NHESSNatural Hazards and Earth System ScienceNHESSNat. Hazards Earth Syst. Sci.1684-9981Copernicus GmbHGöttingen, Germany10.5194/nhess-15-1437-2015Identification of storm surge events over the German Bight from atmospheric reanalysis and climate model dataBefortD. J.daniel.befort@met.fu-berlin.dehttps://orcid.org/0000-0002-2851-0470FischerM.LeckebuschG. C.https://orcid.org/0000-0001-9242-7682UlbrichU.https://orcid.org/0000-0001-7558-6622GanskeA.RosenhagenG.HeinrichH.https://orcid.org/0000-0001-5900-6800Institute of Meteorology, Freie Universität Berlin, Berlin, GermanySchool of Geography, Earth and Environmental Sciences, University of Birmingham, Birmingham, UKDeutscher Wetterdienst (DWD), Hamburg, GermanyGerman Maritime and Hydrographic Agency (BSH), Hamburg, GermanyD. J. Befort (daniel.befort@met.fu-berlin.de)30June20151561437144717April201404June201417May201518May2015This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://www.nat-hazards-earth-syst-sci.net/15/1437/2015/nhess-15-1437-2015.htmlThe full text article is available as a PDF file from https://www.nat-hazards-earth-syst-sci.net/15/1437/2015/nhess-15-1437-2015.pdf
A new procedure for the identification of storm surge situations for
the German Bight is developed and applied to reanalysis and global
climate model data. This method is based on the empirical approach
for estimating storm surge heights using information about wind speed
and wind direction. Here, we hypothesize that storm surge events
are caused by high wind speeds from north-westerly
direction in combination with a large-scale wind storm event
affecting the North Sea region. The method is calibrated for ERA-40
data, using the data from the storm surge atlas for Cuxhaven. It is
shown that using information of both wind speed and direction as
well as large-scale wind storm events improves the identification of
storm surge events.
To estimate possible future changes of potential storm surge events,
we apply the new identification approach to an ensemble of three transient climate
change simulations performed with the ECHAM5/MPIOM model
under A1B greenhouse gas scenario forcing. We find an
increase in the total number of potential storm surge events of about
12 % [(2001–2100)–(1901–2000)], mainly based on changes of moderate events. Yearly numbers of
storm surge relevant events show high interannual and decadal
variability and only one of three simulations shows
a statistical significant increase in the yearly number of potential storm
surge events between 1900 and 2100. However, no changes in
the maximum intensity and duration of all potential events is determined.
Extreme value statistic analysis confirms no frequency change of the most severe events.
Introduction
Storm surges at the German coast have a high socio-economic impact,
as they are the most dangerous hazard for the coastal areas, even
affecting the densely populated urban region of Hamburg.
The factors influencing storm surges are summarized in
and the knowledge about past and possible future
changing storm-surge statistics is reviewed in
. Winds blowing from offshore directions cause
a rise in water levels at the coast, which is particularly relevant
during high tides. Other factors influencing the rise of the water
level during a storm are local water depth and external surges
.
The height of a storm surge is defined by the rise of the water level
above the mean high water level (MHW). Following the definition used
by the German Maritime and Hydrographic Agency (BSH) a storm surge
event at the German North Sea coast with water levels exceeding the
MHW by 1.5 to 2.5 m is called a “storm surge”, an excess of
2.5 to 3.5 m is defined as a “heavy storm surge”, and an event
exceeding 3.5 m above MHW is called a “very heavy storm surge”
. The shape of the German Bight coastline
and its estuaries intensifies the water rise as water that is pushed
by north-western winds into the southern North Sea is impounded.
The BSH uses dynamical atmospheric and hydrological models to forecast
water levels for the German coasts and estuaries. Lately,
a dynamical-statistical forecasting system was developed
. However, until recently, the forecasts
were based on an empirical-statistical approach using a multi-linear
regression . The formula used in the latter
procedure is based on 13 linear equations using wind speed and
direction, surface pressure and its change in time, air and/or water
temperature, and observed water level in Wick (Scotland) as input
data. The factors explaining most of the variability according to
these calculations are wind speed and wind direction.
The optimal wind direction of 295∘, found out for
the town of Cuxhaven, in the centre of the German Bight coast, is
commonly used as a proxy for the whole region . The component of the
observed wind projected on this direction is called the “effective
wind component” subsequently.
Future changes of extreme water levels
during storm surges are both determined by changes of mean sea level and wind
storm intensities. found in an analysis of sea level data from 13
gauges in the German Bight that linear extreme sea level trends
exceeded mean sea level trends in the second half of the 20th century,
indicating that changes in local extreme winds have played an
important role in the recent past. Several recent studies on changes
of storm surges under future greenhouse gas (GHG) conditions found
only minor or no significant changes for the German and Dutch North
Sea coastline
. Other
studies found indications of an increase in storm surge extremes at
the North Sea coast associated with increased GHG concentrations
. This is in line with
, who calculated an increase in insurable losses
due to storm surges under future climate scenarios for this
region. With respect to wave heights under scenario conditions,
pointed at increasing extreme wave heights over
large parts in the southern and eastern North Sea. This is not
a contradiction to findings of , who found no
significant change of projected mean wave heights and periods along
the Dutch coasts as there can be different trends in the means and the
extremes of an atmospheric phenomenon see
e.g.,. Findings from another study focusing on
storm surges affecting the English coasts are also inconclusive as trends in surge
heights cannot be separated from natural variability ().
We do not attempt to review the reasons for the different results
obtained in detail. Rather, we develop a new methodology for
estimating changes in storm surge risks for the German Bight region
solely due to changes in frequency and strength of storms under future
climate scenario conditions. It is designed to be used with coarse
grid data, meeting the current standard of GCM runs in CMIP5 (). In
a first step, the skill of our method is tested by identifying
historic storm surge events in reanalysis data. In a second step, we
apply the method to the output data of the CMIP3 ECHAM5/MPIOM
simulations under recent and future climate conditions as the
requirement of available zonal and meridional wind data at 6 hourly
time steps is fulfilled by this model. To estimate the total future
storm surge risk in the German Bight, these results can be used to
find single events for which corresponding water levels can be
calculated and analyzed with regional hydrodynamic models. Thus,
we split the future storm surge risk into an atmospheric and an
oceanic part and neglect nonlinear interactions of the atmosphere and
the ocean.
Data
For this study we use the storm surge atlas of the station in Cuxhaven
covering the period 1901–2008. It comprises
a total of 166 storm surges with information about the measured total
water level, the calculated astronomical tide and the wind surge. In this study we use the
wind surge height, which was computed by subtracting total water
height and calculated astronomical tide, thus any other factors like external
surges have been neglected. The storm surge atlas only includes those
storm surge events for which water levels exceeded the MHW by at least
1.5 m. In context with the tidal amplitude, which is about 2.9 m
for Cuxhaven, this means that even wind surges of the order of a heavy storm
surge, following the definition of the BSH, are not fully included in the data,
when they occur during low tides. The temporal resolution of time series for total water level,
astronomical tide and wind surge is 5 min.
For the calibration of the storm identification method we use the
reanalysis data of ERA-40 . It covers the period
1957–2002 with a horizontal resolution of 1.125∘
(≈ 125 km; T159).
In order to estimate changes of storm surge risk, the IPCC-AR4 ensemble
simulations of the global coupled atmosphere-ocean model ECHAM5/MPIOM with
a horizontal resolution of 1.875∘ (≈ 210 km; T63) are used.
In this study, we investigate the three transient ensemble simulations driven
with observed GHG forcing for the period 1900 until 2000
and A1B scenario forcing for
the period 2001 until 2100 .
Zonal and meridional surface wind in 10 m height with a time resolution of
6 h for ERA-40 reanalysis and ECHAM5 data are analyzed.
Method
An identification of all wind events potentially leading to extreme
storm surges along the German Bight coast is hindered by the coarse
resolution of the global climate model simulations in which the German
Bight region is represented by about two grid points only. Wind speed
data from these grid points may not be representative for a situation
with a major surge-producing storm, potentially including some small
and short-lasting events.
In this study, we investigate the improvements of the identification of past
storm surge events by additionally taking into account the large-scale wind
field over the North Sea region. The advantage of this procedure is
investigated in Sect. .
Effective wind component
As the basis for the identification of storm surge events we apply a method
which is based on the statistical approach developed by the German Maritime
and Hydrographic Agency . It considers the
effective wind component which is the projection of 10 m winds on the
direction of 295∘ (WNW direction). In this study, the effective wind
component is defined as the mean value calculated using all grid points
within the German Bight region, which consists of seven grid points in ERA-40
reanalysis data and two grid points in ECHAM5 model data
(Fig. a and b). This calculation is done by first calculating
the effective wind component at each grid point before averaging over the
area.
(a) German Bight (red dots) and North Sea region (black
dots) for ERA-40 reanalysis. (b) German Bight (red dots) and North
Sea region (black dots) for ECHAM5 (T63) data. The blue box defines the
boundaries of the North Sea region.
Wind storm identification
Our identification of large-scale wind storms is based on the methodology
described by . This algorithm uses the surface wind
speed in a data set, looking for spatially coherent regions with grid points
exceeding the local 98th percentile of absolute wind speed for each model
time step. Such a storm region is a candidate for a wind storm event if it
has a minimum area of about 150 000 km2 (≈ 20 (7) grid
points in ERA-40 (ECHAM5) at 60∘ N). As it may occur that a wind
field cluster with grid boxes exceeding the 98th percentile is decomposed
into sub-clusters where none of these fulfill the minimum size required, we
use an envelope constructed of the 95th percentile. Thus, even if none of the
individual sub-clusters matches the size criterion solely, it will be counted
as a candidate for a wind field track if those sub-clusters are connected
through the 95th percentile and the total size of all these sub-clusters
exceed the minimum area. These identified clusters are tracked in time using
a nearest-neighbour algorithm, obeying a maximum permitted movement of the
cluster centre of 600 km per 6 h time step, plus an additional allowance for
movements of the centre within the cluster (half of the maximum cluster
extension). Finally, storm events must have a minimum duration of 18 h.
During the period covered by ERA-40 reanalysis data (1957–2002) 83
storm surges are observed, 82 of which occurred between September and May. Thus, our analysis focuses on the
months from September until May as this is the primary season for storm
surges in the German Bight area.
Using ERA-40 reanalysis data we calculate the local 98th percentile of 10 m
wind speed over the whole period from 1957 until 2002. To estimate changes in
storm surge potential, the 98th percentile for ECHAM5 is calculated using all
three ensemble members from the 20C simulations only. The same percentile
value is used to detect storm events during 1900 until 2000 regarding 20C
period and 2001 until 2100 for the A1B scenario period, respectively.
Method to detect relevant storm surge events
As no information about the astronomical tide
is included in the IPCC-AR4 ECHAM5/MPI-OM simulations, our analysis is based
on wind speed and wind direction only. Thus, for observed storm surge events
we use the wind surge data from the storm surge catalogue at the station in
Cuxhaven only.
In the first step, potential storm surges are identified based on effective wind component
values over the German Bight only. In a second step we use additional information
about large-scale wind fields. Thus, an event with storm surge potential is
characterized by its mean effective wind component over the German
Bight region (see Sect. ) and a large-scale
wind storm event, detected by the algorithm explained (see
Sect. ), in the vicinity of the German
North Sea coast.
Events are only considered if the large-scale wind storm is located over
parts of the North Sea region. This region is illustrated in Fig.
a and Fig. b for ERA-40 reanalysis and ECHAM5
model grids, respectively. In total, the region consists of 99 grid points in
ERA-40 and 35 grid points in ECHAM5.
Results
The analysis is divided into four parts. As a first step, we detect storm
surge relevant situations using effective wind components calculated from
ERA-40 reanalysis data solely over the German Bight. Using
the storm surge catalogue from the station in Cuxhaven , we
assign storm surge events to their effective wind components. Secondly, we
use the new method presented in Sect. , combining
effective wind component information and large-scale wind storm events to
identify storm surge relevant events. Comparing both methods gives an
estimate of the additional value of our proposed methodology. In the third
step, this approach is transferred to ECHAM5 model data
.
Detection of potential storm surge events under recent and future climate
conditions gives an estimate of the possible changes in storm surge activity
over the German Bight region. In a last step, extreme value statistics are
applied to the results of the latter section to calculate return levels for
these events, followed by a short conclusion and discussion.
Storm surge events in ERA-40 reanalysis data
Our first attempt in identifying storm surge relevant events in atmospheric
model data is based on an empirically derived relationship between observed
wind surge and wind speed and direction developed at BSH. First, we calculate
effective wind components over the German Bight every 6 h from ERA-40
reanalysis data. Next, we determine the dates of observed storm surge events
from the storm surge atlas of Cuxhaven. As we focus on wind surge, the date
assigned to an observed storm surge event is the one at wind surge maximum.
In general, no wind data from reanalysis (6 hourly) is available at the
exact date of wind surge maximum (every 5 min). Thus, we take
effective wind component data for the time step before and after wind surge
maximum into account. If data with a much higher time resolution were to be
used, one would have to think about a 3 h time lag between wind maximum and
surge maximum (). For this study, this will not affect our
results. To estimate the usability of this method we count the total number
of the 6 hourly time steps in ERA-40 data, for which this threshold of the
effective wind component is exceeded. Out of this, we derive the ratio
between all observed storm surge events at Cuxhaven and the total number of
time steps exceeding this threshold, revealing that only 3.7 % of all time
steps in ERA-40 reanalysis data, which exceed the threshold of the effective
wind component (observed for the storm surges at Cuxhaven), lead to a storm
surge in reality.
As our proposed method is based on the combination of high effective wind
components and a large-scale wind storm event, we first identify large-scale
storm events with a minimum duration, size and strength in ERA-40 data (see
Sect. ). In total, 3541 events are found
between September and May for the period 1957 until 2002. This large number
is due to the extent of the domain in which wind storm events are identified,
covering the region from 35∘ W to 30∘ E and 35 to
75∘ N. 1353 out of 3541 events have at least one time step with an
exceedance of the local 98th percentile over at least one grid point within
the North Sea region. As we hypothesize that a large-scale wind storm over
the North Sea region is obligatory for the occurrence of a storm surge, only
these 1353 events are considered (see Sect. ).
Effective wind components over the German Bight for all 1353 large-scale wind
storm events are derived.
As expected, not all 1353 wind storm events affecting a North Sea grid
point led to a high observed wind surge. We identify those events
which are associated with a high wind surge at Cuxhaven by looking for
a wind field event which is part of an existing wind storm event
affecting the North Sea region at the date of the observed wind surge
maximum.
There are two possibilities to assign an effective wind component from ERA-40
data to the observed wind surge maxima. The first idea is to take the higher
of the two subsequent (6 hourly) values before or after the wind surge
maximum. With this approach a frequency distribution of maximum effective
wind components for all 1353 large-scale wind storm events affecting the
North Sea region and for those which can be assigned to an observed storm
surge, can be calculated (Fig. a). We find that a
large-scale wind storm event can be found for 80 out of 82 storm surge events
(red bars in Fig. a). This data set suggests that
storm surges are characterized by a minimum effective wind component of about
9.45 ms-1 in reanalysis data over the German Bight region. Thus,
all large-scale wind storm events affecting the North Sea region combined
with an effective wind component exceeding 9.45 ms-1 are
regarded as potential storm surge events (see
Sect. ). Note that the maximum effective wind
component is not always found at one of the 6 hourly time steps before or
after the maximum surge. One of the reasons for this fact could be related to
the actual astronomic tide as the effect of high wind speeds is reduced
during high tide compared to low tide . As well, this
minimum effective wind component reflects the mean over seven grid points in
ERA-40 reanalysis data. Therefore, it is smaller than observed wind speeds at
a particular station during storm surges.
(a) Histogram of effective wind component over the German
Bight region for all large-scale wind storm events (grey) and wind storm
events which could be assigned to an observed storm surge (red). Effective
wind component for assigned events are calculated using the maximum of the
ERA-40 time step directly before and after the wind surge maximum.
(b) Same as (a) but here effective wind component for
assigned events (red) is calculated using the maximum effective wind
component during the whole large-scale wind storm track.
In a second approach we assign the highest effective wind component during
the whole wind storm event and do not consider only the maximum of the
two subsequent values before or after the surge maximum. In this
case, the shape of the frequency distribution is slightly changed
compared to the first distribution
(Fig. b).
Similar to the first approach, based on the effective wind component only, we
try to assess the usability of the new method. Therefore, we calculate the
ratio of all potential storm surge events in ERA-40 reanalysis data and all
storm surge events, which could be assigned to a large-scale wind storm.
Large-scale wind storm events with effective wind components below about
10 ms-1 never lead to an observed storm surge at Cuxhaven
(Fig. b). This is reasonable as the wind is
blowing offshore in case of negative effective wind component values. As
illustrated in Fig. b, the ratio of those events
which are assigned to an observed storm surge and all large-scale wind storm
events is increasing with stronger effective wind component values. A
large-scale wind storm event with effective wind components between 10 and
11 ms-1 only leads to a storm surge in one percent of all
cases. In contrast to this, events with effective wind components between
16.5 and 17.5 ms-1 lead in about 50 %, and events with
effective wind components exceeding about 19 ms-1 always lead to
an observed storm surge at Cuxhaven. Thus, wind surge tends to exceed tidal
amplitudes with increasing effective wind components.
The large number of storms with comparably low effective wind components,
which could not be assigned to an observed storm surge, could be caused
by the effect that we neglected the interaction between the tide and the
winds. Furthermore, the storm surge catalogue from the Cuxhaven station
only lists events which led to a minimum total water height of 1.5 m
above MHW. Thus, even if a high wind surge is present but it does not
occur during a favourable tide phase, this threshold for total water
height is not reached and the event is not included in the storm surge
catalogue. Events with effective wind components (larger than
19 ms-1) cause a very high wind surge, thus leading to
a storm surge (exceeding the criterion of 1.5 m
above MHW) independent of the tidal phase.
Two observed storm surge events could not be assigned to a large-scale wind
storm event over the North Sea. One of these events is characterized by a large-scale wind storm event which is too short, lasting only two time steps. In the
second case, two separated wind fields exist whereof one is located over the
North Sea and the other one is located over Scandinavia. Due to the simple
nearest-neighbour algorithm the wind field over Scandinavia is connected to
the existing wind storm event as it is closer to the previous wind field, and
the wind field over the North Sea is neglected.
Overall, the approach based on the combination of high effective wind
components and large-scale wind storm events over the North Sea region
outperforms the approach based on effective wind components solely, which
makes this method well suited for coarse resolved GCM data.
Thus, the new method based on:
a large-scale wind storm event affecting the North Sea region
an effective wind component exceeding 9.45 ms-1 over the German Bight region
is further used to detect potential storm surge relevant events in ECHAM5 20C and A1B simulations.
Potential storm surge events in ECHAM5 20C and A1B
Due to the coarser spatial
resolution of the GCM data with 1.875∘ compared to
1.125∘ in ERA-40 reanalysis data, we apply
some minor changes to the method presented in
Sect. for the model analysis. Thus, the regions for
the German Bight and North Sea used for the identification in ECHAM5
are not the same regions used in ERA-40. Section
shows that storm surge events do occur when the spatial mean of the
ERA-40 effective wind component over the German Bight exceeds
9.45 ms-1 in combination with a large-scale storm field
over the North Sea region. As absolute wind speeds can essentially
differ between spatially lower resolved model and reanalysis data, we
use percentile values rather than absolute wind speed values. An
effective wind component of 9.45 ms-1 in ERA-40 reanalysis
data corresponds to a percentile value of 91.97 %. The 91.97th percentile of effective wind
component over the German Bight in ECHAM5 data corresponds to an absolute
wind speed of 9.84 ms-1. To calculate this value we use
the 20C realizations solely.
Number of potential storm surge events
The number of events within the whole identification area covering parts of
the North Atlantic is statistically significantly lower during the A1B period
with respect to the 20C period (see Table ). However,
the increased number of events with at least one time step within the North
Sea region during the A1B period, found for two out of three ensemble
members, could indicate a shift in the cyclone tracks. Considering the sum of
all three ensemble members of the 20C and A1B period, we find an increase in
potential storm surge events affecting the North Sea region (with effective
wind component above 9.84 ms-1) by 12.4 % in A1B compared to
20C.
Mean number of events per year within the whole identification area
(top line), events with at least one time step within the North Sea (NS)
region (middle line) and the amount of potential storm surge events
(large-scale wind storm affecting North Sea region + effective wind component
above 9.45 ms-1 (ERA-40 ), 9.84 ms-1 (ECHAM5)
respectively, bottom line) for ERA40 (1957–2002) and for each scenario run
of ECHAM5 20C (1900–2000) and ECHAM5 A1B (2001–2100).
DataERA4020C 120C 220C 3A1B 1A1B 2A1B 3Total storm events77.088.288.188.283.183.383.5Events with wind storm over NS29.429.028.729.829.830.128.9Potential storm surge events11.09.610.110.711.412.010.8with wind storm over NS andveff>9.45ms-1 (ERA-40) or,veff>9.84ms-1 (ECHAM5)
We calculate the number of all potential storm surge events (effective wind
component above 9.84 ms-1) in ECHAM5 20C and A1B data for all
ensemble members depending on their effective wind components
(Fig. ). For the majority of classes the number of storm
events in A1B has increased with respect to those of the 20C simulations
(Fig. ).
Number of potential storm surge events with respect to their maximum
effective wind component over the German Bight region for the three ensemble
members of the 20C (1901–2000, green to blue) and A1B (2001–2100, yellow to
red) period. Grey bars denote the boundaries of the four categories used in
Sect. . The black rectangle indicates those potential
storm surge classes, which are used for the extreme value statistic described
in Sect. .
A high increase of at least 10 % is found for events with effective wind
components of 11.8–16.8 ms-1. Extreme events in particular, with
effective wind components of 20.8–21.8 ms-1 become more
frequent. However, the storms with effective wind components of 17.8–20.8
and 21.8–22.8 ms-1 become infrequent for the A1B runs (see
Fig. ). For storms with an effective wind component above
23.84 ms-1, the percentage changes are insignificant due to the
low numbers of events. A Kolmogorow-Smirnow test with an error probability of
α=0.05 reveals that the distributions (for 20C and A1B) cannot
be distinguished.
The annual numbers of all potential storm surge events for all three ensemble
members for the period 1900 until 2100 are shown in
Fig. a. The ensemble mean of the ensemble members
(black lines) of the 20C scenario (green to blue) and the A1B scenario
(yellow to red) reveals a statistically significant linear trend with an
error probability of α=0.01 and an increase in relevant storm surge
events of almost 1.5 events per 200 years within the period 1900–2100.
By considering the three runs separately, the relevance of the simulated
decadal climate fluctuations is shown: run 1 offers a significant increase of
almost three events in 200 years with an error probability of α=0.001. In run 2 a rise is seen as well which, however, is not statistically
significant. Run 3 shows a slight decrease which is also not statistically
significant. However, taken into account the large interannual fluctuations
in the range between two and 28 events per year, the increase of even three events
in 200 years is very moderate.
Annual number of potential storm surge events from 1900 to 2100 for
the scenario runs of 20C (green to blue) and A1B (yellow to red). The black
lines denote the ensemble mean for (a) all potential storm surge
events, (b) weak potential storm surge events (category 1),
(c) moderate potential storm surge events (category 2), (d)
strong potential storm surge events (category 3) and (e) very strong
potential storm surge events (category 4).
As discussed in Sect. , the probability of a potential
wind storm event leading to a storm surge increases for stronger effective
wind components. Thus, we divide all potential storm surge events into four
categories dependent on the effective wind component value: (1) weak (<
12 ms-1), (2) moderate (between 12 ms-1 and
18 ms-1), (3) strong (between 18 ms-1 and
21 ms-1), and (4) very strong (above 22 ms-1). We
find an increase for three categories (1,2,4) as well as for the total number
of potential storm surge events under future climate conditions
(Table ). However, only the total number as well as the
number of moderate events (category 2) differ significantly under the
assumption of a Poisson distributed variable. Using results shown in
Sect. , weak potential storm surge events only lead to a
storm surge in Cuxhaven in 2 % of the cases, and moderate events in
13.7 % of the cases. In contrast, it is found that strong events are in 87.5 % and
very strong events in 100 % of all cases related to a storm surge.
Percental changes in the number of potential storm surge events of all
three ensemble members during 2001–2100 (A1B) compared to the 1901–2000
(20C) for the four categories and all events. Statistically significant
changes are highlighted.
We find no significant trend of the annual number of weak (category 1)
potential storm surge events as well as for strong (category 3) potential
storm surge events (Fig. b and d). For these two
categories none of the three ensemble members shows a significant increase or
decrease. We find for the ensemble mean a statistically significant increase
for moderate events (category 2) of about 1.17 events (with an error
probability of α=0.05) and an increase for very strong events
(category 4) of about 0.29 with an error probability of α=0.01
(Fig. c and e). For these two categories two out of
three ensemble members show a statistically significant increase of the
annual number.
It should be noted that changes in very strong potential storm surge events
are difficult to assess, as the number of events in this category is very low
(20C: 88; A1B: 128) and results might depend on the exact classification of
potential storm surge events. Overall, the increase in moderate events is
more robust due to the higher number of events (20C: 1909 and A1B: 2228).
This indicates that the increase of all potential storm surge events
(Fig. a) is dominated by the changes of moderate
events, the latter leading to a storm surge in ERA-40 reanalysis
data in about 13.7 % of cases.
In Fig. the 30 year moving mean of the total number of
potential storm surge events per year for the considered runs and for the
ensemble mean is shown. The large decadal fluctuations can be clearly seen in
the time series of the means. Therefore, it has to be noted that the increase
in numbers of relevant storms is not monotonic. Furthermore, the 30 year
periods with the largest numbers of storm surge relevant events are not found
at the end of the 21st century for all three ensemble members.
30 year running mean of the annual number of potential storm surge
events from 1915 to 2085 for the three runs (run 1: blue, run 2: green, run
3: red) and for the mean of all ensemble members (black).
Duration of potential storm surge events
In addition to the number of potential storm surge events, we
investigate if the duration of such events changes under future
conditions. Therefore, the number of time steps are counted for which
the storm is located in the region of the North Sea and the effective
wind component of 9.84 ms-1 is exceeded. An increased storm
surge potential can be assumed due to an increased duration of the
events, even if the maximum effective wind component is
unchanged. Figure depicts the percentage of
exceedings for the runs of 20C and A1B with respect to the total
number of tracks in the respective century. However,
analysing the uncertainty of the counts (confidence interval of
the Poisson distribution) for 20C and A1B shows no evidence for a change of the duration (with an error probability of
α=0.05).
Potential storm surge duration distribution (in %) with respect to
the number of time steps which exceed the critically effective wind component
for the three 20C runs (1901–2000, green to blue) and the three A1B runs
(2001–2100, yellow to red).
Return levels of extreme effective wind
component
To investigate the occurrence rate of extreme events, extreme value
statistics (EVS) provide a suitable approach. At this point, only the
basic idea of the theory is covered and reference is made to more
in-depth literature, e.g. . In EVS, extreme values
above a certain threshold can be described by using the generalized
Pareto distribution (GPD) given by:
H(y)=1-1+ξyσ̃-1/ξwith{y:y>0},(1+ξy/σ̃)>0,
with the form parameter, ξ, and the scale parameter, σ̃,
which is dependent on the statistical threshold value used
for restricting the assessment to extremes. The choice of the
threshold is delicate because if the threshold chosen is
too high, not enough values for the statistical analysis exist. For
a peak-over-threshold approach a selection of an appropriate threshold, u,
is vital. Guidance for threshold selection is provided by the mean
residual life plot (mrlp); if the threshold u0 is large enough to
ensure the GPD approximation to be valid then, for all thresholds
u>u0, a linear relationship between the mean of excesses
E(x-u|x>u) and the
threshold u holds ().
The mrlp derived by using all three 20C ensemble members is shown in
Fig. . In addition to the mean excess (continuous line), their
95 % confidence interval is drawn as well (dotted lines). The red
vertical line marks the threshold of 9.84 ms-1 derived from
historical storm surges (cf. Sect. ) with more than
a quarter of the time steps exceeding that threshold. However, based on the
mrlp, this threshold is too low to allow for a reliable approximation with
the GPD; the plot rather suggests a threshold of 18.0 ms-1,
because above this threshold the mrlp is approximately linear in u. Thus, the
following results solely include strong and very strong storm surge events
(category 3 and category 4). It has already been established that a statistically significant and robust change of the total number
of events for these two classes does not exist. Furthermore, we cannot identify a trend of
the yearly number of strong and very strong potential storm surge events for
both categories together. Thus, we use the stationary approach of the GPD.
With the threshold of 18.0 ms-1 we calculate return levels
including all three ensemble members available during A1B and the 20C period
(Fig. ). In this case the return levels specify the effective
wind component in ms-1 which is expected to be exceeded on
average once in a certain return period. As expected, the increased number of
potential storm surge events is also reflected in slightly raised return
levels on the interannual to interdecadal scale (return levels 0.1–7) for
the runs of A1B. The effective wind components, which are expected to be
exceeded several times a year are increased by nearly 0.5 ms-1
in the A1B runs with respect to 20C. On the decadal to multi-decadal scale
the return levels are higher for 20C because of a few more events of the
strongest intensity classes (Fig. ). The increased
uncertainty of the multi-decadal return levels are expressed by wider
confidence intervals. However, the slight differences in the return levels
are not significant. Thus, extreme value statistics confirm that strong and
very strong storm surge events do not reveal interpretable changes in the
future realisation of ECHAM5.
Mean residual life plot for the effective wind component of all
three 20C runs. The mean excess (continuous line) and the 95 % confidence
intervals (dashed lines) are shown. Thresholds derived from historical storm
surge events (red) and derived from the extreme value statistics (blue) are
indicated as vertical lines. In addition, the percentage of these exceedings
are specified.
Conclusions
In this study, changes of the storm surge risk at the German Bight
coast caused by possible changes of meteorological parameters under
future climate conditions are investigated. In order to detect storm
surge events in meteorological reanalysis and climate model data, we
use 10 m wind vectors only. We hypothesize that storm surge events
are attributed to strong near surface wind speeds over the German
Bight projected on a wind direction of 295∘, under the condition of
a large-scale wind storm event affecting the North Sea region.
Using ERA-40 reanalysis data and the storm surge atlas for the station
Cuxhaven, 80 out of 82 observed storm surge events between September and May
could be assigned to a large-scale wind storm event. The reason that one
event could not be assigned to any storm surge event is due to difficulties
arising from the simple nearest-neighbour approach used for the identification
of large-scale wind fields. Another event does not fulfill the minimum
duration criterion of 18 h.
Our analysis shows that storm surge events are characterized by an effective
wind component exceeding 9.45 ms-1 in ERA-40, but not every wind
event with an effective wind component over 9.45 ms-1 leads to
an observed storm surge. This is mainly explained by the fact that tides are
not taken into account in our approach. As the interaction of tides and high
winds are important for the development of a storm surge, it can also be
observed that not every storm causes a storm surge. Furthermore, other
factors, e.g. external surges are included in the wind surge data as this
variable is derived by subtracting the water level and the astronomical
tide. showed that in about 20% of all observed storm
surge events the water level is influenced by external surges. However, in
most of these cases the wind surge was not in phase with the external surge
().
Return levels in ms-1 (y axis) for different
return periods in years (x axis) for 20C (1901–2000, turquoise) and
A1B (2001–2100, orange) + the 95 % confidence intervals (dashed
lines). Only effective wind component values exceeding
18 ms-1 are considered.
Nevertheless, if the effective wind component is higher than about
19 ms-1, then water levels rise above the criterion height for
storm surges, independent of the phase of the tide. In these cases, the
respective tidal phase only determines the severeness of the storm surge. For
ERA-40 winds, all wind storm events with effective wind components exceeding
19 ms-1 can be attributed to an observed storm surge. Thus,
higher effective wind components lead to a reduced importance of the tidal
phase.
The analysis presented in this paper shows that the proposed method using
large-scale wind field information as well as effective wind components over
the German Bight enhances the detection of potential storm surge events
compared to using effective wind component values solely.
We apply our methodology to three runs of the ECHAM5/MPI-OM simulations
forced by observed GHG concentrations for the period 1900
until 2000 and with GHG concentrations following the SRES A1B scenario from
2001 until 2100. As absolute wind speed distributions between model and
reanalysis data differ, we do not use the absolute threshold of
9.45 ms-1 (derived from ERA-40 data) for the identification of
potential storm surge events in ECHAM5 model data. Instead, we use a minimum
effective wind component of 9.84 ms-1 for ECHAM5 data,
corresponding to the same percentile value of 91.97 % as
9.45 ms-1 in ERA-40.
The total number of potential storm surge events identified in all three
simulations under the A1B scenario for the whole period between 2001 and 2100
is increased by 12.4 % with respect to the 20C period (1901–2000). All
three time series of yearly numbers of potential storm surge events display
a high interannual and decadal variability. Only one simulation shows
a significant increase in the number of potential storm surge events per
year, while the two others show either an increasing or a decreasing trend,
which are both insignificant. We find no increase of the highest effective
wind components as well as in the duration of potential storm surge events
under future climate conditions.
We divide potential storm surge events identified in ECHAM5 model data into
four categories according to their maximum effective wind components. A
statistically significant increase of the total number of these events is
found, which is mainly due to a statistically significant increase of
moderate potential storm surge events. According to ERA-40 reanalysis data,
the probability of these moderate events to cause a storm surge in Cuxhaven
is 13.7 %.
Extreme value statistics reveal an increase in effective wind components
with an interannual to interdecadal scale, whereas events with larger return
periods exhibit lower return levels, indicating that the rarest events
will potentially show reduced intensity. However, uncertainties are large and
do not allow for a meaningful interpretation of the changes in the most severe
events.
The increase in the total number of potential storm surge events found in
this study cannot directly be compared to other studies including effects of
the astronomical tide (e.g., ; ).
Nevertheless, found no significant increase in storm surge
events in Cuxhaven, which is in line with our results showing no changes in
the most severe events.
The fact that the proposed method can be easily applied to simulations
carried out with different climate models independent of their spatial
resolution at low computational costs is the main advantage. Thus, for
example, it allows for a pre-selection of potential storm surge events in
coarse resolved data, which can be further dynamically downscaled using
regional climate models to investigate local characteristics of these events
in more detail.
Acknowledgements
The authors thank the LSBG (Landesbetrieb für Straßen,
Brücken und Gewässer) in Hamburg for providing data from the
storm surge atlas for the station Cuxhaven as well as ECMWF for
providing ERA-40 reanalysis data. This work was funded by the KLIWAS
research programme of the Federal Ministry of Transport and Digital
Infrastructure (BMVI). The authors thank all three anonymous referees for their constructive
comments.Edited by: I. Didenkulova
Reviewed by: three anonymous referees
References
Coles, S.: An Introduction to Statistical Modeling of Extreme Values, Springer-Verlag, London, 2001.
Dangendorf, S., Mudersbach, C., Jensen, J., Ganske, A., and Heinrich, H.: Seasonal to decadal forcing of
high water level percentiles in the German Bight throughout the last century, Ocean Dynam., 63, 533–548, 2013.
Debernard, J. B. and Roed, L. P.: Future wind, wave and storm surge climate in the Northern Seas: a revisit, Tellus A, 60, 427–438, 2008.
de Winter, R. C., Sterl, A., de Vries, J. W., Weber, S. L., and Ruessink, G.: The effect of climate change
on extreme waves in front of the Dutch coast, Ocean Dynam., 62, 1139–1152, 2012.Gaslikova, L., Schwerzmann, A., Raible, C. C., and Stocker, T. F.: Future storm surge impacts on insurable
losses for the North Sea region, Nat. Hazards Earth Syst. Sci., 11, 1205–1216, 10.5194/nhess-11-1205-2011, 2011.
Gaslikova, L., Grabemann, I., and Groll, N.: Changes in North Sea storm surge conditions for four transient
future climate realizations, Nat. Hazards, 66, 1501–1518, 2013.
Gönnert, G.: Sturmfluten und Windstau in der Deutschen Bucht – Charakter,
Veränderungen und Maximalwerte im 20. Jahrhundert, Die Küste – Archiv
für Forschung und Technik an der Nord- und Ostsee, Heft 67, 185–365, 2003.
Gönnert, G. and Buß, T.: Sturmfluten zur Bemessung von Hochwasserschutzanlagen, Landesbetrieb Straßen,
Brücken und Gewässer, Hamburg, 2009.
Gönnert, G. and Sossidi, K.: A new approach to calculate extreme storm
surges: analysing the interaction of storm surge components, in: Coastal
Processes II, Vol. 149, 139–150, WIT PRESS, Southhampton, 139–150, 2011a.
Gönnert, G. and Sossidi, K.: A new approach to calculate extreme storm surges, Irrig. Drain., 60, 91–98, 2011b.
Gönnert, G.: Ermittlung des Sturmflutbemessungswasserstandes für den
öffentlichen Hochwasserschutz in Hamburg, Landesbetrieb Straßen,
Brücken und Gewässer, Hamburg, 2012.
Grabemann, I. and Weisse, R.: Climate change impact on extreme wave
conditions in the North Sea: an ensemble study, Ocean Dynam., 58, 199–212,
2008.
Jensen, J., Mudersbach, C., Müller-Navarra, S. H., Bork, I., Koziar, C.,
and Renner, V.: Modellgestützte Untersuchungen zu Sturmfluten
mit sehr geringen Eintrittswahrscheinlichkeiten an der deutschen Nordseeküste,
Die Küste, 71, 123–167, 2006.
Langenberg, H., Pfizenmayer, A., von Storch, H., and Sundermann, J.: Storm-related sea level variations along
the North Sea coast: natural variability and anthropogenic change, Cont. Shelf Res., 19, 821–842, 1999.
Leckebusch, G. C., Renggli, D., and Ulbrich, U.: Development and application of an objective storm severity measure
for the Northeast Atlantic region, Meteorol. Z., 17, 575–587, 2008.
Lowe, J. A., Howard, T., Pardaens, A., Tinker, J., Holt, J., Wakelin, S.,
Milne, G., Leake, J., Wolf, J., Horsburgh, K., Reeder, T.,
Jenkins, G.,Ridley, J., Dye, S., and Bradley, S.: UK climate projections
science report: Marine and coastal projections, Technical report, Met Office
Hadley Centre, Exeter, UK, 2009.
Müller-Navarra, S. H., Lange, W., Dick, S., and Soetje, K. C.: Über die Verfahren der Wasserstands- und
Sturmflutvorhersage: hydrodynamisch-numerische Modelle der Nord- und Ostsee und ein empirisch-statistisches
Verfahren für die Deutsche Bucht, Promet, 29, 117–124, 2003.
Müller-Navarra, S. H., Seifert, W., Lehmann, H.-A., and Maudrich, S.: Sturmflutvorhersage für Hamburg 1962
und heute, Bundesamt für Seeschifffahrt und Hydrographie, Hamburg und Rostock, 2012.
Pinto, J. G., Ulbrich, U., Leckebusch, G. C., Spangehl, T., Reyers, M., and Zacharias, S.: Changes in storm track
and cyclone activity in three SRES ensemble experiments with the ECHAM5/MPI-OM1 GCM, Clim. Dynam., 29, 195–210, 2007.
Röckner, E.: IPCC-AR4 MPI-ECHAM5_T63L31 MPI-OM_GR1.5L40
SRESA1B run no.2: atmosphere 6 HOUR values MPImet/MaD Germany, World Data Center for Climate
(WDCC), 2004a.
Röckner, E.: IPCC-AR4 MPI-ECHAM5_T63L31 MPI-OM_GR1.5L40
20C3M run no.2: atmosphere 6 HOUR values MPImet/MaD Germany, World Data Center for Climate
(WDCC), 2004b.
Röckner, E., Lautenschlager, M., and Esch, M.: IPCC-AR4 MPI-ECHAM5_T63L31 MPI-OM_GR1.5L40
20C3M run no.3: atmosphere 6 HOUR values MPImet/MaD Germany, World Data Center for Climate
(WDCC), 2006a.
Röckner, E., Lautenschlager, M., and Schneider, H.: IPCC-AR4 MPI-ECHAM5_T63L31 MPI-OM_GR1.5L40
SRESA1B run no.1: atmosphere 6 HOUR values MPImet/MaD Germany, World Data Center for Climate
(WDCC), 2006b.
Röckner, E., Lautenschlager, M., and Schneider, H.: IPCC-AR4 MPI-ECHAM5_T63L31 MPI-OM_GR1.5L40
SRESA1B run no.3: atmosphere 6 HOUR values MPImet/MaD Germany, World Data Center for Climate
(WDCC), 2006c.
Röckner, E., Lautenschlager, M., and Schneider, H.: IPCC-AR4 MPI-ECHAM5_T63L31 MPI-OM_GR1.5L40
20C3M run no.1: atmosphere 6 HOUR values MPImet/MaD Germany, World Data Center for Climate
(WDCC), 2006d.Sterl, A., van den Brink, H., de Vries, H., Haarsma, R., and van Meijgaard, E.: An ensemble study of extreme storm
surge related water levels in the North Sea in a changing climate, Ocean Sci., 5, 369–378, 10.5194/os-5-369-2009, 2009.Taylor, K. E., Stouffer, R. J., and Meehl, G. A.: An Overview of CMIP5 and
the Experiment Design, B. Am. Meteorol. Soc., 93, 485–498,
10.1175/BAMS-D-11-00094.1, 2012.Tomczak, G.: Über die Genauigkeit der Sturmflutvorhersage für die deutsche Nordseeküste, Deutsche
Hydrografische Z., 13, 1–13, 10.1007/BF02224731, 1960.Ulbrich, U., Leckebusch, G. C., and Pinto, J. G.: Extra-tropical cyclones in the present and future climate:
a review, Theor. Appl. Climatol., 96, 117–131, 10.1007/s00704-008-0083-8, 2009.
Ulbrich, U., Leckebusch, G. C., Grieger, J., Schuster, M., Akperov, M., Bardin, M. Y., Feng, Y., Gulev, S.,
Inatsu, M., Keay, K., Kew, S. F., Liberato, M. L. R., Lionello, P., Mokhov, I. I., Neu, U., Pinto, J. G.,
Raible, C. C., Reale, M., Rudeva, I., Simmonds, I., Tilinina, N. D., Trigo, I. F., Ulbrich, S., Wang, X. L.,
and Wernli, H.: Are greenhouse gas signals of Northern Hemisphere winter extra-tropical cyclone activity
dependent on the identification and tracking algorithm?, Meteorol. Z., 22, 61–68, 2013.Uppala, S. M., Kållberg, P. W., Simmons, A. J., Andrae, U.,
Bechtold, V. D. C., Fiorino, M., Gibson, J. K., Haseler, J., Hernandez, A.,
Kelly, G. A., Li, X., Onogi, K., Saarinen, S., Sokka, N., Allan, R. P.,
Andersson, E., Arpe, K., Balmaseda, M. A., Beljaars, A. C. M.,
Berg, L. V. D., Bidlot, J., Bormann, N., Caires, S., Chevallier, F.,
Dethof, A., Dragosavac, M., Fisher, M., Fuentes, M., Hagemann, S.,
Hólm, E., Hoskins, B. J., Isaksen, L., Janssen, P. A. E. M., Jenne, R.,
Mcnally, A. P., Mahfouf, J.-F., Morcrette, J.-J., Rayner, N. A.,
Saunders, R. W., Simon, P., Sterl, A., Trenberth, K. E., Untch, A.,
Vasiljevic, D., Viterbo, P., and Woollen, J.: The ERA-40 re-analysis,
Q. J. Roy. Meteor. Soc., 131, 2961–3012, 10.1256/qj.04.176, 2005.
von Storch, H. and Reichardt, H.: A Scenario of Storm Surge Statistics for the German Bight at the Expected Time of
Doubled Atmospheric Carbon Dioxide Concentration, J. Climate, 10, 2653–2662, 1997.von Storch, H. and Woth, K.: Storm surges: perspectives and options, Sustain. Sci., 3, 33–43, 10.1007/s11625-008-0044-2, 2008.
Weisse, R., von Storch, H., Niemeyer, H. D., and Knaack, H.: Changing North Sea storm surge climate:
an increasing hazard?, Ocean Coast. Manage., 68, 58–68, 2012.
Woth, K., Weisse, R., and von Storch, H.: Climate change and North Sea storm surge extremes: an ensemble
study of storm surge extremes expected in a changed climate projected by four different regional climate
models, Ocean Dynam., 56, 3–15, 2006.