SCEC Broadband Platform (BBP) Simulation Methods Validation for NGA-East

Transcription

1 SCEC Broadband Platform (BBP) Simulation Methods Validation for NGA-East BBP Validation Team: N. Abrahamson, P. Somerville, F. Silva, P. Maechling, R. Archuleta, J. Anderson, K. Assatourians, G. Atkinson, J. Bayless, J. Crempien, C. Di Alessandro, R. Graves, T. Hyun, R. Kamai, K. Olsen, R. Takedatsu, F. Wang, K. Wooddell,, D. Dreger, G. Beroza, S. Day, T. Jordan, P. Spudich, J. Stewart and their collaborators Menu du jour Introduction Validation framework and schemes Part A. Validation against recorded ground motions Event and record selection Correcting for site conditions Part B. Validation against GMPEs Evaluation

2 Large collaborative validation of simulations using the SCEC BroadBand Platform Driven by need of seismic hazard projects to supplement recorded datasets South-Western U.S. utilities (SWUS) PEER NGA-East project (new CENA hazard model) PEER NGA-West projects Southern California Earthquake Center (SCEC) BroadBand Platform (BBP) Set of computational tools for ground motion simulations, including post-processing Collaboration of SWUS-SCEC-PEER critical to success. 3 Past validations: Source: Graves and Pitarka (2010) This exercise: Quantitative validation for forward simulations in engineering problems 4

3 Objective relevant to NGA-East Quantitative validation for forward simulations To supplement recorded data for development of GMPEs and hazard analyses Key focus: 5% damped elastic average PSA (f= Hz/ T= s) Validation for mean ground motions only not for standard deviation! 5 Key lessons learned past validations Need more transparency... Need to validate against many events Need clear documentation of fixed and optimized parameters from modelers for each region Need source description that is consistent between methods Use unique crustal structure (V, Q) for all models Consider multiple source realizations Run simulations for reference site conditions correct data with empirical site factors Make all validation metrics computation and plots in uniform units/format implement postprocessing pipeline on BBP Need to tie-in to BBP_v14.3 specific Validation code/bbp Exercise 6 version

4 Validation schemes A. Validation against recorded earthquake ground motions B. Validation against GMPE for generic scenarios Validation allows for development of region-specific rules (source scaling, path) 7 Sims, TS, Ref. Event and stations Obs, TS, Surf. This validation exercise Part A. Site factors IM Processors Part B. IM Processors Sims, IM, Ref. Obs, IM, Ref. Obs, IM, Surf. GMPE, IM, Ref. GOF, Ref. Obs: observed/recorded Sims: simulated TS: time series IM: Intensity measures (e.g. PSA) Ref.: reference rock site Surf.: surface, soil site GOF: goodness-of-fit 8

5 Simulation Methods and Modelers Method Name(s) Composite Source Model (CSM) UCSB EXSIM Graves and Pitarka SDSU (BB Toolbox) Method type Finite fault models Broadband deterministic Stochastic Brune spectrum Hybrid: deterministic LF and stochastic HF Contact(s) and Institution J. Anderson (UNR) R. Archuleta, J. Crempien (UCSB) K. Assatourians, G. Atkinson (UWO) R. Graves (USGS) K. Olsen (SDSU) Point source methods validated outside BBP: SMSIM Silva s stochastic models (also FF) GMPEs used for comparison: NGA-West1: As, BA, CB, CY (2008) CENA: Silva SCVS 2003, Atkinson , Pezeshk et al. (2011) 9 Selection of events and stations Region Event Name Year Mw # Records < 200 km (* <1000 km) Note on Selection # Selected Records Actual (final) WUS Loma Prieta WUS Northridge WUS Landers WUS WUS Whittier Narrows North Palm Springs All stations within 10 km selected All stations within 100 km selected Truncate stations at 40 km JAPAN Tottori JAPAN Niigata WUS Alum Rock WUS Chino Hills CENA Saguenay * CENA Riviere-du- Loup * CENA Mineral, VA * All records selected, use only within 200 km All records selected, use only within 200 km All records selected, use only within 300 km Large dataset (~25 EQs) Many regions & tectonic environments Span wide magnitude range (Mw 4.64 to 7.62) Variety of mechanisms Well-recorded (17 EQs with> 40 records) Select a large subset of stations (~40) that are consistent with mean and standard deviation PSa of the full dataset. 10

6 Selection of stations GOAL: Reduce the number of records (stations) per event while maintaining the mean and standard deviation of the full dataset. CRITERIA: Site V s30 >= 300 m/s to reduce strong non-linear site effects Well-Recorded Earthquakes (40 stations per. event is ideal) Close-in records (<200 km; <1000 km for CENA) METHODOLOGY: 1. Develop simple regression model for all stations for each earthquake (PGA, T=0.2 sec, T=1.0 s) 2. Randomly sample 10 stations from 4 equal distance bins (log space) to generate ~1000 to 10,000 random samples. 3. Perform regressions on samples from population. 4. Develop a penalty function to select station sample with small mean bias and standard deviation over all spectral periods (PGA, T=0.2 sec, T=1.0 sec) Selection of stations Northridge Example 35.4 Northridge Station Locations Latitude Vs30 Full Set Sampled Stations (Vs30 > 300 m/s) Longitude Vs m/s Distance (km)

7 Selection of stations Northridge Example 10 0 PGA (g) Full Model Prediction Sample Prediction PGA Observations Random Samples σ full = σ selected = Distance Northridge (km) Spectral Acceleration at T=0.2 sec (g) Full Model Prediction Sample Prediction T=0.2 sec Observations Random Samples Northridge Distance (km) σ full = σ selected = Spectral Acceleration at T=1.0 sec (g) Full Model Prediction Sample Prediction T=1.0 sec Observations Random Samples σ full = σ selected = Distance (km) Correcting data to reference condition WUS: Recorded data corrected to V s30 =863 m/s Empirical amplification factors from Stewart and Seyhan (Boore et al. NGA-West2 model) F lin =Linear scaling (V s30 effect) F nl =Nonlinear scaling relative to F lin F site =1/exp(F lin +F nl ) Basin correction through Z 1.0 =f(v s30 ) based on Chiou and Youngs 2008 (NGA-West1 model) F basin =1/exp(f(Φ 5,Φ 6, Φ 7, Φ 8 ) Corrected PSA=PSA Rec *F site *F basin

8 Correcting data to reference condition CENA: Recorded data corrected to V s30,ref =1000 m/s Empirical amplification based on GWG memo (Stewart et al.) F lin =Linear scaling (V s30,rec to 760 m/s) F 760-Vs30,ref = Linear scaling using updated AB06 factors (Boore, personal communication) F site =1/exp(F lin +F 760-Vs30,ref ) Corrected PSA=PSA Rec *F site Nomenclature For each scenario, specification of: Source: M w, geometry, location, hypocenter Path: consistent with 1D velocity model Site (as-recorded to reference): empirical site correction factors from Boore et al NGA- West2 For each scenario, seismograms generated for: 50 source realizations ~ 40 stations 2 horizontal dir. 4,000 time series 16

9 Ground motion products Qualitative evaluation of velocity time series and Husid plot based on Arias intensity Northridge PSA, station SCE Vs PSA (g) 17 Ground motion products Goodness-of-fit measures for PSA Average GOF with T for all stations within an event Average GOF with T for all realizations (all stations) Period (s) Period (s) 18

10 Ground motion products Goodness-of-fit measures for PSA Average GOF with distance (all realizations) 19 Ground motion products Goodness-of-fit measures for PSA Map of average GOF (all realizations) 20

11 Ground motion products GOF plots also developed for NGA-West1 (2008) GMPEs SMSIM Allows to see trends/event terms 21 Part B: Scenario selection Scenarios from NGA-West1&2 well constrained by data at 20 and 50 km Rrup M6.2 SS, Vertical, Z tor = 4 km M6.6 SS, Vertical, Z tor = 0 km M6.6 REV, Dip=45 deg., Z tor = 3 km M5.5 REV, Dip=45 deg., Z tor = 6 km 50 realizations of the source, WITH randomized hypocenter location for each Simulations for two velocity models: NorCal and SoCal Part B (comparison with GMPEs) 22

12 Summary of Simulated Events Summary - Parts A and B Landers Loma Prieta Niigata Northridge Tottori North Palm Springs Whittier Saguenay Mineral Alum Rock Chino Hills Riviere-du-Loup Part B: 4 scenarios * 23 Evaluation Evaluation Review panel (June 2013) Douglas Dreger (Chair), UC Berkeley Gregory Beroza, Stanford Steven Day, SDSU Christine Goulet, UC Berkeley Thomas Jordan, USC Paul Spudich, USGS Jonathan Stewart, UCLA Input for review Modeler s documentation and self-assessment BBP results (parts A and B) Part A: criteria based on binned GOF according to M (event), R, T Part B: simple pass-fail 24

13 Evaluation Part A Evaluation Part A 1. Comparison of PSA GOF for each event Mean bias Mean absolute bias Failure threshold is ln(2)=0.69 Thresholds of 0.5 and 0.35 were considered as passing criteria km

14 0.1 to 1 s 1 to 3 s More than3 s km km 5-20 km 0-5 km 0.01 to 0.1 s 27 Evaluation Part A Evaluation Part A 1. Comparison of PSA GOF (mean and mean absolute bias) 2. Combined metric: mean and mean absolute bias n n Used alone Used with GMPEs 3. Evaluation of attenuation bias n Distance dependence slope of zero within 95% confidence interval 28

15 Combined Metric & Comparison with GMPEs Evaluation Part A CGOF = 1 2 ln ( data model ) ln ( data model ) 29 CGOF km km 5-20 km 0-5 km 30

16 Combined Metric & Comparison with GMPEs Evaluation Part A CGOF = 1 2 ln ( data model ) ln ( data model ) CGOFNormalized = CGOF sims CGOFGMPE 31 CGOF Normalized km km 5-20 km 0-5 km 32

17 Evaluation Part A Evaluation Part A 1. Comparison of PSA GOF (mean and mean absolute bias) 2. Combined metric: mean and mean absolute bias Used alone Used with GMPEs 3. Evaluation of attenuation bias Distance dependence slope of zero within 95% confidence interval Also check systematic bias (qualitative) 33 Attenuation Bias Evaluation Part A Fit a line through distance binned GOF values! ln Sa $ obs # " Sasyn & = a + b ln R % ( ) Determine whether slope b=0 lies within 95% confidence interval 34

18 Bias with distance Red shows a ratio of abs(b)/b 95% greater than 1.0, the zero slope does not lie within the 95% CI from simulations. Green shows cases where b=0 lies within the 95% CI from simulations. Smaller numbers are generally controlled by small estimates of slope. Part B (comparison with GMPEs) Part B Design and Evaluation criteria Scenarios from NGA- West1&2 well constrained by data at 20 and 50 km Rrup M5.5 REV M6.2 SS M6.6 SS & REV 50 realizations of the source, WITH randomized hypocenter location for each Simulations for two velocity models: NorCal and SoCal 36

19 Evaluation - Results v14.3 CG Update of Review Panel Findings Details in SRL Focus issue (8 papers) under review The BBP objective of a version-controlled numerical test bed with common post-processing tools was successful in producing results enabling straightforward analysis and review. All of the currently implemented methods should continue to be refined and improved to provide a variety of options for users and to capture epistemic uncertainty. Four methods, EXSIM, G&P, SDSU and UCSB were found to be suitable for simulation of spectral acceleration from T=0.01 to 3 s over R= 0 to 200 km within the validation magnitude range (Mw ) for WUS events. The methods are deemed suitable up to Mw 8 for purposes of assessing relative effects of changes in source geometry, rupture direction, presence of secondary slip on splays, hanging wall effects, etc. Additional work is needed for absolute amplitudes. At T>1 s there is increased bias, and for T> 3 s there are significant deviations from GMPEs, based on WUS events. 37 References Atkinson, G. M., D. M. Boore, K. Assatourians, K. Campbell and D. Motazedian (2009). A guide to differences between stochastic point-source and stochastic finite-fault simulations, Bull. Seism. Soc. Am. 99, Beresnev, I., and G. Atkinson (1998a).FINSIM: a FORTRAN program for simulating stochastic acceleration time histories from finite faults, Seism. Res. Lett. 69, Beresnev, I., and G. Atkinson (1998a).FINSIM: a FORTRAN program for simulating stochastic acceleration time histories from finite faults, Seism. Res. Lett. 69, Boore, D. M. (2005). SMSIM--Fortran Programs for Simulating Ground Motions from Earthquakes: Version 2.3--A Revision of OFR A, U.S. Geological Survey Open-File Report Boore, D. M. (2009). Comparing stochastic point-source and finite-source ground-motion simulations: SMSIM and EXSIM, Bull. Seism. Soc. Am. 99, Brune, J. N. (1970). Tectonic stress and the spectra of seismic shear waves from earthquakes, J. Geophys. Res., 76, Graves, R. A. Pitarka (2010). Broadband ground motion simulation using hybrid approach, Bulletin of Seismological Society of America. Bull. Seism. Soc. Am., 100, 5A, Irikura, K. and H. Miyake (2010). Recipe for Predicting Strong Ground Motion from Crustal Earthquake Scenarios, Pure and Applied Geophysics, DOI /s Liu, P., R. J. Archuleta and S. H. Hartzell (2006). Prediction of broadband ground-motion time histories: Hybrid low/high-frequency method with correlated random source parameters, Bull. Seismol. Soc. Am., 96, , doi: / Mai, P.M., Imperatori, W., Olsen, K.B. (2010), Hybrid Broadband Ground-Motion Simulations: Combining Long-Period Deterministic Synthetics with High-Frequency Multiple S-to-S Backscattering, BSSA, 100(5A), pp Mena, B. and Mai, P. M., Olsen, K. B., Purvance, M. D. and Brune, J. N. (2010). Hybrid Broadband Ground-Motion Simulation Using Scattering Green's Functions: Application to Large-Magnitude Events, Bull. Seism. Soc. Am. 100, Motazedian, D., and G. M. Atkinson (2005). Stochastic finite-fault modeling based on a dynamic corner frequency, Bull. Seismol. Soc. Am. 95, point source in multilayered media, Geophysical Journal International, Volume 148, Issue 3, pp Schmedes, J., R. J. Archuleta, and D. Lavallée (2010). Correlation of earthquake source parameters inferred from dynamic rupture simulations, J. Geophys. Res., 115, B03304, doi: /2009jb Schmedes, J., R. J. Archuleta, and D. Lavallée (2010). Dependency of supershear transition and ground motion on the autocorrelation of initial stress, Tectonophysics, 493, , doi: /j.tecto Schmedes, J., R. J. Archuleta, and D. Lavallée, (2012). A kinematic rupture model generator incorporating spatial interdependency of earthquake source parameters, Geophys. J. Int., doi: /gji/ggs021 Somerville, P. G., Callaghan, S., Maechling, P., Graves, R. W., Collins, N., Olsen, K. B., Imperatori, W., Jones, M., Archuleta, R., Schmedes, J., And Jordan, T.H. (2011). The SCEC Broadband Ground Motion Simulation Platform, SRL, 82(2), p. 275, /gssrl Zeng, Y., J. G. Anderson and G. Yu (1994). A composite source model for computing realistic synthetic strong ground motions, Geophysical Research Letters 21, Zhu, L. & Rivera, L. A. (2002) A note on the dynamic and static displacements from a