Ocean Modelling 196 (2025) 102542

Contents lists available at ScienceDirect

Ocean Modelling
journal homepage: www.elsevier.com/locate/ocemod

On the directional distributions of the observation-based wind input and

dissipation terms in a third-generation wave model
Franklin F. Ayala a,c ,∗, Rubén D. Montoya b , Andrés F. Osorio a,d , Alexander Babanin c
a
Universidad Nacional de Colombia, Sede Medellín, Facultad de Minas, Departamento de Geociencias y Medio Ambiente, Grupo de investigación OCEANICOS, Cra
80 No. 65–223, Bloque M2, Medellín, Colombia
b
Universidad de Medellín, Grupo de Investigación en Ingeniería Civil GICI, Grupo de Investigación en Calidad del agua y Modelación Hídrica GICAMH, Cra 87
No. 30-65, Medellin, Colombia
c
Department of Infrastructure Engineering, University of Melbourne, Australia
d Center of Excellence in Marine Science, CEMARIN, Colombia

ARTICLE INFO ABSTRACT

Keywords: New observation-based directional spreading functions for the wind input term 𝑆𝑖𝑛 and the dissipation term
Directional distribution 𝑆𝑑𝑠 have been investigated in third-generation wave models (WAVEWATCH III, ST6 version). In this study
Spectral wave modeling we evaluate bimodal directional distribution for 𝑆𝑑𝑠 and narrower (cos2 𝜃 and cos4 𝜃) distributions for 𝑆𝑖𝑛 . The
WAVEWATCH III
impact of these new directional functions on the integral wave parameters, as well as the nondirectional and
Wind input term
directional spectra, is analyzed through a set of numerical experiments and their results are compared against
Dissipation term
buoy observations in the Caribbean Sea. Improvements in the estimation of 𝐻𝑠 and 𝑇𝑝 under wind-sea dominant
conditions are presented. For the shape of frequency spectra, the bimodal distribution of 𝑆𝑑𝑠 causes a weaker
energy enhancement by comparison with the narrower distribution of 𝑆𝑖𝑛 , but it generally shows a better
fit than the default functions in the model. The combined effect of both modifications brings larger energy
allocation around the main wave propagation direction, generating both an energy increase and a downshift
at the spectral peak. These findings suggest the relevance of including realistic directional distributions and
reveal the need for further studies of directional distributions of the source terms.

1. Introduction input, 𝑆𝑑𝑠 reflects the wave dissipation due to white-capping and
non-breaking mechanisms, and 𝑆𝑛𝑙 represents the redistribution of the
Third-generation spectral wave modeling has been broadly used in energy over the spectrum by means of nonlinear four-wave interactions.
wave hindcasting and real-time forecasting over the last four decades The source and sink terms are dependent on the frequency-direction
(Komen et al., 1994; Young, 1999). The phase-averaged wave models space.
propagate the wave action density 𝑁(𝑘, 𝜃, 𝑡, 𝐱) based on physical pro- The mechanism of Miles (1957) suggested that the atmospheric
cesses that contribute to the generation, evolution, and dissipation of
input 𝑆𝑖𝑛 is proportional to the wave spectrum itself, which implies an
the wind-waves involved in the radiative balance equation:
exponential growth and ultimately a positive-feedback mechanism:
𝜕𝑁 𝑆 𝑆 + 𝑆𝑑𝑠 + 𝑆𝑛𝑙
+ ∇ ⋅ 𝐱𝑁 = 𝑡𝑜𝑡 = 𝑖𝑛 (1)
𝜕𝑡 𝜎 𝜎 𝑆𝑖𝑛 (𝑓 , 𝜃) = 𝛽(𝑓 , 𝜃)𝑁(𝑓 , 𝜃) (2)
where the left side represents the spatio-temporal evolution of the
All the parametrizations for 𝑆𝑖𝑛 have extensively focused on de-
wave action density 𝑁 = 𝐹 (𝑘, 𝜃, 𝑡, 𝐱)∕𝜎. The variance density spectrum
termining the wave growth function 𝛽(𝑓 ), rather than the directional
𝐹 (𝑘, 𝜃, 𝑡, 𝐱) is defined as a function of the wavenumber 𝑘, direction 𝜃,
time 𝑡 and space 𝐱, while 𝜎 = 2𝜋𝑓 is the radian frequency. In deep distribution 𝐷(𝜃) used to spread the incoming energy in the spectrum
waters, it is broadly accepted that all energy fluxes responsible for the and generally assumed to follow a cosine function:
wind-wave evolution 𝑆𝑡𝑜𝑡 are primarily quantified by the source and
𝛽(𝑓 , 𝜃) = 𝛽(𝑓 )𝐷(𝜃) = 𝛽(𝑓 ) cos(𝜃) (3)
sink terms 𝑆𝑖𝑛 , 𝑆𝑑𝑠 and 𝑆𝑛𝑙 . 𝑆𝑖𝑛 is the growth rate by atmospheric

∗ Corresponding author at: Universidad Nacional de Colombia, Sede Medellín, Facultad de Minas, Departamento de Geociencias y Medio Ambiente, Grupo de
investigación OCEANICOS, Cra 80 No. 65–223, Bloque M2, Medellín, Colombia.
E-mail addresses: [email protected], [email protected] (F.F. Ayala).

https://doi.org/10.1016/j.ocemod.2025.102542
Received 12 July 2024; Received in revised form 2 March 2025; Accepted 17 March 2025
Available online 4 April 2025
1463-5003/© 2025 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

𝛽 was established to depend on the speed and direction of the wind This study aims to evaluate different directional distribution func-
and waves according to (Miles, 1957): tions for the wind input and dissipation source terms using buoy
( ) observations and WWIII model simulations during medium conditions.
𝑈
𝛽∝ cos 𝜃 (4) Details about the new directional distributions implemented and tested
𝑐
where 𝑈 is a reference mean wind speed (e.g. wind speed at 10 m in WWIII are given in Section 2. A description of the model setup
elevation, 𝑈10 ), 𝑐 is the wave celerity, and 𝜃 is the difference between for the simulation under regular conditions is presented in Section 3.
the wave and the wind direction (𝜃 − 𝜃𝑤 ). This dependence was later Section 4 showcases the results of the effect of these directional distri-
confirmed by different experimental studies that led to the parameter- butions on the wave integral parameters, nondirectional spectra, and
izations that constitute the basis for the most widely used wind input directional spectra. Finally, the discussion and conclusions are outlined
packages in today’s wave models: Snyder et al. (1981) and Plant (1982). in Sections 5 and 6, respectively.
Snyder et al. (1981) parametrization gave a temporal growth rate
as follows: 2. Directional spreading functions
( )
𝜌 𝑈10
𝛽(𝑓 , 𝜃) ∝ 𝑎 cos 𝜃 − 1 2𝜋𝑓 (5) A brief representation of the directional distributions for the wind
𝜌𝑤 𝑐
input and dissipation terms is presented below.
where 𝜌𝑎 and 𝜌𝑤 are the densities of air and water, respectively. On
the other hand, Plant (1982) conducted a study that exhibited another 2.1. Wind input term
dependence:
𝜌 ( 𝑢∗ )2 Rogers et al. (2012) defined a parameter 𝑊 used to compute the
𝛽(𝑓 , 𝜃) ∝ 𝑎 2𝜋𝑓 cos 𝜃 (6) wave growth rate under the effect of favorable winds, 𝑊1 (𝑘, 𝜃), and ad-
𝜌𝑤 𝑐
verse winds, 𝑊2 (𝑘, 𝜃), based on earlier studies carried out by (Donelan
where 𝑢∗ is the wind shear velocity. These previous formulations have
et al., 2006).
essentially hypothesized a cosine directional spreading function for the { }
𝑈 ( )
wind input. However, Shabani et al. (2016) proved that the wind input 𝑊1 (𝑘, 𝜃) = max2 0, cos 𝜃 − 𝜃𝑤 − 1 (9)
energy function follows a narrower distribution (cos3.6 𝜃) based on in- 𝑐
{ ( ) }
2 𝑈
situ measurements of the directional distribution of the wind shear 𝑊2 (𝑘, 𝜃) = min 0, cos 𝜃 − 𝜃𝑤 − 1 (10)
𝑐
stress during swell conditions. Even though the baseline formulation
𝑊 = 𝑊1 − 𝑎0 𝑊2 (11)
of Donelan et al. (2006) was also originally based on Snyder et al.
(1981), they incorporated experiment results from Hsiao and Shemdim where 𝑈 is the wind at 10 m elevation, 𝑈10 , 𝑎0 is a tuning parameter
(1983) which showed a quadratic form for the wave growth function to control the growth rate for adverse winds, and 𝜃 and 𝜃𝑤 represent
for a wider range of inverse wave ages. wave and wind direction, respectively. The adverse wind effect can be
( )2 assumed to be a proxy for swell dissipation.
𝜌 𝑈10
𝛽(𝑓 , 𝜃) ∝ 𝑎 cos 𝜃 − 1 2𝜋𝑓 (7) These previous components of the wind input term W are then
𝜌𝑤 𝑐
modified following a narrower directional distribution, described by
This expression was later adopted in the observation-based source
the product of the original cos(𝜃 − 𝜃𝑤 ) term and its absolute value
term (Rogers et al., 2012; Zieger et al., 2015), leading to a narrower
| cos(𝜃 − 𝜃𝑤 )|. This product preserves the sign of the cosine distribution,
cos 𝜃 function. Besides these expressions, the wind input of Janssen
which is important for determining whether the wind is favorable or
(2004), later modified by Ardhuin et al. (2010), utilized a tuning
adverse, and ultimately, whether it represents a positive or negative
parameter 𝑝𝑖𝑛 as a power for the cos 𝜃 term in their parametrizations
wind input in the source term.
based on Miles theory, allowing certain control over the directional
spreading for the wave growth rate. Despite these approaches, neither { }
𝑈 ( ) ( )
narrower cosine functions in the observation-based formulation nor 𝑊1 (𝑘, 𝜃) = max2 0, cos 𝜃 − 𝜃𝑤 | cos 𝜃 − 𝜃𝑤 | − 1 (12)
𝑐
values for 𝑝𝑖𝑛 greater than 2 (Rascle and Ardhuin, 2013; Stopa et al., { ( ) ( ) }
𝑈
2016) have been evaluated in the WAVEWATCH III model (hereafter 𝑊2 (𝑘, 𝜃) = min2 0, cos 𝜃 − 𝜃𝑤 | cos 𝜃 − 𝜃𝑤 | − 1 (13)
𝑐
WWIII).
On the other hand, the dissipation term 𝑆𝑑𝑠 has been also expressed 𝑊cos|cos| = 𝑊1 − 𝑎0 𝑊2 (14)
in terms of a dissipation rate 𝛾𝑑𝑠 : The magnitude of the directional spreading function is thus ap-
proximately equal to |cos(𝜃 − 𝜃𝑤 )2 |. The directional distribution of the
𝑆𝑑𝑠 (𝑘, 𝜃) = 𝛾𝑑𝑠 𝑁(𝑘, 𝜃) (8)
original 𝑊 function and the narrower 𝑊 function (hereafter denoted
Most of the current dissipation rate formulations have typically as 𝑊cos|cos| ) are depicted in Fig. 1, assuming 𝑈 = 10 m∕s, 𝑐 = 2.5 m∕s
been assumed to follow an isotropic distribution (Hasselmann, 1974; and a wind direction of 180◦ .
Tolman and Chalikov, 1996). Ardhuin et al. (2010) defined a direction- As seen, the integral over the directions of the 𝑊 function is higher
dependent part in their dissipation rate expression including a cosine than the integral over the directions for 𝑊cos|cos| , implying a reduction
term to compute the spectral saturation 𝐵(𝑘, 𝜃) and to allocate the in the magnitude of the wave-supported stress 𝜏⃗𝑤 :
maximum dissipation in the mean wave direction. This adjustment was 2𝜋 𝑘𝑚𝑎𝑥 𝑆𝑖𝑛 (𝑘, 𝜃)
necessary because the initial versions of this term were generating a 𝜏⃗𝑤 = 𝜌𝑤 𝑔 (cos 𝜃, sin 𝜃)𝑑𝑘, 𝑑𝜃 (15)
∫0 ∫0 𝑐
narrower spectrum than the measurements (Ardhuin and Boyer, 2006).
On the contrary, field experiments conducted by Young and Babanin where 𝑔 is the gravitational acceleration constant and 𝑘𝑚𝑎𝑥 is the
(2006) evidenced anisotropic behavior of the dissipation rate, as a highest discrete wavenumber. Since the reduction of 𝜏⃗𝑤 is not intended,
greater loss of energy was more concentrated at oblique angles to a factor 𝐹1 was computed to account for the exceedance over 𝑊 and
the main wave propagation direction. Babanin et al. (2010) formu- used to correct the magnitude of the 𝑊cos|cos| function. The factor is
lated a first approach to describe this mechanism following a bimodal determined as a function of the wavenumber 𝑘:
directional spreading function. This feature has been proposed for 2𝜋
∫0 𝑊 (𝑘, 𝜃)𝑑𝜃
the observation-based source term as one of the next developments 𝐹1 (𝑘) = (16)
2𝜋
that could improve the representation of the wind-sea system during ∫0 𝑊cos|cos| (𝑘, 𝜃)𝑑𝜃
extreme wind conditions (Liu et al., 2017), but it has not yet been Finally, the function 𝑊cos|cos|factor meets the required narrowness and
implemented in the latest WWIII version (WW3DG, 2019). satisfies the wave-supported stress constraint, as presented in Fig. 1.

2
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

Fig. 1. Comparison between the directional distribution of the original 𝑊 function (solid black line) and the narrower 𝑊cos|cos| function with and without the correction factor
(dashed and solid black line, respectively). 𝑈 = 10 m∕s, 𝑐 = 2.5 m∕s and a wind direction of 180◦ were assumed.

Fig. 2. Comparison of the directional spreading functions for the dissipation rate. The isotropic formulation is denoted with the solid black line. The anisotropic distribution
represented through a bimodal function is determined using the parameter values 𝜃𝑝 = 40◦ and 𝑝 = 2 (solid red line).

The 𝑊1 and 𝑊2 terms were also computed with an even narrower employed in the ST1, ST2, ST3, and ST6 source term packages in
distribution obtained by multiplying by |cos(𝜃 − 𝜃𝑤 )|3 , similar to the WWIII, the anisotropic distribution will be utilized and evaluated in
process presented in Eqs. (12)–(14). This resulted in a magnitude of this research.
the directional distribution similar to |cos(𝜃 − 𝜃𝑤 )4 |.
3. Data and methods
2.2. Dissipation term
3.1. Study zone
Babanin et al. (2010) proposed a bimodal distribution of the dissi-
pation source term based on the experiments at Lake George reported
by Young and Babanin (2006). This distribution relies on two param- The study area encompasses the Caribbean Sea and the Gulf of
eters: 𝜃𝑝 and 𝑝. The former represents the direction (in radians) where Mexico with a domain longitude interval of 98◦ W–58◦ W and a latitude
the maximum dissipation rate occurs relative to the main propagation interval of 8.5◦ N–32.5◦ N. As it is an enclosed basin, the Lesser and
direction of the waves, and the latter shifts the depth of the middle Greater Antilles block most of the swell coming from the open North
trough in the main propagation direction. The directional distribution Atlantic Ocean, leading to a strong dominance of wind-sea systems
function 𝑉 has the form: during low and moderate winds (Stopa et al., 2013). The wind climatol-
ogy in the Caribbean Sea can be described by two main and distinctly
( ( )2 )
⎧ ( )
atmospheric conditions. First, there are the extreme conditions during
( ) ⎪𝑉1 𝜃, 𝑓 , 𝑈10 ∕𝑐𝑝 = 𝐴(𝑓 ) exp −𝑝 𝜃 + 𝜃𝑝 , 𝜃<0
𝑉 𝜃, 𝑓 , 𝑈10 ∕𝑐𝑝 = ⎨ ( ) ( ( )2 ) the hurricane season in the Atlantic, which runs from June to Novem-
⎪𝑉2 𝜃, 𝑓 , 𝑈10 ∕𝑐𝑝 = 𝐴(𝑓 ) exp −𝑝 𝜃 − 𝜃𝑝 , 𝜃≥0 ber, and cold fronts, with the highest frequency of events occurring
⎩
in January, February, and March (Ortiz-Royero et al., 2013). Second,
(17) there are the so-called average conditions, governed by the dynamics
where 𝐴(𝑓 ) is the directional spectral narrowness, 𝑐𝑝 is the phase speed of the trade winds, which vary throughout the annual cycle.
of the spectral peak, and 𝜃 represents the angle (in radians) relative The trade winds in the Caribbean are primarily influenced by the
to the main propagation direction of the waves. Both 𝜃𝑝 and 𝑝 were movement of the Inter-Tropical Convergence Zone (ITCZ) across the
defined as variables dependent on 𝑈10 ∕𝑐𝑝 and the frequency 𝑓 , but equatorial region, with two maxima in the summer and winter (DJF and
their relationship was not explicitly determined. Therefore they were JJA), and two minima in the fall and spring (MAM and SON) (Wang,
assumed to be constant throughout this study. As an initial approach, 2007; Ruiz-Ochoa and Bernal, 2009). Large-scale phenomena such as
a value 𝑝 = 2 was defined based on the performance of the research the El Niño/Southern Oscillation (ENSO) also affect trade wind patterns
wave model WAVETIME-1 against existing experimental data, as shown in the Caribbean on various spatial and temporal scales. With this con-
by Tsagareli (2009). 𝜃𝑝 was also established to be equal to 40◦ , derived text in mind, the simulation period was defined based on the following
from work of Young and Babanin (2006) (see their Fig. 18). climate conditions: (i) Neutral ENSO year, (ii) Low-energy trade winds,
The directional distributions for the dissipation rate used in this and (iii) Absence of hurricanes and cold fronts. The simulation period
study are outlined in Fig. 2. While the isotropic behavior is currently was defined between 1 May 2020 to 31 May 2020.

3
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

Fig. 3. Bathymetry and orography in the study domain for the case under regular conditions. The red triangles correspond to the NOAA buoy locations with measurements
available for the simulation period.

3.2. Model configuration Table 1

Summary of the directional spreading functions tested in each experiment for the case
under regular conditions.
A rectangular grid was used to propagate the waves over a domain
Experiment 𝑆𝑖𝑛 directional 𝑆𝑑𝑠 directional distribution
with a spatial resolution of 1/8◦ . The spectral discretization consisted name distribution
of 30 frequency steps, starting with a frequency of 0.042 Hz and
ctrl cos Isotropic
increasing by a factor of 1.1. For the direction domain, 36 intervals cos2 cos2 Isotropic
with a resolution of 𝛥𝜃 = 10◦ were also configured. The global, cos4 cos4 Isotropic
propagation, refraction and source terms time steps were 400, 100, bim cos Anisotropic (bimodal 𝑝 = 2,
100 and 30 s respectively. The numerical scheme used was the third- 𝜃𝑝 = 40◦ )
cos4 +bim cos4 Anisotropic (bimodal 𝑝 = 2,
order Ultimate Quickest scheme taken from Leonard (1991). Wind
𝜃𝑝 = 40◦ )
input and dissipation terms were activated from the ST6 source term
package (Rogers et al., 2012; Zieger et al., 2015; Liu et al., 2019)
while nonlinear interactions were solved with Discrete Interaction Ap-
proximation (DIA, Hasselmann et al. (1985)). Furthermore, the depth- where 𝐸𝑓 𝑛 (𝑓 ) ≡ 𝐸(𝑓 )∕𝑚0 , 𝐸(𝑓 ) represents the nondirectional spectrum
induced breaking parametrization by Battjes and Janssen (1978) and and 𝑚0 is the zeroth moment or total energy in the 2D wave spectrum.
the JONSWAP bottom friction formulation were activated. Finally, the effect on the directional spectrum was also evaluated.
The ETOPO1 bathymetry database (Amante and Eakins, 2009) was Two different 𝐻𝑠 hindcast series were compared with the ctrl
utilized to extract depth, mask and obstruction data for the entire experiment: one forced with ERA5 winds and the other with CFSR
domain using the Gridgen software (Chawla and Tolman, 2007). For winds. The data were obtained from Liu et al. (2021) for the study zone
this study, the hourly reanalysis data from ERA5 (Hersbach et al., 2019) and interpolated at the buoy locations.
with a spatial resolution of 1/4◦ was adopted to drive the simulations
described below. 3.4. Measurements

3.3. Description of the experiments Three different buoys (42057, 42058, and 42059) from the National
Data Buoy Center (NDBC) were utilized to validate the case under
A description of the experiments conducted is given in Table 1. All regular conditions. These buoys are located in the Caribbean Sea (see
the experiments used the default values for the tunable parameters in Fig. 3) and record wave integral parameters such as 𝐻𝑠 , 𝑇𝑝 , and peak
ST6, except CDFAC which was set to 1.08 based on a suggestion from direction 𝜃𝑝 (under nautical convention), as well as one-dimensional
an ERA5-driven global wave hindcast (Liu et al., 2021). The original (1D) wave spectra, 𝐸(𝑓 ). Buoy data had a 1-h temporal resolution and
isotropic distribution for the dissipation rate and the cos function for a variable frequency resolution ranging from 0.005 Hz to 0.02 Hz in
the wind input were used in the ctrl experiment. The 𝐜𝐨𝐬𝟐 and 𝐜𝐨𝐬𝟒 the frequency domain.
tests incorporated cos2 and cos4 distributions for the wind input term, Although direct buoy measurements for the 2D spectra 𝐸(𝑓 , 𝜃) are
respectively, while the dissipation term was kept isotropic. The bim not provided, they can be calculated based on a directional spreading
experiment used the original wind input spreading (cos) as in the function 𝐷(𝜃) (Holthuijsen, 2007):
ctrl simulation but modified the dissipation spreading into a bimodal
form (anisotropic). Finally, the 𝐜𝐨𝐬𝟒 +bim run fully incorporated both
𝐸(𝑓 , 𝜃) = 𝐸(𝑓 )𝐷(𝜃) (19)
changes: the cos4 term and the bimodal behavior.
1( ( ) )
The results modeled with the experiments were contrasted with 𝐷(𝑓 , 𝜃) = 0, 5 + 𝑅1 cos 𝜃 − 𝛼1 + 𝑅2 cos(2(𝜃 − 𝛼2 )) (20)
𝜋
observations to evaluate their impact in different ways. Firstly, the
impact on the wave scalar parameters (e.g. significant wave height 𝐻𝑠 The first and second normalized directional Fourier coefficients,
and the dominant wave period 𝑇𝑝 ) was determined, and secondly, the 𝑅1 and 𝑅2 , respectively, along with the mean (𝛼1 ) and principal (𝛼2 )
effect on the nondirectional spectra was assessed, including a variable wave directions, can be directly downloaded from NDBC. It is worth
to quantitatively compare the spectral shape, namely the narrowness mentioning that the validation of the measurements made by the pitch-
in the frequency space 𝑄𝑐 proposed by (Rogers and Vledder, 2013): roll buoys relies on certain assumptions, which are sometimes not
satisfied due to the hull of the buoy not recognizing the true tilt of
𝑄𝑐 = max(𝐸𝑓 𝑛 )∕𝑇𝑚,−1,0 (18) the waves. Additionally, the method used to estimate 𝐷(𝑓 , 𝜃) has some

4
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

limitations derived from the computation of the cross-spectra and the spreading functions for the wind input, although it maintains a better
appearance of unrealistic negative values (Montoya et al., 2013), which fit than the ctrl experiment.
were set to zero in this study. The combined effect of the new directional distributions (𝐜𝐨𝐬𝟒 +bim)
Other more complex approaches to estimate the directional spread- outperforms all the experiments (𝑏 = −0.01 m and RMSE = 0.2 m).
ing, such as Maximum Likehood Method (MLM), Maximum Entropy Differences of up to 0.4 m are evidenced between the ctrl and 𝐜𝐨𝐬𝟒 +bim
Method (MEM), Extended Maximum Entropy Principle (EMEP), and simulation for the maximum 𝐻𝑠 values. Despite this, the modeled 𝐻𝑠
Wavelet Directional Method (WDM) can also be used to spread the is still underestimated for all the experiments for higher 𝐻𝑠 measure-
wave energy in the spectrum (Lin et al., 2022). While the Longuet- ments.
Higgins Method (LHM; Longuet-Higgins et al. (1963)), employed in As shown in Table 2, experiments bim and 𝐜𝐨𝐬𝟐 show slightly less
this study, has been reported to be faster, it can struggle to accurately bias than the ctrl simulation for buoy 42057, while other metrics like
reproduce narrow directional distributions, often resulting in broader 𝜌 and 𝑆𝐼 barely change among the experiments. Meanwhile, experi-
wave spectra compared to other methods (Benoit, 1994). ments for buoy 42059 that involve an individual modification of the
MEM and MLM have also some limitations (Donelan et al., 2015). directional distribution, whether it be 𝐜𝐨𝐬𝟐 , 𝐜𝐨𝐬𝟒 or bim, have the lowest
For instance, MEM tends to generate artificial double peaks on NDBC RMSE values, ranging between 0.15 and 0.16. The combined 𝐜𝐨𝐬𝟒 +bim
dataset and it has lower directional resolution than MLM (Earle et al., overestimates 𝐻𝑠 as indicated, with the highest negative bias error
1999). Additionally, MLM can slightly underestimate wave energy and RMSE of all the experiments. All the experiments tend to show
around the mean wave direction, while it produces overestimation of a decrease in positive bias, implying an overall energy increase due to
energy in the opposite direction (Zhang and Zhang, 2006). Even though the new directional spreading functions, particularly evident in the 𝐜𝐨𝐬𝟒
using other approaches could generate even narrower wave systems and 𝐜𝐨𝐬𝟒 +bim experiments.
than the single-peaked ones presented in Figs. 9 and 10, their primary The peak wave period 𝑇𝑝 mostly ranges between 6 and 10 s, rein-
advantage could be exploited in scenarios involving multiple wave forcing the evidence of the dominance of local wind-sea (see Fig. 5).
systems and where high operational level is required. For all the buoys, the ctrl simulation generally depicts the largest bias
𝑏 and RMSE values (see Table 3). In spite of 𝐜𝐨𝐬𝟐 , 𝐜𝐨𝐬𝟒 and 𝐜𝐨𝐬𝟒 +bim
3.5. Metrics for model validation have similar performances in terms of RMSE, the narrower wind input
experiments 𝐜𝐨𝐬𝟐 and 𝐜𝐨𝐬𝟒 mostly outperform the simulations, excepting
the combined experiment, which achieves the best bias correction in
The following statistical metrics were used to compare the wave
buoy 42057 and the highest correlation coefficient 𝜌 for buoy 42059.
quantities of the model estimates 𝑦 with the buoy observations 𝑥: bias
The bimodal experiment again has a minor effect on the enhancement
𝑏, root-mean-square error 𝑅𝑀𝑆𝐸, correlation coefficient 𝜌, and scatter
of the modeled 𝑇𝑝 similar to the 𝐻𝑠 comparison presented previously.
index 𝑆𝐼:
All the experiments are significantly biased for buoy 42057 compared
1 ∑(
𝑁
) to buoys 42058 and 42059.
𝑏= 𝑦 − 𝑥𝑖 (21)
𝑁 𝑖=1 𝑖 The buoy located in the Central Caribbean Sea (42058) exhibits the
√ strongest correlation coefficient 𝜌 for 𝐻𝑠 and 𝑇𝑝 of all the buoys, with
√
√1 ∑ 𝑁
( )2 the slightest tendency toward overprediction of these variables. The
RMSE = √ 𝑦 − 𝑥𝑖 (22)
𝑁 𝑖=1 𝑖 conditions at this buoy location are expected to be fully developed (or
∑𝑁 almost fully developed) because of the long fetch.
𝑖=1 (𝑦𝑖 − 𝑦)(𝑥
̄ 𝑖 − 𝑥) ̄ It is noteworthy that all the experiments consistently lead to a
𝜌= √ √ (23)
∑𝑁 ∑𝑁 decrease in bias, resulting in an increase in the modeled 𝑇𝑝 compared
𝑖=1 (𝑦𝑖 − 𝑦)̄2 ̄ 2
𝑖=1 (𝑥𝑖 − 𝑥)
√ to the observations. This increase leads to lower peak frequencies and
1 ∑𝑁 [ ]2
a downshift of the spectral peak. Therefore, the modifications on the
𝑁 𝑖=1 (𝑦𝑖 − 𝑦) ̄ − (𝑥𝑖 − 𝑥)
̄
𝑆𝐼 = (24) directional spreading functions of the wind input and dissipation term
𝑥̄
not only affect the amount of energy in the wave spectra but also the
where the bar over 𝑥 and 𝑦 denotes their mean values, and 𝑁 indicates
location of the spectral peak.
the number of data points.
Research studies have shown that errors in estimating directional
wave spectra may go unnoticed when comparing modeled and mea-
4. Results sured data solely based on wave integral parameters such as 𝐻𝑠 and
𝜃𝑝 (Portilla-Yandún et al., 2015). A precise estimation of these scalar
4.1. Effects on scalar wave parameters variables does not necessarily ensure an accurate representation of the
1D or 2D wave spectra, especially in scenarios where locally generated
There is a considerable wind-wave alignment that reflects the wind- waves and swell systems occur within the same sea state. This issue
sea dominance in the Caribbean Sea with deviations between those appears whether in medium or extreme conditions (Fan et al., 2009;
directions no longer than 20◦ (see Fig. 11). Notably, buoy 42058 has Fan and Rogers, 2016; Yang et al., 2022) and therefore the performance
the longest fetch and wind speed values (between 7 and 12 m/s), which of the experiments at estimating the 1D and 2D wave spectra is also
lead to 𝐻𝑠 values of up to 4 m. addressed.
Before comparing the ctrl experiment with the simulations in Table
1, it is worth nothing that other hindcasts show similar performance. 4.2. Effects on the 1D wave spectra
ERA5-based 𝐻𝑠 is less biased than CFSR-based 𝐻𝑠 (Çalışır et al., 2021;
Liu et al., 2021) and 𝐻𝑠 values computed with ctrl for all the buoys, Measured and modeled 1D wave spectra were compared at the two
except the buoy 42057, where the ERA5-based hindcast performs worst. most energetic instances (𝐻𝑠 peaks) for buoy 42058, observed at 13:00
Fig. 4(c)–(f) exhibit that any experiment incorporating a variation in UTC on 10 May 2020 and 18:00 UTC on 17 May 2020. A 3-point
the directional spreading of either the wind input or dissipation term moving average was used to smooth/remove the spikes in the frequency
has a lower RMSE compared to ctrl simulation. In particular, 𝐜𝐨𝐬𝟒 space (see Fig. 6).
demonstrates a better agreement than 𝐜𝐨𝐬𝟐 with a reduction in bias The ctrl simulation exhibits the lowest energy values across the
from 0.1 m to 0.03 m. Additionally, the isolated effect of the bimodal frequency range at both times and the bim, 𝐜𝐨𝐬𝟐 , and 𝐜𝐨𝐬𝟒 simulations
distribution for the dissipation, evaluated through the bim experiment, evidence a significant energy enhancement around the spectral peak
had a weaker influence on the 𝐻𝑠 series than the narrow directional (0.11–0.12 Hz). The bimodal distribution for the dissipation has a

5
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

Fig. 4. Comparison of simulated and observed 𝐻𝑠 values for buoy 42058 between the ctrl experiment (red dots) and (a) ERA5-based hindcast values (cyan dots), (b) NCEP climate
forecast system reanalysis (CFSR)-based hindcast values (purple dots) (c) 𝐜𝐨𝐬𝟐 (dark blue dots), (d) 𝐜𝐨𝐬𝟒 (pink dots), (e) bim (gray dots), and (f) 𝐜𝐨𝐬𝟒 +bim (green dots). Both
hindcast values from (a) and (b) are obtained from Liu et al. (2021). The statistical metrics for bias 𝑏, RMSE, correlation coefficient 𝜌 and scatter index (𝑆𝐼) are given in each
subplot.

Table 2
Error metrics for the modeled-observed 𝐻𝑠 comparison for each experiment and buoy. The model performance indicators are bias 𝑏, RMSE, correlation coefficient 𝜌 and the scatter
index (SI). The best metric values found in the experiments are shown in bold.
Experiment Buoy 42057 Buoy 42058 Buoy 42059
𝑏 RMSE 𝜌 SI 𝑏 RMSE 𝜌 SI 𝑏 RMSE 𝜌 SI
ctrl 0.02 0.13 0.89 0.12 0.21 0.29 0.93 0.09 0.09 0.17 0.88 0.09
hindcast ERA5 −0.23 0.26 0.89 0.13 0.12 0.24 0.93 0.09 −0.03 0.15 0.87 0.09
hindcast CFSR −0.09 0.18 0.85 0.14 0.26 0.35 0.91 0.11 0.14 0.22 0.79 0.1
cos2 −0.01 0.13 0.89 0.12 0.1 0.23 0.93 0.09 0.02 0.15 0.88 0.09
cos4 −0.03 0.14 0.89 0.13 0.03 0.21 0.93 0.09 −0.04 0.16 0.89 0.1
bim 0.01 0.13 0.89 0.12 0.17 0.27 0.93 0.09 0.06 0.16 0.88 0.09
cos4 +bim −0.04 0.15 0.89 0.13 −0.01 0.2 0.93 0.09 −0.07 0.18 0.89 0.1

Table 3
Same as Table 2, but for 𝑇𝑝 . Notice that there are no 𝑇𝑝 hindcast series for the buoys and the comparison is only between the experiments.
Experiment Buoy 42057 Buoy 42058 Buoy 42059
𝑏 RMSE 𝜌 SI 𝑏 RMSE 𝜌 SI 𝑏 RMSE 𝜌 SI
ctrl 1.33 1.97 0.35 0.19 0.26 0.55 0.79 0.06 0.17 0.59 0.53 0.09
cos2 1.23 1.96 0.32 0.2 0.1 0.49 0.79 0.06 0.01 0.57 0.53 0.09
cos4 1.22 2.0 0.3 0.21 −0.03 0.48 0.79 0.06 −0.1 0.57 0.54 0.09
bim 1.32 1.97 0.35 0.19 0.22 0.53 0.79 0.06 0.14 0.58 0.53 0.09
cos4 +bim 1.18 1.97 0.3 0.21 −0.07 0.48 0.79 0.06 −0.13 0.58 0.55 0.09

smaller impact on peak energy. On the other hand, the narrower mostly similar and the experiments do not evidence discrepancies in
distribution of wind input, the higher the energy at the peak, leading to comparison with the ctrl simulation.
a better agreement with the observations. The combined modification All the experiments, including the ctrl simulation, tend to under-
(𝐜𝐨𝐬𝟒 + bim) gives the best fit around the spectral peak, with differences predict the narrowness in the frequency space reported by buoy 42058,
up to 4 m2 s in the peak frequency compared to the ctrl experi- which leads to broader modeled nondirectional spectra than the mea-
ment. The experiments still underestimate the spectral peak energy surements (see Fig. 7). The statistical metrics indicate that the ex-
(the modeled peak is still ∼8 m2 s lower than the measurements). periments barely affect the model performance when computing the
For all the simulations, the energy at the high-frequency tail remains frequency narrowness 𝑄𝑐 (differences up to 0.02 m in RMSE and 0.03

6
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

Fig. 5. Same as Fig. 4 but for 𝑇𝑝 and excluding the comparison with the hindcast-based values.

Fig. 6. 1D wave energy spectra for buoy 42058 at (a) 13:00 UTC 10 May 2020 and (b) 18:00 UTC 17 May 2020. Both axes in each subplot are in log–log scale. The inset in
each subplot represents a zoomed-in area around the spectral peak. The legend follows the same convention as in Fig. 4.

7
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

Fig. 7. Comparison of simulated and observed 𝑄𝑐 values for buoy 42058 between the ctrl experiment (red dots) and (a) 𝐜𝐨𝐬𝟐 (dark blue dots), (b) 𝐜𝐨𝐬𝟒 (pink dots), (c) bim (gray
dots), and (d) 𝐜𝐨𝐬𝟒 +bim (green dots). The statistical metrics for bias 𝑏, RMSE, correlation coefficient 𝜌 and scatter index (𝑆𝐼) are given in each subplot.

in 𝜌). Frequency narrowness values of this order of magnitude were also obtained with the ctrl and 𝐜𝐨𝐬𝟒 +bim do not substantially differ from
reported by Rogers and Vledder (2013). each other (green arrows generally overlap the red ones).
The new directional distributions therefore clearly affect the energy
level at the spectral peak while mainly maintaining the spectral shape 4.3. Effects on the 2D wave spectra
initially registered by the ctrl simulation. This broadening of the nondi-
rectional spectra has been reported and is associated with the use of 4.3.1. Visual comparison of individual 2D spectra
DIA as the nonlinear solver for the four-wave nonlinear interactions The effects of the new directional spreading functions on the di-
term in the phase-averaged wave models (Hasselmann et al., 1985). rectional wave spectra were also evaluated for the two most energetic
To provide further support to the previous interpretations, two instances (𝐻𝑠 peaks) for buoy 42058, as conducted for the analysis
RMSE values were compared: (i) a RMSE calculated between the 1D of the 1D wave spectra. The 2D buoy spectra were also smoothed by
spectra modeled with the ctrl experiment and the observed 1D spectra, a 3-point moving average for each direction bin. In order to remove
and (ii) a RMSE computed between the 1D spectra modeled with 𝐜𝐨𝐬𝟒 unreliable and noisy signals from the measured data, the minimum
+ bim and the observed 1D spectra. This difference is time-dependent energy limit was defined as 0.05% of the peak energy from the buoy.
and it is normalized by the first RMSE mentioned above. As shown in The modeled spectra were interpolated at the buoy frequencies to
Fig. 8, the peak wave direction modeled with the ctrl and 𝐜𝐨𝐬𝟒 + bim calculate the differences between the simulations and the observations.
and the wind direction are represented through a stick plot for each For buoy 42058 at 13:00 UTC 10 May 2020, the observed spectrum
day of the simulation period. shows a peak around 0.1 Hz, and most of the energy is spread over
For the first 20 days of the simulation period, the 𝐜𝐨𝐬𝟒 +bim ex- a directional sector of 90◦ , mainly covering directions from NE to SE.
periment mostly shows lower RMSE values with differences of up to The wave peak direction indicates that the highest incoming energy
60%–70% of the RMSE computed with the ctrl simulation (orange is from the east, and the wind direction shows an alignment with
areas). Intermittent improvement occurs for the time instances that easterly winds (see Fig. 9). A visual inspection easily proves that both
surround the maximum 𝐻𝑠 values (see also the longest wind direc- the directional spreading and the energy for frequencies higher than
tion arrows for those dates). After 20–21 May 2020, the combined 0.25 Hz are greater for the modeled spectra than for the measurements.
experiment performs worse than the ctrl simulation (blue areas). A The ctrl and 𝐜𝐨𝐬𝟒 +bim experiments effectively reproduce the energy
larger deviation between the modeled peak wave direction and the distribution along the frequency-direction space. The bimodal effect
wind direction is evidenced on 22, 27 and 28 May, while the lowest of a lower dissipation rate at ±40◦ from the main wave propagation
wind speeds are also seen on these dates. The peak wave directions direction (around 45◦ ) and a narrower wind input distribution impact

8
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

Fig. 8. (a) Stick plot for buoy 42058 showing the wind direction used in the model (black arrows) and the peak wave direction modeled with the ctrl (red arrows) and the
𝐜𝐨𝐬𝟒 +bim (green arrows) experiments. The arrows point in the direction the waves/winds are going to. (b) Normalized differences of RMSE values between the ctrl and 𝐜𝐨𝐬𝟒 +bim
for the 1D spectra during the simulation period for buoy 42058. Orange values mean positive normalized differences (the RMSE from the ctrl experiment is higher than the RMSE
from 𝐜𝐨𝐬𝟒 +bim) and blue values indicate the opposite.

Fig. 9. Comparison of 2D wave spectrum at 13:00 UTC 10 May 2020 for buoy 42058. The observed spectrum of the buoy is presented in (a) while the spectra modeled with the
ctrl and 𝐜𝐨𝐬𝟒 +bim simulations are shown in (b) and (c), respectively. The spectra showing the differences corresponding to each simulation are depicted in panels (d) and (e)
and the black contour is defined for 0.2. The black and red arrows are pointing in the wind and peak wave directions, respectively, in accordance with meteorological convention.
The wave spectrum was normalized by the peak energy registered in the buoy spectrum.

the amount of energy allocated in and around the spectral peak (see directions. This wind-wave direction alignment reinforces the evidence
a lighter red color for the spectrum showing the differences between of the presence of wind-sea conditions for this buoy.
the buoy and the 𝐜𝐨𝐬𝟒 +bim for the NE-SE directional sector). This joint For the other 𝐻𝑠 maximum, most of the energy is spread further
effect results in a moderate energy increase and better performance is south than the previous time instance and the peak frequency is higher
demonstrated in the directions close to the wave peak direction and for than 0.1 Hz. Neither experiment properly captures the energy dis-
frequencies higher than 0.1 Hz in comparison with the ctrl simulation. tributed over the S-SE directions. However, the 𝐜𝐨𝐬𝟒 +bim experiment
Additionally, during periods longer than 10 s, the energy represented exhibits lower differences than the ctrl experiment in the eastward
for the 𝐜𝐨𝐬𝟒 +bim experiment is greater than the ctrl simulation and direction. Both the modeled spectra are generally wider above the peak
the model indicates a slight northward shift in the wave peak and wind frequency compared to the measurements and show a smoother energy

9
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

Fig. 10. Same as Fig. 9, but at 18:00 UTC 17 May 2022. The black contour in the differences spectra is defined for 0.1.

decay along the frequencies. At this time, the wind-wave alignment is experiments can be easily attributed to two factors: (i) a narrower
weaker than in the first peak but it is still typical of wind-sea conditions. directional distribution for 𝑆𝑖𝑛 generates a higher amount of energy
The energy allocated around the peak frequency is again higher in around directions associated with wind-sea, and (ii) a bimodal dissi-
the combined experiment than the ctrl and a subtle southward shift pation spreading for 𝑆𝑑𝑠 allocates a lower dissipation rate in wind-sea
is present for 𝐜𝐨𝐬𝟒 +bim. directions, resulting in less energy lost. It is important to highlight that
other 𝑆𝑑𝑠 models, such as the one proposed by Romero (2019), have
4.3.2. Comparison of 2D spectra time series also been capable to correct the directional biases from the current
The RMSE differences between the 2D spectra simulated with the ST6 physics by employing an anisotropic distribution for the breaking
ctrl and 𝐜𝐨𝐬𝟒 +bim experiments for buoy 42058 were calculated by con- dissipation. This parametrization conversely assigns greater energy loss
trasting their corresponding nondirectional spectra for each direction. in the peak direction to reproduce better the bimodal lobe separation
This comparison was made by subtracting the RMSE values and scaling and bimodal amplitude of the wind-wave spectrum for waves shorter
them by the RMSE values calculated by the ctrl simulation, as pre- than the dominant scale (Romero and Lubana, 2022).
sented for the one-dimensional analysis. The results are shown in Fig. 𝐜𝐨𝐬𝟒 + bim outperforms all experiments in estimating 𝐻𝑠 values for
11 and provide further evidence of the bimodal effect of the dissipation buoy 42058 at time instances where the ctrl simulation usually pro-
term and the fact the narrower distributions for the wind input allocate duces underpredictions. However, this behavior is not fully consistent
more energy around the main wave propagation direction (mostly among the buoys and it can lead to overestimations of the 𝐻𝑠 , mainly
aligned with the wind direction). During the first half of the simulation presented in the buoys 42057 and 42059. Utilizing other directional
period, this joint effect improves the model simulations by around distributions with less energy in the wind-sea directions such as bim
±30◦ from the wind direction (except around 8 May 2020), which and 𝐜𝐨𝐬𝟐 usually generate a best-fit with the observations. Although
nearly coincides with the interval between the higher lobes in the it remains unclear, the misrepresentation from the joint simulation
dissipation rate determined by parameter 𝜃𝑝 from Eq. (17). After 20–21 could primarily be due to the use of a fixed bimodal form over the
May 2020, the ctrl simulation outperforms the combined experiment time in the dissipation term, while the wave age dependence plays an
around the main propagation direction. The improved performance important role in the computation of directional distribution for the
of the combined experiment for directions below the wind direction wind-wave (Romero and Melville, 2010).
(between 50◦ and 90◦ ) is worth highlighting. This asymmetric behavior Similar to 𝐻𝑠 , 𝑇𝑝 values are better reproduced for buoy 42058 under
around the wind direction was also present between 06 May and 10 the joint simulation, while simulations like 𝐜𝐨𝐬𝟐 show closer agreements
May when the ctrl simulation for directions over 90◦ was showing a for the other buoys. Lower dissipation and higher wind input rates
best-fit. in the wind-sea frequencies are expected to have a strong impact in
𝑇𝑝 , especially because the former feature varies over the frequencies
5. Discussion (e.g. less pronounced at frequencies higher than the peak frequency
𝑓𝑝 ) according to Babanin (2009). At the same time, dissipation rate
Regarding the simulations with individual modifications, the en- starts decaying again at angles larger than 𝜃𝑝 , generating a possible
ergy increase reflected by 𝐻𝑠 increments for the 𝐜𝐨𝐬𝟐 , 𝐜𝐨𝐬𝟒 and bim increase in swell energy. This behavior could explain the increases in

10
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

Fig. 11. Normalized differences of RMSE values between the ctrl and 𝐜𝐨𝐬𝟒 +bim for the 2D spectra during the simulation period for buoy 42058. Red values mean positive
normalized differences (RMSE from ctrl experiment is higher than RMSE from 𝐜𝐨𝐬𝟒 +bim) and blue values indicate the opposite. The black curve indicates the wind direction.

𝑇𝑝 presented in Fig. 5 and the downshifting of the spectral peak in Fig. dissipation rate along the main wave propagation direction can also
6. foster better estimation of directional wave spectra during tropical cy-
When considering the shape of the nondirectional spectra, the en- clones. This is because higher energy values in the wind-sea systems are
ergy increase tends to enlarge the peakedness, but this effect is balanced usually present in the rear quadrants and are currently underestimated
with greater energy values at low frequencies which generate a slightly in numerical wave modeling (Liu et al., 2017). Further experiments are
broader spectrum and ultimately, the spectral narrowness 𝑄𝑐 is mostly needed to validate these changes in the aforementioned regions.
conserved, or even improved, with the new directional distributions.
Additionally, the high-frequency tail is well-represented for all the 6. Conclusions
experiments (see Fig. 6), indicating that the modifications in the di-
rectional spreading were correctly implemented in the WWIII source Recapitulating the research gap addressed in this study, different di-
code. rectional distributions for the wind input and the dissipation term were
When the ctrl is compared with the combined experiment repre- proposed based on in-situ measurements, but their potential effects on
senting the nondirectional spectra during the simulation period, the the current widely-used wave models had not been analyzed until now.
new observation-based directional distributions clearly help improve This paper attempted to evaluate the model performance when these
the capability of the model around the spectral peak. The precise directional spreading functions were implemented. The main findings
reason for the occasional overprediction of the spectral peak cannot are summarized as follows:
be determined by looking at the wind-wave alignment (see Fig. 8) as • A new directional distribution in the wind input term of the ST6
the model performs both better and worse than the ctrl during similar source package was formulated and tested in WWIII under wind-
misalignments. sea conditions in the Caribbean Sea during the non-hurricane
The evaluation of the 2D spectra shows greater wave energy values season. Simulations derived from cos2 and cos4 functions showed
with the new distributions in the directional sector around the mean good agreement with integral and spectral buoy measurements,
wave direction. This helps enhance the wave energy and is the main especially in the higher wind conditions where the control exper-
effect of modifying the directional spreading of the wind input and iment tended to underestimate the wave energy. These modifica-
dissipation term. Despite the fact that the most energetic part of the tions provide an initial confirmation of findings by Shabani et al.
spectrum is better represented, the energy is still underestimated at (2016) where the wind input rate function was found to follow a
peak frequencies in some scenarios. The improvements generated by cos3.6 distribution. These narrower distributions also conserve the
𝐜𝐨𝐬𝟒 +bim are not necessarily symmetric along the wind direction, and momentum balance via the wave-supported stress 𝜏𝑤 .
the combined experiment shows a better fit in directions lower than • A simplified anisotropic distribution for the dissipation rate was
the wind direction while negative RMSE differences indicate a worse also tested in the WWIII model. The bim experiment reflects the
performance above the wind direction. This type of behavior reinforces expected behavior of the new implementation through smaller
the need to compare the directional wave spectra rather than the wave energy losses around the main wave direction under the wind-
integral parameters because they mask some features that can properly sea conditions. This implementation improves the estimation of
describe the sea state. the wave parameters and 2D spectra, it aligns with the field ex-
It is worth mentioning that the open boundary conditions could periments and distributions made by Young and Babanin (2006)
improve the model performance for the most eastern buoys (42059 and Babanin et al. (2010), and it could provide certain flexibility
and 42060) based on the possible existence of swell approaching the to modify the shape of the dissipation rate in the direction space.
Caribbean Sea through the Greater and Lesser Antilles (Jury, 2017; de • The joint effect of the directional distribution of the wind input
Farias et al., 2012). On the other hand, it is also remarkable that the and dissipation term was analyzed. The 𝐜𝐨𝐬𝟒 +bim combination
measured wind speeds and the interpolated data from ERA5 at the buoy outperforms the control and individual tests in most of the sim-
locations are reasonably similar in magnitude and variability. However, ulation period for 𝐻𝑠 at buoy 42058. Generally, the changes
a spatial validation of the wind fields should be done (beyond of the reflect an increase in wave energy at all the buoys, as is expected
scope of this study). under the wind-sea dominance in the Caribbean Sea. Although
These results show that these modifications can be applicable in these increments also led to higher overestimation during pe-
regions where the wind-sea dominates the wave energy and where riods when the ctrl experiment also overestimates the energy,
there are strong alignments between the wind and wave direction. The the implementation of these new directional distributions seems
conditions where the wind and waves are mostly aligned or differ from promising.
each other by less than 30–40◦ can be found in other enclosed basins • All the experiments exhibit similar spectral narrowness 𝑄𝑐 values
(e.g. Arabian Gulf Langodan et al., 2023 and the South China Sea Qian and high-frequency tails in the nondirectional spectra. That is to
et al., 2019). The novel directional spreading functions described here say, they generally keep the spectral shape of the 1D spectra. The
can therefore initially improve the model performance to estimate the experiments show increases in 𝑇𝑝 values compared to those mod-
directional wave spectra in these areas. The amplification of the wave eled with the ctrl experiment, generating a slight downshifting of
energy created by a narrower spreading of the wind input and lower the spectral peak.

11
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

The findings presented here suggest that future research should Benoit, M., 1994. Extensive comparison of directional wave analysis methods from
identify the impact of the aforementioned directional distributions gauge array data. pp. 740–754.
Çalışır, E., Soran, M.B., Akpınar, A., 2021. Quality of the ERA5 and CFSR winds and
through additional simulations under more complex sea states. In this
their contribution to wave modelling performance in a semi-closed sea. J. Oper.
sense, the following topics could be addressed: Ocean. 16 (2), 106–130. http://dx.doi.org/10.1080/1755876X.2021.1911126.
Chawla, A., Tolman, H., 2007. Automated Grid Generation for WAVEWATCH III.
• Incorporating the bimodal distribution as a function of the wind Technical Note 254, NOAA/NWS/NCEP/OMB, p. 71.
speed and wave frequency, as originally proposed by Babanin de Farias, E.G.G., Lorenzzetti, J.A., Chapron, B., 2012. Swell and wind-sea distributions
et al. (2010), could introduce a time dependence of the directional over the mid-latitude and tropical North Atlantic for the period 2002–2008. In:
Behera, S. (Ed.), Int. J. Ocean. 2012, 306723. http://dx.doi.org/10.1155/2012/
spreading behavior, allowing flexibility in the energy loss at 306723.
oblique angles during the simulation period. Donelan, M., Babanin, A., Sanina, E., Chalikov, D., 2015. A comparison of methods for
• Carrying out further tests in severe weather conditions (i.e. trop- estimating directional spectra of surface waves. J. Geophys. Res.: Ocean. 120 (7),
ical cyclones) could evaluate the new directional distributions 5040–5053. http://dx.doi.org/10.1002/2015JC010808.
Donelan, M.A., Babanin, A.V., Young, I.R., Banner, M.L., 2006. Wave-follower field
under adverse winds and analyze the presence of swell systems
measurements of the wind-input spectral function. Part II: Parameterization of
as well as the influence of non-linear interactions. the wind input. J. Phys. Oceanogr. 36 (8), 1672–1689. http://dx.doi.org/10.1175/
JPO2933.1.
CRediT authorship contribution statement Earle, M., Steele, K., Wang, D., 1999. Use of advanced directional wave spectra
analysis methods. Ocean Eng. 26 (12), 1421–1434. http://dx.doi.org/10.1016/
S0029-8018(99)00010-4.
Franklin F. Ayala: Writing – original draft, Visualization, Soft- Fan, Y., Ginis, I., Hara, T., Wright, C.W., Walsh, E.J., 2009. Numerical simulations and
ware, Methodology, Investigation, Data curation. Rubén D. Montoya: observations of surface wave fields under an extreme tropical cyclone. J. Phys.
Writing – review & editing, Validation, Supervision, Formal analysis, Oceanogr. 39 (9), 2097–2116. http://dx.doi.org/10.1175/2009JPO4224.1.
Conceptualization. Andrés F. Osorio: Writing – review & editing, Fan, Y., Rogers, W.E., 2016. Drag coefficient comparisons between observed and model
simulated directional wave spectra under hurricane conditions. Ocean. Model. 102,
Supervision, Project administration. Alexander Babanin: Writing –
1–13. http://dx.doi.org/10.1016/j.ocemod.2016.04.004.
review & editing, Supervision, Resources, Methodology. Hasselmann, K., 1974. On the spectral dissipation of ocean waves due to white capping.
Bound.-Layer Meteorol. 6 (1), 107–127. http://dx.doi.org/10.1007/BF00232479.
Declaration of competing interest Hasselmann, S., Hasselmann, K., Allender, J.H., Barnett, T.P., 1985. Computations and
parametrizations of the nonlinear energy transfer in a gravity-wave spectrum. Part
II: Parametrizations of the nonlinear energy transfer for application in wave models.
The authors declare the following financial interests/personal rela- J. Phys. Oceanogr. 15, 1378–1391. http://dx.doi.org/10.1175/1520-0485(1985)
tionships which may be considered as potential competing interests: 015<1378:CAPOTN>2.0.CO;2.
Two co-authors are guest editors of the special issue to which this paper Hersbach, H., Bell, B., Berrisford, P., Horányi, A., Sabater, J.M., Nicolas, J., Radu, R.,
is sent: Andrés F. Osorio and Alexander Babanin. Schepers, D., Simmons, A., Soci, C., Dee, D., 2019. Global reanalysis: goodbye
ERA-Interim, hello ERA5. ECMWF Newsl. 159, 17–24. http://dx.doi.org/10.21957/
If there are other authors, they declare that they have no known
vf291hehd7.
competing financial interests or personal relationships that could have Holthuijsen, L.H., 2007. Waves in Oceanic and Coastal Waters. Cambridge University
appeared to influence the work reported in this paper. Press.
Hsiao, S.V., Shemdim, O.H., 1983. Measurements of wind velocity and pressure with
Acknowledgment a wave follower during MARSEN. J. Geophys. Res.: Ocean. 88 (C14), 9841–9849.
http://dx.doi.org/10.1029/JC088iC14p09841.
Janssen, P.A.E.M., 2004. The Interaction of Ocean Waves and Wind. Cambridge
The authors are thankful to Dr. Qingxiang Liu from the Ocean Uni- University Press.
versity of China for valuable discussions. Franklin F. Ayala is grateful Jury, M., 2017. Characteristics and meteorology of atlantic swells reaching the
to the Agencia de Educación Postsecundaria de Medellín. Andrés F. caribbean. J. Coast. Res. 34 (2), 400–412. http://dx.doi.org/10.2112/JCOASTRES-
D-17-00029.1.
Osorio acknowledges the Ministerio de Ciencia, Tecnología e Inno-
Komen, G.J., Cavaleri, L., Donelan, M., Hasselmann, K., Hasselmann, S.,
vación (Minciencias), through the Fondo Nacional de Financiamiento Janssen, P.A.E.M., 1994. Dynamics and Modelling of Ocean Waves. Cambridge
para la Ciencia, la Tecnología y la Innovación Francisco José de Caldas University Press.
(80740-538-2020). Langodan, S., Cavaleri, L., Benetazzo, A., Bertotti, L., Dasari, H.P., Hoteit, I., 2023. The
Funding: This research did not receive any specific grant from peculiar wind and wave climatology of the Arabian Gulf. Ocean Eng. 290, 116158.
http://dx.doi.org/10.1016/j.oceaneng.2023.116158.
funding agencies in the public, commercial, or not-for-profit sectors. Leonard, B., 1991. The ULTIMATE conservative difference scheme applied to unsteady
one-dimensional advection. Comput. Methods Appl. Mech. Engrg. 88 (1), 17–74.
Data availability http://dx.doi.org/10.1016/0045-7825(91)90232-U.
Lin, Z., Adcock, T.A., McAllister, M.L., 2022. Estimating ocean wave directional
spreading using wave following buoys: a comparison of experimental buoy and
Data will be made available on request.
gauge data. J. Ocean. Eng. Mar. Energy 8, 83–97. http://dx.doi.org/10.1007/
s40722-021-00218-7.
Liu, Q., Babanin, A., Fan, Y., Zieger, S., 2017. Numerical simulations of ocean surface
References waves under hurricane conditions: assessment of existing model permormance.
Ocean. Model. 118, 73–93. http://dx.doi.org/10.1016/j.ocemod.2017.08.005.
Amante, C., Eakins, B., 2009. ETOPO1 global relief model converted to PanMap layer Liu, Q., Babanin, A.V., Rogers, W.E., Zieger, S., Young, I.R., Bidlot, J.R., Durrant, T.,
format [dataset]. http://dx.doi.org/10.1594/PANGAEA.769615. Ewans, K., Guan, C., Kirezci, C., Lemos, G., MacHutchon, K., Moon, I.J., Rapizo, H.,
Ardhuin, F., Boyer, A.L., 2006. Numerical modelling of sea states: valida- tion of Ribal, A., Semedo, A., Wang, J., 2021. Global wave hindcasts using the observation-
spectral shapes (in French). Annu. Navig. 54, 55–71. based source terms: Description and validation. J. Adv. Model. Earth Syst. 13 (8),
Ardhuin, F., Rogers, E., Babanin, A.V., Filipot, J.F., Magne, R., Roland, A., van der e2021MS002493. http://dx.doi.org/10.1029/2021MS002493.
Westhuysen, A., Queffeulou, P., Lefevre, J.M., Aouf, L., Collard, F., 2010. Semiem- Liu, Q., Rogers, W.E., Babanin, A.V., Young, I.R., Romero, L., Zieger, S., Qiao, F.,
pirical dissipation source functions for ocean waves. Part I: Definition, calibration, Guan, C., 2019. Observation-based source terms in the third-generation wave
and validation. J. Phys. Oceanogr. 40 (9), 1917–1941. http://dx.doi.org/10.1175/ model WAVEWATCH III: Updates and verification. J. Phys. Oceanogr. 49, 489–517.
2010JPO4324.1. http://dx.doi.org/10.1175/JPO-D-18-0137.1.
Babanin, A., 2009. Breaking of ocean surface waves. Acta Phys. Slovaca 59, 305–535. Longuet-Higgins, M.S., Cartwrigth, D.E., Smith, N.D., 1963. Observations of the direc-
http://dx.doi.org/10.2478/v10155-010-0097-5. tional spectrum of sea waves using the motions of a floating buoy. In: Ocean Wave
Babanin, A.V., Tsagareli, K.N., Young, I.R., Walker, D.J., 2010. Numerical investigation Spectra. Prentice Hall, New Jersey, USA, pp. 111–136.
of spectral evolution of wind waves. Part II: Dissipation term and evolution tests. Miles, J.W., 1957. On the generation of surface waves by shear flows. J. Fluid Mech.
J. Phys. Oceanogr. 40 (4), 667–683. http://dx.doi.org/10.1175/2009JPO4370.1. 3, 185–204. http://dx.doi.org/10.1017/S0022112057000567.
Battjes, J.A., Janssen, J., 1978. Energy loss and wave set-up due to breaking of Montoya, R.D., Arias, A.O., Royero, J.C., Ocampo-Torres, F.J., 2013. A wave parameters
random waves. In: Proc. 16th Congerence on Coastal Engineering. ASCE, Hamburg, and directional spectrum analysis for extreme winds. Ocean Eng. 67, 100–118.
Germany, pp. 569–587. http://dx.doi.org/10.1061/9780872621909.034. http://dx.doi.org/10.1016/j.oceaneng.2013.04.016.

12
F.F. Ayala et al. Ocean Modelling 196 (2025) 102542

Ortiz-Royero, J.C., Otero, L.J., Restrepo, J.C., Ruiz, J., Cadena, M., 2013. Cold fronts in Shabani, B., Babanin, A.V., Baldock, T.E., 2016. Observations of the directional
the Colombian Caribbean Sea and their relationship to extreme wave events. Nat. distribution of the wind energy input function over swell waves. J. Geophys. Res.:
Hazards Earth Syst. Sci. 13 (11), 2797–2804. http://dx.doi.org/10.5194/nhess-13- Ocean. 121 (2), 1174–1193. http://dx.doi.org/10.1002/2015JC011225.
2797-2013. Snyder, R.L., Dobson, F.W., Elliott, J.A., Long, R.B., 1981. Array measurements of
Plant, W.J., 1982. A relationship between wind stress and wave slope. J. Geophys. Res. atmospheric pressure fluctuations above surface gravity waves. J. Fluid Mech. 102,
87 (C3), 1961–1967. http://dx.doi.org/10.1029/JC087iC03p01961. 1–59. http://dx.doi.org/10.1017/S0022112081002528.
Portilla-Yandún, J., Cavaleri, L., Van Vledder, G.P., 2015. Wave spectra partitioning Stopa, J.E., Ardhuin, F., Babanin, A., Zieger, S., 2016. Comparison and validation of
and long term statistical distribution. Ocean. Model. 96, 148–160. http://dx.doi. physical wave parameterizations in spectral wave models. Ocean. Model. 103, 2–17.
org/10.1016/j.ocemod.2015.06.008. http://dx.doi.org/10.1016/j.ocemod.2015.09.003.
Qian, C., Jiang, H., Wang, X., Ge, C., 2019. Climatology of wind-seas and swells Stopa, J.E., Cheung, K.F., Tolman, H.L., Chawla, A., 2013. Patterns and cycles in the
in the China seas from wave hindcast. J. Ocean. Univ. China 19, 90–100. http: climate forecast system reanalysis wind and wave data. Ocean. Model. 70, 207–220.
//dx.doi.org/10.1007/s11802-020-3924-4. http://dx.doi.org/10.1016/j.ocemod.2012.10.005.
Rascle, N., Ardhuin, F., 2013. A global wave parameter database for geophysical Tolman, H.L., Chalikov, D., 1996. Source terms in a third-generation wind wave
applications. Part 2: Model validation with improved source term parameterization. model. J. Phys. Oceanogr. 26 (11), 2497–2518. http://dx.doi.org/10.1175/1520-
Ocean. Model. 70, 174–188. http://dx.doi.org/10.1016/j.ocemod.2012.12.001. 0485(1996)026<2497:STIATG>2.0.CO;2.
Rogers, W.E., Babanin, A.V., Wang, D.W., 2012. Observation-consistent input and Tsagareli, K., 2009. Numerical Investigation of Wind Input and Spectral Dissipation in
whitecapping dissipation in a model for wind-generated surface waves: Description Evolution of Wind Waves (Ph.D. thesis). Univeristy of Adelaide, Adelaide, Australia.
and simple calculations. J. Atmos. Ocean. Technol. 29 (9), 1329–1346. http: Wang, C., 2007. Variability of the Caribbean low-level jet and its relations to climate.
//dx.doi.org/10.1175/JTECH-D-11-00092.1. Clim. Dyn. 29, 411–422. http://dx.doi.org/10.1007/s00382-007-0243-z.
Rogers, W.E., Vledder, G.P.V., 2013. Frequency width in predictions of windsea spectra The WAVEWATCH III Development Group (WW3DG), 2019. User Manual and Sys-
and the role of the nonlinear solver. Ocean. Model. 70, 52–61. http://dx.doi.org/ tem Documentation of WAVEWATCH III Version 6.07. Technical Note 333,
10.1016/J.OCEMOD.2012.11.010. NOAA/NWS/NCEP/MMAB, College Park, MD, USA, p. 465.
Romero, L., 2019. Distribution of surface wave breaking fronts. Geophys. Res. Lett. 46 Yang, H., Shao, Z., Liang, B., Wang, Z., Lee, D., 2022. Performance of different input
(17–18), 10463–10474. http://dx.doi.org/10.1029/2019GL083408. and dissipation packages in WAVEWATCH III model during tropical cyclones. Phys.
Romero, L., Lubana, K., 2022. On the bimodality of the wind-wave spectrum: Fluids 34 (10), 107102. http://dx.doi.org/10.1063/5.0120059.
Mean square slopes and azimuthal overlap integral. J. Phys. Oceanogr. 52 (7), Young, I., 1999. Wind generated ocean waves. Elsevier Ocean Engineering Series,
1549–1562. http://dx.doi.org/10.1175/JPO-D-21-0299.1. Elsevier Science.
Romero, L., Melville, W.K., 2010. Airborne observations of fetch-limited waves in the Young, I.R., Babanin, A.V., 2006. Spectral distribution of energy dissipation of wind-
gulf of tehuantepec. J. Phys. Oceanogr. 40 (3), 441–465. http://dx.doi.org/10. generated waves due to dominant wave breaking. J. Phys. Oceanogr. 36 (3),
1175/2009JPO4127.1. 376–394. http://dx.doi.org/10.1175/JPO2859.1.
Ruiz-Ochoa, M.A., Bernal, G., 2009. Seasonal and interannual wind variability into Zhang, S., Zhang, J., 2006. A new approach to estimate directional spreading pa-
the NCEP/NCAR reanalysis data on the Colombian basin, Caribbean sea (in rameters of a cosine-2s model. J. Atmos. Ocean. Technol. 23 (2), 287–301. http:
Spanish). Av. Recur. Hidráulicos 7–20, URL: https://revistas.unal.edu.co/index. //dx.doi.org/10.1175/JTECH1846.1.
php/arh/article/view/14328. Zieger, S., Babanin, A.V., Rogers, W.E., Young, I.R., 2015. Observation-based source
terms in the third-generation wave model WAVEWATCH. Ocean. Model. 96, 2–25.
http://dx.doi.org/10.1016/j.ocemod.2015.07.014.

13