Upload
ngodan
View
217
Download
0
Embed Size (px)
Citation preview
Three-dimensional mechanical models for the June 2000 earthquake sequence in the South Icelandic Seismic Zone
Loïc Dubois a, Kurt L. Feigl a, *, Dimitri Komatitsch b, Thóra Árnadóttir c, and Freysteinn Sigmundsson c
a Dynamique Terrestre et Planétaire CNRS UMR 5562, Université Paul Sabatier, Observatoire Midi-Pyrénées, 14 av. Édouard Belin, 31400 Toulouse, France
b Laboratoire de Modélisation et d'Imagerie en Géosciences CNRS UMR 5212, Université de Pau et des Pays de l'Adour, BP 1155, 64013 Pau Cedex, France
c Nordic Volcanological Center, Institute of Earth Sciences, University of Iceland, Reykjavik, Iceland
Tuesday, May 09, 2023
Submitted to Earth and Planetary Science Letters
Number of characters = 0Number of words = 0Number of words in text only = 6455 Max is 6500 (not abstract, captions or references)Supplementary Material? YesNumber of pages (including figures) = 22Number of figures = 6Number of tables = 2Total number of display items = 8 Max is 10.Number of references = 79 Max is 80.
File name = document.doc
Application nameMicrosoft Word 2004 for MacEndNote Version 10.0 for MacMathType Version 5 for Mac.
*Corresponding author, now atDepartment of Geology and GeophysicsUniversity of Wisconsin-Madison1215 West Dayton StreetMadison, WI53706 [email protected]. +1 608 262-8960Fax. +1 608 262-0693
1
1
2
3
4
5
6
7
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
3233
3435
36
37
38
39
40
41
42
43
44
45
46
47
48
2
Dubois et al. submitted 2023-05-09
TABLE OF CONTENTS
Table of contents 2
Abstract 3
1. Introduction 3
2. Data and Methods 6
2.1 Data 62.2 Numerical Solution using the Finite Element Method 72.3 Modeling poroelastic deformation with an elastic approximation 82.4 Viscoelastic modeling 92.5 Stress transfer 10
3. Results 10
3.1 Estimating the distribution of co-seismic fault slip 103.2 Modeling postseismic deformation with the elastic approximation for poroelastic effects 113.3 Viscoelastic model for post-seismic deformation 123.4 Stress transfer 13
4. Discussion and Conclusions 14
4.1 Co-seismic effects 144.2 Viscoelastic effects 154.3 Poroelastic effects 154.4 Migrating seismicity 16
Acknowledgments 16
Supplementary Material 17
SM.1 Testing against other solutions 17SM.2 Estimating co-seismic slip by joint inversion of GPS and INSAR measurements 17SM.3 Approximating complete poroelastic relaxation by an elastic solution 18
References 20
Tables 26
Last page 26
Page 2
34
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
56
Dubois et al. submitted 2023-05-09
ABSTRACT
We analyze a sequence of earthquakes that began in June 2000 in the South Iceland Seismic Zone (SISZ)
in light of observations of crustal deformation, seismicity, and water level, focusing on the tendency for
seismicity to migrate from east to west. At least five processes may be involved in transferring stress from one
fault to another: (1) propagation of seismic waves, (2) changes of static stress caused by co-seismic slip, (3)
inter-seismic strain accumulation, (4) fluctuations in hydrological conditions, and (5) ductile flow of subcrustal
rocks. All five of these phenomena occurred in the SISZ during the post-seismic time interval following the
June 2000 earthquakes. By using three-dimensional finite element models with realistic geometric and rheologic
configurations to match the observations, we test hypotheses for the underlying geophysical processes. The
coseismic slip distribution for the Mw 6.6 June 21 mainshock is deeper when estimated in a realistic layered
configuration with varying crustal thickness than in a homogenous half-space. Processes (1) and (2) explain
only the aftershocks within a minute of the mainshocks. A quasi-static approximation for modeling poroelastic
effects in process (4) by changing Poisson’s ratio from undrained to drained conditions does not provide an
adequate fit to post-seismic deformation recorded by interferometric analysis of synthetic aperture radar images.
The same approximation fails to explain the location of subsequent seismicity, including the June 21 mainshock.
Although a viscoelastic model of subcrustal ductile flow (5) can increase Coulomb failure stress by more than
10 kPa, it does so over time scales longer than a year. Accumulating interseismic strain (3) in an elastic upper
crust that thickens eastward breaks the symmetry in the lobes where co-seismic stress exceeds the Coulomb
failure criterion. This process may explain the tendency of subsequent earthquakes to migrate from east to west
across the South Iceland Seismic Zone within a single sequence. It does not, however, explain their timing.
1. INTRODUCTION
The South Iceland Seismic Zone (SISZ) is aptly named. Located on the boundary between the North
America and Eurasia plates, it accommodates their relative motion as earthquakes. The plate boundary follows
the spreading mid-Atlantic Ridge, comes ashore on the Reykjanes Peninsula, where it becomes a shear zone to
reach the Western Volcanic Zone at a triple point near the Hengill volcanic center (Fig 1). There, most of the
present-day deformation continues across the SISZ to the Eastern Volcanic Zone. Overall, the SISZ
accommodates sinistral shear along its east-to-west trend, but most of its large (M 6) earthquakes rupture as
right-lateral strike slip on vertical faults striking north. These parallel faults are 10 – 20 km long and ~5 km
apart, analogous to books on a shelf [1]. Indeed, the SISZ deforms as a shear zone, accommodating
18.9 ± 0.5 mm/yr of relative plate motion [2, 3] in a band less than 20 km wide [4]. Consequently, the
interseismic strain rate is ~10–6 per year. Its kinematics can be described as a locked fault zone slipping 16 to
20 mm/yr below 8 to 12 km depth with a dislocation in an elastic half-space [5]. If accumulated elastically over
Page 3
78
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
910
Dubois et al. submitted 2023-05-09
a century, this strain could be released as ~2 m of co-seismic slip. Fortunately, earthquakes in the SISZ do not
rupture its entire 100-km length from east to west; the 20-km length of the north-south faults seems to limit their
magnitude to 6.5 or 7.
The seismic sequence of events began on 17 June 2000, with a main shock of magnitude Mw = 6.6 [6]. It
produced co-seismic deformation that was measured geodetically by both GPS [7] and INSAR [8]. Three and a
half days (81 hours) later, a second earthquake ruptured a distinct fault some 18 km to the west of the first event.
Both earthquakes have strike-slip focal mechanisms on faults dipping nearly vertically [9]. Both earthquakes
ruptured the ground surface in a right-lateral, en echelon pattern with an overall strike within a few degrees of
North [10]. The slip distributions estimated from geodetic data for both earthquakes have maxima over 2 m,
centroid depths around 6 km, down-dip widths of 10 – 12 km, and along-strike lengths of 16 – 18 km [7, 11].
In this study, we build on previous estimates of the distribution of slip that occurred on the rupture
surfaces of the two mainshocks in June 2000 from GPS measurements of displacements [7], INSAR recordings
of range change [8], and a combination of both in a joint inversion [11]. All three of these studies assume an
elastic half space with uniform values of shear modulus (or rigidity) and Poisson’s ratio . Yet other work
STUDY/STUDIES used four times in the last two sentences outside of Iceland suggest that this simple
assumption can bias estimates of slip distribution [12, 13]. For example, the aftershock hypocenters located by
seismological procedures are deeper than the bottom of the fault rupture inferred from geodetic measurements in
the case of the 1994 Northridge earthquake in California [14]. In other words, the seismological and geodetic
procedures for locating co-seismic slip will yield different estimates if they assume different elastic models. A
similar discrepancy between the deepest aftershocks and the bottom of the slip distribution also appears to apply
to the June 2000 sequence in the SISZ. Although the slip diminishes to less than 1 m below 8 km depth [15], the
relocated aftershock hypocenters reach down to 10 km, with a few as deep as 12 km [16]. The hypocentral
depths estimated from seismology using a layered velocity model are 6.3 and 5.0 km for the June 17 and 21
mainshocks, respectively [6]. By using layered models that account for increasing rigidity with depth, we expect
deeper slip distributions [17, 18]. An accurate estimate of the slip distribution is required to test the hypothesis
that the M 6 earthquakes in the SISZ rupture into the lower crust [19]. Accurate estimates of slip distribution are
also important for calculating maps of earthquakes triggered by stress transfer. The latter is sensitive to the
former because the threshold value of 10 kPa (0.1 bar) usually used to evaluate the Coulomb failure criterion is
several orders of magnitude smaller than the stress drop for a typical earthquake [20, 21].
The earthquakes in the 2000 SISZ sequence seem to be causally related. Just after the June 17 mainshock,
additional earthquakes struck the Reykanes Peninsula, as far as 80 km to the west of the June 17 epicenter.
Three of these generated coseismic displacements large enough to measure by INSAR and GPS and to model
with dislocation sources with moments equivalent to Mw = 5.3, 5.5 and 5.9 [22, 23]. These secondary
earthquakes ruptured the surface, broke rocks, and coincided with a drop of several meters in the water level of
Page 4
1112
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
1314
Dubois et al. submitted 2023-05-09
Lake Klevervatn [24]. Three of these events, at 8 s, 26 s, and 30 s after the June 17 mainshock at 15:40:41 UTC
were dynamically triggered by the seismic wave train as it propagated westward [6]. A fourth event occurred 5
minutes after the June 17 mainshock at a distance of almost 80 km [6]. The June 17 earthquake appears to have
triggered the June 21 earthquake, based on the increase in static Coulomb stress calculated at the June 21
hypocenter on a vertical, north-striking fault [25]. The migration of M ≥ 6 earthquakes from east to west across
the SISZ has been observed in previous earthquake sequences in 1732-1734, 1784, and 1896 [1]. Another M ≥ 6
event occurred in 1912 at the east end of the SISZ [26, 27].
Understanding the processes driving the seismicity from east to west is the second objective of this study.
Although static Coulomb stress changes can explain the location of the triggered earthquakes, this simple theory
cannot explain the time delay between the “source” (triggering) and “receiver” (triggered) events, as pointed out
by Scholz [21, 28] and recently reviewed by Brodsky and Prejean [29]. At least five processes may be involved
in transferring stress from one fault to another: (1) propagation of seismic waves, (2) changes of static stress
caused by co-seismic slip, (3) inter-seismic strain accumulation, (4) fluctuations in hydrological conditions, and
(5) ductile flow of subcrustal rocks. All five of these phenomena occurred in the SISZ during the post-seismic
time interval following the June 2000 earthquakes.
(1) The dynamic stresses caused by propagation of seismic waves apparently triggered the earthquakes on
the Reykanes Peninsula in the first minute following the June 17 mainshock, as described above [6].
(2) In the “cascading seismicity” hypothesis, one earthquake triggers the next in a kind of seismic
“domino effect”. Changes in static stress could alter rate- and state-dependent friction, thus producing and
(?????) a finite time delay between successive earthquakes[30].
(3) In terms of moment, the main events in the June 2000 sequence released only about a quarter of the
strain accumulated since 1912, the date of the last major earthquake in the SISZ, assuming a constant strain rate
over the intervening 88 years [4, 15]. Such interseismic strain accumulation could conceivably contribute to
stress transfer. If strain accumulates at the rate of 10-6/yr, it could increase stress by 10 kPa in less than a year,
assuming a Young’s modulus of 50 GPa.
(4) Fluctuations in hydrological conditions occurred in the weeks to months after the June 17 mainshock
[31, 32]. Poroelastic effects can explain about half the post-seismic signal in an interferogram spanning June 19
through July 24 in the SISZ [32]. The same effects might also explain stress transfer in the 2000 SISZ sequence.
Under this hypothesis, the co-seismic perturbation to the hydrologic pressure field also alters the stress
conditions on faults near the mainshock, triggering aftershocks in areas where increasing pressure unclamps
faults near failure [33, 34].
(5) Ductile flow of subcrustal rocks occurs over longer time scales of several years. For example post-
seismic deformation has been recorded by GPS around the June 2000 faults as late as May 2004 [35]. These
centimeter-sized displacements have been modeled using a viscoelastic Burger’s rheology in a semi-analytic
Page 5
1516
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
1718
Dubois et al. submitted 2023-05-09
spherical harmonic formulation [35]. Here, we use a linear Maxwell viscoelastic rheology in a Cartesian finite
element formulation to investigate the effect of geometry, viscosity and other parameters on surface
displacements. Although which one ??? you talk about geometry, viscosity and other parameters in the previous
sentence ductile flow is much too slow to explain the 86-hour time delay between the June 17 and June 21
events, it may contribute to the time delays (~100 years) between earthquake sequences in the SISZ.
All these considerations motivate us to move beyond the first-order in which sense ??? better to suppress
that word I guess approximation of a half-space with uniform elastic properties. In this study, we explore
models with more realistic geometric and rheologic configurations. To do so, we calculate stress and
displacement fields using a finite element method [e.g., 36]. In particular, we use the TECTON code, renovated
by Charles Williams [37-39], as described below. This approach allows us to account for the variations in
crustal thickness (Fig 1 and Fig 2) inferred from earthquake hypocenter locations, seismic tomography, and
gravity modeling [40-44].
2. DATA AND METHODS
2.1 Data
In this study, we consider four types of data, all of which have been analyzed and described previously.
GPS
The complete GPS data set, including the post-seismic time series, has been described elsewhere [5, 35].
The co-seismic displacement vectors are the same as used previously to estimate the distribution of slip on the
faults that ruptured on June 17 and 21 [7, 11].
INSAR
Interferometric analysis of synthetic aperture radar images (INSAR) has recorded the crustal deformation
that occurred between June and September, 2000, as described previously [8, 15, 22, 32]. The data set used here
is the same as analyzed by Pedersen et al. [11]The characteristic relaxation time of the post-seismic deformation
near the June 17 fault is of the order of a month because another interferogram spanning the second post-seismic
month (12 August to 16 September) shows less than 15 mm of differential range change across the fault (blue
curve in Figure 2e of Jónsson et al. [32]). About half of the post-seismic deformation observed within 5 km of
the June 17 fault recorded by INSAR in the first month can be explained by poroelastic effects involving fluid
flowing in the epicentral area [32].
Page 6
Call Fig 1.
1920
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
2122
Dubois et al. submitted 2023-05-09
Water level in wells and boreholes
Fluctuations in hydrological conditions have been observed in periodic measurements of pressure and
water level in wells and boreholes around the SISZ [31, 32]. The polarity of these changes can be well described
by the two mainshock focal mechanisms [31, 32]. In the post-seismic interval, however, the wells recover over
time scales varying from hours to weeks, indicating large variations in the rock-mechanical properties over
distances as short as a few kilometers.
Earthquake locations and focal mechanisms
The earthquakes recorded by the SIL seismological network [19, 45] operated by the Icelandic
Meteorological Office have been relocated using multiplet algorithms [46-48] to establish a catalog of focal
mechanisms and locations with a precision better than 1 km in all three dimensions [16, 49, 50] as part of the
PREPARED project [51].
2.2 Numerical Solution using the Finite Element Method
TECTON is a software package that implements a finite element formulation based on three-dimensional
hexahedral elements [37-39]. We have tested TECTON for accuracy by comparing its results to semi-analytic
solutions, as described in the doctoral thesis of the first author [52] and in the Supplemental Material. We use
the word “configuration” to describe the combination of the geometry of the mesh (topology) and distribution of
mechanical properties (rheology) within it. These configurations appear in Fig 1, Fig 2, Table 1, and Table 2.
Mesh
We have developed two meshes specifically for this problem to reproduce the main interfaces in our
problem: upper/lower crust, mantle/crust, as well as the fault rupture surfaces for the June 17 and June 21
mainshocks. The boundary conditions on the bottom and four sides are fixed while the top surface is free. The
region under study is 180 km by 100 km by 20 km (blue box in Fig 1). To avoid edge effects, we extend the
meshes to 300 km by 300 km by 100 km. The elastic solution achieves numerical convergence with 70 000
nodes, but we use 140 000 nodes to handle the time-dependence in the visc-elastic case [52]. The fault slip is
described as relative, horizontal displacements at 690 “split nodes” [38].
The simplest configuration is a homogeneous half-space labeled HOM. The mesh for HOM includes
horizontal layers with constant thickness. The same mesh defines the geometry of the heterogeneous HET
configuration. A different mesh, used in the TOM and ISL configurations, allows lateral variations in crustal
thickness, and thus gradients in rheologic properties, as sketched in Fig 1B.
Page 7
2324
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
2526
Dubois et al. submitted 2023-05-09
Rheology
After defining the meshes, we now define the rheology for each configuration. All the rheologic
configurations obey an elastic constitutive relation (Hooke’s law), with stress linearly proportional to strain. For
each element in the mesh, we specify the Poisson’s ratio and a shear modulus , as summarized in Fig 2 and
Table 1.
In the HOM and TOM configurations, we assume 0.28 and 30 GPa throughout the half-
space [11]. In the layered HET and ISL configurations, Poisson’s ratio varies from 0.28 in the upper crust
to 0.25 in the lower crust, and then to 0.30 in the mantle, in accordance with previous studies of
Iceland [40, 53, 54]. The thickness of these layers is constant in the HET configuration, with the base of the
upper crust set to Hu = 5 km depth and the base of the lower crust set to Hc = 23 km depth. In the ISL
configuration, the upper and lower crustal layers vary in thickness, as sketched in Fig 2B and Fig 2C. These two
crustal layers both thicken and dip toward the east)ity I don’t like this sentence very much, we should maybe
rephrase i?],) as evident in previous three-dimensio by studies [40, 41, 43, 44, 54]. Given the density, we then
use the P-wave and S-wave velocities [52] to calculate and (under undrained conditions) for each element
(Fig 2).
The shear modulus is set to 30 GPa in the HOM and TOM configurations. In the HET and ISL
configurations, it varies from 30 GPa in the upper crust to 60 GPa in the mantle. The HOF and
HEF configurations add weak fault zones to the HOM and HET configurations, respectively. In the elements
located within a few kilometers of the faults, the HOF and HET configurations include values as low as
3 GPa, i.e. four times smaller than in the surrounding crustal rocks [55]. Such low values can be justified by
previous co-seismic studies that have assumed values as low as 10 GPa [56] and as high as 50
GPa [57]. Following Cocco and Rice [55, 58], we assume that the bulk modulus has the same value in the
damage zone as in the surrounding rock. Hence we set Poisson’s ratio as high as 0.43 within the weak
fault zones in the HEF and HOF configurations.
A thorough sensitivity analysis suggests that the co-seismic displacement field at the surface is most
sensitive to the value of the shear modulus (rigidity) [52]. For this parameter, a perturbation of 10 percent can
increase the misfit to coseismic displacements measured by GPS and INSAR by as much as 50 percent. The
shape of the displacement field is also fairly sensitive to the geometric location of the fault, particularly its
depth. The displacement field, however, is relatively insensitive to the geometry of the subcrustal layers.
Similarly, the absolute value of Poisson’s ratio makes little difference. A displacement field calculated with
= 0.25 is almost (??????)extremely similar to one with = 0.28, in agreement with a previous study of the
June 2000 earthquakes [11].
To describe the post-seismic time interval, we introduce viscosity in the lower crust below a depth of Hu,
and in the upper mantle, below a depth of Hc.
Page 8
2728
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
2930
Dubois et al. submitted 2023-05-09
2.3 Modeling poroelastic deformation with an elastic approximation
To explain the post-seismic interferogram (Fig 5), Jónsson et al. invoke poroelastic effects in the first
month following the June 17 mainshock [32]. To describe the observed signal, they employ a modeling
approach involving changes in Poisson’s ratio . this quasi-static approximation, which we will call the Poisson
Recipe in the rest of this study, originally introduced by Rice and Cleary [59] and described in the
Supplementary Material, uses only two elastic coefficients to solve a problem in a two-phase, poroelastic
medium that requires at least two more coefficients to describe completely. This quasi-static model assumes a
change in elastic properties, specifically Poisson’s ratio, from undrained to drained conditions, as for several
other earthquakes in different tectonic settings [60, 61]. The validity of the assumptions underlying
this simple, static and elastic approximation to what is actually a complex, time-dependent,
poroelastic problem [62] will enter into our discussion below. Using all these assumptions, this single-parameter
description approximates the complete post-seismic displacement fields generated by poroelastic rebound as the
difference between two co-seismic fields calculated with different values of Poisson’s ratio. Consequently, only
the change in Poisson’s ratio undrained – drained controls the shape of the displacement field.
The previous study of the 2000 SISZ sequence [32] makes strong assumptions about the spatial
distribution of the Poisson’s ratio change . As a first approximation, it uses a half-space formulation to
calculate the Green’s functions G such that = 0.04 everywhere [63]. As a second approximation, it develops
a different formulation with horizontal layers, such that the drainage occurs only in the uppermost 1.5 km of the
crust, where = 0.08.
Making a third approximation is one of the goals of this study. Here we allow the Poisson’s ratio change
to vary horizontally as well as vertically. Indeed, the finite element method can calculate the elastic Green’s
function G from an arbitrary geometric distribution of change in Poisson’s ratio
In the Supplementary Material, we conclude that = 0.04 is a reasonable average value for the change in
Poisson’s ratio. We note, however, that this parameter trades off with the thickness of the drainage layer.
Accordingly, we experiment with different values and geometric configurations, as listed in Table 2. To
summarize, we consider three categories of poroelastic configurations, depending on spatial distribution of .
In the first category, is uniform (INV1, FOR1 and FOR2). In the second category is a function of depth
only (FOR3, TRY1, TRY2, TRY3, INV2, and INV3). In the third category, varies with depth and horizontal
position (INV4).
2.4 Viscoelastic modeling
To describe post-seismic deformation, we introduce a linear Maxwell viscoelastic rheology in the lower
crust and upper mantle layers of the geometric configurations described above. Varying the viscosity in a series
of forward models, we seek the value that minimizes the misfit to the horizontal components of the
Page 9
Call Fig 5
3132
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
3334
Dubois et al. submitted 2023-05-09
displacement field measured by GPS in the post-seismic interval from the time of the earthquakes through May
2004 [5, 35].
2.5 Stress transfer
Using the slip distributions estimated as described above, we calculate the Coulomb stress on faults in the
SISZ using the same notation and values as Árnadóttir et al. [25]
σC = Δσ t + μ f (Δσ n − B3
Δσ kk ) (1)
where B = 0.5 is Skempton’s coefficient, f = 0.75 is the effective coefficient of friction on the fault, σt is the
increase in shear stress projected onto the fault plane (taken positive in the sense of the slip vector), σn is the
increase in normal stress (taken positive in tension), and σkk is the trace of the stress tensor.
To evaluate the Coulomb failure criterion, we project the stress tensor onto the horizontal strike slip
vector on a north-striking vertical “receiver” fault that typifies the SISZ focal mechanisms [25]. The result is a
scalar Coulomb stress σC. At a given location, if the value of σC exceeds a critical (threshold) value σC(critical),
then we VERB MISSING HEREconclude that the stressing process favors the occurrence of earthquakes there.
If, in fact, an aftershock occurs at the given location, then we count it as a “triggered” event. In this study, we
assume a threshold value of σC(critical) = 10 kPa, equivalent to the stress induced by a column of water 1 m high.
3. RESULTS
3.1 Estimating the distribution of co-seismic fault slip
Using the method described in the Supplementary material, we estimate the distribution of co-seismic
slip for the two mainshocks in a joint inversion of the GPS and INSAR measurements. The shape and depth of
the slip distribution depend on the geometric and rheologic configurations used in the calculation.
Homogenous configurations
Using the homogeneous geometric configuration (HOM), we estimate the distribution of co-seismic slip
on the June 17 and June 21 fault planesFig 3. It is essentially similar to that estimated using the dislocation
formulation [15], further validating our finite-element approach. The largest differences are in the upper-most
part of the fault, where the dislocation and finite-element WORD MISSING HERE: “methods” maybe?methods
parameterize the slip differently, as described above. The uncertainty in the slip estimates reaches 1 m near the
surface, particularly in the section of the June 17 fault south of the hypocenter.. This result may reflect
Page 10
Call Fig 3
3536
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
3738
Dubois et al. submitted 2023-05-09
discrepancies between the INSAR and GPS measurements for points near the fault. Elsewhere on both faults,
however, the standard deviation of the slip is less than 0.5 m. Most of the slip is shallow, above 6 km depth. The
slip decays to less than 0.5 m below 9 km depth for the June 17 fault and 11 km depth for the June 21 fault. The
centroid depths are 3.7 and 3.8 km, respectively. Some of the aftershock hypocenters are located outside the
high-slip areas: between 8 and 10 km depth in the June 17 fault plane and between easting coordinates -8 and -
6 km in the June 21 fault plane.
Layered configurations
The upper two panels of Fig 3 show the slip distribution estimated using the HET configuration. The
difference between the HET and HOM configurations appears in Fig 4. The slip extends slightly deeper in HET
than in HOM. The centroid is 0.3 km deeper for the June 17 fault and 0.5 km deeper for the June 21 fault. For
the June 21 fault, there is over 1 m more slip at 10 km depth in HET than in HOM. The increase in slip
concentrates around the hypocenter relocated at 5.0 km depth from seismology using a layered velocity model
[6]. This difference is statistically significant with over 95 percent confidence because it is more than two times
larger than the standard deviation of the HET estimate (Fig 4). The surface slip is also significantly larger in the
HET configuration than in the HOM estimate. The slip distributions using the ISL configuration are essentially
the same as those found with HET [52].
Configurations with fault damage zones
The HOF configuration adds a weak damage zone around the mainshock faults to the HOM
configuration. The slip estimated in HOF concentrates in the upper crust, significantly increasing the maximum
slip value around the June 21 hypocenter at 5 km depth (Fig 4). The slip is significantly shallower in HOF than
in HOM. Of all the configurations, HOF shows the steepest slip gradient and thus the highest residual stress.
The aftershock locations seem to correlate with these areas of high slip gradient, particularly for the southern
edge of the June 21 rupture. The aftershocks also concentrate at the edges of the slip distribution where the slip
estimates become smaller than their typical 50-cm uncertainties. Adding a damage zone to the HET
configuration to specify the HEF configuration also changes the slip distribution. The HEF result has
significantly more slip in the uppermost mesh elements and around the hypocenters in both fault planes than
does the HET result.
3.2 Modeling postseismic deformation with the elastic approximation for poroelastic effects
Fluids flowing in the fractured and fissured rock around the June 17 fault seem to be the most likely
explanation for the observed time delays: 86 hours between the two mainshocks and the month-long post-
seismic signature observed in the interferogram (Fig 5). To explore this possibility, we consider poroelastic
Page 11
3940
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
4142
Dubois et al. submitted 2023-05-09
effects by extending the work of Jónsson et al. [32] with a series of model calculations summarized in Table 2.
To begin, we reproduce their results by using the homogeneous geometric configuration HOM and a Poisson’s
ratio contrast = 0.04. This model calculation (FOR1) has a variance reduction of σ2 = 39 percent for the
observed one-month interferogram with respect to a null model. Another such calculation, FOR2, concentrates
the drainage in the uppermost 2 km of crust, similar to the layered model of Jónsson et al. [32]. It increases the
Poisson’s contrast to = 0.06 to achieve a variance reduction of 48 percent. Both FOR1 and FOR2 appear to
fit the data well in a profile across strike (A-A’ in Fig 5c), in agreement with the black and green dashed curves,
respectively, in Figure 2 of Jónsson et al. [32]. In map view and a profile parallel to strike (B-B’ in Fig 5b),
however, the deformation calculated with the finite element approach exhibits small features near the fault that
are not present in the dislocation calculation [32]. We interpret these features as a consequence of the increased
surface slip allowed in the finite element formulation. Another attempt, FOR3, using = 0.06 in the HET
configuration, yields a slightly different result (σ2 = 40 percent), highlighting the importance of layering in
poroelastic models.
We also estimate the Poisson’s ratio change in a trial-and-error procedure. The TRY1 configuration
fits the data (σ2 = 44 percent) as well as any other solution we could find by assigning a high value of = 0.10
to the shallow layer between 1 and 2 km. The TRY2 configuration achieves the same reduction with lower
values of The TRY3 solution is equivalent to FOR3 (Table 2).
Next, we allow horizontal variations in the Poisson’s contrast in an attempt to reproduce the
diminished southwest lobe of the post-seismic fringe pattern. This set of geometric configurations includes a
damage zone extending 2-3 km around the fault. The damage zone can be further subdivided into four quadrants
in two layers, adding as many as eight free parameters that we estimate using a non-negative least squares
method [64]. One such attempt, the INV1 solution with only one free parameter, seems reasonable because it
resembles FOR1 in and σ2. The INV2 solution with two free parameters and INV3 solution with five free
parameters, resemble TRY2 and TRY1 respectively (Table 2). An extreme attempt with 8 free parameters,
INV4, achieves a variance reduction of σ2 = 65 percent by forcing the Poisson’s ratio contrast to the physically
unreasonable value of = 0.244 in the southwest quadrant of the damage zone between 2 and 5 km depth.
3.3 Viscoelastic model for post-seismic deformation
Of the parameters in the viscoelastic model, viscosity has the most influence on the post-seismic
displacement field, based on a sensitivity study that varies each parameter individually [52]. By optimizing the
fit to the horizontal components of the GPS measurements, we estimate the viscosities in the lower crust and
upper mantle, albeit with large uncertainties. For the lower crust, we find the 95 percent confidence interval for
viscosity to fall between 2 and 20 EPa.s during the first post-seismic year, and 9 – 90 EPa.s for the interval from
July 2001 to May 2004. (1 exapascal-second = 1018 Pa.s). The different geometric configurations HOM, HET
Page 12
4344
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
4546
Dubois et al. submitted 2023-05-09
and ISL yield similar results. For the upper mantle, the best-fitting viscosity is 1 – 30 EPa.s for the early post-
seismic interval and the ISL configuration. The other configurations yield lower viscosity estimates with wider
confidence intervals, suggesting that using a realistic geometry is particularly important when continuous GPS
observations are sparse. For the upper mantle, a viscosity of 2 – 100 EPa.s, irrespective of geometric
configuration, fits the 2001-2004 data. These values are compatible with a previous study that used a different
modeling approach to describe the same post-seismic observations [35].
3.4 Stress transfer
Next, we evaluate the consequences of various sources of stress in terms of their ability to trigger
earthquakes.
June 17 and 21 co-seismic stress
Does the June 17 mainshock increase the Coulomb stress on a north-striking, vertical strike-slip fault at
the June 21 hypocenter? Yes, the increase is 100 to 150 kPa, in the three layered configurations (HEF, HET and
ISL), confirming the result of a previous study [25] that assumes a homogeneous half-space equivalent to our
HOM configuration. In all three cases, the stress increase calculated at the June 21 hypocenter exceeds the
conventional threshold of 10 kPa. In map view, the area where earthquakes are favored, within the contour of
10 kPa, is somewhat smaller when calculated with the layered HET configuration than with the uniform HOM
configuration (Fig 6a).
Do the two mainshocks increase the Coulomb stress at the location of the aftershocks? Yes, about half the
aftershocks are located within the 10-kPa threshold contour. Thus we conclude that only half the earthquakes
after the June 17 mainshock, including the June 21 mainshock, can be explained by the co-seismic stress
produced by the June 17 mainshock under the Coulomb failure criterion. In other words, only half the seismicity
is triggered by increases in static stress. The other half must be triggered by some other, presumably dynamic,
process, as considered in the following paragraphs.
“Domino effect” of cascading aftershocks
We generalize the Coulomb stress calculation to include all the 14 388 earthquakes between June 17 and
December 31 for which reliable focal mechanisms have been estimated, except the June 21 mainshock. This
calculation determines the time-dependent changes in the Coulomb stress field on vertical, north-striking strike-
slip faults. (To save calculation time, we use the layered configuration HET and the EDGRN/EDCMP
implementation to calculate the Green’s functions [65]). Fig 6c shows the changes in Coulomb stress at 4 km
depth. This stress increase occurs within less than 10 minutes of the June 17 mainshock and remains roughly
constant until the end of 2000. At the June 21 hypocenter, however, the increase is only 5 kPa. The primary
Page 13
4748
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
4950
Dubois et al. submitted 2023-05-09
contributor to the static stress increase is a magnitude 5 aftershock that occurred some 130 seconds after the
June 17 mainshock at 15:42:50 midway between the two mainshock faults [16, 49]. Accounting for this large
event enlarges the area of increased Coulomb stress on the June 21 fault plane.
Modeling stress transfer with the elastic approximation for poroelastic effects
The elastic approximation for modeling poroelastic effects explains the location of only the aftershocks
nearest to the mainshock faults. For example, even the extreme HET configuration with the INV4 values
does not increase the Coulomb failure stress on the June 21 fault (Fig 6b).
Viscoelastic processes
Viscoelastic processes are also capable of triggering seismicity. After four years of post-seismic
relaxation, the Coulomb failure stress at 4 km depth exceeds the threshold of 10 kPa over most of the SISZ (Fig
6d). The threshold contour for the ISL configurationsis less symmetric than the one for the HET configuration,
highlighting the importance of a realistic geometric configuration in stress calculations.
Interseismic strain accumulation
Stress also builds up during the interseismic phase of the earthquake cycle as elastic strain accumulates
between large earthquakes. If the interseismic strain rate is constant in time and uniform in space, but the rigid
upper crust thickens from west to east, then the stress rate field will exhibit a gradient. At a given depth, say
5 km (on the upper/lower crust boundary near the June 21 mainshock hypocenter), the rigidity increases
westward because the stiffer lower crust is shallower there. Consequently, stress will be higher on the west side
of a fault than on the east side, thereby explaining the tendency of seismicity to migrate from east to west across
the SISZ. To quantify this purely geometric effect, we run a simple finite element model to mimic the constant
interseismic strain rate. This elastic interseismic model applies the far-field boundary conditions as 90 cm of
displacement at N110o on the east and west faces of the mesh to mimic 100 years of plate motion. Since the
strain depends on the dimension of the mesh, we cannot calibrate the absolute stress values. To isolate the stress
contributed by the geometric variation in crustal thickness, we run this model twice, once in the HET geometric
configuration with horizontal layers, and then again in the ISL configuration with the crust thickening toward
the east. After projecting the resultant stresses onto North-striking vertical strike-slip faults, we subtract the two
fields to obtain the geometric contribution to the change in Coulomb stress σC. This stress field exhibits a
strong gradient. It is 40 kPa higher on the June 21 fault than on the June 17 fault (Fig 6e). Combined with the
co-seismic stress imposed by the June 17 mainshock alone, this geometric interseismic effect produces a stress
field that is distinctly asymmetric (Fig 6f). Its threshold contour (10 kPa) includes more of the aftershocks
located between 3 and 5 km depth than for the other models.
Page 14
5152
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
5354
Dubois et al. submitted 2023-05-09
4. DISCUSSION AND CONCLUSIONS
4.1 Co-seismic effects
The slip distributions are significantly deeper in realistic three-dimensional geometric configurations than
in the conventional uniform half-space configuration. The slip diminishes to negligible values less than 50 cm
below 12 km depth, supporting the notion that large earthquakes in the SISZ can penetrate into the lower crust
[19]. The layered HET configuration is well suited for calculating coseismic deformation and stress. The
coseismic stress increase generated by the June 17 mainshock can explain the location of the June 21
mainshock, confirming a previous study [25].
The domino effect of cascading aftershocks following the June 17 aftershock also increases the stress on
the (future) location of the June 21 mainshock. Since static stress migrates at the speed of an elastic wave,
standard Coulomb calculations based on the June 17 mainshock or its aftershocks cannot explain the 86-hour
time delay between the two mainshocks.
4.2 Viscoelastic effects
The simple Maxwell viscoelastic model can explain the postseismic deformation between June 2001 and
May 2004 with viscosities of the order of ~10 EPa.s for both the lower crust and the upper mantle. Although the
uncertainties are large “plus or minus a factor” sounds weird, should be “times a factor” (an order of magnitude
at 68 percent confidence), the acceptable ranges are slightly higher and narrower using realistic geometric
configurations. For the first year, the GPS measurements of post-seismic displacements can be fit with a lower
viscosity, of the order of ~1 EPa.s, albeit with large uncertainties. We conclude that the available data provide
only weak constraints on subcrustal viscosity, partly because the poroelastic and viscoelastic effects have similar
amplitude.
4.3 Poroelastic effects
None of the poroelastic models fit the geodetic data very well, despite drastic increases in the complexity
of the models. It seems that the elastic approximation described in the Supplementary Material is insufficient to
describe the post-seismic observations in this case. Even with drastic increases in the number of free parameters,
this approach achieves only modest reductions in variance. Nor does it account for the location of many
aftershocks.
The explanation for this shortcoming probably involves time-dependent effects. The elastic
approximation described above is valid only after a very long time has elapsed and the fluid flow field has
returned to equilibrium (drained) conditions. The elastic approximation does not reproduce the map view of the
post-seismic interferogram very well, particularly in the fringe lobe southwest of the June 17 rupture. There, the
Page 15
5556
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
5758
Dubois et al. submitted 2023-05-09
half-space model with = 0.04 everywhere predicts about two times more uplift than is observed. To explain
this discrepancy, Jónsson et al. [32] suggest that the poroelastic rebound was faster in this area than elsewhere.
Indeed, if the hydrological state had returned to “drained” conditions in the two days following the June 17
mainshock, then the poroelastic signal would not appear in the interferogram beginning June 19. The water level
in the well at Hallstun (HR-19), just south of the June 17 rupture, lends some support to this interpretation. It
recovers faster than other wells, suggesting locally high permeability (and thus a large change in Poisson’s ratio
in this area. Yet our attempts to account for such a local heterogeneity do not fit the observed fringe pattern
much better than the uniform half-space model. We infer that time dependence, rather than spatial
heterogeneity, is contributing to the misfit. We conclude that the conditions for the static elastic approximation
to be a valid are not fulfilled. A complete time-dependent treatment, however, would require describing the
material parameters with more parameters to account for permeability, porosity, etc. [62, 66, 67].
4.4 Migrating seismicity
We have found that a combination of co-seismic and accumulated interseismic stress can explain the
tendency of seismicity to migrate from east to west across the SISZ during a single sequence of earthquakes.
This new result indicates that realistic geometry, including crustal thickening, is important for modeling
interseismic processes.
This mechanism does not, however, explain the 86-hour delay observed between the June 17 and June 21
mainshocks. The interseismic process is quite slow, increasing Coulomb failure stress at a given point VERB
MISSING HERE by only ~1 Pa per day. To explain the delay, we imagine a two-step process like the “coupled
poroelastic effect” that Scholz suggested to explain two successive earthquakes in the Imperial Valley,
California [68]. In the first step, the source shock instantaneously changes the pressure of pore fluids in the rock
around the receiver fault. In the second (slower) step, the hydrological flow field gradually returns to
equilibrium, increasing the pore pressure, unclamping the receiver fault and triggering the second earthquake.
Other examples of this two-step “unclogging” process have been invoked in California [29, 69, 70], Oregon
[71], and elsewhere [72, 73].
Over longer time scales, in the ~100 year intervals between sequences of earthquakes in the SISZ (e.g.,
1912 to 2000), inter- and post-seismic processes seem the most plausible explanation. Indeed, we have shown
that a linear viscoelastic rheology can increase the Coulomb failure stress by 10 kPa at distances longer than
10 km in only 4 years, as to be expected from the characteristic Maxwell time of 1 to 10 years approximated by
the ratio of a viscosity 1 to 10 EPa.s to a rigidity of 30 GPa. Longer time scales presumably involve more
complicated rheologies.
Page 16
5960
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
6162
Dubois et al. submitted 2023-05-09
ACKNOWLEDGMENTS
We thank Charles Williams for generously providing the source code for his revised version of
TECTON, as well as Frank Roth and his colleagues for the EDGRN/EDCMP and PSGRN/PSCMP codes. We
also thank Eric Hetland, Tim Masterlark, Rikke Pedersen, Kristin Vogfjord, Fred Pollitz, Sandra Richwalski,
Pall Einarsson and Herb Wang for helpful discussions. Grimur Bjornsson kindly explained the well data
graciously provided by Iceland Geosurvey and Reykjavik Energy. Kristin Vogfjord and Ragnar Stefansson
provided the relocated earthquake catalog estimated by the Icelandic Meteorological Office as part of the
PREPARED project. This work was financed by EU projects RETINA, PREPARED and FORESIGHT as well
as French national programs GDR STRAINSAR and ANR HYDRO-SEISMO.
SUPPLEMENTARY MATERIAL
SM.1 Testing against other solutions
To evaluate the accuracy of our finite-element modeling procedure, we compare its predictions to a
standard analytic solution for a rectangular dislocation buried in an elastic half-space, as formulated by Okada
[74] and implemented in the RNGCHN code [75]. Using the coseismic slip distribution estimated in a uniform
half space [15], we calculate the displacement field at the surface by both methods. The average difference is
less than 1 mm in all three components, thus validating our finite-element approach [52].
To “CONDITIONING” could be confusing here because in the finite element literature it always refers to
the conditioning of the matrix system to solve, so let us use another word increase our confidence in the
geometrically complex ISL configuration, we test its mesh in the TOM configuration. Like HOM, the TOM
configuration has a uniform rheology: Poisson’s ratio 0.28 and shear modulus 30 GPa throughout
the half-space. Like ISL, the TOM configuration uses a mesh that accounts for dipping layers with variable
thickness. The co-seismic surface displacement fields calculated using the HOM and TOM configurations differ
by less than 1 mm for all three components.
For the post-seismic viscoelastic models, we compare the surface displacement fields calculated using
two different methods. The first is our finite-element approach using the TECTON software, as described above.
The second is a semi-analytic Green’s function approach using the PSGRN/PSCMP software [76]. The average
difference in surface displacement is less than 3 mm for all three components after complete relaxation [52].
SM.2 Estimating co-seismic slip by joint inversion of GPS and INSAR measurements
To estimate the distribution of slip on the June 17 and June 21 fault planes, we use the same data set and
inversion algorithm as Pedersen et al. [15]. In this study, we allow more realistic geometric and rheologic
configurations by using the three-dimensional finite-element formulation to calculate the elastic Green’s
Page 17
6364
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
6566
Dubois et al. submitted 2023-05-09
functions, i.e. the partial derivatives of the measurable quantities (displacements) with respect to the model
parameters (slip values in discrete fault patches). This difference in approach entails a difference in the fault
parameterization. The dislocation formulation specifies the slip as a constant value inside each rectangular
rupture “patch” on the fault plane. The parameterization in the finite-element approach assigns a slip value to
each “split” node. In other words, the slip distribution is discontinuous in the dislocation formulation, but
continuous (piece-wise planar) in the finite-element formulation. As a result, the latter leads to smoother slip
distributions and smaller stress concentrations than the former. The finite element formulation also allows non-
zero slip at the uppermost node located at the air-rock interface, thus avoiding a singularity in the case of an
earthquake with surface rupture.
Another technical improvement involves choosing the appropriate amount of smoothing to optimize the
trade-off between fitting the data and finding a reasonable slip distribution. Here we choose the value of the
smoothing parameter that minimizes the sum of the absolute values of the slip parameters [67].
To evaluate the uncertainty of the slip estimates, we use a bootstrap algorithm [77]. Accordingly, we
resample Nb = 1000 times the data vector randomly according to its (diagonal) covariance. For each realization
of the data vector, we solve the inverse problem to estimate a population of the estimates m̂ for the model
parameters m. Then the vector of the standard deviations of the model parameter estimates is
σ (m) =(m̂i − E(m̂)[ ]
2
i=1
Nb
∑Nb − 1
(2)
where the E operator denotes the expected value.
SM.3 Approximating complete poroelastic relaxation by an elastic solution
Just after the co-seismic rupture, at time t+q, “undrained” conditions apply because the fluid in a
representative elementary volume has not yet reached equilibrium with an external boundary [62]. Presumably,
the same undrained conditions also apply during the brief time interval when a seismic wave propagates through
a representative elementary volume. According to previous studies [32, 60, 78], the Poisson’s ratio under
undrained conditions is
(tq+ ) = ν undrained B 0.31 (3)
At a much later time t∞, the fluids have reached equilibrium with an external boundary, deformation is
controlled by the solid matrix alone, and the rocks can be considered “drained”. Under these conditions,
Poisson’s ratio is [32, 60, 78]
(t∞ ) = ν drained B 0.27 (4)
Page 18
6768
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
6970
Dubois et al. submitted 2023-05-09
The differential post-seismic displacement accumulated in the interval of time between t+q and t∞ is thus
w u(t∞ −u(tq+ G raie
−G uraie⎡⎣ ⎤⎦u (5)
where G is the Green’s function relating co-seismic slip u on the fault plane to displacement u at the free
surface. We assume the geometric configuration and the slip u to be identical in both terms. Furthermore, we
assume that the rigidity is the same under drained and undrained conditions [55, 58, 62].
In the following, we attempt to determine reasonable values for . Under drained conditions, the value
of Poisson’s ratio is [62]
drained = 3ν − α B(1+ ν undrained )3 − 2α B(1+ ν undrained ) (6)
where we set the Biot-Willis parameter a = 0.28 and the Skempton’s coefficient B = 0.5 to find drained = 0.25.
Under undrained conditions, the value of Poisson’s ratio can be calculated from the ratio of P-wave to S-
wave velocities Vp/Vs estimated from seismological studies
uundrained =12
1−1
Vp
Vs
⎛⎝⎜
⎞⎠⎟
2
−1
⎛
⎝
⎜⎜⎜⎜⎜
⎞
⎠
⎟⎟⎟⎟⎟
(7)
In and around the SISZ, the Vp/Vs ratio is typically between 1.70 and 1.85, as measured by seismic tomography
[40, 79]. Bjarnason et al. [54] suggest a value of 0.29 for the undrained value for Poisson’s ratio in the SISZ.
REFERENCES
Page 19
7172
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
7374
Dubois et al. submitted 2023-05-09
Captions
Fig 1. Map showing (above) the main tectonic features of the study area from Árnadóttir et al. [23]. The
Reykjanes Peninsula (RP), the Hengill triple junction (He), the Western Volcanic Zone (WVZ), the South
Iceland Seismic Zone (SISZ), and the Eastern Volcanic Zone (EVZ) are indicated. The light shaded areas are
individual fissure swarms with associated central volcanoes. Mapped surface faults of Holocene age are shown
with black lines [80]. The epicenter locations of earthquakes on June 17, 2000 are shown with black stars. The 17
June main shock is labeled J17. The Reykjanes Peninsula events are 26 s afterwards on the Hvalhnúkur fault (H),
30 s afterwards near Lake Kleifarvatn (K), and 5 minutes afterwards at Núpshlídarháls (N) as located by geodesy
[23] and timed by seismology [6]. Location of the 21 June 2000 main shock is shown with a white star labeled
J21. The location of Reykjavik, the capital of Iceland, is noted. Inset shows a simplified map of the plate
boundary across Iceland, with the location of the main part of the figure indicated by the black rectangle. The
black arrows show the relative motion between the North America and Eurasian plates from NUVEL-1A [2, 3].
Block diagram (below) showing thickening of the crust from east to west across the SISZ as inferred from the
literature [43, 44]. The dimensions of the mesh are 300 x 300 x 100 km (whole box). Color code shows Iceland
(gray), the SISZ (blue rectangle), water (blue), upper crust (green), lower crust (orange), the upper mantle (red),
fault traces (red) for the June 17 (east) and June 21 (west) faults, and the reference point (gray dot) at the center
of the modeled June 21 fault trace (63.99oN, 20.71oW) for the local easting and northing coordinate system.
Fig 2. Geometric configuration of rheologic parameters. A) Cross section showing, as a function of depth, the
following SENTENCE NOT FINISHED HERE.rheologic parameters: density r (thick dashed line), rigidity
(thick solid line), and Poisson’s ratio (thin solid line). The thicknesses of the rheologic layers depend on the
depths Hu and Hc to the bases of the upper and lower crust, respectively. The configurations are composed of: a
uniform half space (HOM and TOM), horizontal layers with constant thickness (HET), dipping layers with
variable thickness (ISL), a half-space with a damage zone (HOF), and horizontal layers with a damage zone
(HEF). (B) Thickness of the upper crust Hu as parameterized in the TOM and ISL configurations. Contour
interval is 1 km. (C) Depth to the interface between lower crust and upper mantle Hc in the TOM and ISL
configuration. Contour interval is 5 km. Solid lines represent traces of modeled faults for June 17 (east) and June
21 (west). Dot indicates the reference point of the local cartographic coordinate system, centered on the June 21
epicenter.
Fig 3. View of fault planes showing the slip distribution estimated for the June 17 main shock (left column) and
June 21 main shock (right column) in the horizontally-layered HET configuration (upper row). White stars
indicate the hypocenters of the mainshocks. Black dots indicate aftershocks within 1 km of each fault and 24
hours of the June 17 and June 21 mainshocks, respectively, as relocated by the Icelandic Meteorological Office,
Page 20
7576
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
7778
Dubois et al. submitted 2023-05-09
as described in the section on data and methods. The lower panels show the 1-σ uncertainty of the slip estimates
for the June 17 (left) and June 21 (right) events.
Fig 4. View of fault planes of the June 17 main shock (left column) and June 20 main shock (right column)
showing the differences in estimated slip distribution for the HET configuration with respect to HOM (top row),
HOF with respect to HOM (middle row), and HEF with respect to HET (bottom row). The black lines denote the
significance ratio, defined as the slip estimate divided by its 1-σ uncertainty. The contour where this ratio equals
2 thus includes the area where the differences are statistically significant at the level of 95% confidence. For each
panel, the white circle indicates the centroid of the first slip distribution in the pair, while the black circle
indicates the centroid of the second, e.g., white circle is the HET centroid in the first row.
Fig 5. Poroelastic models for a postseismic interferogram, showing range change (the component of
displacement along the line of sight from the ground to satellite) in map view as observed (a) and calculated from
the FOR1 configuration (d). Solid black lines outline the co-seismic portion of the fault plane that ruptured
during the June 17 mainshock. Dashed lines indicate locations of profiles. Panel (b) shows negative range change
as a function of distance along east-west profile A-A’ as observed (points), and calculated from various
configurations: FOR3 (black line), TRY1 (green line), INV2 (blue line), and INV4 (red lines). Panel (c) shows
negative range change as a function of distance along north-south profile B-B’ with the same plotting
conventions. All range changes pertain to the first month of the post-seismic interval, from June 19 to July 24.
Negative range change corresponds to uplift; positive range change corresponds to subsidence.
Fig 6. Change in Coulomb failure stress σC as a result of four candidate processes for stress transfer. The
contours outline the areas where the change in Coulomb failure stress σC resolved on north-striking vertical
faults at 4 km depth exceeds the conventional threshold value of σC(critical) =10 kPa. The sources of stress
include: (A) the co-seismic stresses generated by the June 17 mainshock calculated using configurations HOM
(black line) and HET (blue line); (B) poroelastic effects calculated using the elastic approximation and two
different configurations for the change in Poisson’s ratio : INV2 (green line) and INV4 (blue line); (C) the
domino effect of cascading aftershocks, (D) viscoelastic processes accumulated in mid 2004 calculated from the
HET (blue line) and ISL (green line) configurations (E) geometric contribution to 100 years of interseismic strain
accumulation obtained by subtracting the interseismic stress field calculated using the HOM geometric
configuration from that obtained with the ISL configuration; (F) combination of the co-seismic and the geometric
contribution to interseismic stress integrated over 100 years, calculated as the sum of the HET field (A) and panel
(E). Red lines indicate faults for the June 17 (east) and June 21 (west) mainshocks. Dots in panels (A, B, C, E, F)
indicate the epicenters of earthquakes between June 17 and 21, 2000 [16, 49].
Page 21
7980
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
8182
Dubois et al. submitted 2023-05-09
TABLES
Table 1. Geometric and rheologic parameters for configurations used to estimate distributions of co-seismic slip.
Table 2. Parameter values for poroelastic modeling of post-seismic deformation and the variance reduction. In
the FOR and TRY series, the parameter values are input to forward calculations. In the INV series, the
parameters are estimated from the data shown in Fig 5.
LAST PAGE
Page 22
8384
656
657
658
659
660
661
662
8586