Microetching Techniques for Revealing Printed Wiring Board

Dynamics of dyke intrusion in the mid-crust of Iceland

Robert S. White a,⁎, Julian Drew a, Hilary R. Martens a, Janet Key a, Heidi Soosalu a,1, Steinunn S. Jakobsdóttir b

a Bullard Laboratories, University of Cambridge, Madingley Rd, Cambridge CB3 0EZ, UKb Icelandic Meteorological Office, Bústaðavegur 9, 150 Reykjavík, Iceland

a b s t r a c ta r t i c l e i n f o

Article history:

Received 28 October 2010Received in revised form 21 February 2011Accepted 22 February 2011

Editor: P. Shearer

Keywords:

dykeintrusionmicroearthquakesfault-plane solutionmoment tensor solutionIceland

We have captured a remarkable sequence of microearthquakes showing progressive melt intrusion of a dykemoving upward from a sill at 18 km depth in the mid-crust of the northern volcanic rift zone in Iceland. Two-thirds of the earth's crust is created at mid-ocean rifts. Two-thirds of that crust is formed by intrusion andfreezing before it erupts of molten rock generated within the underlying mantle. Here we show seismicityaccompanying melt intrusion from 17.5 to 13.5 km depth along a dyke dipping at 50° in the mid-crust of theIcelandic rift zone. Although the crust at these depths is normally aseismic, high strain rates as melt intrudesgenerate microearthquakes up to magnitude 2.2. Moment tensor solutions show dominantly double-couplefailure, with fault mechanisms sometimes flipping between normal and reverse faulting within minutes in thesame location, but breaking along fault planes with the same orientations. We suggest several possiblereasons for the flipping fault mechanisms: the breakage of solidified plugs of basalt within the dyke itself asmore melt intrudes; intrusion along sub-parallel fractures or dykelet fingers into the local stress field creatednear the tip of a propagating dyke; or movement on small jogs or offsets between adjacent en echelon dykes.Although the faulting is caused ultimately by melt movement, there is no resolvable volumetric component inthemoment tensor solutions. The inferred fault planes frommicroearthquakes align precisely with the overallplane of the dyke delineated by hypocentres. Melt injection occurs in bursts propagating at 2–3 m/min alongchannels c. 0.2 m thick, producing swarms of microearthquakes lasting several hours. Intervening quiescentperiods last tens to hundreds of hours.

© 2011 Elsevier B.V. All rights reserved.

1. Introduction

Melt generated in the mantle beneath rift systems migrates intothe crust where, on average, two-thirds of it freezes as intrusions andone-third is extruded at fissures or volcanic centres (White et al.,1992, 2008). Although there aremany observations of shallowmagmachambers and of seismicity accompanying dyke propagation in theuppermost 5 km of the crust, earthquakes caused by dyke propagationin the deeper crust are rarely recorded. Limited seismic observations,however, do not necessarily preclude the prolonged presence of meltat depth, as melt may reside aseismically in sills, possibly at multiplelevels. Here we report on seismicity exceptionally well recorded by adense array of seismometers, that accompanies the propagation of adyke within the mid-crust of the Icelandic volcanic rift system.

Over a period of one year (March 2007–March 2008), the Icelandicnational seismic network recorded over 10,000 microearthquakesbelow Mount Upptyppingar in the Kverkfjöll volcanic system (Fig. 1).The seismicity was produced by melt injection along a southwarddipping dyke (Jakobsdóttir et al., 2008). Starting from a depth of c.

18 km below sea level, the deep seismicity migrated in both up- anddown-dip directions and also laterally before stopping at c. 13 kmdepth, where the melt apparently froze without eruption.

The brittle–ductile boundary in the vicinity of Mt. Upptyppingar ismarked by the termination of upper crustal seismicity at 6–7 km depth(Fig. 1b; see also depth histogram of seismicity in Fig. 1b of Key et al.,2011). So the earthquakes at 13–18 km depth caused by theUpptyppingar intrusion are well below the brittle zone, in a normallyaseismic region. Nevertheless their appearance is of earthquakesgenerated by brittle failure, with dominant frequencies around 7 Hz.We postulate that high strain rates produced locally by the magmamovement generated the seismicity. An unusual feature of theUpptyppingar seismicity is rapid alternation between normal andreverse fault mechanisms, often in the same region andwithinminutes.These rapid changes are explained best by magma movement.

1.1. Previous reports of seismicity due to melt movement in the

deep crust

Much deep seismicity elsewhere attributed to the flow of magmacomprises long period (LP) events, with typically a narrow spectrumof energy in the 1–3 Hz range, emergentwaveforms and usually only afew tens of events recorded over long periods which may extend toseveral years (e.g., Pitt et al., 2002). The clearest examples of deep LP

Earth and Planetary Science Letters 304 (2011) 300–312

⁎ Corresponding author. Tel.: +44 1223 337187; fax: +44 1223 360779.E-mail address: [email protected] (R.S. White).

1 Present address: Geological Survey of Estonia, Kadaka tee 82, Tallinn 12618, Estonia.

0012-821X/$ – see front matter © 2011 Elsevier B.V. All rights reserved.doi:10.1016/j.epsl.2011.02.038

Contents lists available at ScienceDirect

Earth and Planetary Science Letters

j ourna l homepage: www.e lsev ie r.com/ locate /eps l

Fig. 1. Location map and Upptyppingar seismicity. (a) Map showing epicentres of seismicity from multi-event double-difference locations (events shown in yellow occurred duringthe 6–24th July 2007 dyke injection sequence discussed in this paper and those in green show preceding events from the start of dyke growth in March 2007). Red triangles are sitesof temporary seismometers during July–August 2007, inverted red triangles are permanent stations of the SIL network. The neovolcanic rift segments of Askja and Kverkfjöll areshaded light yellow, ice caps white, and rivers blue. Orange line labelled Hr shows Hrimalda eruptive fissure with a similar orientation to the strike of the Upptyppingar dyke. Insetshows rift zones coloured yellow cutting across Iceland (from Einarsson and Sæmundsson, 1987). (b) East–west cross-section showing earthquakes in the brittle upper crust during2006–2009 (in brown) (Soosalu et al., 2010) and deep events caused by melt injection into the crust during March–July 2007 (same colour coding as in part a).

301R.S. White et al. / Earth and Planetary Science Letters 304 (2011) 300–312

events in similar locations to brittle failure earthquakes are underKilauea on Hawaii (Power et al., 2004;Wright and Klein, 2006), with apoorer constrained example from Las Cañadas caldera on Tenerife(Almendros et al., 2007). We do not find any clear deep LP eventsbeneath Upptyppingar.

The many thousands of seismic events we map have impulsive,high frequency waveforms characteristic of brittle failure. Brittlefailure earthquakes in the deep, ductile part of the crust attributed tomagmatic activity have been reported from only a small number ofplaces, and from rifts only in continental North America (Smith et al.,2004), and Askja (Key et al., 2011; Soosalu et al., 2010), which isadjacent to the Kverkfjöll rift discussed here (Fig. 1).

1.2. Previous reports of flipping fault planes

One of our main results is that faults can flip from normal toreverse in essentially the same locations and sometimes only minutesapart. Although there are few other studies with the coverage of thefocal sphere provided by our dense array, there are sparse indicationsof rapid changes of polarity in fault plane solutions from a few otherplaces. Themost similar to our study is from an intrusion beneath LakeTahoe in Nevada–California in late 2003 (Smith et al., 2004). A swarmof 1611 earthquakes at 29–33 km depth over a 6-month period isinterpreted as caused by melt injection along a 50° dipping plane. Ofthe 24 events with local magnitudes of 1.5–2.2 for which fault planesolutions could be constructed, 13 were reverse faults, 5 were normalfaults and the remaining 6 were undefined (i.e., one of the nodalplanes passed through the vertical axis of the focal sphere) (seesupplementary Fig. S3 in Smith et al., 2004). Earthquakes withdifferent polarities were interleaved in time, though the relativesparseness of earthquakes from which fault plane solutions could beconstructed meant that the reported events with flipped polaritieswere typically separated by a day or more.

Another example of rapid polarity changes is from seismicityproduced by a dyke propagating at 5 km depth in the brittle crustduring the 1983 Kilauea eruption. The coverage of seismometers wasinsufficient to make fault plane solutions, but one set of 9 eventsrelocated to within a few tens of metres of one another shows on fourseparate seismometers at different azimuths and distances that theseismic traces from 4 events were almost the exact inverse of thosefrom the other 5 (see Fig. 10 of Rubin et al., 1998). This is a strongevidence for rapidly flipping fault plane solutions in essentially thesame location.

In the shallow (b10 km depth) crust, and especially close to thefree surface, local stresses produced by dyke inflation interact withregional stresses. This may cause 90° rotation of the P-axes of faultplane solutions (Bonafede and Danesi, 1997; Roman and Cashman,2006), although it is not likely to cause rapid alternations of the faultplane solutions in a given location. There is anecdotal evidence forrapid alternations of the P and T axes of fault plane solutions inartificial hydrofracture simulations, which are always in the shallow,brittle crust, but we have been unable to find published documen-tation of this.

1.3. Geodetic constraints on Upptyppingar magma volume and

dyke geometry

Independent constraints on the geometry of the Upptyppingardyke are available from surface deformation recorded by continuousGPS receivers at three sites c. 25 km away (Jakobsdóttir et al., 2008),from campaign GPS measurements at a further six sites and fromsatellite radar interferometry (Hooper et al., 2009). Modelling thesedeformation data independently of the seismicity constraints suggestsa similar dip, extent and orientation of the dyke to that outlined by theseismicity. Using an elastic model of the subsurface, the dyke isinferred to reach 1 m thickness in the centre, decreasing to less than

0.2 m toward the flanks. The total volume of melt injected into thedyke from the best fitting model is 40–47 Mm³ over a year-longperiod during 2007–2008 (Hooper et al., 2009). This modelling,however, assumed a very deep melt source in the upper mantle withrapid transfer into the dipping dyke in the mid-crust such thatchanges in the volume of the deep melt source did not affect thesurface deformation. We discuss later the likelihood that the inclineddyke is fed by a horizontal sill at about 18 km depth, which wouldhave contracted as melt was extracted. If the collapse of the sillreservoir were to be included in themodelling of the surface deforma-tion, it would require a larger volume of melt in the dyke to producethe same surface deformation.

The total amount of melt involved in this dyke intrusion is notlarge, although it is still approximately half the estimated total meltvolume from the 2010 Eyjafjallajökull eruption (Gudmundsson et al.,2010) which paralysed much of Europe. Nevertheless, the relativelysmall melt volume in the Upptyppingar intrusion caused considerableseismicity and surface deformation and provides valuable insightsinto the way melt is injected into the crust.

2. Data acquisition

We use an array of 27 seismometers (Fig. 1a) recording at 100samples per second to investigate an intense period of seismicity duringJuly 2007. This includes the largest earthquake (local magnitude 2.2)produced during the year-long period of seismic activity. Six of theseismometers (inverted triangles on Fig. 1a) are permanent installationsrun by the Icelandic Meteorological Office (IMO), mostly with LennartzLE-3D/5s sensors on concrete plinths.Wedeployed a temporary array of21 three-component Guralp CMG-6TD broad-band (0.03–50 Hz)seismometers for the period July–August 2007 (upright triangles onFig. 1a). These seismometers were mostly buried beneath the surface,which here comprises either volcanic ash or hyaloclastite, and packed inposition with fine sand to provide good coupling. Together the 27seismometers provide dense coverage around the seismicity.

The Upptyppingar area of Iceland is remote and seismically quiet.There is little vegetation, very few animals and cultural noise isminimal, with no permanent dwellings. There are no rivers near mostof the stations, and the desert conditionsmean that the stations are alldry. We record earthquakes with local magnitudes as small as −1.0from depths of over 17 km. For this study we use 545 events recordedacross the entire array from the period 6–24 July 2007. The b-value(Jakobsdóttir et al., 2008; Scholz, 1968) is 2.1, characteristic ofseismicity caused by volcanic intrusion.

3. Hypocentre locations

We used a new Coalesence Microseismic Mapping (CMM) methodto determine initial hypocentre locations (Drew, 2010). For consistencywith IMO locations from the period preceding the deployment of ourdense array,weuse the sameone-dimensional, isotropic velocityfield asIMO (see Supplementary data Table S1). In practise the velocity varieslittle over the 13–19 km depth interval within which the events lie, andthe locations are tightly clustered, so the use of a one-dimensionalvelocity model rather than a more complex 3D velocity model makeslittle difference to the relative locations of the events.

We refined the automated CMM locations by manually picking andassigning uncertainties to the onset times of the P- and S-wave arrivals.Hypocentral locationswere then recalculated using a double-differencetechnique (Waldhauser, 2001). Estimated relative location uncertain-ties of the hypocentres are less than 60 m (uncertainties defined byHypoDD output, also see Martens et al. (2010) for more details onlocation uncertainties). Hypocentres for the period preceding the July2007 intrusion used data from the sparser IMO network (invertedtriangles, Fig. 1). Relative spatial errors from multi-event double-difference locations using the IMO array alone are estimated as b200 m

302 R.S. White et al. / Earth and Planetary Science Letters 304 (2011) 300–312

(Jakobsdóttir et al., 2008). To allow comparison between our locationsand those from the IMO, we redatumed the IMO locations using directcomparisons of the positions of 300 well located events (Martens et al.,2010).

The most remarkable feature of the seismicity is that hypocentresshallower than 17.5 km (i.e., those we attribute to intrusion along adyke) lie on a single plane dipping at 50° and striking at 074° (Fig. 2a),

with a root-mean-square misfit of only 115 m. Themisfit is comparableto the precision of the hypocentre locations themselves, although theuncertainties of the locations mean that the hypocentres could liewithin a dipping zone c. 100 m thick. The 3D rotating animation [MovieS1 embedded in electronic version, and available as SupplementaryMovie S1 in the print version] shows the precision of the fit of the 6–24July 2007 hypocentres to this single plane.

The intrusion plane as viewed from the dip direction is shown inFig. 2b. The July intrusion occurred along the western margin of thepreceding February–July 2007 intrusions, (green dots, Fig. 2b). Itreoccupies previously injected crust from 17 to 15 km depth during6–21 July, but breaks into new crust at 15–13.5 km depth during 22–24 July. This is particularly clear in the animation [Movie S2 embeddedin electronic version, and available as Supplementary Movie S2 in theprint version].

There are hints from the normal faults at c.18 km depth, which liebelow and off the plane of the dyke, that these are caused by deflationas a feeder sill (broken grey line, Fig. 2a) collapses and feeds its meltinto the dyke. There are also other, less well constrained events at thesame location as this inferred sill (not shown in Fig. 2), but in generalthe melt evacuation generates little seismicity. Following the Julyintrusion, seismicity migrated laterally at 15–16 km depth beforefinally propagating up to about 12 km depth where it ceased a yearafter its onset, as the melt froze without erupting (Jakobsdóttir et al.,2008). So the inclined dyke apparently links two sub-horizontal sillreservoirs, one at c. 18 km depth and one at 15–16 km depth (Fig. 2b).

The seismicity occurs mainly in localised bursts (Fig. 3), whichpropagate at speeds up to 2–3 m/min for time periods on the order ofhours. If the seismicity is caused by melt movement, then this impliesthat themelt itselfmoves in bursts. Based on the pattern of hypocentres,the melt moves along relatively restricted channels up to severalhundred metres in width and sometimes reoccupies the same region.

The total energy released by all the earthquakes summed togetheris considerably less than the energy required to expand the dykethickness by the amount inferred from the surface deformation.Hence, much of the melt movement must be aseismic, presumably asthe melt flows along open channels. Furthermore, much of theseismicity probably does not represent fracturing of the country rockas the dyke tip propagates. We draw this conclusion from observa-tions that the seismicity sometimes jumps to deeper parts of the dykeand occasionally propagates from shallower to deeper depths, whilestill lying within the plane of the dyke (e.g., the seismicity in the latterpart of 6 July and the early hours of 7 July (Fig. 3)). The animation ofhypocentres viewed along the strike of the dyke [Movie S3 embeddedin electronic version, and available as Supplementary Movie S3 in theprint version] shows migration of seismicity both up and down thedyke, despite an overall progression towards shallower depths duringthe three-week period in July 2007. This behaviour has also beenobserved elsewhere in dyke intrusion. For example, in the 1983intrusion at Kilauea, abundant seismicity often occurred well after theleading edge had passed (Klein et al., 1987; Rubin and Gillard, 1998),with seismicity concentrated in a few ‘hot spots’ in the rift.

4. Moment tensor solutions

4.1. Calculation of moment tensors

Moment tensor solutions are well constrained by impulsivearrivals (Fig. 4) and the dense seismometer network (Figs. 1 and 5).After rotation into transverse and radial components, polarity pickswere made manually from event gathers. We used a moment tensorinversion, taking into account both the polarities and amplitudes ofarrivals. Observed amplitudes were corrected for sensitivity of thegeophones, for geometric spreading and for an assumed quality factor,Qp, of 250, typical of igneous rocks at depth. The maximum amplitude

SILL?

(a)

(b)

Looking along dyke strike

Looking at dyke from dip direction

Horizontal distance in strike direction (km)

Horizontal distance in dip direction (km)

SILL?

0 1 2 3 4 5 6 7

0 1 2 3 4 5 6 7

13

14

15

16

17

18

19

13

14

15

16

17

18

19

SILL?

Fig. 2. Hypocentre locations viewed along strike and from the dip direction of the dyke.(a) Vertical cross section looking along strike of dyke during the 6–24 July 2007 periodof melt injection. Dots show earthquakes with fault plane solutions controlled by 12 ormore polarity picks: red are reverse faults and blue are normal faults. Inset shows faultplanes from 229 double couple reverse solutions viewed along strike and projectedonto an equal area projection orientated vertically and orthogonal to the dyke. Brokengrey line shows speculative location of a feeder sill. (b) View looking from the directionof dip of the dyke plane. Colours as in part (a) with additional small green dots showingIcelandic Meteorological Office double-difference relocations of microearthquakes fromthe Icelandic network during the preceding period of melt injection from March 1 toJuly 6 2007. To allow direct comparison, IMO locations are redatumed to our locationsusing direct comparisons of the positions of 300 events in common, which necessitatedmoving the IMO locations 530 m westward, 479 m southward and 86 m shallower(Martens et al., 2010). Inset shows rose diagram of slip vectors from the same 229reverse earthquakes projected onto the plane of the dyke, with motion of the hangingwall shown. See Supplementary Information for animation of sequential earthquakes.


of the arrival at each seismometer was calculated from the seismicdata after bandpass filtering 2–16 Hz.

Although the moment tensor inversion is capable of using trans-verse and radial S-waves as well as P-waves, we used only the verticalcomponents of the direct P-wave arrivals. The S-wave arrivals areconsiderably noisier (since they arrive in the wake of the P-waves),and often have ambiguous polarities. Given the dense array, theinclusion of noisy S-wave data contributed little extra constraint tothe moment tensor solution and may have actually degraded it. Weobtained a more consistent set of moment tensor solutions by usingonly P-wave arrivals, with at least 12 polarity picks, giving 266 faultplane solutions for the July 6–24 period.

The moment tensor solutions for two earthquakes, shown in Fig. 5,are calculated from the seismic data shown in Fig. 4. One is a normalfault and the other is reverse, but they occurred less than 5 minutesapart and within 100 metres of each other (i.e., within the locationuncertainty), during an intense burst of seismicity. The amplitudesand polarities are well fitted by the moment tensor solutions, asshown by comparison of the open bars at the left-hand end of eachseismic trace in Fig. 4, which shows the observed amplitudes, with thesolid coloured bars showing the calculated amplitudes. There are nodiscrepant polarities, and even the small arrivals near nodal planeshave sufficiently good signal to noise ratios that they are fitted well.We discuss in Section 6.2 the likely physical explanations for thesequickly flipping fault plane solutions, which have nearly identicalnodal planes but opposite polarities.

A moment tensor solution can be decomposed into an isotropicand a deviatoric component (Aki and Richards, 1980). The isotropiccomponent represents a volume change, and can be specifieduniquely, while the deviatoric component can be interpreted in avariety of ways. One of the most common is to divide it into a doublecouple component and a compensated linear vector dipole (CLVD,with zero volume change, zero net force, and zero net moment), usingthe same principal axis system for both (Hudson et al., 1989; Knopoff

and Randall, 1970; Vavrycŭk, 2005). In Sections 4.2 and 4.3 we discussdecomposition of our moment tensor solutions.

4.2. Isotropic (volumetric) component of moment tensors

The isotropic component could range from pure explosive (allcompressional arrivals) to pure implosion (all dilatational arrivals).We might expect melt intrusion in a dyke to produce significantvolumetric change which would be evident in the moment tensorsolutions, particularly since the high strain rates that cause seismicityin the otherwise ductile region are caused by melt intrusion.

However, as shown in Fig. 6a, there is no evidence for significantvolumetric change. The inferred percentage volumetric change for allthe calculated moment tensors is distributed symmetrically aboutzero, with a standard deviation of 8%. Moreover, the normal andreverse faults considered separately are equally distributed about zerovolume change. The scatter is likely to be caused by noise and errorsintroduced in the moment tensor inversion, and is not consideredsignificant. Furthermore, systematic volume changes associated withthe faulting would be likely to be biassed toward expansion producedby the injection of fluid. No evidence of such a bias is observed.

Some reported evidence of volumetric changes exists for geothermalearthquakes elsewhere. Miller et al. (1998) report that 25% of 70 eventsthat occurred in shallow crust near Hengill, Iceland could be explainedsolely by double couple shear failure, whereas the remainder showed anaverage andmaximumvolumetric increase of 20% and 42%, respectively.Only one of their events showed a volumetric decrease, giving adistribution highly skewed to volumetric expansion. Similar results arereported from 26 shallow events in Long Valley caldera, California: 21events show a volumetric increase, with a maximum of 32% volumechange (Foulger et al., 1998). They attribute this to the rapid flow ofwater, steam, or carbon dioxide into opening tensile cracks. The brittleconditions present in upper crustal geothermal areas with open cracks,however, is very different to the conditions prevailing at 15 km depth in

Fig. 3. Locations of microearthquakes during the 6–24 July 2007 period of melt injection. Circles show earthquakes with fault plane solutions controlled by 12 or more polarity picks:red are reverse faults and blue are normal faults. Note that the melt moves upward in two stages, first from 17 to 15 km depth and then from 15 to 13.5 km. Eastings and Northingsare in UTM area 28 format.


the ductile mid-crust under Upptyppingar. It is likely, however, thatcarbon dioxide is present in sufficient concentrations in the basaltic meltto start coning out of solution at depths of 15–20 km (Pan et al., 1991),thusproviding apossibledriver of crackpropagation andmicroseismicitynear the dyke tip.

Results perhaps more similar to the circumstances of mid-crustalmelt injection under Upptyppingar come from two hydrofractureexperiments in boreholes at 5–6 km depth where most cracks areclosed by the ambient pressure (Horálek et al., 2010, Vavryčuk et al.,2008). Moment tensor solutions from those boreholes show noconsistent volumetric component and are dominantly double coupleevents caused by shear failure.

4.3. Deviatoric component of moment tensors

The deviatoric component of the moment tensor solution can beconsidered as a combination of a double couple and a compensatedlinear-vector dipole (CLVD). In Fig. 6b we show a histogram of thedeviation from a double couple solution (Hudson et al., 1989). Themean deviation considering all our moment tensors inversions (0.8%)is extremely close to zero. The standard deviation about the mean is30%, with scatter distributed equally between positive and negative.Independent results for normal (blue, Fig. 6b) and reverse faults (red,Fig. 6b) are not significantly different. We conclude that the seismicevents are double couple, with failure in shear. The scatter in apparentdeviation from double couple probably arises mainly from noise thatinduces variations in signal amplitude, particularly near the nodal

planes where the signal is small. Other well-known effects that couldcause deviations from double couple solutions include anisotropy andinhomogeneities in the velocity structure near the source, variationsin propagation effects and near-source structure, and changes to thelocal rheology during faulting.

In Fig. 6c we use a source-type plot (Hudson et al., 1989) to showthe distribution of moment tensor solutions. Pure double couplesolutions plot in the centre at (0, 0). Pure tensile opening crackswouldplot at the edge of the top left quadrant, and pure tensile closingcracks at the edge of the lower right quadrant (Fig. 6c). The mean ofthe moment tensor solutions is nearly perfectly double couple at(0.00, −0.01), shown as a cross with error bars on Fig. 6c. In theremainder of this paper we therefore use the double couple com-ponents derived from the moment tensor solutions (Strelitz, 1989) inour analysis.

5. Double couple fault plane solutions

In Section 5.1 we discuss the alignment of the pressure (P) andtension (T) axes. Reverse faults are three times more common thannormal faults. The choice of which of the two nodal planes is the faultand which the auxiliary plane is a matter of interpretation, discussedin Section 5.2.

In general the fault plane solutions of the normal faults are slightlyless well constrained than are those from the reverse faults, probablybecause on average the normal faults are slightly smaller in magnitude(mean 1.2±0.2) than the reverse faults (mean 1.4±0.3). Hence the

(a) Normal fault: 2007July 6th 20:47 (b) Reverse fault: 2007 July 6th 20:52

Fig. 4. Seismic traces from normal and reverse faults. (a) Panel showing seismic data recorded by the vertical components of seismometers in our array from a local magnitude 1.3normal earthquake at 20:47 GMT on 6 July 2007. Traces are aligned with picked arrival at 0.0 s. The fault plane solution from this earthquake is shown in Fig. 5a. (b) Panel showingseismic data recorded by the vertical components of seismometers in our array from a local magnitude 1.4 reverse earthquake at 20:52 GMT on 6 July 2007. The fault plane solutionfrom this earthquake is shown in Fig. 5b. Seismometer locations are shown in Fig. 1a, and are displayed in latitude order with themost northerly at the top. Traces coloured red have acompressional first arrival P-wave, while those in blue have a dilatational first arrival. Those omitted have sufficiently poor signal to noise ratio that no polarity has been assigned,and they have not been used in the moment tensor solution. The seismometer at HELI, for example, is adjacent to a campsite and was very noisy during this period, so has not beenused in this fault plane solution. All seismic traces have been band-pass filtered between 2 and 16 Hz and have been normalised to allow comparison. The open box at the left-handend of the trace shows the true maximum amplitude for that trace. The coloured bar at the left hand end shows the theoretical amplitude and polarity from the moment tensorsolution. Note that there is good general agreement between the observed and theoretical amplitudes, and that the nodal planes (small amplitudes) are well displayed.


normal faults have a lower signal to noise ratio, allowing noise to play abigger part in causing scatter of the solutions.

5.1. Pressure (P) and tension (T) axes

P and T axes from all the fault plane solutions cluster tightly andswap orientation almost exactly between normal and reverse faults(Fig. 7a and b). Both the T axes of normal faults and the P axes ofreverse faults have an average dip close to zero. The mean directionsof the T axes of normal faults (167°±30°) and the P axes of reversefaults (168°±25°) are identical. Strikingly, this direction is orthog-onal to the strike of the dyke (074°), and is unrelated to the regionalextension direction of 106° caused by plate spreading (DeMets et al.,1994). We conclude that the fault orientations of the micro-earthquakes are governed by local stress changes caused by the dyke.

5.2. Fault planes

Poles to the nodal planes of each earthquake divide into twodistinct fields for both the normal and reverse faults. For each faulttype, there is one tightly clustered northward dipping group of poleswhich is perpendicular to the plane of the dyke inferred from thehypocentres (Fig. 7c and d), and one southward dipping group ofpoles which is dispersed in the plane of the dyke (Fig. 7e and f). Thereare two possible choices for which of the nodal planes is the faultplane. We demonstrate how we choose which nodal plane is the fault

plane in Fig. 8 using data from the reverse faults: identical argumentsapply for the normal faults.

First, if we assume that the southward dipping planes, shown inFig. 8a as great circles, are the fault planes, then as shownschematically in Fig. 8c, the individual fault planes align closely withthe macroscopic plane of the dyke. The slip vectors for each event arethen the poles to the auxiliary planes shown in Fig. 8b (these poleswould be those shown in Fig. 7f). These slip vectors lie within theplane of the dyke, but vary considerably in direction, so are not all in adip-slip direction. On the other hand, if we were to assume that thenorthward dipping planes shown in Fig. 8b are the fault planes, thenthe slip vectors would be the poles to what would then be theauxiliary planes in Fig. 8a. These slip vectors would be very tightlyaligned (they would be the poles shown in Fig. 7d), and lie orthogonalto the plane of the dyke as shown by the arrows in Fig. 8d. Although itmight be thought attractive that the closing (for reverse faults) oropening (for normal faults) direction is orthogonal to the dyke, aconsequence of this choice of fault planes would be that the actualfault breaks would all be northward dipping with a wide range ofstrikes. Every event (and there are 10,000 recorded) would then breaka new section of crust at a different angle, as shown schematically inFig. 8d. Dykes exhumed at the surface do not show such high anglepervasive fracturing of the host rock. On the contrary, dykes emplacedin igneous bedrock often show abundant sub-parallel fracturingadjacent to the dyke (Kavanagh and Sparks, 2011). We thereforeidentify the tightly clustered south-dipping planes (Fig. 8a) as thefault planes. The neutral axes, defined as the mutual trace of each pair

Fig. 5. Moment tensor and double couple solutions for normal and reverse faults shown in Fig. 4. The moment tensor solutions at the top were calculated from the polarities andamplitudes of seismic data shown in Fig. 4. The double couple fault plane solutions at the bottom are derived from the moment tensor solutions, with nodal planes shown as solidlines. Broken red line shows the plane of the dyke defined by hypocentre locations (074/50°). Circles show negative polarity arrivals, crosses show positive polarities, locations ofseismometers shown in Fig. 1. P and T mark positions of pressure and tension axes. Lower hemisphere equal area projection.


of focal planes, lie at the intersections of the planes shown in Fig. 8aand b. These are distributed around the interpreted mean fault plane,thus adding support to this interpretation of the south-dipping planesas the fault planes.

The fault plane orientations are remarkably similar to the macro-scopic orientation of the dyke derived from the hypocentral locations.This is perhaps even clearer on the projection of fault planes shown inthe inset of Fig. 2a by comparison with the dip of the dyke defined byhypocentral locations. The reverse faults have amean strike and dip of072/49° (red great circle, Fig. 7d), indistinguishable from the overall

dyke orientation of 074/50°. The normal fault planes are slightly morevariable (Fig. 7c), but the average strike and dip are still consistentwith the dyke orientation. The slip vectors of the reverse faultsprojected onto the plane of the dyke show that motion during theseismic events is preferentially in one of two directions oblique to thedip direction (rose diagram, Fig. 2b inset), rather than being in adirectly up-dip direction. There are two dominant directions. One is atabout 30° to the right of the directly up-dip direction, which coincideswith the orientation of the melt channels, shown for example by thetrend of the red and blue hypocentres between 15.5 and 17 km depthon Fig. 2b and on animation S2. The other direction is at about 30° tothe left of the directly up-dip direction, which we interpret asrepresenting fracturing into new rock away from the edge of thepreceding dyke intrusion as delineated by green dots on Fig. 2b.

In Fig. 7 g and h we show rose diagrams of the P and T axesprojected onto a vertical plane perpendicular to the plane of the dyke.Although the P and T axes are directions of maximum instantaneousstrain, they apparently also provide a good representation of theprincipal stress directions for these small faults. The reverse faults inparticular, with their well constrained fault plane solutions, showtextbook distributions of the P and T axes with respect to the averagefault plane parallel to the dyke.

5.3. Dynamics of fracturing and magma flow

If, as we infer from the fault plane solutions, the fracture planes aresub-parallel to the macroscopic plane of the dyke, much of theseismicity is likely to be caused by fractures re-breaking regionswhereearliermelt has solidified against the dykewalls. Thesewould beweakhorizons, particularly where magma is re-intruding a melt channelafter a period of a fewweeks, as is apparently the case for the first partof the July 2007 intrusion episode (Section 3 and Fig. 2b). Theseismicity often occurs deeper than the tip (seeMovie S2), presumablymarking places where constriction of the flow is causing fracturing asmelt forces its way along the channel. Some of the faultingmay also be

(a)

(b)

(c)

Fig. 6. Isotropic and deviatoric components from moment tensor solutions. (a) Theisotropic volumetric component (i.e., explosion or contraction) calculated from themoment tensor solutions for all the earthquakes discussed here, expressed as apercentage. The averages of the normal and reverse faults considered separately ortogether are all are close to zero, with the results symmetrically distributed around zero.The scatter of ±8% (1−σ) is characteristic of the influence of random noise on themoment tensor solutions, so we conclude that there is no evidence for a systematicisotropic volumetric change in themicroearthquakes. Means and standard deviations ofvolumetric changes are as follows: 3±10% from 67 normal faults; −3±7% from 205reverse faults; and −1±8% from all 272 faults considered together. (b) The deviationfrom pure double couple faulting in the deviatoric component for all the earthquakesdiscussed here, expressed as a percentage (the parameter T as defined by Hudson et al.,1989): 0% is pure double couple,−100% and+100% are pure compensated linear vectordipole (CLVD) components. The averages of the normal and reverse faults consideredseparately or together are all are close to zero,with the results symmetrically distributedaround zero. The scatter of ±30% (1−σ) is characteristic of the influence of randomnoise on the moment tensor solutions, so we conclude that there is no evidence for asystematic non-double couple component in the microearthquakes. Meansand standard deviations of the deviation from double couple solutions are as follows:9±34% from 67 normal faults;−3±26% from 205 reverse faults; and 0±29% from all272 faults considered together. (c) Distribution of isotropic and deviatoric componentsof all the moment tensor solutions displayed on a Hudson et al. (1989) source type plot:blue crosses are from normal faults, red crosses from reverse faults. The vertical axis isthe volumetric component defined by the Hudson et al. (1989) parameter ‘k’ and thehorizontal axis is the deviation from a double couple source mechanism at zero volumechange, defined by the Hudson et al. (1989) parameter ‘tau’. A double couple (DC)mechanism plots at the centre (k, tau=0, 0); pure explosive or implosive sources at thetop and bottom apices, respectively; and tensile opening or closing faults at the pointsmarked on the upper left and lower right edges, respectively. Black cross shows theaverage of all the source mechanisms, with 1−σ error bars: this plots at (0.00,−0.01).We infer that the best inference is that all the faults are double couple, with the scatterdue to random noise in the moment tensor inversion.


(a) (b)

(c) (d)

(e) (f)

(g) (h)

Fig. 7. Earthquake fault plane solutions. (a) Lower hemisphere equal area projection of P (filled red circles) and T (open blue circles) axes from all normal faults during 6–24 July. Taxes are contoured by percentage of points contained within a cone subtending a 1% surface area using scale shown in box between diagrams (same contour colour scale applies topanels a–d). (b) Same for all reverse faults during 6–24 July. (c) Lower hemisphere equal area projection of poles to fault planes from all normal faults, contoured using scale shownin box between diagrams, together with average fault plane in red. Strike/dip of mean fault plane from these poles is 070/58°, mean resultant vector length (Mardia, 1972) is 0.90.(d) Same for all reverse faults. Strike/dip of mean fault plane is 072/49°, mean resultant vector length 0.94. (e) Lower hemisphere equal area projection of poles to auxiliary planesfrom all normal faults, (i.e., the slip vectors) together with girdle fit to poles shown in red. (f) Same for all reverse faults. (g) Equal area projection onto vertical plane orthogonal tothe dyke showing mean fault plane (red), and orientations of P axes (red) and T axes (blue) for all normal faults. (h) Same for all reverse faults.


caused by melt intrusion or faulting between en echelon dykes,multiple sub-parallel dykes, or jogs in dykes (Weinberger et al., 2000).

Faulting at dyke tips as they propagate has often been assumed tocause tensional failure (Lister and Kerr, 1991; Maccaferri et al., 2010;Roman et al., 2004), or failure at a high angle to the propagating crack.In contrast, we find that reverse faults are dominant and there is noevidence for faulting at a high angle to the dyke. Reverse faulting hasalso been reported from several other dykes. In dykes mapped inIceland, Khodayar and Einarsson (2004) show that reverse faultingcan occur both as a result of uplift above the dyke tip and in placeswhere there are local bends and irregularities in the shape of the dyke.Fault plane solutions from seismicity accompanying deep intrusionsbeneath Eyjafjallajökull in Iceland exhibit predominantly reversemechanisms (Dahm and Brandsdottir, 1997), as do earthquakes fromthe Lake Tahoe deep intrusion (Smith et al., 2004). As we discuss inSection 6, propagation of the tip of an inclined dyke can cause bothreverse and normal senses of faulting near the tip, due to theasymmetric stress field above and below the dyke tip. Away from thedyke tip, both normal and reverse faults can be caused by melt re-breaking regions where plugs of melt have solidified within the dyke.

We can put some broad constraints on the melt flow rate and thethickness of the melt channels from our estimates of the seismicpropagation rate, v, of up to 2–3 m/min. We assume that, at leastlocally, the seismic propagation rate represents the melt propagationrate. Similar estimates of melt propagation speeds at depth are

reported in Klein (1982) based on the 1982 Kilauea eruption, wherethe seismicity migrated at about 1000 m per day (~1 m/min).

The volumetric flow rate, Q, is related to the half thickness, h, of themelt channel by

v = Q = 2 h:

Following Lister and Kerr (1991) and Taisne and Jaupart (2009)wecan derive another relation between Q and h by balancing thebuoyancy forces derived from the density difference, Δρ, the force ofgravity, g, and the dip of the dyke, θ, against the drag caused by theviscosity, μ, of the intruding magma:

h = 3μQð Þ= 2Δρ g sinθð Þ½ �1=3

:

Solving for both equations and using values for basaltic melt inTable 1, with a propagation rate of 2–3 m/min, we deduce that themelt channel has a thickness of 0.15–0.20 m. Since the seismicity islikely to occur in places where the flow is restricted, it is possible thatthere is also aseismic flow of melt in other, thicker channels.

The along-strike width of the dykes, estimated from the lateralspread of seismicity during any given melt intrusion episode, isseveral hundred metres. The observed propagation rates and inferredchannel thicknesses along a 400 m wide channel gives a volumetricflow rate of 2–4 m³/s, which is 2–3 times the year-long average flow

(a) (b)

(c) (d)

Fig. 8. Interpretationsofpossible faults inferred fromeachof thenodal planes. (a) and (b) showthe twopopulationsof stackednodal planes fromall the reverse faults on lowerhemispherestereographic plots. (c) is the schematic diagram looking along strike showing the orientations of fault breaks (thin black lines) if the stacked nodal planes in panel (a) represent the faultplanes. Red line is the trace of the dyke calculated from hypocentres and green arrows show slip vectors in the plane of the dyke. (d) is the schematic diagram for alternative explanationthat nodal planes in (b) represent the fault breaks. Faults are nowat a high angle to the dyke and slip vectors are orthogonal to theplane of thedyke.We interpret (a) as the fault planes and(b) as the auxiliary planes.


rate in the dyke, deduced from the total inferred volume. This isconsistent with melt intrusion occurring primarily in short-livedbursts (Fig. 3).

A further approximate calculation is the time taken for the meltchannel to freeze by conduction. Following Jaeger (1968), and assumingthat melt is injected at a temperature near its melting point, then thetime, t, to completely solidify a melt channel containing stationarymeltis given by:

t = h2= 4 λ

2κ

� �

where λ is calculated from an error function and κ is the thermaldiffusivity. Assuming that the thermal properties of the melt and solidare identical, since the crust is formed bymultiple basaltic injections, a0.15–0.20 m thick basaltic melt layer, intruded at 500 °C above theambient temperature of the country rock, could solidify in about anhour. If the melt was only 250 °C hotter than the country rock, thenthe time to freeze the melt channel is increased to 1.5–2.5 h. Even adyke 1 m thick would take only 1.5–3 days to freeze if the melt wasstationary. These are very short periods, though they would increasesomewhat if the melt was intruded well above its melting temper-ature, if previous melt injections have preheated the country rocklocally, or if the melt is continually moving. Nevertheless, it isconsistent with the short periods of a few hours over which the burstsof seismicity occur that we attribute to melt movement (Fig. 3).

Another consequence of the short time taken to freeze the melt,and the fact that the timescale of conductive cooling scales as thesquare of the dyke thickness, is that small variations in dyke thicknesscan cause rapid changes in solidification rates. Underground mappingof kimberlite dykes intruded into dolerite 2–3 km below the surfaceshows variations in dyke thickness, both up-dip and along-strike,from 0.2 to 1.6 m on length scales of a few hundredmetres (Kavanaghand Sparks, 2011). These variations in dyke thickness are thereforelikely to lead to marked changes in flow rates and local solidificationrates as the channel thickness varies. This could be an important factorin creating localised ‘hot spots’ of recurrent seismic activity.

6. Discussion

6.1. Dyke orientation

Although it is often assumed that dykes carrying melt through thelower and middle crusts are likely to be sub-vertical, in Upptyppingarnot only does the dyke dip at 50°, but it is also 60° oblique to theregional strike of the neovolcanic zone rift axis and its fabric of faultsand fissures. Why should this be? There are few reports anywhere ofmicroseismicity in the normally ductile part of the crust caused bymelt movement, but in one of the few other documented cases,beneath Lake Tahoe, seismicity at 29–33 km depth on a plane with an

identical dip of 50° is attributed to melt intruding a dipping dyke(Smith et al., 2004). Recent tank experiments and numericalmodelling have shown that once a dyke has started to open with aparticular dip, it is energetically easier in a homogenous elasticmedium to extend by crack failure along the same dip (Maccaferriet al., 2010). This may explain the remarkably planar nature of thedyke. Whatever the reason for starting intrusion at a high angle, thissuggests that it is likely to continue at that angle until it reaches achange in the properties of the rock being intruded.

The reason for the highly oblique orientation of the dyke withrespect to the rift axis is uncertain and is beyond the scope of thispaper. We speculate that it might be due to rapid elevation of thelandmass to the south, where melting of the Vatnajökull ice cap iscausing unloading and isostatic rebound at a rate measured by GPS ofup to 25 mm/a (Arnadóttir et al., 2009). This rebound rate is of thesame order as the N106°E–N74°W directed extension rate of 20 mm/acaused by plate motions; hence, it may cause the minimum stressdirection to rotate. Circumstances are now similar to those when theIcelandic ice caps melted 10,000 yr ago, although the total isostaticunloading at that time was even greater. Some local eruptive fissuresexhibit highly oblique orientations similar to that of the Upptyppingardyke, such as the 5 km long east–west trending Hrimalda fissure(labelled Hr in Fig. 1) in the Askja rift segment. Hence, the presentdyke may be reoccupying a mid-crustal structure formed at the end ofthe last ice age with the same orientation. Imposition of a similarstress field as the Vatnajökull ice cap melts at the present day makes itfavourable for the melt to intrude at a similar oblique orientation.

6.2. Flipping normal–reverse faults

Melt intruding along the dyke plane, even in small quantities,creates weak, low-friction interfaces that can be moved readily insmall faults. The coincidence of individual fault plane orientationswith the macroscopic orientation of the dyke suggests that this is amajor mechanism for generating microearthquakes. For magnitude1.0–1.5 earthquakes, scaling laws suggest that with a displacement tolength ratio of 10−4

–10−5, the displacement would be a few milli-metres over a distance of a few tens of metres (Kim and Sanderson,2005; Rubin and Gillard 1998). The occurrence of reverse faults as theinclined dyke extends is aided by the collapse of the feeder sillreservoir, which lies directly beneath it (Fig. 2a). A mass balance overthe volume of the sill and the dyke suggests that the sill collapses byabout 2 m as the dyke inflates, thus providing accommodation spacethat would favour the dominant reverse faulting.

The rapid alternation in fault types from normal to reverse, butwith the same orientation of P and T axes (albeit swapped) is a newobservation, although the modification of the stress field by meltintrusion and resultant rotations in earthquake fault planes have beendocumented from the brittle upper crust beneath other volcanoes(Roman et al., 2004). In our data, sometimes the fault polarity flipswithin a fewminutes and at the same locationwithin our resolution ofc. 70 m. On other occasions there are clusters of ten or moreconsecutive normal faults in essentially the same place.

We make four speculative suggestions to explain the flipping faultplanes. The first is that they occur on fractures parallel to the dykeformed either near the tip of the propagating dyke, or nearconstrictions in the dyke (Fig. 9a). The stress field near the magmatictip of a dyke can generate closely spaced dyke-parallel joints (Delaneyet al., 1986; Rubin, 1993). As the dyke propagates this may generate amoving ‘process zone’with associatedmicroearthquakes ahead of andadjacent to the dyke tip, or the fractures may be reactivated later asmelt advances into and fills out the dyke tip. The stress pattern oneither side of the change in dyke thickness behind the dyke tip isstrongly asymmetric for a 50° dipping dyke, thus providing conditionssuitable for creating closely spaced faults of opposite polarity (Fig. 9a).

Table 1

Parameters used for modelling of magma flow and cooling.

Description Symbol Value

Flow rate in melt channel per unit width Q 5–9×10−3m³/s/mHalf thickness of melt channel h 0.07–0.09 mMelt velocity v 0.033–0.05 m/sTime to solidify melt channel t 2.6–4.3×103sDensity difference (country rock–melt) Δρ 200 kg/m³Force of gravity g 9.81 m/s²Viscosity of magma μ 100 Pa sThermal diffusivity of magma and country rock κ 0.71×10−6m²/sLatent heat of magma L 420 kJ/kgSpecific heat of magma c 1.0 kJ/kg/°CMagma temperature 1250 °CInitial country rock temperature 750 °CDip of dyke θ 50°


A second possible mechanism is that the microearthquakes arecaused by fracture of plugs of solidified melt. When the plug breaksand moves, it creates a normal fault if the upper boundary breaks(Fig. 9b), or a reverse fault if the bottom boundary breaks (Fig. 9c). Itmight be argued that the cross-sectional area of the solidified melt istoo small for the liquid melt beneath it to exert sufficient force tobreak it free. However, even tiny amounts of melt percolating into theputative fault break would weaken it hugely so that only smalloverpressures might be required. A further argument against thismechanism might be that it is difficult to imagine moving a solidifiedplug with lateral dimensions of tens of metres as a coherent block.However, it is possible that the displacement–length scaling ratio of10−4

–10−5 derived from observations of typical tectonic earthquakeswould not apply in the same way here. The amount of energy radiatedby a fault break is proportional to the product of the fault area and theslip, so in the case of the dyke, a similar sized earthquake would beproduced by slip of a few tens of millimetres on a fault with dimen-sions of a fewmetres. This is entirely reasonable in the context of meltre-opening an existing dyke channel.

A third possibility to explain flipping fault mechanisms is thatdykes often occur in sub-parallel swarms or en echelon segments withsimilar overall orientations, although locally variable. Since the

intrusions are in the ductile crust, modification of the local elasticstress field created by the previous intrusion will dominate theorientation of the stresses experienced by a new intrusion. There is anasymmetry in the sense of shear that will be dominant depending onwhether the new intrusion is above or below the previous one, whichcould give rise to opposite fault mechanisms that are closely spacedboth spatially and temporally. Closely related to this possibility is afourth scenario, which is that jogs or offset in en echelon dyke seg-ments may create fractures in the country between them which canhave either normal or reverse senses of motion in closely similarlocations (Weinberger et al., 2000).

6.3. Building the igneous crust with dykes and sills

The pattern of seismicity suggests that the igneous crust is builtfrom sills that pond at multiple depths and are fed by relatively short-lived dyke intrusions between them, consistent with inferences frompetrological estimates of intermediate depths of crystallisation inerupted magmas (Kelley and Barton, 2008; Maclennan et al., 2001).Beneath Upptyppingar we have captured melt moving from a sill at c.18 km depth to another at c. 15–16 km. The melt eventually froze inthe subsurface. In general, the minority of melt that sometimes erupts

(a)

(b) (c)

Fig. 9. Schematic diagrams of possible dyke-parallel fault mechanisms. (a) Faulting can occur ahead of dyke tip, and movements in opposite senses can occur on fault-parallelfractures in the country rock on either side of the leading edge of a dipping dyke, or near locations where there are abrupt changes in the dyke width. (b) Normal faulting within thedyke may occur when a plug of solidified melt is broken from the upper chilled margin of the dyke and moved by the pressure of magma from below. (c) Reverse faulting may occurwhen the lower chilled margin of a plug of solidified melt in a dyke is broken.


has probably passed through several episodes of ponding anddifferentiation in the crust before it is extruded at the surface.Seismicity in this normally ductile part of the mid-crust is caused byhigh strain rates resulting from rapid bursts of melt movement.However, it is likely that much of the magmatic flow, and indeedmuch of the inflation of the dyke, is aseismic. We draw this conclusionbased on the moment tensor solutions, which can be explained bypurely double couple failure with no volumetric component. Theoccurrence of flipping normal–reverse faulting with the P and T axesswapped points to local control of the stress field by melt movementand elastic deformation of the country rock.

Supplementarymaterials related to this article can be found onlineat doi:10.1016/j.epsl.2011.02.038.

Acknowledgements

Seismometers were borrowed from the Natural EnvironmentResearch Council SEIS-UK (loan 842), who also archive the data. Wethank M. Coffin, J. Eccles, D. Hawthorn, J.-C. Molina Santana and A.Nowacki for the fieldwork assistance, H. Tuffen, J. Maclennan and M.Edmonds for the discussions of intrusion processes, K. Vogfjörd and R.Slunga for the information on the velocity field used by IMO forlocations, and J. Kavanagh for a preprint of her work on fracturing neardykes and several referees including A. Agnon for their comments onan earlier draft. Dept. Earth Sciences, Cambridge contribution numberESC1979.

References

Aki, K., Richards, P.G., 1980. Quantitative Seismology. : Theory and Methods, vol. 1. W. H.Freeman and Company, San Francisco. ISBN 0-7167-1058-7, 557 pp.

Almendros, J., Ibáñez, J., Carmona, E., Zandomeneghi, D., 2007. Array analyses ofvolcanic earthquakes and tremor recorded at Las Cañadas caldera (Tenerife Island,Spain) during the 2004 seismic activation of Teide volcano. J. Volcanol. Geotherm.Res. 160, 285–299.

Arnadóttir, T., Lund, B., Jiang,W., Geirsson, H., Björnsson, H., Einarsson, P., Sigurdsson, T.,2009. Glacial rebound and plate spreading: results from the first countrywide GPSobservations in Iceland. Geophys. J. Int. 177, 691–716. doi:10.1111/j.1365-246X.2008.04059.x.

Bonafede, M., Danesi, S., 1997. Near-field modifications of stress induced by dykeinjection at shallow depth. Geophys. J. Int. 130, 435–448.

Dahm, T., Brandsdottir, B., 1997. Moment tensors of microearthquakes from theEyjafjallajökull volcano in South Iceland. Geophys. J. Int. 130, 183–192.

Delaney, P.T., Pollard, D.D., Ziony, J.I., McKee, E.H., 1986. Field relations between dikesand joints: emplacement processes and paleostress analysis. J. Geophys. Res. 91,4920–4938.

DeMets, C., Gordon, R.G., Argus, D.F., Stein, S., 1994. Effect of recent revisions to thegeomagnetic reversal time scale on estimates of current plate motions. Geophys.Res. Lett. 21, 2191–2194.

Drew, J., 2010. Coalescence microseismic mapping: an imaging method for thedetection and location of seismic events. PhD dissertation, University of Cambridge.

Einarsson, P., Sæmundsson, K., 1987. Earthquake epicenters 1982–1985 and volcanicsystems in Iceland. In: Sigfússon, Þ.I. (Ed.), Íhlutarins eðli, Map accompanyingFestschrift for Þorbjörn Sigurgeirsson. Menningarsjóður, Reykjavík.

Foulger, G.R., Julian, B.R., Hill, D.P., Pitt, A.M., Malin, P.E., Shalev, E., 1998. Non-double-couple microearthquakes at Long Valley caldera, California, provide evidence forhydraulic fracturing. J. Volcanol. Geotherm. Res. 132, 45–71.

Gudmundsson, M.T., Pedersen, R., Vogfjörd, K., Thorbjarnardóttir, G., Jakobsdóttir, S.S.,Roberts, M.J., 2010. Eruptions of Eyjafjallajökull volcano, Iceland. EOS 91, 190–191.

Hooper, A., Ófeigsson, B., Sigmundsson, F., Geirsson, H., Einarsson, P., Sturkell, E., 2009. Ismagma generated due to retreating ice caps, likely to erupt? Constraints on lower-crustal stress in IcelandfromInSAR.ProceedingsFRINGEWorkshop,Frascati,2009.

Horálek, J., Jechumtálová, Z., Dorbath, L., Šílený, J., 2010. Source mechanisms of micro-earthquakes induced in a fluid injection experiment at the HDR site Soultz-sous-Forêts (Alsace) in 2003 and their temporal and spatial variations. Geophys. J. Int.181, 1547–1565. doi:10.1111/j.1365-246X.2010.04506.x.

Hudson, J.A., Pearce, R.G., Rogers, R.M., 1989. Source type plot for inversion of themoment tensor. J. Geophys. Res. 94, 765–774.

Jaeger, J.C., 1968. Cooling and solidification of igneous rocks. In: Hess, H.H., Poldevaart,A. (Eds.), Basalts: the Poldevaart Treatise on Rocks of Basaltic Composition, NewYork, Interscience, pp. 503–536.

Jakobsdóttir, S.S., Roberts, M.J., Guðmundsson, G.B., Geirsson, H., Slunga, R., 2008.Earthquake swarms at Upptyppingar, north-east Iceland: a sign of magma intrusion?Stud. Geophys. Geod. 52, 513–528.

Kavanagh, J.L., Sparks, R.S.J., 2011. Insights of dyke emplacement mechanics fromdetailed 3-D dyke thickness datasets. J. Geol. Soc. Lond. doi:10.1144/0016-76492010-137..

Kelley, D.F., Barton, M., 2008. Pressures of crystallization of Icelandic magmas. J. Pet. 49,465–492. doi:10.1093/petrology/egm089.

Key, J., White, R.S., Soosalu, H.E., Jakobsdóttir, S.S., 2011. Multiple melt injection along aspreading segment at Askja, Iceland. Geophys. Res. Lett. 38, L05301.

Khodayar, M., Einarsson, P., 2004. Reverse-slip structures at oceanic diverging plateboundaries and their kinematic origin: data from tertiary crust of west and southIceland. J. Struct. Geol. 26, 1945–1960.

Kim, Y.-S., Sanderson, D.J., 2005. The relationship between displacement and length offaults: a review. Earth Sci. Rev. 68, 317–334.

Klein, F.W., 1982. Earthquakes at Loihi submarine volcano and the Hawaiian Hot Spot.J. Geophys. Res. 87, 7719–7726.

Klein, F., Koyanagi, R.Y., Nakata, I.S., Tanigawa, W.R., 1987. The seismicity of Kilauea'smagma system. In: Decker, R.W., Wright, T.L., Stauffer, P.H. (Eds.), Volcanism inHawaii: US Geol. Surv. Prof. Pap., 1350, pp. 1019–1186.

Knopoff, L., Randall, M.J., 1970. The compensated linear-vector dipole — a possiblemechanism for deep earthquakes. J. Geophys. Res. 75, 4957–4963.

Lister, J.R., Kerr, R.C., 1991. Fluid-mechanical models of crack propagation and theirapplication to magma transport in dykes. J. Geophys. Res. 96, 10,049–10,077.

Maccaferri, F., Bonafede, M., Rivalta, E., 2010. A numerical model of dyke propagation inlayered elastic media. Geophys. J. Int. 180, 1107–1123. doi:10.1111/j.1365-246X.2009.04495.x.

Maclennan, J., McKenzie, D., Gronvold, A., Slater, L., 2001. Crustal accretion undernorthern Iceland. Earth Planet. Sci. Lett. 191, 295–310.

Mardia, K.V., 1972. Statistics of Directional Data. Academic Press, London. 357 pp.Martens, H.R., White, R.S., Key, J., Drew, J., Soosalu, H., Jakobsdóttir, S., 2010. Dense

seismic network provides new insight into the 2007 Upptyppingar dyke intrusion.Jökull 60, 47–66.

Miller, A.D., Julian, B.R., Foulger, G.R., 1998. Three-dimensional seismic structure andmoment tensors of non-double-couple earthquakes at the Hengill–Grensdalurvolcanic complex, Iceland. Geophys. J. Int. 133, 309–325.

Pan, V., Holloway, J.R., Hervig, R.L., 1991. The pressure and temperature dependence ofcarbon dioxide solubility in tholeiitic basalt melts. Geochem. Cosmochim. Acta 55,1587–1595.

Pitt, A.M., Hill, D.P., Walter, S.W., Johnston, M.J.S., 2002. Midcrustal, long-periodearthquakes beneath northern California volcanic areas. Seismol. Res. Lett. 73,144–152.

Power, J.A., Stihler, S.D.,White, R.A., Moran, S.C., 2004. Observations of deep long-period(DLP) seismic events beneath Aleutian arc volcanoes; 1989–2002. J. Volcanol.Geotherm. Res. 138, 243–266.

Roman, D.C., Cashman, K.V., 2006. The origin of volcano-tectonic earthquake swarms.Geology 34, 457–460.

Roman, D.C., Moran, S.C., Power, J.A., Cashman, K.V., 2004. Temporal and spatialvariation of local stress fields before and after the 1992 eruptions of Crater Peakvent, Mount Spurr volcano, Alaska. Bull. Seismol. Soc. Am. 94, 2366–2379.

Rubin, A.M., 1993. Tensile fracture of rock at high confining pressure: implications fordike propagation. J. Geophys. Res. 98, 15,919–15,935.

Rubin, A.M., Gillard, D., 1998. Dike-induced earthquakes: theoretical considerations.J. Geophys. Res. 103, 10,017–10,030.

Rubin, A.M., Gillard, D., Got, J.-L., 1998. A reinterpretation of seismicity associated withthe January 1983 dike intrusion at Kilauea Volcano, Hawaii. J. Geophys. Res. 103,10,003–10,015.

Scholz, C.H., 1968. The frequency–magnitude relation of microfracturing in rock and itsrelation to earthquakes. Bull. Seismol. Soc. Am. 58, 399–415.

Smith, K.D., von Seggern, D., Blewitt, G., Preston, L., Anderson, J.G., Wernicke, B.P., Davis,J.L., 2004. Evidence for deep magma injection beneath Lake Tahoe, Nevada–California. Science 305, 1277–1280.

Soosalu, H., Key, A.J., White, R.S., Knox, C., Einarsson, P., Jakobsdóttir, S.S., 2010. Lower-crustal earthquakes caused by magma movement beneath Askja volcano on thenorth Iceland rift. Bull. Volcanol. 72, 55–62. doi:10.1007/s00445-009-0297-3.

Strelitz, R.A., 1989. Choosing the ‘best’ double couple from a moment-tensor inversion.Geophys. J. Int. 99, 811–815.

Taisne, B., Jaupart, C., 2009. Dike propagation through layered rocks. J. Geophys. Res.114, B09203. doi:10.1029/2008JB006228.

Vavrycŭk, V., 2005. Focal mechanisms in anisotropic media. Geophys. J. Int. 161,334–346.

Vavryčuk, V., Bohnhoff, M., Jechumtálová, Z., Kolář, P., Šílený, J., 2008. Non-double-couple mechanisms of microearthquakes induced during the 2000 injectionexperiment at the KTB site, Germany: a result of tensile faulting or anisotropy ofa rock? Tectonophysics 456, 74–93.

Waldhauser, F., 2001. HypoDD: A computer program to compute double-differencehypocenter locations: U.S. Geol. Surv. Open File Rep., 01-113. 25 pp.

Weinberger, R., Lyakhovsky, V., Baer, G., Agnon, A., 2000. Damage zone around en-echelondike segments in porous sandstone. J. Geophys. Res. 105, 3115–3133.

White, R.S., McKenzie, D., O'Nions, R.K., 1992. Oceanic crustal thickness from seismicmeasurementsandrareearthelementinversions. J.Geophys.Res.97,19,683–19,715.

White, R.S., Smith, L.K., Roberts, A.W., Christie, P.A.F., Kusznir, N.J., iSIMM Team, 2008.Lower-crustal intrusion on the North Atlantic continental margin. Nature 452,460–464 plus supplementary information at www.nature.com.

Wright, T.L., Klein, F.W., 2006. Deep magma transport at Kilauea volcano, Hawaii. Lithos87, 50–79.


Microetching Techniques for Revealing Printed Wiring Board

Documents

Transcript of Microetching Techniques for Revealing Printed Wiring Board