Illuminating the Pre‐, Co‐, and Post‐Seismic Phases of the 2016 M7.8 Kaikōura Earthquake With 10 Years of Seismicity

Abstract

The 2016 M7.8 Kaikōura earthquake is one of the most complex earthquakes in recorded history, with significant rupture of at least 21 crustal faults. Using a matched‐filter detection routine, precise cross‐correlation pick corrections, and accurate location and relocation techniques, we construct a catalog of 33,328 earthquakes between 2009 and 2020 on and adjacent to the faults that ruptured in the Kaikōura earthquake. We also compute focal mechanisms for 1,755 of the earthquakes used as templates. Using this catalog we reassess the rupture pathway of the Kaikōura earthquake. In particular we show that: (a) the earthquake nucleated on the Humps Fault; (b) there is a likely linking offshore reverse fault between the southern fault system and the Papatea Fault, which could explain the anomalously high slip on the Papatea Fault; (c) the faults that ruptured in the 2013 Cook Strait sequence were reactivated by the Kaikōura earthquake and may have played a role in the termination of the earthquake; and (d) no seismicity on an underlying subduction interface is observed beneath almost all of the ruptured region suggesting that if deformation did occur on the plate interface then it occurred aseismically and did not play a significant role in generating co‐seismic ground motion.

Full text

1. Introduction
The November 2016 Kaikōura M 7.8 earthquake ruptured at least 21 faults in the Marlborough Fault Zone at
the transition from subduction on the Hikurangi subduction zone to on-land transpression (Figure1; Hamling
et al., 2017; Kaiser et al., 2017; Litchfield et al., 2018). This complex earthquake involved a wide range of
co-seismic faulting styles, producing dextral, sinistral, reverse and normal surface ruptures (Clark etal.,2017).
In addition to the extensive crustal faulting, the underlying subduction interface may have slipped co-seismically
(Bai etal.,2017; T. Wang etal.,2018), although regional data show little evidence for this (Hamling etal.,2017;
Holden etal.,2017).
The transpressional rupture cascade resulted in significant surface rupture of multiple previously known and
unknown faults (Litchfield etal.,2018). The complexity of the earthquake rupture (Hamling etal.,2017) has
to date precluded the robust constraint of the role of individual faults within the rupture sequence (e.g., Holden
et al., 2017) and the dynamics of the rupture propagation and termination (Ando & Kaneko, 2018; Ulrich
etal.,2019). When modeling such complex ruptures, the identification of all major participating faults has a signif-
icant impact on where the inferred slip is concentrated (e.g., Hamling etal.,2017) and the propagation sequence
Abstract The 2016 M7.8 Kaikōura earthquake is one of the most complex earthquakes in recorded
history, with significant rupture of at least 21 crustal faults. Using a matched-filter detection routine, precise
cross-correlation pick corrections, and accurate location and relocation techniques, we construct a catalog
of 33,328 earthquakes between 2009 and 2020 on and adjacent to the faults that ruptured in the Kaikōura
earthquake. We also compute focal mechanisms for 1,755 of the earthquakes used as templates. Using this
catalog we reassess the rupture pathway of the Kaikōura earthquake. In particular we show that: (a) the
earthquake nucleated on the Humps Fault; (b) there is a likely linking offshore reverse fault between the
southern fault system and the Papatea Fault, which could explain the anomalously high slip on the Papatea
Fault; (c) the faults that ruptured in the 2013 Cook Strait sequence were reactivated by the Kaikōura earthquake
and may have played a role in the termination of the earthquake; and (d) no seismicity on an underlying
subduction interface is observed beneath almost all of the ruptured region suggesting that if deformation did
occur on the plate interface then it occurred aseismically and did not play a significant role in generating
co-seismic ground motion.
Plain Language Summary The 2016 Kaikōura earthquake in the South Island of New Zealand, is
one of the most complex earthquakes reported. While extensive geological work has been undertaken to map
the surface faulting in the earthquake, it remains unclear how these faults are linked together at depth. In this
paper we document the construction of a dense, long-duration catalog of earthquakes that occurred on and
around the faults that slipped in the Kaikōura earthquake. Using this catalog of 33,328 earthquakes we are able
to illuminate likely sub-surface links between faults and investigate how these faults slipped before and after
the Kaikōura earthquake. We show that offshore faults provide a link between the southern faults, where the
earthquake started, and the northern faults, where the highest slip occurred. We also show that the earthquake
stopped on faults that had previously slipped in the 2013 Cook Strait earthquakes, and which likely played
a role in earthquake arrest. Finally we see no evidence for elevated seismicity on the underlying subduction
interface beneath the faults that slipped in the Kaikōura earthquake.
CHAMBERLAIN ETAL.
© 2021. The Authors.
This is an open access article under
the terms of the Creative Commons
Attribution-NonCommercial-NoDerivs
License, which permits use and
distribution in any medium, provided the
original work is properly cited, the use is
non-commercial and no modifications or
adaptations are made.
Illuminating the Pre-, Co-, and Post-Seismic Phases of the 2016
M7.8 Kaikōura Earthquake With 10Years of Seismicity
C. J. Chamberlain1 , W. B. Frank2 , F. Lanza3 , J. Townend1 , and E. Warren-Smith4
1School of Geography, Environment and Earth Sciences, Victoria University of Wellington, Wellington, New Zealand,
2Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA, USA,
3Swiss Seismological Service, ETH Zürich, Zürich, Switzerland, 4GNS Science, Lower Hutt, New Zealand
Key Points:
• 10-year long matched-filter derived
catalog of 33,328 earthquakes
surrounding the 2016 Kaikōura
earthquake
• Observed offshore reverse faulting
provides a direct and viable rupture
pathway
• No detectable seismicity occurs on
the subduction interface, and any
deformation there is aseismic
Supporting Information:
Supporting Information may be found in
the online version of this article.
Correspondence to:
C. J. Chamberlain,
calum.chamberlain@vuw.ac.nz
Citation:
Chamberlain, C. J., Frank, W. B.,
Lanza, F., Townend, J., & Warren-
Smith, E. (2021). Illuminating the
pre-, co-, and post-seismic phases of
the 2016 M7.8 Kaikōura earthquake
with 10years of seismicity. Journal
of Geophysical Research: Solid Earth,
126, e2021JB022304. https://doi.
org/10.1029/2021JB022304
Received 24 APR 2021
Accepted 18 JUL 2021
Corrected 15 JUL 2022
This article was corrected on 15 JUL 2022.
See the end of the full text for details.
10.1029/2021JB022304
RESEARCH ARTICLE
1 of 24

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
2 of 24
from one fault to the next, exemplified by the different rupture pathways modeled by Ando and Kaneko(2018)
and Ulrich etal.(2019). The implications of these models are wide ranging: from a general understanding of how
earthquakes are able to propagate through complex fault systems, to more local implications for seismic hazard
in central New Zealand.
Figure 1. Main panel: GeoNet short-period and broadband seismographs (orange inverted triangles) used in this study for detection and picking, temporary
seismographs (orange squares) used solely for picking, and continuous GNSS receivers (green triangles) active during the Kaikōura post-seismic period. Dashed lines
mark the modeled subduction interface from C. A. Williams etal.(2013), and solid black lines mark faults of the NZ Active Fault Database (R. Langridge etal.,2016).
Red lines mark the mapped surface ruptures of the Kaikōura earthquake (Clark etal.,2017), with fault names labeled. Inset: Regional setting of the Kaikōura region
showing additional seismographs used for detection and location as inverted orange triangles. The location of the main panel is outlined as a red box, the region studied
by Lanza etal.(2019) is shown as a blue box, and solid and dashed lines are the active fault database and modeled subduction interface respectively.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
3 of 24
Almost all published models of co-seismic and post-seismic deformation in the Kaikōura earthquake have been
based on simplified fault models derived from near-surface geological data (e.g., Clark etal.,2017; Hamling
et al., 2017; Holden et al., 2017; T. Wang etal., 2018; Xu et al.,2018). While these data provide essential
controls, they do not provide robust information on the fault structure at depth, where most of the slip happens
during earthquakes. Accurate earthquake catalogs provide a viable tool to constrain fault geometry at depth (e.g.,
Plesch etal.,2020), but have thus far been unavailable for the Kaikōura region, apart from the relatively small
catalog developed by Lanza etal.(2019), and sparse moment tensor analysis by Cesca etal.(2017) and spatially
non-uniform relocations of Mouslopoulou etal. (2019). Such catalogs of seismicity can also help illuminate other
elements of the Kaikōura earthquake, including its relationship to prior seismicity in New Zealand, and how the
various faults respond post-seismically.
1.1. Co-Seismic Kinematics and Rupture Propagation
Kinematic rupture models (Holden et al., 2017) show that the Kaikōura rupture started slowly on the
Humps-Hundalee Fault system (Nicol etal.,2018; J. N. Williams etal., 2018). However, hypocenter estimates
of the Kaikōura earthquake vary from being consistent with nucleation on the Humps Fault (Lanza etal.,2019;
Nicol etal.,2018), to being as much as 7–15km off the Humps Fault (according to the GeoNet (www.geonet.
org.nz/earthquake/technical/2016p858000, last accessed April 24, 2021) and USGS solutions respectively (earth-
quake.usgs.gov/earthquakes/eventpage/us1000778i, last accessed April 24, 2021)).
Once initiated, the rupture propagated north-east towards the Hope Fault, but only produced a minor surface
rupture (Hamling et al., 2017; Litchfield et al., 2018) of this fault, which previous paleoseismic studies have
indicated to have a high Quaternary slip-rate (Litchfield etal.,2018). The rupture then stepped onto the Jordan
Thrust-Kekerengu system where the maximum co-seismic surface offset of 11.8m dextral occurred on the
Kekerengu Fault (Kearse etal.,2018). The dominantly N-S-striking Papatea Fault, which intersects the junction
between the Jordan Thrust and the Kekerengu Fault, also ruptured with up to 9.5m of uplift and 6.1m of sinistral
motion (R. M. Langridge etal., 2018). Previous authors (e.g., Hamling etal.,2017; Holden et al., 2017) have
noted that the high slip on the short (c. 19km long) Papatea fault cannot be fit by elastic rupture models. The
Papatea Fault intersects the Jordan Thrust-Kekerengu system at the point where dextral slip increases from the
Jordan Thrust to the Kekerengu, and on-fault dip-slip motion changes sense, from normal on the Jordan Thrust to
reverse on the Kekerengu (Kearse etal.,2018). This normal motion (NW down) on the Jordan Thrust appears not
to be the dominant long-term sense of motion, with higher mountains on the NW side attesting to the dominantly
oblique-reverse motion on the Jordan Thrust and Upper Kowhai Faults on geological timescales (Van Dissen &
Yeats,1991).
The details of the rupture pathway between the southern Humps-Hundalee fault system and the Kekerengu
Fault are not well-resolved and two main pathways have been postulated. First, the offshore route, from the
Hundalee Fault to the Papatea Fault via mostly unmapped offshore thrust faults. This trajectory appears consist-
ent with a range of observations including off-fault damage at the Papatea-Jordan Thrust-Kekerengu junction
(Klinger etal.,2018), kinematic (Holden etal.,2017) and dynamic rupture simulations (Ulrich etal.,2019), and
tsunami modeling (Bai etal.,2017; Gusman etal.,2018). The second scenario involves rupture jumping from the
Hundalee Fault to the Upper Kowhai Fault and onto the Jordan Thrust and Kekerengu Faults with limited slip on
the intermediate Whites (Ando & Kaneko,2018) and inferred Snowflake Spur Faults (Zinke etal., 2019). The
lack of resolution of the fault network and possible inter-connections at depth inferred from surface observations
alone mean that it remains unclear which scenario actually occurred.
The rupture continued to propagate northwards onto the Needles Fault and other faults in the Cape Campbell
region before terminating near Cape Campbell itself (Kearse etal.,2018), in the region of the 2013


6.6 Cook
Strait and Lake Grassmere earthquakes (Hamling etal.,2014). This northward rupture propagation resulted in
strong shaking in New Zealand's capital city, Wellington (Bradley etal.,2017; Kaiser etal.,2017). The reasons
for rupture terminating near Cape Campbell, despite the availability of faults straddling Cook Strait (Kearse
etal.,2018), remains unclear. Dynamic rupture models (Ando & Kaneko,2018; Ulrich et al., 2019) are able
to capture most of the major features of the Kaikōura rupture, including the absence of slip on the Hope Fault,
maximum co-seismic offset, and the termination near Cape Campbell. However, how these two models achieve
termination at Cape Campbell differs: Ando and Kaneko(2018) accounted for the termination by a c. 10° rotation
in the prevailing stress field, which is indicated by focal mechanism inversions using data from prior to the Cook

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
4 of 24
Strait earthquakes (Balfour etal.,2005; Townend etal.,2012). In contrast Ulrich etal. (2019) did not invoke a
stress rotation, and instead artificially reduce the stress on the Needles Fault. It is also possible that the Cook
Strait sequence invoked an as-yet unconstrained rotation in the stress field, resulting in the pre-Kaikōura stress
field differing from that used by Ando and Kaneko(2018). Neither modeling study included the more favorably
oriented faults that ruptured in the 2013 Cook Strait sequence.
In addition to the upper crustal faulting complexities, it remains unclear what role the underlying subduction
interface played in the Kaikōura earthquake (Hamling,2020). Lamb et al. (2018) suggested that the pattern
of strain accumulation on the interface can explain the diversity of crustal faulting, but it is not clear that the
interface played an active co-seismic role. Different models and data suggest differing contributions from the
subduction interface to the co-seismic moment budget of the Kaikōura earthquake. Generally, models derived
from regional data (e.g., Hamling etal.,2017; Gusman etal.,2018; Holden etal.,2017) require negligible seismic
moment on the underlying interface. In contrast, studies using teleseismic data tend to favor more slip occurring
on the subduction interface (e.g., T. Wang etal.,2018; Bai etal.,2017). Whether the subduction interface beneath
the northern South Island can slip seismically is fundamentally important to understanding seismic hazard in this
populous region of New Zealand (Wallace etal.,2018).
1.2. Post-Seismic Response
Afterslip inferred using geodetic data from the Kaikōura fault system for the months following the earthquake
shows significant afterslip on the faults known to have ruptured (Mouslopoulou etal.,2019; Wallace etal.,2018)
accompanied by afterslip or triggered slow-slip on the underlying subduction interface (Mouslopoulou etal.,2019;
Wallace etal.,2017; Yu etal.,2020) and triggered slow-slip in other regions of the Hikurangi margin (Wallace
etal.,2017). However, these models have used a relatively simple model of crustal faulting that does not capture
the spatial extent of aftershocks, in part due to a lack of a dense, high-precision aftershock catalog.
Romanet and Ide(2019) observed tremor occurring prior to the Kaikōura earthquake, near the zone of mapped
subduction interface afterslip, and suggested that the tremor may be related to interface slip.However, it is also
possible that the tremor locates on the downdip extent of faults in the Marlborough Fault Zone. Further work
is underway to better constrain these observations. Few aftershocks have yet been reliably linked to slip on the
subduction interface (Lanza etal.,2019).
The Kaikōura earthquake generated a significant and ongoing aftershock sequence (Kaiser etal.,2017) and trig-
gered earthquakes throughout New Zealand (Peng etal.,2018; Yao etal.,2021). However, it was relatively unpro-
ductive compared to average statistics for it's magnitude (Chamberlain etal.,2020; Christophersen etal.,2017)
resulting in an over-estimation of aftershock rates early in the sequence when average aftershock behavior was
used in forecasting (www.geonet.org.nz/earthquake/forecast/kaikoura, last accessed 22/01/2021). This relatively
low-productivity aftershock sequence is in contrast to the similarly complex Ridgecrest earthquake, which was
highly productive (Liu etal.,2019). Liu etal.(2019) suggested that the complexity of the Ridgecrest earthquake
may have promoted productivity due to strong stress concentrations around fault step-overs. However, that expla-
nation does not explain why the Kaikōura earthquake was relatively unproductive despite the involvement of
significant stepovers and presumably associated stress concentrations.
1.3. Unresolved Questions
Most models of co- and post-seismic slip around the Kaikōura earthquake have used multi-fault models of fault
ruptures, but these models have generally restricted the available faults to those with significant surface rupture,
or simplifications thereof. The only study that we are aware of that used aftershocks to better define the rupture
geometry focused on a small number of moment tensor solutions fixed at epicenters computed by GeoNet (Cesca
etal.,2017). We demonstrate in this paper that these GeoNet locations are poorly constrained due to the use of
the IASP91 (Kennett & Engdahl,1991) 1D velocity model (as also found by Lanza etal.,2019; Yao etal.,2021),
rendering them too inaccurate for use in defining fault structures.
Previous analysis of Kaikōura aftershocks (Lanza etal.,2019) has demonstrated the diffuse nature of aftershocks
around the step-over and Cape Campbell regions, which suggests slip occurred on additional crustal faults. In this
paper we expand on this aftershock catalog to explore the diversity of faulting around the faults that ruptured in

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
5 of 24
the Kaikōura earthquake, with the goal of shedding light on the pre-, co- and post-seismic faulting processes. We
particularly focus on several fundamental aspects of the rupture that remain unresolved:
1. Rupture Initiation (Section4.1): Where and how did the Kaikōura earthquake nucleate and were there observ-
able precursory signals?
2. Rupture Pathway (Section4.2): What was the likely rupture pathway between the southern fault system and
the high-slip Kekerengu fault and how was this step-over accommodated kinematically?
3. Subduction Interface (Section4.3): What was the seismogenic role of the subduction interface beneath the
known crustal fault ruptures of the Kaikōura earthquake?
4. Termination (Section4.4): Why did the rupture terminate at Cape Campbell and what was the significance of
the previous 2013


6.6 Cook Strait and Lake Grassmere earthquakes on this termination?
5. Post-seismic (Section4.5): How did such co-seismic complexity affect post-seismic afterslip?
2. Data and Methods
To obtain a more detailed picture of the fault geometry at depth, and the pre- and post-seismic evolution of
fault slip, we conducted a matched-filter search to generate a more complete representation of the seismicity.
We analyzed
𝐴
10years of continuous data using earthquakes that occurred on the faults that ruptured co- and
post-seismically in the Kaikōura earthquake as template events.
We used the catalog of 2,654 aftershocks and the mainshock picked and located by Lanza etal.(2019) as template
events to provide a methodologically consistent set of phase-picks. This catalog includes every event of

≥3
cataloged by GeoNet that occurred between November 13 and May 13, 2017 (UTC) in a rectangular region
between latitudes −43.00°and −40.80°and longitudes 172.75°and 175.20°, apart from 110 earthquakes that had
poorly constrained depths. We previously attempted to use the GeoNet catalog directly to construct templates but
found that the phase pick-quality was too variable, and the paucity of S-picks hindered our detection capability:
the resulting catalog contained excessive false detections. The Lanza etal.(2019) catalog contains the dominant,
moderate-to-large magnitude seismicity recorded in the seven months following the Kaikōura mainshock.
We constructed templates using data from 21 GeoNet broadband and short-period seismographs (Figure1). We
excluded strong-motion instruments from our analysis due to their variable timing quality (S. Bannister pers.
comm.). Note that these stations were included in the analysis of Lanza etal. (2019) and may have degraded
location quality in this prior work. We did not include temporary stations (e.g., from the STREWN network, as
analyzed by Lanza etal.(2019)) in our detection effort to exclude bias in detections arising from variations in
network geometry and station density.
Templates were made using EQcorrscan (Chamberlain etal.,2018). Continuous day-long data were detrended,
resampled in the frequency domain to 30.0Hz to reduce computational load, filtered using a 4th-order Butterworth
bandpass filter between 1.5 and 12Hz, and trimmed to 4s length around the P and S phase-picks on the vertical
and horizontal channels respectively. We tested a range of filters and template lengths and found that using a
higher low-cut frequency resulted in additional false detections likely related to correlations with high-frequency
noise, whereas using a lower low-cut frequency resulted in a degradation of correlations with true detections and
an increase in background (e.g., noise) correlation sums. Increasing the length of templates resulted in excessive
phase-overlap and compromised our ability to conduct later phase-picking analysis of detections. We removed
channels with a signal-to-noise ratio less than four, where we computed signal-to-noise ratio using the ratio of the
maximum amplitude in the template to the root-mean-squared amplitude of 100s of pre-template noise. Finally
we removed templates containing data from fewer than five stations, leaving a set of 2,584 templates.
We computed detections between January 1, 2009 and January 1, 2020 using the EQcorrscan package
(Chamberlain etal.,2018) which computes the network-wide stack of the normalized cross-correlation between
template waveforms and continuous data across multiple channels. We used the efficient FFTW (Fastest Fourier
Transform in the West, Frigo and Johnson(1998)) backend that implements the chunked-correlation algorithm of
Senobari etal.(2019), and the FMF (Fast-Matched-Filter, Beaucé etal.(2018)) GPU-based routine when a GPU
was available. Note that in compiling this catalog we implemented full normalization in the FMF code to ensure
compatibility with other correlations (Full-normalization in FMF implemented in pull request 38: github.com/
beridel/fast_matched_filter/pull/38,lastaccessedJuly292021).

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
6 of 24
Detections were made when the summed correlations exceeded 10
×
the median absolute deviation of the day-long
stack of correlations, and had at least an average normalized correlation above 0.15. To cope with degraded corre-
lations at the end of correlation epochs (in this case days) due to the delay-and-stack approach taken to compute
the summed correlations, we overlapped each day of correlation by the maximum moveout in the templates.
Detections from individual templates were required to be at least 4s apart. To remove duplicate detections (e.g.,
detections of the same event by different templates), we retained only the detections with the highest average
correlation if multiple detections occurred within 1s of each other.
To enable location of the detected events and further remove false detections we computed cross-correlation
derived phase-picks, following the methodology outlined by Warren-Smith etal.(2017). For each detection, the
relevant channel of the template and continuous data were correlated in a short window of ±0.5s around the
assumed pick-time based on a time-shifted version of the template phase-pick. A pick was made at the maximum
of this 1 s-long correlogram, if the maximum normalized correlation exceeded 0.4. Following this step, detec-
tions with picks on fewer than five stations were removed. This provided a catalog of 33,343 events comprising
899,460 phase-picks. In this picking step we incorporated the four temporary STREWN stations around Cape
Campbell, and GeoNet station CRSZ, deployed after the Kaikōura earthquake, to enhance our locations without
biasing our detections.
Because most of our detections were made during the active aftershock sequence of the Kaikōura earthquake,
some of the correlation picks we made were associated with the wrong event due to overlapping events from
different parts of the aftershock region. To combat this we undertook an additional quality-control step in which,
for each event, we located the event using HYPOCENTER (Lienert & Havskov,1995) and the 1D velocity model
of Okada etal.(2019). If the root-mean-squared (RMS) travel-time residual of the location exceeded 1s the pick
with the highest residual was removed and the event located again. We repeated this process until either the RMS
fell below 1s, or picks from fewer than five stations remained. If the events RMS did not drop below 1s with
five or more stations, the event was discarded. This removed 30 events leaving us a total of 33,328 events and
896,727 phase picks.
We located the detected earthquakes using the NonLinLoc software of Lomax et al. (2000) and the New
Zealand-wide 3D (NZ3D) velocity model of Eberhart-Phillips et al. (2017), version 2.2, which includes the
updated tomography around the Cook Strait region conducted by Henrys etal.(2020). We note that the issues
encountered by Lanza etal.(2019) in using NonLinLoc were rectified here by changing a flag in the NonLinLoc
Grid2Time3D source-code. We also tested using SIMUL2014 (Eberhart-Phillips & Bannister,2015) and found
that the fit to the data was degraded compared to our NonLinLoc locations. We suspect that this reduced quality
is because our events frequently contain S-picks without a corresponding P-pick, which SIMUL2014 cannot
use. This is because S-phases usually correlate better than P-phases due to their high amplitudes. We were able
to locate all events, but only 32,939 events are considered here because 389 occurred outside the study region
(Figure2).
Following this location step, we made automatic amplitude picks for all events and used these to compute local
magnitudes. We used the EQcorrscan (Chamberlain etal.,2018) amplitude-picking routines which picks half the
maximum peak-to-trough amplitude on a filtered, Wood-Anderson-simulated trace and corrects for the applied
filter. Comparison of these automatic picks with GeoNet amplitude picks for similar events (both those within
the template set and not in the template set) shows good agreement. We then computed local magnitudes by
inverting for a local magnitude scale that maps to moment magnitude, following the methodology of (Michailos
etal.,2019), taken from the moment tensor catalog maintained by GeoNet (https://github.com/GeoNet/data/tree/
main/moment-tensor,lastaccessedJuly292021) and based on the work of Ristau(2013).
We subsequently undertook relative relocation of all earthquakes using the GrowClust software (Trugman &
Shearer,2017) and HypoDD (version 2.1b) (Waldhauser & Ellsworth,2000). For GrowClust we used an average
1D velocity model extracted from the NZ3D velocity model (between 72–110km in X and
−
100–80km in Y in
the coordinate system of Eberhart-Phillips and Bannister(2015), TableS1) used for initial location. For HypoDD
we used the NZ3D model version 2.2 (Eberhart-Phillips & Bannister,2015; Henrys etal.,2020). We found little
difference between the two location methods, and so report the GrowClust locations here because they provide
robust, bootstrapped uncertainties (Trugman & Shearer,2017). We were able to relocate 27,431 earthquakes in
total.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
7 of 24
Finally, we computed first-motion-derived focal mechanism solutions for template events. To compute template
focal mechanisms we undertook manual polarity determination of the automatically determined P arrivals from
Lanza etal.(2019). We included stations from the STREWN network, and strong-motion stations in the GeoNet
network (station locations are plotted in FigureS8), but note that we did not use the timing of these phase arrivals
in our location calculations. We then inverted for the best-fitting focal mechanisms of all template events with
polarity picks at more than 8 stations (n=1,754) using the Bayesian algorithm developed by Walsh etal.(2009).
We used our NonLinLoc derived location estimates and uncertainties to compute takeoff angle and azimuth
posterior density functions.
Figure 2. Earthquakes on and around the faults (red lines) that ruptured in the Kaikōura earthquake plotted as circles colored by depth. Earthquakes deeper than 20km
are plotted in green. Dashed contours mark the depth to the modeled subduction interface (C. A. Williams etal.,2013). The dashed cyan line, labeled A–A′ is the
cross-section line shown in Figure7. Dashed dark blue boxes mark the bounds of the relevant figures. The gold star marks the mainshock hypocenter computed here.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
8 of 24
3. Results
We detected and located 33,328 earthquakes that occurred between January 1, 2009 and January 1, 2020 asso-
ciated with the regions active during the aftershock sequence of the 2016 Kaikōura M7.8 earthquake. Of these
earthquakes, we were able to compute precise relative relocations for a suite of 27,431 earthquakes (Figure2).
Our NonLinLoc locations have median 68% confidence uncertainties of between 1.8km and 3.0km (minimum
and maximum confidence ellipsoid lengths) and 2.8km in depth (FigureS6). Our GrowClust relocations have
median relative uncertainties of 0.2km in horizontal and depth directions.
As found by Lanza etal.(2019), but not by GeoNet, our hypocenter location for the Kaikōura mainshock (latitude
−
42.624, longitude 172.989, depth: 12.5km) lies almost directly beneath the Humps Fault, about 8.2km NNW
from the GeoNet location (beyond the bounds of uncertainties of either location) and c. 2km north of the loca-
tion obtained by Nicol etal.(2018). We were not able to relocate the mainshock hypocenter (using Growclust or
HypoDD) due to the complexity and clipping of the waveforms and resulting low correlations with other events.
This mis-location by GeoNet is likely due to the use of an inappropriate velocity model (ISAP91: www.geonet.
org.nz/earthquake/technical/2016p858000, last accessed September 7 2020). We discuss the variation in hypo-
center location further in Section4.1.
We obtain magnitudes ranging from 0.2–6.3 (Figure3). We note that the maximum magnitude of 6.3 was computed
for the


7.8 mainshock, which is beyond the range at which we would expect reliable amplitude-based local
magnitudes (see Figure S5). The largest aftershock magnitude we calculated is


5.9 30minutes after the
mainshock, for which GeoNet provide a magnitude of


6.2. In general our local magnitude scale gives
lower magnitudes than GeoNet at high magnitudes (Figure S5). We were unable to calculate magnitudes for
50 earthquakes due to insufficient amplitude picks of sufficient quality. The completeness of our catalog is
strongly variable in time: as noted by Hainzl(2016), during periods of high-rate seismicity the magnitude of
Figure 3. Upper panel: Local magnitudes for all earthquakes in our catalog (blue) and magnitude of completeness computed by goodness-of-fit (Wiemer etal.,2000)
(red). Magnitude of completeness was computed using a sliding window of 1,000 events. Magnitude of completeness is only shown when at-least 300 magnitudes
were above the best fitting completeness, and the fit was above 98%. Lower panel: Earthquakes projected onto the A–A′ cross-section (Figure2), and plotted against
origin-time. Earthquakes are colored by depth. Earthquakes deeper than 20km are plotted in green and the gray ellipse outlines the deep normal-faulting sequence
discussed in the text. The Lake Grassmere, Cook Strait and Kaikōura earthquakes are plotted as gold stars. Right panels show zoomed in views of the two weeks
following the Kaikōura mainshock, marked as vertical dashed black lines in the left panels.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
9 of 24
completeness increases, and we observe this after the Kaikōura mainshock. Before and within a few months after
the mainshock, our magnitude of completeness is around


1.2, however in the hours after the mainshock the
completeness becomes as elevated as


3.8 (Figure3). One of the main causes of elevated completeness, despite
the ability of the matched-filter method to detect earthquakes with overlapping waveforms, is the restriction in our
workflow to only detect events separated by at least 1s.
The vast majority of earthquakes in our catalog are aftershocks of the Kaikōura earthquake (30,652 events, or
92%, occurred after the mainshock). The earliest aftershock we detect occurred 2minutes and 48s after the main-
shock origin time, approximately 45–65s after the completion of the mainshock rupture (Holden etal.,2017).
However our catalog also includes aftershocks of the Cook Strait earthquakes, with 2,326 earthquakes between
the start of the Cook Strait sequence on the 18th of July 2013 and the Kaikōura mainshock. Some events in our
catalog appear to be associated with failure within the subducted plate. The sequence of earthquakes visible
in Figure3 at c. 125 km along the section occur at c. 25km depth and have focal mechanisms consistent with
normal-faulting in the subducted slab. Interestingly this family of earthquakes culminated in a sequence of eight
earthquakes in the seven days prior to the Kaikōura mainshock. We also detect limited earthquakes associated
with slip on the subduction interface made by templates representing likely interface events reported by Lanza
etal.(2019) near Cape Campbell. Most (28,768 or 86% or absolute locations and of 24,568 or 90% relative relo-
cations) of our earthquakes are found to have been shallower than 15km.
4. Discussion
This updated and expanded catalog of earthquakes on and surrounding the faults that ruptured in the Kaikōura
earthquake serves as the basis to re-evaluate some of the outstanding questions regarding this complex earth-
quake. Here we discuss the key questions outlined previously and highlight some key fault structures that have
previously been poorly resolved or unknown.
4.1. Rupture Initiation
Multiple hypocenter locations for the Kaikōura earthquake are now available and, as demonstrated by Nicol
etal.(2018), there is some inconsistency between them. In our locations we find that the mainshock hypocenter
locates almost directly beneath the surface trace of the Humps Fault, at a depth of 12.5±5.8km (Figure4). The
first-motion-derived focal mechanism of the mainshock that we construct here (strike/dip/rake of 250°/78°/−174°)
is consistent with dextral slip on a steeply dipping plane similar to the strike of the Humps Fault. A Gaussian fit to
the NonLinLoc uncertainties at the
1
level provides a horizontal uncertainty ellipse oriented at 96° with a maxi-
mum length of 2.3km and minimum length of 1.8km. Our location is slightly different (but within uncertainty)
from that of the previous solution of Lanza etal.(2019), whose phase picks we use here, and notably different
from the Geonet location that does not place the hypocenter on the Humps Fault. The GeoNet hypocenter could
indicate that an initial rupture on a separate fault to the south occurred, which subsequently triggered slip on the
Humps Fault as suggested by Ando and Kaneko(2018) to explain some of the mismatch in the initial rupture
speed between their model and observations. However, we are confident that the rupture did in fact nucleate on
the Humps Fault, and discuss possible causes of the discrepancies in locations below.
In this work we have not used picks on the strong-motion sites with known timing problems. We also use an
updated velocity model, and a different location method compared to Lanza etal.(2019). When we use the same
location method (using the software SIMUL) and/or use the same velocity model as Lanza etal.(2019), we obtain
a similar result to our preferred solution, suggesting that the main source of error in the previous location of Lanza
etal.(2019) was from the inclusion of picks from sites with problematic timing.
The GeoNet preferred location for the mainshock hypocenter (at the time of writing this, April 24, 2021, was
at −42.693°N, 173.022°E and 15.11km depth) lies 8.2 km to the south of our location, beyond the combined
uncertainties in our location and the quoted horizontal uncertainty in the GeoNet location (2.3km in latitude and
3.4km in longitude). The GeoNet solution is computed using the IASP91 (Kennett & Engdahl,1991) global 1D
velocity model and the LOCSAT location program (Bratt & Nagy,1991). When we locate the mainshock using
the GeoNet pick times in NonLinLoc using the NZ3D 2.2 velocity model used here we obtain a similar location
to our location (within uncertainty). We suggest that the use of the global 1D velocity model is inappropriate for
accurate location of crustal seismicity in New Zealand, and results in incorrect locations and under estimated

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
10 of 24
Figure 4. Nucleation region of the Kaikōura earthquake. Upper panel: map of relocated earthquakes (circles colored by
depth and scaled by magnitude) and focal mechanisms of template events, colored by depth. Earthquakes deeper than 20km
are plotted in green. Mainshock location is marked by a star: note that this is an absolute location rather than a relocation
for reasons explained in the text. The first-motion derived focal mechanism of the mainshock is shown in red. Alternative
mainshock locations are plotted as blue stars and labeled as Lanza, Nicol, GeoNet and USGS for the Lanza etal.(2019),
Nicol etal.(2018) GeoNet and USGS solutions respectively. Mapped surface ruptures are plotted as red lines, and other
faults of the NZ active faults database are plotted in black. Dashed black contours mark the modeled subduction interface
from C. A. Williams etal.(2013). The dashed cyan line shows the cross-section line plotted in the lower panel. Lower panel:
Cross-section (SW to NE) of relocated hypocenters projected onto the cyan line in the upper panel. Earthquakes are colored
by time since 30s prior to mainshock, note that the colorscale is logarithmic. Earthquakes are scaled by magnitude. The star
marks the absolute location of the mainshock.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
11 of 24
location uncertainties, as also shown in central North Island by Illsley-Kemp etal. (2021). Similar issues are
likely to apply to other location solutions for the Kaikōura mainshock that do not use an appropriate velocity
model.
The location computed by Nicol etal.(2018) is within the uncertainty of our location, and was computed using
a similar method to that used here. However, the aftershock relocations computed by Nicol etal. (2018) use
GeoNet locations as starting locations, which are inaccurate due to the use of the IASP91 velocity model. As
such, relocation from these inaccurate starting locations is the likely cause of difference between the relocations
of Nicol etal.(2018) and those presented here, which here delineate a nearly vertical structure consistent with
our mainshock focal mechanism. The south-dipping lineation extending through the subduction interface shown
by Nicol etal.(2018) is not visible in our relocations, probably due to more robust starting locations used here.
We note that a foreshock c. 7s prior to the mainshock (FigureS1) may also have contributed to inaccuracies in
mainshock location: if picks were made on the much smaller foreshock P-phases for the four GeoNet stations
that they are visible on then these arrival times would bias the location. This foreshock is located close to the
mainshock, but the mainshock obscures the S-phase on most stations and the P-phase is only visible on four
stations due to the size of the foreshock, and the resulting location we obtain has high uncertainties. We did not
detect this foreshock with our matched-filter detector due to the poor signal on most stations, and it is therefore
not included in our catalog.
In summary, our more accurate mainshock location and focal mechanism confirm that the Kaikōura earthquake
most likely nucleated as a dextral strike-slip rupture of the Humps Fault, and confirm that the Humps Fault here
is steeply dipping (c. 80°) to the North. This suggests that off-fault triggering did not play a strong role in the
nucleation of the Kaikōura earthquake, and other factors must be the cause of the early long-duration release of
seismic energy. Ulrich et al. (2019) were able to reproduce the slow initial phase of the rupture through the
Humps-Hundalee system in their dynamic rupture simulation. Finally, it is worth noting that any seismic back-
projections that compute the location of high-frequency radiation sources relative to the mainshock may be biased
by the use of inaccurate hypocenters (e.g., Tan etal.,2019; D. Wang etal.,2018).
We do not observe precursory seismicity in our catalog aside from the foreshock approximately 7s prior to the
mainshock which we did not detect by matched-filter and is not included in our final catalog. This includes no
seismicity in the epicentral region following any of the 2010 Darfield earthquake, 2011 Christchurch earth-
quakes, or the 2013 Cook Strait sequence, which are likely to have induced dynamic stress changes in the epicen-
tral region of the Kaikōura earthquake. We attempted to run a focused matched-filter search using GeoNet data
and the 7s foreshock as a template, but this did not make any further reliable detections. We note that our catalog
is likely biased by being constructed using only aftershocks as templates, and the presence of at least one visible
foreshock should motivate further analysis of foreshock activity here.
4.2. Rupture Pathway
The Kaikōura earthquake involved substantial rupture (
𝐴
1.5 m surface slip) of at least 13 faults (Litchfield
etal., 2018). Initial observations suggested that large stepovers (up to 20km), particularly between the south-
ern faults (Humps-Hundalee system) and the high slip Kekerengu Fault, were present (Hamling etal., 2017;
Kaiser etal.,2017). Such large stepovers commonly correspond to rupture termination points (Harris etal.,1991;
Wesnousky,2006). More recently, additional faults, including the Point Keen or other offshore reverse faults, and/
or links between the Hundalee and Jordan Thrust/Upper Kowhai Faults (via the Leader and Whites Faults) have
been postulated to explain the rupture sequence (e.g., Ando & Kaneko,2018; Zinke etal.,2019). In particular,
the dynamic rupture model of Ando and Kaneko(2018) has rupture propagating from the Hundalee Fault to the
Upper Kowhai and Jordan Thrust Faults with limited slip on the linking Whites Fault (Figure6), and suggests that
this step-over was accommodated mostly by transient dynamic stresses or elastic waves. In contrast, the dynamic
rupture model of Ulrich etal.(2019) has rupture propagating from the Hundalee Fault onto the offshore reverse
faults before triggering slip on the Papatea Fault, which then caused rupture of the Jordan Thrust and Kekerengu
Faults.
Although we do not have direct co-seismic evidence in our catalog of the rupture pathway, our earthquake loca-
tions help to illuminate the structure of these linking faults at depth (Figure5). Two key faults emerge: (a) an
offshore, dominantly reverse, structure similar to the Point Keen Fault modeled by Ulrich etal.(2019); Hamling

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
12 of 24
Figure 5.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
13 of 24
et al. (2017) and (b) a previously unidentified strike-slip, near-vertical structure linking the Papatea-Jordan
Thrust-Kekerengu-Fidget junction to the inland, unruptured Clarence Fault. We herein refer to this second new
fault as the Snowgrass Creek Fault, named after a nearby stream. The Snowgrass Creek Fault strikes approx-
imately 140°, has a near vertical dip, and a surface length of approximately 12km. Note that this fault is not
associated with any reported surface rupture. There is also a continuous trend of earthquake locations spanning
the gap between the southern fault system and the Jordan Thrust, suggesting that either the offshore route, via
the offshore thrust system, or the onshore route, via the Whites Fault are viable options for rupture propagation.
Several key observations provide further constraints on the most likely rupture route for the Kaikōura earth-
quake, principally the occurrence of a small, localized tsunami (Gusman etal.,2018), and the inverted motion
of the Jordan Thrust, which hosted normal motion rather than the reverse motion, as would be expected from the
geological record (Howell etal.,2020; Van Dissen & Yeats,1991). We propose that these two factors, alongside
our observation that offshore thrust faulting spans the gap between the Hundalee Fault and the Papatea Fault,
require that the earthquake propagated via the offshore route (Figure6). In addition the observation of a tsunami
requires some co-seismic offshore deformation which would be provided by offshore thrust faulting (Gusman
etal.,2018), and the normal (inverted) sense of slip on the Jordan Thrust Fault can be explained by our preferred
model. This is in agreement with recent modeling studies by Ulrich etal.(2019) and Klinger etal.(2018).
In our preferred rupture scenario we suggest that the offshore thrust fault (or faults, here labeled as the Point Keen
Fault for consistency, despite the opposite sense of slip compared to the geologically recognised Point Keen Fault
(Litchfield etal.,2018)), the Papatea Fault, and extending into the newly discovered Snowgrass Creek Fault acted
as one thrust block with a sinistral north-western edge (Figure6). Within this thrust block, the normal motion
of the usually reverse Jordan Thrust Fault occurs as a consequence of the eastward motion of eastern side of the
block (normally the footwall). In other words, the coastal side of the Jordan Thrust is extended seawards relative
to the pinned inland side resulting in normal motion.
This scenario can also help to explain the high slip on the Papatea Fault. In this scenario, the Papatea Fault sits
at the corner between dominantly thrust motion offshore, to dominantly sinistral-normal oblique motion onshore
on the Snowgrass Creek Fault. Not only does this scenario provide additional fault length for the combined
Papatea-Snowgrass Creek-Point Keen Fault system, meaning that co-seismic displacements scale more consist-
ently with fault length, but also that the Papatea Fault acts in a similar style to a restraining bend, for exam-
ple, with large co-seismic strain exceeding the long-term accumulated elastic strain, which other authors have
suggested is insufficient to explain the slip amplitude on the Papatea Fault (e.g., Diederichs etal.,2019).
We use the same equations, converted to SI units, as R. M. Langridge etal.(2018), after Stirling etal.(2012),
namely:
= 2∕3log+ 4∕3log−1.82,
(1)
where

is fault width and

is fault length, both in meters, and

0
= 𝜇
(2)
where

0
is the seismic moment in N
⋅
m,

is the shear modulus, which Stirling etal.(2012) assume to be
3×10
10
Pa,

and

are as before, and

is the single-event displacement in meters.

0
is calculated using:
log0=9.05+1.5

.
(3)
Figure 5. Earthquake locations around the transition from southern/epicentral faults to the Kekerengu fault. Top: map view of relocated earthquakes plotted as circles
colored by depth and scaled by magnitude. Earthquakes deeper than 20km are plotted in green. Thrust focal mechanisms (45°
𝐴
rake
𝐴
135°) for template events are also
plotted, colored by depth. Active faults are plotted in black, and faults with known surface rupture during the Kaikōura earthquake are plotted in red. Black dashed
contours mark the depth to the interface model of C. A. Williams etal.(2013). The cyan dashed line marks the cross-section line shown in the lower panel. The green
solid line marks the inferred location of the newly identified Snowgrass Creek fault (labeled). Note that the surface dip of the Clarence Fault is c. 70°NW (Rattenbury
& Isaac,2012), and the Snowgrass Creek fault appears to terminate at the Clarence Fault at depth. Bottom: Cross-section perpendicular to the dominant strike of
reverse focal mechanisms. Earthquakes within 7.5km of the cross-section are projected onto the line. Solid straight lines mark the locations and dips of cross-section
intersecting faults from Litchfield etal.(2018). The solid curved line at depth marks the subduction interface model of C. A. Williams etal.(2013). Note that the broad
cluster of earthquakes at the down-dip end of the Upper Kowhai Fault is likely associated with projecting earthquakes on a fault striking obliquely to the cross-section.
Similarly, our preferred arcuate geometry of offshore thrusting, and variable dip provides an explanation for the broad region of earthquakes below the inferred Point
Keen Fault.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
14 of 24
This way, we are able to estimate single-event displacements for various fault combinations. We deduce that R.
M. Langridge etal.(2018) adopted a fault width of 18.5km based on the magnitude they compute. Using this
fault width and a combination of the Papatea and Snowgrass Creek faults (which adds approximately 15km to
the length when incorporating the dip of the Clarence Fault and hence additional length of the Snowgrass Creek
Fault at depth) we find a single-event displacement of 2.3m. Incorporating the Point Keen Fault in our preferred
geometry results in an 83km total length and average displacement of 5.8m. Finally, including the section of the
Hundalee Fault between the coast and the Stone Jug Fault increases the length to 93km and slip to 6.5m. The
average net slip on the Papatea Fault was measured to be 6.4±0.2 (R. M. Langridge etal.,2018), reinforcing our
proposed combined fault system explanation.
The existence of the Snowgrass Creek Fault also helps to explain the drop in slip across the Kekerengu-Jordan
Thrust junction, despite the similar strikes of these two faults. A simple model of this junction is that of a quadru-
ple junction between the Jordan Thrust, Papatea, Kekerengu and Snowgrass Creek Faults (discounting the Fidget
Figure 6. Schematic, not-to scale cartoon illustrating links between faults in the stepover region between the southern faults and the high slip Kekerengu Fault, and
how the Papatea Fault may operate as a restraining pop-up. Faults in gray represent major through-going structures of the Marlborough Fault Zone (the Hope and
Clarence Faults) which did not have significant co-seismic rupture, but which may have localized slip at depth near fault junctions as indicated by darker gray shading.
Note that the Hope Fault is truncated for visibility, but extends further offshore than plotted. Colored, outlined arrows on faults show sense of co-seismic motion,
approximately scaled by size to show relative slip magnitudes between different faults. The thin red line with arrows shows preferred inland rupture route of Ando and
Kaneko(2018) via the Whites Fault (inferred, dashed line) and triggered slip on the Papatea (also denoted by dashed line). The thin black line with arrows shows our
preferred offshore rupture route, with bi-lateral rupture originating from the Papatea-Kekerengu-Snowgrass Creek-Jordan Thrust junction. Inset shows simplified map
view of faults, colored as in main plot, illustrating how the Papatea-Point Keen connection forms an offshore compressional bend with anticipated vertical motion.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
15 of 24
Fault that has mapped surface rupture away from the junction, but not nearby (Litchfield etal.,2018)). By aver-
aging the InSAR derived coseismic displacement field (Hamling,2020) in blocks around the fault system (see
FiguresS2 and S3) we estimate the strike-parallel and perpendicular components of motion on the Snowgrass
Creek to be 1.3m sinistral and 3.4m of extension. The resulting sinistral transtensional motion is consistent with
the dominant aftershock focal mechanisms (FigureS8). The strong change in the InSAR-derived North-South
displacement field aligns with the strike of the Snowgrass Creek Fault constrained by our earthquake locations.
Including the Snowgrass Creek Fault as a separation between the western side of the Kekerengu Fault and the
western (inland) side of the Jordan Thrust reduces the required dextral motion from 6.2m on the Kekerengu to
3.3m on the Jordan Thrust. The difference in these estimated offsets corresponds well with the difference in
dextral offsets measured by Kearse etal.(2018), which rise from c. 1–8m on the Jordan Thrust, and are gener-
ally between 10–12m on the Kekerengu Fault (see FiguresS2 and S3). Without the Snowgrass Creek Fault,
block offsets require 5.1 and 5.0m of dextral offset on the Kekerengu and Jordan Thrust Faults, which does not
allow for change in the change in dextral offset observed. Our estimates do not capture the total slip on the faults
because we use spatially averaged displacements in off-fault blocks to capture the general kinematics. Never-
theless, the change in slip between the Kekerengu and Jordan Thrust cannot be accommodated without some
additional deformation, and the Snowgrass Creek Fault provides a viable structure for this deformation.
We suggest, therefore, that the Kaikōura earthquake propagated from the Hundalee Fault onto the offshore
thrust system, which then activated the Papatea and Snowgrass Creek Faults, which in turn triggered slip on the
Kekerengu Fault. In this model, the role of the Jordan Thrust is minor, and the extension of aftershocks between
the Jordan Thrust to the Whites Fault is a consequence of the underlying thrust system. This scenario agrees
with the dynamic rupture simulation of Ulrich etal.(2019), but is at odds with that of Ando and Kaneko(2018)
whose model did not result in significant slip on the Papatea Fault. We note that both Ando and Kaneko(2018)
and Ulrich etal.(2019) have used a shallower dip on the offshore thrust system than the 45–60° dip found here,
which results in a reduced possible stress-drop in the model of Ando and Kaneko(2018), making it a less favora-
ble rupture pathway in their model.
The Snowgrass Creek Fault also appears to link with the Clarence Fault, a key component of the Marlborough
Fault system (Van Dissen & Nicol, 2009) that did not rupture in the Kaikōura earthquake. One of the earliest
aftershocks we detected, a


4.8 within nine minutes of the mainshock origin time, occurred at the junction of
the Snowgrass Creek and Clarence Faults, suggesting that the Clarence Fault may have been active early in the
aftershock sequence. That neither the Hope nor the Clarence Faults had significant co-seismic rupture despite
evident triggered aftershocks, remains an intriguing observation.
4.3. Subduction Interface
We observe no earthquakes consistent with slip on the subduction interface beneath the majority of the upper-plate
faults (Figure7). The few earthquakes observed close to the subduction interface (e.g., at 23km depth in Figure5)
show normal-faulting mechanisms, consistent with extension in the down-going plate, and were active prior to
the Kaikōura earthquake. Some earthquakes consistent with subduction interface slip occur beneath the Cape
Campbell region, as shown by Lanza etal.(2019) and here (Figure8), but not all show mechanisms consistent
with interface slip here. It may be that the northern-tip of South Island is the point where the subduction interface
becomes seismically active, as proposed by Henrys etal.(2020).
When considering the significance of a lack of aftershocks in our catalog on the subduction interface it is impor-
tant to restate the limitations of matched-filter catalogs. Such catalogs by definition only contain earthquakes
similar to those in the template data set: if we do not have any subduction interface earthquakes in our template
set then we should not be surprised to see no subduction related events in the final catalog. However, our template
catalog is composed of all earthquakes in the GeoNet catalog between November 13, 2016 and May 12, 2017
larger than


3 (Lanza etal.,2019). As such, any missing seismicity should be of small magnitude and therefore
likely contributed minimally to the total (post-seismic) moment release.
Our data set provides no direct constraints on whether the subduction interface slipped co-seismically, but by
more accurately mapping crustal seismicity we are able to robustly demonstrate the existence of offshore thrust
faulting south of the Kekerengu Fault. Such offshore faulting has been previously used in models that recreate
co-seismic data without the need for significant slip on a subduction source (e.g., Clark etal., 2017; Gusman

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
16 of 24
etal., 2018). Incorporating more realistic models of crustal faulting at depth, derived from our catalog, may
provide greater constraints on the co-seismic role of the subduction interface.
The lack of aftershocks on the subduction interface does not preclude afterslip on the interface because this after-
slip could be aseismic. However, it seems unlikely that if the subduction interface is aseismic in the post-seismic
period that it would have contributed significantly to the co-seismic seismic wavefield. The published models of
post-seismic slip have used simple models of crustal faulting (for instance Wallace etal.(2018) use four crustal
fault sources attempting to simulate the Humps, Kekerengu/Jordan Thrust, Needles and an offshore thrust fault).
The simplicity in crustal faults may lead to inaccurate mapping of slip onto the underlying subduction interface.
For example, in the Cape Campbell area, at the northern tip of South Island, strong co- and post-seismic uplift
occurred (Wallace etal., 2018). This uplift includes a large short-wavelength component: the uplift at GNSS
station CMBL is more than triple that at station WITH (Figures1 and9), within a few tens of kilometers. WITH
and CMBL are separated by the faults that ruptured in the Lake Grassmere earthquake (Hamling etal.,2014),
which were re-invigorated during the Kaikōura aftershock sequence (Figure8). These faults are more shallowly
dipping than the Needles Fault, and have a significant reverse component (Hamling etal.,2014), but the pattern
of uplift observed in the Kaikōura earthquake is the reverse of that in the Lake Grassmere earthquake (Hamling
etal.,2014). This suggests that either the Lake Grassmere and Cook Strait Faults were reactivated with a normal
sense of motion (but we do not observe normal focal mechanisms in this region), or other reverse faults dipping
to the East, such as the London Hills Fault, were responsible for this short-wavelength uplift. No faults between
WITH and CMBL with this sense of motion were included in the afterslip model of Wallace etal.(2018). Inclu-
sion of these faults, which have a strong aftershock signature (Figure8) may reduce the need for interface slip
beneath Cape Campbell.
Figure 7. Along-strike earthquake distribution, along line A–A′ shown on Figure2. Top: Aftershock moment density (blue) computed in 1km bins perpendicular
to the cross-section, and slip density derived by Ulrich etal.(2019) (red). Note that the projection of all slip in this 3D fault geometry onto a single plane results in
the summation of slip across multiple fault strands. The peak in slip around 65km along the section occurs at the corner between the Stone Jug and Hundalee faults
and is likely unrealistic, and in part due to the projection of slip on a single plane. Bottom: Aftershock locations colored by time since 30s prior to the the Kaikōura
mainshock. Note that the color-scale is logarithmic. The location of the epicenter of the mainshock is shown by a gold star, and the depth of the interface from C. A.
Williams etal.(2013) is shown as a solid line. The purple dashed contour marks the Qs=200 contour from the NZW3D 2.2 model (Henrys etal.,2020).

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
17 of 24
Figure 8. Earthquake locations near the termination of the Kaikōura earthquake. Top: Map view of earthquake relocations
colored by depth. Earthquakes deeper than 20km are colored green. Earthquakes with black outlines mark events that
occurred prior to the Kaikōura mainshock, including events triggered by the Cook Strait and Lake Grassmere earthquakes
in 2013, which are plotted as gold stars. Active faults without surface rupture from the Kaikōura earthquake are plotted
as black lines, and those with surface rupture are plotted in red. Black dashed contours show the model of the Hikurangi
subduction interface from C. A. Williams etal.(2013). The teal oval outlines the events close to the subduction interface
that have mechanisms possibly related to slip on the interface as identified by Lanza etal.(2019). The dashed cyan line
marks the cross-section plotted below, and the width of the swath (10km) is shown at each end of the cross-section line.
Bottom: Cross-section of earthquake locations colored by time after 30s prior to the Kaikōura mainshock within 5km of the
cross-section line. The subduction interface is shown as a curved solid black line, and the projections of the Needles (surface
dip of 70°, (Litchfield etal.,2018)) and London Hills (surface dip of 70°(R. Langridge etal.,2016)) faults to 10km depth are
shown.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
18 of 24
Figure 9. GPS time-series and cumulative aftershock density for regions around the Kaikōura afterslip region. Regions are ordered north (top) to south. GPS
displacements for sites CMBL, WITH, KAIK and MRBL have a long-term gradient removed (calculated between 2015/01/01 to 2016/11/1). Sites LOK1, GLOK,
MUL1 and LOOK have not had any gradient removed because they were not active prior to Kaikōura. Data from stations GLOK and LOOK have been shifted to have
matching displacements at the end of the recording periods of LOK1 and MUL1 respectively. Note that the overlap is imperfect, but provides a representative view of
displacement in the region. In general the evolution of the aftershock sequence matches the evolution of the displacement for these regions, however there are strong
differences across the regions highlighting different amounts of afterslip.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
19 of 24
Incorporating more realistic and complex crustal faulting is unlikely to completely remove the need for slip on
the underlying subduction interface: crustal faults are likely to help to explain short-wavelength geodetic features,
but not the long-wavelength features seen in both the post-seismic InSAR and GNSS data (Wallace etal.,2018).
Recent modeling work by Eberhart-Phillips etal.(2021), constrained by seismic attenuation modeling results,
shows that deformation between the subducted Pacific plate and overlying Australian plate is likely to be ductile
with no clear interface structure. In this scenario, ductile deformation rather than interface slip may be controlling
the long-wavelength post-seismic signature. Such ductile deformation would likely be aseismic, consistent with
both the geodetic signature and the lack of aftershocks.
4.4. Termination
The Kaikōura earthquake terminated near Cape Campbell, at the north-eastern tip of South Island. Surface
ruptures were mapped on the Needles (offshore, but without rupture of the nearby/adjoining Boo Boo Fault
(Kearse etal.,2018)), Marfells Beach, Cape Campbell Road and Lighthouse Faults (Litchfield etal.,2018). The
rupture terminated despite the existence of multiple other pre-existing mapped faults in the region. The Cape
Campbell region also hosted the 2013 Cook Strait earthquake sequence, including the


6.6 Cook Strait earth-
quake on July 21, 2013, and the subsequent


6.6 Lake Grassmere earthquake on August 16, 2013 (Hamling
etal.,2014). This region is also close to the modeled southern rupture extend of the

8 1855 Wairarapa earth-
quake (Darby & Beanland,1992; Rodgers & Little,2006).
Dynamic rupture simulations have been able to simulate arrest on the Needles Fault (Ando & Kaneko,2018;
Ulrich et al., 2019), either by invoking a small (10° clockwise) rotation in the regional stress field (Ando &
Kaneko,2018), or by enforcing reduced pre-stress on the Needles Fault while rotating the stress field in the oppo-
site direction (Ulrich etal.,2019). The two shallow (
𝐴
25km)


estimations from Townend etal.(2012) in the
region (their clusters 16 and 11) suggest a possible clockwise rotation as used by Ando and Kaneko(2018). The
counter-clockwise rotated cluster in Cook Strait (cluster 18) has a centroid at 42km depth and is likely related
to stresses associated with subduction interface beneath. We therefore favor a clockwise rotation to an


orientation of c. 110° which reduces the potential stress drop on the Needles Fault and leads to the spontaneous
termination in the model of Ando and Kaneko(2018). This rotation is also consistent with the earlier work of
Balfour etal.(2005).
Neither of the above-mentioned dynamic rupture models (Ando & Kaneko,2018; Ulrich etal.,2019) includes
slip on other faults around Cape Campbell, despite the mapped surface ruptures (Litchfield etal.,2018) and the
diffuse aftershocks mapped here and by Lanza etal.(2019). Importantly, the inferred rupture plane of the Cook
Strait earthquakes is rotated c. 9° clockwise of the average strike of the Needles Fault (Hamling etal.,2014),
resulting in a more favorable orientation for slip on these faults in the regional stress-field. Interestingly we see a
general paucity of earthquakes on the Needles Fault (Figure8) compared to faults directly beneath Cape Camp-
bell despite the co-seismic rupture of the Needles Fault. We suggest that this may be due to the unfavorable orien-
tation of this fault. We also favor a more steeply dipping (near-vertical) Needles Fault, with much of the reverse
component of deformation taken up by shallower dipping faults to the West.
Because the templates we use, despite having been constructed exclusively from aftershocks of the Kaikōura
earthquake, detect aftershocks of the Cook Strait sequence (but not the mainshocks), the Kaikōura aftershock
sequence must include re-rupture of favorably oriented faults that were active during the Cook Strait aftershock
sequence. Focal mechanisms of aftershocks in this region include multiple dextral-reverse mechanisms striking
c. 055°, similar to the Cook Strait mainshocks.
We consider two possibilities for the cause of the activation of the Cook Strait sequence fault(s) by the Kaikōura
earthquake: (a) the Kaikōura earthquake co-seismically ruptured the more favorably oriented “Cook Strait Fault”;
(b) seismicity on the “Cook Strait Fault” was triggered post-seismically. As computed by Ulrich etal.(2019), the
maximum Coulomb failure stress (
Δ
CFS) reduction on the Needles Fault due to the Cook Strait sequence is small
(c. 0.1MPa), and is strongly heterogeneous. However, the stress drops on the “Cook Strait Fault” itself due to the
Cook Strait and Lake Grassmere earthquakes are 1 and 3.5MPa respectively. We hypothesize that this resulted
in reduced pre-stress on the “Cook Strait Fault,” ensuring that the Kaikōura earthquake could not generate signif-
icant rupture through this more favorably oriented fault, either co-seismically or post-seismically. Changes in

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
20 of 24
frictional properties on the “Cook Strait Fault” may also act to inhibit rupture, but we have no direct observations
of the frictional properties, nor how they vary in time for these faults.
Our aftershock locations do not show clear evidence for a structural boundary within Cook Strait as the control
for rupture termination. Instead we observe a consistent migration of aftershocks away from the inferred rupture
termination point into Cook Strait (see Section4.5, Figures7 andS9). Nevertheless, the aftershocks do concen-
trate within the region of low Q (high seismic attenuation), as demonstrated by Henrys etal. (2020). Henrys
etal.(2020) suggested that the change in seismic properties in Cook Strait may be linked to changes in interface
coupling, upper-plate deformation and strain-accumulation, which may play a role in rupture termination. In
general the aftershocks are found to have occurred within regions of low Q, which may be indicative of regions
of higher fracturing or damage, more capable of hosting seismicity (Henrys etal.,2020).
We suggest that a combination of an unfavorably oriented Needles Fault, reduced pre-stress due to prior rupture
of other nearby faults, and the presence of diffuse faulting around Cape Campbell, served to terminate the rupture
near Cape Campbell.
4.5. Post-Seismic
The catalog we present here is dominated by aftershocks providing important information on deformation
processes following complex co-seismic slip. Spatially, several key features are apparent in the post-seismic
period (Figure7). First, the peak aftershock densities occur at the rupture termination point near Cape Campbell,
and in the step-over region between the southern and northern rupture domains. Strong aftershock activity near
rupture terminations where there are elevated stress concentrations is common (King etal., 1994), and we do
see many aftershocks surrounding the Needles Fault (Figure8): however, the majority of aftershocks around
Cape Campbell occur in a distributed region between the Needles Fault and the location of the 2013 Cook Strait
sequence. The patch of aftershocks around Cape Campbell expands in time, following a roughly log-time expan-
sion, and seems to expand bilaterally (Figures7 andS9).
As previously reported, there are very few aftershocks associated with the Papatea Fault and the highest-slip
patch of the Kekerengu Fault, which we interpret to be segments that experienced near-total stress-drop. The
high-slip patch of the Kekerengu Fault separates the two regions of high aftershock density and may provide a
limiting control to the aftershock sequence.
In the south, we see a continuation of aftershocks beyond the southern rupture termination point, and clustered
triggered off-fault seismicity. We also note that, although there are aftershocks on the Leader and surrounding
faults, we also see a continuous trend of aftershocks joining the Humps and Hundalee Faults, effectively cutting
off this block, and potentially accommodating block rotation as proposed by T. Wang etal.(2020).
Comparison of GNSS displacements with earthquake rates in regions surrounding the GNSS site shows that after-
shock rates are generally proportional to displacement rates (Figure9). The catalog presented here is sufficiently
detailed to map earthquakes to individual faults, but the published post-seismic slip models do not have suffi-
ciently detailed crustal fault resolution to directly compare aftershocks with afterslip.Because of the complexity
of the earthquake, GNSS displacement measured at a single site is likely to correspond to slip on multiple fault
sources, rendering direct comparison of geodetic data with seismicity non-unique. Nevertheless, despite the range
of faulting and co-seismic slip, it appears that aftershock distributions correlate well with geodetically determined
displacements, suggesting that aftershocks are driven by local afterslip (Frank etal.,2017; Perfettini etal.,2018).
5. Conclusions
The 2016 M 7.8 Kaikōura earthquake is widely regarded as one of the most complex earthquakes in recorded
history (Hamling,2020). Detailed mapping of seismicity around the faults that ruptured in the Kaikōura earth-
quake further emphasizes this complexity: at-least in the post-seismic period, multiple faults that did not have
surface rupture are activated including two of the high slip-rate and high hazard Marlborough Faults (the Clar-
ence and the Hope Faults). However, the additional faults observable through this mapping may also simplify
some of the kinematics of the rupture by providing additional structures to host variations in slip between nearby
fault segments.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
21 of 24
To address the original outstanding questions outlined in Section1.3, and as discussed in Sections4.1–4.5, our
conclusions are as follows:
1. The mainshock unequivocally nucleated on the Humps Fault. Previous scatter in published locations can be
attributed to inappropriate location methods or data quality issues which we have thoroughly addressed in
this study.
2. We do not observe any precursory activity in our catalog, but this is likely in-part a limitation of using the
aftershock-derived template set. We do observe one foreshock 7s prior to the mainshock, however the sparsity
of seismic stations limits our ability to investigate further.
3. Offshore thrust faulting illuminated by aftershocks suggests a physical connection between the Hundalee and
Papatea Faults, which may explain anomalously high slip on the Papatea Fault and provides a likely southern/
offshore rupture route.
4. The Snowgrass Creek-Papatea-Jordan Thurst-Kekerengu system acts as a quadruple junction providing a
means of distributing the drop in slip between the Kekerengu and Jordan Thrust Faults.
5. Both the Hope and Clarence Faults were active post-seismically and produced aftershocks, though these were
not laterally extensive, and occur near fault junctions or transitional zones.
6. We observe very few aftershocks on the subduction interface. A proportion of the afterslip previously mapped
onto the subduction interface may instead be accommodated by unmodelled upper crustal faults, such as
the previously unidentified Snowgrass Creek Fault, the Clarence Fault and diffuse faulting characterized by
abundant aftershocks near Cape Campbell. However crustal faults are unlikely to remove the need for deep
deformation to explain the long-wavelength signature in the geodetic data, but this deformation likely occurs
aseismically.
7. The rupture terminated near the epicenters of the Lake Grassmere and Cook Strait 2013 earthquakes, and
likely re-ruptured these faults. The Cook Strait and Lake Grassmere faults are more favorably oriented for
slip than the co-seismically ruptured Needles Fault, and we propose that the combination of unfavorable
orientation of the Needles together with reduced pre-stress on the Lake Grassmere and Cook Strait faults was
sufficient to cause the rupture to terminate here.
8. Aftershocks concentrate at step-overs in faulting, and at the rupture termination near Cape Campbell. The
patch of high co-seismic slip on the Kekerengu Fault has few aftershocks and potentially experienced near
total stress drop, and may separate patches of afterslip reducing aftershock productivity.
Considering all of the above, we infer that the Kaikōura earthquake nucleated without significant detectable
precursory seismicity on the Humps Fault before transitioning through the Leader/Stone Jug system and onto
the Hundalee Fault. The rupture then continued directly onto the offshore fault system characterized by reverse
slip, elsewhere called the Point Keen Fault. Slip then transitioned onto the Papatea Fault, likely by directly
linked faults at depth in a thrust block bounded by sinistral faulting on the Papatea and Snowgrass Creek Faults
(Figure6). Within this block, the Jordan Thrust Fault was reactivated in an extensional stress regime giving rise
to normal motion (in contrast to the long-term motion on this fault), and the difference in slip between the Jordan
Thrust and Kekerengu Faults is accommodated by buried slip on the previously unknown Snowgrass Creek Fault.
Slip then transitioned onto the Kekerengu Fault, which experienced near-total stress-drop in the high slip patch
identified by other authors (e.g., Kearse etal.,2018), and characterized here by a lack of aftershocks. The rupture
then propagated onto the Needles Fault and other faults around Cape Campbell that were previously ruptured in
the 2013 Cook Strait earthquakes. A combination of an unfavorable stress orientation on the Needles Fault and
reduced pre-stress due to recent slip on the Cook Strait and Lake Grassmere faults resulted in the termination of
the Kaikōura earthquake at Cape Campbell. We see no evidence for seismic slip on an underlying subduction
interface, apart from a small cluster of interface related seismicity near Cape Campbell. We therefore suggest that
the boundary between the overriding Australian plate and subduction Pacific plate may be ductile beneath much
of the Kaikōura earthquake fault system as suggested by Eberhart-Phillips etal.(2021).
Data Availability Statement
All waveform data for GeoNet stations were downloaded from GeoNet via their FDSN client (last accessed April 20,
2021). All data from the STREWN network (code Z1) were downloaded from the IRIS FDSN Client (last accessed
6 June 2021). The catalog generated here is available at https://zenodo.org/record/6763130#.YrpLdDVBzd4 (last

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
22 of 24
accessed April 24, 2021) in QUAKEML and CSV format. All code used to generate this catalog is open-source,
and the scripts to complete the workflow are available on at https://zenodo.org/record/6763130#.YrpLdDVBzd4
(last accessed July 1, 2021).
References
Ando, R., & Kaneko, Y. (2018). Dynamic rupture simulation reproduces spontaneous multifault rupture and arrest during the 2016 Mw 7.9
Kaikōura earthquake. Geophysical Research Letters, 45(23), 875–12. https://doi.org/10.1029/2018GL080550
Bai, Y., Lay, T., Cheung, K. F., & Ye, L. (2017). Two regions of seafloor deformation generated the tsunami for the 13 November 2016, Kaikōura,
New Zealand earthquake. Geophysical Research Letters, 44(13), 6597–6606. https://doi.org/10.1002/2017GL073717
Balfour, N., Savage, M., & Townend, J. (2005). Stress and crustal anisotropy in Marlborough, New Zealand: Evidence for low fault strength and
structure-controlled anisotropy. Geophysical Journal International, 163(3), 1073–1086. https://doi.org/10.1111/j.1365-246x.2005.02783.x
Beaucé, E., Frank, W. B., & Romanenko, A. (2018). Fast matched filter (FMF): An efficient seismic matched-filter search for both CPU and GPU
architectures. Seismological Research Letters, 89(1), 165–172. https://doi.org/10.1785/0220170181
Bradley, B. A., Razafindrakoto, H. N., & Polak, V. (2017). Ground-motion observations from the 14 November 2016 M w 7.8 Kaikōura,
New Zealand, earthquake and insights from broadband simulations. Seismological Research Letters, 88(3), 740–756. https://doi.
org/10.1785/0220160225
Bratt, S., & Nagy, W. (1991). The LocSAT program. Science Applications International Corporation.
Cesca, S., Zhang, Y., Mouslopoulou, V., Wang, R., Saul, J., Savage, M., etal. (2017). Complex rupture process of the Mw 7.8, 2016, Kaikōura
earthquake, New Zealand, and its aftershock sequence. Earth and Planetary Science Letters, 478, 110–120. https://doi.org/10.1016/j.
epsl.2017.08.024
Chamberlain, C. J., Hopp, C. J., Boese, C. M., Warren-Smith, E., Chambers, D., Chu, S. X., etal. (2018). EQcorrscan: Repeating and near-repeating
earthquake detection and analysis in python. Seismological Research Letters, 89(1), 173–181. https://doi.org/10.1785/0220170151
Chamberlain, C. J., Townend, J., & Gerstenberger, M. C. (2020). RT-EQcorrscan: Near-real-time matched-filtering for rapid development of
dense earthquake catalogs. Seismological Society of America, 91(6), 3574–3584. https://doi.org/10.1785/0220200171
Christophersen, A., Rhoades, D., Gerstenberger, M., Bannister, S., Becker, J., Potter, S., & McBride, S. (2017). Progress and challenges in opera-
tional earthquake forecasting in New Zealand. In New Zealand society for earthquake engineering annual technical conference.
Clark, K., Nissen, E., Howarth, J., Hamling, I., Mountjoy, J., Ries, W., etal. (2017). Highly variable coastal deformation in the 2016 MW7.8
Kaikura earthquake reflects rupture complexity along a transpressional plate boundary. Earth and Planetary Science Letters, 474, 334–344.
https://doi.org/10.1016/j.epsl.2017.06.048
Darby, D. J., & Beanland, S. (1992). Possible source models for the 1855 Wairarapa earthquake, New Zealand. Journal of Geophysical Research:
Solid Earth, 97(B9), 12375–12389. https://doi.org/10.1029/92jb00567
Diederichs, A., Nissen, E., Lajoie, L., Langridge, R., Malireddi, S., Clark, K., etal. (2019). Unusual kinematics of the Papatea fault (2016
Kaikōura earthquake) suggest anelastic rupture. Science advances, 5(10), eaax5703. https://doi.org/10.1126/sciadv.aax5703
Eberhart-Phillips, D., & Bannister, S. (2015). 3-D imaging of the northern Hikurangi subduction zone, New Zealand: Variations in subducted
sediment, slab fluids and slow slip. Geophysical Journal International, 201(2), 838–855. https://doi.org/10.1093/gji/ggv057
Eberhart-Phillips, D., Bannister, S., & Reyners, M. (2017). New Zealand wide model 2.1 seismic velocity model for New Zealand. Zenodo. https://
doi.org/10.5281/zenodo.1043558
Eberhart-Phillips, D., Ellis, S., Lanza, F., & Bannister, S. (2021). Heterogeneous material properties–As inferred from seismic attenuation - Influ-
enced multiple fault rupture and ductile creep of the Kaikōura Mw 7.8 earthquake, New Zealand. Geophysical Journal International. https://
doi.org/10.1093/gji/ggab272
Frank, W. B., Poli, P., & Perfettini, H. (2017). Mapping the rheology of the Central Chile subduction zone with aftershocks. Geophysical Research
Letters, 44(11), 5374–5382. https://doi.org/10.1002/2016gl072288
Frigo, M., & Johnson, S. G. (1998). FFTW: An adaptive software architecture for the FFT. In Proceedings of the 1998 Ieee International Confer-
ence on Acoustics, Speech and Signal Processing, Icassp’98 (cat. no. 98ch36181) (Vol. 3,pp.1381–1384).
Gusman, A. R., Satake, K., Gunawan, E., Hamling, I., & Power, W. (2018). Contribution from multiple fault ruptures to tsunami generation during
the 2016 Kaikōura earthquake. Pure and Applied Geophysics, 175(8), 2557–2574. https://doi.org/10.1007/s00024-018-1949-z
Hainzl, S. (2016). Rate-dependent incompleteness of earthquake catalogs. Seismological Research Letters, 87(2A), 337–344. https://doi.
org/10.1785/0220150211
Hamling, I. J. (2020). A review of the 2016 Kaikura earthquake: Insights from the first 3 years. Journal of the Royal Society of New Zealand,
50(2), 226–244. https://doi.org/10.1080/03036758.2019.1701048
Hamling, I. J., D'Anastasio, E., Wallace, L. M., Ellis, S., Motagh, M., Samsonov, S., etal. (2014). Crustal deformation and stress transfer during
a propagating earthquake sequence: The 2013 Cook Strait sequence, central New Zealand. Journal of Geophysical Research: Solid Earth,
119(7), 6080–6092. https://doi.org/10.1002/2014JB011084
Hamling, I. J., Hreinsdóttir, S., Clark, K., Elliott, J., Liang, C., Fielding, E., etal. (2017). Complex multifault rupture during the 2016 Mw 7.8
Kaikōura earthquake, New Zealand. Science, 356(6334), eaam7194. https://doi.org/10.1126/science.aam7194
Harris, R. A., Archuleta, R. J., & Day, S. M. (1991). Fault steps and the dynamic rupture process: 2-D numerical simulations of a spontaneously
propagating shear fracture. Geophysical Research Letters, 18(5), 893–896. https://doi.org/10.1029/91gl01061
Henrys, S., Eberhart-Phillips, D., Bassett, D., Sutherland, R., Okaya, D., Savage, M., & others. (2020). Upper plate heterogeneity along the South-
ern Hikurangi Margin, New Zealand. Geophysical Research Letters, 47(4), e2019GL085511. https://doi.org/10.1029/2019gl085511
Holden, C., Kaneko, Y., D'Anastasio, E., Benites, R., Fry, B., & Hamling, I. J. (2017). The 2016 Kaikura earthquake revealed by kinematic
source inversion and seismic wavefield simulations: Slow rupture propagation on a geometrically complex crustal fault network. Geophysical
Research Letters, 44(22), 11320–11328. https://doi.org/10.1002/2017GL075301
Howell, A., Nissen, E., Stahl, T., Clark, K., Kearse, J., Van Dissen, R., etal. (2020). Three-dimensional surface displacements during the 2016
MW 7.8 Kaikura earthquake (New Zealand) from photogrammetry-derived point clouds. Journal of Geophysical Research: Solid Earth,
125(1), e2019JB018739. https://doi.org/10.1029/2019JB018739
Hunter, J. D. (2007). Matplotlib: A 2D graphics environment. Computing in Science & Engineering, 9(3), 90–95. https://doi.org/10.1109/
MCSE.2007.55
Illsley-Kemp, F., Barker, S. J., Wilson, C. J., Chamberlain, C. J., Hreinsdóttir, S., Ellis, S., etal. (2021). Volcanic unrest at Taupō volcano in 2019:
Causes, mechanisms and implications. Geochemistry, Geophysics, Geosystems, 22, e2021GC009803. https://doi.org/10.1029/2021GC009803
Acknowledgments
Chamberlain and Townend are grateful
to New Zealand's Earthquake Commis-
sion (EQC) for funding through the
EQC Programme in Seismology and
Tectonic Geodesy at Victoria University
of Wellington, and an EQC Biennial
Grant (18/753), and to the Royal Society
of New Zealand for funding through the
Marsden Fast-Start project 17-VUW-121.
Frank, Townend and Chamberlain also
received funding for this collaboration
through the New Zealand Royal Societies
Catalyst Grant scheme. Chamberlain and
Warren-Smith are grateful to the New
Zealand Ministry of Business, Innovation
and Employment for funding through the
Endeavour programme: ’Rapid Charac-
terisation of Earthquakes and Tsunamis
(RCET)'. Warren-Smith was funded
through GNS Science Ministy of Business
Innovation and Employment strategic
science investment funding (GNS-SSIF).
Initial compute time was provided by
the New Zealand eScience Infrastructure
(NeSI) for which we are grateful. Later
computations were run on Amazon Web
Services, with data handling assistance
from GeoNet. We acknowledge the New
Zealand GeoNet project and its sponsors
EQC, GNS Science, LINZ, NEMA and
MBIE for providing data used in this
study. ObsPy (Krischer etal.,2015) was
used extensively throughout this work,
figures were made using matplot-
lib (Hunter,2007) and cartopy (Met
Office,2010–2015), and we are very
grateful to the developers and maintainers
of these open-source projects who enable
complex science. We are grateful to the
Editor, Rachel Abercrombie, Daniel
Trugman and Thorne Lay for their helpful
reviews. Finally we are grateful to Tim
Little, Andy Howell, Carolyn Boulton,
James Crampton, Rob Langridge, Laura
Wallace, Ian Hamling and others in the
New Zealand geoscience community for
informative discussions during the prepa-
ration of this work. We are also grateful
to Fengzhou Tan who identified an error
in our initial focal mechanism catalogue
and assisted the authors in correcting
the catalogue. This manuscript has been
updated to incorporate the corrected focal
mechanisms.

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
23 of 24
Kaiser, A., Balfour, N., Fry, B., Holden, C., Litchfield, N., Gerstenberger, M., etal. (2017). The 2016 Kaikura, New Zealand, earthquake: Prelim-
inary seismological report. Seismological Research Letters, 88(3), 727–739. https://doi.org/10.1785/0220170018
Kearse, J., Little, T. A., Van Dissen, R. J., Bar nes, P.M., Langridge, R., Mountjoy, J., etal. (2018). Onshore to offshore ground-surface and
seabed rupture of the Jordan–Kekerengu–Needles fault network during the 2016 Mw 7.8 Kaikura earthquake, New Zealand. Bulletin of the
Seismological Society of America, 108(3B), 1573–1595. https://doi.org/10.1785/0120170304
Kennett, B., & Engdahl, E. (1991). Traveltimes for global ear thquake location and phase identification. Geophysical Journal International,
105(2), 429–465. https://doi.org/10.1111/j.1365-246x.1991.tb06724.x
King, G. C., Stein, R. S., & Lin, J. (1994). Static stress changes and the triggering of earthquakes. Bulletin of the Seismological Society of Amer-
ica, 84(3), 935–953.
Klinger, Y., Okubo, K., Vallage, A., Champenois, J., Delorme, A., Rougier, E., etal. (2018). Earthquake damage patterns resolve complex rupture
processes. Geophysical Research Letters, 45(19), 10–279. https://doi.org/10.1029/2018gl078842
Krischer, L., Megies, T., Barsch, R., Beyreuther, M., Lecocq, T., Caudron, C., & Wassermann, J. (2015). ObsPy: A bridge for seismology into the
scientific Python ecosystem. Computational Science & Discovery, 8(1), 014003. https://doi.org/10.1088/1749-4699/8/1/014003
Lamb, S., Arnold, R., & Moore, J. D. (2018). Locking on a megathrust as a cause of distributed faulting and fault-jumping earthquakes. Nature
Geoscience, 11(11), 871–875. https://doi.org/10.1038/s41561-018-0230-5
Langridge, R., Ries, W., Litchfield, N., Villamor, P., Van Dissen, R., Barrell, D., etal. (2016). The New Zealand active faults database. New
Zealand Journal of Geology and Geophysics, 59(1), 86–96. https://doi.org/10.1080/00288306.2015.1112818
Langridge, R. M., Rowland, J., Villamor, P., Mountjoy, J., Townsend, D. B., Nissen, E., etal. (2018). Coseismic rupture and preliminary slip
estimates for the Papatea fault and its role in the 2016 Mw7.8 Kaikura, New Zealand, earthquake. Bulletin of the Seismological Society of
America, 108(3B), 1596–1622. https://doi.org/10.1785/0120170336
Lanza, F., Chamberlain, C. J., Jacobs, K., Warren-Smith, E., Godfrey, H. J., Kortink, M., etal. (2019). Crustal fault connectivity of the Mw
7.8 2016 Kaikura Earthquake constrained by aftershock relocations. Geophysical Research Letters, 46(12), 6487–6496. https://doi.
org/10.1029/2019GL082780
Lienert, B. R., & Havskov, J. (1995). A computer program for locating earthquakes both locally and globally. Seismological Research Letters,
66(5), 26–36. https://doi.org/10.1785/gssrl.66.5.26
Litchfield, N. J., Villamor, P., Dissen, R. J. V., Nicol, A., Barnes, P.M., Barrell, A., etal. (2018). Surface rupture of multiple crustal faults in
the 2016 M w 7.8 Kaikōura, New Zealand, Earthquake. Bulletin of the Seismological Society of America, 108(3B), 1496–1520. https://doi.
org/10.1785/0120170300
Liu, C., Lay, T., Brodsky, E. E., Dascher-Cousineau, K., & Xiong, X. (2019). Coseismic rupture process of the large 2019 ridgecrest earth-
quakes from joint inversion of geodetic and seismological observations. Geophysical Research Letters, 46(21), 11820–11829. https://doi.
org/10.1029/2019GL084949
Lomax, A., Virieux, J., Volant, P., & Berge-Thierry, C. (2000). Probabilistic earthquake location in 3D and layered models. In Advances in seismic
event location (pp.101–134). Springer. https://doi.org/10.1007/978-94-015-9536-0_5
Met Office. (2010–2015). Cartopy: A cartographic python library with a Matplotlib interface [computer software manual]. Exeter, Devon.
Retrieved from https://scitools.org.uk/cartopy
Michailos, K., Smith, E. G., Chamberlain, C. J., Savage, M. K., & Townend, J. (2019). Variations in seismogenic thickness along the Central
Alpine Fault, New Zealand, revealed by a decade's relocated microseismicity. Geochemistry, Geophysics, Geosystems, 20(1), 470–486. https://
doi.org/10.1029/2018gc007743
Mouslopoulou, V., Saltogianni, V., Nicol, A., Oncken, O., Begg, J., Babeyko, A., etal. (2019). Breaking a subduction-termination from top to
bottom: The large 2016 Kaikura Earthquake, New Zealand. Earth and Planetary Science Letters, 506, 221–230. https://doi.org/10.1016/j.
epsl.2018.10.020
Nicol, A., Khajavi, N., Pettinga, J. R., Fenton, C., Stahl, T., Bannister, S., etal. (2018). Preliminary geometry, displacement, and kinematics of
fault ruptures in the epicentral region of the 2016 Mw 7.8 Kaikura, New Zealand, Earthquake. Bulletin of the Seismological Society of America,
108(3B), 1521–1539. https://doi.org/10.1785/0120170329
Okada, T., Iio, Y., Matsumoto, S., Bannister, S., Ohmi, S., Horiuchi, S., etal. (2019). Comparative tomography of reverse-slip and strike-slip seis-
motectonic provinces in the northern South Island, New Zealand. Tectonophysics, 765, 172–186. https://doi.org/10.1016/j.tecto.2019.03.016
Peng, Z., Fry, B., Chao, K., Yao, D., Meng, X., & Jolly, A. (2018). Remote triggering of microearthquakes and tremor in New Zealand following the
2016 Mw 7.8 Kaikura Earthquake. Bulletin of the Seismological Society of America, 108(3B), 1784–1793. https://doi.org/10.1785/0120170327
Perfettini, H., Frank, W., Marsan, D., & Bouchon, M. (2018). A model of aftershock migration driven by afterslip. Geophysical Research Letters,
45(5), 2283–2293. https://doi.org/10.1002/2017gl076287
Plesch, A., Shaw, J. H., Ross, Z. E., & Hauksson, E. (2020). Detailed 3D fault representations for the 2019 Ridgecrest, California, earthquake
sequence. Bulletin of the Seismological Society of America, 110(4), 1818–1831. https://doi.org/10.1785/0120200053
Rattenbury, M., & Isaac, M. (2012). The QMAP 1: 250 000 geological map of New Zealand project. New Zealand Journal of Geology and
Geophysics, 55(4), 393–405. https://doi.org/10.1080/00288306.2012.725417
Ristau, J. (2013). Update of regional moment tensor analysis for earthquakes in New Zealand and adjacent offshore regions. Bulletin of the Seis-
mological Society of America, 103(4), 2520–2533. https://doi.org/10.1785/0120120339
Rodgers, D., & Little, T. (2006). World's largest coseismic strike-slip offset: The 1855 rupture of the Wairarapa Fault, New Zealand, and impli-
cations for displacement/length scaling of continental earthquakes. Journal of Geophysical Research: Solid Earth, 111(B12), B12408. https://
doi.org/10.1029/2005jb004065
Romanet, P., & Ide, S. (2019). Ambient tectonic tremors in Manawatu, Cape Turnagain, Marlborough, and Puysegur, New Zealand. Earth, Plan-
ets and Space, 71(1), 1–9. https://doi.org/10.1186/s40623-019-1039-1
Senobari, N. S., Funning, G. J., Keogh, E., Zhu, Y., Yeh, C.-C. M., Zimmerman, Z., & Mueen, A. (2019). Super-efficient cross-correlation
(SEC-C): A fast matched filtering code suitable for desktop computers. Seismological Research Letters, 90(1), 322–334. https://doi.
org/10.1785/0220180122
Stirling, M., McVerry, G., Gerstenberger, M., Litchfield, N., Van Dissen, R., Ber ryman, K., etal. (2012). National seismic hazard model for New
Zealand: 2010 update. Bulletin of the Seismological Society of America, 102(4), 1514–1542. https://doi.org/10.1785/0120110170
Tan, F., Ge, Z., Kao, H., & Nissen, E. (2019). Validation of the 3-D phase-weighted relative back projection technique and its application to the
2016 M w 7.8 Kaikōura Earthquake. Geophysical Journal International, 217(1), 375–388. https://doi.org/10.1093/gji/ggz032
Townend, J., Sherburn, S., Arnold, R., Boese, C., & Woods, L. (2012). Three-dimensional variations in present-day tectonic stress along the
Australia–Pacific plate boundary in New Zealand. Earth and Planetary Science Letters, 353, 47–59. https://doi.org/10.1016/j.epsl.2012.08.003

Journal of Geophysical Research: Solid Earth
CHAMBERLAIN ETAL.
10.1029/2021JB022304
24 of 24
Trugman, D. T., & Shearer, P. M. (2017). GrowClust: A hierarchical clustering algorithm for relative earthquake relocation, with applica-
tion to the Spanish Springs and Sheldon, Nevada, earthquake sequences. Seismological Research Letters, 88(2A), 379–391. https://doi.
org/10.1785/0220160188
Ulrich, T., Gabriel, A.-A., Ampuero, J.-P., & Xu, W. (2019). Dynamic viability of the 2016 Mw 7.8 Kaikōura earthquake cascade on weak crustal
faults. Nature Communications, 10(1), 375–388. https://doi.org/10.1038/s41467-019-09125
Van Dissen, R., & Nicol, A. (2009). Mid-late Holocene paleoseismicity of the eastern Clarence Fault, Marlborough, New Zealand. New Zealand
Journal of Geology and Geophysics, 52(3), 1213. https://doi.org/10.1080/00288300909509886
Van Dissen, R., & Yeats, R. S. (1991). Hope fault, Jordan thrust, and uplift of the seaward Kaikōura Range, New Zealand. Geology, 19(4),
393–396. https://doi.org/10.1130/0091-7613(1991)019<0393:hfjtau>2.3.co;2
Waldhauser, F., & Ellsworth, W. L. (2000). A double-difference earthquake location algorithm: Method and application to the northern Hayward
fault, California. Bulletin of the Seismological Society of America, 90(6), 1353–1368. https://doi.org/10.1785/0120000006
Wallace, L. M., Hreinsdóttir, S., Ellis, S., Hamling, I., D'Anastasio, E., & Denys, P. (2018). Triggered slow slip and afterslip on the south-
ern hikurangi subduction zone following the Kaikura Earthquake. Geophysical Research Letters, 45(10), 4710–4718. https://doi.
org/10.1002/2018GL077385
Wallace, L. M., Kaneko, Y., Hreinsdóttir, S., Hamling, I., Peng, Z., Bartlow, N., etal. (2017). Large-scale dynamic triggering of shallow slow slip
enhanced by overlying sedimentary wedge. Nature Geoscience, 10, 765–770. https://doi.org/10.1038/NGEO3021
Walsh, D., Arnold, R., & Townend, J. (2009). A Bayesian approach to determining and parametrizing earthquake focal mechanisms. Geophysical
Journal International, 176(1), 235–255. https://doi.org/10.1111/j.1365-246x.2008.03979.x
Wang, D., Chen, Y., Wang, Q., & Mori, J. (2018). Complex rupture of the 13 November 2016 Mw 7.8 Kaikōura, New Zealand earthquake:
Comparison of high-frequency and low-frequency observations. Tectonophysics, 733, 100–107. https://doi.org/10.1016/j.tecto.2018.02.004
Wang, T., Jiao, L., Tapponnier, P., Shi, X., & Wei, S. (2020). Space imaging geodesy reveals near circular, coseismic block rotation during the 2016
Mw 7.8 Kaikōura earthquake, New Zealand. Geophysical Research Letters, 47(22), e2020GL090206. https://doi.org/10.1029/2020gl090206
Wang, T., Wei, S., Shi, X., Qiu, Q., Li, L., Peng, D., etal. (2018). The 2016 Kaikōura earthquake: Simultaneous rupture of the subduction inter-
face and overlying faults. Earth and Planetary Science Letters, 482, 44–51. https://doi.org/10.1016/j.epsl.2017.10.056
Warren-Smith, E., Chamberlain, C. J., Lamb, S., & Townend, J. (2017). High-precision analysis of an aftershock sequence using matched-filter
detection: The 4 May 2015 ML 6 Wanaka earthquake, Southern Alps, New Zealand. Seismological Research Letters, 88(4), 1065–1077.
Wesnousky, S. G. (2006). Predicting the endpoints of earthquake ruptures. Nature, 444(7117), 358–360. https://doi.org/10.1038/nature05275
Wiemer, S., & Wyss, M. (2000). Minimum magnitude of completeness in earthquake catalogs: Examples from Alaska, the western United States,
and Japan. Bulletin of the Seismological Society of America, 90(4), 859–869. https://doi.org/10.1785/0119990114
Williams, C. A., Eberhart-Phillips, D., Bannister, S., Barker, D. H. N., Henrys, S., Reyners, M., & Sutherland, R. (2013). Revised interface geom-
etry for the Hikurangi subduction zone, New Zealand. Seismological Research Letters, 84(6), 1066–1073. https://doi.org/10.1785/0220130035
Williams, J. N., Barrell, D. J. A., Stirling, M. W., Sauer, K. M., Duke, G. C., & Hao, K. X. (2018). Surface rupture of the Hundalee Fault during the
2016 Mw 7.8 Kaikura earthquake. Bulletin of the Seismological Society of America, 06(3B), 1540–1555. https://doi.org/10.1785/0120170291
Xu, W., Feng, G., Meng, L., Zhang, A., Ampuero, J. P., Bürgmann, R., & Fang, L. (2018). Transpressional rupture cascade of the 2016 Mw 7.8
Kaikōura earthquake, New Zealand. Journal of Geophysical Research: Solid Earth, 123(3), 2396–2409. https://doi.org/10.1002/2017JB015168
Yao, D., Peng, Z., Kaneko, Y., Fry, B., & Meng, X. (2021). Dynamic triggering of earthquakes in the North Island of New Zealand following the
2016 Mw 7.8 Kaikōura earthquake. Earth and Planetary Science Letters, 557, 116723. https://doi.org/10.1016/j.epsl.2020.116723
Yu, C., Li, Z., & Penna, N. T. (2020). Triggered afterslip on the southern Hikurangi subduction interface following the 2016 Kaikōura earthquake
from InSAR time series with atmospheric corrections. Remote Sensing of Environment, 251, 112097. https://doi.org/10.1016/j.rse.2020.112097
Zinke, R., Hollingsworth, J., Dolan, J. F., & Van Dissen, R. (2019). Three-dimensional surface deformation in the 2016 MW 7.8 Kaikōura, New
Zealand, earthquake from optical image correlation: Implications for strain localization and long-term evolution of the Pacific-Australian plate
boundary. Geochemistry, Geophysics, Geosystems, 20(3), 1609–1628. https://doi.org/10.1029/2018gc007951
Erratum
The originally published version of this article included an error in the initial focal mechanism catalogue. The
error did not affect the overall conclusions of the article. Figures 4 and 5, as well as the Supporting Information,
have been replaced with corrected versions; the data and software archive at Zenodo has been updated; and minor
text changes have been made to the Introduction and Section 4.1. This may be considered the official version of
record.