Jökull - 01.01.2011, Blaðsíða 45

Location accuracy of Eyjafjallajökull earthquakes
the ’wrong’ velocity model VM2 is used to invert the
synthetic arrival times generated using VM1, Hypoin-
verse does find best-fit locations which are shallower
than the synthetic sources, if the true source depth
is between 3–6 km depth (Figure 8c). This is the
depth range with the greatest contrast between the two
velocity models, with VM2 having higher velocities
than VM1 for these depths (Figure 2e). Conversely,
when VM1 is used to invert synthetic arrivals gener-
ated using VM2, the resultant hypocentre depths are
consistently ’too deep’ for synthetic sources shallower
than ∼10 km (Figure 8d). Synthetic source depths of
∼7 km or deeper are recovered to within ±1 km when
inverted using either velocity model, suggesting that
deeper real sources should be well constrained and
relatively insensitive to the assumed velocity model.
DISCUSSION
Effects of changing network geometry
Our synthetic tests and inversion of real data all show
that when data from the temporary stations is ex-
cluded from the location inversion, best-fit hypocen-
tral depths are deeper than those obtained when all
stations are used in the inversion. This is a clear il-
lustration of the importance of proximal stations in
constraining the depth of earthquakes. The temporary
stations also improve the azimuthal coverage of the
network for seismicity in the active region under the
northeast flank of the glacier.
The exclusion of the permanent station VES has
a further, consistent ’deepening’ effect on hypocentral
solutions for true sources deeper than ∼3 km and is
likely to reflect better the reality for most events in the
active region. This effect on depth is minor (<0.5 km)
if the temporary stations are included in the network,
but acts to increase the best-fit hypocentral depth by
1–2 km if only the permanent IMO stations are used.
At shallow depths beneath the glacier, where there
are no stations directly overhead to provide good
depth constraint, network geometry becomes a lim-
iting factor even if all available stations are included
in the hypocentral inversion. Figure 8 illustrates that
there is a characteristic depth, determined by the ve-
locity model (4 km for VM1, 3 km for VM2), shal-
lower than which source depths cannot be well re-
solved by the Eyjafjallajökull network for earthquakes
occurring under the northeast flank of the glacier.
Besides having a dominant effect on the depth
of hypocentral locations, the more proximal tempo-
rary stations also provide better azimuthal coverage.
The configuration of the network when the temporary
stations are excluded leaves an azimuthal gap to the
southwest of ∼125◦ between the relatively proximal
stations MID and ESK, which is filled only by the
more distant station VES (Figure 1). The result of ex-
cluding data from the temporary seismometers is to
shift epicentres to the southwest from the true syn-
thetic location in every case we have shown (Figures 6
and 7). Even with ’perfect’ synthetic arrival times, the
relative weakness in coverage on the southwest side of
the network appears to preclude good constraint along
a northeast-southwest axis. The hypocentres of the
real test events obtained using only the sparser perma-
nent network are also located to the southwest of those
obtained when data from all seismometers are in-
cluded (Figure 5). This southwestward shift can also
be seen clearly for the wider catalogue of seismicity
in this locality during this period (Figure 2c vs. 2a).
When data from station VES are also excluded,
the constraints on the position along this northeast-
southwest axis become markedly poorer. When this
is the case, the inversion tends to converge on appar-
ent epicentral locations even further to the southwest,
∼1.5 km away from the true source’s epicentre for the
synthetic sources shown at 5 km and 10 km depth in
Figures 6 and 7.
At what depth are the real earthquakes?
Our results show that hypocentral depths of 7–11 km
for our test events are artificially deep and are a con-
sequence of sub-optimal network geometry, only ob-
tained when data from the temporary stations are ex-
cluded (Figures 4–6 and 8). The only scenario in
which true depths of 7–11 km could be consistent with
the real and synthetic results is if the true 3-D veloc-
ity structure is sufficiently different to both VM1 and
VM2 in such a way that the velocity structure along
ray-paths to the proximal temporary stations means
that their inclusion in the inversion skews results to ar-
JÖKULL No. 61, 2011 45