European Geosciences Union General Assembly 2006 Vienna, Austria,, 02 07 April 2006 FORECASTING THE M6.9 KYTHERA EARTHQUAKE OF 8 JANUARY 2006: A STEP FORWARD IN EARTHQUAKE PREDICTION? Andreas Tzanis Department of Geophysics and Geothermy, University of Athens,, Greece Filippos Vallianatos Department of Natural Resources, Engineering, Technological Educational Institute of Crete, Chania Branch, Crete, Greece.
BACKGROUND If seismicity is viewed as a sequence of cycles culminating in some s kind of critical point, then it is expected that an increase (acceleration) of seismic activity ty and long range correlation between events will precede large earthquakes. d The acceleration assumes the form Ω( t) n = k( t t) α Ω( t) = K + A( t t) (1) d t where logω=cm+d, t c is the time at which a critical state is attained), A <0, n<1 and K = Ω @ t=t c. IMPLEMENTATION c c Ω is the cumulative Benioff strain E i (t) : the energy of the i th event s.t N(t) : the total number of events at time t ε( t) = = E ( ) i 1 i t s.t. log 10 E i (t) = 4.8 + 1.5M S. Power-law behaviour tested with the Curvature C < 1, if power-law a good approximation N ( t) where C = (Power law fit RMS) (Linear fit RMS) Compute (1) using all earthquakes within concentric circular areas as Find radius at which C is minimum. Do this on a regular grid and construct maps of curvature, critical cal exponent, critical time and predicted magnitude to study.
THE DATA The The most detailed (but not most accurate) catalogue of Greek seismicity smicity is compiled by the Geodynamic Institute of NOA, containing over 55000 events.
DATA REDUCTION I Magnitude of Completeness is M L = 3.6 Detailed study of study area shows that magnitude of completeness is M L = 3.9-4 Rate change detected in Magnitude Reporting as of 1995.5, using method of Reasenberg, (1985) Rate change corrected, using method of Zuniga & Wyss,, (1995)
DATA REDUCTION II NOA reports M L but existing empirical relations converting magnitude to Moment, Energy and Benioff Strain require M S. Using the common events contained in the respective catalogues, NOA M L is converted to ISC M S : S (ISC) = 1.68M L(NOA) 3.35 M S (ISC)
APRIL 2002: CURVATURE AND CRITICAL EXPONENT Curvature shows areas of stronger or weaker power-law behaviour, but Power-law behaviour observed both when seismicity is accelerating (n<1) or decelerating (n>1) The distribution of the critical exponent shows a well structured butterfly pattern with nearly sharp boundaries between exponents greater or smaller than unity!!!
POSSIBLE INTERPRETATION Borrowing and expanding an idea from Bowman King (2001) and King and Bowman (2001), thet observed characteristics of distributed accelerating / decelerating seismicity can possibly be understood in terms of a model combining simple elastic rebound and stress transfer. Failure threshold Self-organization of the regional fault system / Regional damage mechanics at work here Acceleration observed here Deceleration observed here Regions of acceleration / deceleration are defined by the stress field required to rupture a fault with a specified orientation and rake.
APRIL 2002: STRESS MODEL FOR THE POWER-LAW BEVAVIOUR I Define a fault with ϕ=45º, δ=80º, λ=150º, capable of producing a M=7 earthquake in a medium with Young s s modulus 7 1010 7 Pa. Compute failure stress at the depth of 10km. Failure stress is herein defined as (σ( 1 + σ 3 )/2 fault bars
APRIL 2002: STRESS MODEL FOR THE POWER-LAW BEVAVIOUR II Two fault models: One at upper crustal depths and one at intermediate depths: known seismicity trends in the area are also of intermediate depth. The observed distribution of curvature correlates better with the stress bright spots of the intermediate depth fault (location and pear shape of the western accelerating lobe, quiescent channel, location and shape of the eastern accelerating lobe).
APRIL 2002: STRESS MODEL FOR THE POWER-LAW BEVAVIOUR III Time dependence of acceleration and predicted parameters: How big : M = 7.2 7.4 When : T c = 2003.2-2004 Where : Near Antikythira island, SW Hellenic Arc, between Crete and the Peloponnese
APRIL 2002: DISCUSSION Q Do we have an earthquake prediction? A In the absence of a concrete case-history, it is very difficult, if not impossible to assess. If so, Q Are the predicted parameters, especially the magnitude, fairly reasonable? A Yes,, for the area, but power-law acceleration is a renormalizationr process, so that when a new element is added, (i.e. a large pre-schock schock), the sequence is renormalized and the predicted model parameters may change significantly (recall the observed time-history of acceleration). If, however, it is correct that pre-shock magnitudes get progressively larger with approaching to failure, recent seismicity ity patterns indicate that the critical point may be relatively nigh! It is not at all necessary that a large earthquake will occur at the end of the process. The stored energy may be dissipated with low rate aseismic event(s) ) or with a series of smaller earthquakes. It is always possible that we are wrong!
JANUARY 2006: AN EARTHQUAKE Recently, a large earthquake did occur in the area, with the following parameters: Time: 2006-01-08 at 11:34:53.1 Magnitude: M w = 6.7 (NOA M L = 6.9) Location: 36.31 N; 23.25 E Depth: 60 km Fault: N-NE to NE oblique slip thrust ϕ = 198 ; δ = 50 ; λ = 52 or ϕ = 68 ; δ = 56 ; λ = 125 Was this the earthquake? Figure courtesy of the European-Mediterranean Seismological Centre @ URL http://www.emsc-csem.org
JANUARY 2006: CURVATURE AND CRITICAL EXPONENT Curvature 1 Critical Exponent 2.2-50 -100 0.9-50 -100 2 1.8 Northing (km) -150-200 -250-300 -350-400 100 200 300 400 500 600 Easting (km) 0.8 0.7 0.6 0.5 0.4 Northing (km) -150-200 -250-300 -350-400 100 200 300 400 500 600 Easting (km) 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 Distribution of curvature same as in 2002, with areas of stronger or weaker power-law behaviour. Lower curvatures (better powerlaw) at, and to the NW of Kythira. The distribution of the critical exponent is the same as in 2002, exhibiting a butterfly pattern with nearly sharp boundaries between exponents greater or smaller than unity!!!
JANUARY 2006: STRESS MODEL OF THE ACTUAL FAULT -50 Failure stress (bars) > 0.3 0.25 0.2-50 Failure stress (Predicted fault, April 2002) > 0.3 0.2 0.2-100 0.15-100 0.1-150 0.1-150 0.1 Northing (km) -200-250 0.05 0-0.05 Northing (km) -200-250 0.0 0-0. -300-0.1-300 -0. -350-0.15-350 -0. -400 100 200 300 400 500 600 Easting (km) < -0.3-0.2-0.25-400 100 200 300 400 500 600 Easting (km) < -0.3-0. -0. Left: The 8 January 2006 fault with ϕ = 68 ; δ = 56 ; λ = 125 and epicentre at (36.11ºN, 23.36ºE). Right: The predicted fault with ϕ =45º, δ =80º, λ =150º,, located by trial and error within 30 km of the actual fault. Both: Magnitude M s = 6.9; Depth to: 45 km; Young s s modulus 18 10 7 Pa. Figures show failure stress at the depth of 20 km. Patterns of positive / negative stress transfer somewhat different but with little effect on the outcome because they have overall similar characteristics. Such simple models could anyhow be only gross approximations of reality.
JANUARY 2006: TIME DEPENDENCE OF ACCELERATION Time-to to-failure analysis applied in exactly the same way as in April 2002. 2. We show representative results from a sample of earthquakes in the t positive stress transfer domain (drafted out of several different populations with similar results). Curvature 2010 Critical Time Stable, almost accurate determination of parameters, 1 0.8 0.6 0.4 0.2 2009 2008 2007 2006 2005 2004 2003 but note the shift in the predicted critical time after 2004 and the jerk at the end of 2006 0 1998 2000 2002 2004 2006 Time 2002 1998 2000 2002 2004 2006 Time 0 8 Predicted Magnitude 0.5 Critical Exponent -50 0.45-100 7.5 0.4 0.35-150 7 0.3 0.25-200 6.5 0.2 0.15-250 6 1998 2000 2002 2004 2006 Time 0.1 1998 2000 2002 2004 2006 Time -300 0 100 200 300 400
JANUARY 2006: THE END OF THE CYCLE QUIESCENCE? Benioff strain 6 x 107 5 4 3 2 1 C=0.496; N=154; m=0.237; M=7.4; T c =2007.424 0-50 -100-150 -200 0-1 1970 1975 1980 1985 1990 1995 2000 2005 2010 x 10 7 Time -250-300 0 100 200 300 400 5.4 Benioff strain 5.2 5 4.8 QUIESCENCE? 4.6 2003 2003.5 2004 2004.5 2005 2005.5 2006 Time There are no earthquakes in the positive stress transfer area between early 2004 and late 2005, with 2 events appearing after 2005.5, clearly not fitting into the accelerating pattern and producing the anomalies observed in previous slide! It is possible that the critical state was attained as early as 2004 The earthquake broke out 1.5 years later meaning that additional (unknown) factors delayed the global transition (failure). In the meanwhile, the area went quiescent!
AN INDEPENDENT TEST : THE FRACTAL DIMENSION I If the stress-transfer transfer model leading to the above analysis is representative of o seismogenetic processes, then: The acceleration of seismic release rates would necessitate increasingly stronger clustering,, therefore, a persistently decreasing fractal dimension. Conversely, deceleration of seismic release rates (relaxation) would imply a stable fractal dimension, (or even an increasing fractal dimension). To investigate we use the ISC catalogue Better for this objective as it has better epicentral and hypocentral determinations. NOA depth estimates are notoriously inaccurate. Data available only up to 2004.24, just as needed! Compute D 3 using the Correlation Integral.
AN INDEPENDENT TEST: THE FRACTAL DIMENSION II Temporal Variation of the D-value D-value versus b-value 3 2.8 D-value 2.5 2.7 2.6 D=2b 2 1996 1997 1998 1999 2000 2001 2002 2003 2004 Time in years 1.5 Temporal Variation of the b-value D-value 2.5 2.4 2.3 b-value 1 2.2 0.5 1996 1997 1998 1999 2000 2001 2002 2003 2004 Time in years 3 Temporal Variation of the D-value. 2.1 2 D = 1.09b +/- 0.09b 1.14 1.16 1.18 1.2 1.22 1.24 1.26 b-value D-value 2.5 2 1.5 1 1996 1997 1998 1999 2000 2001 2002 2003 2004 Time in years 1.5 Temporal Variation of the b-value Statistically significant and corresponding decrease of D 3 and b -values observed in the area of positive stress transfer (acceleration). The relationship between D 3 and b is linear b-value 1 0.5 1996 1997 1998 1999 2000 2001 2002 2003 2004 Time in years No change in D 3 and b-values observed in area of negative stress transfer (deceleration).
DISCUSSION Q A A Q A A A A Q A Was this the earthquake? Two independent lines of compelling evidence suggest that it probably WAS! More importantly, the evidence lend support to a physical model of earthquake preparation that is both tenable and testable! Were the predicted parameters, especially the magnitude and time,, fairly reasonable? The epicentral area was accurately estimated. The predicted faulting mechanism was fairly guessed. The predicted magnitude was overestimated by 0.4-0.5 at 2002. Re-evaluation evaluation on the basis of the actual fault improved estimation to within 0.2-0.3 of the observed magnitude. The predicted time was wrong by 2 years at 2002 subsequently it improved (e.g. Tzanis and Vallianatos,, 2004). Re-evaluation evaluation on the basis of the actual fault, improved the prediction to within 0.3 years of the actual time of occurrence. The error in the 2002 estimation could possibly be due to contamination of earthquake population used for estimation by unrelated events The acceleration process is non-linear and therefore very sensitive! Are we absolutely sure? NO! In the absence of a concrete case-history, it is very difficult, if not impossible to answer affirmatively. We must be very cautious! The acceleration-by-stress-transfer transfer model may soon be put to the test! The acceleration