PROBLEMATIC NON-SHEAR MECHANISM OF MODERATE

							PROBLEMATIC NON-SHEAR
MECHANISM OF MODERATE
   EARTHQUAKES IN
   WESTERN GREECE

      J. Zahradník, E. Sokos
     Charles University Prague
   National Observatory of Athens
    Data source: broad-band
Le-3D/20sec waveforms from the
 National Observatory of Athens
Motivation: W. Greece M>5 events
                     DC% < 60%
                       sometimes
                         reported
                         by major
                         agencies
                     • collision
                     • transform f.
                     • subduction
              Vartholomio
            (near Zakynthos)
               Dec. 2, 2002

              ETH-SED:
               Mw=5.7
              DC%=55 !
              (HRV: DC%=58,
Zakynthos
              Mednet: DC%=44)
      Questions to be answered
• Can the data be explained with DC%=100 ?
• If we accept an explanation with a non-DC
  component, is it so large as reported by major
  agencies ?
• Can we estimate the uncertainty of the DC%?
• Is there any relation between the DC% error
  bars and the quality of the match between
  data and synthetics?
• Can the multiple source explain the non-DC ?
                  Method
• moment-tensor inversion (minimization of
  the L2 waveform misfit by the weighted
  least-square method)

• optimization of the source position and time
  (maximization of the correlation by the
  space-time grid search)

• a single point source for f < 0.1 Hz, and
  multiple point sources for f < 0.3 Hz
                                     6 NOA stations,
                                     f=0.05 to 0.1 Hz




blue: data
black: synthetics
for crustal model of Haslinger et al. (1999)

                       weights proportional to 1/A were applied
Are you surprised by the quality
of the match although we work
       only with T < 20 ?

  A trick: After obtaining a preliminary
        solution, we refine it using
      artificially aligned waveforms
      (i.e. a station and component-
   dependent shift by a few seconds) !

      Justification: a synthetic test.
   Forward modeling: uncertain location
 and crustal model justifies the waveform
alignment (artificial shift by a few seconds)




    epicenter shift 5 km    epicenter shift + 3
               to N or E           crustal models
Resolving depth and fault-plane
solution




 opt. depth 17 km:   very stable strike, dip, rake
 Mo=0.16e18 Nm       for all this depth range,
 Mw=5.4              much less stable DC%
100% DC matches data also well
(only 0.05 worse)




     we cannot see
     the difference
           Going into large details:
             Optimum correlation
                             is not
                        compatible
                               with
                         100% DC



trial time shift
     Forcing DC% to be > 90%
decreases correlation. What is an
                    “acceptable”
                         range ?
                    top curves: DC%


                    bottom curves:
                          correlation


                    thick curves refer
                    to the optimum depth
 Uncertainty: (6 data subsets by
repeatedly removing one station)
                        red: „error bars‟
                        (a relative measure)


                        +/- one sigma
                        taken (formally) as
                        acceptable solutions

                        however, we need
                        them “in 2D”, i.e.
                        for a range of
                        depths and shifts
   “2D” acceptable solutions and
their distribution reveal the DC%
                       uncertainty
                      green: DC-percentage
                      (with red „error bars‟)

                      blue: correlation
                     For optimum depth
                     of 17 km we get
                     DC%= 77 to 95 %

                     However, 17 km is not
                     strongly preferred ...
                     so DC%= 72 to 97%
 Compare the uncertain
 DC percentage
 with very stable
 strike-dip-rake

DC%: 72 to 97 %

black nodal lines:
all “2D” acceptable solutions
(a range of depths and time shifts)
Without artificial waveform alignment:
lower DC%, more uncertainty




                   DC%= 60 to 95%
                   (65 to 90% at optimal depth)
How the result may change in a
  different crustal model ?




                         the artificial time shifts kept
           fixed as for Haslinger et al. model, i.e. not
                    optimal for Novotny et al. model!
The uncertainty of DC% is still 72 to
  97%, but the depth is problematic
  In any case, for this event we
        systematically find
           DC% > 70%
(i.e. higher than reported for this
 earthquake by major agencies).
     Can we explain the DC% by
     means of a multiple source ?
Fixing the opt. source position and
increasing frequency (f < 0.3 Hz):
3 subevents




   2-sec time delay between sub 1 and 2; sub 3 is unstable
Subevents 1 and 2: similar strike
     and dip, but different rake

               Consider sub 1 and 2 as 100% DC
               (but unequal !), and sum up their
               moment tensors:

               Result: sub 1+ 2 provides DC%
               77 to 93%, analogous to the
               previous single-source study.

               Multiplicity seems to explain
               the non-DC mechanism.
 We have to understand the
space-time complexity of the
          source
 We searched multiple point sources
in both possible fault-planes 1 and 2,
    passing through the epicenter,
    but the results were not good.
Innovation: 3D (volume) point
     source optimization

We allow the single-source position to vary
not only in depth, but also in a horizontal
                  plane.
 We do not find an optimized hypocenter,
          but a major slip patch!
The new trial fault plane is given
           by the patch (point 3)
                           and the
                           known
           old epicenter
                             strike
                                  o
   patch                    (303 )
       5 trial source positions   the old epicenter now
       at each depth              appears slightly off
       (16,17, and 18 km)         the plane
                                  (location error)
the subevent time separation is stable
 (2 + 2 sec), and the focal mechanism
                                as well,
                             incl. sub 3



           2 + 2 sec
The optimized “fault plane”                                 RLS
stabilized the solution a lot.
                                                           VLS

 Sub 1,2,3 are not separated                                ITM
 more than ~2 km from
 each other, but the delay                                  EVR
 is 2+2 seconds.
 = small distance, large delay.
                                                            JAN

 Rupture propag. with arrest ?                              KEK
 A multiple event !
                                  trial vertical plane 8 x 2 km
               Conclusions
• The data can be explained with DC%=100.
• Statistically, the DC% is 70 to 95%, much
  larger than reported by major agencies.
• It cannot be excluded that (for this event) a
  better crustal model can still increase DC%.
• Vartholomio earthquake consisted of 3 nearby,
  but enough delayed subevents (we have “3
  earthquakes”, not 1 with 3 patches).
• Multiplicity provides a partial explanation of
  the non-DC mechanism.
            http://seis30.karlov.mff.cuni.cz
http://seis30.karlov.mff.cuni.cz



                        Thank you !