Section 4:

Extreme Value Analysis

An Introduction

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

Motivation

What can happen in extreme

cases?

Design of weather exposed

constructions (a pylon, wind
power plant, )
Plan constructions for the
protection against natural
disasters.
Storm surge, The Netherlands Feb. 1953

Sea dams in the Netherlands

are dimensioned for the one in
10000 year event. (?!)

Society is risk-averse !
Storm Kyrill, Northern Europe Jan. 2007
Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

Outline

Theoretical Background

Modelling Block Maxima

Modelling Peaks over Threshold

Additional Remarks

Material based on:

Coles 2001 (Chap. 1- 4)

Hosking 1990, Hosking and Wallis 1993, Zwiers and Kharin 1998,
Martins and Stedinger 2000, Katz et al. 2002, Palutikof et al. 1999.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

Section 4: Extreme Value Analysis An Introduction

Theoretical Background

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

Return Value and Return Period

X(T): Return Value X of Return Period T

If T is measured in years: X is the threshold that is exceeded in one year with

a probability of 1/T. (One or more exceedances!)

If T is very large (T >> 1 year) this is equivalent to saying that X is exceeded

on average once in T years.

X(T): Amplitude as function of rareness (the quantile function)

Example: Engelberg (daily rainfall)

The T=5-year return value rainfall is X=70 mm. <=> A rainfall of 70 mm is
exceeded in one year with a probability of 20%.

The rainfall that occurred on August 21, 1954 (X=89.6 mm) has a return
period of T=15 years. <=> Such an event (or larger) is expected on average
every 15 years.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

What if wed use data only?

Extreme 2-day precipitation in Entlebuch (1901-2004/5)
Empirical return period

200

180

2004

2005

177 mm, 2005.08.21-22

T=?

T=105

150 mm, 1984.08.09-10

T=104

T=53

133 mm, 1946.07.05-06

132 mm, 1974.08.22-23

T=52
T=35

T=35
T=27

160

140

120

100
Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

Extreme Value Analysis

Purpose

Find reliable estimates of X(T) for large T (i.e. rare events),

even for T larger than the period of observation,

including estimates of the uncertainty of X(T).

Procedure

Choose an appropriate parametric distribution function

Calibrate it such that it describes available data well
Extrapolate distribution function

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

Distribution of Maxima

Independent identically distributed random vars:

X1, X 2 , X3, ..., X n

X k ~ F(x) iid

E.g. representative for daily precipitations in a year (n=365)

F(x):= prob(Xk x), the parent distribution (CDF)

The Maximum

M n = max ( X1, X 2 , X3, ..., X n )

E.g. the largest 24-hour total in a year

A random variable with:

M n ~ F n (x)

Because:

prob ( M n ! x ) = prob ( X1 ! x, ..., X n ! x )

= prob ( X1 ! x ) " ... " prob ( X n ! x ) = F n ( x )

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

10

Illustration

Distributions of Maxima (Mn)

of the exponential distribution

Distributions of Maxima (Mn)

of the normal distribution

For large n the distribution of Mn

converges to one shape.

Asymptotic shapes for the normal and

the exponential parent distributions
are the same. (The Gumbel Distr.)

Convergence is fast for exponential

but slow for normal parent
distribution.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

11

Extremal Types Theorem

If distribution of Mn converges with large n

I.e. if there exist an > 0, bn, and
a non-degenerate distribution G(x)
such that:

" x ! bn %
F $
**
( G ( x)
' *n()
a
# n &
n

then the limit distribution G(x) is one of

The Gumbel distribution

The Frchet distribution
The Weibull distribution

independent of the parent distribution F(x).

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

12

Extremal Types Distributions

The Gumbel distribution (CDF)

G ( x ) = exp (!exp (!x ))
Emil Julius Gumbel
1891-1966

The Frchet distribution (CDF)

#
0
x!0
%
G ( x) = $
"!
%& exp ("x ) x > 0, ! > 0

Maurice Ren
Frchet
1878-1973

The Weibull distribution (CDF)

#
!
% exp ! (!x )
G ( x) = $
%&
1

x < 0, ! > 0
x"0

Ernst Hjalmar
Waloddi Weibull
1887-1979

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

13

Asymptotic Distributions Laws

The Central Limit Theorem:

The mean of a large number of iid random variables is distributed like the
Normal Distribution independently of the parent distribution.

The Extremal Types Theorem:

The maximum of a large number of iid random variables is distributed like the
Gumbel or Frchet or Weibull Distributions independently of the parent
distribution ( if there is convergence at all).

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

14

Convergence?

In theory

No general assurance of convergence but

convergence is warranted for continuous parent distributions under fairly
general regularity assumptions (well-behaviour of the tail).

In practice

It is justified to presume that data from natural phenomena have a parent

distribution the maxima of which are converging.
Much more care is required with the assumption that the asymptotic limit is
applicable:
slow convergence with some parent distributions
threshold processes may influence the tail

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

15

Sketch of Proof

Max - Stability

A distribution function G(x) is max-stable if there exist ak (>0), and bk so that:

" x ! bk %
Gk ( x) = G $
', k = 1, 2, 3, ...
# ak &

If G(x) is a limit distribution then it must be max-stable:

X1,1

X1,2 ! X1,n

#n!"
##
!

X 2,1

X1,2 ! X 2,n

#n!"
##
! M (n),2 $ G

"
X k,1

"
! X k,n

X k,2

M (n),1 $ G
!

#n!"
##
! M (n),k $ G

n!"

M (n!k ) " G

M (n!k ) " G k

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

16

Sketch of Proof

Gumbel, Frchet and Weibull are max-stable.

Can be verified with simple algebra.

E.g. Gumbel

G k ( x ) = exp (!exp (!x ))

= exp (!k " exp (!x ))

= exp !exp (!x + log ( k ))

)
))

= exp !exp ! ( x ! log ( k )) = G ( x ! log ( k ))

There are no other max-stable distributions than Gumbel, Frchet

and Weibull.

Proof is complex (function theory).

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

17

Generalized Extreme Value Distribution

The GEV Distribution (CDF)

!1 "
.0 (
" x ! %+ 20
GEV (x; , ! , " ) = exp /!*1+ " $
'- 3
# ! &, 04
01 )
where : 1+ " 5 x ! > 0
!

A combined parametrisation of
all three limit distributions

Three parameters:
Location (), Scale (),
Shape ()

=0; =1

= 0: Gumbel, unbounded
> 0: Frchet, lower bound
< 0: Weibull, upper bound

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

18

Which one is appropriate ?

The GEV

takes a special role in statistics.

is theoretically appropriate to modelling
extremes (minima, maxima).

Independent of the nature of original data

Wind gusts,
Precipitation,
Stock market changes,
etc.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

19

History of Extreme Value Statistics

Ronald Aylmer Fisher
& L.H.C. Tippett
First statement of extremal types theorem

Boris V. Gnedenko
1912-1995

Unification / extension of Extreme Value Theory

Emil Julius Gumbel

1891-1966

Statistical application of theory to estimate extremes

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

20

Section 4: Extreme Value Analysis An Introduction

Modelling of Block Maxima

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

22

The Block Maxima Approach

Estimate X(T) (for rare extremes)

by parametric modelling of Maxima
taken from large blocks of
independent data.

Jenkinson 1955, Gumbel 1958

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

23

Procedure

Build Blocks

Extract Block Maxima

Divide full dataset into equal sized chunks of data

E.g. yearly blocks of 365/366 daily precipitation measurements

Determine the Max for each block

Fit GEV to the Max and estimate X(T)

Estimate parameters of a GEV fitted to the block maxima.

Calculate the return value function X(T) and its uncertainty.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

24

The Gumbel Diagramm

Y

The Gumbel Diagramm X(T)

T-axis is transformed such that

Gumbel-Distribution is a straight line.

estimated CDF

F ( x ) = GEV ( x; , ! , " )

Y ( x ) = ! log ! log ( F ( x ))

Gumbel Variate

rank = 108
111
T!k =
= 37a
3

Horizontal axis is linear in Y.

T ( x ) = 1 (1! F(x)) Return period

Block Maxima

xk

k = 1,.., N

T!k =

N +1
N +1! rank(xk )

plotting points of
block maxima xk

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

25

The Gumbel Diagram

Extrapolation:
200-year return
value: 145 mm

Return value:
10-year return
value: 82 mm

Return period:
Amounts fallen on
2005.08.22 have a
return period of 46
years.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

26

Parameter Estimation

Maximum Likelihood (ML) Estimation

Numerically maximize the Likelihood Function

Find , , such that the probability of drawing {xk, k=1,.., N}

as random sample from the GEV would be largest.
Pro: Obtain approximate confidence intervals (e.g. Coles 2001).

Cons: Robustness for small N, numerical procedure

L-Moments Estimation
Set , , so that the first 3 sample L-Moments are equal to the

first 3 L-Moments of the GEV.

Formuli for GEV L-moments from Hosking 1990
Pros: Robustness, analytical solution.
Cons: Best estimate only. Uncertainties (CIs) to be inferred separately.

Hosking 1990, Coles 2001

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

27

Parameter Estimation (Example)

L-Moments
= 53.2 mm
= 12.6 mm
= 0.055
x(T=200) = 131 mm

Max-Likelihood
= 53.5 mm
= 12.5 mm
= 0.038
x(T=200) = 127 mm

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

28

Assumptions

Original data are identically distributed

Original data are independent

May be violated by serial correlation

Asymptotic limit. Maxima from a large number of originals

May be violated by seasonality, trends,

Need large blocks

Need enough independent episodes from the season with largest

values!

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

29

Assessing Goodness of Fit

Engelberg
1-day totals
1901-2010

QQ-plot

Check if GEV can adequately describe the data

Too small block size and non-stationarities may manifest in inconsistencies
Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

30

Uncertainty?

How accurate are parameter estimates ?

How accurate are return values ?

How far is extrapolation justified ?

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

31

Lets roll storms

X100 = 183 km/h

T(Lothar) = 16 a

Yearly maximum
wind gusts in
Zurich 1981-1999

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

32

Lets roll storms

X100=205 km/h
X100 = 183 km/h
X100=165 km/h

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

33

Lets roll storms

60%
95%

Parametric Resampling
Confidence Intervals
X100: 145 - 250 km/h

Large sampling
uncertainty.
Uncertainty of X(T)
increases with T.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

34

Confidence Intervals

Resampling Confidence Intervals

Simulate random samples (same size) and fit GEV

Parametric: random samples from estimated distribution.
Non-parametric: Draw samples (with replacement) from the original data.

Pros: Generally accurate, applicable to any estimation method

Cons: Computationally demanding.

Asymptotic Maximum Likelihood Confidence (Delta-Method)

Likelihood-Theory: For large samples the sampling distribution of parameters

is multivariate Normal. Variances/covariances are inferred from curvature of
Likelihood surface.
Pros: fast, analytic
Cons: applicable to MLE only, not very accurate for small samples

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

35

Confidence Intervals
90% ML confidence

90% resampling confidence

symmetric

asymmetric

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

36

What goes wrong here?

ML-CI (Delta Method)
Vivian,
Feb 1990

Index for Storminess

in Europe.
ERA40, 1958-2002

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

37

Confidence Intervals

Likelihood-Profile Confidence

Similar to asymptotic ML-CI. Exploits higher moments in the shape of the

likelihood function.
Converges to asymptotic ML-CI for large samples.

Pros:

Cons:

more accurate for small samples, strong shapes, large T

makes better use of information in available data,
also accurate for negative shape.
only for ML estimates, computationally demanding.

Coles 2001
Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

39

ML vs. Likelihood Profile CI

Vivian,
Feb 1990

Likelihood-Profile
ML (Delta Method)

Index for Storminess

in Europe.
ERA40, 1958-2002

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

40

Block Size: What is large enough ?

A trade-off between biases (too small block size) and large sampling
errors (large blocks, small number of blocks).

Desirable block size depends on parent distribution

Smaller for precipitation (close to exponential distribution)

Larger for temperature (close to Normal distribution)

For the Exponential (Normal) parent distributions GEVs from block

sizes >20 (>50) provide a reasonably bias free approximation for
return periods between 5 and 100 times the block size.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

41

Other Issues in Practice

Trends violate iid assumption!

Solution: Parametrization of trends. GEV parameters vary

smoothly with time (Katz et al. 2002).

GEV does not make sense for seasonal means (e.g. summer 2003).
Not a Maximum of a large block!

In a network with many stations you will find events with very large
local return periods every now and then. Citing those results only
is misleading. (Dont fall for sensation journalism!)

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

42

Example: August 2005

MeteoSwiss 2006
Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

43

Section 4: Extreme Value Analysis An Introduction

Modelling of Peaks over Threshold

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

45

The Peak-over-Threshold Approach

Estimate X(T) (for rare extremes)

by parametric modeling of independent

exceedances above a large threshold.

Note: Hydrologists tend to call this the method of Partial Duration Series

Davison & Smith 1990

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

46

Distribution of Exceedances

Independent identically distributed random variables:

X1, X 2 , X3, ..., X N

X k ~ F(x) iid

E.g. daily maximum of wind speed over 30 years of observation

F(x):= prob(Xk x), the parent distribution

The Distribution of Exceedances

u a threshold (e.g. u=10 m/s, about Beaufort 6 or larger)

Exceedances: y=xu
CDF of exceedances Eu(y):

Eu ( y) := prob ( X < u + y X > u)

Eu ( y ) =

F ( x = u + y ) ! F ( x = u)
1! F ( x = u)

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

47

Clarification
prob ( X > u + y X > u) = 1! Eu ( y)
prob ( X > u + y X > u) =
" Eu ( y ) =

1! F ( x = u + y)
1! F(x = u)

F ( x = u + y ) ! F ( x = u)
1! F ( x = u)

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

48

Asymptotic Theorem for Exceedances

If Block Maxima of F(x) asymptote to GEV

prob ( M n < z ) ! GEV ( z; , ! , " )

for n " #

then the distribution of exceedances Eu(y) asymptotes (for large u)

to a limit distribution:
Eu ( y) ! GPD ( y;!! , " )

for u " #
$1 "

%
y(
GPD ( y;!! , " ) = 1$ '1+ ! *
&
!! )

with !! = ! + " + (u $ ) > 0

GPD the Generalized Pareto Distribution

GPD and GEV shape parameters are identical
GPD and GEV scale parameters are related.
See outline of proof in Coles 2001, p.76
Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

49

Generalized Pareto Distribution

Vilfredo Federico Damaso Pareto

1848-1923

The GPD (CDF)

!1 !

"
y%
GPD ( y;! , " ) = 1! $1+ ! '
#
"&
y ( 0, 1+ ! y " ( 0

=1

Two parameters:
Scale (), Shape ()

= 0: Exponential Distribution
< 0: upper bound at /

> 0: no upper bound

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

50

Procedure

Select a threshold u

Extract the exceedances from the dataset

n values out of the total data values

Should be large enough to be in asymptotic limit (see later)

Exceedances need to be mutually independent (see later)

Fit GPD to exceedances, yields conditional distribution:

prob ( X > x X > u) = 1! GPD ( x ! u;! , " )

Estimate unconditional distribution and return values

prob ( X > x ) = prob ( X > u) ! (1" GPD ( x " u;! , " ))

with prob(X>u) estimated as n/N (the third model parameter)

Return values X(T) from the unconditional distribution

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

51

Estimation and CIs

Exactly as with GEV

Parameter Estimation

Maximum Likelihood Method (find maximum of Likelihood Function numerically)

Method of L-Moments (analytical formuli, see Hosking 1990)

Confidence Intervals

Resampling
Asymptotic ML confidence intervals (Delta Method)
Likelihood profile

Hosking 1990, Coles 2001

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

52

The Exceedance Diagram

90% confidence

640 independent
exceedances

Threshold:
30 mm

Max-Likelihood
= 11.3 mm
= 0.07
x(T=200) = 133 mm

Axis in log(T):
Exponential Distribution
( = 0) is a straight line.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

53

Assumptions

Exceedances are identically distributed

Exceedances are independent

May be violated by serial correlation

Much more critical than for block maximum approach
In general resolved by declustering of original data

E.g.: Exceedances should be separated by at least m days (runs-declustering).

May be violated e.g. by seasonality, by trends

May be resolved by stratification, explicit modeling

Sufficiently large threshold (asymptotic limit)

Depends on the parent distribution (fast/slow convergence)

Use additional diagnostics for appropriate choice (see later)

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

54

Threshold Selection

Parameter Dependence on Threshold

If exceedances of threshold u are GPD then exceedances of threshold v (>u)

are also GPD with:

GPD(y, ! v , " v ) =

GPD((v ! u) + y, ! u , " u )
GPD((v ! u), ! u , " u )

for all y > 0

Can only be satisfied if:

! v = ! u, " v ! ! " v = " u ! ! " u

At sufficiently large thresholds u :

(a) shape is independent of threshold
(b) modified scale ( u) is independent of threshold

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

55

Threshold Selection
Diagnostics for
threshold selection

approx. stable
within limits

approx. stable
within limits

Engelberg,
daily precipitation (mm)
1901-2010

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Be aware:
Stability within
confidence limits
does not warrant
true stability.
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

56

Threshold Selection
Mean Residual Life Plot:

If exceedances are GPD

prob(Yu < y) = GPD ( y, ! u , ! )

Then expected value of
exceedances is:

linear
within limits

Mean of exceedance (beyond threshold)

of daily precipitation peaks
Engelberg, 1901-2010

E (Yu ) =

!u !" "u
1+ "

I.e. for sufficiently large

thresholds, the mean of the
exceedances depends linearly
on threshold.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

57

Block-Max vs. Peak-over-Thresh

Block Maximum:
GEV, 110 maxima

X(100)
152 mm

Peak-over-Threshold:
GPD, 640 peaks > 30 mm

X(100)
138 mm

127 mm
101 mm

120 mm
105 mm

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

58

Block-Max vs. Peak-over-Thresh

Block Maximum Approach

Peak-over-Threshold Approach

Pros

Pros

Theoretical assumptions are

less critical in practice.
Independence of maxima can
be achieved by selecting large
block size.
More easy to apply.

Cons

Cons

Estimation uncertainties can be

large because sample size is
small.

Smaller CIs if a small

threshold is justified. (More
independent exceedances than
block maxima.)

Independence assumption is
critical in practice. Need
declustering techniques.
Needs diagnostics for threshold
selection. Choice somewhat
ambiguous in practice.
Less easy to apply in practice.

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

59

Section 4: Extreme Value Analysis An Introduction

Additional Remarks

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

60

Additional Remarks
Precipitation Extremes

Dependence of return levels on duration is commonly described by an

Intensity-Duration-Frequency Curves (also Depth-Duration-Frequency) using
scaling relationships between durations.
Bern

Log Precipitation Rate (Intensity)

Return Period (Frequency)

Log Measurement Interval (Duration)

Koutsoyannis et al. 1998; Geiger, Zeller, Rthlisberger, 1990
Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

61

Additional Remarks

Hydrological Extremes

Hydrological processes involve threshold processes. Only very few events

may have been observed from the extreme tail. Very large block size needed.
Statistics alone may be inadequate. Physical modelling of scenarios instead.

Hourly runoff extremes, Langeten

Liechti, 2008
Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

62

Additional Remarks

Spatial Extremes

Utilize concepts of spatial dependence or regional pooling of data to reduce

uncertainties in estimation.
Combine extreme values analysis and spatial statistics to estimate extremes
for unobserved locations.

Cooley et al. 2007, Alila 2007; Davison et al. 2012

Analysis of Climate and Weather Data | Extreme Value Analysis An Introduction | HS 2015 | christoph.frei [at] meteoswiss.ch

63