OTC 7683

Response Based Design Conditions in the North Sea:

Application of a New Method
Peter S. Tromans and Luc Vandersohuren, Shell Research, Rijswijk, The Netherlands

Copyright 1995, Offshore Technology Conference

This paper was presented st the 27th Annual OTC in Houston, Texas, U. S.A., 1-4 May !995.

This paper was selected for presentation by the OTC Program Committee following review of information contained In sn abstract submitted by the author(s). Contents of the paper,
w presented, have not been reviewed by the Offshore Technology Conference and are subject to cerrectlon by the author(s). The material, as presented, does not nscesssrily reflect
any position of the Offshore Technology Conference or its officers. Permission to copy is restricted to an sbstract of not more thsn 300 words. Illustrations may not be copied. The abstract
should contain conspicuous acknowledgment of where snd by whom the paper is presented.

SUMMARY from a combination of waves, currents and wind, though waves

The estimation of rare, extreme environments from a relatively are generally the dominant factor. The problem of specifying
short oceanographic database is crucial to the calculation of meteorological and oceanographic (met-ocean) design
loads and responses of offshore structures. A novel and conditions is one of estimating these environmental variables
consistent approach has been developed that makes use of a corresponding to some return period, typically 100 years, on the
number of asymptotic properties of extremes. A storm can be basis of measured or hindcast time series extending over a
adequately characterised by its most probable extreme wave or relatively short period, 5 to 25 years. For the North Sea and
the resultant structural response. This allows us to treat storms, some other areas, a wi&ly used approach to obtaining design
rather than sea states, as the essential random, independent wave heights has involved fitting cumulative distributions to the
events; to account correctly for uncertainty in the largest wave sign~lcant wave heights of successive three hour sea states
and structural response within a storm; and to deal with the [1,2]. U is common to neglect both the correlation between
uncertainty in the seventy of randomly arriving storms, successive sea states and uncertainty in the extreme wave of a
including the rare storms that are more severe than those in the sea state. In spite of this, the distributions are used to estimate a
database. The method provides the long term load statistics 100 year return value. Similar processes are used separately for
essential to reliability analysis and to the calibration of each met-ocean variable. The consequent 100 year wind, 100
environmental Ioad factors in design codes. year wave and 100 year current are conservatively assumed to
occur simultaneously and act in the same direction. For a
We demonstrate the application of the method to the prediction typical jacket structure, this will lead to a “design load” that is
of extreme waves, loads and “response based” environmental much more severe than the “true” 100 year load.
design conditions for a drag dominated structure in the North
Sea. Some rather different methods of estimating extreme waves,
originally developed for the analysis of hurricane data in the
INTRODUCTION Gulf of Mexico, are summarised by Haring and Heideman [3].
In all these approaches the assumption of independence of sea
The estimation of extreme waves states is avoided by using storms as the independent events. A
The loads induced by extreme storms are critical in the design storm may be loosely defined as a continuous period, typically
of offshore structures for location in severe seas, The load arises 12 to 36 hours, of severe sea characterised by a development
phase during which the significant wave height increases, a
References and figures at end ofpaper peak and a subsequent decay phase in which the sea moderates,

387
2 Response Based Design Conditions in the North Sea OTC 7683
as shown in Figure 1. In one method a storm is specified by the the way it characterises the storms. Our method uses the most
significant wave height at its peak. The other sea states of the probable extreme individual wave (or structural response) of the
storm and the uncertainty in the extreme wave of a storm are storm history rather than the peak sign ~lcant wave height. This
described by statistics conditional only on the peak significant distinction is an important one. The most probable extreme
wave height. The method was applied by Jahns and Wheeler individual wave height is a function of severaI of the most
[4]. Discarding all the significant wave heights except for the severe sea states of the storm. Thus, our method uses more of
storm peaks appears to involve some information loss. the data and should therefore be less sensitive to noise.
Moreover, the storm peak is likely to be sensitive to the
measurement and sampling noise in measured data and the Outline of the present paper
smoothing of hindcast data. In this paper we summarise the results of the statistical theory
Another method avoids this problem. The desired long term that allows a storm to be characterised by its most probable
distribution is obtained by calculating the distribution of extreme wave or structural response. This provides a model for
extreme wave height at each location for each storm in the the uncertainty of the extreme within a storm, the “short time”
database and then averaging these distributions over all the scale statistics. The “long time” scale statistics can then be
locations and storms. Ward et al [5] and Fornstall et al [6] have modeiled by fitting a distribution to the most probable extreme
used this approach. It is useful only if long met-ocean time wave height or response calculated for the historical storms in a
series are available over an homogeneous area of ocean that is hindcast database. Convolution of the short and long time scale
large compared with the length scales of meteorological models provides a complete long term distribution. The bulk of
conditions. This restricts it to the analysis of seas such as those the paper explains the application of the method to hindcast
generated by hurricanes in the Gulf of Mexico. met-ocean data from the Northern European Storm Study
(NESS) [13]. An extreme wave is estimated. The method is
Joint met-ocean or response based design conditions used together with generic load models to develop long term
Gordon et al [7], Heideman et al [8], Forristall et al [6] and load statistics and response based design conditions. Finally, we
Prior-Jones and Beiboer [9] have summarised much of the discuss the reductions in design loads that arise, the potential
literature on the joint probability of environmental variables, consequences for cost saving and structural reliability, and the
particularly wind, wave and current. Studies for waters off appropriate environmental load factors for a load resistance
Europe, USA and Australia suggest that accounting for joint factor design (LRFD) format.
probability in the specification of environmental conditions THEORETICAL BASIS - ASYMPTOTIC PROPERTIES
reduces design loads on drag dominated structures by 10 to
OF EXTREMES
40%. The environmentaf variables used for the design of a
structure should be based on its response to the environment In this section we summarise the basic statistical results on
[10,1 1]. The generation of joint met-ocean or response based which the method is founded. The analysis will be discussed in
design conditions requires a generic load model. This model a future paper. These are extensions of results for time histories
efficiently combines the environmental variables into a near high maxima in random processes and distributions of
structural response, such as base shear force. extremes [14]. Assuming that the time series of significant wave
height H$(() can be transformed into a Gaussian variable x(r), it
Methods of obtaining joint met-ocean conditions fall into two
is possible to show that the behaviour of the history of H$
categories. In the fiit, the generic load model defines a surface
exhibits some self-similarity in the region near a high
in the multi-variate space of met-ocean variables (see Coles and
maximum. As a consequence the distribution of the extreme
Tawn [12]) and the extremely complex problem of performing
individual wave height, H, or crest elevation, a, of a storm
joint statistics of all the met-wean variables including time
converges to an asymptotic form conditional on the most
scales and directions arises. Methods in the second catego~
probable extreme value. Thus,
avoid this diftlculty. In these, we use the generic load model to
combine the environmental variables directly into a structural P(HIHmP)- exp - exp-in N((H/HmP)2- 1) (1)
response and proceed to analyse the statistics of this response.
Such an approach was used, together with averaging over many P(alaw) - exp - exp-in N((alam~ - 1) (2)
storms by Forristall et al [6] to derive oceanographic criteria for where N = TJT and h N = 8 for northern European winter
the Gulf of Mexico. storms. T~is a time scale of the storm and T is a wave period.
Here, we demonstrate a new method that avoids most of the The theory shows that the most probable maximum wave
shortcomings of the existing approaches. It is equally applicable height, HmP, and crest elevation, amp, are controlled by sea
to the prediction of extreme waves, response based, joint met- states within eighty per cent of the ped H~.
ocean conditions and long term load statistics. It employs These results explain, substantiate or are compatible with
storms as the independent events. It has some similarity to the several other author’s works [2, 15,16, 17]. Equations (1) and (2)
method of Jahns and Wheeler, but is distinguished from it by are fundamental to the remainder of the paper.

388
 OTC 7683 P.S. Tromans and L. Vanderschuren
3
PREDICTION OF EXTREME WAVES If there are ni waves in interval i, then the distribution of the
This section illustrates the application of the method to the largest of these ni waves is
prediction of an extreme wave for a NESS grid point in the P(Hli) = [R(HIH$i)]ni = exp -niQi
northern North Sea. The process follows the steps in Figure 2.
If the interwds of the storm, s, are numbered over -1< i <J,
The starting point is a time series of significant wave height.
where i = O is the peak of the storm, then the probability
Identification of storms and directional sectors distribution of the largest wave of the whole storm is
In the prediction of extreme conditions, moderate seas play no P(Hls) = .lHJ P(Hli) = exp - X.lJ niQi
role and we discard ~1 time interv~s in the database with
The range (-IJ) should k chosen such that the excluded
significant wave height less than 30 to 40 70 of the largest
intervals do not influence P(HIs).
vaiue. This reduces the data to a manageable quantity and
breaks the data base into “storms.” However, a very small This procedure is repeated to produce P(HIs) for each storm, s =
number of the resulting storms have twin peaks. If the trough in 1 to S, in the time series.
significant wave height between the two peaks is less than 80 %
of the lowest peak, they are broken at the trough to form two The “short time” variability
storms. Otherwise they are maintained as one. Application of The distributions, P(HIs) should converge to an asymptotic
these steps to 25 years of data for a North Sea location produces form, conditional only on HmP, the most probable value of the
several hundred storms. extreme individual wave height of the storm. The HmPof each
Winter storms come from different directions. Some directions storm is easily obtained since P(HIs) = l/e when H = HmP.This
correspond to more severe winds, longer fetches and deeper result is a consequence of equat ion (1).
waters and produce the most severe seas. Thus, we divide the The theory is tested for this location by re-plotting the
storms into directional sectors. The sectors can be selected using probability distributions against H/Hmy. Figure 4 shows the
a polar plot of the peak significant wave height against direction good agreement between the distributions normalised in this
of the storms. We define the direction of a winter storm by the way and the theoretical model given by equation (I).
mean wave direction at the storm peak. The directions bounding
the chosen sectors should be checked against charts to confirm The “long time” statistics of storms
consistency with fetch limits, Figure 3. The sectors should be To specify the long time statistics of storms we must obtain a
sufficiently narrow to resolve directional sensitivity, but probability distribution for HmP. There is no theo~ @ indicate
sufficiently wide to contain a statistically sign~icant number of the appropriate form of this distribution. The sample values of
storms, typically order 100.
HW (calculatti above) do not directly provide information
We now have a sample of storm histories for each directional about the rarer, more severe storms with return periods greater
sector. For this example, we shall continue the calculation for than the duration of the original time series. To predict the
the most severe sector. upper tail of the distribution it is necesstuy to make some
assumption about the “true” distribution that the data obey. We
The probability distribution of the extreme wave of a shall assume the tail of the distribution asymptotes to a limiting
given storm history form. We have used both the Weibull and the generalised Pareto
distribution. The choice is case dependent. For this example, we
It is straightforward to combine the probability distributions of
shall continue the dew-iption with the generalised Pareto
individual waves and obtain the probability distribution of the
distribution (GPD) [21]. For a variable x, this is
extreme wave of a ston-n of given history, such as that in Figure
1. There are several ways of describing the individual wave and
l–(l+yx/c)-’h, foro>O, y#O
its statistics. The present theory can accommodate any choice of GPD(x;cs,y) =
variable or probability distribution; wave height or crest l–exp(-x/cr), forts >0, y =0
{
elevation with a Rayleigh distribution or the empirical
distributions of Fornstall [18] or Krogstad [19], or Haring et al with x >0 and 1+ y x/cT>0. Depending on the vahe of y, the
[20]. Here we shall use wave height and treat it as Rayleigh GPD represents a tail that decays algebraically or exponentially
distributed. or a tail with an upper bound. The last property is a convenient
one since it is reasonable to expect fetch and bathymetry limits
Consider three hour interval i of storm s extracted from a time to impose an upper bound on storm intensity for some locations.
series of met-ocean data. The probability distribution of
individual wave height in interval, i, which has a significant The parameters y, CTmust be estimated using the sample values
wave height of Hsi, is, according to Rayleigh, of HmP.It is ciear that the largest vaks are most relevant in
fitting a tail. The question arkx how many values should we
P(HIH~i)= R(HIH~i)= 1- exp - 2(HlH~if = 1- Qi USe? If we use too many we shall include some that do not fit

389
4 Response Based Design Conditions in the North Sea OTC 7683
the distribution and the estimates will be biased. Too few values angle between mean wave and current directions, 19C,and a
will lead to uncertainty due to small sample size. A criterion
between mean wave and wind directions, 0 ~. Over the dum
that helps our engineering judgement in these circumstances is
of a sea state, all the met-ocean variables are treated
the mean square error,
constants with the exception of crest elevation which descr
MSE = [Bias(~(r))]2 +Var(f(r)) the individual wave. Base shear is given by

where ~(r) is the estimate of y based on r values. A4SE is F = ~u’ +~uaTOcos6c +A3@2a2 +A4uOa2 COSOC
IT
minimised when r = k and the smallest of the k storms has H~P +~@2a’ ]T2 +A6@2a2T’ +~W2 Cosfiw
= Hmpmin. rn the present calculations the bias and variance have
been estimated by “bootstrapping” [22]. where ~ to A7 depend on the configuration of the structure
the attack direction. The dominant term in the expressio
Figure 5 shows the result of fitting the GPD to Hmp data. The A,@’a*, the component of drag arising from wave kinem:
plot shows the probability of exceedance of Hmp, obtained from
below mean water line. A similar expression can be obtaina
the GPD together with the ordered sample data. The
overturning moment, M, with ~ to A7 replaced by ~, to
logarithmic scale is chosen since this would display an
Generic models of this general form can be used to represent
exponential tail as a straight line. However, this exaggerates the
many space frame structures [24-29].
importance of the highest vaiues. Differentiation of the GPD
gives, p(HrnP), the probability density function of HmPbased on The constants ~ to A17have been obtained by using a classical
the k stormswith Hmp> Hmpmin. least square method to fit the GLM’s for base shear and
overturning moment. The numerical values were generated
The distribution of the extreme wave of a random
using Newwave kinematics [30], a 1/7 power law current profile
storm and of an interval T
and a structure consisting of a one metre diameter column
Combining the short time scale model, P(HIH~p), given by roughened from mud-line to sea level and smooth above. A
equation (1), with the long time scale model, p(Hmp), deduced range of input variables was selected to provide a representative
from the data, we can obtain the distribution of the extreme set of around one thousand values of each response. Figure 8
wave height of any random storm (r.s.). Thus, shows loads obtained from the generic load model plotted
against numerical predictions. The coefficient of correlation
squared is greater than 0.99.

Distribution of extreme response within a storm

A random storm is any storm with HmP> H~Pm., but otherwise
unspecified. F’(Hlr.s.) is plotted in Figure 6. The mean arrival We next obtain an asymptotic form for the distribution of the
rate of random storms can be estimated from the sample as ~ = extreme response, X, within a storm. X may represent base
(k+ 1)/~, where 7 is the total duration of the time series of met- shear force, F, or overturning moment, M. According to
ocean data. Uncertainty in the arrival rate has no significant equation (2) above, for severe storms, the probability
consequence so long as k is reasonably large. Treating storm distribution of the extreme crest, a, of the storm is given by
arrivals as a Poisson process, the extreme wave of an interval T P(alamp) - exp - exp-in N((a/amp~ - 1) with h’V = 8. (2)
has the probability distribution,
The dominant term in the GLM determining the distribution of
P(HIT) = [P(Hlr.s.)JoT for ~Tlarge. extreme response is A,@*a*, wave drag below mean water line.
A plot of extreme wave height against return period is drawn in This term will govern the shape of the distribution of extreme
Figure 7. response. Thus, for severe storms, the distribution of the
extreme response X is of the form
THE GENERIC LOAD MODEL AND THE EXTREME
P(XLYmp)- exp - exp-h N(X/Xmp- 1) with lnN = 8. (3)
RESPONSE TO A STORM
The generic load model (GLM) is used to transform statistics of PREDICTION OF EXTREME LOAD STATISTICS
ocean surface elevation into structural response statistics. This section summarises the prediction of extreme load
Though other approaches have been developed [6,23], the statistics, which follows a very similar procedure to that for
GLM’s used here are based on the stick models for statically wave height. The met-ocean time series is broken into storms
responding, drag dominated structures described by Tromans et exactly as in calculation of extreme waves.
al [24]. These provide expressions for base shear and mud-line
overturning moment in terms of linear crest elevation, a,
directional spreading factor, UI, zero crossing period, T, depth
integrated current, u, one minute sustained wind speed, W,

390
 OTC 7683 P.S. Tromans and L. Vanderschuren 5
The probability distribution of the extreme load in a
storm histoty
P(xwlr.s.) = J:“- P(xlxw)P(&#xw
T-

Individual linear crest elevations within a stationary sea state The extreme load of an interval T has the probability
obey a Rayleigh distribution. The generic load model can ke distribution,
used, with the appropriate wave period, current and directions
for the sea state, to transform crest statistics into response P(XIT) =[P(Xlr.s.)];~ for ;Tlarge.
statistics. For intervat i with sign~lcant wave height of H$j of
This gives load for any desired return period. A plot of extreme
storm s extracted from a time series of met-ocean data, the
load against return period is drawn in Figure 11.
probability distribution of individual wave crest elevation is
RESPONSE BASED DESIGN CONDITIONS
P(aRsi) = R(al H.j) = 1- exp - 8(al H# = 1- Qi
Thus, the probability distribution for the largest load produced At this point we assume that the present methods have been
by the ni waves in the interval is applied to individual wave height, crest elevation and the
statistics of four response variables: base shear force and mud-
P(xli) = [R(X,:l (X)1 HJ/~i = exp -niQi(X,:l (X)) line overturning moment, both with and without wind forces.
For each a 100 year return value can be estimated Hlw, al~,
where Xis For Mand X. Xi(a) is given by the GLM with the
Flw, MIGU, Flmw, Mlww. The subscript w indicates the
appropriate sea state properties and a = X,:’(X) is its inverse. inclusion of wind load. Since the return values of responses
Combining the intervals - f to J, the probability distribution of pertain to the generic structure, they are of no direct value for
the largest load of the whole of storms is offshore structural design. However, they can be used with the
generic load model to deduce response based design conditions
P(xls) = .,lTJ P(xli) (wave, in-line current and wind). These conditions will produce
loads close to the true 100 year values when applied to a
where X is given by the GLM as above.
structure of the same generic type.
Thus, P(Xk) is calculated for each storm, s = 1 to S.
Without wind load, the two GLM equations are
The “short time” variability
F100=A,u2 +A2uaTQ+A3@2a2 -tA4u@a2 /T
The distributions, P(XIS), should converge to equation (3), the (4)
+A$02a3 IT* +A6@2a2T’
asymptotic form, conditional only on X~P, the most probable
extreme individual load in the storm. The value of XMPof each M,w = A,,u2 +A,2uaT@-t. A,3cD2a2+A,4u@a2 f T
storm is easily obtained since, if the asymptotic form is valid, (5)
+A,502a3 JT’ +-A1602a2T2
P(n) = l/e when X = XmP. The theory is tested for this
location by re-scaling the probability distributions on XIX~P. In addition, we know the values of FIW, Mlw, and aim. The
Figure 9 showsthe goodagreementbetweenthe distributions kinematics reduction factor, 0, for wave spreading is usually
normalised in this way and the theoretical model. estimated from offshore measurements. The mean zero crossing
period,T, is an important element of the joint met-ocean design
We now see that equation (3) describes the short time variability conditions. In the process of deriving the 100 year wave height,
and we have a sample set of values of XmP.
the 100 year storm will also have been specified in terms of
The “long time” statistics of stems Hmplw, the most probable extreme wave of that storm. T can be
estimated from alw and Hmpzm by means of correlations. Thus,
To specify the long time statistics of storms requires a equations (4) and (5) are two equations with two unknowns:
probability distribution for XmP. me procedure, usingGpl),is linear crest elevation, a, and in-line current, u.
the same as used for the Hmpvalues above. Figure 10 shows the
result of fitting a GPD to X~P data. Differentiating this curve In Figure 12 we have used equations (4) and (5) to plot a
gives p(XmP). the probability density of the X~P values greater against u for fixed, 100 year values of base shear and
th~ Xmp~in.The number of storms with X~P > XMP~inis k. overturning moment for a typical North Sea location. The line
ata= alw indicates the 100 year crest elevation. Which point
The distribution of the extreme load of a random of intersection on the plot provides an appropriate 100 year
storm and of an interval T environmental design condition? We shall see later on, when we
present some results, that the choice is not of great consequence.
P(XIXmp),The distribution of the extreme response of a storm of The intersection of FIW, and Ml~ seems suitable since these
given X~P is given by equation (3). Thus, the probability are both response variables. However, these two lines are so
distribution of the extreme response of any random storm (r.s.),
close to parallel that finding their intersection may be an ill
of unspecified XmP,with XmP> Xmpnincm now ~ c~culat~ ~
conditioned problem. Thus, the intersection of al~ with either
.-.
Flw or MIOOis generally recommended. The difference in

391
6 Response Based Design Conditions in the North Sea OTC 7683
currents and consequent loads between choosing Flm or ikfl~ is RP2A-LRFD, would allow the weight of the sub-structure and
insignificant for jackets in the North Sea. These practices are piles to decrease by 19 per cent. The failure rate of the structure
consistent with the recommendations of API [31]. would increase from 10-8 per year, but would still be less than
The design wind speed can be obtained by calculating the in- 10-5per year.
line wind required to account for the difference between either In the re-design of Tern, the load factor applied to
Mlm and MIooW or F1~ and Fi~w. Again, the choice of environmental load was 1.35, as recommended in API RP2A-
overturning moment or base shear generally has little LRFD. This value was developed by the API to ensure adequate
consequence for the result. reliability in the Gulf of Mexico and based on experience there
with working stress design. Does a load factor of 1.35 together
RESULTS
with the explicit and implicit factors on resistance lead to
In the table below, we give some example results for met-ocean appropriate levels of structural reliability in all locations?
design conditions derived using the present methods for three
If the most recent practices for calculating wave load [32,33]
locations: one in each of the northern (NNS), central (CNS) and
are used together with response based design conditions, then
southern (SNS) areas of the North Sea. These wave and current
the design load is close to the true 100 year load. The safety of
values should not be regarded as representative of the whole
the structure is determined by the ratio of its ultimate strength
areas. Obviously, the validity of any long term load statistics
to the true 100 year load and the sensitivity of load to return
and design conditions depends on the adequacy of the met-
period at its location. Using API LRFD with a load factor of
ocean time series. These nx.ults were derived from the NESS
1.35 and accounting for resistance factors and material strength
hindcast database which is presently being reasseswd and this
bias, the ultimate strength of an optimised structure should not
may result in some changes.
be less than about 1.8 times the 100 year load. Studies have
We also present plots of normalised load versus return period shown that this factor is likely to be closer to 2.2 [28]. We list
for these locations and for the Gulf of Mexico (GOM) in Figure below the annual probabilities of failure for structures with
13. The GOM line was obtained from results presented by expected ultimate strengths of 2.2 times the 100 year load and
Forristall et al [6]. Figure 13 shows that across the North Sea an uncertainty in strength described by a normal distribution
locations there are very large differences in the sensitivity of with a standard deviation of 10 percent of the ultimate strength.
load to return period. More exposed areas may display greater The values are calculated using the extreme load return period
sensitivity than the northern North Sea. relations of Figure 13.
Location Wave Height [m] Surface current [m/s] Location Probability of failure per year
1 I
I
100 year wave and current to produce 100 year base shear SNS 1X1O-7
NNs 28.5 0.2 NNS 8X10-6
CNS 21.4 0.2 GOM 3X1O-5
SNS 15.2 0.5
Though the reliability levels might well be regarded as
Wave and current producing 100 year base shear and moment
satisfactory, they are not uniform. Moreover, it is conceivable
NNs 28 0.3 that locations exist where the load return period curve is steeper
CNS 21 0.2 than for the Gutf of Mexico. If desired it would be possible to
SNS 14.3 0.7 make the probabilities of failure uniform by tuning the load
factors for each location.
DISCUSSION
Response based design conditions are significantly less severe CONCLUSIONS
than those obtained from conventional North Sea practices: Our main conclusions are:
wave heights are hardly affected; currents are significantly
reduced, by a factor of three to few, wind speeds are reducedby 1. A number of asymptotic properties of extreme values can be
a small amount. The consequence for environmental design applied in the analysis of oceanographic time series,
loading is a reduction of order one quarter. This is in addition to 2. The probability distribution of the extreme wave height of a
other decreases in design loads, relative to values used in the storm and the extreme structural response to a storm are of
early 1980’s. The consequences for weight and cost saving and standard forms and are characterised by their most probable
structural reliability are discussed elsewhere [28]. For Tern, a extreme values.
launched steel jacket installed in the northern North Sea in
1988, the total effect of a revised extreme storm loading practice 3. A consistent, storm based, method of analysis of the statistics
is to reduce the design environmental loads by a factor of two. of extreme loads and structural responses has been formulated,
Utilisation of this in are-design of the structure, based on API accounting for uncertainty of the extreme within a storm and
the uncertainty in storm severity.

392
 OTC 7683 P.S. Tromans and L. Vanderschuren 7
4. The method can be used to calculate response based (joint) 11. Battjes, J.A. “Encounter probability of extreme structural
met-ocean design conditions. response values based on multi-parameter descriptions of the
5. The method can be used to develop a location dependent physical environment”, BOSS Conference Proceedings, 1979.
approach to load factors. 12. Coles, S.G. and Tawn, J.A. “Statistical methods for multi-
6. Response based design conditions offer potentially large cost variate extremes: an application to structural design,” Appl.
savings and an improved approach to structural safety. S[atist., v. 43, no. 1, Royal Statistical Society, 1993.
13. Peters, D.J., Shaw, C.J., Grant, C.K., Heideman, J.C. and
ACKNOWLEDGEMENTS
Szabo, D. “Modelling the North Sea through the Northern
The authorsthank C.J. Shaw of SLPM and M. Efthymiou of European Storm Study; OTC 7130, 25th OTC, Houston, 1993.
Shell U.K. for their support, and G.Z. Fornstall of Shell Dev.
14. Leadbetter, M.R., Lindgren, G. and Rootzen, H. Extremes
Co., J.C. Heideman of EPRCO., and J.W. van de Graaf of Shell
and related properties of random sequences and processes,
Research for their advice and discussions. The development of
Springer-Verlag, New York, 1983.
these methods was assisted by M.R. Kraneveld, B. Heijermans
and P.P. de Wolf, students of T.U. Delft, and their professom 15. Bergman, L.E., “probabilities for highest waves in a
J.A. Battjes and P. Groeneboom. hurricane,” ASCE J. Wways, Harbours and Coastal Engg.,
v.99(WW2), pp. 185-207, 1973. See also Proc. 12th Conf. on
REFERENCES
Coastal Engg. Washington (1970) VOI.1,pp. 53-64.
1. Hogben, N. “Long term wave statistics,” in The sea v.9, pt. A
16. Ronold, K.O. and Haver, S., “Foundation safety of gravity
Ocean engineering science, ed. by Ix Mehaute, B. and Hanes, based systems during severe storms,” OMAE Conf., 1991.
D., Wiley, New York, 1990, pp. 293-333.
17. Sobey, R.J., Chandler, B.D. and Harper, B.A., “Extreme
2. Tucker, M.J. Waves in ocean engineering, Ellis Horwood,
waves and wave counts in a hurricane,” Proc. Twenty-Second
New York, 1991.
Coastal Engineering Conf., ASCE, Delft, 1990.
3. Haring, R.E. and Heideman, J.C., “GuIf of Mexico rare wave
18. Fornstall, G.Z.: “On the statistical distribution of wave
return periods,” J. Per. Tech., 1980, pp. 35-47.
heights in a storm,” J. Geophys. Res. (1978) 83,2353-2358.
4. Jahns, H.O. and Wheeler, J.D., “Long term wave
19. Krogstad, H.E., “Height and period distributions of extreme
probabilities based on hindcasting of severe storms”, J. Pet.
waves,” Applied Ocean Res. v 7, pp. 158-165.
Tech. pp. 473486, April 1973.
20. Haring, R.E., Osborne, A.R. and Spencer, L.P., “Extreme
5. Ward, E.G., Bergman, L.E. and Cardone, V.J., “Statistics of
wave parameters based on continental shelf storm records,”
hurricane waves in the Gulf of Mexico,” J. Pet, Tech., 1979, pp.
Proc. of 15th Coastal Eng. Conf., ASCE, pp. 151-170, 1976.
632-642.
21. Pickands III, J., “Statistical inference using extreme order
6. Forristall, G.Z., Larrabee, R.D. and Mercier, R.S.,
statistics,” Ann. Statist. 3, pp. 119-131, 1975.
“Combined oceanographic criteria for deepwater structures in
the Gulf of Mexico”, OTC 6541, 23rd OTC, Houston, 1991. 22. Hall, P., The bootstrap and Edgeworth expansion, Spnnger-
Verlag, New York, 1992.
7. Gordon, R.L., DaM, F.-E., and Peters, D.J.H. “NOCDAP
results: On the relationship of extreme waves and currents for 23. Heideman, J.C. “Parametric response model for wave/
design of offshore structures”, Ocean Engg., 12,4, pp.293-308, current joint probability”, submitted to American Petroleum
1985. Institute Technical Advisory Committee TAC 88-20, Jan,,
1980.
8. Heideman, J.C., Hagen, O., Cooper, C. and Dahl, F.-E. “Joint
probability of extreme waves and currents on the Norwegian 24. Tromans, P.S., Hagemeijer, P.M. and Wassink, H.R. “The
Shelf”, ASCE J. of Wway, Coastal and Ocean Engg., 115, 5~pp. statistics of the extreme response of offshore structures”, Ocean
534-546,1989. Engg., 19,2, pp. 161-181, 1992.
9. Prior-Jones, R.L., and Beiboer, F.L. “Use of joint probability 25. van de Graaf, J.W. and Tromans, P.S. “Statisticti
in deriving environmental design criteria”, S.U.T. Conf. on verification of predicted loading and ultimate strength against
Environmental Forces on Offshore Structures, London 1990; in observed storm damage for an offshore structure”, OTC 6573,
Advances in Underwater Technology, Ocean Science and 23rd OTC, Houston, 1991.
Offshore Engineering, Kluwer, Dordrecht.
26. Tromans, P.S. and van de Graaf, J.W. “A substantiated risk
10. Marshall, P.W. and Bea, R.G. “Failure modes of offshore assessment of a jacket structure”, OTC 7075, 24th OTC,
platforms”, BOSS Conference Proceedings, 1976. Houston, 1992.

393
8 Response Based Design Conditions in the North Sea
OTC 7683
27. van de Graaf, J.W., Tromans, P.S., Vanderschuren, L, and
Jukui, B.H. “Failure probability of a jack-up in the central North
Sea”, 4th Int. Jack-up Conf,, City Univ., London. Sept., 1993.
28. van de Graaf, J.W., Tromans, P.S. and Efthymiou, M. “The [m]
reliability of offshore structures and its dependence on design
code and environment”, OTC 7382, 26th OTC, Houston, 1994.
29. Si Boon-Ing, M., Vanderschuren, L., van de Graaf, J.W.
and Tromans, P.S., “Frdlure probability of a southern North Sea
platform under environmental loading”, Int. Offshore and Polm
Engg. Conf., ISOPE-93, Singapore, June, 1993.
30. Tromans, P.S., Anaturk, A.R. and Hagemeijer, P.M. “A new
model for the kinematics of large ocean waves - application as a
design wave”, Int. Off. and Polar Engg. Conf., Edinburgh,
1991.
31. APl recommended practice RP2A-I!.RFD, first edition, July
1993, American Petroleum Institute, Washington, DC, USA.
32. Tromans, P.S., Efthymiou, M., van de Graaf, J.W.,
Vanderschuren, L. and Taylor, P.H. “Extreme storm loading on
fued offshore platforms”, BOSS Conf., Imperial college,
London, July, 1992. -15 -12 -9 -6 -3 0 3 6 9 12 15
33. Petrauskas, C., Heideman, J.C. and Berek, E.P. “Extreme Time [hours]
wave force calculation procedure for the 20th edition of API
RP2A”, OTC 7153, 25th OTC, Houston, 1993.
Fig. 1- A storm history.

s, S2

“L4 “ -Ezz
E ‘p -. 1
Hindcast + J; u)
-x
data base
.~ 2 A a ~1 -— ___
fiz
Time I
Identify storms x I’rpl x
X*

——
—__
___
_
Calculate P(x) and XmPfor each storm

%
Lc-
-? 1

x ~1 -— __

1
+

X/xw

Determine short time variability

\
+

Calculate P(x)
of VT random storms
Calculate P(x)
ofany random storm
1- -i?
~
a
Xm
Determine long time variability

Fig. 2- Evaluation of extreme response statistics.

OTC 7683 P.S. Tromans and L. Vanderschuren 9

l-P(Hmp)
1.000

---- --- 7

t- t

0.100

----
--
-.---
---- ----
●✼
---- ---- ----
A ----- )

0.010 ✎✍✎
✎✍✍✍✍
✍✍✍✍
✍ ✍
✍✍✍
✍ ✍✍✍
✍✍✍✍
✍✍✍
✍ ✍✍✍
✍
✍✍✍✍
✍✍✍
--------
----
✍✍✍✍
✍✍✍✍
---- ---------------~
✍✍✍✍✍✍✍✍✍✍✍✍
1----- ----
t--------
I ----
--------------
----- ---
--.x
l---------,--’-------
----- ---- ----- ----- ----- -
t t
I I I
0.001
10 15 20 25
Hmp [m]

Fig. 3- Map of the Northern North Sea with directional sectors. Fig. 5- The long time scale variability: probability distribution
of extreme storms characterised by Hmp.

1.0 1.0
P(HIHmp) P(Hlr.s.)
0.9
c
0.8 ,J!-- 0.8
● ●“
,
., 0.7
,,
,
0.6 ‘0.6
— Mean
0.5
. .. . . .. 95°A limits

0.4 0.4
H Model

0.3

0.2 0.2
/
#.
+ 0.1

0.0 0.0 ●

0.8 0.9 1.0 1.1 1.2 1.3 1.4 10.0 15.0 20.0 25.0 30.0
H/Hmp H [m]

Fig. 4 - Comparison of the probability distributions of Fig. 6- The probability distribution of the extreme wave height
normalised extreme wave height with the theoretical model, of a random storm.
equation (l).

395
 10 Response Based Design Conditions in the North Sea OTC 7683

35 1.0
Wave height [m]
P(FIFmp)
.C
32 0.8

29 0.6
.
. . . .-”” — Mean
-.. ..-
-.. ---
26 ‘------ 95Y01imits
,.- ..”. 0.4
E Model

23 0.2
,,
,
. ,
20 m I 0.0
10 100 1000 0.60.70 .80.91.01.1 1.21.31 .41.51.61.71.8
Return period [years] FIFmp

Fig. 7- The dependence of extreme individual wave height (full Fig. 9 - Comparison of the probability distributions of
line) and storm seventy (broken line), characterised by HmD, on normalised extreme base shear with the theoretical model,
. .
return period. equation (3).

Generic load model values [kNJ l-P(Fmp)

1000 I I 1 I 1.000 .-..---: . ::::: ----
---- ----
---- ----
---- --
----
---- . ----. --- - ----
----
----
----
.-.
-----
. -
---- --- --- ---- ---- ----
---- --- --- ---- ---- .--- -
800 ---- ---- --- ---- ---- ---- -
---- ---- --. ---- ---- ---- -

600
0.100
----
----
----

---- ----
------------

----
m ----
----
----
---- ::::
---.-
--
----
----
----
---
.--::
-----

I----
----
---- - ---- ----
- ---- ---- -

---- ---- ----

400
t
0.010 -------------
---- ---- -.:::

HI
----
---- ----
---- ---- ---- ----
---- ---- ---- ---- ---
200 ---- ---- ----
---- ---- ----
---- ---- .-. _ ---- ---- . ---

0 0.001
0 200 400 600 800 1000 0 100 200 300 400 500 600
Numerical Values [kN] Fmp [kN]

Fig. 8- Comparison of generic load model for base shear with

Fig. 10- The long time scale variability probability distribution
numerical simulation for 160 metre water depth.
of extreme storms characterised by Fmp.

396
 P.S. Tromans and L. Vanderschuren
11

2.0
Normalized load 2.5
Normalised load

GOM
,’
,/ ,
.
2.0
,. ,. ‘“ NNS,
,
#, .’
.“
1.5
,’
,“ “’CNS
,.

●
.’ #

1.5 ,
,.” “”+ -.
#.-
..0
,.

1.0
1.04
100 1000 10000 1.0E+02 1.0E+03 1.0E+04 1.0E+05
Return period [years]
Return period [years]
Fig. 11- ‘l%edependence of extreme load on return period.
Fig. 13 - Normalised load versus return period for different
areas.
16.0 .

.\
Crest elevation [m]

15.0

-.
,— —-- .
14.0

3!
‘.
‘..
‘.
‘.
‘.
‘..
●.
13.0

.
12.0<
0.00 0.10 0.20 0.30 0.40 0.50
In-1ine current speed [m/s]

Fig. 12- Evaluation of the response based design condition.

397
I