Jökull - 01.01.2017, Side 17

Zeinab Jeddi et al.
The eastern events have low magnitude, mostly
between -0.5 and 0.5 ML and a high b-value of
1.6±0.1. Estimates of the b-value in volcanic earth-
quake populations are often anomalously high as in
this case (Roberts et al., 2015). Based on their fre-
quency content (4–25 Hz) and clear P and S ar-
rivals, the eastern events are classified as VT (McNutt,
2005).
We identified two main families of waveforms
sharing the same frequency content, but with differ-
ent amplitude ratios of P and S phases on different
components. Therefore, their mechanisms might dif-
fer. However, the SIL system identifies only a few
events with a double-couple mechanism and those are
not consistent. This may be due to amplitude distor-
tions at individual stations due to local topographic
effects. Their temporal evolution indicates that they
often occur in swarms of tens to hundreds of events
within several days. Two main swarms occurred in
November 2010 and December 2011. However, be-
cause the seismic network was densified in the period
2011–2013, we are able to analyze the seismicity in
detail only during this time period. Before this time,
the seismic station coverage in the area (the perma-
nent monitoring network of the IMO) was not dense
enough to detect and provide good locations of such
small-magnitude events. Therefore, we are not able to
infer when the area became seismically active.
We have been able to locate the hypocenters at an
average absolute depth of around 3.5 km with a non-
linear location method. The combined distribution of
their depth and depth error suggests that they do not
occur near the surface and, therefore, they cannot be
due to glacial processes.
We see no clear temporal migration of our detailed
locations. However, the vast majority of events oc-
curred during a short time span (less than two weeks).
We cannot rule out the possibility that the source re-
gion may have migrated on a time scale longer than
the deployment of our temporary network (2 years).
VT events are often associated with shear move-
ments on near-planar faults, similar to tectonic earth-
quakes. This can sometimes be inferred from
the shape of seismic clusters after relative location
(Hjaltadóttir, 2009, Got et al., 1994). We applied
a relative-location technique to the two families of
events. The resulting distribution of hypocenters is
small and does not clearly indicate that they lie on
a plane. However, there is some suggestion that the
central cluster of events in family 1 is elongate with
a strike of NNE/SSW and dipping towards the north.
This suggestive strike is intermediate between that of
the EVZ and the inferred orientation of the southern-
most sections of the Eldgjá fissure near Kriki (see Fig-
ure 1), where a hyaloclastite deposit has been associ-
ated with the Eldgjá fires (Larsen, 2000). The events
are mostly confined to a 400 m wide depth interval,
but some are distributed over a 1 km depth range be-
low that. This distribution differs significantly from
that of the absolute locations of Jeddi et al. (2016)
who found a cylindrical distribution over the depth
range from 0 to 6 km with an average depth uncer-
tainty of 300 m. We note that 35% of the events
located in the tomographic study were not included
in the relative locations due to low correlation coef-
ficients and/or outlying data (presumably because of
cycle skipping in the differential-time measurements).
This may partly explain the difference, either in terms
of larger errors due to poor correlation or because the
poor correlation may be a result of the events being
located away from the master events of the two iden-
tified families. However, the events near the center
of the distribution of Jeddi’s et al. (2016) absolute
locations all yield differential measurements with a
high correlation coefficient. Furthermore, the lack of
consistency between these absolute locations and our
relative locations can be explained if their uncertain-
ties are underestimated. This may very well be the
case. In both cases, the error estimation is at best sim-
plistic as is generally the case in earthquake location.
First, it should be noted that in both cases the error es-
timate includes only random errors and ignores any
bias that may arise, e.g., from an erroneous veloc-
ity model. Second, the error estimates are in general
based on the inconsistencies of relatively few data and
are, therefore, not very robust. Third, Gaussian statis-
tics are assumed by applying a least-squares optimiza-
tion, while this may not be justified. In fact, we find
that the residual distribution after relative location is
more akin to an exponential distribution than a Gaus-
12 JÖKULL No. 67, 2017