Chiou and Youngs PEER-NGA Empirical Ground Motion Model for ...
Chiou and Youngs PEER-NGA Empirical Ground Motion Model for ... Chiou and Youngs PEER-NGA Empirical Ground Motion Model for ...
MODEL PARAMETER DEVELOPMENT The general form of the ground motion model used to assess the model parameters is shown in Equation (16): f Source f = c ( Path f HW 2 4 9 [ y ] [ 1+ exp{ c ( c − M ) } ] 3 M n M = c ln b 2 2 [ RRUP + c5 cosh{ c6 max( M − cHM , 0 ) } ] + ( c4a − c4 ) × ln[ RRUP + cRB ] = φ × [ c + c / cosh{ max( M − c , 0} ] tan / 2 ) × ⎧ ⎡ VS 30 ⎤ ⎫ = φ1 min⎨ln ⎢ 0⎬ + ⎩ ⎣1130⎥ , b ⎦ ⎭ 2 γ 1 1130 ⎛ c2 − c ⎞ − 6 ) + ⎜ ⎟ × ln ⎝ cn ⎠ f ⎧ W cos( δ ) ⎫ ⎨ ⎬ ⎩2( ZTOR + 1) ⎭ ⎧ × ⎨1− π / 2 ⎩ R { ln[ y ] } [ exp{ φ ( min( V , 1130 ) − 360) } − expφ ( 1130 − 360) ] 3 ] = c γ 2 2 = c cos ( δ ) × tanh( R Site Site ln ln[ y Surface RUP = ln[ y 1 + f S 30 1130 Source ] + f −1 R ⎫ JB ⎬ + 0. 001⎭ − 4 ) C&Y2006 Page 33 + Site f Path Site + γ 3 + σ ⋅ z f HW + c + τ ⋅ z 1a × R ij F RUP 3 RV + c 1b F NM RUP ⎡ exp 1130 + φ ⎤ 4 ln⎢ ⎥ ⎣ φ4 ⎦ i + c ( Z The parameter y1130 is the ground motion on the reference site condition (VS30 = 1130 m/sec). Its level is based on the source scaling function fSource, the path scaling function, fPath, the hanging wall function fHW, and a random effect τ zi that is modeled as a Gaussian random variate with inter-event standard deviation τ. The log of the ground motion at a site is the sum of the log of the reference motion and a nonlinear amplification, fSite, that is a function of VS30 and the level of the reference motion. The ground motions at the site also include a random Gaussian variate with intra-event standard deviation σ. The inter-event component of randomness in included when computing the site amplification. Also note that the site amplification function uses the reference motion for the same spectral period. The additional parameters in Equation (16) are: RRUP , closest distance to the rupture plane (km); RJB , Joyner-Boore distance to the rupture plane (km); δ , rupture dip; W , rupture width (km); ZTOR , depth to top of rupture (km); FRV , reverse faulting factor equal to 1 for 30º ≤ λ ≤ 150º, and 0 otherwise; FNM , normal faulting factor equal to 1 for -120º ≤ λ ≤ -60º, 0 otherwise; λ , slip rake angle; VS30 , average shear wave velocity for top 30 m (m/s). Note, fHW applies to all faulting styles. The model parameters were obtained by fitting the model to the selected PEER-NGA data using the nonlinear mixed effects method nlme implemented in the statistical packages S- Plus and R. The process used to obtain these parameters is described below. 7 TOR + (16)
EFFECT OF DATA TRUNCATION The initial analyses of the PEER-NGA data suggested that the anelastic attenuation term γ in Equation (11) was 50-percent larger in absolute value for earthquakes from Taiwan than for earthquakes in California or the other active tectonic regions represented in the selected database (Table 2). For pga the estimated values of γ were approximately -0.006 for California earthquakes and approximately -0.009 for Taiwan earthquakes. This would imply that Q for Taiwan was significantly lower than Q for California or the other regions. However, review of the literature failed to produce studies that confirmed this result. The comparison of Q estimates was made more difficult by the varying assumptions about geometric spreading made by the various investigators. In addition, the estimates of γ obtained from the broadband data for the three southern California earthquakes 0163, 0167 and 0170 were in the range of -0.012 to -0.014 (Figure 13). These results led us to consideration of the effects of missing response data on the estimation of ground motion model parameters. There are two forms of missing response data treated in the literature: censoring and truncation. Censoring occurs when a known set of instruments is triggered, but the response is only known to be below ytruncation. In this case, the number of censored observations and predictor variables for these observations (e.g. M, RRUP, and VS30) are known for the censored sample. Truncation occurs when the observed sample is truncated at some response level ytruncation such that no responses are reported below this level. The number of missing values and values of the predictor variables for these observations are unknown. The two forms of missing response variables lead to different forms of the sample likelihood function in fitting models by maximum likelihood. Censoring/truncation of the strong motion database occur due to the occurrence of ground motions below the trigger threshold for the recording instruments and from the selective processing of recordings favoring those with larger amplitudes. Censoring/truncation occurs also from record processing as frequency bands with low signal/noise ratios are filtered out of the process records. Although the number of possible recordings in a given earthquake is knowable, some instruments may have malfunctioned and there is not a complete accounting of the possible instruments available. Therefore, the truncation model is considered the most appropriate condition. Some evidence of truncation of the data is seen in Figure 2, but it is more apparent when examining individual earthquakes. An obvious case is shown in Figure 27 for the three small southern California earthquakes analyzed previously. The left hand plot shows the 660 pga values for the three earthquakes obtained from the broad band recordings and the right hand plot shows the 119 pga values for the processed records that are in the PEER-NGA database for these earthquakes. Approximate truncation levels are indicated by the dashed horizontal lines. The curves indicate fits to the data using truncated regression techniques with random effects as follows. C&Y2006 Page 34
- Page 1 and 2: Chiou and Youngs PEER-NGA Empirical
- Page 3 and 4: data are consistent with strong mot
- Page 5 and 6: Figure 1: Magnitude-distance-region
- Page 7 and 8: Figure 2: Empirical ground motion d
- Page 9 and 10: EQID Earthquake M Table 3: Inferred
- Page 11 and 12: Site Average Shear Wave Velocity: A
- Page 13 and 14: Figure 6: Relationship between VS30
- Page 15 and 16: 1 ) ∝ C2 × M + ( C2 − C ) × l
- Page 17 and 18: Figure 9: Peak acceleration data fr
- Page 19 and 20: C4+C5M slowly and the value of the
- Page 21 and 22: allows the interpretation of the co
- Page 23 and 24: Figure 13: Coefficients resulting f
- Page 25 and 26: the top of rupture located at x = 0
- Page 27 and 28: Figure 18: Intra-event residuals fo
- Page 29 and 30: Figure 21: Variation of HW* with ma
- Page 31 and 32: The interpretation of the parameter
- Page 33: to the PEER-NGA pga data selected f
- Page 37 and 38: term [ 1 Φ( y ( θ ) + τ ⋅ z ,
- Page 39 and 40: Table 4: Estimate of Anelastic Atte
- Page 41 and 42: data truncated at a maximum distanc
- Page 43 and 44: faulting earthquakes at long period
- Page 45 and 46: Slope -1.5 -1.0 -0.5 0.0 0.5 1.0 0.
- Page 47 and 48: C&Y2006 Page 46 Table 5: Coefficien
- Page 49 and 50: c1 of T0.010S c1 of T1.000S MODEL R
- Page 51 and 52: esid 1 0 -1 -2 resid resid 1 0 -1 -
- Page 53 and 54: esid resid resid 1 0 -1 -2 1 0 -1 -
- Page 55 and 56: esid 2 1 0 -1 -2 SCEC Version 2 0 2
- Page 57 and 58: Amplification w.r.t. Vs30 = 1130 m/
- Page 59 and 60: Sa(g) Sa(g) 10 1 0.1 0.01 10 1 0.1
- Page 61 and 62: Sa (g) Sa (g) 1 0.1 0.01 0.001 0.00
- Page 63 and 64: Sa (g) Sa (g) 1 0.1 0.01 0.001 1 0.
- Page 65 and 66: EXAMPLE CALCULATIONS FORTRAN routin
- Page 67 and 68: Table 6: Example Calculations Perio
- Page 69 and 70: REFERENCES Abrahamson, N.A., and Si
- Page 71 and 72: Frankel, A., A. McGarr, J. Bicknell
- Page 73 and 74: Appendix A Recordings from PEER-NGA
- Page 75 and 76: RSN EQID Earthquake M Station No, S
- Page 77 and 78: RSN EQID Earthquake M Station No, S
- Page 79 and 80: RSN EQID Earthquake M Station No, S
- Page 81 and 82: RSN EQID Earthquake M Station No, S
- Page 83 and 84: RSN EQID Earthquake M Station No, S
EFFECT OF DATA TRUNCATION<br />
The initial analyses of the <strong>PEER</strong>-<strong>NGA</strong> data suggested that the anelastic attenuation term γ in<br />
Equation (11) was 50-percent larger in absolute value <strong>for</strong> earthquakes from Taiwan than <strong>for</strong><br />
earthquakes in Cali<strong>for</strong>nia or the other active tectonic regions represented in the selected<br />
database (Table 2). For pga the estimated values of γ were approximately -0.006 <strong>for</strong><br />
Cali<strong>for</strong>nia earthquakes <strong>and</strong> approximately -0.009 <strong>for</strong> Taiwan earthquakes. This would imply<br />
that Q <strong>for</strong> Taiwan was significantly lower than Q <strong>for</strong> Cali<strong>for</strong>nia or the other regions.<br />
However, review of the literature failed to produce studies that confirmed this result. The<br />
comparison of Q estimates was made more difficult by the varying assumptions about<br />
geometric spreading made by the various investigators. In addition, the estimates of γ<br />
obtained from the broadb<strong>and</strong> data <strong>for</strong> the three southern Cali<strong>for</strong>nia earthquakes 0163, 0167<br />
<strong>and</strong> 0170 were in the range of -0.012 to -0.014 (Figure 13).<br />
These results led us to consideration of the effects of missing response data on the estimation<br />
of ground motion model parameters. There are two <strong>for</strong>ms of missing response data treated in<br />
the literature: censoring <strong>and</strong> truncation. Censoring occurs when a known set of instruments<br />
is triggered, but the response is only known to be below ytruncation. In this case, the number of<br />
censored observations <strong>and</strong> predictor variables <strong>for</strong> these observations (e.g. M, RRUP, <strong>and</strong><br />
VS30) are known <strong>for</strong> the censored sample. Truncation occurs when the observed sample is<br />
truncated at some response level ytruncation such that no responses are reported below this level.<br />
The number of missing values <strong>and</strong> values of the predictor variables <strong>for</strong> these observations are<br />
unknown. The two <strong>for</strong>ms of missing response variables lead to different <strong>for</strong>ms of the sample<br />
likelihood function in fitting models by maximum likelihood.<br />
Censoring/truncation of the strong motion database occur due to the occurrence of ground<br />
motions below the trigger threshold <strong>for</strong> the recording instruments <strong>and</strong> from the selective<br />
processing of recordings favoring those with larger amplitudes. Censoring/truncation occurs<br />
also from record processing as frequency b<strong>and</strong>s with low signal/noise ratios are filtered out<br />
of the process records. Although the number of possible recordings in a given earthquake is<br />
knowable, some instruments may have malfunctioned <strong>and</strong> there is not a complete accounting<br />
of the possible instruments available. There<strong>for</strong>e, the truncation model is considered the most<br />
appropriate condition.<br />
Some evidence of truncation of the data is seen in Figure 2, but it is more apparent when<br />
examining individual earthquakes. An obvious case is shown in Figure 27 <strong>for</strong> the three small<br />
southern Cali<strong>for</strong>nia earthquakes analyzed previously. The left h<strong>and</strong> plot shows the 660 pga<br />
values <strong>for</strong> the three earthquakes obtained from the broad b<strong>and</strong> recordings <strong>and</strong> the right h<strong>and</strong><br />
plot shows the 119 pga values <strong>for</strong> the processed records that are in the <strong>PEER</strong>-<strong>NGA</strong> database<br />
<strong>for</strong> these earthquakes. Approximate truncation levels are indicated by the dashed horizontal<br />
lines. The curves indicate fits to the data using truncated regression techniques with r<strong>and</strong>om<br />
effects as follows.<br />
C&Y2006 Page 34