S1-02 DUROUCHOUX-HUMBERT Probabilistic seismic hazard ...€¦ · Qualification of dynamic analyses...
Transcript of S1-02 DUROUCHOUX-HUMBERT Probabilistic seismic hazard ...€¦ · Qualification of dynamic analyses...
Session : Session 1 Qualification of probabilistic seismic hazard assessment
Probabilistic seismic hazard assessment
Christophe DUROUCHOUX – EDF/TEGG
Nicolas HUMBERT – EDF/CIH
International Symposium
Qualification of dynamic analyses of dams and their equipments
and of probabilistic assessment seismic hazard in Europe
31th August – 2nd September 2016 – Saint-Malo
Saint-Malo © Yannick LE GAL
SUMMARY
1.What is « seismic hazard » ?The components of the hazard
Probabilistic Method
Epistemic uncertainties
The Uniform hazard spectrum (UHS)
2.Research & developpment for seismic hazard assessement.Conditionnal Spectra : rigorous use of the Uniform hazard spectrum
Qualification of PSHA : Bayesian inference
2Title | 2016
What is « seismic hazard » ?
� Tell me what will happen
� Tell me what can happen
� Tell me what’s already happened and is likely to occur again
� Tell me what the regulation tells me to do
3Probabilistic seismic hazard assessment | 2016
the components of the hazard
4
AttenuationAttenuationAttenuationAttenuation
Site EffectSite EffectSite EffectSite Effect
Behavior of the Behavior of the Behavior of the Behavior of the installation & installation & installation & installation &
SSISSISSISSI
Seismic Seismic Seismic Seismic SourceSourceSourceSource
Probabilistic seismic hazard assessment | 2016
Probabilistic Method
5
date M19-juil-63 2.104-févr-61 6.425-août-68 5.209-nov-73 5.406-déc-63 4.810-juil-67 6.1
26-août-67 4.203-janv-68 4.026-nov-59 6.812-mai-76 5.110-avr-75 4.1
12-mars-82 2.6… …
Earthquake catalog
Site
Attenuation :Log(a) = f(M, D) +/- σ
Seismicsource
D
Occurrence Model
Log(N) = a - b.MGütenberg-Richter
Probabilistic seismic hazard assessment | 2016
Probabilistic Method
� What is the annual probability that the
acceleration on the site exceeds the value of:
6
A = 0.1 g
Probabilistic seismic hazard assessment | 2016
Probabilistic Method
7
0 0.05 0.1 0.15 0.2 0.25
pro
ba
bil
ity
acceleration
probability that acceleration exceeds 0.1 g
1 2 3 4 5
M M+dM N(M,M+dM) Acc P(a>A) N(a>A)
5.4 5.8 0.0008 0.082 0.3827 0.0003
Magnitude range
Number of earthquakes in the class
Log(N) = a – b.M Median Acceleration Acc = f(M,R)
Probability that the acceleration A is exceeded
on the site by an earthquake in the
range of magnitude
(LogNormal)
0.0001
0.001
0.01
0.1
1
3 3.4 3.8 4.2 4.6 5 5.4 5.8 6.2 6.6Classe de Magnitude
nom
bre
Ann
uel
Average annual number of
earthquakes inducing an acceleration
exceeding A, for the magnitude
class
ALEATORY UNCERTAINTY
Probabilistic seismic hazard assessment | 2016
Probabilistic Method
8
M M+dM N(M,M+dM) Acc P(a>A) N(a>A)
3.0 3.4 0.1903 0.015 0.0021 0.0004
3.4 3.8 0.0758 0.019 0.0075 0.0006
3.8 4.2 0.0302 0.026 0.0225 0.0007
4.2 4.6 0.0120 0.035 0.0572 0.0007
4.6 5.0 0.0048 0.046 0.1247 0.0006
5.0 5.4 0.0019 0.061 0.2342 0.0004
5.4 5.8 0.0008 0.082 0.3827 0.0003
5.8 6.2 0.0003 0.109 0.5511 0.0002
6.2 6.6 0.0001 0.145 0.7106 0.0001
6.6 7.0 0.00005 0.194 0.8369 0.00004
N = 0.004
Return Period :
T = 1/N = 250 years
What is the annual probability that the acceleration on the site exceeds the value of:
Probabilistic seismic hazard assessment | 2016
Probabilistic Method
9Probabilistic seismic hazard assessment | 2016
475
0.1231
10
100
1 000
10 000
0 0.05 0.1 0.15 0.2 0.25 0.3
Re
turn
Pe
rio
d (
ye
ars)
Acceleration (g)
Probabilistic Method
10Probabilistic seismic hazard assessment | 2016
SourceZones
SiteR
∑ ∑∞
= =+ <=
0
,
max
min
)(].[R
M
MM
dMMMTOTAL
AaPNN
Probabilistic Method
11Probabilistic seismic hazard assessment | 2016
SismotectonicZonations
Probabilistic Method
12Probabilistic seismic hazard assessment | 2016
Hazard Map
Acceleration cm/s² T=475 ans
2525
2525
25
25
25
25
25
25
25
25
50
50
50
50
50
50
50
50
50
50
50
50
50
75
75
75
75
75
100
100
100100
125
125
125150
PARIS
MARSEILLE
BÂLE
LYON
TOULOUSENICE
NANTES
STRASBOURG
BORDEAUX
ST.ETIENNE
LE HAVRE
RENNES
MONTPELLIER
TOULON
REIMS
LILLE
BREST
GRENOBLE
CLERMONT FER.
LE MANS
DIJON
LIMOGES
TOURSANGERS
AMIENS
NIMES
BESANCON
CAEN
ROUEN
METZ
NANCY
PERPIGNAN
ORLEANS
AVIGNON
PAU
DUNKERQUE
POITIERS
CALAIS
BOURGES
LA ROCHELLE
TROYES
CANNES
LORIENT
VALENCE
COLMAR
MONTLUCON
QUIMPER
ROANNE
CHAMBERY
CHATEAUROUX
ANNECY
BRIVE
ARLES
BLOIS
ANGOULEME
VANNES
MACON
CHARTRES
AUXERRE
PERIGUEUX
AGEN
GAP
LE PUY
0
12.5
25
50
75
100
125
150
200
250
Probabilistic Method
13Probabilistic seismic hazard assessment | 2016
P = Annual probability of exceeding a given value
Probability of exceeding the given value
A LEAST ONCE IN « D » YEARS
1 - (1- P)D = RISK
=> P = 1 - (1 - R)1/D
Poisson process: P = 1 - e-N = 1 - e-1/T T = -D/Ln(1-R)
Life Time (D)
10 50 100
Risk (R)
0.1 % 9 995 49 975 99 950
1 % 995 4 975 9 950
10 % 95 475 949
63 % 10 50 100
Probabilistic Method: epistemic uncertainties
14Probabilistic seismic hazard assessment | 2016
zoning EQ catalogOccurrence
ModelMmax
Attenuation Model
Attenuation Model
Attenuation Model
Logic Tree
0.3
0.5
0.2
Probabilistic Method: epistemic uncertainties
15Probabilistic seismic hazard assessment | 2016
zoning EQ catalogOccurrence
ModelMmax
Attenuation Model
Attenuation Model
Attenuation Model
Logic Tree
0.3
0.5
0.2
Probabilistic Method: epistemic uncertainties
16Probabilistic seismic hazard assessment | 2016
Logic Tree
zoning EQ catalogOccurrence
ModelMmax
Attenuation Model
Attenuation Model
Attenuation Model
Attenuation Model
Attenuation Model
Attenuation Model
Mmax
Attenuation Model
Attenuation Model
Attenuation Model
Mmax
0.30.5
0.2
Probabilistic Method: epistemic uncertainties
17Probabilistic seismic hazard assessment | 2016
Logic Tree
zoning catalog
Occurrence Model
Mmax Attenuation Model
Attenuation Model
Attenuation Model
Mmax Attenuation Model
Attenuation Model
Attenuation Model
Mmax Attenuation Model
Attenuation Model
Attenuation Model
Occurrence Model
Mmax Attenuation Model
Attenuation Model
Attenuation Model
Mmax Attenuation Model
Attenuation Model
Attenuation Model
Mmax Attenuation Model
Attenuation Model
Attenuation Model
zoning catalogOccurrence
ModelMmax
Attenuation Model
k calculations corresponding to the k branches of the logic tree, each result with it’s weight.
Probabilistic Method: epistemic uncertainties
18Probabilistic seismic hazard assessment | 2016
Probabilistic Method: epistemic uncertainties
19Probabilistic seismic hazard assessment | 2016
475
0.1230.092 0.171
10
100
1 000
10 000
0 0.05 0.1 0.15 0.2 0.25 0.3
Re
turn
Pe
rio
d (
year
s)
Acceleration (g)
The result is composed of the whole confidence interval
UHS : uniform hazard spectrum
UHS : uniform hazard spectrum
20Probabilistic seismic hazard assessment | 2016
� The probabilistic calculation is done for each spectral frequency
independently
� Each point of the UHS has the same probability of exceedance
� An UHS doesn’t correspond the spectrum of a real earthquake : the
different parts of the UHS is generated by different types of earthquakes.
UHS : uniform hazard spectrum
21Probabilistic seismic hazard assessment | 2016
0.01
0.1
0.1 1 10 100
Acc
ele
rati
on
(g)
Frequency (Hz)
UHS
De-aggregation of the hazard
PSHA : some difficulties
22Probabilistic seismic hazard assessment | 2016
� The availability of data, validated, with known uncertainties.
� The meaning of s in the GMPEs ?
� How to take into account expert judgment / how to weight branches in
the logic trees ?
� How to incorporate site effect into the probabilistic scheme ?
� How to set the Maximum magnitude ? Is the Gütenberg-Richter model
still valid for rare events ?
Conclusion of part 1
23Probabilistic seismic hazard assessment | 2016
� The (probabilistic) seismic hazard can not be a single value
� Several choices have to be done :
� Return period
� Level on confidence
� Avoid confusion between :
� The probability for an earthquake of a given magnitude to occur in the region,
� The probability for a given level of ground motion to occur on the site.
Conditional Spectra : rigorous use of the UHS
24Probabilistic seismic hazard assessment | 2016
� What is the return period of a UHS?
A UHS is obtained by the observation of seismic activity during a given return period
(here 10 000 years)
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10 000
Conditional Spectra : rigorous use of the UHS
25Probabilistic seismic hazard assessment | 2016
� What is the return period of a UHS?
For a 10 000y hazard study, the UHS is the max of the recorded spectra during 10 000 year of
observation
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10 000
0
1
2
3
4
5
6
0.1 1 10 100
acc
ele
rati
on
(m/s
²)
Frequency (Hz)
0
1
2
3
4
5
6
0.1 1 10 100
acc
ele
rati
on
(m/s
²)
Frequency (Hz)
0
1
2
3
4
5
6
0.1 1 10 100
acc
ele
rati
on
(m/s
²)
Frequency (Hz)
0
1
2
3
4
5
6
0.1 1 10 100
acc
ele
rati
on
(m/s
²)
Frequency (Hz)
0
1
2
3
4
5
6
0.1 1 10 100
acc
ele
rati
on
(m/s
²)
Frequency (Hz)
Conditional Spectra : rigorous use of the UHS
26Probabilistic seismic hazard assessment | 2016
� What is the return period of a UHS?
A UHS is not a single even
Using the UHS as one single even leads to a much higher return period!
0
1
2
3
4
5
6
0.1 1 10 100
acce
lera
tio
n (
m/s
²)
Frequency (Hz)
Conditional Spectra : rigorous use of the UHS
27Probabilistic seismic hazard assessment | 2016
� What is the return period of a UHS?
Using the UHS as one single even leads to unphysical accelerometric time motion
+ + + = ????
Conditional Spectra : rigorous use of the UHS
28Probabilistic seismic hazard assessment | 2016
� Concept of the conditional spectra method
To transform the UHS in several physical scenarii
0
1
2
3
4
5
6
0.1 1 10 100
acc
ele
rati
on
(m
/s²)
Frequency (Hz)
0
1
2
3
4
5
6
0.1 1 10 100
acc
ele
rati
on
(m/s
²)
Freq uency (Hz)
0
1
2
3
4
5
6
0.1 1 10 100
acce
lera
tio
n (
m/s
²)
Frequency (Hz)
0
1
2
3
4
5
6
0.1 1 10 100
acce
lera
tio
n (
m/s
²)
Freq uency (Hz)
0
1
2
3
4
5
6
0.1 1 10 100
acce
lera
tio
n (
m/s
²)
Freq uency (Hz)
Conditional Spectra : rigorous use of the UHS
29Probabilistic seismic hazard assessment | 2016
� The method of the Conditional Spectra� Go back to more physic input motion than the broad band spectra provided by modern codes
� By dividing the spectrum in several scenarii, lead to a higher number of computations
� Extensively published
� "Conditional Spectra" Lin & Baker - Encyclopedia of Earthquake Engineering
� Baker, 2011. "Conditional Mean Spectrum: Tool for ground motion selection." Journal of Structural Engineering,
� Already used for industrial studies
� Diablo canyon nuclear power plant
� Partial conclusion:
never use the hazard from a probabilistic seismic assessment without coming back to realistic
Qualification of PSHA : Bayesian inference
30Probabilistic seismic hazard assessment | 2016
� Context & Motivations
� In the specific case of moderate and low seismicity areas, the lack of strong motion data
lead to select an attenuation model built on data coming from high seismicity regions.
� Surprisingly, in that context of lack of data, the local seismic recording are not frequently
used to calibrate the attenuation model.
� The updating technique hereinafter try to answer this issue by a systematic method
Low return period
10 000 Years
Qualification of PSHA : Bayesian inference
31Probabilistic seismic hazard assessment | 2016
� Example of uncertainty (Yucca Mountain)
First observation: the uncertainties are extreme, at high & LOW return period
Usually, at low return period, the uncertainties à limited by the fact that this type of events
are frequent and consequently well known � it highlights the fact that local data are not
used to fit the hazard assessment
Qualification of PSHA : Bayesian inference
32Probabilistic seismic hazard assessment | 2016
However there is local data, not used in PSHA
� Exemple 1: the French broadband and accelerometric permanent network
� more than 100 stations,
“PSHA Updating Technique with a Bayesian Framework: Innovations” N. HUMBERT et Al 2015
Qualification of PSHA : Bayesian inference
33Probabilistic seismic hazard assessment | 2016
However there is local data, not used in PSHA
� Exemple 2: CEA velocimetric network
� since 1950 - 40 velocimetric stations
Qualification of PSHA : Bayesian inference
34Probabilistic seismic hazard assessment | 2016
However there is local data, not used in PSHA
� Exemple 3: Historical feedback:
� Sisfrance: 1300 � 2007
� 6000 earthquakes
http://www.sisfrance.net/
UPDATING OF A PSHA BASED ON BAYESIAN INFERENCEWITH HISTORICAL MACROSEISMC INTENSITIESE. Viallet(1), N. Humbert(2), P. Mottier(3)
Qualification of PSHA : Bayesian inference
35Probabilistic seismic hazard assessment | 2016
However there is local data, not used in PSHA
� Exemple 4: Geological unstable structures:
Qualification of PSHA : Bayesian inference
36Probabilistic seismic hazard assessment | 2016
Methods of updating are described in 20 publications presented in PAVIE
� Nicolas Kuehn - Non-Ergodic Seismic Hazard: Using Bayesian Updating for Site-Specific
and Path-Specific Effects for Ground-Motion Models
� Roger Musson- Statistical tests of PSHA models.
� Pierre Labbé, - A method for testing PSHA outputs against historical seismicity at the
scale of a territory; example of France
� Jacopo Selva- Probabilistic Seismic Hazard Assessment: Combining Cornell-Like
Approaches and Data at Sites through Bayesian Inference.
� …..
Recommendation of OECD (PAVIE 2015)
Recommendation 2.1 – A state-of-the-art PSHA should include a testing (or scoring) phase against any available local obse rvation (including any kind of observation and any period of observation) and should include testing not only against its median results but als o against its whole distribution (percentiles).
Title | 2016 37
THANK YOU FOR
YOUR ATTENTION