You are on page 1of 22

Earthquake Engineering & Engineering Seismology

Fundamentals of Probabilistic Seismic


Hazard Analysis

Dr Peter J. Stafford

Friday 5th December, 2008

Outline

• The objective of PSHA

• Some basic thought experiments

• Brief revision of probability theory and probability distributions

• Formal representations of PSHA


• Point sources
• Fault sources
• Areal sources

Objective of PSHA

The purpose of PSHA is to provide rational basis via


which earthquake actions may be specified for
seismic design in structural and geotechnical
earthquake engineering applications

The most common types of earthquake actions are:


• 5%-damped acceleration response spectral ordinates for
relatively simple equivalent static design approaches
• Acceleration time-histories for relatively complex inelastic
time domain analyses

1
Specified Design Actions

475-yr Design Spectrum

Two very different ways of


representing seismic actions

Both ways have one crucial


thing in common
Hazard Consistent
 they share the same Accelerogram
Return Period

What is a Return Period?

Hazard Curve Design Spectrum

The return period for a particular level of ground 1


motion is the reciprocal of the Mean Rate of RP 
Exceedance of that ground motion level MRE

The Hazard Curve – a key output of PSHA

The mechanics of PSHA are


all related to constructing a
Hazard Curve

The hazard curve provides the


ground-motion level
corresponding to any return
period

2
A thought experiment…

Consider one site affected by a single earthquake source a


distance R away

This source only produces one type of earthquake – a magnitude


M event

This event occurs, on average, once every two years

When this event occurs it causes ground motions at the site that
exceed a level of Sa*(T) 50% of the time

How often is the level Sa*(T) exceeded???

The thought experiment extended…

Imagine another site that is affected by two earthquake sources


– one of which is the same as that in our previous experiment

The second source is further away and also only produces one
type of earthquake with a certain magnitude M

This source is less active and has an event once every 10 years

When this event occurs there is a 10% chance that the ground
motions at the site will exceed the level of Sa*(T)

Now how often is the Sa*(T) exceeded?

If we keep on extending…

What if these sources can produce earthquakes of more than one


type?

What if these sources have earthquakes occurring at multiple


positions?

What if we add more sources?

What if we consider more ground-motion levels?

What if we just let maths summarise all of these thoughts?

3
Brief Revision of Basic Probability Theory and
Common Probability Distributions

Introduction

Formal presentations of the PSHA framework make use of


continuous probability distributions and a series of integrals

In reality, these integrals are too complicated to be evaluated


directly  they are evaluated using discrete approximations

To really understand the framework and to be able to use PSHA


to its full potential you must be able to appreciate how the
distributions operate

Discrete Probabilities

Processes for which there are a finite number of possible


outcomes are generally modelled as discrete variables

Classic examples that are used a are a dice and a coin

• A fair die has six possible outcomes, all of which are equally
likely (in theory)

• A theoretical coin has two possible outcomes, each of which is


equally likely

4
Discrete Normal Distribution

Discrete
approximation to the
standard normal
distribution used for
modelling the ground
motion variability

Properties of Discrete Variables

The sum of discrete probabilities must equal unity

If there are N possible states and each is equally likely then the
probability of any one state being observed is simply 1/N

The probabilities of discrete multivariate distributions should


also sum to unity

  P x , x   1
J K

j k
j 1 k 1

Continuous Variables

If a system had an infinite number of states then the probability


of any given value would be zero according to the statement of
discrete distributions just presented

Rather than deal with discrete probabilities we deal with


probability densities

Rather than summing to unity, the integral over probability


density functions must equal unity

5
Discrete & Continuous Normal Distribution

It is correct to say
that a particular
epsilon value has a
certain probability
when modelling this
as a discrete variable,
but not when
modelling as a
continuous variable

The equivalent is:

Properties of continuous distributions


Cumulative distribution function:

Probability of exceeding some value, z:

Probability of being between two values of z:

Conditional Probability (two parameters)

Probability of two simultaneous discrete observations:

Independent events:

Continuous versions:

6
Conditional Probability (three parameters)

Probability of three simultaneous discrete observations:

Continuous version:

This page is particularly important!!! Remember M, R, 


scenarios for later…

Common Distributions

Magnitude-frequency
Locations of hypocentres
distributions

Ground-motion distributions
Rupture dimensions, Area,
Width, Length
Maximum Magnitudes for
sources

7
Composite
Distributions

Youngs &
Coppersmith
(1985)
Truncated
Exponential
Uniform

Formal Representations of PSHA

8
How it began…

The general outline of probabilistic seismic


hazard analysis was first published in 1968 by
Allin Cornell (then at MIT, now at Stanford)

Many features have since been added but the


underlying procedure remains the same

Primary Outputs of PSHA

The uniform hazard spectrum is a common output, but it is


constructed from the true building blocks of hazard curves

Hazard curves provide information regarding how often a


particular level of ground motion is exceeded on an annual basis

It is important to note that the hazard curve is relevant for any


given year – NOT next year

That is a completely different question

PSHA
2
ce
ur
So

Log(PGA)

ε
D 1A
A
Source 1
Log(D)
Number
Log(N)

M M

9
PSHA

2
ce
ur
So

Log(PGA)
D 1A
A
Source 1
Log(D)
Number
Log(N)

M M

PSHA
2
ce
ur
So

Log(PGA)

B
D1
B

Source 1
Log(D)
Number
Log(N)

M M

Mathematical Formulation

The rate at which the ground motion level gm* is


exceeded can be calculated from the following general
expression:

The key component of this representation is the joint


probability density function in magnitude, distance and
epsilon

10
PSHA in textbooks (Kramer, 1996)

The procedure that is outlined in Kramer (1996) is one


way of dealing with this joint probability density function
(JPDF)

However, in general, this procedure is not correct. It is


not consistent with real ground motion models or with real
earthquakes

The procedure that is given is correct if one is dealing with


point sources

Point Source Representation

Recall from earlier that a joint pdf may be decomposed using


our knowledge of conditional probabilities

If these variables are independent then we may make use of:

For point sources the assumption is made that all of the


variables are independent:

Breaking up the JPDF for Point Sources

The general expression:

May then be written as follows under the assumption of independence

This in turn may be written as follows (as in Kramer, 1996)

11
Breaking up the JPDF for Point Sources

Now, last time we saw that the probability that the ground
motion exceeds a target level gm* given some magnitude and
distance may be found from the ground motion model

Discrete Approach

Now recall from earlier that:

Therefore,
Probability from the
Total number of Magnitude- Probability of each
And earthquakes rupture location
frequency relations
above Mmin

Sa(T|) Annual
M Rrup Period  p() n(m) Sa(T) T (g) Rate

5.5 72 0.01 -3 0.0062 0.00155 0.01727 0.6325 0.00259 9.651E-06

pi()
5.5 72 0.01 -2 0.0606 0.00155 0.01727 0.6325 0.00488 9.418E-05

5.5 72 0.01 -1 0.2417 0.00155 0.01727 0.6325 0.00918 3.757E-04

5.5 72 0.01 0 0.3829 0.00155 0.01727 0.6325 0.01727 5.952E-04

What is the problem with this approach?

Modern EGMMs use distance metrics


related to the fault rupture

The fault rupture dimensions are a


function of the earthquake magnitude.
This means that the distributions of
magnitude and distance are not
independent

The only ‘real’ point sources that occur


in reality are from very small
earthquakes and we don’t consider
these for design anyway

12
Extended Fault Sources

Defining the Rupture Scenario

Position of hypocentre
Rupture area, width & length
Strike & dip of fault
Seismogenic limits

How many different rupture scenarios?

During our thought experiment and also in the coursework we are


thinking about a small number of scenarios.

How many should we really consider? ALL RELEVANT SCENARIOS


This term is really just an indicator
function – once the scenario
(including epsilon) is defined the
ground motion either is, or is not,
greater than the target

More on the limits of integration shortly…

13
Areal Sources

These sources are more complicated once again


(if treated properly)

In some (many) cases, background sources will just be specified


to pick up the “left-overs” from a comprehensive fault-based
source model

In these sorts of cases the maximum magnitudes that occur in


these background sources tend to be rather low

In such cases it may be sufficient to simply assume that the


point source approach holds, i.e., rupture sizes are negligible

Areal Sources

When the areal sources are genuinely capable of generating


large earthquakes then we have to consider extended fault
ruptures

This time though, we don’t know what the orientation of the


fault plane is

For some regions we may have an idea of the likely orientation


of these faults from knowledge of the regional tectonics

In such cases we can use this knowledge to specify possible


ranges of orientations

If we know nothing then we should consider all physically


realistic orientations

Extended Ruptures in Areal Sources

Now we have five additional integrations:


(1) over a range of hypocentral positions (3D)
(2) over a range of dip values
(3) over a range of strike values

14
Further Extensions…

By now you should have understood that the equations that


govern the PSHA process can start to look rather complicated

Further extensions can be made, such as for rupture directivity,


and these involve further integrals

All of these integrals are evaluated through discrete


approximation in reality

All we are doing is considering more and more combinations of


variables that define our rupture scenarios

In this sense the method is no more complicated than the


coursework. It just takes longer!

The Limits of Integration

The limits of integration for the rupture models are relatively


straightforward – although some thought regarding truncation
should be given

Magnitude always goes from some minimum magnitude


earthquake thought possible of causing damage to engineered
structures (~5.0)

Maximum magnitude is based upon various methods, geological


considerations, previous observations, statistics (extreme value
theory), source scaling relationships,…

Distance really takes care of itself as the contributions to hazard


at large distances are very small

Limits on Epsilon

Many models have magnitude


dependent standard
deviations

There are cases where the


probability of exceeding a
particular ground motion level
is higher for a small
magnitude event due to its
larger standard deviation!!

Note that this occurs at quite


high epsilon values (AS97)

15
Limits on Epsilon
Such high epsilon values do occur in reality!!

Strasser et al. (2007)

Maximum Ground Motions

2 Maximum recorded
Predicted upper bound
PGA (g)

Predicted limiting value


1.5

0.5

0
1940 1960 1980 2000
Year
300
PGV (cm/s)

200

100

0
1940 1960 1980 2000
Year

Truncating the Ground Motion Distribution

16
Effect of Minimum Magnitude

Influence of Minimum Magnitude

Very different
records

Those with sharp


“spikes” have high
PGA but are very
unlikely to cause
damage

Need to find a way


to filter these out

Possible solution is
CAV – not ideal
though

17
The CAV Filter

Deterministic Seismic Hazard Analysis


(DSHA)

Maximum Grounds Motions and DSHA

Prior to PSHA hazard


analyses were
deterministic (DSHA)

There is still a lot of


debate, from some
people, regarding the
use of DSHA

18
DSHA

Supposed to be conservative – the “worst case scenario”

What about the ground motion? Take median plus one standard
deviation!

In general the discussion regarding whether or not to use DSHA


or PSHA is not very helpful

DSHA vs PSHA

DSHA considers a small number of scenarios and


assumes that these scenarios will occur

PSHA considers all scenarios (including that


considered by DSHA) and accounts for the
likelihoods of these scenarios occurring

People for DSHA argue that you cannot identify design scenarios
from PSHA.

However, as you will see in the coursework, disaggregation does


this for us quite nicely

Disaggregation

Store all of the


scenarios and their
contributions to the
hazard and use these to
identify the scenario(s)
for which time histories
should be obtained

Mean or Modal
scenario?

19
4D Disaggregation

Scenario Spectra

Also arguments about the meaning of a uniform hazard spectrum

20
Seismic Actions

Current seismic design around


the world is based upon the
specification of a 5%-damped
elastic acceleration response
spectrum [pseudo-spectral
acceleration 2Sd(T)]

The spectrum is generally a


type of UHS (uniform hazard
spectrum) that corresponds to
some return period, e.g., 475
years

Where does 475 years come from?

Typical design life of structures is ~50 years, engineers


wanted to design for most scenarios, but not the most
extreme. Leave a 10% chance of exceeding design level
over this life

Use a Poisson model for the occurrence of earthquakes


(and therefore, also ground motions)

Poisson Distribution

21
Arbitrariness of Return Periods

Better to think of return periods and annual rates of exceedance


rather than say 10% probability of exceedance in 50 years

22

You might also like