Nat. Hazards Earth Syst. Sci.

, 7, 165–175, 2007
www.nat-hazards-earth-syst-sci.net/7/165/2007/ Natural Hazards
© Author(s) 2007. This work is licensed and Earth
under a Creative Commons License. System Sciences

Changing European storm loss potentials under modified climate

conditions according to ensemble simulations of the
ECHAM5/MPI-OM1 GCM
J. G. Pinto1 , E. L. Fröhlich1,2 , G. C. Leckebusch3 , and U. Ulbrich3
1 Institut für Geophysik und Meteorologie, Universität zu Köln, Germany
2 Institut für Physik der Atmosphäre, Johannes Gutenberg-Universität Mainz, Germany
3 Institut für Meteorologie, Freie Universität Berlin, Germany

Received: 18 September 2006 – Revised: 18 December 2006 – Accepted: 30 January 2007 – Published: 8 February 2007

Abstract. A simple storm loss model is applied to an ensem- 1 Introduction

ble of ECHAM5/MPI-OM1 GCM simulations in order to es-
timate changes of insured loss potentials over Europe in the Along with floods (e.g. Elbe River 2002, cf. Ulbrich et al.,
21st century. Losses are computed based on the daily maxi- 2003a,b), winter storms associated with the passage of in-
mum wind speed for each grid point. The calibration of the tense extra-tropical cyclones are one of the primary natural
loss model is performed using wind data from the ERA40- hazards over Europe (SwissRe, 2000). During December
Reanalysis and German loss data. The obtained annual losses 1999, three severe storms hit Europe (e.g., Ulbrich et al.,
for the present climate conditions (20C, three realisations) re- 2001), causing insured losses above 10 billion EUR (Mu-
produce the statistical features of the historical insurance loss nichRe, 2001). The total economic losses were roughly
data for Germany. twice as much. The insured loss attributed to only one of
those storms (Lothar, 26 December 1999) amounted to 5.9
The climate change experiments correspond to the SRES-
billion EUR (11.3 billion EUR for economic loss), primar-
Scenarios A1B and A2, and for each of them three realisa-
ily in France (MunichRe, 2001). In fact, MunichRe (1999,
tions are considered. On average, insured loss potentials in-
2001) estimate that storm events were responsible for more
crease for all analysed European regions at the end of the
than 50% (60%) of total economic (insured) losses over Ger-
21st century. Changes are largest for Germany and France,
many for the period 1970–1999. Furthermore, a significant
and lowest for Portugal/Spain. Additionally, the spread be-
increase in losses has been detected during the second half of
tween the single realisations is large, ranging e.g. for Ger-
the 20th century (Berz, 2001). Even though this loss increase
many from −4% to +43% in terms of mean annual loss.
is mainly ascribed to socio-economic factors (e.g., economic
Moreover, almost all simulations show an increasing inter-
and demographic growth), variations of meteorological con-
annual variability of storm damage. This assessment is even
ditions can also contribute considerably to modify both the
more pronounced if no adaptation of building structure to
number and strength of natural disasters (e.g., Berz, 2001;
climate change is considered. The increased loss poten-
Emanuel, 2005; Webster et al., 2005; SwissRe, 2006). More-
tials are linked with enhanced values for the high percentiles
over, there are first indications that losses in Europe could in-
of surface wind maxima over Western and Central Europe,
crease during the 21st century (SwissRe, 2006; Leckebusch
which in turn are associated with an enhanced number and
et al., 2007).
increased intensity of extreme cyclones over the British Isles
and the North Sea. Temporal changes of storm activity over Europe can be
quantified by analysing the frequency distributions of cy-
clones and/or local wind speeds. This aspect has been ex-
plored in many recent publications based on observational
datasets: While some authors (e.g., Schinke, 1993; Lam-
Correspondence to: J. G. Pinto bert, 1996; Geng and Sugi, 2001; Paciorek et al., 2002) iden-
(jpinto@meteo.uni-koeln.de) tified significant changes in extra-tropical cyclone activity

Published by Copernicus GmbH on behalf of the European Geosciences Union.

166 J. G. Pinto et al.: Changing European storm loss potentials

during the latter half of the 20th century in the Northern Hence, we analyse changes in insured loss potentials over
Hemisphere, other authors (e.g., von Storch et al., 1993; Europe between the current and future climate based on the
WASA-Group, 1998; Bärring and von Storch, 2004) objected SRES scenarios A1B and A2 (Nakićenović et al., 2000) con-
that these trends may be unreliable because of the inhomo- sidering three simulations per experiment. These simulations
geneities in data series, and that the recent increase in activity have been analysed in terms of changes of synoptic activity
could be a part of the strong natural inter-decadal variations and cyclone characteristics in Pinto et al. (2007).
of the storm climate. The storm loss model is calibrated with recent ERA40-
Within this context, it is important to analyse to what ex- Reanalysis data. As investigations of Palutikof and Skellern
tent frequency and intensity of winter storms may change (1991) and Klawa and Ulbrich (2003) imply that the 98th
under the enhanced GHG forcing conditions in the current percentile of the daily maximum wind speed is a reason-
(21st) century (IPCC, 2001). Future changes of storm activ- able threshold for initiation of damage, special attention was
ity have been found in global climate model (GCM) simula- given to changes in this percentile of local wind speed in
tions (e.g., Lambert, 1995; Carnell et al., 1996; Knippertz Pinto et al. (2007). Their results show an increase of the 98th
et al., 2000; Lambert et al., 2002; Geng and Sugi, 2003; percentile values over Western Europe for both the A1B and
Raible and Blender, 2004; Leckebusch and Ulbrich, 2004; A2 experiments. Here, we further explore the consequences
Lambert and Fyfe, 2006; Leckebusch et al., 2006; Bengts- for these enhanced wind speeds to insurance losses over Eu-
son et al., 2006; Pinto et al., 2006) conducted with pre- rope, considering the following regions: Germany, United
scribed enhanced greenhouse gas forcing according to differ- Kingdom, France, Norway/Sweden and Portugal/Spain.
ent SRES scenarios (IPCC, 1996, 2001; Nakićenović et al., Information about data sets is given in Sect. 2, followed
2000). Many of these climate change studies point to an in- by a detailed description of the storm loss model in Sect. 3.
crease in the number of intense low pressure systems over the Section 4 analyses extreme winds and storm loss values for
North Atlantic, while the frequency of weak cyclones may Germany for present climate conditions in both for Reanaly-
be reduced. However, uncertainties with respect to the mag- sis and GCM data. Changes in cyclone activity and extreme
nitude of the changes remain, as the results are sensitive to winds in a future climate are presented in Sect. 5. The corre-
both the choice of data as well as the choice of methods. Typ- spondent changes in storm losses are analysed in Sect. 6. A
ically, the extreme wind gusts are associated with the occur- summary and a short discussion conclude this paper.
rence of deep and intense cyclones. Knippertz et al. (2000)
associated the enhanced number of extreme wind situations
in one ECHAM4 GHG simulation with an augmented num- 2 Data
ber of deep cyclones (particularly under 970 hPa) over North-
ern Europe adjacent ocean areas. More recently, Leckebusch 2.1 Climate simulations of ECHAM5/MPI-OM1
et al. (2006) have juxtaposed the changes in the 5% strongest
cyclones (measured by the intensity of circulation) and for The analysed climate change experiments were performed
all systems, and associated these with changes in higher with the ECHAM5/MPI-OM1 GCM, a coupled atmosphere-
wind percentiles for several GCMs (ECHAM4, ECHAM5, ocean model developed at the Max-Planck-Institute (MPI)
HadCM3, HadAM3P). Results for the 5% strongest cyclones for Meteorology in Hamburg (Germany). The atmospheric
indicate enhanced cyclone activity for western parts of Cen- component is the spectral model ECHAM5 (Roeckner et al.,
tral Europe in all GCM simulations, which is consistent with 2003) at T63 horizontal resolution, which corresponds to a
enhanced frequency of wind extremes over large parts of Eu- spatial resolution of circa 1.875◦ ×1.875◦ (∼200 km). The
rope (Leckebusch et al., 2006). oceanic component MPI-OM1 includes a dynamic sea ice
While a wide range of publications focuses on the past and model (Marsland et al., 2003). Compared to older versions
future climatology of storms, little has been publicly avail- of the coupled ECHAM-model, the ECHAM5/MPI-OM1 is
able in terms of interdisciplinary approaches, e.g. combining able to provide a stable climate without flux adjustment.
meteorological and insurance aspects of storms (Klawa and An ensemble of realisations with different forcings is con-
Ulbrich, 2003). This fact may be attributed to the lack of sidered. For the recent climate, the simulations (hereafter
loss data (which is required to estimate the extent of dam- 20C) were computed with radiative forcing according to his-
age caused by severe storms) available for researchers. The torical GHG and aerosol concentrations for the period 1860–
methodology used here was initially developed by Klawa and 2000. The initial conditions for the three runs are three dif-
Ulbrich (2003) and has recently been used by Leckebusch ferent states of the 500-yr-long pre-industrial control simu-
et al. (2007) to assess changes of loss potentials over Ger- lation, which was computed with constant 1860 GHG con-
many and England in a multi-model approach. While that centrations. For future climate conditions, two experiments
study considered only one simulation per GCM, we focus are considered, following the SRES scenarios A1B and A2
here on an ensemble of climate change simulations of the (Nakićenović et al., 2000) for the period 2001–2100. The
ECHAM5/MPI-OM1 GCM in order to estimate the sensi- CO2 concentrations increase from 367 ppm (year 2000) to
tivity of the results against the GCMs internal variability. 703 ppm and 836 ppm by the year 2100 for scenarios A1B

Nat. Hazards Earth Syst. Sci., 7, 165–175, 2007 www.nat-hazards-earth-syst-sci.net/7/165/2007/

J. G. Pinto et al.: Changing European storm loss potentials 167

and A2, respectively. The scenario simulations are initial- in 1990 (with five storms in January and February, including
ized with the final states of the 20C simulations, and there- “Daria“, 26 January 1990) stand out with very high values
fore they may be considered as extensions of the previous (cf. Sect. 4). No significant temporal trend of storm damage
runs. Analysis period is the winter half year (ONDJFM), is revealed in the time series.
when intense storms typically occur. This is an appropriate The population density distribution is used to estimate the
simplification due to the fact that 98% of all storm damage is distribution of insured values within a single region/country
ascribed to winter storms (MunichRe, 1999). Climate signals (see next section). Information about worldwide population
refer to the changes between end of the 21st century (2060– numbers are provided by the United Nations Environment
2100) and recent climate conditions (1960–2000). Program (UNEP/GRID, 2006) resolved on scale of 1◦ ×1◦
For the wind data, particular attention is given to the grid boxes.
daily maximum near-surface (10 m) wind speed (hereafter
wimax), which is based on the internal time step of the at-
mospheric model (circa 15 min). Additionally, the largest 3 Methods
daily value of the four 6-hourly wind speed values (at re-
3.1 Storm loss model
ported time, based on a 10 min average, not maximum of
the last 6 h, hereafter w6h) is analysed for comparison with The statistical model used to estimate monetary storm loss
ERA40 data. to buildings is based on the daily maximum wind speed.
A complete description can be found in Klawa and Ulbrich
2.2 ERA40-reanalysis (2003), but a short description of its main features is given
below. The construction of the storm loss model is based on
Near-surface (10 m) wind data of the ERA40-Reanalysis
four main assumptions about the occurrence of storm dam-
(hereafter ERA40, cf. Uppala et al., 2004) provided by
age:
European Centre for Medium Range Weather Forecasts
(ECMWF) are used for model calibration and validation. 1. Storm damages are produced when a certain, locally
The meteorological surface data from ERA40 is available adapted threshold of wind speed is exceeded. This as-
on gaussian resolution N80 (∼1.125◦ ×1.125◦ ) for the period sumption accounts for the adaptation of buildings to the
September 1957 to August 2002 with a time step of 6 h. The local conditions. This threshold is approximately equal
period 1960 to 2000 is considered. Unfortunately, no daily to the local 98th percentile (v98 ) of the daily maximum
maximum wind speed (similar to GCM wimax) is available. wind speed.
Therefore, the largest daily value of the four 6 hourly wind
speed values of a day is taken for model calibration. See 2. Loss extent of storms increases with the cube of the
Sects. 3 and 4 for further discussion of this subject. maximum daily wind speed above the threshold. This
exponent is in line with physical argument (the cube of
2.3 Insurance data and population density wind speed is proportional to the advection of kinetic
energy) and empirical evidence (MunichRe, 2001).
Our loss model is calibrated based on data provided by
the German Insurance Association (“Gesamtverband der 3. Storm damages depend on the local wind climate, and a
Deutschen Versicherer“, GDV), which provided yearly sums certain excess of the maximum wind over the percentile
of loss ratio in Germany for the period 1970 to 1997. They value will have less effect where high wind speeds are
include storm loss related to private buildings and are col- common. Thus, the daily maximum wind speed excess
lected from the member organisations of GDV. The relation- is scaled by the local 98th percentile (v98 ) value.
ship between insured claims and totally insured values is re- 4. Insured loss depends on the spatial distribution of in-
ferred to as loss ratio. The time series is nearly inflation- sured property values. As this information is not avail-
adjusted, although the German reunification of 1990 may able, we assume that insured property values are pro-
contribute to slight changes in the loss sums in the early portional to population density (pop).
1990s (Klawa and Ulbrich, 2003). Large losses are usu-
ally caused by intense storm events which affect large areas. Based on these assumptions, a so-called daily loss index is
Summer wind storm events play a negligible role in the an- computed:
nual loss sums used (Klawa and Ulbrich, 2003), and other  3
local events (torrential rain, landslides) are not considered in v(area)
loss(area) ∝ pop(area) · −1 (1)
this investigation. Hence, the time series of annual loss ra- v98 (area)
tio is mainly characterised by the occurrence or absence of
| {z }
= loss index
such severe storm events within each individual year. Con-
sequently, the loss ratios in 1972 (“Niedersachsen-Orkan“, 3 As a consequence, wind speeds which are scaled with the lo-
November 1972), 1976 (Gale “Capella“, 3 January 1976) and cal 98th percentile provide a measure of the storm intensity

www.nat-hazards-earth-syst-sci.net/7/165/2007/ Nat. Hazards Earth Syst. Sci., 7, 165–175, 2007

 damage (derived from the ERA40 wind data) for each grid
point was accumulated both in time (over one year) and space standards), the climate signals can be expected to give at least
(over all grid points in Germany), resulting in an annual loss a rough estimate of the expected changes. The alterations of
time series. The linear regression of observed annual loss loss potentials for these regions due to climate change will
ntials 168
ratio of storm damage to buildings in Germany and the ac- 3 be
J. regarded in al.:
G. Pinto et section six. European storm loss potentials
Changing
cumulated loss index are given in Fig. 1. The obtained re-
80
gression coefficient A (6.82·10−2 ) is assumed to be approxi- 80
annual loss ratio [0.01 € per 1000 €]

observations (GDV)
70
mately the factor of proportionality in Eq. 1, and the constant 70
model output (ERA40)
nalysis regression
B60(6.309) is the axis equation: For
intercept. y = 0.0682*x+6.309
the estimation of annual

loss ratio [0.01 € per 1000 €]

60
ed by storm
50
loss values for Germany (Ger), the complete model
ecasts can be summarised by the following equation: 50
40
dation. 40
0 1
30
spec- Byear „ Ger
B X v(area, day)
«3 C 30
20 X C
period loss ≈ A· pop(area)·B − 1 C+B (2)
10
B v 98 (area) C 20
h. The area @ day | {z }A
0 for v > v98 else 0 10
o daily 0 100 200 300 400 500 600 700 800
ilable. Loss potentials are annual
givenlossinindex for Germany
terms of the[~vloss
max/v98] ratio, defined as
0

1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980

1981

1982
1983
1984
1985
1986
1987
1988

1989

1990
1991
1992
1993
1994
1995
1996
1997
y wind the ratio between insured claims and totally insured values,
. See Fig.
and1.1.
Fig. Linear regression
Linear
typically regression
have between
between
an order yearly and
yearly
of magnitude andofregionally
regionally
10−5 (0.01 accumu-
accumu-
EUR
lated loss indices (ERA40) and annual loss ratio in Germany (GDV) Fig. 2. Annual accumulated loss ratio in Germany for the time pe-
lated loss indices (ERA40) and annual loss
damage per 1000 EUR insured). These relative amounts ratio in Germany (GDV)
are
forthe
the period1970-1997.
1970–1997. 1970-1997. Comparison between historical loss data (GDV)
riod of 1970–1997.
for
more period
relevant to the insurance companies than absolute val- and ERA40 derived estimations.
ues as they are not affected by inflation rates and hence al-
low long-term comparisons. Here, we compare the obtained
ed by which
four mainis largely independent
assumptions about of
thelocal conditions
occurrence in a particu-
yearly sums of storm loss in Germany with theof storm
insurance dam-
data lation betweenwinds
4 Extreme both (correlation
and storm loss coefficient:
for the present is iden-
r=0.87)climate
nd der lar area. Note that the use of the cube of the excess over the
age:
of the GDV in the period 1970-1997 (Fig.2). A high corre- tified, corresponding to an explained variance of 76%.
y sums threshold v98 also implies that an absolute increase of wind
lation between both (correlation coefficient: r = 0.87) is In The
thisstorm
section,
lossresults
model as based on simulated
described in Eq. (2) winds
can nowof thebe
They speeds
1. Storm above that threshold
damages are producedwill have when a strong
a certain,(and locally
highly
identified, corresponding to an explained variance of 76 %. present to
applied climate
GCM (20C)data. As arethevalidated
model was against those with
calibrated obtained
w6h
e col- non-linear)
adapted effect in loss.
threshold of wind speed is exceeded. This as-
The storm loss model as described in Eq. 2 can now be ap- with we
data, ERA40 data. As pointed
first considered the sameout in thefor
variable introduction,
the GCM. Sec- the
lation- sumption accounts for the adaptation
plied to GCM data. As the model was calibrated with w6h of buildings to the 98th percentile
ondly, we applied of local wind isdata.
it to wimax particularly
As the important
results of theto eval-
loss
3.2 local
Model adjustment
conditions. This
s is re- data, we first considered thethreshold
same variable is approximately
for the GCM.equal Sec- uate changesbased
calculations in theonstorm
both GCMclimate,windasvariables
it corresponds
showedtoneg- the
flation- to the local 98th percentile (v
ondly, we applied it to wimax data. As the results98 ) of the daily maximum
of the loss threshold
ligible beyond which
differences losses
(see next may beweexpected.
section), decided to Therefore,
concen-
0 may The wind
loss index
speed. (which can be considered a “raw” damage)
calculations based on both GCM wind variables showed neg- we first
trate compare
on wimax the 98th
because percentile
it includes thefields
actual ofinformation
ERA40 to the on
early must be calibrated with loss data from historical storms. As
ligible differences (see next section), we decided to concen- ensemble
the averages
gust wind speeds of (unlike
20C forw6h).the period
The 1960-2000
model validationin terms is
reliable
2. Loss loss data from
extent individual
of storms eventswith
increases are not available,
cube of the
sually trate on wimax because it includes the actualthe information on of the w6h
presented in wind
the next(Fig 3a,b). The spatial distributions are
section.
available
maximum yearly dataset
daily wind is used.
speed The
above daily raw damage (de-
reas. the gust wind speeds (unlike w6h). Thethe model threshold.
validation This
is mainly influenced
Estimations by land-sea-contrasts
of storm damage for otherwith largestregions
European values
he an- rivedexponent
from theisERA40 in line windphysical
with data) forargumenteach grid point
(the cubewasof
presented in the next section. overperformed
are the Norththe Atlantic
followingstorm way:track region.
First, Thewind
the local differences
clima-
accumulated
wind speed bothisinproportional
time (over one year) and spaceof(over all
other Estimations of storm damage fortoother the advection
European kinetic
regions are between and
tologies bothpopulation
datasets aredensities
small, with are the
used GCM overestimat-
(following steps
grid energy)
points inand Germany), resulting
empiricalway: evidence in an annual loss time se-
ered in performed the following First,(MunichRe,
the local wind 2001). clima- ing the
1–4 values over
described the North
above). Atlantic
Second, the storm
modeltrack and under-
calibration ob-
oss ra- ries. The linear regression of observed annual loss ratio of
tologies and population densities are used (following steps estimating
tained them over(i.e.
for Germany the adjusted
Eurasiantocontinental
the German areas
loss(range
values) is
storm damage
3. Storm to buildings
damages depend inonGermany
thethe local and
wind theclimate,
accumulateda
nce of 1-4 described above). Second, model calibrationand ob- between
is used (cf.+/-Fig.
4 m/s), particularly
1). This occurs due overtohigh orography.
the lack of loss Fur-
data
loss certain
index are givenofinthe
excess Fig.maximum
1. The obtained regression coeffi-
Con- tained for Germany −2 (i.e. adjusted to the Germanpercentile
wind over the loss val- thermore,
for large deviations
the analysed countries. In areprecise
perceptible
terms,along the Euro-
this means that
cientvalue
A (6.82×10
will(cf. )less
haveFig. is assumed
effect to be approximately
where highdue wind the fac-
speeds
rkan“, ues) is used 1). This occurs to the lackareof pean the
only coast lines which
transfer function result primarely
between normed fromwinddifferent
to lossspa-is
tor ofcommon.
proportionality in Eq. (1), and the
Thus, the daily maximum wind speed excess constant B (6.309)
n 1990 derived from German data, all other information are those
is theisaxis
scaledintercept.
by the localFor the98th estimation
percentileof(vannual storm loss
98 ) value.
Daria“, which should be applied for this region. Even though this
values for Germany, the complete model can be summarised
ection simplification produces loss assessments which may differ
by4.the following
Insured loss equation:
depends on the spatial distribution of in-
is re- from real loss values (e.g., because of different building stan-
sured property values.  As this information  is not avail-
dards), the climate signals can be expected to give at least a
able, we assume that insured property values are pro-
ate the rough estimate of the expected changes. The changes of loss
Ger to population  year density
3 
portional (pop).
X 
ountry
X v(area, day)
potentials for these regions will be regarded in section six.

loss≈A · pop(area)·  −1  +B (2)
ulation area

day
v98 (area) 
Based on these assumptions,| a so-called
 {z daily }loss index is

nment computed: for v>v98 else 0
1◦ x1◦ 4 Extreme winds and storm loss for the present climate
Loss potentials are given„in v(area)
terms of the loss
«3 ratio, defined as
loss(area) ∝ pop(area)
the ratio between · claims and−totally
insured 1 insured values,(1) In this section, results based on simulated winds of the
v98 (area)
and typically|have an order {z of magnitude of} 10−5 (0.01 EUR present climate (20C) are validated against those obtained
loss indexThese relative amounts are
damage per 1000 EUR=insured). with ERA40 data. As pointed out in the introduction, the
more
As relevant to the
a consequence, insurance
wind speeds companies than absolute
which are scaled with theval-
lo- 98th percentile of local wind is particularly important to eval-
cal 98th percentile provide a measure of the storm hence
ues as they are not affected by inflation rates and al-
intensity uate changes in the storm climate, as it corresponds to the
m loss low long-term comparisons. Here, we compare the
which is largely independent of local conditions in a particu-obtained threshold beyond which losses may be expected. Therefore,
speed. yearly
lar area.sums
Noteofthat
storm
theloss
useinofGermany
the cubewith theexcess
of the insurance
overdata
the we first compare the 98th percentile fields of ERA40 to the
Ulbrich threshold v also implies that an absolute increase ofcorre-
of the GDV in the period 1970–1997 (Fig.
98
2). A high wind ensemble averages of 20C for the period 1960–2000 in terms
given speeds above that threshold will have a strong (and highly
sed on non-linear) effect
Nat. Hazards in Syst.
Earth loss. Sci., 7, 165–175, 2007 www.nat-hazards-earth-syst-sci.net/7/165/2007/
 very high spatial coherence, e.g. differences over Germany attribut
are smaller than 2% (data not shown). with w
As expected, wimax is larger than w6h for the whole study semble
area (cf. Fig 3b,c). Differences between the 98th percentiles values
J. G. Pinto et al.: Changing European storm loss potentials of the two GCM wind variables are largest over the main 169 m/s) ar
North Atlantic storm track region, while they are almost neg- the GD
ligible
Table 1. Validation of loss ratios for Germany based on observations (GDV) over continental
(1970–1997), areas(1960–2000)
ERA40-data e.g. Central andEurope. As the
ECHAM5/MPI- ratio de
normalised
OM1 GCM data for present climate conditions (1960–2000). ea means ensemble wind(runs
average distributions
1–3). Units(with 98th
in 0.01 percentile)
C per 1000 C. for indi- 4. Plea
vidual grid points also show similar statistical characteristics than Fi
Observations (GDV) for both variables (data
ERA40 not shown), it is acceptable to use the
ECHAM5 two run
model calibrated with w6h data with wimax data. the app
run 1 2 3 ea
Next, we consider the annual loss estimations over Germany. respond
mean 16.97 16.98 18.87 18.70 17.99 18.52
standard deviation 15.72
The mean13.12
loss ratio of the13.79
14.26
GCM simulations
13.73
is on average sults in
13.27
loss da
250

of the w6h wind (Figs. 3a, b). The spatial distributions are (a)

loss ratio [0.01 € per 1000 €]

mainly influenced by land-sea-contrasts with largest values 200

over the North Atlantic storm track region. The differences

between both datasets are small, with the GCM overestimat- 150

ing the values over the North Atlantic storm track and under-
estimating them over the Eurasian continental areas (range is 100

between +/−4 m/s), particularly over high orography. Fur-

thermore, large deviations are perceptible along the Euro- 50

pean coast lines which result primarely from different spa-

tial resolutions between ERA40 and the GCM. The 98th per- 0

centile distribution for the three 20C ensemble runs show a

1960
very high spatial coherence, e.g. differences over Germany
are smaller than 2% (data not shown).
(b)
Fig. 4. T
As expected, wimax is larger than w6h for the whole study simulati
area (cf. Figs. 3b, c). Differences between the 98th per- conditio
centiles of the two GCM wind variables are largest over the
main North Atlantic storm track region, while they are al-
most negligible over continental areas e.g. Central Europe.
As the normalised wind distributions (with 98th percentile) 5 Ch
for individual grid points also show similar statistical charac-
In this
teristics for both variables (data not shown), it is acceptable
ditions
to use the model calibrated with w6h data with wimax data.
to the
Next, we consider the annual loss estimations over Ger-
many. The mean loss ratio of the GCM simulations is on (c) the cha
sider th
average circa 10% larger than both the observational and
over th
ERA40 derived loss data (Table 1). This discrepancy cannot
storm s
be solely attributed to the use of wimax, as the 20C mean loss
and int
value with w6h is also larger than ERA40 and GDV values
sults w
(ensemble average: 18.10×10−5 C). In terms of standard de-
conside
viation, values for both wimax and w6h (ensemble average:
scribed
13.07×10−5 C) are close to the ERA40 values, and slightly
be foun
lower than the GDV data (cf. Table 1). The time series of
change
annual loss ratio derived for the first ensemble member is
clones
shown in Fig. 4. Please note that this figure has a different
tion giv
vertical scale than Fig. 1. Its main features are comparable
clones
to the other two runs and no temporal trends are detected. In Fig.
summary, the application of the storm loss model to the GCM Fig. 3.
3. 98th
98th percentile
percentile of
of 10
10 m
m wind
wind speed
speed for
for (a)
(a) ERA40
ERA40 data
data (in-
(in- flanks,
terval:
terval: 3 [m/s]) (b) ECHAM5/MPI-OM1 GCM data for the present
3 [m/s]) (b) ECHAM5/MPI-OM1 GCM data for the present fore, it
data correspondent to recent climate conditions provides re- climate
climate (1960–2000), w6h, ensemble
(1960-2000), w6h, ensemble average
average (c)
(c) same
same as
as(b) but for
b) but for and int
alistic results in comparison to observational near data and wimax. Panel (c) adapted from Pinto et
wimax adapted from Pinto et al. (2006b). al. (2007).
change
historical loss data.

5 Changing storms in a future climate

In this section, GCM results derived from future climate con-

ditions (end of the 21st century) are analysed and compared

www.nat-hazards-earth-syst-sci.net/7/165/2007/ Nat. Hazards Earth Syst. Sci., 7, 165–175, 2007

 the application of the storm loss model to the GCM data cor-
ermany. respondent to recent climate conditions provides realistic re-
average sults in comparison to observational near data and historical
170
loss data. J. G. Pinto et al.: Changing European storm loss potentials
250

ECHAM5 run 1 (20C) Typically, loadings are higher than in A1B experiments. The
zonally elongated area of increased extreme winds reaching
loss ratio [0.01 € per 1000 €]

200

from Europe into the Asian continent is found in all ensem-

150
ble members. Hence, Western Europe is expected be more
exposed to the influence of extreme wind storms than under
the present climate conditions. This result is attributed to the
100
changes detected for extreme cyclones near the British Isles
(Figs. 5a, e).
50
Considering now other regions, the most striking feature is
a widespread decrease of the extreme wind speeds at lower
0
latitudes (e.g. Mediterranean) and at polar latitudes (e.g., be-
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
tween Greenland, North Scandinavia and Spitsbergen Isles).
This feature is detected in all simulations (with changing
Fig. 4. Time series of annual loss ratios for Germany, first ensemble magnitudes), showing a very coherent spatial pattern. The
simulation from ECHAM5/MPI-OM1 GCM with present climate
changes in the polar regions are also related to alterations in
(1960-2000).
conditions (1960–2000).
the surface roughness which in turn evolves from the retreat
of sea ice.

to
5 the results from
Changing storms theinprevious
a futuresection.
climate Before analysing
the changes in storm loss potentials for Germany, we con- 6 Changing loss in a future climate
sider
In thisthe changes
section, GCM in storm
resultsactivity
obtained andfor
offuture
the extreme
climatewinds
con-
over
ditionsthe(end
North of Atlantic region and
the 21st century) areEurope.
analysedIn andterms of the
compared Based on these wind speed signals, increased losses over
storm
to the system
results activity,
from thethe changes
previous in cyclone
section. Beforetrack density
analysing Western and Central Europe and decreased losses for north-
and intensityin
the changes ofstorm
cycloneslossare of particular
potentials interest. Such
for Germany, we con-re- ern Scandinavia and the Mediterranean may be intuitively ex-
sults
sider were derivedinbystorm
the changes Pintoactivity
et al. (2007)
and offorthethe simulations
extreme winds pected. This train of thought implicitly assumes that build-
considered
over the North in this paper region
Atlantic with theand cyclone
Europe. trackInalgorithm
terms of de-
the ings will have the same characteristic weak points with re-
scribed
storm systemin Pinto et al. (2005).
activity, the changesFurther technical
in cyclone details
track can
density spect to wind as today, and no adaptation takes place to a
be
andfound
intensityin those references.
of cyclones are ofHere, we concentrate
particular interest. Suchon the
re- changed wind climate. The alternative, a rapid and complete
changes
sults were ofderived
frequency and intensity
by Pinto of thefor
et al. (2006b) 5%thestrongest cy-
simulations adaptation to a changing wind climate, must also be taken
clones
considered(hereafter
in this “extreme
paper withcyclones“)
the cyclone following the defini-
track algorithm de- into account. Both possibilities are considered using the loss
tion
scribedgiven in Leckebusch
in Pinto et al.Further
et al. (2005). (2006).technical
Such extreme cy-
details can model: Without adaptation, the 98th wind percentiles remain
clones
be found typically
in thosefeature strongHere,
references. pressure gradients onontheir
we concentrate the those of the present day climate, while with adaptation those
flanks,
changeswhich in turn produce
of frequency the large
and intensity wind
of the 5%gusts. There-
strongest cy- of the 21st century are used. The results for ensemble av-
fore,
clonesit (hereafter
is reasonable to assume
“extreme that changes
cyclones“) followingin the
thenumber
defini- erages consider the three individual runs as one single ex-
and
tion intensity of such extreme
given in Leckebusch et al.cyclones
(2006a). are responsible
Such extreme cy-for tended simulation with 120 years of data. Changes in loss
changes in the higher
clones typically featurewind percentiles.
strong pressureThe connection
gradients be-
on their potentials for Germany considering both individual runs and
data (in- tween
flanks, deeper
which cyclones and higher
in turn produce extreme
the large windwind speeds
gusts. There- in ensemble means are presented in Table 2, giving percentage
present GCMs
fore, it has been previously
is reasonable to assume documented
that changes e.g. inbythe
Knippertz
number changes of the mean and standard deviation of the loss time
) but for et
andal.intensity
(2000) and of Leckebusch
such extremeet cyclones
al. (2006).are responsible for series. Significant differences (Students t-Test 95% confi-
changes
Figure 5 includes the climate signal ofThe
in the higher wind percentiles. connection
cyclone be-
character- dence level) between the present and future loss data are de-
istics and 98th wind percentile for members of the A1B en- noted by shaded boxes. If adaptation is assumed, there is no
semble (upper four panels). Changes of cyclone track den- significant change in terms of mean values, though deviations
sity and cyclone intensity of extreme cyclones for the first are positive in most cases (ranging from −6% to +45%).
run (Fig. 5a) show an increase in the number of intense cy- In ensemble average, the changes are +6.3% for A1B and
clones between Scotland and Scandinavia, which goes very +13.3% for A2. If no adaptation is considered, changes are
well with the significantly enhanced values of the 98th wind positive for all simulations, and the changes are significant in
percentile over Western and Central Europe (Students t-Test ensemble average, reaching +40.2% for A1B and +49.1% for
95% confidence level) for the same run (Fig. 5b), particu- A2. More important, the changes of the standard deviation
larly over the North and Baltic Sea and the adjacent areas. are larger and significant in 5 from 6 cases with full adapta-
The second run shows a similar pattern of change with lower tion, even exceeding 100% for the first A2 simulation. The
loadings (wind percentiles shown in Fig. 5c), while the third changes in ensemble mean reach circa +50% and +74% for
run shows little significant results (Fig. 5d). In spite of these the A1B and A2 experiments. As expected, if no adaptation
differences, it appears that the signal structure is similar. The to future storm conditions is considered, the changes of the
impression of similarity is further supported by the results standard deviation of annual loss potentials are even more
obtained from the A2 experiments (Fig. 5, lower four panels). pronounced (over 100% for the ensemble averages).

Nat. Hazards Earth Syst. Sci., 7, 165–175, 2007 www.nat-hazards-earth-syst-sci.net/7/165/2007/

J. G. Pinto et al.: Changing European storm loss potentials 171
6 J. G. Pinto et al.: Changing European storm loss potentials

(a) (b)

(c) (d)

(e) (f )

(g) (h)

Fig. 5. Climate signals for characteristics of extreme cyclones and 98th percentile of 10 m wind speed (wimax) between present (1960–2000)
and future
Fig. 5.(2060–2100)
Climate signalsclimate conditions
for track densities,based on the
intensities of ECHAM5/MPI-OM1
extreme cyclones and 98th simulations.
percentile of(a)
10cyclone track between
m wind speed density present
A1B first run minus
(1960-
20C first run and
2000) (interval: 0.2 [cycloneclimate
future (2060-2100) days/winter]).
conditionsRed isolines
based on thelimit areas where a significant
ECHAM5/MPI-OM1 simulations.change of cyclone
(a) cyclone trackintensities
density A1B is detected
first run (0.05:
95%-confidence levels;
minus 20C first run 0.01 same0.2for
(interval: 99%-confidence
[cyclone days/winter]). level). (b) 98th
Red isolines percentile
limit for awimax
areas where wind
significant speedof(interval:
change 0.3 [m/s]),
cyclone intensities A1B first run
is detected
minus (0.05:
20C first95%-confidence
run (c) samelevels;as (b) 0.01 same
but for A1Bfor second
99%-confidence level).as(b)
run (d) same (b)98th
but percentile for wimax
for A1B third run (e)wind speed
same (interval:
as (a) but for0.3A2[m/s]), A1B(f) same
first run.
firstfor
as (b) but runA2minus
first20C first
run. (g)run (c) same
same as (f)asbut
(b) for
but for
A2A1Bsecondsecond
run.run
(h)(d)same
sameasas(f)
(b)but
but for
for A1B third run.
A2 third run (e)
Onsame as (a) butareas
all panels, for A2 firstsignificance
with run.
(f) same
differences as (b)
(95%- andbut99%-
for A2 first run. (g)
confidence sameare
levels) as (f) but for (t-test
in colour A2 second run. (h)
on winter same as (f) but for A2 third run. On all panels, areas with
basis).
significance differences (95%- and 99%- confidence levels) are in colour (t-test on winter basis).

www.nat-hazards-earth-syst-sci.net/7/165/2007/ Nat. Hazards Earth Syst. Sci., 7, 165–175, 2007

172 J. G. Pinto et al.: Changing European storm loss potentials

Table 2. Climate Change Signal of loss ratio for Germany (%). Differences between present (1960-2000) and future (2060-2100) climate
conditions based on the ECHAM5/MPI-OM1 simulations. ea means ensemble average (runs 1–3). Significant differences (95%- confidence
level) are shaded gray (t-test on annual basis).

ECHAM5
scenario A1B A2
run 1 2 3 ea 1 2 3 ea
mean +6.9% +12.1% -0.7% +6.3% +42.9% -3.7% +1.8% +13.3%
with adaption
mean +60.4% +50.7% +8.2% +40.2% +110% +27.0% +11.7% +49.1%
no adaption
standard deviation +85.5% +35.0% +24.7% +49.5% +127% +43.8% +39.3% +73.9%
with adaption
standard deviation +189% +84.6% +34.4% +112% +221% +96.4% +54.9% +137%
no adaption

In order to further explore these changes, we analyse indi- way/Sweden, nonetheless making use of the calibration for
vidual time series of annual loss ratios for Germany. As an Germany. This approximation is acceptable as all five re-
example, time series of some of A1B and A2 runs are shown gions feature a similar economical development, and hence
in Fig. 6. Compared with the time series for the 20C (cf. similar insured values (per capita) may be expected (see also
Fig. 4), the stronger year to year variability is immediately considerations in section three). The climate change signals
apparent: e.g., the time series of the first A1B run (Fig. 6a) for the mean and standard deviation are presented in Table 3
features an exceptional year (2079). The amount of dam- in terms of ensemble averages for A1B and A2. The first
age is about three times the maximum annual loss ratio in impression is that the results are comparable to those for
the control period when adaptation is taken into account, and Germany: There is typically a small increase in mean val-
five times as high without. Note that large loss ratio val- ues (though often not significant), while changes in standard
ues in individual years can also be found in the other runs deviation are significant in almost all cases. In fact, the re-
(e.g. the A2 runs in Figs. 6b, c), though they typically fea- sults for France and the United Kingdom are very similar to
ture smaller magnitudes. These exceptionally high loss ra- those for Germany. Nevertheless, note that for the A2 exper-
tios could arise both from the occurrence of single extraordi- iment a minor (not significant) decrease in loss potentials is
nary storm events (unknown under the present climate condi- obtained for the United Kingdom if full adaptation in con-
tions) or from a multiplicity of strong winter storms similar to sidered. If the corresponding single runs are considering, the
those detected in the present climate. We looked at the daily changes are always smaller than minus 10%.
“raw” damage data e.g. during the year 2079 for the first A1B For Portugal/Spain and Norway/Sweden, no significant
run and compared its characteristics to other years and other changes for average loss are detected, and changes range
GCM-derived time series, particularly those from the 20C only up to +/−10%. Please note that the calculations for
runs. We found that such exceptional years feature an in- Portugal/Spain have the particularity that the 98th percentiles
creased number of intense storms, noticeable as single peaks in the scenario periods are actually lower than the ones for
in the daily time series. However, these storms do not differ 20C. Considering full adaptation, this would mean that at the
drastically in terms of their typical characteristics from other end of the 21st century the basic structure of buildings would
storm events (irrespective of the large magnitudes). There- be worse than today. This assumption may be regarded as
fore, it can be assumed that increases in loss potentials may realistic, as by longer absence of intense storms the incen-
be primarily associated with an increased number of intense tive to maintain (or enhance) the structure of the buildings
storms (agreeing with the results for intense cyclones, cf. sec- (and hence their resistance to storms) is reduced. Changes
tion five) and not necessarily with the occurrence of single in standard deviation are significant for A1B alone. For Nor-
extraordinary events with intensity unknown in the present way/Sweden, a significant increase in standard deviation is
climate. However, this conclusion must be regarded as pre- identified for both scenarios with and without adaptation.
liminary, and more detailed studies are necessary. These as-
pects, particularly in terms of changes in the magnitude of
storms (both in terms of intensity and affected areas) and 7 Summary and conclusions
consequently on their associated return periods will be ad-
dressed in future investigations. We have analysed an ensemble of climate experiments with
In order to estimate climate signals of loss potentials for the ECHAM5/OM1 GCM in order to infer changes in storm
the other European regions, we apply the storm loss model induced loss potentials over Europe. This investigation is
to the United Kingdom, France, Portugal/Spain and Nor- done primarily for Germany, and then extended to other

Nat. Hazards Earth Syst. Sci., 7, 165–175, 2007 www.nat-hazards-earth-syst-sci.net/7/165/2007/

J. G. Pinto et al.: Changing European storm loss potentials 173

Table 3. Climate Change Signal of loss ratio for different European regions (%). Differences between present (1960–2000) and future (2060–
2100) climate conditions based on the ECHAM5/MPI-OM1 simulations. ea means ensemble average (runs 1–3). Significant differences
(95%- confidence level) are shaded gray (t-test on annual basis).

ECHAM5 (ensemble averages)

Portugal / Norway /
region Germany U.K. France
Spain Sweden
scenario A1B A2 A1B A2 A1B A2 A1B A2 A1B A2
mean +6% +13% +8% -4% +3% +34% +8% +10% +1% +4%
with adaption
mean +40% +49% +43% +24% +11% +44% -5% -5% +7% +10%
no adaption
standard deviation +50% +74% +37% +19% +25% +132% +74% +36% +29% +66%
with adaption
standard deviation +112% +137% +119% +85% +37% +156% +63% +11% +58% +89%
no adaption

countries of Western Europe. We used a simple storm loss acteristics to account for the changing local wind character-
model based on the following assumptions (a): Losses in- istics induced by climate change.
crease with the cube of the maximum daily wind speed (b):
The storm loss model has recently been applied to single
Losses only occur if a high climatological wind percentile
runs with the ECHAM4/OPYC3, HadAM3P and HadCM3
(98th) is exceeded (c): Losses are proportional to the differ-
models by Leckebusch et al. (2007). The results indicate
ence between the occurred daily maximum wind value and
an increase in loss potentials for Germany and the United
the 98th percentile (d): Insured loss depends on the spatial
Kingdom with the ECHAM4 and HadCM3 GMs, while the
distribution of insured property values.
HadAM3P has less pronounced changes and even negative
Based on the results for five regions, it can be concluded
changes over Germany. The main difference to the ECHAM5
that a higher year-to-year variability of loss potentials may
results is that the changes in terms of standard deviation are
be expected in a future climate over Western and Central Eu-
typically lower for those three GCMs (except for HadCM3
rope. Without adaptation, average loss tolls increase signif-
for the United Kingdom). Causes for the differences are not
icantly for France, the United Kingdom and Germany. In
clear yet, but a strong coherence between the changes in loss
most cases, the increases are not statistically significant with
potentials and the frequency of extreme cyclones was iden-
adaptation. Also, there is a high spread between the changes
tified for the individual GCMs (cf. Leckebusch et al., 2006):
in storm loss as estimated from the individual runs for the
e.g. the HadAM3P show more moderate changes in terms
same experiment. Considering the changes in cyclone counts
of extreme cyclones than ECHAM5. Hence, the different
and intensity as well as those in daily maximum wind speeds,
climate signals in terms of cyclone activity detected in the
the different runs show many common features (Fig. 5) e.g.
individual GCMs are surely one of the factors corroborating
an enhancement of extreme cyclone intensities over Western
to the dissimilar signals in storm losses obtained for the four
Europe (cf. also Pinto et al., 2007), which in turn induces
GCMs.
an increase of the 98th wind percentiles in a zonally elon-
gated area across Europe. The relationship between wind With respect to the insecurities of our approach, we note
percentiles and storm loss numbers is, however, highly non- that our loss model implicitly assumes that there is no change
linear (cf. Klawa and Ulbrich, 2003), so significant changes in socio-economic factors (e.g. changes in demography and
in extreme wind speeds do not necessarily imply similar re- hence density of insured values) over a period of 100 years.
sults for the accumulated storm losses. However, such factors will surely undergo changes in the fu-
One of the most pertinent results of our study is the in- ture. Missing perspectives of these factors may constitute a
crease in interannual variability of insured losses in a modi- source of uncertainty which is as large as the meteorologi-
fied climate, which in turn indicates a diminution in the re- cal aspects investigated in the current paper. Further studies
turn period of extreme rare events. This effect seems to be as- are necessary in order to assess changes of loss potentials
sociated with an increased probability for years with multiple derived from Regional Circulation Models, which feature a
events and hence of large loss numbers, which may severely higher resolution than GCMs.This is ongoing research e.g.
affect management strategies of the insurance companies. in the ENSEMBLES project. Last but not least, the envi-
This result is valid for all five regions and is particularly large ronmental politics of the industrialized countries will be pri-
for Germany, the United Kingdom and France. The larger marily responsible whether we are heading for a restricted
signals obtained for no-adaptation calculations further indi- climate change (e.g. B1, A1B scenarios) or for a strong
cate the need of a change in political/management terms in change (e.g. A2, A1F scenarios) in the earth’s climate whose
relation of risk strategies and changes in infrastructure char- consequences can hardly be foreseen at present.

www.nat-hazards-earth-syst-sci.net/7/165/2007/ Nat. Hazards Earth Syst. Sci., 7, 165–175, 2007

174
8 J.J. G.
G. Pinto
Pinto et
et al.:
al.: Changing
ChangingEuropean
Europeanstorm
stormloss
losspotentials
potentials

250 work. The authors

which however dowould like todrastically
not differ thank S. Zacharias,
in termsM. of Karremann
their typ-
(a) ECHAM5 run 1 (A1B, w. A.)
ECHAM5 run 1 (A1B, no A.)
and S. Ulbrich for preparing some of the figures.
ical characteristics from other storm events irrespective of
loss ratio [0.01 € pro 1000 €]

200
their large magnitudes. Therefore, it can be assumed that in-
Edited by: A. Mugnai
creases in loss potentials may be primarily associated with
150 Reviewed by: I. Kan and two other referees
an increased number of intense storms (agreeing with the re-
sults for intense cyclones, cf. section five) and not neces-
100
sarily with the occurrence of single extraordinary event with
References
intensity not know in the present climate. However, this con-
50
clusion
Bärring, must
L. andbe regarded
von Storch, as preliminary,
H.: Scandinavianand morestorminess
detailed
studies
sinceareabout
necessary.
1800, These aspects,
Geophys. Res.particularly
Lett., 31, in terms of
L20 202,
0
doi:10.1029/2004GL020441,
changes in the magnitude of storms 2004. (both in terms of inten-
2060
2062
2064
2066
2068
2070
2072
2074
2076
2078
2080
2082
2084
2086
2088
2090
2092
2094
2096
2098
2100
Bengtsson,
sity L., Hodges,
and affected areas)K.and I., and Roeckner, E.:
consequently Stormassociated
on their Tracks and
250 Climate
return Change,
periods will J.beClimate,
addressed 19, 3518–3543, 2006.
in future investigations.
(b) ECHAM5 run 1 (A2, w. A.)
ECHAM5 run 1 (A2, no A.)
Berz,
In orderG.: toNaturkatastrophen
estimate climateimsignals 21. Jahrhundert – Trends for
of loss potentials und
loss ratio [0.01 € pro 1000 €]

200
Schadenpotentiale, DKKV, Extreme Naturereignisse, Bonn-
the other European regions, we apply the storm loss model
Leipzig, pp. 253–264, 2001.
to the United Kingdom, France, Portugal/Spain and Nor-
Carnell, R. E., Senior, C. A., and Mitchell, J. F. B.: An assessment
150 way/Sweden,
of measures nonetheless
of storminess:making simulated usechanges
of the incalibration for
northern hemi-
Germany. This approximation is acceptable
sphere winter due to increasing CO2 , Clim. Dyn., 12, 467–476, as all five re-
100 gions feature a similar economical development, and hence
1996.
similar
Emanuel, insured
K.: values (per capita)
Increasing may also beofexpected
destructiveness tropical (see
cy-
50 alsoclones
considerations
over the in pastsection
30 years,three).Nature,
The climate change
436, 686–688,
signals for the mean and standard
doi:10.1038/nature03906, 2005. deviation are presented in
0 Geng, in
Tab.3 Q.terms
and Sugi, M.: Variability
of ensemble averagesoffor theA1BNorth Atlantic
and A2. The Cy-
2060
2062
2064
2066
2068
2070
2072
2074
2076
2078
2080
2082
2084
2086
2088
2090
2092
2094
2096
2098
2100

clone Activity in Winter Analyzed from NCEP-NCAR

first impression is that the results are comparable to those for Reanaly-
sis data, There
Germany: J. Climate, 14, 3863–3873,
is typically a small2001. increase in mean values
250 Geng, Q. and Sugi, M.: Possible change of extratropical cyclone
(c) ECHAM5 run 2 (A2, w. A.)
ECHAM5 run 2 (A2, no A.)
(though often not significant), while changes
activity due to enhanced greenhouse gases and sulfate aerosols –
in standard de-
loss ratio [0.01 € pro 1000 €]

200
viation are significant in almost all cases. In fact, the results
study with a high-resolution AGCM, J. Climate, 16, 2262–2274,
for France and the United Kingdom are very similar to those
2003.
150
for Germany.
IPCC: Climate Nevertheless,
Change 1995: The notescience
that forofthe A2 experiment
climate change, editeda
minor
by: (not significant)
Houghton decrease
J. T., Meira Filho,inL.lossG.,potentials
Callander, is B.obtained
A., et al.,
100
forIntergovernmental
the United Kingdom Panel ifonfull adaptation
Climate Change,in considered.
Cambridge If
Univer-
thesity
corresponding
press, 1996. single runs are considering, the negative
50
changes are always
IPCC: Climate Change. smaller than 10 Basis,
The scientific %. edited by: Houghton,
ForJ. T.,
Portugal/Spain
Ding, Y., Griggs, andD.Norway/Sweden,
J., et al., Intergovernmentalno significant
Panel on
Climatefor
changes Change,
average Cambridge
loss areUniversity
detected,press, and 2001.
changes range
0
Klawa,
only upM.to and+/- Ulbrich,
10%. Please U.: A model note thatfor the
theestimation
calculationsof storm
for
2060
2062
2064
2066
2068
2070
2072
2074
2076
2078
2080
2082
2084
2086
2088
2090
2092
2094
2096
2098
2100

losses and thehave

Portugal/Spain identification of severe that
the particularity winter
thestorms in Germany,
98th percentiles
Nat.scenario
in the Hazard Earthperiods Syst.areSci.,actually
3, 725–732, lower 2003.
than the ones for
Fig. 6. TimeTime series
series of
of annual
annual loss
loss ratios
ratios for
for Germany
Germany based
based Knippertz, P., Ulbrich, U., and Speth, P.: Cyclone activity and wind
20C. Considering full adaptation, this would mean that at the
on ECHAM5/MPI-OM1 GCM data for future climate climate conditions
conditions speeds in a transient GHG experiment, Clim. Res., 15, 109–122,
end of the 21st century the basic structure of buildings would
(1960–2000).
(1960-2000). (a) (a) A1B
A1B scenarios,
scenarios, first
first ensemble
ensemble member.
member. Grey
Grey 2000.
be worse S.than
Lambert, J.: today.
The Effect This of assumption may be regarded
Enhanced Greenhouse Warming as on
columns for values with adaptation
adaptation,, black columns for values
values with
with
no-adaption. (b) as a) but for A2 scenario, first ensemble
ensemble member
member realistic, as by longer absence of intense
Winter Cyclone Frequencies and Strengths, J. Climate, storms the8,incen-
1447–
(c) as (a) butfor
a) but forA2
A2scenario,
scenario,second
secondensemble
ensemblemember.
member. tive1452,
to maintain
1995. (or enhance) the structure of the buildings
(and hence
Lambert, their
S. J.: resistance
Intense to storms)
extratropical northern is hemisphere
reduced. Changes
winter cy-
in standard deviation
clone events: 1899–1991,are significant
J. Geophys. forRes.,
A1B alone.
101, For Nor-
21 319–21 325,
way/Sweden,
1996. a significant increase in standard deviation is
Acknowledgements.
in Fig. 6b,c), though Thisthey
worktypically
was supported
feature by the European
smaller magni- Lambert, S.
identified forJ.both
and scenarios
Fyfe, J. C.: withChanges in winter
and without cyclone fre-
adaptation.
Union
tudes. Programme Energy, Environment
These exceptionally high loss ratiosand could
Sustainable De-
arise both quencies and strengths simulated in enhanced greenhouse warm-
velopment under contracts EVK2-CT-2001-00118 (MICE)
from the occurrence of single extraordinary storm events (un- and ing experiments: results from the models participating in the
GOCE-CT-2003-505593-ENSEMBLES. We acknowledge the 7 IPCC
Summary and exercise,
diagnostic Conclusions
Clim. Dyn., 26, 713–728, doi:10.
known under the present climate conditions) or from a mul-
provision of annual loss data by the German Insurance Association 1007/s00382-006-0110-3, 2006.
tiplicity of strong winter storms similar to those detected in
GDV. We would like to kindly thank Erich Roeckner and the MPI We have S.
Lambert, analysed anJ.,ensemble
J., Sheng, and Boyle,ofJ.climate
S.: Winterexperiments with
cyclone frequen-
the present climate. We looked at the raw daily loss data e.g. thecies
ECHAM5/OM1 GCM in order toininfer changes in storm
for Meteorology (Hamburg, Germany) by order of the Federal in thirteen models paricipating the Atmospheric Model
during the year
Environment 2079for
Agency forsupport
this runandandproviding
compared theitsECHAM5
charac- induced loss potentials
Intercomparison Projectover Europe.
(AMIP1), Clim.This
Dyn.,investigation is
19, 1–16, 2002.
teristics to other years and other GCM-derived
data and the DKRZ/WDCC (Hamburg, Germany) for computer time series, done primarily
Leckebusch, G. C.for
andGermany,
Ulbrich, U.:and then
On the extendedbetween
relationship to other
cy-
particularly
capacity. J. G.those
Pinto from
and E.the 20C runs.
L. Fröhlich We found
contributed that tosuch
equally this countries of Western
clones and Europe. events
extreme windstorm We used overa Europe
simpleunder
stormclimate
loss
exceptional years feature an extraordinary number of intense model based on the following assumptions (a): Losses in-
storms (noticeable as single peaks in the daily time series), crease with the cube of the maximum daily wind speed (b):
Nat. Hazards Earth Syst. Sci., 7, 165–175, 2007 www.nat-hazards-earth-syst-sci.net/7/165/2007/
J. G. Pinto et al.: Changing European storm loss potentials 175

change, Global Planet. Change, 44, 181–193, 2004. Raible, C. C. and Blender, R.: North Hemisphere Midlatitude cy-
Leckebusch, G. C., Koffi, B., Ulbrich, U., Pinto, J. G., Spangehl, T., clone variability in GCM-simulations in different ocean repre-
and Zacharias, S.: Analysis of frequency and intensity of winter sentations, Clim. Dyn., 22, 239–248, 2004.
storm events in Europe on synoptic and regional scales from a Roeckner, E., Bäuml, G., Bonaventura, L., Brokopf, R., Esch,
multi-model perspective, Climate Research, Special Issue to the M., Giorgetta, M., Hagemann, S., Kirchner, I., Kornblueh,
Mice project, 31, 59–74, 2006. L., Manzini, E., Rhodin, A., Schlese, U., Schulzweida, U.,
Leckebusch, G. C., Ulbrich, U., Fröhlich, E. L., and Pinto, J. G.: and Tompkins, A.: The atmospheric general circulation model
Property loss potentials for European mid-latitude storms in a ECHAM5. Part I: Model description, MPI Report, Nr. 249, 2003.
changing climate, Geophys. Res. Lett., accepted, 2007. Schinke, H.: On the Occurence of Deep Cyclones over Europe and
Marsland, S. J., Haak, H., Jungclaus, J. H., Latif, M., and Roske, the North Atlantic in the Period 1930–1991, Beitr. Phys. Atmo-
F.: The Max-Planck-Institute global ocean/sea ice model with sph., 66, 223–237, 1993.
orthogonal curvilinear coordinates, Ocean modelling, 5, 91–127, SwissRe: Storm over Europe – An underestimated risk, Swiss Re
2003. publishing, Zürich, 2000.
MunichRe: Naturkatastrophen in Deutschland - Schadenerfahrun- SwissRe: The effects of climate change: Storm damage in Europe
gen und Schadenpotentiale, Publication of Munich Reinsurance on the rise, Swiss Re publishing, Focus Report, Zürich, Can be
Company, 1999. downloaded from http://www.swissre.com, 2006.
MunichRe: Winter storms in Europe (II) – Analysis of 1999 losses Ulbrich, U., Fink, A., Klawa, M., and Pinto, J. G.: Three extreme
and loss potentials, Publication of Munich Reinsurance Com- storms over Europe in December 1999, Weather, 56, 70–80,
pany, 2001. 2001.
Nakićenović, N., Alcamo, J., Davis, G., de Vries, B., Fenhann, Ulbrich, U., Brücher, T., Fink, A., Leckebusch, G., Krüger, A., and
J., Gaffin, S., Gregory, K., Grübler, A., Jung, T. Y., Kram, T., Pinto, J. G.: The Central European Floods in August 2002. Part I:
Rovere, E. L. L., Michaelis, L., Mori, S., Morita, T., Pepper, Rainfall periods and flood development, Weather, 58, 371–377,
W., Pitcher, H., Price, L., Raihi, K., Roehl, A., Rogner, H.-H., 2003a.
Sankovski, A., Schlesinger, M., Shukla, P., Smith, S., Swart, R., Ulbrich, U., Brücher, T., Fink, A., Leckebusch, G., Krüger, A., and
Rooijen, S. V., Victor, N., and Dadi, Z.: Emission Scenarios. Pinto, J. G.: The Central European Floods in August 2002. Part
A Special Report of Working Group III of the Intergovernmen- II: Synoptic causes and considerations with respect to climate
tal Panel on Climate Change, Cambridge University press, Cam- change, Weather, 58, 434–442, 2003b.
bridge, UK, 2000. UNEP/GRID: The United Nations Environment Programme/
Paciorek, J. C., Risbey, J. S., Ventura, V., and Rosen, R. D.: Mul- Global Resource Information Database, http://www.na.unep.net/
tiple Indices of Northern Hemisphere Cyclonic Activity, Winters datasets/datalist.php, 2006.
1949–99, J. Climate, 15, 1573–1590, 2002. Uppala, S., Kallberg, P., Hernandez, A., Saarinen, S., Fiorino, M.,
Palutikof, J. P. and Skellern, A. R.: Storm Severity over Britain, A Li, X., Onogi, K., Sokka, N., Andrae, U., and Da Costa Bechtold,
Report to Commercial Union General Insurance, Climatic Re- V.: ERA-40: ECMWF 45-year reanalysis of the global atmo-
search Unit, School of Environmental Sciences, University of sphere and surface conditions 1957–2002, ECMWF Newsletter,
East Anglia, Norwich (UK), 1991. 101, 2–21, 2004.
Pinto, J. G., Spangehl, T., Ulbrich, U., and Speth, P.: Sensitivities von Storch, H., Guddal, J., Iden, K., Jónsson, T., Perlwitz, J., Reis-
of a cyclone detection and tracking algorithm: individual tracks tad, M., de Ronde, J., Schmidt, H., and Zorita, E.: Changing
and climatology, Meteorol. Z., 14, 823–838, 2005. Statistics of storms in the North Atlantic, MPI Report, Nr. 116,
Pinto, J. G., Spangehl, T., Ulbrich, U., and Speth, P.: Assessment p. 18, 1993.
of winter cyclone activity in a transient ECHAM4-OPYC3 GHG WASA-Group: Changing Waves and Storms in the Northeast At-
experiment, Meteorol. Z., 15, 279–291, 2006. lantic?, Bulletin of the American Meteorological Society, 79,
Pinto, J. G., Ulbrich, U., Leckebusch, G. C., Spangehl, T., Reyers, 741–760, 1998.
M., and Zacharias, S.: Changes in barocline wave activity in an Webster, P. J., Holland, G. J., Curry, J. A., and Chang, H.-
ensemble of ECHAM5/MPI-OM1 simulations for three different R.: Changes in Tropical Cyclone Number, Duration, and In-
SRES scenarios, Clim. Dyn., in press, doi:10.1007/s00382-007- tensity in a Warming Environment, Science, 309, 1844–1846,
0230-4, 2007. doi:10.1126/science.1116448, 2005.

www.nat-hazards-earth-syst-sci.net/7/165/2007/ Nat. Hazards Earth Syst. Sci., 7, 165–175, 2007