Jökull - 01.01.2010, Page 27

The 2008 South Iceland Seismic Zone aftershock sequence
posterior probability for the time and location of an
event. This has been verified experimentally (Drew et
al., 2005; Drew, 2010). For this data, an STA/LTA ra-
tio of 0.1/ 0.4 s was used to detect P phases and 0.2/0.6
s to detect SH phases.
Table 1. LOKI stations deployment overview.
– Uppsetning færanlegra jarðskjálftamæla.
Station Name Deployment Station code
GUFU 29/5/2008 21:46 9881
SOLV 29/5/2008 23:16 987A
ORLH 30/5/2008 00:30 988B
HEIDI 30/5/2008 01:13 9870
HOLL 30/5/2008 03:10 987B
RAUH 30/5/2008 18:00 9886
RANI 30/5/2008 19:00 9879
HOLT 30/5/2008 21:02 987C
HALS 30/5/2008 22:34 A01A
BURF 31/5/2008 16:53 A06F
VNES 31/5/2008 19:02 A06E
UGAR 21/6/2008 16:31 986F
The program exists in two forms, CMM Detect
and CMM Locate. Both programs make use of a Look
Up Table (LUT) of modelled travel times from each
station to every cell of a specified grid (i.e. a search
area). CMM Detect performs a systematic exhaustive
search over the grid, with candidate events identified
as the time and location that the coalescence function
exceeds a pre-set threshold (Figure 3). CMM Locate
takes a list of event times, and evaluates the objective
function over short time windows, returning an up-
dated estimate of the time and location for each event.
This is much quicker and easier than re-running CMM
Detect, allowing the effects of changes to input pa-
rameters to be quickly tested.
A cell size of 320 m3 was used for most of the
runs, then reduced to 230 m3 for the final run of CMM
Locate. Using smaller cells is computationally more
costly, without necessarily resulting in any real gains
in accuracy. The optimal spatial sampling is governed
by the shape (smoothness) of the mapped objective
function, which is in turn governed by the input data
(the constraints on arrival times) and the misfit be-
tween data and model (residual).
The time corresponding to the local maximum of
the STA/LTA signal, will be shifted in time with re-
spect to an interactive arrival time pick. As long as
the shift in time is consistent between the time picks
for each of the phase arrivals, the result is a shift in
the estimated time of the event, not the estimated lo-
cation. As the CMM program exhaustively evaluates
for all times, it will also evaluate a second peak of
emergent P energy arrivals. However, as long as the
signal-to-noise ratio of the primary arrival is bigger,
and the primary arrival fits the modelled traveltime
equally well, the CMM estimated time and location
will correspond to the primary arrival.
Figure 3. Cartoon showing coalescence of signals
from three stations. The dashed lines mark equal
ray travel time from each of the stations. As they
are migrated back towards their respective stations,
they eventually intersect at the event location. Fig-
ures from Drew (2010). – CMM forritið skynjar og
staðsetur jarðskjálfta með því að nema útslag og bera
saman reiknaðan ferðatíma til hverrar stöðvar.
Utilizing a cluster, with each day of data processed
as a separate run, the time required to processes the
dataset was between 24 and 36 hours. In each run the
data was band-pass filtered 2–12 Hz for S waves and
4–20 Hz for P waves.
JÖKULL No. 60 27