You are on page 1of 57

Ground motion prediction and intensity

conversion relations for the European region


Mathilde B. Srensen and the SAFER WP4 team

Real time shake maps
Data
Source parameters
Site effects
Ground motion prediction equations
PGx vs. I relations
Work done
Cairo:
Intensity prediction
relation (GFZ, NRIAG)
Istanbul:
Intensity prediction relation (GFZ)
PGx vs. I (GFZ)
PGx vs. I (KOERI)
Bucharest:
Intensity prediction relation (GFZ)
PGx vs. I (GFZ)
Naples:
Intensity prediction relation (GFZ)
Ground motion prediction relations (AMRA)
PGx vs. I (INGV)
Iceland:
Ground motion prediction relations (WP5 IMOR)
PGx vs. I (IMOR)
Athens:
Ground motion prediction relations (NKUA)
Outline
Athens
Ground motion prediction
Bucharest
Intensity prediction
PGx vs. I
Cairo
Intensity prediction
Iceland
Ground motion prediction
PGx vs. I
Istanbul
Intensity prediction
PGx vs. I
Naples
Intensity prediction
Ground motion prediction
PGx vs. I
Comparison
Athens
Athens
Partner: NKUA
Selected 397 records from 73 events
recorded to more than 4 stations
1
3
5
7
1 10 100 1000
HYPOCENTRAL DISTANCE (km)
M
A
G
N
I
T
U
D
E

(
M
l
)
Two step stratified regression technique (Fukushima and Tanaka 1990,
Joyner and Boore 1981) was used in order to derive site depended
attenuation relationships.
log
10
(PGA) = 0.65 -1.61 log
10
+ 0.20S + 0.71
log
10
(PGV) = 0.83 -1.44 log
10
+ 0.15S 1.74
log
10
(PSA
0.1
) = 0.55 -1.57 log
10
+ 0.13S + 1.24
log
10
(PSA
0.2
) = 0.59 -1.23 log
10
+ 0.15S + 0.54
log
10
(PSA
0.5
) = 1.16 -2.12log
10
+ 0.16S - 0.54
log
10
(PSA
1
) = 1.14 -1.63log
10
+ 0.23S - 1.74
log
10
(PSA
2
) = 1.2 -1.46log
10
+ 0.21S - 2.88
log
10
(PSA
3
) = 1.15 -1.37log
10
+ 0.23S - 3.25
Bucharest
Vrancea region Intensity prediction relation
Based on digitized intensity maps from 5 large Vrancea earthquakes
Account for anisotropy by introducing spatial site correction function


where dI is a site correction function


( ) ( ) dI M f h h R b
h
h R
a e h d M c I
W W
+ +
+
+ + =
2 2
2
2 2
log log
| | ( )

=
+ + =
5
1
2
, 2 , 4 , 2 , 1 , 5
2
, 1 , 3 , 6
) ( ) )( ( 2 ) ( exp ) , (
j
j j j j j j j j
p p p p p p p p dI u u u
Derive relation for three
distance measures
Partner: GFZ
Vrancea region - Intensity prediction relation
Epicentral distance Rupture distance J-B distance
Similar regression error (~0.6) for the three distance measures
Partner: GFZ
Vrancea region PGx vs. I
Strong motion data from 1977 (ESD), 1986, 1990a, 1990b (NIEP-Uni Karlsruhe) events
Test four weighting schemes (unweighted, weighted, log(mean(PGM)), mean(log(PGM)))
Parameter Relation Mean
regression
error
PGA, raw data I=1.76log(PGA)+6.56 0.6726
PGA, weighted data I=2.76log(PGA)+6.63 0.6726
PGA, average I=4.48log(PGA)+6.55 0.3572
PGA, log average I=4.24log(PGA)+6.70 0.3748
PGV, raw data I=2.10log(PGV)+8.42 0.5359
PGV, weighted data I=2.33log(PGV)+8.58 0.5359
PGV, average I=2.84log(PGV)+8.93 0.3587
PGV, log average I=2.77log(PGV)+8.97 0.3398
Partner: GFZ
Cairo
Cairo Intensity prediction
Partner: GFZ, NRIAG
Based on intensity point data from
7 earthquakes
Unified magnitudes using empirical
relations Mw(Ms)
Mean regression error: o=0.58
Compare to digitized isoseismal
lines from the 1992 Cairo
earthquake (not included in
regression)
( ) ( ) h h R
h
h R
h M I
W
+
+
+ =
2 2
2
2 2
0035 . 0 log 90 . 1 33 . 5 log 01 . 3 86 . 0
Iceland
46 earthquakes in southwest Iceland used derive attenuation relations (D5.2)
Partner: IMOR
Ground motion prediction equations for southwest Iceland (D5.2)
Partner: IMOR
PGV values from 46 earthquakes PGA values from 46 earthquakes
95 . 4 05 . 1 ) 10 00309 . 0 ( log 70 . 1 ) ( log
619 . 0
10 10
+ + = M r PGV
M
72 . 3 27 . 1 0434 . 0 ) 10 262 . 0 ( log 29 . 2 ) ( log
2 287 . 0
10 10
+ + = M M r PGA
M
Attenuation models for a M6.5 event (D5.2)
Partner: IMOR
Thick line: PGA attenuation
model developed in SAFER,
based on velocity records from
the national seismic network, SIL.

Red dashed: (Olafsson and
Sigurbjrnsson) theoretical model
derived for Iceland

Solid fucshia (Halldorsson and
Sveinsson) based on strong
motion data from 6 earthquakes
in Iceland.

The data poinst are PGA values
obtained for the M6.5 June 17
2000 earthquake in the South
Iceland Seismic Zone.
PGx vs. I relations based on 5 earthquakes in SW Iceland
Partner: IMOR
MMI = 1.9 log10(PGV) + 7.7 MMI = 1.6 log10(PGA) + 5.7
Istanbul
Marmara Sea region Intensity prediction relation
Based on data from digitized isoseismial maps from 7 earthquakes
Mean regression error: o=0.672
( ) h h R
h
h R
M I
JB
JB
W S
+
+
+ =
2
2
2
2
2
0020 . 0 log 656 . 2 913 . 5 376 . 0
1912 1953
1999
Partner: GFZ
Marmara Sea region PGx vs. I
Strong motion data from 1983 Biga and 1999 Izmit earthquake from the European
Strong-Motion Database (ESD)
Test four weighting schemes (unweighted, weighted, log(mean(PGM)), mean(log(PGM)))
Parameter Relation Mean
regression
error
PGA, raw data I=3.20log(PGA)+6.33 0.8062
PGA, weighted data I=3.62log(PGA)+6.51 0.8062
PGA, average I=4.52log(PGA)+6.38 0.4469
PGA, log average I=4.29log(PGA)+6.51 0.4859
PGV, raw data I=3.23log(PGV)+9.37 0.8267
PGV, weighted data I=4.00log(PGV)+10.18 0.8267
PGV, average I=5.14log(PGV)+11.08 0.7068
PGV, log average I=5.04log(PGV)+11.15 0.6245
Partner: GFZ
Marmara Sea region PGx vs. I
Partner: KOERI
Derive PGx vs. I relation based on
Observed intensities from 58 earthquakes
Estimated PGA or PGV from ground motion prediction equations of Akkar
and Bommer, Boore and Atkinson, Campbell and Bozorgnia and Glkan and
Kalkan

I
akkar
=4.9 * log(PGV) + 0.02 =0.8
I
boore
=4.35 * log(PGV) + 1.57 =0.7
I
campbell
=4.24 * log(PGV) + 1.96 =0.7

I
kalkan
=6.6 * log(PGA) 7.28 =0.9
I
boore
=4.5 * log(PGA) 2.73 =0.8
I
campbell
=4.8 * log(PGA) 2.98 =0.7

Marmara Sea region PGx vs. I
Partner: KOERI
The PGV-Intensity relationships obtained
by Boore and Atkinson (2006) and
Campbell and Bozorgnia (2006)
attenuation relationships agree well with
the Wald et al. (1999)s PGV-Intensity
relationship for I<6,

The PGV-Intensity relationship obtained by
Akkar and Bommer (2006) attenuation
relationship is considerably different than
Wald et al. (1999)s PGV Intensity
relationship for I<8.

All PGA-Intensity relationships obtained in
this study are considerably different than
Wald et al. (1999)s PGA-Intensity
relationship.
Marmara Sea region PGx vs. I
Partner: KOERI
Using the PGA/PGV distributions obtained for the Kocaeli earthquake with
Boore and Atkinson (2006) attenuation relationships, intensity distribution
for the Kocaeli earthquake was obtained as in the following figure.
Naples
Campania region - Data
Intensity points for 9 earthquakes from DBMI04 online database
Source parameters taken mainly from Gasperini et al. (1999)
Associate the source parameters with uncertainty
Partner: GFZ
Campania region Intensity prediction relation, Monte Carlo approach
Earthquake Lon (E) Lat (N) M
w
Strike () Dip ()
1456 14.71 10km 41.30 10km 7.0 0.3 125 27 45 30
1688 14.56 10km 41.28 10km 6.7 0.3 118 14 45 30
1694 15.34 10km 40.88 10km 6.9 0.3 121 12 45

30
1702 14.99 10km 41.12 10km 6.3 0.3 107 189 45 30
1732 15.06 10km 41.08 10km 6.6 0.3 92 96 45 30
1805 14.47 10km 41.50 10km 6.6 0.3 124 27 45 30
1857 15.84 10km 40.35 10km 7.0 0.3 127 11 45 30
1930 15.41 10km 41.06 10km 6.4 0.3 110 15 55 15
1980 15.33 10km 40.78 10km 6.9 0.3 135 15 41 15
Perform 1 Mio. regressions sampling source parameters
within the given uncertainty bounds
Compare to result of standard regression
Intensity points for 9 earthquakes from DBMI04 online database
Associate the source parameters with uncertainty
Partner: GFZ
Campania region - Results
( )
* 2 * 2
2 *
2 * 2
*
log h h R b
h
h R
a e M c I
W S
+
+
+ =
Variation in regression paramteres compensated
Similar regression error indicating uncertainty in
intensity data
Regression c e
*
a b h
*

R
JB
(MC) 0.658 5.127 3.991 0.0012 10.761 0.948
R
JB
(Std) 0.986 3.151 3.309 0.0024 5.960 0.941
R
epi
(MC) 0.690 5.277 6.001 -0.0026 19.665 0.971
R
epi
(Std) 1.556 -0.428 5.518 -0.0020 15.550 0.972
Partner: GFZ
Campania region - Results
Effect of propagating uncertainties through Monte Carlo approach is
negligible







This indicates that the effect of uncertainties in source parameters is
negligible in comparison to the spread in the intensity data
This implies that intensity data for Italy can be predicted only within 1
intensity unit
Partner: GFZ
Attenuation relationships for PGA and PGV in southern Apennines
Partner: AMRA
The dataset used to retrieve the
attenuation relationship consisted of an
integrated observed and synthetic
strong-motion database that was
obtained using the stochastic approach
proposed by Boore (1983). The input
parameters for the simulation technique,
i.e., the average static stress-drop values
and attenuation parameters (geometric
and anelastic), were obtained through
spectral analysis of waveforms from
earthquakes recorded by the Istituto
Nazionale di Geofisica e Vulcanologia
(INGV) seismic network for a magnitude
range Md (1.5, 5.0) over the last 15
years.
Attenuation relationships for PGA and PGV in southern Apennines
Partner: AMRA
c + + + = log log
2 2
10 10
h R c bM a Pgx
Pgx a b c h o
Pga (m/s
2
) -0.559 0.383 -1.4 5.5 0.155
Pgv (m/s) -3.13 0.570 -1.4 5.0 0.185
PGA PGV
Attenuation relationships for PGA and PGV in southern Apennines
Partner: AMRA
Synthetic databases for PGA and PGV as functions of the epicentral distances for M 5, 6
and 7.
Crosses refer to the data of November 1980/18:34 M 6.9 Irpinia earthquake
01 December 1980/19:04 M 4.6 aftershock
16 January 1981/00:37 M 4.7 aftershock
Continuous black lines refer to the local attenuation relationships retrieved in this
project while dotted and bold dashed lines refer respectively to the SP96 and CA97
attenuation relationships.
Italy PGx vs. I
Data:
The Italian strong motion database, ITACA (Luzi et al., 2008)
The Macroseismic Database of Italy, DBMI08 (Stucchi et al., 2007)
266 PGM-I
MCS
data pairs (three times larger than those adopted previously for
Italy; time period 1976-2004)
Earthquakes
PGM-I
MCS
pairs
(From Faenza and Michelini, in publication)
Partner: INGV
Italy PGx vs. I
Partner: INGV
Relation obtained through Orthogonal Distance Regression (ODR) allowing for
Inclusion of uncertainties for both independent and dependent variables
Direct inversion between PGM and I
The regression has been applied to a binned data set, using the geometric mean
Single-line regression is sucient to t the data
PGV single-line regression for I
MCS
VI
I
MCS
= 5.11 0.07 + 2.35 0.09 log
PGV

PGA single-line regression for I
MCS
VI
I
MCS
= 1.68 0.22 + 2.58 0.14 log
PGA
Italy PGx vs. I
Partner: INGV
PGV PGA
Example: M
W
6.3, April 6, 2009, LAquila main shock in Central Italy
Partner: INGV
Preliminary
Macroseismic Field
Courteously from QUEST
Instrumental Data ShakeMap
Macroseismic Data
Shakemap
- Good match between predicted and reported
macroseismic data
- The regressions can be used to predict realistic
ground motions from intensity data alone
Comparison
Comparison of relations intensity prediction
M=7
M=6
Comparison of relations ground motion prediction
M=7
M=6
Comparison of relations PGx vs. I
Comparison of relations PGx vs. I
Comparison of relations PGx vs. I
Regression model
( ) ( ) h h R b
h
h R
a e h d M c I
W
+
+
+ + =
2 2
2
2 2
log log
Epicentral
intensity (I
0
)
Geometrical
spreading
Energy
absorption
Apply weighting scheme so each intensity class enters with the same weight
Solve weighted least squares problem:

) ( min
1
x A I W
x

Physically based ground motion prediction equations


Adjusted relations to fit the characteristics of the given region
(important in early warning applications and for special regions)
Easy to implement for the user
Epicentral distance (distance to the epicenter, on the surface)
Joyner-Boore distance (shortest distance to surface projection of fault)



Rupture distance (shortest distance to the fault)
For rupture distance the functional relation must be updated:



( ) ( ) h h R b
h
h R
a e h d M c I
W
+
+
+ + =
2 2
2
2 2
log log
( ) ( ) h R b
h
R
a e h d M c I
W
+ + = log log
Three distance measures
Marmara Sea region - Results
1912 1953
1999
Uncertainties

1
) * (

= J J C
T
A W J *
1
=
m n
x A I W

=

2
1
2
) (
o
) ( ) , 2 / ) 1 ((
1
C diag m n t x x
c
=

o
Cy y m n t I
T
error
+ =
2 1
) , 2 / ) 1 (( o o
|
|
.
|

\
|
+
+
=
c
c
= h h R
h
h R
h M
x
Ax I
y
w
T 2 2
2
2 2
, log , 1 ), log( ,
) (
The uncertainties in the estimated parameters x and in predicting a new intensity I for given predictor values M
w
, R,
and h are connected with the covariance matrix C of the parameter estimates
with
and the mean squared regression error
where m is the dimension of x (the number of model parameters). For a specified level
of certainty , the confidence bounds x
c
for the fitted parameters x are given by
where t
-1
(p,) is the inverse of the cumulative t-distribution for the corresponding probability p and degrees of
freedom. For 40, t
-1
(p,) N
-1
(p), the inverse of the cumulative standard normal distribution at p. In this case a
certainty level of 68.3% (=0.683) corresponds to the standard deviation (1) of normally distributed errors.
Much more interesting in this study is the error of a new intensity prediction I of the estimated model. For given
predictor values M
w
, R, and h, this can be expressed by
where y is the Jacobian of I-Ax at the predictor values:
Marmara Sea region Comparison to previous relations
Earthquake This study Erdik and Eren (1983) Erdik et al. (1985) Bse (2006)
1912 0.7603 0.9305 0.6342 1.7123
1935 0.3848 0.5402 0.4697 0.7356
1953 0.6014 0.9611 0.8710 1.5566
1963 0.6324 1.4904 1.2882 2.6040
1964 0.5634 0.6976 0.6078 1.4825
1967 0.7601 0.9393 1.1987 1.5151
1999 0.6365 0.9307 1.2108 0.8418
Average 0.6198 0.9271 0.8972 1.4926
Erdik and Eren (1983) I
MSK
=0.34+1.54M
S
-1.24ln(R)
Erdik et al. (1985) I
MSK
=-3.92+2.08M
S
-0.98ln(R)
Bse (2006) Ln(I)=0.8089+0.2317M
W
-0.1073ln(R
JB
+0.6M
W
)-0.0052R
JB
+c
B

Vrancea region previous relations
Jianu (1992) Only intensity curves are presented
Marza and Pantea (1994) I
MSK
=I
m
+0.800-0.411log(D
h
) (average relation for all azimuths)
Enescu et al. (2004) Only intensity curves are presented
Sokolov (in review) I
MSK
=-7.0288+0.014807h+(2.4249-6.484010
-3
h) M
W

+(2.515510
-2
-4.536210
-3
h+(-8.501210
-3
+1.549210
-3
h) M
W
)R
+(5.773510
-5
-2.841710
-7
h+(-1.933710
-5
+9.886710
-8
h) M
W
)R
2

+c
S
(Rock site)
Ardeleanu et al. (2005) I
MSK
=I
0
-3log(D
h
/h)-1/30.001O( D
h
-h)
Zsiros (1996) I=I
0
-3log(D
h
/104)-30.0037ln(e) ( D
h
-104) (I
0
=7)
I=I
0
-3log(D
h
/95)-30.0025ln(e) ( D
h
-95) (I
0
=8)
Various zonatins or lacking information makes direct
comparison to previous relations difficult
Advantages of our relation: Physical basis and easy
implementation
Marmara Sea region - Data
Digitized isoseismial maps from 7 earthquakes
1999 event from zmen (2000)
Remaining from Eyidogan et al. (1991)
Source parameters taken from various publications, mostly on the indvidual
events

zmen, 2000
Vrancea region - Data
Digitized intensity maps from 5 important Vrancea earthquakes
Intensity maps of University of Karlsruhe
Anisotropic intensity distribution
Source parameters taken from various publications, mostly on the indvidual
events
Bonjer, pers. comm., 2007
Vrancea region - Results
Event Rupture distance Epicentral distance Joyner-Boore distance
1940 0.805 0.833 0.828
1977 0.660 0.604 0.612
1986 0.477 0.487 0.485
1990a 0.558 0.546 0.551
1990b 0.595 0.599 0.606
Combined 0.600 0.585 0.589
Regression error
Close to error in a new intensity estimate using the
relation
Similar error for the three distance measures
PGM vs I relations - Results
Marmara Sea Vrancea Campania
Weighted regression seems to represent average relation
Other relations are within the 68% prediction bounds of the weighted
relation
The spread indicates the level of uncertainty due to few data with large
spread

You might also like