BSSA12 GMPEs. Dave Boore, Jon Stewart, Emel Seyhan, Gail Atkinson. NGA- West2 Public MeeCng Berkeley, California 15 November 2012

Transcription

1 BSSA12 GMPEs Dave Boore, Jon Stewart, Emel Seyhan, Gail Atkinson NGA- West2 Public MeeCng Berkeley, California 15 November

2 Model DescripCon BA08 funcconal form for M and R jb scaling Change from BA08: Modified V s30 - scaling from NGA- West2 Task 8. 2

3 The GMPE (same as used in BA08, except for F s ) ln Y = F ( M) + F ( R, M) + F ( V, R, M) + εσ M D JB S S30 JB For F M M h : M, F D, F S are magnitude, distance, and site funccons. ε is the fracconal number of standard deviacons of a single predicted value of away from the mean value of (e.g., would be 1.5 standard deviacons smaller than the mean value). All terms, including the coefficient σ, are period dependent. σ is computed using the equacon 2 2 σ = φ + τ where φ is the intra- event aleatory uncertainty and τ is the inter- event aleatory uncertainty. 3

4 The Distance and Magnitude FuncCons Distance dependence: F ( R, M) = [ c + c ( M M )]ln( R/ R ) + c ( R R ) D JB 1 2 ref ref 3 ref where R= R + h 2 2 JB 4

5 The Magnitude FuncCon Magnitude Dependence (Primary): For M M h : For M>M h : F ( M) = e SS + e NS + e RS + e ( M M ) + e ( M M ) M h 5 h F ( M) = ess+ e NS+ e RS+ e ( M M ) M ( e6 0.0) h 2 Mechanism SS NS RS strikeslip normal reverse

6 The Site AmplificaCon FuncCon FS = FLIN + FNL Linear Amplification: F = b ln( V / V ) LIN lin S 30 ref ( V ref = 760 m/s) The coefficient b lin depends on period. We use the Stewart and Seyhan (2012) model. 6

7 The Site AmplificaCon FuncCon FS = FLIN + FNL Nonlinear Amplification: F NL PGA760m/s + 0.1g = bnl ln 0.1g The coefficient b nl depends on period and V S30. We use Stewart and Seyhan s (2012) model 2, where b nl is called f 2. 7

8 DirecCvity Basin depth Not Included Hanging wall (using R JB accounts for this to some extent) Other possible predictor variables (e.g., dip, Ztor, etc.) 8

9 DeterminaCon of Coefficients (2- stage regression) Select data: no basement or large structure records, etc; R JB < 80 km event class 1, as determined by CR JB =10 km Adjust observacons to V s30 =760 m/s using SS12 site amps Constrain c 3 (anelascc term) to the BA08 values Regress for other coefficients, including pseudodepth h 9

10 Stage 2 regressions All but T=3.0, 10 s: quadracc to 6.75; linear beyond, not allowed to go negacve T=3.0,10 s: quadracc 10

11 IM vs R, shorter periods (except PGV) Compare BSSA12 and BA08 (for BA08 almost no data were available for M<4.2) 11

12 IM vs R, longer periods Compare BSSA12 and BA08 (for BA08 almost no data were available for M<4.2) 12

13 Interevent residuals vs M (T=0.2s) Note that τ seems to be overescmated for larger M 13

14 Interevent residuals vs M (T=1.0s) Note that τ seems to be overescmated for larger M 14

15 Intraevent residuals vs R JB (T=0.2s) Note that φ seems to be overescmated for small R JB 15

16 Intraevent residuals vs R JB (T=1.0s) Note that φ may be overescmated for small R JB 16

17 Intraevent residuals vs V S30 (T=0.2s) 17

18 Intraevent residuals vs V S30 (T=1.0 s) 18

19 RE Residuals Analysis Purpose: evaluate model performance using broader data set (> 80 km considered, C1 & C2 events, recordings) Procedure: compute c and η i from following RE regression (i is event index, j is record index): R i, j = c +η i + ε i, j η i examined relacve to source parameters examined relacve to R jb and V s30 For evaluacon of site effects, also look at rock residuals, r (computed senng V s30 = 760 m/s). 19

20 M and R jb Data Screening 20

21 Event Term Trends PGA T=0.2s η i C1 & C2 CR jb 10 km T=1s T=3s η i M M 21

22 Event Term Trends PGA T=0.2s η i C1 CR jb 10 km T=1s T=3s η i M M 22

23 Event Term Trends PGA T=0.2s η i C2 CR jb 10 km T=1s T=3s η i Will use to define a8ershock modifica>on factors M M 23

24 Event Term Trends PGA T=0.2s η i C1 & C2 CR jb 10 km T=1s T=3s η i Hypo. Depth (km) Hypo. Depth (km) 24

25 Within- Event Residuals PGA T=0.2s All data C1 & C2 CR jb 10 km T=1s T=3s Rjb (km) Rjb (km) 25

26 Within- Event Residuals PGA T=0.2s California C1 & C2 CR jb 10 km T=1s T=3s Rjb (km) Rjb (km) 26

27 Within- Event Residuals PGA T=0.2s China C1 & C2 CR jb 10 km T=1s T=3s Rjb (km) Rjb (km) 27

28 PGA T=0.2s T=1s T=3s Rjb (km) Rjb (km) 28

29 Within- Event Residuals PGA T=0.2s Japan C1 & C2 CR jb 10 km T=1s T=3s Rjb (km) Rjb (km) 29

30 Within- Event Residuals PGA T=0.2s Taiwan C1 & C2 CR jb 10 km T=1s T=3s Rjb (km) Rjb (km) 30

31 Within- Event Residuals PGA T=0.2s T=1s T=3s M<4 M<5 5=<M<6 6=<M<7 7=<M<8 Rjb (km) Rjb (km) Rjb (km) Rjb (km) 31

32 Within- Event Residuals PGA T=0.2s All data C1 & C2 CR jb 10 km T=1s T=3s V s30 (m/s) V s30 (m/s) 32

33 Within- Event Rock Residuals PGA T=0.2s r All data C1 & C2 CR jb 10 km T=1s T=3s r V s30 (m/s) V s30 (m/s) 33

34 Summary and Next Steps Simple model, retains principal aspects of funcconal form from BA08 What s the trade off for this simplicity? PotenCally higher sigma Not applicable for certain condicons AnCcipated future changes: Final coefficients from RE regressions AnelasCc apenuacon re- evaluated from SMM data Aqershock (C2) correccon Regional correccons for anelascc apenuacon & V s30 - scaling AddiConal residuals analysis and adjust as needed 34