Professional Documents
Culture Documents
2007 Forristall Wave Crest Heights and Deck Damage in Hurricanes Ivan, Katrina, and Rita
2007 Forristall Wave Crest Heights and Deck Damage in Hurricanes Ivan, Katrina, and Rita
Wave Crest Heights and Deck Damage in Hurricanes Ivan, Katrina, and Rita
George Z. Forristall, Forristall Ocean Engineering
based on single point measurements they under-predict the surge raises this estimate to 16.66 m. This estimate is slightly
maximum crest affecting a deck. higher than the 16.45 m elevation of the bottom of the deck
In the remainder of this paper, we consider these girders, but unlikely to have been high enough to cause the
mechanisms in detail. Then predictions of second order crest observed damage.
heights over deck areas are compared with the elevation of
observed damage at Petronius and other structures. Effect of Unusual Wave Spectra
Crossing seas have been suggested as a possible cause of freak
Second Order Crest Heights waves. By freak waves, we mean individual wave or crest
Crest heights in steep waves are higher than those predicted by heights that cannot be reasonably explained by standard
Gaussian theory because of non-linearities. The distribution statistical theories. For example, Donelan and Magnusson [6]
produced from simulations of second order waves by Forristall argued that mixed or crossing seas could increase the
[5] accounts for most of the non-linearity. It is a Weibull probability of high wave crests. They show a hindcast
distribution of the form indicating broad spreading that coincided with a shipwreck,
η β but no measurements were available to prove that freak wave
P(η 2 ) = exp − (1)
crests occurred.
α H S The Oceanweather fields files give two measures of
angular spreading from the directional wave spectra. The
where Angular Spreading Function [7] is the mean value over the
α = 1/ 8 + 0.2568S1 + 0.0800Ur 552 bins of cos(θ-ψ), where ψ is the vector mean direction.
(2) The sum is weighted by the variance component in each bin.
β = 2 − 1.7912S1 − 0.5302Ur + 0.2824Ur 2 If the angular spectrum is uniformly distributed over 360º, this
statistic is zero. If the distribution is uniform over 180º, this
The mean steepness and Ursell number are given by statistic is 2/π. If all variance is concentrated at the VMD, it is
1.0.
2π H s The In-Line Variance Ratio [8] is computed as
S1 =
g T12 (3)
Hs IVR =
∑∑ cos (θ −ψ )dS
2
(6)
Ur =
k12 d 3 ∑∑ dS
The maximum crest height during a storm can be estimated by If spectral variance is uniformly distributed over the entire
integrating this short term crest height distributions over the compass, or over a semicircle, IVR = 0.5. If the spectral
storm’s sea state history. If the probability that the wave or variance is confined to one angular band, or to two bands that
crest height exceeds η is given by P(η), then the probability are 180º apart, IVR = 1.0. According to Haring and
that the height will not exceed η in N waves is given by Heideman, cos2 spreading corresponds to IVR = 0.75.
Fairly broad wave spreading is typical during hurricanes.
N (4) Whether such spreading is enough to be a possible cause of
P(ηmax < η ) = 1− P(η ) unusual or freak waves can be studied by examining
measurements of individual hurricane waves. Extreme waves
For a sequence of records during a storm, the probability of were measured at the Marlin TLP and the Medusa truss spar
non-exceedance becomes during Hurricane Ivan. These measurements were described
by Cooper et al. [9]. At Marlin, the significant wave height
k Ni
(5) came close to 16 m. At Medusa, it was close to 12 m.
P(ηmax < η ) = ∏ 1− Pi (η ) Directional spectra and the two measures of wave spreading in
i =1
Ivan are available from the Oceanweather hindcast. The
significant wave height and directional spreading at Marlin are
We applied equation (5) to the Oceanweather hindcasts of shown in Figure 2. At both Medusa and Marlin (not shown),
Hurricanes Ivan, Katrina and Rita. The mean wave period T1 the spreading measures were close to 0.75 at the peak of the
was calculated from the integral and first moment of the wave
storm. At both, the spreading increased very rapidly after the
spectrum. The water depth was taken from the Oceanweather close passage of the center of the storm by these sites.
grid files. The crest height statistics at Marlin are shown in Figure 3.
The Oceanweather hindcasts include the storm surge
The crest height populations are well represented by the
calculated using a two dimensional hydrodynamic model. We second order distribution in equation (1). There are no
calculated the total surface elevation by adding the storm significant outliers to the fitted crest distributions so there is
surge to the second order crest heights in each record of the no evidence of freak waves in the measurement sets.
hindcast model. We then applied equation (5) to find the Figure 4 shows the angular spreading function and in-line
expected maximum of the total surface elevation at each grid variance ratio at Petronius from the Oceanweather hindcasts.
point in each storm. These measures of directional spreading are very similar to
The expected value of the maximum second order crest those at Marlin and Medusa. The spreading measures were
height in Ivan at Petronius is 16.28 m. Including the storm close to 0.75 through the peak of the storm and the spreading
OTC 18620 3
increased after the passage of the storm center. The crest over a deck is higher than that the wave crest at any one point.
height distributions showed no unusual features at Marlin and But the area inundated by that crest is not necessarily a large
Medusa for similar directional spectra, so we do not expect portion of the deck.
unusual crests as a result of crossing seas at Petronius. Piterbarg [13] found asymptotic distributions for Gaussian
Narrow spectra have also been suggested as a cause of processes over large multi-dimensional spaces. Krogstad et al.
freak waves. Benjamin and Feir [10] found that high regular [14] applied Piterbarg’s theorem to the estimation of
waves suffered a modulational instability as they progressed maximum wave crests. Consider a Gaussian random process
down a long wave tank. A similar instability exists for over an n-dimensional space with standard deviation σ.
random waves when the spectral bandwidth is sufficiently According to Krogstad et al., the probability that the
narrow. The envelope of the wave amplitude oscillates. As a maximum of a multi-dimensional Gaussian process is less
result, the highest waves are higher than they would be if there than sσ is:
were no instability. Osborne et al. [11] demonstrated this 1
effect both theoretically and experimentally. Pr(η < sσ ) = exp −Qn ( s) exp − ( s 2 − hN2 ) (9)
2
Janssen [12] has proposed the Benjamin-Feir Index (BFI)
as a quantitative measure of the power of the modulational
where
instability. The waves must be both high and narrow-banded
in frequency space. The latest version of the index from ( n −1) / 2
(−1) m (n − 1)!
Janssen’s 2005 paper is Qn ( s ) = h1N− n ∑
m =0 m !2m(n − 1 − 2m)!
s n − 2 m −1 (10)
Krogstad et al. [14] used Piterbarg’s equations to estimate Equation (20) is shown as the red line in Figure 7. For small
the maximum crest height over a 100 km by 100 km area. areas, the equivalent number of waves is thus proportional to
With a mean zero crossing period of 10 seconds, the length of the side rather than to the area. We speculate
E (ηmax ) = 1.32 H s (18) that this is because the maximum value of the wave surface is
likely to be on the edge of the area when the edge is shorter
This very large expected value of the linear crest height may than one wavelength.
help explain the very large crests that have been reported in The results are robust. Equations (19) and (20) also match
studies of images made by orbiting synthetic aperture radars. simulations with peak wave periods of 8 and 12 seconds.
We are more interested in time series over smaller areas on Changing the horizontal grid size from 5 m to 2.5 m makes
the order of 20 – 100 m on a side. It is reasonably easy to test less than 1% difference in the crest heights. If the deck area is
the theoretical predictions of linear crest heights using placed at a 45° angle to the direction of wave propagation, the
simulated time series. Forristall [15] made simulations over a simulated crest heights are the same as for when the waves
128 by 128 point grid with a horizontal spacing of 5 m. The propagate parallel to the deck.
time series were 1024 seconds in length with a time step of Non-linear interactions make crests in steep waves higher
0.25 sec. The simulations were repeated 100 times. than linear theory predicts. Equation (1) gives a good match
Maximum values of crest height were found over the whole to the distribution of second order crest heights at a point.
area and over smaller sub-regions of the area. This technique Second order simulations over an area require substantial
yielded many samples of the maximum crest heights. computer time, but they are not necessary. The effects of non-
Averaging the results gave stable estimates of the expected linearity and deck area are multiplicative. The linear crest
value of the crest height over different areas. heights at each point over the area of the deck are amplified by
Figure 7 shows the results of simulations of a Jonswap the factor between the distribution in equation (1) and the
spectrum with a peak period of 10.0 sec. The mean period Rayleigh distribution appropriate for linear theory.
was thus 8.35 sec and the associated wave length was 109 m. For reasonably large numbers of waves N, the maximum
The waves were directionally spread using the spreading second order crest height at a point is given by
functions for fetch limited waves given by Ewans [16]. The
blue diamonds show the empirically determined expected E (ηmax ) = (b + 0.5772 / a)σ (21)
value of the maximum crest over a square with the side length
given on the abscissa. The crest heights are normalized by the where
standard deviation σ of the time series. a = ( β / 4α )(log N )1−1/ β (22)
The expected crest height maxima over an area are
b = 4α (log N )1/ β
substantially higher than those at one point. That is true even
for relatively small areas. The expected maximum at one
point for 1024 seconds of 8.35 sec waves is 3.29σ. For a and α and β are the coefficients given in equation (2). For the
square 60 m on a side, the expected maximum is 3.944σ, or example of a Jonswap spectrum with a peak period of 10 sec
20% higher. If Hs = 8 m, the maximum crest somewhere and Hs = 8 m in infinitely deep water, we get α = 0.3725 and β
under a square 60 m deck would be 7.88 m compared to 6.58 = 1.8684. The expected value of the highest crest in 1024
m for the linear crest height at one wave staff. seconds is then 3.68σ. The increase of the second order crest
For the directional spreading used in our simulations, λ2 ≈ over the first order crest is 3.68/3.29 = 1.118. Applying that
3λ1. Fitting equations (11), (14) and (17) to the simulations factor to the expected value of a linear crest over a 60 m
for the larger areas gave an empirical estimate of 1.25 for the square deck gives crest height of 4.41σ = 8.82 m. Table 1
correlation correction factor. The expected number of crests summarizes the example calculations.
would then be given by
Calculation Method Maximum Crest Height
Linear at point 6.58
N = 2π AT /(1.25λ1λ2Tz ) (19) Second order at point 7.36
Linear over area 7.88
Second order over area 8.82
The effect of using Equation 19 to define the number of wave
crests is shown is shown by the magenta curve in Figure 7. It
falls well below the simulations for the smaller deck areas. Table 1. Expected values of crest heights in 1024
Piterbarg’s theorem is asymptotic in the sense that it applies seconds for Hs = 8 m, Tp = 10 sec.
for large areas. It is not surprising that it fails for side lengths Figure 8 shows the result of calculations using equation
smaller than one wavelength. For such small areas, it is likely (20) to calculate N and equation (21) to calculate the expected
that a local maximum of the surface will not appear in the area maximum crest height. The calculations were made for a
at any given time. range of deck areas. For each deck area, a curve shows the
Empirically, Forristall [15] found that for small areas a maximum normalized crest height expected in the number of
good fit to the simulations can be found by taking waves shown on the abscissa of the graph. The increase in
expected maximum crest height is large even for small deck
L T (20) areas. Then as the area increases, the crest heights increase
N =2
λ1 Tz more slowly.
OTC 18620 5
Consideration of the deck area makes it much more likely Neglecting changes in spectral shape, that one table gives a
than previously recognized for a wave crest to strike some part good guide for all conditions. For example, if the peak wave
of the deck. This does not necessarily mean that the structural period was 12 seconds, 1.54 = 5.1, so the area inundated at
reliability is less than previously calculated. If only a small 0.90 of the crest height would be 272 m2.
part of the deck experiences green water then the additional To a good degree of approximation, the area inundated is
force on the structure will not be too great. given by
The shape of the highest crests can again be found by A = 0.62Tp4 (1 − r )1.7 (23)
simulation. Forristall [15] made 200 simulations of a Jonswap
spectrum with at peak period of 8 sec. The simulations were where r is the fraction of the maximum crest height. Equation
1024 sec long. The horizontal grid spacing was 2.5 m to give (23) is for the particular case of Ewans’ [16] directional
good resolution around the wave peak. The wave from each spreading function. The inundation areas for other spreading
simulation at the time of the maximum crest was normalized functions can be found by computing the autocorrelation of
by the maximum crest height and centered on the position of the directional spectrum.
the maximum crest. Then the normalized wave shapes were Equations (21) and (23) can be combined to estimate the
averaged. Figure 9 shows the resulting average wave shape as deck elevation for which a given area will be inundated.
a contour plot. Figure 11 shows the results for a deck 50 m square and a
It is also possible to derive the average shape of the Jonswap spectrum with a peak period of 12 sec. The deck
maximum wave from theory. The idea that the average shape elevations are normalized by Hs. The dashed curve in the
of the highest wave in a time series tends to the figure shows the elevation of the expected maximum crest on
autocorrelation function of the time series has been a single wave staff. We see that if the platform design was
independently discovered many times. Lindgren [17] and based on the crest elevation statistics of a single wave staff,
Boccotti [18] give detailed mathematical descriptions of the the expected area of inundation would be about 400 m2.
behaviour of a Gaussian process near a high local maximum.
Tromans et al. [19] introduced the idea to a wide audience of Platform Damage Cases
ocean engineers and called it NewWave. Fedele and Arena
[20] have now extended the concept to second order waves. Petronius in Hurricane Ivan
Essentially the same theory applies to two spatial dimensions. Table 3 summarizes the crest height calculations for the
The autocorrelation function of waves in two spatial Petronius tower shown in Figure 1.
dimensions is the average shape of the highest waves. This
idea was developed by Phillips et al. [21]. Phillips et al. [22] Bottom of Steel 16.45 m
then applied the theory to measurements from a scanning radar Second order crest height 16.28 m
altimeter. Second order crest plus surge 16.66 m
The autocorrelation function for a Jonswap spectrum with Maximum crest from equation (14) 18.69 m
2
a peak period of 8 seconds and Ewans’ [16] directional Area inundated from equation (23) 727 m
Maximum crest from simulations 18.93 m
spreading function is shown in Figure 10. As expected, the Area inundated from simulations 410 m
2
E (η max,2 nd ) = 1.125i16.28 + 0.38 = 18.69 (27) Maximum crest from simulations 17.12 m
2
Area inundated from simulations 100 m
offshore platforms. Some of the damage cannot be explained deck area in a thorough study of the data from these
with standard analysis tools. The purpose of this study was to hurricanes.
investigate this unexplained damage by applying new
methods. Acknowledgement
The standard practice for calculating the maximum crest This work was sponsored by the Climatology and Simulation
height in a storm is to integrate the short term distribution of of Eddies (CASE) JIP. We appreciate their support and
second order crest heights over the significant wave heights permission to publish the results.
hindcast for the storm. We performed that integration for each
grid point in the Oceanweather hindcasts. In many cases, the References
crest heights calculated by the standard methods were not high
enough to have caused the observed damage. 1. Cardone, V.J., A. T. Cox, F. Counillon, and D. Szabo (2005),
The hindcast waves were spread rather broadly. But the Hindcast of winds, waves and currents in northern Gulf of
waves at the damaged platforms were no more broadly spread Mexico in Hurricane Ivan (2004), Proc. Offshore Tech. Conf.,
than at locations where the crest height distributions OTC 17736.
2. Forristall, G.Z. (1978), On the statistical distribution of wave
conformed to second order theory. Freak waves produced by heights in a storm, J. Geophys. Res., 83, 2353-2358.
crossing seas did not do the unexplained damage. 3. Nielsen, F.G. (2003), Comparative study on airgap under
Instabilities caused by the combination of steep waves and floating platforms and run-up along platform columns, Marine
a narrow directional spectrum were not a cause of freak waves Structures, 16, 97-134.
that damaged platforms. The Benjamin-Feir Index (BFI) is a 4. Van Iperen, E.J., G.Z. Forristall, J.A. Battjes, and J.A. Pinkster
measure of that combination. BFI values at the sites of the (2004), Amplification of waves by a concrete gravity sub-
damaged platforms were similar to those at the locations structure: Linear diffraction analysis and estimating the extreme
where the crest height distributions conformed to second order wave height, Proc. 23rd Int. Conf. on Offshore Mech. and Arctic
Eng, OMAE 2004-51022, Vancouver.
theory.
5. Forristall, G.Z. (2000), Wave crest distributions: Observations
Predicting maximum crest heights over the area of the and second order theory, J. Phys. Oceanogr., 30, 1931-1943.
decks rather than at a single point explains almost all of the 6. Donelan, M.A., and A.K. Magnusson (2005), The role of
observed damage. The highest crest over the area of a meteorological focusing in generating rogue wave conditions,
platform deck is almost always higher than the highest crest at Proc. Aha Huliko’a Hawaiian Winter Workshop, University of
a single point. The statistical theory and simulations of this Hawaii at Manoa.
effect show that it is surprisingly and significantly large. For 7. Gumbel, E.J., J.A. Greenwood, and D. Durand (1953), The
the deck areas and directional spectra studied, the maximum circular normal distribution: theory and tables, J. Amer. Statist.
over the decks was 10 - 14% higher than the expected Ass., 48, 131-152.
8. Haring, R.E. and J.C. Heideman, 1978. Gulf of Mexico rare
maximum at a single point.
wave return periods. Proc. Offshore Tech. Conf., OTC 3230.
We recommend that an evaluation of the difference 9. Cooper, C., J. Stear, J. Heideman, M. Santala, G. Forristall, D.
between local and global crest heights is included in structural Driver, and P. Fourchy (2005), Implications of Hurricane Ivan
design. The crests that cause the local damage on decks affect on deepwater Gulf of Mexico metocean design criteria, Proc.
only a relatively small area because of dispersion and Offshore Tech. Conf., OTC 17740, Houston.
directional spreading. They can cause the observed local 10. Benjamin, T.B., and J.E. Feir (1967), The disintegration of
damage even though they do not engulf the entire deck or wavetrains in deep water, J. Fluid Mech., 27, 417-430.
cause global structural failure. Whether or not it is 11. Osborne, A.R., M. Onorato, and M. Serio (2005), Nonlinear
economically optimal to accept the possibility of local damage Fourier analysis of deep-water, random surface waves:
Theoretical formulation and experimental observations of rogue
will depend on the specifics of each platform design.
waves, Proc. Aha Huliko’a Hawaiian Winter Workshop,
Evaluating the difference between local and global crest University of Hawaii at Manoa.
heights makes informed choice possible. 12. Janssen, P.A.E.M. (2005), Nonlinear four-wave interaction and
We recommend detailed numerical and experimental freak waves, Proc. Aha Huliko’a Hawaiian Winter Workshop,
studies of wave diffraction and run-up on structures with large University of Hawaii at Manoa.
columns. Two of the structures with deck damage were 13. Piterbarg, V.I. (1996), Asymptotic methods in the theory of
tension leg platforms. They received minor damage in the Gaussian processes and fields, AMS 1 Transl. of Math.
decks despite the fact that their decks are higher than those of Monographs, 148, Providence, R.I.
fixed platforms. Diffraction and run-up may explain this 14. Krogstad, H.E., Liu, J., Socquet-Juglard, H, Dysthe, K.B., and
Trulsen, K. (2004), Spatial extreme value analysis of nonlinear
damage. The question is: how large an air gap is needed to
simulations of random surface waves, Proc. 23rd Int. Conf. on
avoid this kind of damage to floating structures such as Offshore Mechanics and Arctic Engineering, OMAE 51336.
tension leg platforms and spars? 15. Forristall, G.Z. (2006), Maximum wave heights over an area and
We recommend that future calculations of the probability the air gap problem, Proc. 25th Int. Conf. on Offshore Mechanics
of waves in decks should include predictions of the highest and Arctic Engineering, OMAE2006-92022.
crest over the area of the deck. The results of this study are 16. Ewans, K.C. (1998), Observations of the directional spectrum of
based mainly on observations from Hurricane Ivan. Excellent fetch-limited waves, J. Phys. Oceanogr., 28, 495–512.
damage documentation from Hurricanes Katrina and Rita is 17. Lindgren, G. (1970), Some properties of a normal process near a
becoming available. Information on deck heights, deck areas, local maximum, Ann. Math. Statist., 4, 1870-1883.
18. Boccotti, P. (1989), On mechanics of irregular gravity waves,
and orientation should be included in the data collection. We
Atti Acc. Naz. Lincei, Memorie, 19, 11-170.
recommend applying the calculations of crest heights over
8 OTC 18620
5.500
5.000
4.500
Cmax/sigma
4.000
3.500
Figure 4. Significant wave height and wave spreading functions
at Petronius during Hurricane Ivan.
3.000
10 100 1000
Size of deck (m)
1.2000
1.1000
Single point
20 x 20 m
30 x 30 m
1.0000
40 x 40 m
50 x 50 m
Crest/Hs
0.9000 60 x 60 m
70 x 70 m
80 x 80 m
0.8000 90 x 90 m
100 x 100 m
0.6000
10.00 100.00 1000.00
T/Tz
Figure 9. Average wave shape at the time of the maximum crest Figure 12. Probability of exceedance of the maximum crest
as derived from simulations. height plus surge over the Petronius deck area in Hurricane Ivan.
Deck 50 x 50 m, Tp = 12 sec
1.1000
1.0000
Single staff
A=0
Deck Elevation/Hs
A = 25
0.9000
A = 50
A = 100
A = 200
0.8000 A = 400
A = 800
0.7000
0.6000
0 1000 2000 3000 4000 5000 6000 7000 8000 Figure 14. Damage to generators on the Virgo cellar deck after
T(sec)
Hurricane Katrina.
Figure 11. Deck elevation for which the area (m2) given on the
legend will be inundated
OTC 18620 11