Jökull - 01.06.2000, Side 36

Foulger and Field
ons at the surface. The model contains no informati-
on on near-surface areas between stations. This factor
explains why the refraction proíiles of Brandsdóttir et
al. (1997), which involved sensors deployed at 200 m
intervals across the caldera floor, detected much lower
surface velocities than are obtained by lateral extra-
polation of the LET model at the surface.
Other significant discrepancies between the LET
model and the shallow refraction results, highlighted
by Brandsdóttir et al. (1997), are at large distances
from Víti along the profiles. Because rays from earth-
quakes may travel for part of their paths outside the
study volume (Figure 4) a velocity structure there
must be defined for the LET inversion. However, this
velocity structure is so poorly sampled by rays that it
is held fixed in the inversion process. The perimeter
nodes of ,the study volume, by their very nature, are
also very sparsely sampled by rays and only at a few
of them are the velocities resolved (Figure 4). For
these reasons it is invalid to calculate by extrapolation
velocities either outside the grid or in the outer rows
of blocks in the study volume.
This means that the parts of the refraction profi-
les of Brandsdóttir et al. (1997) that are resolved by
the LET are limited to distances from Víti of up to
~5.5 km to the west, up to ~1 km to the east, up to
3 km to the north and up to ~10 km to the south. With
the exception of the lower velocities detected at very
shallow depth by Brandsdóttir et al. (1997), explained
above, the agreement between the LET model and
the refraction results is remarkably good within these
ranges.
The LET study of the Krafla area was rather
typical of this kind of study. The volume of int-
erest was parameterised at 2-3 km intervals in the
horizontal and 1 km intervals in the vertical. The
inversion process determined the best fit velocities
at nodes spaced at those intervals, assuming linear
variations in velocity in between and a smooth gener-
al model. Such a method returns a broad, average
model of the velocity variations in the area. It is not
designed to resolve small bodies nor velocity discont-
inuities. This must be appreciated when comparing
velocities determined by interpolation between grid
nodes with methods that involve relatively localised
and precise sampling such as small-scale seismic
refraction experiments or drilling. Notwithstanding
the imperfect earthquake distribution, a reasonable
inversion result was obtained, with a data variance
reduction of 84% from the starting model. The
primary features imaged were high-velocity bodies
beneath the caldera ring fault (Figures 4a-c). These
were interpreted as gabbroic bodies intruded up the
caldera fault. A smaller, high-velocity body was
detected at shallow depth beneath Leirhnjúkur, and a
low-velocity body to the SW of the caldera.
GRAVITY DATA
Hengill-Grensdalur
Þorbergsson et al. (1984) measured gravity at 315
stations with average spacings of ~1.5 km covering
an area ~450 km2 in the Hengill-Grensdalur area.
The measured values of gravity at most stations have
estimated uncertainties of <0.5 mGal. The data have
been discussed in detail by Hersir et al. (1990).
Krafla
A gravity survey of 393 stations in the Krafla caldera
is described by Karlsdóttir et al. (1978). The estimated
accuracy is 0.5 mGal for most of the stations. Many
of the stations are clustered around geothermal featur-
es rather than being uniformly distributed throughout
the area and the survey design was thus not ideal for
comparison with the LET results.
METHOD
The approach adopted here is to convert the LET
velocity field to density and then to calculate a
“simulated” Bouguer anomaly field. This is compared
with the “observed” Bouguer anomaly field calculated
from the gravity data. The velocity-density relations-
hip derived from measurements in the Reyðarfjörð-
ur drillhole was used here (Christensen and Wilkins,
1982), which is
p = 1530 + 230Vp (1)
where Vp is the P-wave velocity in km/s and p is the
rock density in kg/m3. The LET models were divided
into cubes 0.25 km on a side and velocities interpola-
ted linearly to determine an average velocity for each
34 JÖKULL No. 48