the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Wave boundary layer model in SWAN revisited
Rodolfo Bolaños
Xiaoli Guo Larsén
Mark Kelly
In this study, we extend the work presented in Du et al. (2017) to make the wave boundary layer model (WBLM) applicable for real cases by improving the windinput and whitecapping dissipation source functions. Improvement via the new source terms includes three aspects. First, the WBLM windinput source function is developed by considering the impact of waveinduced wind profile variation on the estimation of wave growth rate. Second, the whitecapping dissipation source function is revised to be not explicitly dependent on wind speed for real wave simulations. Third, several improvements are made to the numerical WBLM algorithm, which increase the model's numerical stability and computational efficiency. The improved WBLM windinput and whitecapping dissipation source functions are calibrated through idealized fetchlimited and depthlimited studies, and validated in real wave simulations during two North Sea storms. The new WBLM source terms show better performance in the simulation of significant wave height and mean wave period than the original source terms.
The accuracy of spectral ocean wave models depends on the forcing from wind, water level, currents, etc. It also depends on the source terms and numerical methods (Ardhuin, 2012). In deep water conditions, the source terms are reduced to the windinput source function (S_{in}), wavebreaking dissipation source function (S_{ds}), and nonlinear fourwaveinteraction source function (S_{nl}). In a previous study (Du et al., 2017), a wave boundary layer model (WBLM) was implemented in the thirdgeneration Simulating WAves Nearshore (SWAN) ocean wave model (Booij et al., 1999) to improve the windinput source function of Janssen (1991, hereafter JANS); this was done by considering the momentum and kinetic energy conservation at each level in the wave boundary layer. It was shown that the new S_{in} improves wave simulations in idealized fetchlimited conditions. Because the evolution of wave spectra depends on the difference between source and sink terms, the change of S_{in} has to be followed by the tuning of the parameters in S_{ds} (Cavaleri, 2009). Du et al. (2017) simply recalibrated the whitecapping dissipation parameters of Komen et al. (1984, hereafter KOM) to be proportional to the WBLM S_{in} (Babanin et al., 2010) and wind speed at 10 m (U_{10}) (Melville and Matusov, 2002). Such a method works in idealized fetchlimited conditions when the winds do not change over time. However, in real cases, wind speed and direction vary in time. Also, wave breaking is related to wave properties such as wave steepness, rather than explicitly dependent on wind speed (e.g., Komen et al., 1994). Moreover, in coastal areas, the bottom friction and depthinduced breaking dissipation become important and they influence the shape of wave spectra. Consequently, S_{in} and S_{ds} are also modified by the shallowwater effect. Therefore, the description of the new S_{in} and S_{ds} in shallow water also needs to be investigated before they are used in real simulations.
Theoretical models of wavebreaking dissipation have been extensively reviewed by Komen et al. (1994), Young and Babanin (2006a), and Cavaleri et al. (2007), and can be classified into whitecapping models (Hasselmann, 1974), saturationbased models (e.g., Phillips, 1985), probabilistic models (e.g., LonguetHiggins, 1969; Yuan et al., 1986; Hua and Yuan, 1992), and turbulent models (Polnikov, 1993). Among them, whitecapping and saturationbased models are widely used in ocean wave models such as WAM (Komen et al., 1994), Simulating WAves Nearshore (SWAN) (Booij et al., 1999), WAVEWATCH III (Tolman and Chalikov, 1996), and MIKE 21 SW (Sørensen et al., 2005). Whitecapping models consider the effect of downwardmoving whitecaps doing work against the upwardmoving waves. Parameterization of whitecapping dissipation can be found in, e.g., Komen et al. (1984), Bidlot et al. (2007), and Bidlot (2012); the dissipation at all frequencies is taken to be proportional to the mean wave steepness defined by a mean wave number and the significant wave height. The saturationbased models assume saturation exists in the equilibrium range of the wave spectrum, and the dissipation rate is proportional to the saturation at any given frequency. Therefore, the dissipation at each frequency is proportional to the local wave steepness or local saturation. Latter studies, however, suggest a twophase behavior of wavebreaking dissipation (Babanin and Young, 2005; Young and Babanin, 2006a): S_{ds} should be a function of the spectral peak plus a cumulative frequencyintegrated term at higher frequencies due to dominant wave breaking. Considering the complexity of wavebreaking processes, recent studies tend to combine the two types of S_{ds} together. Alves and Banner (2003) and van der Westhuysen et al. (2007) used a saturationbased model multiplied by a KOMshaped model to account for the longwave–shortwave and wave–turbulence interactions. Banner and Morison (2010) introduced a breaking probability function to the saturationbased model of Phillips (1985). Ardhuin et al. (2010), Babanin et al. (2010), and Zieger et al. (2015) added a cumulative term to a saturationbased model. Such combined S_{ds} values are proven to be robust in wave simulations, globally to coastal areas (Ardhuin et al., 2012; Ardhuin and Roland, 2012; Leckler et al., 2013).
However, as more physical processes are being taken into account, expressions of S_{ds} become more complex and need more tuning parameters; e.g., the S_{ds} of Ardhuin et al. (2010) needs up to 18 parameters, which makes it difficult to adjust when there is modification of other source terms. The present study aims at finding a proper dissipation source function that is suitable for the new WBLM S_{in}. Therefore, instead of introducing more physics into S_{ds}, numerical adjustment is applied to the KOM dissipation (Komen et al., 1984). The reason that we chose the KOM S_{ds} is that it has been shown to be successful when used with different windinput source functions in SWAN (Snyder et al., 1981; Komen et al., 1984; Janssen, 1991; Larsén et al., 2017), and because the formulation is so flexible that its total magnitude and spectral distribution can be easily tuned with C_{ds} and Δ in Eq. (10). Du et al. (2017) has shown that numerical adjustment to the KOM S_{ds} can be used for the WBLM S_{in}, to reproduce the fetchlimited wave growth curve of Kahma and Calkoen (1992). Moreover, Ardhuin (2012) showed that S_{ds} of the KOM type and saturationbased type (Phillips, 1985) can be adjusted to give very similar behavior. However, we found that using only the KOM S_{ds} within the WBLM produces a toohigh energy level at frequencies higher than the spectral peak (f>f_{p}), and this problem can be solved by using a cumulative dissipation term according to Ardhuin et al. (2010).
In this paper, the improvement of WBLM S_{in}, the revised S_{ds}, and the numerical algorithm changes to the model are presented in Sect. 2. Then, the new pair of S_{in} and S_{ds} is calibrated in idealized fetchlimited and depthlimited studies, and validated in realcase storm simulations in the North Sea. These numerical experiments are described in Sect. 3, and the results are presented in Sect. 4. Wave parameters such as significant wave height, mean period, peak wave period, and spectral shape are validated using point measurements of deep and shallow waters. S_{in} and S_{ds} of KOM and JANS are also examined as a benchmark reference for these storms. As mentioned before, wave growth depends on the difference between source and sink terms as well as numerical discretization methods, especially in real cases. Therefore, it is difficult to distinguish whether an improvement of a specific wave parameter, such as significant wave height, is due to the improvement of S_{in} or S_{ds}. An evaluation of the WBLM S_{in} was presented in Du et al. (2017) in an idealized fetchlimited wave growth study, where the wave growth rate and the drag coefficient calculated from S_{in} are compared with measurements. However, this paper mainly focuses on the application of WBLM for real applications, so that the overall effect of the new WBLM S_{in} and S_{ds} is emphasized.
2.1 The windinput source function
According to Du et al. (2017), the growth rate (β_{g}) of the WBLM windinput source function $\left({S}_{\text{in}}={\mathit{\beta}}_{\mathrm{g}}\left(\mathit{\sigma},\mathit{\theta}\right)N\left(\mathit{\sigma},\mathit{\theta}\right)\right)$ is expressed as
where N(σ,θ) is the action density spectrum, θ and θ_{w} are the wave and wind directions, C_{β} is the Miles parameter (Miles, 1957), ρ_{w} is the water density, and c is the phase velocity of waves. τ_{t}(z) is the local turbulent stress, which is equal to the total stress, τ_{tot}, minus the waveinduced stress, τ_{w}(z). The Miles parameter C_{β} is described as a function of the nondimensional critical height, λ:
where κ=0.41 is the von Kármán constant, and J=1.6 is a constant. In Du et al. (2017), the expression of the nondimensional critical height λ for Miles' parameter (Eq. 2) is derived by the assumption of a logarithmic wind profile following Janssen (1991), and it is expressed as
where g is the gravity acceleration, α=0.008 is a wave age tuning parameter according to Bidlot (2012), z_{0} is the roughness length. However, it is found that using Eq. (3) causes numerical instability in some cases. This is because, within the WBL, the wind profile is not logarithmic under the impact of waves (Du et al., 2017). Using a logarithmic wind profile (Eq. 3) not only slows down the computation but could also fail in converging in some cases. Therefore, when applying WBLM S_{in}, the expression of λ also needs to be changed to adjust to the new wind profile. Here, we follow Miles (1957)'s procedure to drive an approximate expression for λ. In Miles (1957), the nondimensional critical height is defined as
where k is the wave number, z_{c} is the critical height where the wave phase velocity (c) equals the wind speed component in the phase velocity direction $u\left({z}_{\mathrm{c}}\right)\cdot \mathrm{cos}\left(\mathit{\theta}{\mathit{\theta}}_{\mathrm{w}}\right)$,
where θ−θ_{w} is the angular separation between wind and wave directions. We assume that, in the vicinity of the critical height (z_{c}), the wind profile can be approximately described as locally logarithmic:
where ${u}_{*}^{\mathrm{l}}=\sqrt{{\mathit{\tau}}_{\mathrm{t}}/{\mathit{\rho}}_{\mathrm{a}}}$ is the local friction velocity. In the vicinity of the critical height, wind speed at any other heights z can be expressed as
where ${z}_{\mathrm{0}}^{\mathrm{l}}$ is a local effective roughness. Introducing Eqs. (7)–(5), we have wind speed at the critical height:
The critical height is calculated by combining Eqs. (7) and (8):
Considering the shallowwater dispersion relation, $k=(g/{c}^{\mathrm{2}})\mathrm{tanh}\left(kh\right)$, with h the water depth, the combination of Eqs. (4) and (9) results in the nondimensional critical height for any direction:
It is found that Eq. (10) tends to underestimate wave growth at low frequencies. We used the same method as that used in WAM (Bidlot, 2012) and added a wave age tuning parameter (α=0.011) to shift the wave growth towards lower frequencies:
2.2 Whitecapping dissipation source function
The whitecapping dissipation expression of KOM (Komen et al., 1984; Janssen, 1991; Bidlot et al., 2007) in SWAN is written as
where $\mathit{\varphi}\left(\mathit{\sigma},\mathit{\theta}\right)=\mathit{\sigma}N\left(\mathit{\sigma},\mathit{\theta}\right)$ is the frequency spectra. Since the WBLM windinput and dissipation source functions are designed mainly for the wind sea, it is necessary to reduce the swell impact to the whitecapping dissipation. In this study, the mean radian frequency 〈σ〉 and mean wave number 〈k〉 are modified according to Bidlot et al. (2007) to put more emphasis on the high frequencies:
where ${m}_{\mathrm{0}}=\int \int \mathit{\varphi}\left(\mathit{\sigma},\mathit{\theta}\right)\text{d}\mathit{\theta}\text{d}\mathit{\sigma}$ is the variance of the sea surface elevation. The choice of the two dissipation parameters, C_{ds} and Δ, is different for different windinput source functions. For example, for KOM S_{in}, C_{ds}=2.5876 and Δ=1; for JANS S_{in}, C_{ds}=4.5, and Δ=0.5; for WBLM S_{in} in Du et al. (2017), Δ=0.1, and C_{ds} in S_{ds} is related to S_{in} to make sure
where
where U_{10} is wind speed at 10 m, and $\stackrel{\mathrm{\u0303}}{E}={m}_{\mathrm{0}}{g}^{\mathrm{2}}/{U}_{\mathrm{10}}^{\mathrm{4}}$ is a nondimensional energy. As discussed in the introduction, a dissipation parameter that is strongly dependent on wind speed as in Eq. (12) only works in idealized fetchlimited cases but will in principle not work in real cases because wave breaking depends on wave state rather than wind. Here, we explore the use of some wave parameters to replace U_{10} and S_{in} in Eqs. (11) and (12) to get rid of the direct dependence on wind speed. We derive the relationship between U_{10}, m_{0}, peak frequency (f_{p}), and fetch (x) from the three nondimensional parameters, namely nondimensional energy ($\stackrel{\mathrm{\u0303}}{E}$), nondimensional peak frequency (${\stackrel{\mathrm{\u0303}}{F}}_{\mathrm{p}}={f}_{\mathrm{p}}{U}_{\mathrm{10}}/g$), and nondimensional fetch ($\stackrel{\mathrm{\u0303}}{x}=xg/{U}_{\mathrm{10}}^{\mathrm{2}}$). The fetch dependence of $\stackrel{\mathrm{\u0303}}{E}$ and ${\stackrel{\mathrm{\u0303}}{F}}_{\mathrm{p}}$ can be written as
where, in Kahma and Calkoen (1992) (composite condition), $A=\mathrm{5.2}\times {\mathrm{10}}^{\mathrm{7}}$, B=0.9, C=2.1804, $D=\mathrm{0.27}$. According to Eq. (12), U_{10}, $\stackrel{\mathrm{\u0303}}{E}$, and $\stackrel{\mathrm{\u0303}}{x}$ can be expressed as functions of m_{0}, f_{p}, and g:
where U_{10}, $\stackrel{\mathrm{\u0303}}{E}$, and $\stackrel{\mathrm{\u0303}}{x}$ are replaced by U^{′}, E^{′}, and x^{′} since they are parameterized variables. The dissipation coefficient C_{ds} in Eq. (10) can be obtained by fitting the C_{ds} calculated from Eqs. (11) and (12) with U^{′} and x^{′} from Eq. (13):
The form and parameters in Eq. (14) will be obtained in the fetchlimited study in Sect. 4.1. To increase the robustness of the wave modeling in the case of unusualshaped spectra, the peak frequency f_{p} in Eq. (13) is replaced by 0.866〈f〉 according to Komen et al. (1994) who uses k_{p}=0.75〈k〉, where $\u2329f\u232a=\u2329\mathit{\sigma}\u232a/\mathrm{2}\mathit{\pi}$ is the mean frequency. The use of 〈f〉 may cause uncertainty in estimating f_{p}, especially in, e.g., bimodal wave cases. Considering the model performs quite well during the two storm simulations in Sect. 4.3, during which bimodal waves probably exist, we assume that this uncertainty is relatively small.
To reduce the energy level at high frequencies, a cumulative term is added to the dissipation source functions. The cumulative dissipation term (${S}_{\mathrm{ds}}^{\text{c}}$) follows Ardhuin et al. (2010), but the directional dependence of dissipation rate is not considered:
where C_{cu}=1.0 is a dissipation parameter, B_{r}=0.0012 is a saturation threshold, r_{cu}=0.5 is the ratio of the maximum frequency where dissipation of long waves influences short waves, C_{g} is the group velocity, and B(f) is the local saturation (van der Westhuysen et al., 2007):
2.3 Improvement of the numerical algorithm
Considering the expensive cost of WBLM code in Du et al. (2017), improvement of the numerical algorithm of the WBLM (Du et al., 2017) was done in the following aspects.

The unnecessary calculations in the high frequencies were reduced. The WBLM uses 10 Hz as the maximum frequency, which is only being used for very young waves. Usually, the WBLM does not have to solve such high frequencies when the energy is so small that their contribution to the total waveinduced stress is negligible. Therefore, in the new code, the WBLM only solves the active frequency range which is dynamically changing with the wave spectrum. Although the maximum frequency is dynamically changing, all the active frequencies are solved, so there is no influence on the result. Such an adjustment reduces approximately half of the computation time in the idealized fetchlimited study in Sect. 3.1.

The standard calculation in SWAN, a sweeping technique is used for the directional propagation of the waves, needs to be swept four times for each time step. Such a sweep is not necessary for the calculation of WBLM because the WBLM has to integrate over all directions of the spectra. Therefore, WBLM only calculates once per time step.
With the above mentioned refinement, the WBLM is now about 5 times faster than the previous version in Du et al. (2017). It is still slower than KOM and JANS, but it is fast enough for real applications. The averaged calculation time during an idealized fetchlimited study is listed in Table 1, including KOM, JANS, the previous WBLM of Du et al. (2017) (WBM Old), and the new WBLM in this paper. All the setups are the same as those in Sect. 3.1, except JANS uses frequency number ranges from 38 to 54 for different wind speed conditions, while the other schemes use 73 frequencies.
3.1 Idealized fetchlimited study
The revised dissipation parameter (Eq. 14) in S_{ds} together with the new nondimensional critical height as in WBLM S_{in} were first calibrated in the idealized fetchlimited wave growth experiments with the same model setup as in Du et al. (2017). Here, we briefly describe the model setup. We use the onedimensional SWAN. The fetch is between 0 and 3000 km, with the spatial resolution changing gradually from 0.1 to 10 km. The wave spectrum ranges from 0.01 to 10.51 Hz, and the frequency discretization was logarithmic with a frequency exponent of 1.1, which results in 73 frequencies. We use 36 directional bins with a constant spacing of 10^{∘}. SWAN initializes from zero spectrum and runs for 72 h with a time step of 1 min. U_{10} ranges from 5 to 40 ms^{−1} and keeps constant during the 72 h simulation.
The calibration process is carried out in three steps. In Step 1, we run the model using the whitecapping dissipation parameter as in Du et al. (2017) (Eqs. 11 and 12) in the idealized fetchlimited study. Since we added a cumulative dissipation source term, the parameters in Eq. (12) have to be changed into Eq. (16) so as to best fit to the Kahma and Calkoen (1992) fetchlimited wave growth curves:
In Step 2, the real form and parameters in Eq. (14) will be obtained by analyzing and fitting the C_{ds} with x^{′} and U^{′}, which are calculated in Step 1. Finally, the best fit in Step 2 may not be the best choice for the wave model. Therefore, we selected several representative fitting parameters out of dozens of groups through idealized fetchlimited study as the final choice.
3.2 Idealized depthlimited study
In addition to the idealized fetchlimited study, the revised WBLM source terms are also tested in depthlimited wave growth experiments to check if the new source terms perform well with the interaction of the other source terms in the wave model, including the bottom friction and depthinduced wavebreaking dissipation source functions. Following SWAN (2014), we use JONSWAP (Hasselmann et al., 1973) bottom friction with a bottom friction coefficient, C_{b}=0.038 m^{2} s^{−3}, and Battjes and Janssen (1978) depthinduced wavebreaking source function with a breaker parameter, γ=0.8.
In the depthlimited wave growth experiments, we take the measurements of Young and Babanin (2006b) as reference for validation, because they not only provided detailed wind, wave, and water depth information but also provided wave spectrum measurement from capacitance wave probes up to 10 Hz (Young et al., 2005). Zijlema et al. (2012) conducted similar depthlimited wave growth experiments for the calibration of the bottom friction parameter in SWAN, but they did not compare the wave spectrum. Zijlema et al. (2012) selected 10 representative cases from the data presented in Young and Babanin (2006b). We add three more cases because the wave spectra in these three cases are also presented in Young and Babanin (2006b). This ends up with 13 cases in all, which are numbers 1, 11, 17, 23, 28, 30, 32, 58, 61, 63, 82, 83, and 87 in Table 1 of Young and Babanin (2006b). Among them, numbers 1, 11, 28, 32, and 87 have wave spectrum records for model validation. In all 13 cases, the water depth ranges from 0.89 to 1.1 m, and the wind speed ranges from 5.7 to 15 ms^{−1}. The fetch is set to 20 km with a spatial resolution of 0.1 km to ensure the fetch is long enough, and the wave growth is limited by the water depth. The frequency and directional discretization are same as those in the fetchlimited study. SWAN initializes from zero spectrum and runs for 24 h (we found 24 h is long enough for the wave development in the shallowwater conditions in this study), with a time step of 1 min. Two pairs of S_{in} and S_{ds} are tested, namely KOM (Snyder et al., 1981; Komen et al., 1984) and the revised WBLM in this study.
3.3 Real case study in the North Sea
The new WBLM S_{in} and S_{ds} are validated during two winter storms in the North Sea. Wind and wave measurements are obtained during the “Reducing the uncertainty of nearshore wind estimations using wind lidars and mesoscale models” (RUNE) project (Floors et al., 2016b). Simultaneous wind and wave measurements from lidar and buoys are available from November 2015 to January 2016. The experiment was conducted on the west coast of Jutland, Denmark, with an average water depth of 16.5 m. Details about the wind and wave measurements can be found in Floors et al. (2016b, a, c), Bolaños (2016), and Bolaños and Rørbæk (2016). Besides the standard wave parameters such as significant wave heights, peak wave period, and mean wave periods, the twodimensional wave spectra E(f,θ) are also available, which allows us to validate more detailed aspects of the source functions. During the RUNE period, two storms passed the measurement site on 28 November 2015 and 12 August 2015. Wave simulations were done during this period with the three pairs of source terms (KOM, JANS, and WBLM). Besides the measurement from the RUNE project, point wave measurements at Fjaltring, Hanstholm, A121, Väderöarna, and Heligoland north from NOOS (https://noos.bsh.de/, last access: 15 June 2017) FINO1 and FINO3 from FINO (http://fino.bsh.de, last access: 3 April 2018) and SleipnerA and Ekofisk from eKlima (http://sharki.oslo.dnmi.no, last access: 17 November 2016) during the simulation period are also used for model validation. The locations of these measurement sites are shown in Fig. 1a, b, and c.
SWAN is forced by the National Centers for Environmental Prediction (NCEP) Climate Forecast System version 2 (CFSv2) 10 m wind (Saha et al., 2011). The quality of CFSv2 10 m wind speed and direction is evaluated with measurements in Fig. 6a–b, and it has been shown to be good quality for wave simulations in the North Sea in many previous studies (Bolaños et al., 2014). Therefore, the hourly CFSv2 data may be considered reasonable wind forcing, though it may not be accurate in the presence of highly fluctuating wind on scales smaller than 1 h, e.g., Larsén et al. (2017). SWAN uses three nested domains, with a resolution downscaling from 9 to 3 km and 600 m (Fig. 1a). Open boundaries for SWAN are set to 0. The shortest fetch of the observation points is larger than 1000 km according to the wind direction during the storm. According to the fetchlimited study in Sect. 4.1, the fetch is long enough for the waves to develop. The bottom friction and depthinduced wavebreaking source functions for the real case study are same as the depthlimited wave growth studies. Bathymetry data are obtained from the EMODnet digital terrain model (DTM) with a spatial resolution of 1∕8 arcmin. Note that most of the observations are located in the coastal areas in the North Sea, except SleipnerA, A121, and Ekofisk (Fig. 1b, c). The water depth of most sites is less than 30 m, except SleipnerA and Ekofisk, which are relatively deeper, with water depth around 80 and 60 m, respectively. SWAN initializes from zero spectrum, and the first 24 h output are not included in our analysis. The first storm peak is about 72 h after the model initializes to make sure the duration of the simulation is long enough. The frequency discretization was logarithmic with a frequency exponent of 1.1, and the lowest frequency was set to 0.03 Hz. For the KOM and WBLM source terms, a cutoff frequency of 10.05 Hz is used, giving a total of 61 frequencies; for JANS source terms, the cutoff frequency is set to 0.57 Hz to make sure the simulation is stable, giving a total of 31 frequencies. We used 36 directional bins and a 5 min time step. A summary of the constants and model setups used for all the experiments in Sect. 3 is given in Tables 2 and 3, respectively.
4.1 Idealized fetchlimited study
The significant wave height (H_{m0}) calculated from Step 1 of the idealized fetchlimited study using Eqs. (11) and (16) is shown in Fig. 2a. The Kahma and Calkoen (1992) wave growth curves are also shown as solid black lines for reference. Note that Kahma and Calkoen (1992) curves come from data with limited wind speeds (U_{10}≤25 ms^{−1}) and fetches (x≤300 km), and we linearly extend them to broader ranges. The H_{m0} calculated from Step 1 (solid colored lines in Fig. 2a) agrees with Kahma and Calkoen (1992) wave growth curves for fetches x≤10 km. But it tends to underestimate H_{m0} for fetches x>10 km. Therefore, in Step 2, we only fit the C_{ds} in the first 10 km and extend its application for longer fetches. The C_{ds} calculated from Step 1 for wind speed U_{10}=5 ms^{−1} and U_{10}=20 ms^{−1} is shown in Fig. 2b as black circles and black rectangles. Here, we try to speculate the form of Eq. (14) based on the distribution of C_{ds} from Step 1 in Fig. 2b. First, C_{ds} has a clear dependence on U_{10}. We found that in 10 ms^{−1} wind speed conditions, the simulated H_{m0} follows the Kahma and Calkoen (1992) curves quite well in all fetches, while in the other wind speed conditions, the model underestimates H_{m0} when U_{10}<10 ms^{−1} or overestimates H_{m0} when U_{10}>10 ms^{−1}. Therefore, we take 10 ms^{−1} as reference, and there should be a $\left(\frac{{U}^{\prime}}{\mathrm{10}}\right)$ term in the C_{ds} equation. Second, C_{ds} depends on the fetch; considering the fetch dependence is logarithmic (the horizontal coordinate of Fig. 2b is logarithmic), there should be a ln(x) term in the C_{ds} equation. Considering that a logtransformed quantity must be unitless, we use the nondimensional fetch ln(x^{′}) instead of ln(x). Therefore, we assume C_{ds} in Eq. (14) has the following form:
where C1, C2, C3, and C4 are parameters to be determined by fitting the C_{ds} calculated from Step 1. Directly using the four parameters may be easier to fit, but it requires more effort to use or change it in real applications. Therefore, instead of finding a random combination of the four parameters that gives the least square error, we prefer to reduce the number of fitting parameters based on our understanding of the role of the two terms, namely ln(x^{′}) and $\left(\frac{{U}^{\prime}}{\mathrm{10}}\right)$ in Eq. (16). We use fixed value for the power on ln(x^{′}) and $\left(\frac{{U}^{\prime}}{\mathrm{10}}\right)$. By testing $\mathrm{1}\le C\mathrm{2}\le \mathrm{4}$ with a resolution of 1, and $\mathrm{0}\le C\mathrm{4}\le \mathrm{1}$ with a resolution of 0.1, we choose C2=2 and 4, C4=0.5 and 1 as representative values. With the values of C2 and C4 provided, we fit C1 and C3 with the data from Step 1 and end up with the first three groups of fitting parameters listed in Table 4 (FIT1 to FIT3). FIT4 in Table 4 is an improvement for FIT3 which will be described latter. The fitted C_{ds} using FIT1 to FIT4 at different wind speed conditions is shown in Fig. 2b as colored lines with circles and rectangles representing wind speed.
The four groups of parameters (FIT1 to FIT4 in Table 4) are applied in SWAN in the fetchlimited study, and the results are shown in Fig. 2c, d. The effect of the ${\left(\frac{{U}^{\prime}}{\mathrm{10}}\right)}^{C\mathrm{4}}$ term in Eq. (16) can be seen from the comparison between FIT1 and FIT2. The fitted C_{ds}, simulated H_{m0}, and f_{p} of FIT1 and FIT2 are shown in Fig. 2b, c, d as blue solid and red solid lines with circles and rectangles representing wind speed. Although significant difference of C_{ds} between C4=1.0 (FIT1) and C4=0.5 (FIT2) is seen in Fig. 2b, the resulting H_{m0} and f_{p} show a relatively small difference. In low wind speed conditions (U_{10}=5 and 10 ms^{−1}), C4=1.0 (blue solid lines in Fig. 2c, d) results in larger H_{m0} (smaller f_{p}) than C4=0.5 (red solid lines in Fig. 2c, d). In high wind speed conditions (U_{10}=20 and 40 ms^{−1}), C4=1.0 underestimates H_{m0} (overestimates f_{p}) more than C4=0.5 in long fetches (x>10 km).
The effect of the ln(x^{′})^{C2} term can be seen from the comparison between FIT2 and FIT3, the red solid and magenta dashed lines in Fig. 2b, c, d. The impact of C2 to H_{m0} and f_{p} is much weaker than C4. Using C2=2 (magenta dashed lines in Fig. 2c, d) results in slightly larger H_{m0} (smaller f_{p}) than C2=4 (red solid lines in Fig. 2c, d) in long fetches (x>10 km), which results in larger whitecapping dissipation, smaller H_{m0}, and larger f_{p}.
FIT1–FIT3 tend to underestimate H_{m0} (overestimate f_{p}) in long fetches (x>10 km). In the real case study in Sect. 4.3, it will be shown that such an underestimation of H_{m0} failed in capturing large waves in real storm simulations. Therefore, we continuously reduce the value of C2 and C4 until the large waves are captured in both the idealized fetchlimited study and real case study, which results in the parameters of FIT4 (hereafter WBLM). In FIT4, C4=0.0 means that $\left(\frac{{U}^{\prime}}{\mathrm{10}}\right)$ term is disappeared in Eq. (16). From Eq. (13), we found that both U^{′} and x^{′} could be written in the form of ${A}_{\mathrm{1}}{m}_{\mathrm{0}}^{{A}_{\mathrm{2}}}{f}_{\mathrm{p}}^{{A}_{\mathrm{3}}}$, which indicate that the function of the U^{′} term could be replaced by changing the parameters in the x^{′} term. This can be seen from the green lines in Fig. 2b that, without the $\left(\frac{{U}^{\prime}}{\mathrm{10}}\right)$ term, C_{ds} still follows different curves in different wind speed conditions.
Figure 3a shows the wave spectrum from WBLM (solid lines) in 10 ms^{−1} wind speed conditions after 72 h simulation in comparison to the spectrum parameterization of D85 (Donelan et al., 1985) (dashed lines) and JONSWAP (Hasselmann et al., 1973) (dotted lines) with the fetch dependence from Kahma and Calkoen (1992). Detailed equations for D85 and JONSWAP are listed in Appendix Appendix A. The results from the WBLM source term generally follow the shape of the D85 and JONSWAP spectra, but they tend to underestimate the energy at low frequencies at fetches x≥10 km.
The D85 spectra at short fetches (e.g., 1 km) have less energy at high frequencies (e.g., f>1 Hz) than at long fetches (e.g., 3000 km). On the contrary, JONSWAP and WBLM spectra have more energy at short fetches than at long fetches at high frequencies. The D85 spectra have a f^{−4} shape at high frequencies, while JONSWAP has a f^{−5} shape. The highfrequency part of WBLM spectra is between f^{−4} and f^{−5}. Figure 3b shows the source term distribution of windinput (S_{in}, solid lines), whitecapping dissipation (S_{ds}, dashed lines), and 5 times the cumulative dissipation ($\mathrm{5}{S}_{\mathrm{ds}}^{\text{c}}$, dotted lines) source functions at different fetches indicated by color legends in Fig. 3a. $\mathrm{5}{S}_{\mathrm{ds}}^{\text{c}}$ instead of ${S}_{\mathrm{ds}}^{\text{c}}$ is used to better visualize the cumulative dissipation source term. As the waves grow older, the S_{in} values at high frequencies become smaller, and ${S}_{\mathrm{ds}}^{\text{c}}$ values become larger, which may explain why the WBLM spectra have more energy at short fetches than at long fetches at high frequencies. Figure 3c shows the corresponding stress distribution within the wave boundary layer. Here, we also use $\mathrm{5}{\mathit{\tau}}_{\mathrm{w}}^{\mathrm{l}}$ (dotted lines) instead of ${\mathit{\tau}}_{\mathrm{w}}^{\mathrm{l}}$ for the purpose of visualizing the local waveinduced stress. Short fetch waves contribute more surface stress than long fetch waves at high frequencies, while they contribute less stress than long fetch waves at low frequencies. The total waveinduced stress depends on the integration of ${\mathit{\tau}}_{\mathrm{w}}^{\mathrm{l}}$ at all frequencies, which results in waves with a fetch of 5–15 km having the highest total waveinduced stress, waves with a fetch of 1 km having lower total waveinduced stress, and waves with a fetch of 3000 km having the lowest waveinduced stress. Accordingly, total wind stress (τ_{tot}, thick solid lines) at 5–15 km is larger than that at 1 and 3000 km because of the impact of the waves. Figure 3d shows the wind profiles within the wave boundary layer calculated by WBLM. Wind profiles are rather different in different wave development stages, which reveals that it is necessary to take the waveinduced wind profile variation into account in the estimation of the critical height in Sect. 2.1.
4.2 Idealized depthlimited study
Figure 4 shows the nondimensional wave energy as a function of nondimensional depth for fully developed waves in shallow waters after 24 h simulation, with the measurements of Young and Babanin (2006b) as reference. In comparison, results from KOM source terms are also shown. Both the WBLM (red crosses) and KOM (blue squares) show close agreement with the measurements (black circles).
The onedimensional wave spectrum in the depthlimited experiment is further examined after 24 h simulation in Fig. 5a–e for different wind speed and depth conditions, with the measurements of Young and Babanin (2006b) as reference. Both models capture the peak of the wave spectrum. However, KOM (blue lines) tends to underestimate the energy level at high frequencies. On the contrary, the energy level of WBLM (red lines) at high frequencies closely follows the measurements.
4.3 Two storms during the RUNE project
During the two RUNE storms on 28 November 2015 and 12 August 2015, wave simulation was done with SWAN forced by CFSv2 wind. The performance of KOM, JANS, and WBLM source terms are evaluated with buoy measurements in terms of significant wave height H_{m0}, mean wave direction D_{mean}, peak period T_{p}, mean period T_{m01}, and onedimensional wave spectrum. Figure 6 shows the simulated time series of H_{m0}, D_{mean}, T_{p}, and T_{m01} in comparison to buoy measurements at RUNE. To see the impact of different parameters of the WBLM whitecapping dissipation source function to the wave simulation, results from FIT1 to FIT3 are also shown. Similar to the conclusions in the idealized fetchlimited study in Sect. 4.1, these parameters significantly underestimate high waves. Only FIT4 (here WBLM) can be used for real wave simulations to capture the high waves.
Now we compare the performance of the new WBLM with KOM and JANS source terms. For H_{m0}, D_{mean}, and T_{p}, all the modeled time series generally follow the general trends of measurement data. The biggest error of H_{m0} happens at the two storm peaks. The three source terms overestimate the H_{m0} during the peak at about 1 m (15 %). WBLM gives slightly better H_{m0} during the peak than KOM and JANS. But it tends to underestimate T_{p} during the storm peaks. WBLM predicts T_{m01} significantly better than KOM and JANS. Note that the buoy T_{m01} is calculated from the frequency range of 0.005 to 0.64 Hz, JANS is from 0.03 to 0.58 Hz, and KOM and WBLM are from 0.03 to 0.63 Hz. A summary of the statistics is listed in Table 5, and the definitions of the statistics are given in Appendix Appendix B. WBLM generally provides better results for H_{m0} and T_{m01} than KOM and JANS. All the three source terms have similar accuracy in predicting D_{mean}. WBLM is slightly less accurate in predicting T_{p} than KOM and JANS.
The time series of H_{m0} at another nine measurement sites, including Fjaltring (FG), Hanstholm (HM), A121 (A1), Väderöarna (VA), Heligoland north (HN), Fino1 (F1), Fino3 (F3), SleipnerA (SA), and Ekofisk (EK) in the North Sea during the two storm simulations, are shown in Fig. 7. The relative statistics are listed in Table 6. Considering the statistics of mean difference (bias), root mean square error (RMSE), and scatter index (SI), WBLM generally gives better H_{m0} than KOM and JANS for most of the sites. However, in contrast to RUNE, WBLM tends to underestimate the largest waves during storm peaks in the openocean sites, e.g., the storm peak at A121 (Fig. 7c), SleipnerA (Fig. 7g), and Ekofisk (Fig. 7i).
The onedimensional wave spectrum during the whole simulation period at the RUNE site is presented in Fig. 8. The colored lines in Fig. 8a show the data from buoy measurements, and the black envelope lines are used to mark the upper and lower bounds of the measurement data. The colored lines in Fig. 8b, c, d are from SWAN simulations using KOM, JANS, and WBLM source terms, and the black envelope lines are the same as those in Fig. 8a. The three simulations generally capture the shape of the measured spectrum. In comparison to the measurements, KOM and JANS tend to overestimate the energy around the spectral peak, while WBLM gives better energy estimation around the spectral peak. Both KOM and JANS show a leveling off of energy at frequencies higher than about 0.3 Hz, while the measurement and WBLM do not, which may explain the failure of KOM and JANS in simulating T_{m01}. However, seemingly WBLM tends to overestimate the energy at frequencies higher than the peak, which needs further investigation.
This study first calibrates the WBLM windinput and dissipation source terms in idealized fetchlimited cases and further validates the model in idealized depthlimited cases and two real storm cases. In the selected cases, it is proven that the revised WBLM source terms can be used for real cases and can provide certain wave properties better than the original ones in SWAN, such as KOM and JANS. However, two storm cases do not represent all the wave conditions in the ocean; e.g., bimodal waves, slant waves, and swells are not analyzed in detail in this study. Therefore, more comprehensive analysis and validations from different data resources such as satellite data are still necessary in further studies.
The main difference between WBLM and previous windinput source functions in SWAN is that the WBLM explicitly considers physics such as the growth rate reduction of short wind waves in the presence of long waves. This effect mainly affect the young waves. Moreover, the modification of the dissipation coefficient is also focused on the young wind waves. Therefore, the introduction of WBLM source terms to SWAN mainly improves the young wind waves which are usually found in the coastal areas. This can be seen in Fig. 7. The significant wave height (H_{m0}) at the storm peak at the opensea sites, including A121 (Fig. 7c), SleipnerA (Fig. 7h), and Ekofisk (Fig. 7i), is underestimated by WBLM in comparison to measurements, while at the other coastal sites, the H_{m0} at the storm peak is captured quite well by WBLM. Relating the underestimation of H_{m0} at the storm peak at the opensea sites and the underestimation of T_{p} at RUNE (Fig. 6e), these defects maybe due to the inaccuracy of the calculation of nonlinear fourwave interactions at low frequencies or the overestimation of swell dissipation. Here, we simply added a positive wave age tuning parameter α in Eq. (10) following Bidlot et al. (2007) to shift the wave growth towards a lower frequency, which may not enough to overcome these defects. A more exact method for the calculation of nonlinear fourwave interactions and an independent swell dissipation such as that used by Ardhuin et al. (2010) is needed in the future study.
As mentioned in Du et al. (2017), one of the biggest strengths of WBLM is in the estimation of the air–sea momentum flux. Since this study mainly concerns its behavior in the wave simulations, the air–sea momentum flux (or roughness length/drag coefficient) is not included in the analysis. A future study with the focus on its momentum flux estimation was carried out and presented in Du (2017) chap. 8, and it was found that the WBLM method provides reliable roughness length estimation in terms of the magnitude and the spatial distribution of it in coastal shallow water in comparison to point measurements and the Envisat Advanced Synthetic Aperture Radar (ASAR) backscatter.
The WBLM source terms is found to improve the prediction of the mean period significantly. Through analyzing the frequency spectra, it is speculated to be caused by an improved description of the highfrequency part of the spectrum. However, the energy from WBLM at high frequencies seems too high in comparison to measurements. Therefore, the energy distribution in the highfrequency range needs to be further investigated. One possible way of reducing this overestimated energy at high frequencies is by tuning the parameters in the cumulative dissipation source function (e.g., C_{cu} in Eq. 15). However, the tuning of these parameters has to be followed by the tuning of the other parameters in the source terms including windinput, whitecapping, nonlinear fourwave interactions, etc. to make sure that both the stress estimation and the wave simulation are all satisfied.
Janssen (1991)'s windinput source function was found by Du et al. (2017) and Du (2017) chap. 5 to be wrongly implemented in SWAN. We found that in the calculation of the contribution of highfrequency tail to wave stress code, the judgment of whether the growth rate coefficient of a frequency (β_{g}) over all directions is zero, is called before calculating β_{g} in all the directions. This may disregard some of the highfrequency waves and therefore reduces the calculated wave stress. We also tried our best to change the code following the description of Bidlot et al. (2007) and by implementing some functions from WAM code (https://github.com/mywave/WAM, last access: 21 June 2017) to SWAN. Since this study is mainly focused on the usage of WBLM, a detailed analysis of Janssen implementation in SWAN is out of the scope of this paper.
This study aims at applying the WBLM source functions of Du et al. (2017) in SWAN for real wave simulations. Several improvements on the WBLM windinput and whitecapping dissipation source functions are realized. Firstly, the WBLM windinput source function is modified by considering the wind profile change in the estimation of the nondimensional critical height. Secondly, a revised whitecapping dissipation source function is applied, which enables the WBLM method to be used for varying wind conditions. Thirdly, a few refinements of the numerical algorithms of WBLM in SWAN are done to improve the model stability and efficiency, which make it possible to be used for largedomain and highresolution simulations.
The new pair of WBLM windinput and dissipation source functions is calibrated with fetchlimited and depthlimited simulations. It is proven to be able to reproduce the benchmark wave growth curve of Kahma and Calkoen (1992), the energy level, and the onedimensional wave spectrum measured by Young and Babanin (2006b) in the depthlimited study.
The WBLM windinput and dissipation source functions are validated with several point measurements during two storms over the North Sea. Results show that, in comparison to the original windinput and dissipation source functions in SWAN, namely Komen et al. (1984) and Janssen (1991), WBLM improves the prediction of significant wave height and mean wave period in comparison to measurements.
The source code for SWAN used in this study is freely available at http://swanmodel.sourceforge.net (last access: 21 June 2017). The bathymetry data is obtained from EMODnet (http://www.emodnetbathymetry.eu/, last access: 20 November 2014). The observational wave data is downloaded from NOOS (https://noos.bsh.de/, last access: 15 June 2017), FINO (http://fino.bsh.de, registration needed, last access: 3 April 2018), and eKlima (http://sharki.oslo.dnmi.no, registration needed, last access: 17 November 2016). CFSv2 10 m wind speed is download from NCEP Climate Forecast System Version 2 (CFSv2) Selected Hourly TimeSeries Products (Saha et al., 2011). The model data and source code modifications can be obtained from the corresponding author.
The D85 (Donelan et al., 1985) spectra are described by
where f is the frequency and f_{p} is the frequency at the spectral peak. α_{D} is a equilibrium range parameter which is written as
where U_{10} is 10 m wind speed. c_{p} is the phase velocity at the spectral peak. In deep water conditions, ${c}_{\mathrm{p}}=\frac{g}{\mathrm{2}\mathit{\pi}{f}_{\mathrm{p}}}$, where g is the gravity acceleration. γ_{D} is a peak enhancement factor:
and σ_{D} is a peak width parameter, which is written as
The JONSWAP (Hasselmann et al., 1973) spectra are described by
where the equilibrium range parameter is written as
where ($\stackrel{\mathrm{\u0303}}{x}=xg/{U}_{\mathrm{10}}^{\mathrm{2}}$) is a nondimensional fetch, and x is the fetch. Parameterization for the peak enhancement factor (γ_{J}) for JONSWAP spectra is not provided by Hasselmann et al. (1973). According to Hasselmann et al. (1973), γ_{J} is scatted between 1.5 and 6.0, with an average value of 3.3. So, we use the same equation as D85 (Eq. A2) with a limit of $\mathrm{1.5}\le {\mathit{\gamma}}_{J}\le \mathrm{6.0}$. The peak width parameter is written as
For both D85 and JONSWAP spectra, the peak frequency (f_{p}) for a given wind speed (U_{10}) and fetch (x) is calculated from the fetchlimited wave growth relationship of Kahma and Calkoen (1992):
Taking X as the observation value and Y as the modeled value, the mean difference is defined as
The root mean square error is defined as
The scatter index is defined as
The updated WBLM model was developed by JD. JD, RB, and XGL designed all the numerical experiments, which where performed by JD. RB provided the measurement data of RUNE. The manuscript was written by JD with the assistance of RB, XGL, and MK. All authors worked on the paper.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Coastal modelling and uncertainties based on CMEMS products”. It is not associated with a conference.
This project has received funding from the Danish Forskel project XWiWa (PSO12020), the
EU CEASELESS project (H2020EO2016730030), and the National Key Research and Development
Program of China (2017YFC1404200). We are grateful to Jean Raymond Bidlot from ECMWF,
Anna Rutgersson from Uppsala University, and Henrik Bredmose from DTU Wind Energy for
helpful discussions and inputs. Furthermore, we would like to thank Rogier Floors from
DTU Wind Energy for providing the measurement data during Danish ForskEL project RUNE
(12263).
Edited by: Manuel Espino Infantes
Reviewed by: one anonymous referee
Alves, J. H. G. M. and Banner, M. L.: Performance of a SaturationBased DissipationRate Source Term in Modeling the FetchLimited Evolution of Wind Waves, J. Phys. Oceanogr., 33, 1274–1298, https://doi.org/10.1175/15200485(2003)033<1274:POASDS>2.0.CO;2, 2003. a
Ardhuin, F.: Dissipation parameterizations in spectral wave models and general suggestions for improving on today's wave models Fabrice Ardhuine, in: ECMWF Workshop on Ocean Waves, June, 113–124, 2012. a, b
Ardhuin, F. and Roland, A.: Coastal wave reflection, directional spread, and seismoacoustic noise sources, J. Geophys. Res.Oceans, 117, 1–16, https://doi.org/10.1029/2011JC007832, 2012. a
Ardhuin, F., Rogers, E., Babanin, A. V., Filipot, J.F., Magne, R., Roland, A., van der Westhuysen, A., Queffeulou, P., Lefevre, J.M., Aouf, L., and Collard, F.: Semiempirical Dissipation Source Functions for Ocean Waves, Part I: Definition, Calibration, and Validation, J. Phys. Oceanogr., 40, 1917–1941, https://doi.org/10.1175/2010JPO4324.1, 2010. a, b, c, d, e
Ardhuin, F., Roland, A., Dumas, F., Bennis, A.C., Sentchev, A., Forget, P., Wolf, J., Girard, F., Osuna, P., and Benoit, M.: Numerical wave modelling in conditions with strong currents: dissipation, refraction and relative wind, J. Phys. Oceanogr., 42, https://doi.org/10.1175/JPOD110220.1, 2012. a
Babanin, A. V. and Young, I. R.: Twophase behaviour of the spectral dissipation of wind waves, in: Proc. Ocean Waves Measurements and Analysis, Fifth Intern, Symposium WAVES2005, December 2016, 2005. a
Babanin, A. V., Tsagareli, K. N., Young, I. R., and Walker, D. J.: Numerical Investigation of Spectral Evolution of Wind Waves, Part II: Dissipation Term and Evolution Tests, J. Phys. Oceanogr., 40, 667–683, https://doi.org/10.1175/2009JPO4370.1, 2010. a, b
Banner, M. L. and Morison, R. P.: Refined source terms in wind wave models with explicit wave breaking prediction, Part I: Model framework and validation against field data, Ocean Model., 33, 177–189, https://doi.org/10.1016/j.ocemod.2010.01.002, 2010. a
Battjes, J. A. and Janssen, J. P. F. M.: Energy loss and setup due to breaking of random waves, in: Proceedings of 16th International Conference on Coast. Eng., New York, American Society of Civil Engineers, 569–587, https://doi.org/10.1017/CBO9781107415324.004, 1978. a, b
Bidlot, J.R.: Present Status of Wave Forecasting at ECMWF, in: ECMWF Workshop on Ocean Waves, Shinfield Park, Reading RG2 9AX, United Kingdom, 25–27 June 2012, 2012. a, b, c
Bidlot, J.R., Janssen, P. A. E. M., and Abdalla, S.: A revised formulation of ocean wave dissipation and its model impact, ECMWF Technical Memorandum, 509, http://www.citeulike.org/group/11419/article/6354018 (last access: 14 October 2014), 2007. a, b, c, d, e
Bolaños, R.: D3.3 Metocean Conditions and Wave Modelling, Tech. Rep. November, available at: http://orbit.dtu.dk/files/155595377/D3p3_final.pdf (last access: 10 April 2017), 2016. a
Bolaños, R. and Rørbæk, K.: D3.1 Metocean Buoy Deployment, Tech. Rep. November, available at: http://orbit.dtu.dk/files/155595376/D3p1_final.pdf, (last access: 10 April 2017), 2016. a
Bolaños, R., Larsén, X. G., Petersen, O., Nielsen, J., Kelly, M., KofoedHansen, H., Du, J., Sørensen, O., Larsen, S., Hahmann, A., and Badger, M.: Coupling atmospherere and waves for coastal wind turbine design, Proceedings of the Coast. Eng. Conference, 1–11 January 2014, 2014. a
Booij, N., Ris, R., and Holthuijsen, L.: A thirdgeneration wave model for coastal regions, IModel description and validation, J. Geophys. Res., 104, 7649–7666, https://doi.org/10.1029/98JC02622, 1999. a, b
Cavaleri, L.: Wave ModelingMissing the Peaks, J. Phys. Oceanogr., 39, 2757–2778, https://doi.org/10.1175/2009JPO4067.1, 2009. a
Cavaleri, L., Alves, J. H. G. M., Ardhuin, F., Babanin, A., Banner, M., Belibassakis, K., Benoit, M., Donelan, M., Groeneweg, J., Herbers, T. H. C., Hwang, P., Janssen, P. A. E. M., Janssen, T., Lavrenov, I. V., Magne, R., Monbaliu, J., Onorato, M., Polnikov, V., Resio, D., Rogers, W. E., Sheremet, A., McKee Smith, J., Tolman, H. L., van Vledder, G., Wolf, J., and Young, I.: Wave modelling – The state of the art, Prog. Oceanogr., 75, 603–674, https://doi.org/10.1016/j.pocean.2007.05.005, 2007. a
Donelan, M. A., Hamilton, J., and Hui, W. H.: Directional Spectra of WindGenerated Waves, Philos. T. R. Soc. Lond., 315, 509–562, https://doi.org/10.1098/rsta.1985.0054, 1985. a, b, c
Du, J.: Coupling atmospheric and ocean wave models for storm simulation, Ph.D. thesis, Technical University of Denmark, https://doi.org/10.11581/DTU:00000020, 136 pp., 2017. a, b
Du, J., Bolaños, R., and Larsén, X.: The use of a wave boundary layer model in SWAN, J. Geophys. Res.Oceans, 122, 42–62, https://doi.org/10.1002/2016JC012104, 2017. a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s
Floors, R., Hahmann, A., Peña, A., and Karagali, I.: Estimating nearshore wind resources EReport DTU Wind Energy, Vol. 0116, available at: http://orbit.dtu.dk/files/126992882/final.pdf (last access: 10 April 2017), 2016a. a
Floors, R., Lea, G., Peña, A., Karagali, I., and Ahsbahs, T.: Report on RUNE's coastal experiment and first intercomparisons between measurements systems, Vol. 0115, available at: http://orbit.dtu.dk/files/127277148/final.pdf (last access: 10 April 2017), 2016b. a, b
Floors, R., Peña, A., Lea, G., Vasiljevic, N., Simon, E., and Courtney, M.: The RUNE experimentA database of remotesensing observations of nearshore winds, Remote Sens., 8, https://doi.org/10.3390/rs8110884, 2016c. a
Hasselmann, K.: On the spectral dissipation of ocean waves due to white capping, BoundLay. Meteorol., 6, 107–127, https://doi.org/10.1007/BF00232479, 1974. a
Hasselmann, K., Barnett, T. P., Bouws, E., Carlson, H., Cartwright, D. E., Enke, K., Ewing, J. A., Gienapp, H., Hasselmann, D. E., Kruseman, P., Meerburg, A., Muller, P., Olbers, D. J., Richter, K., Sell, W., and Walden, H.: Measurements of WindWave Growth and Swell Decay during the Joint North Sea Wave Project (JONSWAP), Ergänzungsheft zur Deutschen Hydrographischen Zeitschrift Reihe, A8, p. 95, 2710264, 1973. a, b, c, d, e, f, g
Hua, F. and Yuan, Y.: Theoretical Study of Breaking Wave Spectrum and its Application, in: Breaking Waves, Springer Berlin Heidelberg, Berlin, Heidelberg, 277–282, https://doi.org/10.1007/978364284847630, 1992. a
Janssen, P. A. E. M.: Quasilinear Theory of WindWave Generation Applied to Wave Forecasting, J. Phys. Oceanogr., 21, 1631–1642, https://doi.org/10.1175/15200485(1991)021<1631:QLTOWW>2.0.CO;2, 1991. a, b, c, d, e, f
Kahma, K. K. and Calkoen, C. J.: Reconciling Discrepancies in the Observed Growth of Windgenerated Waves, 22, https://doi.org/10.1175/15200485(1992)022<1389:RDITOG>2.0.CO;2, 1992. a, b, c, d, e, f, g, h, i, j
Komen, G. J., Hasselmann, S., and Hasselmann, K.: On the Existence of a Fully Developed WindSea Spectrum, J. Phys. Oceanogr., 14, 1271–1285, https://doi.org/10.1175/15200485(1984)014<1271:OTEOAF>2.0.CO;2, 1984. a, b, c, d, e, f, g
Komen, G. J., Cavareli, L., Donelan, M., Hasselmann, K., Hasselmann, S., and Janssen, P. A. E. M.: Dynamics and modelling of ocean waves, Cambridge University Press, https://doi.org/10.1017/CBO9780511628955, 532 pp., 1994. a, b, c, d
Larsén, X. G., Du, J., Bolaños, R., and Larsen, S.: On the impact of wind on the development of wave field during storm Britta, Ocean Dynam., 67, 1407–1427, https://doi.org/10.1007/s1023601711001, 2017. a, b
Leckler, F., Ardhuin, F., Filipot, J. F., and Mironov, A.: Dissipation source terms and whitecap statistics, Ocean Model., 70, 62–74, https://doi.org/10.1016/j.ocemod.2013.03.007, 2013. a
LonguetHiggins, M. S.: On Wave Breaking and the Equilibrium Spectrum of WindGenerated Waves, P. Roy. Soc. AMath. Phy., 310, 151–159, https://doi.org/10.1098/rspa.1969.0069, 1969. a
Melville, W. K. and Matusov, P.: Distribution of breaking waves at the ocean surface., Nature, 417, 58–63, https://doi.org/10.1038/417058a, 2002. a
Miles, J. W.: On the generation of surface waves by shear flows, Part 2, J. Fluid Mech., 3, 568–582, https://doi.org/10.1017/S0022112059000830, 1957. a, b, c
Phillips, O. M.: Spectral and statistical properties of the equilibrium range in windgenerated gravity waves, J. Fluid Mech., 156, 505–531, https://doi.org/10.1017/S0022112085002221, 1985. a, b, c
Polnikov, V. G.: On a description of a windwave energy dissipation function, in: The AirSea Interface: Radio and Acoustic Sensing, Turbulence and Wave Dynamics, edited by: Donelan, M. A., Hui, W. H., and Plant, W. J., Rosenstiel School of Marine and Atmospheric Science, University of Miami, 277–282, 1993. a
Saha, S., Moorthi, S., Wu, X., Wang, J., Nadiga, S., Tripp, P., Behringer, D., Hou, Y., Chuang, H., Iredell, M., Ek, M., Meng, J., Yang, R., Mendez, M. P., van den Dool, H., Zhang, Q., Wang, W., Chen, M., and Becker, E.: NCEP Climate Forecast System Version 2 (CFSv2) Selected Hourly TimeSeries Products. Research Data Archive at the National Center for Atmospheric Research, Computational and Information Systems Laboratory, updated monthly, https://doi.org/10.5065/D6N877VB, last access: 17 November 2016, 2011. a
Snyder, R. L., Dobson, F. W., Elliott, J. A., and Long, R. B.: Array measurements of atmospheric pressure fluctuations above surface gravity waves, J. Fluid Mech., 102, 1–59, https://doi.org/10.1017/S0022112081002528, 1981. a, b
Sørensen, O. R., KofoedHansen, H., Rugbjerg, M., and Sørensen, L. S.: A thirdgeneration spectral wave model using an unstructured finite volume technique, Proceedings of the 29th Intern. Conf. on Coastal Eng., 1, 894–906, https://doi.org/10.1142/97898127019160071, 2005. a
SWAN: Swan Scientific and Technical Documentation, SWAN Cycle III, Version 41.01, 1–132, 2014. a
Tolman, H. L. and Chalikov, D.: Source Terms in a ThirdGeneration Wind Wave Model, 26, https://doi.org/10.1175/15200485(1996)026<2497:STIATG>2.0.CO;2, 1996. a
van der Westhuysen, A. J., Zijlema, M., and Battjes, J. A.: Nonlinear saturationbased whitecapping dissipation in SWAN for deep and shallow water, Coast. Eng., 54, 151–170, https://doi.org/10.1016/j.coastaleng.2006.08.006, 2007. a, b
Young, I. R. and Babanin, A. V.: Spectral Distribution of Energy Dissipation of WindGenerated Waves due to Dominant Wave Breaking, J. Phys. Oceanogr., 36, 376–394, https://doi.org/10.1175/JPO2859.1, 2006a. a, b
Young, I. R. and Babanin, A. V.: The form of the asymptotic depthlimited wind wave frequency spectrum, J. Geophys. Res.Oceans, 111, 1–15, https://doi.org/10.1029/2005JC003398, 2006b. a, b, c, d, e, f, g, h, i
Young, I. R., Banner, M. L., Donelan, M. A., Babanin, A. V., Melville, W. K., Veron, F., and McCormick, C.: An integrated system for the study of windwave source terms in finitedepth water, J. Atmos. Ocean. Tech., 22, 814–831, https://doi.org/10.1175/JTECH1726.1, 2005. a
Yuan, Y., Tung, C. C., and Huang, N. E.: Statistical Characteristics of Breaking Waves, in: Wave Dynamics and Radio Probing of the Ocean Surface, Springer US, Boston, MA, 265–272, https://doi.org/10.1007/978146848980418, 1986. a
Zieger, S., Babanin, A. V., Erick Rogers, W., and Young, I. R.: Observationbased source terms in the thirdgeneration wave model WAVEWATCH, Ocean Model., 96, 2–25, https://doi.org/10.1016/j.ocemod.2015.07.014, 2015. a
Zijlema, M., Van Vledder, G. P., and Holthuijsen, L. H.: Bottom friction and wind drag for wave models, Coast. Eng., 65, 19–26, https://doi.org/10.1016/j.coastaleng.2012.03.002, 2012. a, b