Uncertainty analysis of river flood management in the Netherlands

M. Kok

HKV Consultants, Lelystad, Netherlands ([email protected])

Delft University of Technology, Faculty of Civil Engineering and Geosciences, Delft, Netherlands

J.W. Stijnen

HKV Consultants, Lelystad, Netherlands ([email protected])

W. Silva

Ministry of Transport, Public Works and Water management, RIZA, Arnhem, Netherlands
([email protected])

ABSTRACT:
The current flood defense design practice along the major rivers in the Netherlands is to include only the natural
variability of water levels (or the discharge) in assessing the exceedance frequency. Other sources of uncertainty which
could cause flooding (such as the roughness of the riverbed or the discharge distribution at the bifurcation points) are
ignored.
In this paper we will show the influence of other uncertainties on the probability of flooding. Instead of the traditional design method (the exceedance frequency of water levels, using only the river discharge as random variable), we
will consider the exceedance probability of (wave) overtopping of the flood defense. We have investigated the failure
frequencies of dike sections and not the flood frequency of dike rings, which always consist of a number of failure
mechanisms, dike sections and hydraulic structures. Therefore the number of random variables remains small enough so
that numerical integration can be used to calculate the frequencies.
It is shown that other sources are a major contribution to the calculated safety against flooding. These uncertainties
also influence the efficiency of measures which reduce the risks of flooding, such as the use of retention areas (for example in emergency situations). In the traditional approach this measure seems highly efficient, but if all uncertainties
are taken into account this measure is less efficient. However, the attractiveness using retention areas depends on the
costs and benefits of this measure, and the approach in this paper is an essential ingredient to assess the benefits. It is
recommended to use the approach in this paper in a cost benefit analysis, and to investigate the influence of the assumptions.

1 INTRODUCTION
The Netherlands are situated in the delta of three
of Europes main rivers: the Rhine, the Meuse and
the Scheldt. As a result of this, the country has been
able to develop into an important, densely populated
nation. But living in the Netherlands is not without
risks. Large parts of the Netherlands are below mean
sea and water levels which may occur on the rivers
Rhine and Meuse. High water levels due to storm
surges on the North Sea, or due to high discharges of
these rivers are a serious threat to the low-lying part
of the Netherlands. Proper construction, management and maintenance of flood defences are essential to the population and further development of the
country.
Without flood defences much of the Netherlands
would be flooded on a regular basis. The influence
of the sea would mainly be felt in the West. The in-

fluence of the waters of the major rivers has a more

(but limited) geographic impact. Along the coast,
protection against flooding is predominantly provided by dunes. Where the dunes are absent or too
narrow, or where the sea arms have been closed off,
flood defences in such as sea dikes or storm surge
barriers have been constructed. Along the full length
of the Rhine and along parts of the Meuse protection
against flooding is provided by dikes. For an overview of the current safety standards along the coast
and major rivers, see Brinkhuis-Jak et al, 2003.
2 DESIGN METHODS
The current safety standard has been set after the
big 1953 flood disaster in the Netherlands. After this
flood the design method of flood protection was improved considerably because of the scientific approach. This approach was invented by the Delta
Committee (Delta Committee 1960, Dantzig 1956).
The default approach for designing flood protection

structures that has been used until then, was based

on the highest recorded water level. In relation to
this water level a certain safety margin (varying
from 0.5 to 1.0 meter) was maintained. The Delta
Committee recommended that a certain desired
safe water level be taken as a starting point. The
safety standards should be based on weighting the
costs of the construction of flood protection structures against the possible damage caused by floods.
An econometric analysis was undertaken by the
Delta Commission for Central Holland. Based on information from 1960 this led to an optimum policy
of 8 10-6 per year. For practical design this was
converted into a design water level with a frequency
of exceedance of 1/10000 per year. These design
frequencies are used for the dike ring areas along the
coast. For the major rivers, however, less strict design frequencies are demanded in the Act of Flood
Defences, because the consequences of flooding in
these riverine areas are less severe than a flood along
the coast. The design frequency of flood defences
along the major rivers has been set to 1/1250 per
year.
The predominantly deterministic determination
assumes the normative Design Water Level (DWL)
that the dike must be able to retain (TAW, 1998).
This water level may include wind effects on the local water level. In addition, the wave run-up is subsequently the most important parameter in the determination of the crest height. In the traditional
deterministic design method the wave run-up is calculated based on certain wind characteristics at the
design water level and the corresponding waves, and
taking into account the geometry of the water defence system. Settlement of the soil body over a
certain planning period is also taken into account to
avoid repair actions, which may be needed if the
height of the dike is below the required reference
level.

In this paper we will follow a probabilistic approach.

3 ASSUMPTIONS
The following assumptions have been made (see
also Stijnen et al, 2002):
a. We investigated only one failure mechanism:
overflow and wave overtopping. Other mechanisms (for example sliding of the inner slope,
piping and micro instability, see TAW, 1998) are
not included. These mechanisms may be important, but in the study TAW, 2001 it is concluded
that overflow and wave overtopping is the dominant mechanism in the probability of flooding,
assuming that the possible weak spots are
strengthened;
b. We investigated the following six locations
along the major Rhine river branches (see for a
map Figure 2):
- Lobith, Upper Rhine river (km 862)
- Millingen, Waal river (km 868)
- Tiel, Waal river (km 915)
- Opijnen, Waal river (km 929)
- Amerongen, Lower Rhine river (km 918)
- Duursche Waarden, IJssel river (km 961)

Figure 2. Overview of the Rhine Branches in the Netherlands

Figure 1. Design of a river dike (TAW, 1998)

In a probabilistic approach the results of the more

deterministic method described above are still
needed as input to the probabilistic calculations. In a
probabilistic approach we are interested in the inundation probability of a dike ring area. Inundation is
caused by failure of the flood defence system. Here,
the whole range of water levels and waves are included in the analysis (TAW, 1998).

c. We did not consider the true dike heights, but the

dikes as they should have been designed according to the design rules of the Technical Advisory Committee on Water Defences;
d. In the assessment of flood defences it is useful to
distinguish between failure and collapse of a
structure. Failure is defined as not fulfilling one
or more water defence functions (the crest of a
part of the flood defence is too low, for example). Collapse means the loss of cohesion or

large deformations in geometry. In this paper we

only handle failure of the water defence.
e. The reliability function Z of the failure mechanism wave overtopping is: Z = qc - Q (Hs , h ),
where qc stands for the critical overtopping discharge (which may be stochastic, but in this paper we will assume that it has a deterministic
value of 0.001 m3/s/m, which is equivalent to 1
l/s/m), Hs is the wave height and h is the water
level.
4 UNCERTAINTY IN PARAMETERS AND
THEIR DISTRIBUTIONS
When designing the height and strength of a water
defense section, there are many (stochastic) factors
which have to be taken into account. It is important
to realize that we are not only dealing with uncertain
parameters, but that each of these parameters has a
distribution of its own that is unknown. Think about
the natural processes such as the river discharge
waves (height and shape), the resulting water levels,
precipitation, wind (speed and direction), etc. On the
other hand, there are also a number of uncertainties
in the creation of models: a hydraulic or hydrological model is never perfect, and neither are the required parameters in these models. Finally, we often
use measurement data. When we use these measurements to estimate parameters, or distributions,
more uncertainties are introduced.
There are basically two categories in which we
can place these uncertainties (Noortwijk et al.,
2002): natural variability and epistemic uncertainties.
4.1 Natural variability
This is sometimes also called inherent uncertainty,
and represents the unpredictability of physical processes. This concerns both uncertainties in time as
well as in space. Uncertainties that are a direct consequence of the variability of natural processes fall
into this category. Think about the direction or velocity of the wind, but also the local or downstream
hydraulic roughness and the discharge distribution
near a river bifurcation point. Uncertainties in the
discharge itself (both the height of the peak and the
shape) are part of this category as well.
4.2 Epistemic uncertainty
The category of epistemic uncertainty (also called
knowledge uncertainty) is a large one, and can be
further subdivided into statistical uncertainty, model
uncertainty and planning uncertainty.
Uncertainties that play a role when determining
the water level on the river, or the discharge into a
retention area belong in the subcategory of model

uncertainties. They arise from prediction models for

the river and the retention area. Other examples of
uncertainties that play a role within the hydraulic
model, are the flow pattern near the inlet construction, and the slope across the inlet construction.
These uncertainties maintain a certain amount of
subjectivity (Cooke, 1991), because their size and
relevance are hard to determine. It is also possible
for these results to be influenced by new research results.
Statistical uncertainty arises when there are not
enough data to estimate the parameters of a probability distribution of a random variable (Kok et al.,
1996, Appendix B). The more data, the smaller becomes the statistical uncertainty. The uncertainties
regarding the choice of the type of probability distribution also fall into this subcategory. Examples are
the probability distributions for the discharge, and
temporal and spatial correlations between the various random variables.
The decision to use a retention area brings with it
a number of uncertainties from these different categories. There are, however, uncertainties that fall in
yet another subcategory, dealing with the organisational side of a measure, especially in the case of
retention. This is closely related to the ability to predict the duration of a discharge wave on a short term
(in the order of days to a week). Of special interest
in this case are questions that concern the actual use
of a retention area. When should the retention area
be flooded? Which retention area should be used?
There are also social and economic aspects surrounding the decision whether to use a retention area
or not, but these will not be discussed in this paper.
5 RESULTS OF DETERMINISTIC AND
PROBABILISTIC CALCULATIONS
For each of the six locations mentioned in Section 3 the failure probabilities for the mechanism
overflow and wave overtopping have been calculated. We also investigated what happens to these
failure probabilities when retention is used as a
measure to increase the safety of dike ring areas. In
each case we investigated the resulting failure probability with and without the measure. This in turn
enabled us to define the term efficiency of the
measure retention as follows:
Efficiency of retention =

Failure probability without retention

Failure probability with retention

With the aid of this definition it is possible to obtain insight in the actual safety benefit of a measure.
In the computations we used the following random
variables and distributions:

The discharge, with actual exceedance probabilities of the discharge peak according to the
working line (Parmet et al, 2002).
The wind direction, with actual statistics for the
measurement station of Schiphol Airport (Geerse
et al, 2002).
The wind speed, with actual exceedance probabilities for the measurement station of Schiphol
Airport (Geerse et al, 2002).
Water level, where a normal distribution is assumed. This is a result from uncertainties around
the river bifurcation points, the geometry, hydraulic roughness and lateral inflow (Stijnen et
al, 2002).

The results that are presented here are based on a

recent study (Stijnen et al, 2002). We made the
computations including the entire shape of the discharge wave. Given the peak of the discharge wave,
the entire shape is assumed to be known. With the
peak of the wave we mean in this case the highest
discharge within a single wave that has a constant
value for a period of 12 hours (an example wave is
shown in Figure 3).

ter level plus an additional minimum safety margin

of 0.5 meter (see also Section 2).
In this paper we present the results with respect to
one flood management measure: the use of retention
areas in case of emergency situations. The efficiency
of other measures (such as Room for the River and
dike heightening) is studied in Stijnen et al, 2002,
but are not presented in this paper. With regard to
retention, we investigated a single area with a volume of 250 Mm3, near the city of Lobith. This volume is inspired by the ideas in a recent advice of the
committee Emergency Retention Areas. (Commissie
Noodoverloopgebieden, 2002). The inlet construction is considered to be ideal, meaning that no restrictions are posed on the amount of inflow, etc.
The inlet sill is kept at a fixed level of 16000 m3/s
(the level it should be according to the current design practice).
In the case of retention the shape and peak of a
discharge wave become important, because they determine the volume of water that needs to be withdrawn from the river. The peak of the wave when
retention is used, can vary in height per location. It
is also possible that the time at which the peak occurs shifts (Figure 4).

17.5
17
16.5
16
15.5
15

-5

0
Time [days]

Figure 3. An example of a discretised water level wave for a

discharge wave with a peak of 16000 m3/s at Lobith, with independent blocks of 12 hours.

Because the number of random variables is relatively small, this enabled us to use numerical integration, instead of other approximation techniques.
The design discharge (the discharge with an annual probability equal to the safety standard of
1/1250) is equal to 16000 m3/s for the Rhine river
(Parmet et al, 2002). With the design discharge, the
design water levels (DWLs) along the river branches
are known. For every location along the Rhine
branches, so-called QH-relations are available that
couple the discharge (Q [m3/s]) with the local water
levels (H [m]). In order to obtain a consistent set of
computations, the height of the dikes at the investigated locations are assumed equal to the design wa-

Peak discharge [m3/s]

Water level near Lobith [m + NAP]

18

Peak without retention

Peak with retention

Time [days]

Figure 4. Illustration that shows the shift of the peak of a

discharge wave in a situation with and without retention. The
peak in the situation without retention occurs at time 0 days,
while in the situation with retention the peak occurs at time t
days.

Five sets of computations have been performed

for each of the six locations. In each of these five
sets another random variable has been added, in order to observe the impact that each additional ran

dom variable has on the failure probabilities. In the

next subsections we will more closely examine these
five computations.
5.1 Random variable: discharge (height of dike
section: DWL [m+NAP])
In this first calculation, we primarily wanted to
establish a base for the rest of the computations. The
results are straightforward, and can be found directly
from discharge frequency function (commonly known as
the working line). The annual failure probability that
is found for each location is equal to the current,
designated safety level of the dikes in the upper river
branches region: 1 / 1250. This is indeed equal to the
annual exceedance probability of the design discharge of 16000 m3/s. In this situation, where we
have only a single random variable, it is possible to
select a single discharge for which the dike section
fails for the first time. A discharge that is higher than
the 16000 m3/s level, will cause a dike section to fail
immediately.
When we make use of the measure retention, the
failure probability for each of the five locations decreases substantially, and is reduced to 1 / 4548 per
year. In each case we also calculated the corresponding critical discharge, which is the lowest discharge that causes failure of the dike section. After
retention the critical discharge is no longer equal to
16000 m3/s but has increased to 17700 m3/s (see also
Figure 5). Even though the retention area starts to
fill up, there is still a positive effect visible on the
water levels downstream. The impact of retention is
no longer noticeable for discharges above 18700
m3/s, where the two working lines are equal again.
-1

10

Working line
"Working line" after the retention inlet point

-2

Exceedance frequencies

10

-3

10

1 / 1250
1 / 4548

-4

10

1 / 9722

-5

10

18700 m3/s
-6

10
10000

12000

14000 16000 18000 20000

Rhine river discharges [m3/s]

22000

Figure 5. The impact of retention on the exceedance frequencies of the Rhine discharge downstream the inlet construction

Obviously, there is a very large, positive impact

of retention for this calculation. The efficiency of the

measure retention on the failure probability in this

case is equal to Eret = 4548 / 1250 = 3.6.
5.2 Random variable: discharge (dike height: DWL
+ 0.5 [m+NAP])
Again, we focus only on the discharge as a random
variable, but this time the dike sections include an
additional margin of 0.5 meter (which is needed for
local wind waves). For each location, the discharge
that is required to raise the water level to this additional height is different. Hence, the results are different for each location, because the relations between the discharge and the water level (the QHrelations) become important. The failure probabilities are collected in Table 1, and the corresponding
critical discharges are shown in Table 2.
Table 1. Failure probabilities, caused by overflow of a dike
section
with height (DWL + 0.5 [m+NAP]).
_________________________________________________
Location
Failure probability [-]
Efficiency
___________________________
________
Without retention With retention
_________________________________________________
Lobith
1 / 3907
1 / 5712
1.46
Millingen
1 / 6163
1 / 6900
1.12
Tiel
1 / 4215
1 / 5712
1.36
Opijnen
1 / 3907
1 / 5712
1.46
Amerongen 1 / 28160
1 / 28160
1.00
D.W.
1 / 7441
1 / 8352
1.12
_________________________________________________
Table 2. Critical discharges in the situation of overflow of a
dike
section with height (DWL + 0.5 [m+NAP]).
____________________________________________
Location
Critial discharge [m3/s]
____________________________
Without retention With retention
____________________________________________
Lobith
17500
18000
Millingen
18100
18300
Tiel
17600
18000
Opijnen
17500
18000
Amerongen
20100
20100
D.W.
18400
18500
____________________________________________

The first thing we notice is that the failure probabilities without retention have become smaller
(compared to the previous section). From the results
it becomes clear that the QH-relation at Amerongen
is very flat: a huge discharge is required to cause
overflowing of the dike when it has been heightened
by 0.5 meter. The steepest QH-relation is the one for
Opijnen, and consequently this location also has one
of the lowest critical discharges.
The results for retention vary somewhat. Clearly
the effect of retention on the failure probability is
positive, although not nearly as large as in the computation of Section 5.1. This can also be noticed in
the limited efficiency. The reason for this diminishing effect is largely due to the fact the inlet sill is
kept at a fixed level of 16000 m3/s. For the location
of Amerongen, the retention area has already been
filled before the peak of the discharge wave arrives,
and is therefore useless unless the inlet sill is raised.

Ideally, the discharge level for which water

should be drawn-off into the retention area is equal
to the critical discharge for that specific location (see
Table 2).
5.3 Random variables: discharge and water level
(height of dike section: DWL + 0.5 [m+NAP])
In addition to the discharge, we now also include
the water level as a random variable. The height of
the dike section is again equal to DWL + 0.5
[m+NAP]. The total discharge wave is now included.
With two random variables it is possible that failure
of a dike section can occur not only at the peak of
the discharge waves (as was the case in Sections 5.1
and 5.2), but also at discharge levels below the peak.
It is assumed that the uncertainty surrounding the
bifurcation points of the river can be modeled by a
normal distribution with a standard deviation that is
equal to 1% of the inflow at a bifurcation point. The
mean of the distribution is assumed to be equal to
the local water level.
Because of lateral inflow and differences in
roughness and geometry, the uncertainty in the water
level is different for specific branches of the river.
For each of the river branches, the used standard deviations can be found in Table 3. The branches that
start after the two bifurcation points in the Rhine
(the Lower Rhine and the IJssel) have the largest
standard deviation, which is caused by the fact that
the uncertainties surrounding these bifurcation
points add up.
When retention is introduced as a measure, an
additional uncertainty is incorporated that is related
to the uncertainties of retention. It is assumed that
this uncertainty is different for each of the branches
of the Rhine. The adapted standard deviations per
branch in the case of retention are also shown in Table 3.
Table 3. The used standard deviations of the water level per
river
branch
____________________________________________
River branch
Standard deviation [m]
____________________________
Without
retention With retention
____________________________________________
Upper Rhine
0.11
0.15
Waal
0.12
0.14
Lower Rhine
0.17
0.17
IJssel
0.25
0.25
____________________________________________

The results of the calculations can be found in

Table 4. The failure probabilities in the situation
without and with retention are both shown, as well
as the resulting efficiency factors.
Table 4. Failure probabilities, caused by overtopping of a dike
section, including both discharge and water level as random
variables
(DWL + 0.5 [m+NAP]).
_________________________________________________
Location
Failure probability [-]
Efficiency
___________________________
________
Without retention With retention
_________________________________________________

Lobith
1 / 3265
1 / 5119
1.57
Millingen
1 / 4295
1 / 5965
1.39
Tiel
1 / 3311
1 / 5248
1.59
Opijnen
1 / 3151
1 / 5107
1.62
Amerongen 1 / 7387
1 / 9752
1.32
D.W.
1
/
2494
1
/
3547
1.42
_________________________________________________

We see that the impact of the additional uncertainty in the water level has a distinctly negative impact on the failure probability (compared to the previous sections). In particular for the locations with
large standard deviations the effects are quite large
(such as Amerongen and the Duursche Waarden).
The efficiency of retention has even increased
compared to the previous section. The reason for this
is the combination of uncertainties in the water level
and the use of the shape of the discharge wave. In
the case of overflowing of a dike section without uncertainties in the water level, it is possible to select a
single discharge for which the dike section will
overflow. A location such as Amerongen, for which
the retention area has already been completely filled
before the peak of the discharge wave passes, will
never profit from retention. When the uncertainties
in the water level are included there are multiple
times within one discharge wave at which overflowing may occur. This increases both the failure
probabilities and the efficiency, because other discharges (water levels) besides the peak of the wave
are important.
5.4 Random variables: discharge, wind direction
and wind speed (height of dike section: DWL +
0.5 [m+NAP])
In this calculation we did not only look at overflowing of a dike section, but at wave run-up due to
the effect of the wind as well. So instead of uncertainties in the water level, we now added the speed
and direction of the wind as random variables, besides the discharge. Again, the effect of the entire
shape of the discharge wave is important, because
failure of a dike section may occur not just at the
peak. The results for both the failure probabilities
and the efficiency can be found in Table 5.
Table 5. Failure probabilities caused by overtopping of a dike
section, with discharge, wind direction and wind speed as random
variables (DWL + 0.5 [m+NAP]).
_________________________________________________
Location
Failure probability [-]
Efficiency
___________________________
________
Without retention With retention
_________________________________________________
Lobith
1 / 1708
1 / 2398
1.40
Millingen
1 / 1741
1 / 2254
1.29
Tiel
1 / 1856
1 / 2691
1.45
Opijnen
1 / 1149
1 / 1293
1.13
Amerongen 1 / 5054
1 / 7357
1.46
D.W.
1
/
2091
1
/
2819
1.35
_________________________________________________

The failure probabilities have increased significantly in comparison to Table 3, even more so than

in Table 4. A location that seems particularly vulnerable to effects of wind-induced waves is Opijnen,
where the failure probability even drops below the
safety standard of 1 / 1250. The effect of retention
is also very poor for this location.
5.5 Random variables: discharge, water level,
wind direction and wind speed (height of dike
section: DWL + 0.5 [m+NAP]).
The computations in this section are a combination
of the random variables in Sections 5.3 and 5.4.
Again, we investigated only the failure mechanism
of overtopping due to overflowing of a dike and due
to wave run-up. The results for both the failure
probabilities and the efficiency can be found in Table 6.
Table 6. Failure probabilities caused by overtopping of a dike
section, with discharge, water level, wind direction and wind
speed
as random variables (DWL + 0.5 [m+NAP]).
_________________________________________________
Location
Failure probability [-]
Efficiency
___________________________
________
Without retention With retention
_________________________________________________
Lobith
1 / 1583
1 / 2055
1.20
Millingen
1 / 1514
1 / 1808
1.19
Tiel
1 / 1678
1 / 2219
1.32
Opijnen
1 / 1064
1 / 1169
1.10
Amerongen 1 / 2059
1 / 2458
1.19
D.W.
1
/
1015
1
/
1106
1.09
_________________________________________________

We see that besides the location of Opijnen, the

location of Duursche Waarden now drops below the
annual safety standard of 1 / 1250 as well. Both locations are sensitive to wind effects and uncertainties in the water level, and these properties translates itself into an unfavorable efficiency for
retention. The impacts for the other locations can be
less clearly distinguished, although clearly the location of Amerongen benefits from a flat QHrelation.
5.6 Random variables: discharge, water level,
wind direction, wind speed and the parameters
of the frequency-discharge relation (height of
dike section: DWL + 0.5 [m+NAP]).
As a further extension to the calculations of Section
5.5, the parameters in the discharge frequency function are considered to be uncertain. For details we
refer to Stijnen et al, 2002). The results are shown in
Table 7.
Table 7. Failure probabilities caused by overtopping of a dike
section, with discharge, water level, wind direction, wind speed
and the parameters of the frequency-discharge relation as random
variables (DWL + 0.5 [m+NAP]).
_________________________________________________
Location
Failure probability [-]
Efficiency
___________________________
________
Without retention With retention
_________________________________________________
Lobith
1 / 1129
1 / 1422
1.26

Millingen
1 / 1101
1 / 1296
1.18
Tiel
1 / 1185
1 / 1515
1.28
Opijnen
1 / 824
1 / 903
1.10
Amerongen 1 / 1469
1 / 1737
1.18
D.W.
1
/
807
1
/
880
1.09
_________________________________________________

For each of the investigated locations the failure

probability increases with approximately 30% to
40%. This is roughly consistent with an increase in
the design discharge of 500 m3/s.
Regarding the efficiency, we see that in this case
not much has changed compared to the results of
Table 6. This can be explained by realizing that the
effect of the uncertainties in the frequency discharge
relation is present in both the situation with and
without the measure retention.
6 DISCUSSION
The results of Section 5 are perhaps somewhat surprising: the retention area seems to have a big influence on the exceedance frequency of water levels,
(see section 5.1), but not on the failure probability of
wave overtopping. It is important to realize that
(many) people and cattle are living in the retention
areas, and in order to actually use the areas they
have to be evacuated. A flood forecast of 1-2 days is
needed for evacuation of the area. The river discharge can be forecasted reasonably accurate, but
the other sources of uncertainty cannot. In the calculations it is assumed that the retention area is used
when the river discharge exceeds the level of the design discharge, 16000 m3/s. However, since the
dikes are higher than the water levels that correspond to the design discharge, the retention area is
sometimes used unnecessary. On the other hand, the
system may fail for discharges that remain lower
than the design discharge (due to uncertainties in the
water level).
In order to interpret more detailed the results in
Section 5, we distinguish three different classes in
which locations can differ: the slope of the QHrelation, the sensitivity to wind-induced waves and
the uncertainty in the water levels. For each of the
three classes we have made a rough indication (see
Table 8).
Table 8. The influence of the QH-relation (1), the uncertainty
on the water level (2), and wind-induced wave effects (3). The
meaning
of the symbols is given in the text.
__________________________________________________
Location
Impact on exceedance probabilities
(small/large)
___________________________
___________
(1)
(2)
(3)
__________________________________________________
Lobith
ooo
o
o
Millingen
oo
oo
oo
Tiel
ooo
o
o
Opijnen
ooo
o
oo
Amerongen o
ooo
ooo
D.W.
oo
ooo
ooo
__________________________________________________

An explanation of the symbols that are used in Table

8 above is given below:
(1) For the QH-relation a o indicates a relatively
flat gradient, while a ooo stands for a relatively steep gradient. It is in fact a comparison
of the failure probabilities of Table 1.
(2) The uncertainties in the water level are closely
related to the uncertainties regarding the two bifurcation points. The Duursche Waarden and
Amerongen are located after the two bifurcation
points and the impacts are therefore the largest
for these two locations. In this case we have
compared Table 3 with Table 5.
(3) The effect of waves is largely influenced by different effective fetches and the orientation of
the dike section. A o means that a location is
not particularly sensitive to the influence of the
wind (and waves). In contrast, a ooo indicates
a location that is sensitive to wind effects. Examples of such locations are the Duursche
Waarden en Opijnen. Now we compared the results of Table 3 and Table 6.
In comparison, the weight of the slope of the
QH-relation is larger than that of the other two criteria. This follows for example from the results of
Amerongen, which is a location that is influenced
substantial by uncertainties in the water level as well
as by waves, but still comes out as a relatively safe
location.
7 CONCLUSIONS & RECOMMENDATIONS
The following conclusions and recommendations
can be drawn:
a. The safety standard (as it is given in the Flood
Protection Law) is an exceedance frequency of
water levels, and along the river Rhine this is
equal to 1/1250. In the actual design method
only one random variable is included: the discharge at the boundary with Germany (Lobith).
If we include uncertainties, the failure probability depends on the properties of the location, but
is still in the same order of magnitude of the
safety standard (range: 1/800 1/2000).
b. The efficiency of retention is defined by the failure probability without retention divided by the
failure probability including retention as an option. This efficiency strongly depends on the inclusion of all the uncertainties which may cause
failure or collapse of a dike. If we take only a
single random variable into account (the peak of
the river discharge) this factor may be equal to
3.6, whereas if we take all relevant uncertainties
into account this factor falls in the range of 1.1 to
1.3. The efficiency reduces significantly if we
take these uncertainties into account, but the retention areas still have a positive impact on the
failure probability.

c. It is possible that the conclusions depend on the

assumptions that were made in the study. We
have used, for example, the heights of dikes as
designed by the actual design rules (instead of
the real heights of the dikes). Another assumption that has been made is that the retention area
is used as soon as the river discharge rises above
16.000 m3/s. It is also very well possible that, instead of such a simple control strategy, more advanced control strategies are desirable. Moreover, the different sources of uncertainty may be
reduced with additional measures, which may
increase the efficiency of retention. We recommend that the influence of these assumptions on
the results of this study are investigated further.
d. In a decision-analysis framework, the efficiency
factor as such is not important. However, this
factor can be used to assess the benefits of flood
management measures with greater accuracy. In
a cost-benefit analysis (see for example Vrijling,
1990 or Brinkhuis-Jak et al, 2003) the costs and
benefits of measures are optimized. We recommend to use the efficiency factor in a costbenefit analysis of (different sorts of disaster
management) measures to reduce the expected
flood damage.
Disclaimer
Any opinions expressed in this paper are those of the authors
and do not necessarily reflect the position of the Dutch Ministry of Transport, Public Works and Water Management.

REFERENCES
Brinkhuis-Jak, M., Holterman, S.R., Kok, M. & Jonkman, S.N.
2003. Cost benefit analysis and flood damage mitigation in
the Netherlands. ESREL conference, Maastricht.
Cooke, R.M., 1991. Experts in Uncertainty. Opinion and subjective probability in Science. London: Oxford University
Press.
Commissie Noodoverloopgebieden 2002. Gecontroleerd overstromen (in Dutch). The Hague.
Dantzig, D. van 1956. Economic decision problems for flood
protection. Econometrica, 276-287, New haven.
Delta Committee 1960. Report of the Delta Committee (in
Dutch) The Hague, The Netherlands.
Geerse, C.P.M., Duits, M.T., Kalk, H.J. & Lammers, I.B.M.,
2002. Wind-waterstandstatistiek Hoek van Holland (in
Dutch) HKV CONSULTANTS & RIZA, juli 2002.
Kok, M., Douben, N., Noortwijk J.M. van, & Silva, W., 1996.
Integrale Verkenning inrichting Rijntakken. Report 12:
Veiligheid (in Dutch). WL & RWS Riza, Arnhem.
Noortwijk, J.M. van & Kalk, H.J. 2002. Onzekerheden in
terugkeertijden van rivierafvoeren.(in Dutch). Client:
RWS-DWW. HKV CONSULTANTS, PR560, September 2002.
Parmet, B.W.A.H., Langemheen, W. van den, Chbab, E.H.,
Kwadijk, J.C.J., Diermanse, F. & Klopstra, D, 2002. Analyse van de maatgevende afvoer van de Rijn te Lobith (in
Dutch). Rijkswaterstaat, RIZA report 2002.012, 2002.
Stijnen, J.W., Kok, M. & Duits, M.T. 2002. Onzekerheidsanalyse Hoogwaterbescherming Rijntakken, onzekerheidsbron

nen en gevolgen van maatregelen (in Dutch), HKV

CONSULTANTS, November 2002, PR464.
TAW, 1998. Fundamentals on Water Defenses. Technical Advisory Committee on Water Defenses, Delft.
TAW, 2000. Van overschrijdingskans naar overstromingskans
(in Dutch), Technical Advisory Committee on Water Defenses, Delft.
Vrijling, J.K. & Beurden, I van, 1990. Sealevel rise: a probabilistic design problem. Coastal Engineering, 1990.
Vrouwenvelder, A.C.W.M., Steenbergen, H.M.G.M &
Slijkhuis, K.A.H. 2001. Theoriehandleiding PC-Ring, Deel
A: Mechanismenbeschrijving (in Dutch). TNO, Delft.
Vuuren, W. van, 2002. Onzekerheden in de WAQUA voorspellingen voor de MHWs op de Rijntakken en Maas (in
Dutch). Rijkswaterstaat RIZA, Arnhem, Memo WSR 2002006, February 2002.