Jökull - 01.01.2017, Blaðsíða 15

Zeinab Jeddi et al.
Figure 7. Non-linear absolute locations of the eastern events presented as the combined probability density of
all 132 events normalized by its maximum. Black contours correspond to 0.9 – 0.1 with 0.1 spacing. Gray con-
tours are at 0.05, 0.03, 0.01. – Dreifing staðsetninga skjálftanna austan í Kötlu sýnd sem heildardreififall þeirra
(summa dreififalla allra 132 skjálfta). Svartar jafnhæðarlínur svara til hlutfallslegrar stærðar dreififallsins við
0.9 - 0.1 med millibilinu 0.1. Gráar hæðarlínur eru teiknaðar við hlutfallslegar stærðir 0.05, 0.03 og 0.01.
egy described in Sgattoni et al. (2016b), we solved
a weighted least squares problem where the relative
weights are the relative uncertainty estimates of the
differential times. We solved this explicitly (not with
an iterative equation solver). This allowed us to prop-
agate the data uncertainty through the calculation and
estimate the covariance matrix of relative locations.
If the data were appropriately explained (within
their uncertainty), we would expect the misfit scaled
by the data covariance to equal the number of degrees
of freedom in the residual data (i.e., the number of
data minus the number of independent location pa-
rameters). We do not know the absolute errors, but we
have estimated their relative sizes based on the corre-
lation coefficients. We scale the error estimates such
that the normalized misfit becomes its expectation,
i.e., unity, and thereby estimate data errors in absolute
terms. Thus, the initial normalized misfit is around
280 and it decreases to 1 after one iteration, i.e., the
normalized misfit is reduced by 99.7%. The inversion
converges in one iteration suggesting that the problem
is effectively linear, indicating that the distribution of
the sources is small.
Figure 8 shows the relative location results for all
selected events of family 1 (Figure 8b) and family 2
(Figure 8c) together with absolute location of master
events using non-linear location method (Figure 8a).
The horizontal distribution of the 132 events of
family 1 is concentrated in a small area, about 400 m
wide, with an average horizontal uncertainty of ∼25
m. Their depth distribution is concentrated around the
master event over an area ∼400 m deep except for
a few events, which are 500–700 m deeper than the
template. The average depth uncertainty is ∼80 m.
The events in family 2 are also located in a very
small area, less than 150 m wide, around the template,
with uncertainties of ∼10 and ∼30 m in the horizontal
and vertical directions, respectively.
DISCUSSION AND CONCLUSIONS
We have analyzed and described a seismic cluster lo-
cated on the eastern flank of Katla volcano near the
tip of Sandfellsjökull glacier. We were able to de-
tect these small events due to the dense seismic net-
work that was operating on and around Katla between
2011 and 2013. A few events were registered by the
IMO monitoring network before 2011, but the area
has not previously been identified as a persistent seis-
mic source at Katla. We can now add the eastern
seismic source area to the better known caldera and
Goðabunga seismic source areas, and the recently ac-
tive south flank area.
10 JÖKULL No. 67, 2017