Analyses of Extreme Wave

Heights in the Gulf of Mexico for

Vijay Panchang
Texas A&M University,
Maritime Systems Engineering Department,
Offshore Engineering Applications
200 Seawolf Parkway, The 2004–2008 hurricane season in the Gulf of Mexico (GOM) saw several exceedances

Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
Galveston, TX 77553 of what was regarded, prior to that period, as the 100-year significant wave heights
e-mail: [email protected] (SWHs) that are used for the design of offshore oil and gas facilities. As a result, these
facilities sustained considerable damage and disrupted U.S. energy supplies. The wave
Chan Kwon Jeong climatology in the GOM is therefore studied in detail. A 51-year database of SWHs was
Dockwise USA, constructed by using a combination of wind and wave models, and both individual wave
16340 Park Ten Place, heights and statistical measures were validated, to the extent possible, using buoy data.
Suite 200, Analyses of the modeled data show that there is an increasing trend in the annual maxi-
Houston, TX 77084 mum SWHs in the eastern part of the GOM; the maximum trend is approximately 5.6 cm/
e-mail: [email protected] year, which is of the same magnitude as that reported for the U.S. west coast. The western
part; on the other hand, shows a decreasing trend. The maximum estimated 100-year
Zeki Demirbilek1 SWHs (denoted by SWH100) are 19.1 m, 22.6 m and 26.7 m for the Gumbel, Weibull, and
U.S. Army Engineer Research the GEV distributions, respectively. The estimates obtained here using the Weibull distri-
and Development Center, bution are comparable to those obtained independently by API (API—American Petro-
Coastal & Hydraulics Laboratory, leum Institute, 2007, “Interim Guidance on Hurricane Conditions in the Gulf of
3909 Halls Ferry Road, Mexico,” API Bulletin No. 2INT-MET). However, the use of objective criteria to identify
Vicksburg, MS 39180 the optimal distribution suggests that the GEV estimates are to be preferred if the engi-
e-mail: [email protected] neer wishes to emphasize the upper tail where extremes are likely to occur. The maximum
increase in the SWH100 due to the 2004–2008 season is of the order of 0.9 m to 2.7 m
(depending as the distribution). Information generated here is intended to supplement the
design recommendations provided by API (American Petroleum Institute, 2007, “Interim
Guidance on Hurricane Conditions in the Gulf of Mexico,” API Bulletin No. 2INT-MET).
[DOI: 10.1115/1.4023205]

1 Introduction measured SWHs are associated with a recent storm, the name has
been included. It is clear that, at seven of the buoy locations
Until recently, significant wave heights (SWHs) corresponding
(42001, 42,003, 42,007, 42,019, 42,035, 42,039, and 42,040) max-
to the 100-year return period used for the design of various oil
imum measurements for the 2004–2008 hurricane seasons are
and gas facilities in the Gulf of Mexico (GOM) were of the order
larger than previous maxima. While the differences are marginal
of 11 m (API [1]; Palao et al. [2]; Panchang et al. [3]). During
at some of the buoy locations, they are substantial at three loca-
the years 2004–2008, however, the GOM (shown in Fig. 1) expe-
tions (42039, 42,040, and 42,007): the previous largest SWH
rienced waves of unusual height relative to the “extreme” condi-
recorded at these locations were 9.3 m (in 1998), 10.8 m (in 1998),
tions. For instance, during Hurricane Ivan in 2004, SWH’s of
and 4.9 m (2002), respectively. Although two of these locations
the order of nearly 16 m were recorded by a National Data Buoy
have relatively short datasets, the differences are notable at the
Center (NDBC) buoy before it malfunctioned; and, SWHs as large
as 17.9 m (corresponding to “maximum” wave heights of approxi-
mately 27.9 m) were recorded by Wang et al. [4], who suggest
that even larger waves may have occurred. Comparably large
(and at some locations, larger) wave heights, well in excess of the
100-year return period estimate, were recorded again during
Hurricanes Katrina (Sept. 2005), Dennis (July 2005), Rita (Sept.
2005), and Ike (Sept. 2008). The extreme storm surge, wind, and
wave conditions, which have been documented and analyzed in
part by Hovis [5] (2005) and by Panchang and Li [6] (2006),
caused extensive damage to the numerous oil and gas facilities in
the Gulf of Mexico [7].
The National Data Buoy Center (NDBC) maintains eleven
wave buoys in the GOM (Fig. 2) that provide wave data for
time periods ranging between 7 and 33 years. Table 1 provides a
summary of these data and includes the highest and second high-
est SWHs recorded at the buoy locations. When the maximum

1
Corresponding author.
Contributed by the Ocean, Offshore, and Arctic Engineering Division of ASME
for publication in the JOURNAL OF OFFSHORE MECHANICS AND ARCTIC ENGINEERING.
Manuscript received May 14, 2012; final manuscript received September 9, 2012;
published online May 2, 2013. Assoc. Editor: Hideyuki Suzuki. Fig. 1 Gulf of Mexico, map and bathymetry (m)

Journal of Offshore Mechanics and Arctic Engineering AUGUST 2013, Vol. 135 / 031104-1
C 2013 by ASME
Copyright V
 The magnitude of the increases and their implications for engi-
neering call for a more comprehensive investigation of the issue,
since the API calculations do not cover the entire Gulf and, more
importantly, such extreme value estimates can be sensitive to both
the data and to the statistical methods used to generate them. The
importance of this was highlighted by the events associated with
Hurricane Ike (2008), when SWH’s that equaled or exceeded
API’s new SWH100 estimates were recorded at some locations.
(As noted earlier, the API [15] study does not include storms after
2006.) In an effort to augment the work described in API [15], we
revisit the problem by independently generating 51 years of

Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
Fig. 2 Approximate locations of NDBC buoys in the Gulf of
Mexico detailed numerical simulation data and using multiple statistical
methods. Additional estimates obtained here are intended to help
the engineer make informed decisions and to assess the uncer-
locations of buoys 42,007 and 42,040; at the latter buoy location, tainty associated with the results.
extremely large differences (relative to previous maxima) were The quality of data is fundamentally important for metocean
recorded during two events: Hurricanes Ivan and Katrina. studies affecting risk-based design of offshore structures.
The measurements, summarized above, would suggest the Although measured data are probably the best source, individual
possibility of an increasing trend in the SWHs in the GOM. measurement sites are widely distributed and the lengths of
Such trends in the wave climate are being increasingly reported. records are relatively short or discontinuous. As a result, measured
Wang et al. [8] used a variety of models and projected a data are usually insufficient to characterize waves over large
60–100 cm increase in the fall 20-year wave heights in the Nor- ocean areas. An alternative method of data development is to
wegian Sea between 1990 and 2080; for the northeast Atlantic, use numerical models to calculate the wave conditions based on
they project a 5–35 cm increase in the mean winter wave heights measured or modeled winds. This method of developing wave
in 2070–2099 relative to 1961–1990. Off the California coast, data for subsequent probabilistic or climatological analyses has
Graham [9] reported an increasing trend in the largest modeled been used; for example, by Pontes et al. [16] to develop a near-
SWHs of about 1.0–2.5 m for the 1949–1999 period. Examining shore atlas for wave energy calculations in Portugal, Millar et al.
buoy data, Komar and Allan [10] reported an increase of 1.7 cm/ [17] to examine the impacts of wave energy farms off the UK
year off the east coast; and Ruggiero et al. [11] reported an coast, Panchang et al. [18] for aquaculture applications, and
increase of about 7.1 cm/year in the largest SWHs since the Cieślikiewicz and Paplińska-Swerpel [19] to develop a 44-year
mid-seventies off the U.S. northwest coast. An increasing trend wave hindcast in the Baltic.
in the GOM, if one were to exist, could be of considerable Yet, synthetic data are potentially vulnerable to many
import, given the enormity of the engineering enterprise in these modeling-related errors (e.g., Rogers et al. [20]; Cardone et al.
waters. Trend estimates are needed for nonlinear approaches to [21], and probabilities can be affected by the choice of events
extremal analyses and have been used; for example, by Warner chosen for modeling. Berek et al. [14] have generated the data by
and Tissot [12] and Obeysekera et al. [13] in the context of hindcasting “select” storms for the period 1950–2006 using pro-
water levels off the GOM coast. Following the approach taken prietary modeling tools, and while some details of their error anal-
by Graham [9], this issue is examined for the GOM using sev- ysis are available (e.g., Forristall [22]) provides an assessment of
eral decades of (modeled) data. their models for four storms), the multifarious nature of the data
As a result of the occurrence of the extremely large waves, the generation process demands the use of alternative tools and mod-
American Petroleum Institute (API) initiated efforts to reexamine els. Here we have used alternative numerical models to create a
the specification of design conditions for offshore structures in the continuous 51-year (1958–2008) hindcast. Further, we have used
GOM. These efforts, summarized by Berek et al. [14], were based all publicly-available wind and wave data to validate the model
on a combination of synthetic (hindcast) wind and wave informa- results, to the maximum extent possible, and to make appropriate
tion obtained with proprietary models and the Weibull distribution adjustments to both the models and to the results prior to prepar-
(with the peak over threshold method) for hurricanes that occurred ing a final dataset for statistical analysis. In particular, we have
prior to 2007. In some regions of the GOM, these new estimates used a modified parametric wind field model (Jeong et al. [23])
indicate substantial increases in the 100-year (design) wave that includes improvements based upon a comparison of such
height and wind speed; relative to API’s previous estimates, the models against “HWIND” data (Powell et al. [24]) for a large
maximum increase is as much as 6.4 m at a given location in the number of storms covering nearly 30 years. Also, nearly 35 years
maximum wave height and 5 m/s in the wind speeds. According of wave buoy data were used to validate not only individual storm
to API [1], the maximum estimated 100–year SWH (denoted by simulations but also the SWH100 estimates.
SWH100) in the GOM was approximately 12 m, whereas the In addition to data reliability, the choice of the dataset itself is
revised estimate is 15.8 m (Berek et al. [14]). important for characterizing extreme events (e.g., Xu and Huang

Table 1 Wave data description in the Gulf of Mexico

Buoy Data duration (year) Depth (m) Maximum SWH (m) 2nd largest SWH (m)

42001 19762008 (33) 3246.0 11.63 (Sept. 2005)Rita 11.2 (Oct. 2002)
42002 19762008 (33) 3566.2 9.70 (Sept. 1988) 8.4 (Nov. 1980)
42003 19772008 (32) 3233.0 11.04 (Sept. 2004)Ivan 10.7 (Nov. 1985)
42007 19812008 (28) 14.0 9.09 (Sept. 2004)Ivan 5.64 (Aug. 2005)Katrina
42019 19902008 (19) 82.3 6.3 (Sept. 2008)Ike 5.92 (Sept. 2005)Rita
42020 19902008 (19) 88.1 8.20 (Aug. 1999) 6.79 (Oct. 1996)
42035 19932008 (16) 13.7 7.1 (Sept. 2008)Ike 4.7 (Sept. 2005)Rita
42036 19942008 (15) 54.5 8.60 (Oct. 1995) 6.9 (Jan. 1994)
42039 19952008 (14) 291.4 12.05 (Sept. 2004)Ivan 10.6 (July 2005)Dennis
42040 19952008 (14) 443.6 16.91 (Aug. 2005)Katrina 15.96 (Sept. 2004)Ivan
42041 19992005 (7) 1055.7 12.31 (Oct. 2002) 8.56 (Sept. 2004)Ivan

031104-2 / Vol. 135, AUGUST 2013 Transactions of the ASME

[25]). Also, the n-year return period estimates can be highly sensi- corresponding to a n-year event, a rule of thumb (Borgman [28])
tive to the chosen statistical model, and to the “threshold” in the states that the dataset should preferably be at least n/3 years long;
peak-over-threshold method that is commonly used for extreme this precludes the exclusive use of two (“NCEP” and “HWIND”)
estimation (Van Vledder et al. [26]). While Berek et al. [14] have of the three sources in Table 2 for the development of SWH100
chosen, a priori, the Weibull distribution, the question of whether estimates. And, while the “Reanalysis” dataset (Kalnaya et al.
this is the best distribution has not been addressed, nor has the [29]) is available for a longer time duration and possesses suffi-
sensitivity of their results to the threshold. cient temporal resolution (12 mins) and has hence been used for
In view of these considerations, a comprehensive study was other large ocean basin wave studies (e.g., Music and Nickovic
conducted for estimating the variability inherent in the SWH100 [30]; Cieślikiewicz and Paplińska-Swerpel [19]), its spatial resolu-
estimates resulting from the use of different sets of data and tion has been found to be too coarse for simulating the spatial
different statistical methods for the entire Gulf of Mexico. In par- details of hurricane waves (Jeong et al. [23]).

Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
ticular, we are concerned with the following questions that can An alternative is to use “parametric models” to reconstruct the
influence design choices: wind fields on the basis of a limited set of hurricane parameters. A
hurricane dataset (HURDAT) pertaining to a limited set of storm
(1) How do SWH100 values obtained by modeling compare
parameters developed by the NOAA from 1851 to present (Jeong
with those obtained from buoy data? What is the effect of
et al. [23]). However, the dataset is substantially detailed only for
modeling errors on the SWH100? What is the effect of data
the post-1950 period. This dataset provides information at 6-hour
length on the estimates? These issues are of concern since
intervals, information such as the location of the storm center,
some estimates in the past (e.g., Panchang and Li [6]) have
storm direction (hs), storm speed (Vs), maximum wind speed (Vm),
been obtained using buoy data which have the benefit of
and storm central pressure (Pc). Details are provided in Jeong
being free of modeling-related errors and can be used either
et al. [23], and because the parametric models were used in this
for design (and are used here for validating modeled esti-
work, a brief summary is provided below.
mates). Buoy data are available for varying time periods,
We considered three commonly-used parametric models, viz.
whereas model results were developed here for 51 years.
the Rankine Vortex (RV) model, the SLOSH model, and the
(2) What is the variability associated with the choice of other
Holland model (with asymmetries; see MacAfee and Pearson
distributions? Three commonly used extreme value distri-
[31] for details) to obtain the wind speed V(x,y) as a function of
bution functions (Gumbel, Weibull, and generalized
HURDAT parameters Vm and Pc and also of other model-
extreme value) were evaluated to select the optimal one for
calculated quantities such as radius to maximum wind (Rm) and
extremal estimation. Further, selection of the most appro-
sea level pressure at the last closed isobar (Pn). The models
priate model for different areas is usually a difficult and
were applied to 56 HURDAT snapshots covering a wide range
subjective problem. Recently Li et al. [27] provided robust
of hurricane development phases/levels. The results, documented
and quantitative means for identifying the most appropriate
in Jeong et al. [23], indicate that the models frequently produced
model based on the engineer’s preference, rather than limit-
considerable mismatch in velocity contours relative to the
ing oneself to any preselected model.
HWIND data. An example, using the RV model, is shown in
(3) Are there trends in the wave climate in the GOM? What is
Figs. 3(a)–3(c) for Hurricane Ike (1998). Wind speed errors are
the effect of the recent (2004–2008) extreme hurricane sea-
as large as 15 m/s; similar errors were encountered with the
son on the estimates of SWH100?
other two wind models. Jeong et al. [23] found that for Hurri-
The paper is organized as follows. Section 2 provides details cane Katrina, the RV model winds led to an underestimate of
about available wind and wave information in the GOM. It also the modeled wave heights by nearly 5 m at the peak of the
examines the performance of three commonly used parametric storm.
wind models when applied to recent storms and briefly describes Other researchers (e.g., Xie et al. [32], MacAfee and Pearson
adjustments to the Rankine vortex model to obtain more reliable [31]) have also noted discrepancies between parametric wind
hurricane windfields (Jeong et al. [23]). The wave modeling meth- models and data, and made adjustments to suit individual cases or
odology used to obtain wave characteristics at an appropriately regions. For example, Xie et al. [32] made adjustments to the
fine resolution, modeling errors, adjustments, and validation of Holland model based on an analysis of four mid-Atlantic hurri-
the calculated wave heights are described in Sec. 3. A portion of canes. For the GOM Jeong et al. [23] made adjustments to the RV
the results, relating to trends, is presented in Sec. 4. Section 5 model, using observed differences between modeled and meas-
summarizes the statistical methods used here for extreme wave ured data to develop empirical correction factors to modify the
height estimation. Validation of the statistical estimates using RV model.
model data and buoy data is presented in Sec. 6; this involves lim- While the reader is referred to Jeong et al. [23] for details of the
iting the model dataset to the length of the buoy data. In Sec. 7, modifications as well as validation, by way of example, we dem-
the full 51 years of model data are used to obtain SWH100, using onstrate in Fig. 3(c) the improvement obtained with the modified
the three distributions and the behavior of each is examined. This RV model as regards wind speed contours. The wave fields result-
helps identify the most appropriate SWH100 estimate at each grid ing from the two wind fields are shown in Figs. 3(d) and 3(e). For
point. Section 8 summarizes the results of this study. both wind and wave fields, although the maximum values are
largely the same, the spatial patterns are different, leading to a
maximum difference of about 3.4 m in SWH estimates at some
2 Modeling of Wind Fields locations. Since such large differences have the potential to affect
Three sources of wind fields, described in Table 2, are available extreme wave height statistics, the modified RV wind model was
to force a wave model. For the generation of estimates used for modeling the wave-fields.

Table 2 Windfield datasets

NCEP HWIND Reanalysis wind

Source Numerical model Measured data Numerical & available data

Resolution 0.25  0.25 6 km  6 km 2.5  2.5
Availability 1999present 1994present 1948present
References http://polar.ncep.noaa.gov/waves/index2.shtml? Powell et al. [24] Kalnaya et al. [29]

Journal of Offshore Mechanics and Arctic Engineering AUGUST 2013, Vol. 135 / 031104-3
 Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
Fig. 3 Wind speeds and SWH’s during Hurricane Ike (at 1200 UTC 9 Sept. 2008): (a) RV-model wind, (b) H*Wind, (c) modified
RV-model wind, (d) SWH (m) using RV-model wind, (e) SWH (m) using modified RV-model wind, (f) the difference SWH (m)
between (d) and (e)

3 Modeling of Wave-Fields As noted earlier, the spatial resolution of the reanalysis wind-
field was found to impede the modeling of the spatial details of
Wave data for the GOM are in fact developed by Texas Coastal
hurricane winds. The modified RV model winds were therefore
Management Program (NOAA), using the model WAVE-
used to replace the reanalysis winds in the hurricane area covering
WATCH, as part of a large-scale simulation for the entire western
a 4  4 region at a resolution of 0.05  0.05 . However, the
North Atlantic, at a grid resolution of about (0.25 ). Since this
benefits of incorporating “background” wind fields (viz., the wind
is insufficient for representing the intricacies of hurricane condi-
fields outside the hurricane and also the wind fields that existed
tions, NOAA developed the “North Atlantic Hurricane” model
before the onset of the hurricane) have been indicated by Liu et al.
but results are available only from 2000. Hindcasting with other
[39]. Therefore we constructed a composite wind field (Fig. 4) by
methods is thus needed to provide data for extreme wave estima-
tion for long return periods.
Toward that end, we used the third-generation wave-prediction
model SWAN (Booij et al. [33]; Ris et al. [34]), which is based on
the spectral action balance equation and includes the total rate of
change of wave action, frequency shift and refraction induced by
depth and currents the effects of generation, dissipation (due to
breaking, bottom-friction, and white-capping), and nonlinear
wave-wave interactions. Reflections and diffraction are incorpo-
rated in an approximate manner (Booij et al. [35]). Spectral or
parametric wave input conditions may be specified along the
boundaries. The model has been widely used and validated by
several investigators (e.g., Booij et al. [33]; Zubier et al. [36];
Chen et al. [37]; Singhal et al. [38]).
A grid covering the entire Gulf of Mexico from 18 N to 32 N
and from 80 W to 100 W at a spatial resolution of 0.2  0.2
(101  76 grid points) was developed. The bathymetry, obtained
from the National Geodetic Data Center, had a resolution of
2 mins (601  451 grid points). The model was run with 576 dis-
crete spectral components (Dh ¼ 15 and Df varying between 0.04
to 0.4 Hz with a logarithmic increment). Fig. 4 Integration of windfields

031104-4 / Vol. 135, AUGUST 2013 Transactions of the ASME

 Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
Fig. 5 SWH contour (a) and SWH comparison at buoy 42,001(b)

merging the parametric model winds with the reanalysis winds simulation. A more likely source of errors may be due to inaccura-
(the latter are interpolated onto a 0.05 grid). cies in the wind-fields in the near-shore areas arising from the
presence of islands (near buoy 42,007). Other possible reasons for
modeling errors are inaccurate representations of the bathymetry
3.1 Basic Results: Individual Storms. Even with the modi- (including subgrid-scale topographic features) and model limita-
fied RV model winds, some difficulties remain while modeling tions (e.g., Rogers et al. [20]). There can also be data problems.
the hurricanes. The HURDAT dataset provides the location of the Bender et al. [42] reported possible errors in SWH measurements
storm center at 6-hour intervals. Here we have connected these for buoy 42,007 by approximately 11% during Hurricane Katrina
locations (denoted by (a) and (b) in Fig. 5) by a straight line, possibly due to swell in shallow water, heeling of the buoy caused
which is assumed to be the storm track, and simulations are made by winds and currents, and a failure to tilt-correct the accelerome-
for intermediate times by placing the storm center on this line by ter data. Unfortunately, no generalized means to assess the NDBC
interpolation. This leads to problems in some instances. For exam- buoy data and to correct them are readily available.
ple, Fig. 5 shows that, during Hurricane Ike, the simulations did Additional comparisons of model SWHs and historical data are
not successfully capture the two peaks seen in the SWH data at shown in Fig. 8. The data were collected from a fixed platform
the location of buoy 42,001, the occurrence of which indicates near Louisiana during Hurricane Camille (1969), which passed
that the buoy is close to the actual storm track (possibly with the within 23 km of the platform. These data are older than all the
eye of the storm passing in the vicinity). However, the assumption NDBC data. The measured maximum SWH is 13.45 m and mod-
of line AB as the assumed storm track yields a simulation contain- eled SWH is 14.65 m [43]. Note that data are available only for a
ing only one peak, and hence substantial errors in SWHs, at the part of the hurricane. The comparisons are reasonable, especially
buoy location. Thus, the actual storm track could be off by as
much 30 km (based on this example).
To address the problem of storm track uncertainty, Heideman
and Mitchell [40] resorted to a technique called “grid point
pooling,” which involves collecting estimates from several points
in a preselected area (see also Jonathan and Ewans [41]), and then
using them for statistical analysis. However, as they acknowledge,
this violates the requirement of statistical independence because
each storm is included multiple times. They recommended
some general guidelines to ameliorate this issue, including a selec-
tion of spatially homogeneous regions, which is subjective and
difficult.
Since the SWHs show considerable variability in the vicinity of
the storm center, we applied a smoothing to the simulated SWHs
by using model data from 2 grid points on all four sides of a given
point (i.e., a nine-point smoothing). Sample results are shown in
Figs. 6 and 7 for Hurricane Ivan. Comparisons with data, shown
in Fig. 7, indicate that the model simulations are fairly realistic,
and also show the occurrence of SWHs larger than 10 m. At buoy
42,007, though, the measured SWHs are larger than model data.
Several factors could adversely affect the modeling effort and the
comparisons. The first is the storm surge effect (i.e., wind-induced
water level changes and surface currents), which can modify
the waves in shallow water. This effect is not included in the Fig. 6 Modeled SWHs for Hurricane Ivan

Journal of Offshore Mechanics and Arctic Engineering AUGUST 2013, Vol. 135 / 031104-5
 Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
Fig. 7 SWH comparisons

northwards from Corpus Christi in water depths between 8.7 m

and 15.8 m. Again, the comparison is reasonable (generally the
differences between the modeled SWHs and the data are less than
0.5 m).

3.2 Generation of SWH Database. For the 1958–2008

period, HURDAT snapshots are available for 145 hurricanes in
the GOM. However, the data are incomplete for 18 hurricanes;
therefore only the remaining 127 hurricanes were included. The
modified RV model was used to reconstruct the hurricane
windfields, which, as noted earlier, were supplemented with the
Fig. 8 SWH comparison during Hurricane Camille (Aug. 1969)
reanalysis winds; the resulting wave conditions modeled as
described in preceding paragraphs. Since it is the maxima in spe-
Table 3 Maximum SWH’s along Texas coast during Hurricane cific time intervals that influence the extreme wave statistics, it is
Ike necessary to assess the quality of such maxima. Here, we use the
monthly maxima at the location of the eleven buoys for assessing
Gage Gauge max. SWH (m) Modeled max. SWH (m) the simulations. Figure 9 shows a comparison involving a total of
2760 pairs. It must be noted that the comparison involves a small
R 4.5 3.99 subset of the overall model results, since the buoy data-lengths are
S 2.9 3.36
U 4.9 4.6
much smaller than the overall modeling period and there were
V 4.5 4.85 instances where the buoy stopped recording. For the comparison,
W 5.1 5.3 model data from the grid point closest to the buoy has been used.
X 5.6 6.1 Figure 9(a) shows that most of the model data and buoy data are
Y 4.8 5.3 fairly close to the slope-one line (slope of the best-fit line is
Z 5.6 5.55 0.82), indicating that the model data are, for the most part, rea-
sonable surrogates and can be used for further (statistical) analy-
ses. However, closer examination reveals that in the case of larger
if one takes the storm track uncertainty into consideration. Finally, wave heights the predicted SWHs appear to be larger than data.
turning to coastal regions, we show in Table 3 a comparison of Since larger wave heights are of importance in the extreme
modeled SWHs with data from the eight gauges deployed by value statistics, we separated the 2760 points into two sets, those
Kennedy et al. [44] during Hurricane Ike. The gauges, denoted larger than 7 m and those smaller than 7 m and the best fit line
by the letters R through Z, were located along the Texas coast obtained for each (Fig. 10). The formula from the best-fit line can

031104-6 / Vol. 135, AUGUST 2013 Transactions of the ASME

 Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
Fig. 9 Comparison of monthly maxima for the eleven NDBC buoys

Fig. 10 Best-fit lines for two groups of monthly maxima

be used to modify the simulated SWHs in order to improve the plotted the annual SWH maxima as a function of time for each
surrogate dataset. The modified results are shown in Fig. 9(b) for grid point and determined the slope of the best-fit line. The results,
all eleven buoy locations and the comparison appears fairly rea- summarized in Fig. 12, indicate that there is a decreasing trend in
sonable. As a result, the correction was applied to all modeled the western half of the GOM (spatial average ¼ 1.6 cm/year),
results. while the eastern half shows an increasing trend (spatial
average ¼ þ2.05 cm/year). The maximum increasing trend in the
eastern half is of the order of 5.6 cm/year, comparable to the esti-
4 Trends in the Wave Climate mates reported by Graham [9] and Ruggiero et al. [11] for the
Using the simulations for the pre-2004 period and for the period U.S. west coast for the Pacific. If only data after the mid-seventies
including the 2004–2008 hurricane season, the maximum SWH is examined, the maximum trend estimates are approximately
obtained from the simulations at each of the 7676 grid points was 4.8 cm/year (for 1975–2008) and 7.95 cm/year (for 1976–2008);
determined. Figure 11 shows the number of times the SWH and if the recent extreme hurricane season is excluded, they are
exceeded 10 m in the Gulf of Mexico; the frequency and spatial 1.77 cm/year (for 1975–2003) and 5.90 cm/year (for 1975–2003).
extent of these large waves clearly increased during 2004–2008.
The maximum number of events exceeding 10 m in SWH during
first 46 years is 8, but is of the order of 12 or more the last four 5 Statistical Methods
years (2004–2008). The largest modeled SWHs occurred in 1969 Estimating the extreme wave statistics at each location is usu-
(Hurricane Camille), with a simulated value of approximately ally done by fitting a distribution to the extremes in successive
18 m in the offshore areas (27 50 N, 87 80 W). time intervals and then using the distribution to estimate the prob-
As mentioned in Sec. 1, several researchers have noted ability of occurrence associated with a specified SWH. The most
an increasing trend in the wave climate in various parts of the common distributions appear to be the Gumbel, the Weibull, and
Atlantic and the Pacific oceans. Identification of such trends is (more recently) the generalized extreme value (GEV) distributions
important for establishing design criteria based on nonlinear (e.g., Xu and Huang [25]; Panchang et al. [18,3]; Neelamani et al.
(i.e., time-dependent) statistical models (e.g., Wang et al. [8]; [46]; Perez et al. [47]; FEMA [48]; Pontes et al. [16]). The param-
Jeong and Panchang [45]; Ruggiero et al. [11]). For the GOM, we eters associated with these distributions were estimated for a given

Journal of Offshore Mechanics and Arctic Engineering AUGUST 2013, Vol. 135 / 031104-7
 Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
Fig. 12 Trend distribution

to identify the preferred model. Not only is this tedious, but this
also constrains one to the chosen model when comparing multiple
time-frames. We follow recent developments by Li et al. [27] who
have provided more robust and quantitative ways based on the
jackknife and the bootstrap resampling methods along with an
error norm for such identification. Thus, the most appropriate
model can be selected by the engineer, rather than limiting oneself
to any preselected model.
In the bootstrap method, a large number of alternative but
equally plausible datasets are created by resampling the original
dataset. The Jackknife method can also be used to create
resampled datasets, but the bootstrap method is selected because
the degree of randomness in the bootstrap resampling is appa-
rently greater and the stability of the optimized parameters has
been rigorously tested by Li et al. [27]. The weighted error norms
of the maximum likelihood optimizations for each of the three cu-
mulative density functions are calculated as the average of
Fig. 11 Number of times SWH > 10 m: 195822003 (a), and 1958– K ¼ 500 bootstrap replicates based on the difference D between
2008 (b) the empirical and the model probabilities (Li et al. [27]):
*  h +
    i  h 

set of data (say x1, x2, x3, …, xn) by the method of maximum likeli- E Di Qj ¼ E max   GQi rij 
1im m þ 1 
hood. The probability associated with a specified “recurrence
interval” or the “return period” N (in years) is
where E stands for the expected value, Q is optimized parameter,
1 GQ is the cumulative density function under consideration, r1,
PðSWH < SWHN Þ ¼ 1  r2, …, rm are the ordered data X in increasing magnitude, and
SN
j ¼ 1, 2, …, K represents the individual sample number.
(where S is the number of data points per year and SWHN is the The distribution with the smallest error norm is to be preferred.
SWH associated with a return period of N years), and the corre- The parameter h controls the emphasis that we place on different
sponding SWHN is determined by equating this probability to the parts of the data while fitting the distribution to the data. When
cumulative density function. As stated earlier, a rule of thumb h ¼ 1, both the lower tail and the upper tail are equally important.
(Borgman [28]) permits extrapolation to duration that is approxi- For h < 1, emphasis is placed on the lower tail, while h > 1
mately three times as long as the length of the data. To create the emphasizes the upper tail where ‘extreme’ events occur. Compu-
dataset (x1, x2, x3,…, xn) we have followed the classical approach tations using these methods were made at all grid points in the
which involves the use of a subset corresponding to the annual GOM and the preferred distributions based on the error norm are
maxima (e.g., Carter and Challenor [49]); this can generally be discussed in Sec. 7.
expected meet to the requirement of the extremal distributions
that the extremes be drawn from samples (in this case, annual 6 Application of Probability Distributions to Model
data) which belong to the same population. We did not model the
effects of directionality or seasonality as described by Jonathan Data and Buoy Data: Validation
et al. [50] and Jonathan and Ewans [41,51] since the expected Models are an effective way to create a long-term (51 years in
improvements are still a matter of debate (Mackay et al. [52]; this case) grid-based database for extreme wave height estimation,
Jonathan and Ewans [53]. allowing the creation of a dataset consisting of annual maxima
As to the choice of distribution, most engineering studies con- required in the classical method of extremal analysis. However,
fine themselves to a preselected distribution (e.g., Neelamani et al. even when using the best models, errors are inevitable, as seen in
[45]; Berek et al. [14]); if multiple models are used, a number of Sec. 3, and generally the numerical results are checked against
subjective measures (e.g., Petrauskas and Aagaard [54]) are used data, when possible, for individual events. The cumulative effects

031104-8 / Vol. 135, AUGUST 2013 Transactions of the ASME

 Table 4 Summary of NDBC buoy data

Buoy Total data available Missing data Comments

42001 19762008 (33) 19761978 (Poor quality) NA

2007 (Jan.Feb.)
40002 19762008 (33) 1976 (Jan.Sept.) Three large SWHs (10 m) are missing
(1977, 1980, and 2005).
1977 (Aug.Sept.) At other times, when data are missing,
1978 (Aug.)1979 (Nov.) model results indicate mostly small SWH’s,
1980 (Aug.) but sometimes of the order of 5 m.

Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
1999 (Jan.Feb.)
2001 (Jan.Aug.)
2005 (July)
42003 19772008 (32) 1977 (Jan.Jun.) One large SWH (10 m) is missing (2008).
1986 (Nov.)1987 (Apr.) At other times, when data are missing,
1987 (Dec.)1988 (Apr.) model results indicate mostly small SWH’s,
1994 (Sept.)1995 (Jan.) but sometimes of the order of 5 m.
1996 (Sept.)1997 (May)
2005 (Aug.Sept.)
2008 (Aug.Nov.)
42007 19812008 (28) 1981 (Oct.)1983 (Jan.) When data are missing, model results indi-
1983 (Apr.Dec.) cated fairly small SWH’s (< 3.5 m).
1987 (Jan.Apr.)
1990 (Aug.)1996 (Sept.)
42019 19902008 (19) 1990 (Jan.May) NA
2000 (Jan.Jun.)
42020 19902008 (19) 1990 (Jan.May) NA
1993 (Aug.Nov.)
1997 (Feb.–Oct.)
42035 19932008 (16) NA 40 km buoy drift during Hurricane Ike.
42036 19942008 (15) 2005 (July) One large SWH (10 m) is missing (2005).
2007 (Feb.Sept.)
42039 19952008 (14) 1995 (all) Annual maximum data from 1996 are used.
42040 19952008 (14) 1995 (Jan.Nov.)
42041 19992005 (7) 1999 (Jan.Nov.) Almost half the data are missing.
2001 (March)2002 (April)
Disestablished after 2005 (April)

of the errors on the estimated SWH100 are rarely examined. Of errors are very large, one can expect a greater impact on the esti-
course, if the overall modeled dataset is long, one can expect the mated SWH100. However, the effect of the errors on the estimated
effect of individual modeling errors (i.e., inaccurate data) on SWH100 must be quantified before proceeding to an eventual
SWH100 to be small; and if the dataset is short, or if even a few application of the statistical models to the full modeled dataset.

Table 5 Estimated SWH100 (m)

(a) Before (2004) (b) Including 2004–2008

Buoy no. Data D (km) No. of year Gumbel Weibull GEV Gumbel Weibull GEV

42001 Buoy 20 28/33 10.62 10.95 12.84 12.64 12.32 14.74

Model 11.65 12.63 14.2 12.90 14.31 16.1
42002 Buoy 12.4 28/33 10.36 9.49 10.47 9.80 9.12 9.82
Model 12.11 12.45 14.56 11.95 12.76 14.25
42003 Buoy 16.1 27/32 10.64 11.24 12.62 12.44 12.43 14.25
Model 12.27 13.24 15.03 13.44 14.92 16.09
42007 Buoy 3.6 23/28 5.75 7.02 5.78 7.61 10.4 10.53
Model 6.71 8.2 8.07 7.16 8.0 7.92
42019 Buoy 5.4 14/19 7.18 6.20 6.26 7.25 6.80 6.88
Model 8.52 8.98 10.0 8.51 9.30 9.71
42020 Buoy 12.8 14/19 8.05 9.40 10.90 9.02 9.94 10.47
Model 12.37 11.26 14.83 12.59 12.90 17.19
Model* 8.03 8.1 10.0 9.9 10.5 13.5
42035 Buoy 7.7 11/16 5.94 7.55 6.59 6.76 7.53 7.44
Model 5.37 5.67 6.4 6.24 6.75 7.39
42036 Buoy 9.2 10/15 9.9 11.52 10.28 9.57 8.90 8.84
Model 9.21 10.77 9.36 10.8 13.08 10.78
42039 Buoy 10.1 8/13 11.0 11.9 13.01 14.49 19.2 17.8
Model 11.3 12.39 13.53 14.15 19.24 17.3
42040 Buoy 9.8 8/13 13.2 16.35 19.45 21.9 31.2 31.8
Model 13.02 14.85 19.1 18.48 23.45 24.56

Journal of Offshore Mechanics and Arctic Engineering AUGUST 2013, Vol. 135 / 031104-9
 Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024

Fig. 13 Modeled and buoy annual maxima

Fortunately, as noted earlier, there are 11 NDBC wave buoys in of time, the dependence of the errors on the length of the dataset
the GOM (Fig. 2) and some of them have been recording data can be assessed.
since 1976 (Table 1), creating a database that is sufficiently long While buoy data may appear to have no problems such as mod-
for SWH100 estimation. In some cases the rule of thumb for eling errors, other problems associated with them can influence a
extrapolating to duration three times the length of the data is satis- comparison of data-derived and model-derived SWH100 estimates.
fied. In fact, Palao et al. [2] and Panchang et al. [3] have used Typical problems found were as follows. First, there are frequent
these data to obtain such estimates. Thus these data can be used to gaps in the data (Table 4). The gaps are sometimes for several
estimate SWH100 as a means to cross-check model-derived esti- months (which include the hurricane season) or for several hours
mates. And, since the buoy data are available for varying lengths (including in some cases the peak of a storm). Second, there can

031104-10 / Vol. 135, AUGUST 2013 Transactions of the ASME

 Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
Fig. 14 Estimated SWH100 using the Gumbel method and 46 Fig. 15 Estimated SWH100 using the Weibull method and 46
years (a) and 51 years (b) of data years (a) and 51 years (b) of data

be uncertainties about the measurement location. This is some- useful for estimation of SWH100. A summary of the examination
times due to a deliberate relocation of the buoy by NOAA, and at is provided in Table 4. On the basis of information contained
other times due to a considerable drift of the buoy in a storm. Dur- therein, data from buoy 42,041 were discarded. For the other ten
ing Hurricane Ike (2008), for example, buoy 42,035 drifted buoys, the annual maxima were used with the Gumbel, the Wei-
approximately 40 km from its original location. If the buoy is bull, and the GEV distributions as described in Sec. 4. Estimates
located near the storm track, where the wave height variability is of SWH100 were obtained for the period excluding the years
greater, the effect of such location changes on SWH100 compari- 2004–2008 and then by including them. Results were obtained
sons at a prechosen point (we used a single location for each using both buoy data and model data. For the model data, length
buoy) can be substantial. Finally, the quality of the measurements of the dataset chosen corresponded to the earliest and latest buoy
may have changed over the years and some measurements could data availability.
be erroneous (Bender et al. [42]). Therefore, even though the Several features are apparent in the results which are presented
buoys provide ‘data’, estimates obtained from these cannot be in Table 5. First, at most locations, the SWH100 estimates obtained
regarded as the ‘absolute’, but perhaps as a generally more ‘reli- using buoy data and model data are close to each other for the
able’ value compared with models. A further point to be noted is most part, especially for the Gumbel and Weibull distributions.
that when making comparisons between model and buoy-derived The differences are of the order of 10%. Frequently, the buoy-
SWH100 estimates is that the buoy location does not usually coin- derived SWH100 is smaller than model SWH100, which in some
cide with a model grid point. Here, the model grid point closest to cases is probably a result of missing buoy data. This is likely in
the buoy was used, and the distance between the two varied the case of buoys 42,001, 42,002, and 42,003 where at times an-
between 3.6 km and 20 km. In addition, if the location happens to nual maxima of the order of 7 m (based on model calculations)
be near the storm track, the variability in the wavefields can create were missing. The differences between buoy and model SWH100
large differences in the two estimates. are certainly larger in the case of 42,020. The reasons for this
Data from all eleven buoys were carefully examined and were examined. Figure 13 shows a plot of the modeled and buoy
obvious problems were identified. When data during hurricane annual maximum. It appears that one data point is responsible for
months was missing, model results for those months were used this mismatch, i.e., year 1999, where the modeled SWH is 12 m as
to assess the usability of a particular buoy. If model results (or against the buoy SWH of 8 m. However, on examination this buoy
HURDAT data) suggested that very large SWH’s may have was found to be close to the track of Hurricane Bret, and in a
occurred during such periods, the remaining buoy data are not region of considerable wave height variability as seen in model

Journal of Offshore Mechanics and Arctic Engineering AUGUST 2013, Vol. 135 / 031104-11
 Table 6 Estimated maximum SWH100 values

1958–2003 1958–2008

Gumbel 18.2 m 19.1 m

Weibull 19.9 m 22.6 m
GEV 25.4 m 26.7 m

Overall, the similarity of the model-derivedand buoy-derived

Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
SWH100 values shown in Table 5 for the Gumbel and Weibull dis-
tributions for both periods is a validation of the modeling proce-
dure. Further, we see that the effect of individual errors
diminishes as the data length increases. In consequence, the
three distributions are applied using all the model data developing
Sec. 3 for 51 years.

7 Results and Discussion: SWH 100 Estimates for the

GOM
After all the simulations were completed, a dataset for each
grid point consisting of 51 annual maxima was developed. For
each of the 7676 grid points, the 3 distributions described in
Sec. 5 were fitted to the data and the SWH100 was estimated,
and the bootstrap technique was used to identify the preferred
distribution.
The SWH100 estimated by three different distributions are
shown in Figs. 14–16 using both the pre-2004 data and the data
that includes the 2004–2008 seasons. The maximum estimate
(spatially) obtained with the Gumbel distribution is the smallest
and that with GEV is the largest (Table 6). In these two cases, the
effect of the 2004–2008 seasons is to increase maximum SWH100
estimate by 0.9 m and 1.3 m, respectively. The effect of the
2004–2008 season is more pronounced in the case of the Weibull
distribution (maximum difference of 2.7 m) and the spatial
variability of the differences is also greater, suggesting that the
Weibull distribution has much greater sensitivity to the data than
the other two.
Fig. 16 Estimated SWH100 using the GEV method and 46 years
The estimates of SWH100 provided in API [1,15] and Berek
(a) and 51years (b) of data
et al. [14] appear to suggest a maximum difference in the SWH100
of the order of 3.8 m. Calculations provided in Figs. 14–16 and
results. (Recall the issues pertaining to track uncertainty noted in Table 6 indicate that the effect of the 2004–2008 hurricane season
Sec. 3; furthermore, the buoy location and the closest grid point is somewhat less. This would suggest that the differences in the
are separated by about 12.8 km.) As an experiment, the effect of SWH100 described in the two API publications may not be entirely
replacing the modeled value at this location with the buoy value due to the 2004–2008 hurricane season but could also be an effect
was studied (denoted by model* in Table 5). The resulting model of the approach taken to obtain the estimates, i.e., differences in
SWH100 were then much closer to buoy SWH100, as shown in the datasets as well as the distributions used may have contributed
Table 5. This demonstrates the considerable effect of modeling to the differences.
errors when the dataset is a relatively short, as opposed to much To summarize, Figs. 14–16 provides a range of SWH100’s in the
smaller effect in the case of buoy 42,001, 42,002, and 42,003 Gulf of Mexico. The range for the maximum SWH100 is as shown
where the longer dataset better corresponds to the rule of thumb. in Table 5. It is interesting to note from Figs. 14–16 that these
The difference for buoy 42,020 between the two models is maxima occur just south of Louisiana; no estimates for the area
smaller as the duration increases to include the 2004–2008 period have been provided by API [15] (2007). However, estimates
(Table 5, right), indicating the diminishing importance of isolated obtained by MacAfee and Wong [55] in this region are of the order
large errors. The table also shows that the differences between of over 20 m, which are consistent with the results in Table 6.
buoy and model SWH100 are greater for the GEV distributions While multiple estimates have been developed in Table 6, the
than for the other two distributions. The reason for this is that for question of which distribution is to be preferred remains. While
the GOM, the GEV appears to emphasize the ‘upper tail’, as there can be no absolute answer to this question, as explained in
shown later in Sec. 6. Sec. 4, the bootstrap technique provides some guidance. 500
Finally, we see that the effect of the 2004–2008 periods is to samples were generated using the data at each grid point and the
increase the SWH100 at most locations, generally by about 2 m or error norm calculated for six values of h (0.35, 0.5, 0.75, 1, 1.25,
less (based on the buoy data). At the location of buoy 42,040, and 1.5); the distribution with the least error norm is selected. The
however, the effect is much greater (e.g., using the Gumbel distri- results, shown in Fig. 17, indicate that for small values of h the
bution the buoy SWH100 increases from 13 m to 22 m). This Gumbel distribution is the preferred one, and for h > 1, the GEV
can be attributed to two reasons: very large wave heights were is the preferred distribution. As h increases from small values, at
recorded in 2004–2008, and the data length is short. Adding the many locations, the Weibull and the GEV are the preferred distri-
high SWH period of 2004–2008 increases the data length from butions (Fig. 17).
only 8 years to 13 years. When the datasets are longer the effect Recalling that small values of h correspond to an emphasis on
of the 2004–2008 seasons is smaller. the lower tail and h > 1 corresponds to an emphasis on the upper

031104-12 / Vol. 135, AUGUST 2013 Transactions of the ASME

 Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
Fig. 17 Variation of optimal distribution with h (GEV: red, Gumbel: blue)

plots in API [15] near the locations of the eleven buoys.) The
GEV estimates are, for the most part, higher than the API esti-
mates. Note that in shallow areas (e.g., near buoys 42,007 and
42,035, where the depth is  14 m), the addition of a 5 m storm
surge could result in a 1 m increase in the SWH100. It is interesting
that the Weibull results obtained here (for 51 years) are mostly
close to (but somewhat larger than) the API results [15]. (With the
exception buoy 42,041, the difference is less than about 2 m.) The
API study also used the Weibull distribution and though their
dataset and modeling methods are different and their statistical
analysis could be sensitive to the threshold chosen, it is encourag-
Fig. 18 SWHs corresponding to different return periods ing that the two sets reinforce each other. While our maximum
Weibull estimate (22.6 m) is considerably greater than the maxi-
mum (15.8 m) reported by Berek et al. [14], it must be noted that
tail, the smooth progression between the distributions observed in the API [15] does not provide results for certain areas where our
Fig. 17 implies that the Gumbel and the GEV are the two calculations show large SWH100 values (south of Louisiana).
extremes. This allows the engineer to make a reasoned selection, Also, as noted above, the Weibull results appear to be the
i.e., if one wants to emphasize the lower tail (where most of the most sensitive to the data (suggesting perhaps that the other
data lie), then the Gumbel distribution is to be preferred, as illus- distributions may be more robust), and the GEV, which provides a
trated by Fig. 18. However, if a better fit to the larger data values better fit to the larger data values (e.g., Fig. 18), may provide a
is to be emphasized, the GEV would be the preferred choice reasonable and conservative alternative to the engineer. In the
(Fig. 18). case of the GEV, though, it must be noted that a better fit to the
Table 7 shows a comparison between the three different meth- higher data values implies larger estimates if N (the return period)
ods and API [15] estimates. (The latter are estimated from the is much greater than the data length. Because of the nonlinear

Journal of Offshore Mechanics and Arctic Engineering AUGUST 2013, Vol. 135 / 031104-13
 Table 7 Comparison of estimated SWH100 (m) at NDBC buoy locations

GEV Gumbel Weibull API

Buoy no. 46 year 51 year 46 year 51 year 46 year 51 year RP2-WSD 2INT-MET MAX SWH

42001 16.53 17.10 13.35 13.83 14.30 14.97 11.9 15.2 11.63
42002 13.82 13.65 11.88 11.78 11.81 11.69 11.9 12.3 9.7
42003 15.78 16.58 12.42 13.17 13 14.08 11.9 13.03 11.04
42007 7.79 7.87 6.51 6.81 7.07 7.62 NA NA 9.09
42019 13.37 12.89 10.84 10.65 11.31 11.19 10.5 11.6 6.3
42020 15.48 15.92 11.8 12 11.81 12.19 10.6 11.8 8.2

Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
42035 6.68 6.88 5.75 5.97 5.78 6.08 5.3 7.44 7.1
42036 12.63 12.8 10.5 10.85 11.81 12.34 11.8 10.2 8.6
42039 16.43 18.84 12.75 14.34 14.05 15.79 11.8 13.77 12.05
42040 19.09 20.37 14.28 15.44 16.24 18.33 11.9 NA 16.91
42041 18.78 19.61 14.58 15.26 15.32 16.22 11.9 12.44 12.31

nature of the curve in Fig. 18, it is not reasonable to obtain esti- alternative estimates. In addition, efforts were directed
mates for return periods of the order of 200, 500 and even 1000 toward identifying the preferred distribution. SWH100 esti-
years as provided in API [15]. mates were obtained using 46 and 51 years of model data.
The Weibull estimates were similar to (but somewhat larger
8 Summary and Conclusions than) the revised API estimates (also obtained with the
Weibull distribution but using different data and statistical
In this study, SWH estimates at 7676 grid points in the GOM methods). The bootstrap method indicated that, for all grid
were obtained using a combination of the modified Rankine points, the GEV distribution is to be preferred if the empha-
Vortex model (Jeong et al. [23]) and the reanalysis dataset, along sis is on the larger wave heights where extreme occur, and
with the SWAN wave model, for the 1958–2008 periods. The the Gumbel distribution is to be preferred if the emphasis is
modeled wind and wave height data were validated using nearly on the lower tail. The maximum increase caused by the
all the data available. Analyses of these data lead to the following 2004–2008 season in the SWH100 is approximately 2.7 m
conclusions: (for the Weibull distribution), 1.3 m (for the GEV distribu-
(1) Modeled monthly maximum SWH values, when compared tion), and 0.9 m (for the Gumbel distribution). These are
with corresponding buoy data (available up to 33 years), smaller than the difference between the old and new API
yielded a slope of 0.82 (based on 2760 points), indicating a guidelines. The 2004–2008 seasons resulted in greater
high correlation. The mismatch was due partly to storm spatial variability of the SWH100 estimates. Table 5 and
track uncertainty, separation between the chosen grid point Figs. 16–18 provided alternative estimates for SWH100 in
and the buoy, and the large wave height variability near the the GOM. The maximum SWH100 estimates using the three
storm track. methods are 19.1 m, 22.6 m, and 26.7 m (Gumbel, Weibull,
(2) Model results for 51 years at 7676 grid points show that the and GEV); while these estimates are larger than the revised
maximum modeled SWH is 18 m. The effect of the API estimates, it is noted that the latter are not available for
2004–2008 seasons is not large on the maximum SWH in all areas of the GOM.
the GOM, but the frequency of large waves (SWH > 10 m)
increased substantially. Also, the eastern part of the GOM
shows an increasing trend while the western part a decreas-
ing trend. In the eastern part, the maximum trend is approx- Acknowledgment
imately 5.6 cm/year in the maximum SWH (based on 51
This work is an outgrowth of coastal wave prediction work sup-
years), 1.77 cm/year based on the 1975–2004 period, and
ported by the Texas Coastal Management Program (NOAA). Zeki
4.84 cm/year based on the 1975–2008 period. The numbers
Demirbilek is grateful for the technical support received in this
that include the latest time periods are comparable to esti-
activity from two research programs of the U.S. Army Corps of
mates obtained for the west coast of the U.S. (e.g., Graham
Engineers: Coastal Inlets Research Program (CIRP), and Dredg-
[9]; Ruggiero et al. [11]). The trends can be used for obtain-
ing Operations and Environmental Research (DOER). Permission
ing time-dependent estimates of extreme wave estimates;
to publish this paper was granted by the Chief, Army Corps of
this type of estimation is receiving greater prominence
Engineers.
owing to sea level rise and climatic changes (e.g., Obeyse-
kera et al. [13]; Warner and Tissot [12]).
(3) SWH100 were obtained using annual maximum obtained by References
modeling and from buoy data. The buoy data lengths varied [1] American Petroleum Institute (API), 2000, “Recommended Practice for Plan-
from 10 years to 33 years. The SWH100 estimates mostly ning, Designing and Constructing Fixed Offshore Platforms—Working Stress
Design,” API Recommended Practice 2A-WSD, Washington, DC.
increased upon inclusion of the 2004–08 hurricane season. [2] Palao, I. M., Teng, C. C., and Brown, D. A., 1994, “An Extremal Analysis of
The SWH100 estimates obtained using the two datasets are Significant Wave Height Data Measured From NDBC Buoys,” Report No.
mostly consistent, with a maximum difference of 2 m (for 1804-07.01, NDBC, Stennis Space Center, MS, p. 46.
the most part). This consistency is a validation of the [3] Panchang, V. G. L., Zhao, L., and Demirbilek, Z., 1999, “Estimation of
Extreme Wave Heights Using GEOSAT Measurements,” Ocean Eng., 26, pp.
model-derived data for statistical analysis. Using buoy 205–225.
42,020 as an example, it was seen that errors in individual [4] Wang, D. W., Mitchell, D. A., Teague, W. J., Jarosz, E., and Hulbert, M. S.,
modeled SWHs have a decreasing effect on SWH100 as the 2005, “Extreme Waves Under Hurricane Ivan,” Science, 309, p. 896.
data length increased from 1990–2003 to 1990–2008. [5] Hovis, G. T., 2005, “Analysis of Storm Surge Measured at Water Level Stations
From Hurricanes Charley, Frances, Ivan, and Jeanne,” Proc. Solutions to
(4) The revised API manual [15] notes that statistical Coastal Disasters 2005, ASCE, Reston, VA, p. 1–15.
methods other than those used in their study can yield [6] Panchang, V. G., and Li, D., 2006, “Large Waves in the Gulf of Mexico Caused
slightly different results. This study explored a range of by Hurricane Ivan,” Bull. Am. Met. Soc., 87(4), p. 481–489.

031104-14 / Vol. 135, AUGUST 2013 Transactions of the ASME

 [7] Clayton, B., 2007, “Gearing up for Gulf Hurricanes,” Houston Chronicle, Apr [31] MacAfee, A. W., and Pearson, G. M., 2006, “Development and Testing of
29. Tropical Cyclone Parametric Wind Models Tailored for Midlatitude Applica-
[8] Wang, X. L., Zwiers, F. W., and Swail, V. R., 2004, “North Atlantic Ocean tion—Preliminary Results,” J. Appl. Meteor. Climatol., 45, pp. 1244–1260.
Wave Climate Change Scenarios for the Twenty-First Century,” J. Climate, [32] Xie, L., Bao, S., Pietrafesa, L. J., Foley, K., and Fuentes, M., 2006, “A Real-
17(12), pp. 2368–2383. Time Hurricane Surface Wind Forecasting Model: Formulation and Ver-
[9] Graham, N., 2005, “Coastal Impacts of North Pacific Winter Wave Climate ification,” Mon. Weather Rev., 134, pp. 1355–1370.
Variability: The Southern California Bight and the Gulf of Farallones,” Report [33] Booij, N., Ris, R. C., and Holthuijsen, L. H., 1999, “A Third-Generation Wave
No.CEC-500-2005-018, Scripps Institution of Oceanography and Hydrologic Model for Coastal Regions: 1. Model Description and Validation,” J. Geophys
Research Center, State of California Energy Commission, San Diego, CA, Res., 104(C4), pp. 7649–7666.
p. 34. [34] Ris, R. C., Holthuijsen, L. H., and Booij, N., “A Third-Generation Wave Model
[10] Komar, P. D., and Allan, J. C., 2007, “Higher Waves Along US East Coast for Coastal Regions: 2. Verification,” J. Geophys Res, 104, pp. 7667–7681.
Linked to Hurricanes,” EOS Trans., 88(30), p. 301. [35] Booij, N., Holthuijsen, L. H., Doorn, N., and Kieftenburg, A. T. M. M., 1997,
[11] Ruggiero, P., Komar, P. D., and Allan, J. C., 2010, “Increasing Wave Heights “Diffraction in a Spectral Wave Model,” Ocean Wave Measurements and Anal-

Downloaded from http://asmedigitalcollection.asme.org/offshoremechanics/article-pdf/135/3/031104/6243397/omae_135_3_031104.pdf by Hong Kong Univ. Sci & Tec user on 02 December 2024
and Extreme Value Projections: The Wave Climate of the U.S. Pacific North- ysis, Proc. 3rd Internat’l Sympos. (Waves97), Virginia, ASCE, New York, pp.
west,” Coastal Eng., 57, pp. 539–552. 243–255.
[12] Warner, N. N., and Tissot, P. E., 2012, “Storm Flooding Sensitivity to Sea [36] Zubier, K., Panchang, V. G., and Demirbilek, Z., 2003, “Simulation of Waves
Level Rise for Galveston Bay, Texas,” Ocean Eng., 44, pp. 23–32. at Duck (North Carolina) Using Two Numerical Models,” Coastal Eng. J,
[13] Obeysekera, J., Park, J., Irizzary-Ortiz, M., Barnes, J., and Trimble, P., 2013, 45(3), pp. 439–469.
“Probabilistic Projection of Mean Sea Level and Coastal Extremes,” J. Water- [37] Chen, Q., Zhao, H. H., Hu, K. L., and Douglass, S. L., 2005, “Prediction of
ways, Port, Coastal Ocean Eng., 139, pp. 135–141. Wind Waves in a Shallow Estuary,” J. Waterway, Port, Coastal., Ocean Eng.,
[14] Berek, E. P., Cooper, C. K., Driver, D. B., Heideman, J. C., Mitchell, D. A., 1341(4), pp. 137–148.
Stear, J. D., and Vogel, M. J., 2007, “Development of Revised Gulf of Mexico [38] Singhal, G., Panchang, V. G., and Lillibridge, J., 2010, “Reliability Assessment
Metocean Hurricane Conditions for Reference by API Recommended for a Wave Forecasting System for Prince William Sound, Alaska,” J. Water-
Practices,” Offshore Technology Conference (OTC), Houston, TX. way, Port, Coastal. Ocean Eng., 136(6), pp. 337–349.
[15] American Petroleum Institute (API), 2007, “Interim Guidance on Hurricane [39] Liu, H., Xie, L., Pietrafesa, L. J., and Bao, S., 2007, “Sensitivity of Wind
Conditions in the Gulf of Mexico,” API Bulletin 2INT-MET, Washington, DC. Waves to Hurricane Wind Characteristics,” Ocean Modelling, 18, pp. 37–52.
[16] Pontes, M. T., Aguiar, R., and Oliveira Pires, H., 2005, “A Nearshore Wave [40] Heideman, J. C., and Mitchell, D. A., 2009, “Grid Point Pooling in Extreme
Energy Atlas for Portugal,” ASME J. Offshore Mech. Arct. Eng., 127(3), pp. Value Analysis of Hurricane Hindcast Data,” J. Waterway, Port, Coastal, Ocean
249–255. Eng., 135, pp. 31–37.
[17] Millar, D. L., Smith, H. C. M., and Reeve, D. E., 2006, “Modeling Analysis of [41] Jonathan, P., and Ewans, K., 2011, “A Spatiodirectional Model for Extreme
the Sensitivity of Shoreline Change to a Wave Farm,” Ocean Eng., 34(5–6), pp. Waves in the Gulf of Mexico,” ASME J. Offshore Mech. Arct. Eng., 133(1),
884–901. p. 011601.
[18] Panchang, V. G., Jeong, C., and Li, D., 2008, “Wave Climatology in Coastal [42] Bender, L. C., Guinasso, N. L., Walpert, J. N., and Howden, S. D., 2010, “A
Maine for Aquaculture and Other Applications,” Estuaries Coasts, 31(2), pp. Comparison of Methods for Determining Significant Wave Heights—Applied
289–299. to a 3-m Discus Buoy During Hurricane Katrina,” J. Atmos. Ocean. Technol.,
[19] Cieślikiewicz, W., and Paplińska-Swerpel, B., 2008, “A 44-Year Hindcast of 27(6), pp. 1012–1028.
Wind Wave Fields Over the Baltic Sea,” Coastal Eng., 55, pp. 894–905. [43] Earle, M. D., 1975, “Extreme Wave Conditions During Hurricane Camille,” J.
[20] Rogers, W. E., Kaihatu, J. M., Hsu, L., Jensen, R. E., Dykes, J. D., and Holland, Geophys. Res., 80(3), pp. 377–379.
K. T., 2007, “Forecasting and Hindcasting Wave With the SWAN Model in the [44] Kennedy, A. B., Gravois, U., and Zachry, B., 2011, “Observations of Landfal-
Southern California Bight,” Coastal Eng., 54, pp. 1–15. ling Wave Spectra During Hurricane Ike,” J. Waterway, Port, Coastal, Ocean
[21] Cardone, V. J., Greenwood, C. V., and Greenwood, J. A., 1992, “Unified Pro- Eng., 137(3), pp. 142–145.
gram for the Specification of Hurricane Boundary Layer Winds Over Surfaces [45] Jeong, C. K., and Panchang, V. G., 2008, “Measurement-Based Estimates of
of Specified Roughness,” Army Engineer Waterways Experiment Station, Con- Extreme Wave Conditions for the Gulf of Mexico,” Oceans 2008 Conference,
tract Report No. CERC-92-1, Vicksburg, MS. Quebec, MTS/IEEE.
[22] Forristall, G. Z., 2007, “Comparing Hindcasts With Wave Measurements From [46] Neelamani, S., Al-Salem, K., and Rakha, K., 2007, “Extreme Waves for
Hurricanes Lili, Ivan, Katrina and Rita,” Proc. 10th International Workshop on Kuwaiti Territorial Waters,” Ocean Eng., 34, pp. 1496–1504.
Wave Hindcasting and Forecasting and Coastal Hazards Symposium, North [47] Perez, O. M., Telfer, T. C., and Ross, L. G., 2003, “On the Calculation of Wave
Shore, Oahu, HI, Nov. 11–16. Climate for Offshore Cage Culture Site Selection: A Case Study in Tenerife,”
[23] Jeong, C. K., Panchang, V. G., and Demirbilek, Z., 2012, “Parametric Adjust- Aquacultural Eng., 29, pp. 1–21.
ments to the Rankine Vortex Wind Model for Gulf of Mexico Hurricanes,” [48] Federal Emergency Management Agency, 2004, “Final Draft Guidelines for
ASME J. Offshore Mech. Arctic Eng., 134(4), p. 041102. Coastal Flood Hazard Analysis and Mapping for the Pacific Coast of the United
[24] Powell, M. D., Houston, S. H., Amat, L. R., Morisseau-Leroy, N., 1998, “The States,” FEMA Study Contractor: Northwest Hydraulic Consultants, Inc.
HRD Real-Time Hurricane Wind Analysis System,” J. Wind Eng. Indust. Aero- [49] Carter, D. J. T., and Challenor, P. G., “Estimating Return Values of Environ-
dyn, 77–78, pp. 53–64. mental Parameters,” Q. J. R. Meteorol. Soc., 107, pp. 259–266.
[25] Xu, S., and Huang, W., 2011, “Estimating Extreme Water Levels With Long- [50] Jonathan, P., Ewans, K., and Forristall, G., 2008, “Statistical Estimation of
Term Data by GEV Distribution at Wusong Station Near Shanghai City in Extreme Ocean Environments: The Requirement for Modelling Directionality
Yangtze Estuary,” Ocean Eng., 38(2–3), pp. 468–478. and Other Covariate Effects,” Ocean Eng., 35(11–12), pp. 1211–1225.
[26] Van Vledder, G., Goda, Y., Hawkes, P., Mansard, E., Martin, M. J., Mathiesen, [51] Jonathan, P., and Ewans, K., 2011, “Modeling the Seasonality of Extreme
M., Peltier, E., and Thompson, E., 1993, “Case Studies of Extreme Wave Anal- Waves in the Gulf of Mexico,” ASME J. Offshore Mech. Arct. Eng., 133(2),
ysis: A Comparative Analysis,” Proc. 2nd International Sympos. Ocean Wave p. 021104.
Measurement and Analysis, ASCE, New York, pp. 978–992. [52] Mackay, E. B. L., Challenor, P. G., and Bahaj, A. S., 2010, “On the Use of Dis-
[27] Li, Y., Simmonds, D., Reeve, D., 2008, “Quantifying Uncertainty in Extreme crete Seasonal and Directional Models for the Estimation of Extreme Wave
Values of Design Parameters With Resampling Techniques,” Ocean Eng., 35, Conditions,” Ocean Eng., 37(5–6), pp. 425–442.
pp. 1029–1038. [53] Jonathan, P., and Ewans, K., 2011, “Discussion of ‘On the Use of Discrete Sea-
[28] Borgman, L. E., 1975, “Extremal Statistics in Ocean Engineering,” Civil Engi- sonal and Directional Models for the Estimation of Extreme Wave Conditions’
neering in the Oceans Conference, New York, pp. 117–133. by Edward B. L. Mackay, Peter G. Challenor, AbuBakr S. Bahaj [Ocean Eng.,
[29] Kalnaya, M. K. E., Kistler, R., Collins, W., Deaven, D., Gandin, L., Iredell, M., 37(5–6), 2010, pp. 425–442],” Ocean Eng., 38(1), 1, p. 205.
Saha, S., White, G., Woollen, J., Zhu, Y., Leetmaa, A., Reynolds, R., Chelliah, [54] Petrauskas, C., and Aagaard, P. M., 1971, “Extrapolation of Historical
M., Ebisuzaki, W., Higgins, W., Janowiak, J., Mo, K. C., Ropelewski, C., Storm Data for Estimating Design-Wave Heights,” Soc. Pet. Eng. J., 11, pp.
Wang, J., Jenne, R., Joseph, D., 1995, “The NCEP/NCAR 40-Year Reanalysis 23–37.
Project,” Bull. Am. Met. Soc., 77, pp. 437–471. [55] MacAfee, A. W., and Wong, S. W. K., 2007, “Extreme Value Analysis of
[30] Music, S., and Nickovic, S., 2008, “44-year Wave Hindcast for the Eastern Tropical Cyclone Trapped-Fetch Waves,” J. Appl. Meteor. Climatol., 46, pp.
Mediterranean,” Coastal Eng., 55, pp. 872–880. 1501–1522.

Journal of Offshore Mechanics and Arctic Engineering AUGUST 2013, Vol. 135 / 031104-15