (2011), Fault location and slip distribution of the 22 February 2011 Mw 6.2 Christchurch, New Zealand, earthquake from geodetic data, Seismol. Res. Lett., 82(6), 789–799, doi:10.1785/gssrl.82.6.789. Cox, S., and R. Sutherland (2007), Regional geological framework of South Island, New Zealand, and its significance for understanding the active plate boundary, in A Continental Plate Boundary: Tectonics at South Island, New Zealand, Geophys. Monogr. Ser., vol. 175, edited by D. Okaya, T. Stern, and F. Davey, pp. 19–46, AGU, Washington, D. C., doi:10.1029/175GM03.

15 of 16

B08305

ATZORI ET AL.: THE CANTERBURY SEQUENCE FROM INSAR

Cubrinovski, M., J. D. Bray, M. Taylor, S. Giorgini, B. Bradley, L. Wotherspoon, and J. Zupan (2011), Soil liquefaction effects in the central business district during the February 2011 Christchurch earthquake, Seismol. Res. Lett., 82(6), 893–904, doi:10.1785/gssrl.82.6.893. de Lange, R., A. Luckman, and T. Murray (2007), Improvement of satellite radar feature tracking for ice velocity derivation by spatial frequency filtering, IEEE Trans. Geosci. Remote Sens., 45(7), 2309–2318, doi:10.1109/ TGRS.2007.896615. Elliott, J. R., E. Nissen, P. C. England, J. A. Jackson, S. Lamb, Z. Li, M. Oehlers, and B. E. Parsons (2012), Slip in the 2010–2011 Canterbury earthquakes, New Zealand, J. Geophys. Res., 117, B03401, doi:10.1029/ 2011JB008868. Forsyth, P. J., D. J. A. Barrell, and R. Jongens (2008), Geology of the Christchurch area, Geol. Map 16, scale 1:250,000 Inst. of Geol. and Nucl. Sci., GNS Sci., Lower Hutt, N. Z. Franceschetti, G., and R. Lanari (1999), Synthetic Aperture Radar Processing, 328 pp., CRC Press, Boca Raton, Fla. Fry, B., R. Benites, and A. Kaiser (2011), The character of accelerations in the Mw 6.2 Christchurch earthquake, Seismol. Res. Lett., 82(6), 846–852, doi:10.1785/gssrl.82.6.846. Gledhill, K., J. Ristau, M. Reyners, B. Fry, and C. Holden (2010), The Darfield (Canterbury) earthquake of September 2010: Preliminary seismological report, Bull. N. Z. Soc. Earthquake Eng., 43(4), 215–221. Gray, A. L., and K. E. Mattar (2000), Influence of ionospheric electron density fluctuations on satellite radar interferometry, Geophys. Res. Lett., 27(10), 1451–1454, doi:10.1029/2000GL000016. Gray, A. L., N. Short, K. E. Mattar, and K. C. Jezek (2001), Velocities and flux of the Filchner ice shelf and its tributaries determined from speckle tracking interferometry, Can. J. Remote Sens., 27(3), 193–206. Green, R. A., C. Wood, B. Cox, M. Cubrinovski, L. Wotherspoon, B. Bradley, T. Algie, J. Allen, A. Bradshaw, and G. Rix (2011), Use of DCP and SASW tests to evaluate liquefaction potential: Predictions vs. observations during the recent New Zealand earthquakes, Seismol. Res. Lett., 82(6), 927–938, doi:10.1785/gssrl.82.6.927. Guidotti, R., M. Stupazzini, C. Smerzini, R. Paolucci, and P. Ramieri (2011), Numerical study on the role of basin geometry and kinematic seismic source in 3D ground motion simulation of the 22 February 2011 MW 6.2 Christchurch earthquake, Seismol. Res. Lett., 82(6), 767–782, doi:10.1785/gssrl.82.6.767. Harris, R. A. (1998), Introduction to special section: Stress trigger, stress shadows, and implication for seismic hazard, J. Geophys. Res., 103(B10), 24,347–24,358, doi:10.1029/98JB01576. Holden, C. (2011), Kinematic source model of the 22 February 2011 Mw 6.2 Christchurch earthquake using strong motion data, Seismol. Res. Lett., 82(6), 783–788, doi:10.1785/gssrl.82.6.783. Holden, C., J. Beavan, B. Fry, M. Reyners, J. Ristau, R. Van Dissen, and P. Villamor (2011), Preliminary source model of the Mw 7.1 Darfield earthquake from geological, geodetic and seismic data, paper presented at Ninth Pacific Conference on Earthquake Engineering: Building an Earthquake-Resilient Society, N. Z. Soc. for Earthquake Eng., Auckland, 14–16 April. Iizuka, H., Y. Sakai, and K. Koketsu (2011), Strong ground motion and damage conditions associated with seismic stations in the February 2011 Christchurch, New Zealand, earthquake, Seismol. Res. Lett., 82(6), 875–881, doi:10.1785/gssrl.82.6.875. Joughin, I. (2002), Ice-sheet velocity mapping: A combined interferometric and speckle-tracking approach, Ann. Glaciol., 34, 195–201, doi:10.3189/ 172756402781817978. Lungarini, L., C. Troise, M. Meo, and G. De Natale (2005), Finite element modelling of topographic effects on elastic ground deformation at Mt. Etna, J. Volcanol. Geotherm. Res., 144, 257–271, doi:10.1016/j. jvolgeores.2004.11.031. Massonnet, D., M. Rossi, C. Carmona, F. Adragna, G. Peltzer, K. Feigl, and T. Rabaute (1993), The displacement field of the Landers earthquake mapped by radar interferometry, Nature, 364, 138–142, doi:10.1038/ 364138a0.

B08305

Menke, W. (1989), Geophysical Data Analysis: Discrete Inverse Theory, Academic, San Diego, Calif. Merryman Boncori, J. P., and J. J. Mohr (2008), A tunable closed form model for the structure function of tropospheric delay, IEEE Geosci. Remote Sens. Lett., 5(2), 222–226, doi:10.1109/LGRS.2008.915738. Mohr, J. J., and J. P. Merryman Boncori (2008), An error prediction framework for interferometric SAR data, IEEE Trans. Geosci. Remote Sens., 46 (6), 1600–1613, doi:10.1109/TGRS.2008.916213. Mucciarelli, M. (2011), Ambient noise measurements following the 2011 Christchurch earthquake: Relationships with previous microzonation studies, liquefaction, and nonlinearity, Seismol. Res. Lett., 82(6), 919–926, doi:10.1785/gssrl.82.6.919. Noir, J., E. Jacques, S. Bekri, P. M. Adler, P. Tapponier, and G. C. P. King (1997), Fluid flow triggered migration of events in the 1989 Dobi earthquake sequence of central Afar, Geophys. Res. Lett., 24, 2335–2338, doi:10.1029/97GL02182. Okada, Y. (1985), Surface deformation due to shear and tensile faults in a half-space, Bull. Seismol. Soc. Am., 75(4), 1135–1154. Orense, R. P., T. Kiyota, S. Yamada, M. Cubrinovski, Y. Hosono, M. Okamura, and S. Yasuda (2011), Comparison of liquefaction features observed during the 2010 and 2011 Canterbury earthquakes, Seismol. Res. Lett., 82(6), 905–918, doi:10.1785/gssrl.82.6.905. Pritchard, H., T. Murray, A. Luckman, T. Strozzi, and S. Barr (2005), Glacier surge dynamics of Sortbrae, east Greenland, from synthetic aperture radar feature tracking, J. Geophys. Res., 110, F03005, doi:10.1029/ 2004JF000233. Quigley, M., et al. (2010), Surface rupture of the Greendale fault during the Mw 7.1 Darfield (Canterbury) earthquake, New Zealand: Initial findings, Bull. N. Z. Soc. Earthquake Eng., 43, 236–242. Quigley, M., R. Van Dissen, N. Lichtfield, P. Villamor, B. Duffy, D. Barrell, K. Furlong, T. Stahl, E. Bilderback, and D. Noble (2011), Surface rupture during the 2010 Mw 7.1 Darfield (Canterbury) earthquake: Implications for fault rupture dynamics and seismic-hazard analysis, Geology, 40, 55–58, doi:10.1130/G32528.1. Raucoules, D., and M. De Michele (2010), Assessing ionospheric influence on L-band SAR data: Implications on coseismic displacement measurements of the 2008 Sichuan earthquake, IEEE Geosci. Remote Sens. Lett., 7(2), 286–290, doi:10.1109/LGRS.2009.2033317. Ristau, J. (2008), Implementation of routine regional moment tensor analysis in New Zealand, Seismol. Res. Lett., 79, 400–415, doi:10.1785/ gssrl.79.3.400. Sibson, R., F. Ghisetti, and J. Ristau (2011), Stress control of an evolving strike-slip fault system during the 2010–2011 Canterbury, New Zealand, sequence, Seismol. Res. Lett., 82(6), 824–832, doi:10.1785/ gssrl.82.6.824. Stirling, M. W., G. H. McVerry, and K. R. Berryman (2002), A new seismic hazard model for New Zealand, Bull. Seismol. Soc. Am., 92, 1878–1903, doi:10.1785/0120010156. Stramondo, S., C. Kyriakopoulos, C. Bignami, M. Chini, D. Melini, M. Moro, M. Picchiani, M. Saroli, and E. Boschi (2011), Did the September 2010 (Darfield) earthquake trigger the February 2011 (Christchurch) event?, Nat. Sci. Rep., 1, 98, doi:10.1038/srep00098 Strozzi, T., A. Luckman, T. Murray, U. Wegmuller, and C. L. Werner (2002), Glacier motion estimation using SAR offset-tracking procedures, IEEE Trans. Geosci. Remote Sens., 40(11), 2384–2391, doi:10.1109/ TGRS.2002.805079. Wallace, L. M., R. J. Beavan, R. McCaffrey, K. R. Berryman, and P. Denys (2007), Balancing the plate motion budget in the South Island, New Zealand using GPS, geological and seismological data, Geophys. J. Int., 168(1), 332–352, doi:10.1111/j.1365-246X.2006.03183.x. Zhan, Z., B. Jin, S. Wei, and R. W. Graves (2011), Coulomb stress change sensitivity due to variability in main shock source models and receiving fault parameters: A case study of the 2010–2011 Christchurch, New Zealand, earthquakes, Seismol. Res. Lett., 82(6), 800–814, doi:10.1785/ gssrl.82.6.800.

16 of 16