Insights on stress- drop magnitude- dependency and variability from the analysis of accelerometric data.

Transcription

1 Insights on stress- drop magnitude- dependency and variability from the analysis of accelerometric data. Fabrice Co+on - GFZ German Research Centre for Geosciences, Potsdam Boumediene Derras - ISTerre, Grenoble Stephane Drouet - Observatório Nacional, Rio de Janeiro ChrisFan Molkethin, Potsdam University, Potsdam

2 (Brune) Dynamic stress drop can be viewed as a modeling parameter that determines the strength of high- frequency radiafon in the ground- mofons Δτ = 200bars Δτ = 100bars Δτ = 100bars Δτ = 200bars Boore,

3 Ground- mofon simulafons all depend, directly or indirectly, on the selected dynamic stress drop (Δτ) and its distribufon. PGA Δτ = 100bars Δτ = 200bars Boore,

4 QuesFon 1. What is the variability of the stress drop? Allmann & Shearer 2009

5 QuesFon 2. Are stress- drop magnitude- dependent? Edward and Faeh, 2013

6 QuesFon 3: do mainshock and a]ershocks have the same stress- drop?

7 Why is it so difficult? Uncertainty in stress drop is dominated by uncertainty in the corner frequency

8 Fourier Amplitude Spectrum Why is it so difficult? Strong trade- off between the corner frequency and a+enuafon (kappa, fmax) for small earthquakes Frequency, Hz Mw=4.5 stress drop=1.0 (1e5 MPa) Kappa=0.03 (s) Mw=4.5 stress drop=10.0 (1e5 MPa) Kappa=0.03 (s) Mw=4.5 stress drop=100.0 (1e5 MPa) Kappa=0.03 (s) Mw=5.5 stress drop=1.0 (1e5 MPa) Kappa=0.03 (s) Mw=5.5 stress drop=10.0 (1e5 MPa) Kappa=0.03 (s) Mw=5.5 stress drop=100.0 (1e5 MPa) Kappa=0.03 (s) Mw=6.5 stress drop=1.0 (1e5 MPa) Kappa=0.03 (s) Mw=6.5 stress drop=10.0 (1e5 MPa) Kappa=0.03 (s) Mw=6.5 stress drop=100.0 (1e5 MPa) Kappa=0.03 (s)

9 New : high quality accelerometric databases (e.g NGA- west2) are available! Ancheta et al., 2014

10 Our method: infer stress- drop properfes from the stafsfcal properfes of large, high quality, accelerometric dataset Ancheta et al., 2014

11 New accelerometric data : the opportunity to analyze ground- mofon variabilifes PGA: 5.01 cm/s 2 Acceleration (cm/s 2 ) PGA: cm/s 2 PGA: cm/s 2 PGA: 27.5 cm/s 2 PGA: cm/s 2 PGA: cm/s 2 Kiknet records (Japan) Mw=5.5, R=50km, Vs30= m/s Time (s)

12 Ground- MoFon Empirical EquaFons (GMPE s) logarithmic residuals conform to a normal distribufon with mean 0 and standard deviafon σt log(pga) M log(distance)

13 QuesFon 1. What is the variability of the stress drop? Allmann & Shearer 2009

14 Stress-drop variability from source -studies (fc, M0) QuesFon 1. What is the standard deviafon (sigma) of the stress drop? (classical source studies) Stress-drops follow a log-normal distribution σln(δτ) 1.5 Source study Region Mean Brune stress-drop (!! ) (MPa) Allmann and Shearer, 2009 Allmann and Shearer, 2009 Oth et al, 2010 Rietbrock et al., 2012 Edwards and Fah, 2012 Edwards and Fah, 2012 Shearer et al., 2006 Margaris and Hatzidimitriou, 2002 Stress-drop variability Ln(!! ) 0 C No. earthquakes Interplate 5.5 M W * Intraplate 1.50* M W 8 Japan (crustal) M JMA 8 UK Switzerland (foreland) Switzerland (alpine) Southern California 1.6 M L 3.1 Greece 5.2 M W * Johnston et al., 1994 Intraplate ? *Published results are divided by 3.95 to take into account the difference between a Madariaga (1976) corner frequency/source radius compared to that of Brune (1970,1971) and the difference in shear wave velocity.

15 Histograms are from 200 runs of SMSIM (Boore, 2003) with a lognormal distribufon of stress drop with sigma of 1.5 (typical value from seismological studies) 1.5 SMSIM: σ ln (Δτ) = 1.5 Probability density ln(pga) Chiou & Youngs 2008 Abrahamson & Silva 2006 Rodriguez-Marek et al ln(pga) Cotton et al., 2013

16 Histograms are from 200 runs of SMSIM (Boore, 2003) with a lognormal distribufon of stress drop with sigma of 1.5 (typical value from seismological studies) and SMSIM: σ ln (Δτ) = 1.5 SMSIM: σ ln (Δτ) = 0.5 Probability density ln(pga) ln(pga) Chiou & Youngs 2008 Abrahamson & Silva 2006 Rodriguez-Marek et al Cotton et al., 2013

17 Stress- drop variability from GMPE s Stress-drops follow a log-normal distribution σln(δτ) 0.5 GMPE Rodriguez-Marek et al., 2011 Abrahamson and Silva, 2008 (M=5) Abrahamson and Silva, 2008 (M=6) Abrahamson and Silva, 2008 (M=7) Akkar and Bommer, 2010 Boore and Atkinson, 2008 Campbell and Bozorgnia, 2008 Chiou and Youngs, 2008 (M=5) Chiou and Youngs, 2008 (M=6) Chiou and Youngs, 2008 (M=7) Table 2. Between-event variability of recent GMPE s Observed Between Inferred Stress Drop Database Event Variability Variability! Ln( PGA)! Ln("# ) Japan NGA NGA NGA Europe NGA NGA NGA NGA NGA Zhao et al., 2006 Japan

18 The stress drop determined from measuring corner frequency and seismic moment has a variability (ln scale) that is, on average, roughly 3-4 Fmes larger than the variability implied by the GMPEs Variability derived from source studies σln(δτ) 1.5 Variability implied by GMPE s σln(δτ) 0.5

19 QuesFon 2. Are stress- drop magnitude- dependent? QuesFon 3. Do mainshocks and a]ershocks have the same stress- drop? Edward and Faeh, 2013

20 Method: analyzing the magnitude scaling of ground- mofons Response PSA (m/s ) 3.33 Hz 10 0 Rhyp=30km Δσ Data All (NGA- west2) Δσ Model var Drouet 1 el al., 2015 Δσ Model cst 10 2 MPA M w )

21 Methods ingredients A high quality accelerometric database : NGA- West2 A fully data- driven ground- mofon analysis method (ArFficial Neural Network) SensiFvity analysis of response spectral amplitudes on seismological parameters (Mw, Q, Stress- drop, kappa): choice of the opfmal frequency StochasFc simulafons of ground- mofons using various magnitude- dependent stress- drops models

22 Target distance: 30 km (to limit the impact of regional a+enuafon variafons and near- source effects) Ancheta et al., 2014

23 PSA ( 10-4 Mainshock Avoiding the specificafon of any a priori funcfonal form, Aftershock 10-4 Mainshock Aftershock ArFficial M w Neural Networks (ANNs) M w provide fully data- M w driven predicfve models ! PSA ( PSA (m Mainshock Aftershock !

24 ArFficial neural networks (ANNs) predicfons compared to the recent Boore et al.nga- 2 GMPE 10 0 T= 0.0 sec 10-1 T= 0.2 sec (g) (g) T= 1.0 sec BSSA14 Boore et al, 2014 ANN(All V,V ) s30 s30 T= 2.0 sec Rhyp=30km M w M w

25 ln y max / σ ln y max / Q 0 ln y max / κ 0 ln y max / M W WNA Mw=6 WNA Mw=7 ENA Mw=6 ENA Mw= fosc [Hz] OpFmal frequency to analyse the magnitude- scaling of stress- drops? fosc [Hz] SensiFvity of response spectral amplitudes on seismological parameters using algorithmic differenfafon (Molkenthin et al., 2014)

26 ln y max / σ ln y max / Q 0 ln y max / κ 0 ln y max / M W WNA Mw=6 WNA Mw=7 ENA Mw=6 ENA Mw= fosc [Hz] At 2-7Hz : large impact of the stress drop and low impact of kappa, Q and Mw fosc [Hz] OpFmal frequency? Selected frequency: 3.33 Hz

27 Do a]ershocks and mainshocks show a different ground- mofon magnitude scaling (rock sites, Vs30>500m/s)? 8 Aftershock events 8 Mainshock events 7 7 M w R (km) hyp M w R (km) hyp

28 AMN predicfve models derived with A]ershocks and Mainshocks data show the same ground- mofon magnitude scaling 10 1 Mainshock Aftershock PSA 3.33 Hz Rhyp=30km M w

29 10 1 PSA 3.33 Hz Aftershocks data Mainshocks data Model NGA west2 AMN predicfve model (R=30km) and data (27km<R<33km) M w

30 amplitude StochasFc simulafons of ground- mofons (SMSIM, Boore, 2003) γ ± number of values α 0.28 ± frequency (Hz) frequency (Hz) kappa (sec) lues amplitude frequency (Hz) log10 (Q0 ) 2.52 ± 0.12 Rrup 4000

31 StochasFc simulafons of ground- mofons (SMSIM, Boore, 2003) with various stress- drop models Brune s stress drop (MPa) Drouet & Cotton (2015) Yoo & Mayeda (2013) Constant stress drop (2.5, 5 and 10 MPa) M w

32 data (27km<R<33km) 10 1 ANN model based on simulations Simulations ANN model based on NGA west2 Data NGA west2 PSA 3.33 Hz M w Variable stress-drop Constant stress drop (10 MPa) M w

33 Constant stress- drop models NGA west2 10 Mpa 5 Mpa 2.5 Mpa PSA 3.33 Hz Rhyp=30km Drouet & Cotton (2015) Yoo & Mayeda (2013) Constant stress drop (2.5, 5 and 10 MPa) M w M w

34 ParFal derivafve of the response spectrum in funcfon of magnitude log 10 (Y max )/ M 3.33 Hz NGA west2 Variable stress-drop Constant stress-drop Variable stress-drop Constant stress drop (10 MPa) M w M w

35 Constant stress- drop models NGA west2 10 Mpa 5 Mpa 2.5 Mpa PSA 3.33 Hz Rhyp=30km Drouet & Cotton (2015) Yoo & Mayeda (2013) Constant stress drop (2.5, 5 and 10 MPa) M w M w

36 Comparison between the magnitude scaling of observed (NGA- west2) and simulated (variable stress- drop) ground- mofons PSA 3.33 Hz NGA west2 Drouet & Cotton (2015) Yoo & Mayeda (2013) 6 5 Drouet & Cotton (2015) Yoo & Mayeda (2013) Constant stress drop (2.5, 5 and 10 MPa) M w M w

37 Conclusions Stress- drop, like ground- mofon accelerafons, is following a log- normal distribufon (σln(δτ) 0.5) Stress- drop, determined by classical source studies, has greater uncertainty than that implied by accelerometric data and GMPE s Stress- drop variafons have a significant impact on parfal derivafve of accelerometric ground- mofons (response spectrum) in funcfon of magnitude (frequency range 2-7Hz) NGA- west2 a]ershocks and mainshocks ground- mofons show the same magnitude scaling Our results do not confirm a stress- drop increase for Mw<4.5 but rather suggest an increase of stress- drop for earthquakes of magnitude larger than 5.0

38 ArFficial neural networks (ANNs) predicfons compared to the recent Boore et al.nga- 2 GMPE T= 0.2 sec Kink due to the the functional form ( Magnitude hinge Mh choice)?

39 1,4 NGA west2 Drouet & Cotton (2015) Yoo & Mayeda (2013) log 10 (Y max )/ M 3.33 Hz 1,2 1 0,8 0,6 0, M w

40 Uncertainty in stress drop is dominated by uncertainty in the corner frequency log 10 Δτ = log 10 # $ 7 16(kV S ) 3 σ ζ 2 = σ χ 2 + 9σ λ 2 2 σ log10 (Δτ ) 2 = σ log10 ( M 0 ) % & + log 10 M 0 + 3log 10 f C 2 + 9σ log10 ( f C )! σ 2 ζ = (0.3) 2 2 χ + 9(0.3) λ σ 2 ζ = σ ζ = σ log10 Δτ = 0.95 σ LnΔτ = 2.3*σ log10 Δτ = 2.2 Suppose we use 0.3 for the standard deviation for both the normal distribution of the seismic moment and corner frequency in log10 space

41 Stress-drop variability from source -studies (fc, M0) QuesFon 1. What is the standard deviafon (sigma) of the stress drop? Stress-drops follow a log-normal distribution σln(δτ) 1.5 Two-third of the values (68%) are within a factor 20 Source study Region Mean Brune stress-drop (!! ) (MPa) Allmann and Shearer, 2009 Allmann and Shearer, 2009 Oth et al, 2010 Rietbrock et al., 2012 Edwards and Fah, 2012 Edwards and Fah, 2012 Shearer et al., 2006 Margaris and Hatzidimitriou, 2002 Stress-drop variability Ln(!! ) 0 C No. earthquakes Interplate 5.5 M W * Intraplate 1.50* M W 8 Japan (crustal) M JMA 8 UK Switzerland (foreland) Switzerland (alpine) Southern California 1.6 M L 3.1 Greece 5.2 M W * Johnston et al., 1994 Intraplate ? *Published results are divided by 3.95 to take into account the difference between a Madariaga (1976) corner frequency/source radius compared to that of Brune (1970,1971) and the difference in shear wave velocity.

42 Several studies have suggested a magnitude- dependence of the Brune stress- drop for Mw<4.5 Edward and Faeh, 2013

43 Magnitude Motivation Such results generate a large epistemic uncertainty and the inflafon of hazard logic tree branches in regions where models based on small earthquakes have to be extrapolated to large earthquakes (e.g. Pegasos project, Switzerland) Stress drop, bars

44 Magnitude- dependence of ANN models based on various data selecfon (R<30km, R<40km, R<60km) Rhyp=30km

45 Magnitude Motivation Such results generate a large epistemic uncertainty and the inflafon of hazard logic tree branches in regions where models based on small earthquakes have to be extrapolated to large earthquakes (e.g. Pegasos project, Switzerland) Stress drop, bars

46 Magnitude- dependence of ANN models based on various data selecfon (R<30km, R<40km, R<60km) Rhyp=30km

47 Ground- mofon variability : lessons from the last 30 years ~300 records Year ~3000 records Strasser et al., 2006