Estimation of Strong Ground Motion: Aleatory Variability and Epistemic Uncertainty

Transcription

1 Estimation of Strong Ground Motion: Aleatory Variability and Epistemic Uncertainty 5th National Conference on Earthquake Engineering 1st National Conference on Earthquake Engineering and Seismology Bucharest, Romania 19th-20th June 2014

2 McGuire et al. (1995) In this age of tight budgets and competing resources, it is just as unacceptable to promote an overlyconservative seismic design or retrofit of an engineered facility as it is to allow an unconservative design or retrofit. Defendable decisions on seismic issues will be made only when unbiased estimates of median ground motions are developed, accounting for all current seismological knowledge, when uncertainties are accurately represented so that the range of possible ground motions for a given earthquake can be established, and when an appropriate, explicit degree of conservatism is adopted in the choice of design or retrofit ground motion. > 2/29

3 ios Seismic hazard assessment Seismic hazard analysis determines the nature and intensity of the ground motions that could be generated at a site by future earthquakes 1. Assessment of event parameters, e.g. 2. what is the probability of a given earthquake occurring on a given fault? Assessment of site parameters, e.g. what is the expected ground motion at site assuming the occurrence of this earthquake? + annual probability of exceedance of a ground motion level = annual rate of earthquake probability that shaking will exceed certain ground motion Earthquake engineering > 3/29

4 History of ground-motion prediction Boxes indicate those methods often used in research and/or practice Douglas and Aochi (2008) > 4/29

5 Ground-motion prediction equations (GMPEs) log10(pga) = a1 + a2mw + (a3 + a4mw)log10 (rjb2+a52) + a6ss + a7sa+ a8fn + a9ft +a10fo Where: PGA: peak ground acceleration a1 a10: regression coefficients Mw: moment magnitude rjb: Joyner-Boore distance in km SS and SA site conditions: SS = 1 for soft soil sites and 0 otherwise SA = 1 for stiff soil sites and 0 otherwise FN, FT and FO faulting parameters: FN = 1 for normal faulting FT = 1 for thrust faulting FO = 1 for odd faulting and 0 otherwise a1 = a2 = a3 = a4 = a5 = 8.1 a6 = (soft soil) a7 = (stiff soil) a8 = (normal) a9 = (thrust/reverse) a10 = (odd/oblique) Ambraseys et al. (2005) > 5/29

6 Aleatory variability and epistemic uncertainty Northridge (1994) Essentially, all models are wrong, but some are useful George Box Epistemic uncertainty of median Logic Many curves almost equally likely trees Aleatory variability σ of GMPEs Scatter around the median Bommer and Boore (2004) > 6/29

7 Digital All Number of records 5000 Mw 5 & Repi 200km 100 Few records from M>7 and R<10km 4000 Most records from few regions Very few records from stable areas Date Number of GMPEs Analogue (Douglas, 2011) Increasing European data and GMPEs 0 > 7/29

8 Mw 6 strike-slip earthquake at rjb=20km on a stiff soil site Factor ~ 20 Pass basic QA criteria Which dot is the correct estimate of the median PGA? Douglas (2010, 2012) > 8/29

9 Quality assurance criteria (update of Cotton et al.) Bommer et al. (2010) 8 models for active and SCRs > 9/29

10 Testing of GMPEs (Scherbaum et al., 2004, 2009) Douglas and Mohais (2009) > 10/29

11 Logic trees to capture centre, body and range Doubt is an uncomfortable condition, but certainty is a ridiculous one. Le doute n'est pas un état bien agréable, mais l'assurance est un état ridicule. Selection and weights by expert judgement and testing Voltaire SHARE FP7 (active crustal) Delavaud et al. (2012) > 11/29

12 4 E-W 3 2 [m/s2] ya.raw [m/s2] VERT 35 0 VERT [s] [s] [s] VERT001718za.raw za.raw Have strong-motion data a nationality? za.raw > 12/ [s] 40 [m/s2] [m/s2][m/s 2 ]

13 Available data Zone Station coverage (1 1 ) All data Mw 5 and Repi 200km All Europe 17% 12% High hazard 44% 34% Low hazard 1% 0% Local National Pan-European > 13/29

14 At present, there is no doubt that these relations are different for different seismic regions, and region and sitespecific models should be developed on the basis of available strong ground motion records. From Sokolov (2000) We have found for peak accelerations remarkable agreement between Europe and western North America and we are as yet unconvinced by apparent regional differences such as are found in Central America and Japan. From Ambraseys et al. (1997) > 14/29

15 Comparing GMPEs Japan California Europe/Middle East Local (Turkey, Italy, Greece, Iran and NZ) Mw 7 rrup=5km rjb=5km repi=16.3km rhypo=22.1km strike-slip > 15/29

16 The problem with local models (intra-regional dependence) Mw 6, focal depth 10km Douglas (2007) > 16/29

17 Turkey (Ulusay et al., 2004) v Italy (Sabetta and Pugliese, 1987) 95% confidence limits (Sabetta and Pugliese, 1987) 95% confidence limits (Ulusay et al., 2004) Mw 6.5 Douglas (2007) > 17/29

18 True for weak motions for strong motions too? Southern California natural logs Central California Magnitude (Mw) Difference here But no difference here Chiou et al. (2010) > 18/29

19 Some ways forward > GEM Working Group on Tectonic Regionalization: Seek to develop global map of tectonic regimes Objectively combine evidence from various global resources: Global plate boundary map Global strain rate map Global Q map > Better GMPEs Remove pseudo-regional dependency Models with regional (e.g. Q) terms (e.g. NGA West 2) With G. Weatherill, M. Pagani, F. Cotton and others > Rapidly developing networks: Broadband networks Accelerometric networks Citizen observatories, e.g.: Quake-Catcher (cheap MEMS accelerometers) Did you feel it (macroseismic intensity)? > 19/29

20 Influence of sigma on PSHA results What is the correct σ? Bommer and Abrahamson (2006) Factor ~ 2 > 20/29

21 No decrease in sigma over time Strasser et al. (2009) based on Douglas (2003) > 21/29

22 Pure error analysis (data binned by M-R) Intrinsic variability without assuming functional form Douglas and Smit (2001) > 22/29

23 Adding additional terms to GMPEs Factor ~ 1.6 Immature faults generate higher motions Radiguet et al. (2009) > 23/29

24 Ideal situation for site-specific hazard analysis Many earthquakes Single site Travel path Anderson and Brune (1999) > 24/29

25 Because of short time interval, what we actually have Make ergodic assumption: replacement of time by space Fewer earthquakes No data from site of interest Many sites Anderson and Brune (1999) > 25/29

26 Removing ergodic assumption Single-station-single-path σ Site adjustment Single-station σ Sites with multiple recordings Atkinson (2006) > 26/29

27 Impact of single-station sigma on PSHA Selfoss Hospital Ornthammarath et al. (2011) > 27/29

28 Average ofusing all observed PGAs ingmpe a strong-motion database Prediction using simple complex GMPE GMPE (e.g.or NGA) Prediction aa very complex simulations What is the design PGA (or other IM)? Aleatory Epistemic Different magnitudes Representativeness of database Different distances Different tectonics/region to database Different VS30 Magnitude Different deeper site effects (e.g. basin) Distance Different kappas VS30 Different rupture direction Basin Different location w.r.t. hanging wall Different style of faulting (mechanism) Different stress drops Different travel paths Other effects Location of faults Style of faulting (mechanism) Integration over possible values Kappa Stress drop Rupture history Travel paths > 28/29

29 Estimation of Strong Ground Motion: Aleatory Variability and Epistemic Uncertainty 5th National Conference on Earthquake Engineering 1st National Conference on Earthquake Engineering and Seismology Bucharest, Romania 19th-20th June 2014