3.1 Onset of breaking and local wave geometry

Wave breaking is considered as a threshold process, with criteria for predicting breaking onset falling into three categories: geometric, kinematic and energetic (Barthelemy et al. Reference Barthelemy, Banner, Peirson, Fedele, Allis and Dias2018). The geometric threshold involves wave steepness, wave asymmetry and maximum theoretical steepness, while the kinematic threshold involves horizontal crest particle velocity and the wave phase speed, $C$ . The sequential images of breaking onset in Qiao & Duncan (Reference Qiao and Duncan2001), Diorio et al. (Reference Diorio, Liu and Duncan2009) and Duncan et al. (Reference Duncan, Qiao, Philomin and Wenz1999) suggested that the geometrical threshold based on the occurrence of a vertical tangent on the forward face of the crest may not be particularly robust. Nevertheless, examination of kinematic criteria is non-trivial because of the difficulties in determining the horizontal crest particle velocity and the ambiguity in defining $C$ for highly unsteady, rapidly evolving breaking crests (Perlin et al. Reference Perlin, Choi and Tian2013). For the energetic threshold, Barthelemy et al. (Reference Barthelemy, Banner, Peirson, Fedele, Allis and Dias2018) proposed a criterion $B_{th}=0.85$ , where $B$ is defined as the ratio between the local energy flux and the local energy density projected on the wave propagation direction and $B_{th}$ is the threshold value of $B$ at which the wave begins to break. At the free surface, $B$ effectively is the horizontal velocity at the crest divided by the forward translation speed of the crest (Barthelemy et al. Reference Barthelemy, Banner, Peirson, Fedele, Allis and Dias2018). Derakhti, Banner & Kirby (Reference Derakhti, Banner and Kirby2018) further suggested that the strength of breaking depends on the rate of change of $B$ at the particular moment when $B$ is up-crossing the critical value of $B_{th}=0.85$ . They numerically obtained the formulation $b=0.034(\unicode[STIX]{x1D6E4}-0.30)^{2.5}$ , where $b$ is the breaking strength parameter, $\unicode[STIX]{x1D6E4}$ is a normalized parameter defined as $\unicode[STIX]{x1D6E4}=T_{b}\,\text{d}B/\text{d}t|_{B_{th}}$ and $T_{b}$ is the local wave period at breaking.

As shown in figure 3, the present experimental data agree with the Barthelemy et al. (Reference Barthelemy, Banner, Peirson, Fedele, Allis and Dias2018) simulations in which $B$ sharply increases near the breaking onset. Note that, in the present study, $B$ was calculated based on the measured horizontal velocity at the crest peak $x_{peak}$ and the phase speed $C$ from the linear wave theory. The location of $x_{peak}$ was first identified at each time instant from the BIV images. Each data point $B$ in figure 3 was then calculated as the horizontal particle velocity at the corresponding location divided by the phase speed $C$ . We examined BIV images and videos during those multiple instances when $B$ increases after breaking (e.g.  $t/T\sim 0.3$ , 0.9 and 1.2). Interestingly, the instantaneous BIV images depict that ‘small impinging/splash-up processes’, similar to those in the plunging breakers in Lim et al. (Reference Lim, Chang, Huang and Na2015) but with a much smaller physical scale, occur during this period. So these small impingements and splash-ups could be the possible cause of the multiple threshold crossing of $B$ . Banner & Fooks (Reference Banner and Fooks1985) observed disturbances in the free-surface measurements that were caused by the presence of the rearward-travelling shear-layer waves. However, these rearward waves are unlikely to be the cause of the multiple increases of $B$ in the present study because the breaking waves used in Banner & Fooks (Reference Banner and Fooks1985) are approximately one order of magnitude smaller in wavelength than those in the present study, and their aeration level is also much lower.

Figure 3.

Temporal evolution of $B$ at the surface calculated based on the wave propagation direction plotted against the threshold value of $B_{th}=0.85$ . The vertical thick grey ‘line’ indicates the range of breaking point. The inset depicts the breaking point, which was geometrically determined from the mean PIV images when a vertical tangent segment appears.

Note that, if $C$ is replaced by the forward translation speed of the crest in calculating $B$ in figure 3, as stated in Barthelemy et al. (Reference Barthelemy, Banner, Peirson, Fedele, Allis and Dias2018), this speed may be obtained following Saket et al. (Reference Saket, Peirson, Banner, Barthelemy and Allis2017) based on the BIV images. In the present study, the horizontal location of the crest point $x_{peak}$ at each time interval of 0.01 s was identified in the BIV images. By linearly fitting the data (the data trend is linear), the averaged crest point speed $x_{peak}/t$ was found to be $1.06L/T$ or $1.06C$ . We prefer the use of $C$ instead of $x_{peak}/t$ because $C$ is readily available. The discrepancy between the averaged crest point speed and $C$ is approximately 6 %.

In figure 3, the approximate range of breaking onset (the shaded region) was determined by visual examination of the PIV images when the wave front became vertical. The inset in figure 3 shows the mean PIV image at the breaking onset when a vertical tangent segment appears. As discussed earlier, using visual examination of the wave front becoming vertical to determine the onset of breaking is non-trivial due to the fact that breaking waves are inherently highly irregular and unsteady near the breaking point and deform rapidly after breaking. However, the geometrical criterion used in the present study can be justified by the following reasons: (i) the variation was less than 1 % of the wavelength (and only 5 % of the wave height) among the 20 repeated runs; (ii) the approach does not involve the ambiguity of defining phase speed; and (iii) the approach directly used the measurements in a spatial domain, and thus the ambiguity in transferring a temporal measurement to a spatial measurement is not an issue. We would also like to bring up the difference between spilling breakers used in the present study and the gentle spilling breakers in Qiao & Duncan (Reference Qiao and Duncan2001) and Diorio et al. (Reference Diorio, Liu and Duncan2009) – the present breaking waves are of high intensity ( $H=0.265~\text{m}$ , high air entrainment with lots of bubbles and a high aeration level due to strong impingement jet-like impact during the process), whereas the breakers in Qiao & Duncan and Diorio et al. (wave heights of the order of $H\sim O(0.01~\text{m})$ ) are mostly very weak breakers with a very low amount of air entrainment and a very low aeration level. It is relatively easy to identify the vertical tangent line in the present study.

The breaking strength parameter $b$ and the normalized parameter $\unicode[STIX]{x1D6E4}$ in the present study are 0.012 and 1.09, respectively, which are consistent with the values reported in Derakhti et al. (Reference Derakhti, Banner and Kirby2018). Noticeably, in the present study, multiple increases of $B$ exceeding 0.85 were observed. Through examining the corresponding BIV images, the occurrences of very weak impingements of the overturning jets (of a single breaking wave as it propagated) were coincident with those multiple exceedances. This may suggest that multiple weak jet impingements occurred in the spilling breaker and these impingements caused $B$ to increase. Interestingly, the rate of change of $B$ for those increases is very similar, implying that the strength of breaking (i.e. impingements of the jets) may also be similar. We also observed that $B$ exceeds 0.85 near the breaking point under plunging breaking waves (not shown) using the data from Lim et al. (Reference Lim, Chang, Huang and Na2015). Therefore, the breaking criterion was applied to both spilling and plunging breakers.

Figure 4 shows the trajectory of the dimensionless breaking wave crest. As shown by the linear least-squares fit, the crest height decreased continuously with a constant slope of $-0.08$ over the entire breaking process. Through laboratory experiment, Tian et al. (Reference Tian, Perlin and Choi2012) showed that the crest height of spilling waves is more uniform compared to that of plunging waves. They related the difference in the growth/decay rate of the crest height to the breaking strength parameter, $b$ . That is in agreement with Diorio et al. (Reference Diorio, Liu and Duncan2009), while the reason for spillers to have a lower growth rate is their weaker breaking strength. Furthermore, Tian et al. (Reference Tian, Perlin and Choi2012) observed that the crest height of plungers decreases rapidly following the wave breaking, corresponding to a significant loss of potential energy. However, no abrupt decrease of the crest height was observed for the spillers.

Figure 4.

Spatial variation of the dimensionless crest profiles and trajectory. The crest profiles were identified using the BIV images while the solid circles indicate the local maxima. The solid line is the linear least-squares fit of the local maxima with a slope of $-0.08$ .

3.2 Flow kinematics, vorticity and turbulence

The instantaneous velocity fields measured using PIV were ensemble-averaged to obtain the mean velocity fields $(U,W)$ from 20 repeated runs, i.e.

(3.1) $$\begin{eqnarray}\displaystyle & \displaystyle U(x,z,t)=\frac{1}{N}\mathop{\sum }_{i=1}^{N}u_{i}(x,z,t), & \displaystyle\end{eqnarray}$$

(3.2) $$\begin{eqnarray}\displaystyle & \displaystyle W(x,z,t)=\frac{1}{N}\mathop{\sum }_{i=1}^{N}w_{i}(x,z,t), & \displaystyle\end{eqnarray}$$

where $u_{i}$ and $w_{i}$ are the instantaneous horizontal and vertical velocities, the subscript $i$ represents the $i$ th measurement, and $N$ is the total number of repeated runs ( $N=20$ in the present study). The averaged r.m.s. velocity fluctuations are $u_{rms}=0.016~\text{m}~\text{s}^{-1}$ and $w_{rms}=0.011~\text{m}~\text{s}^{-1}$ before breaking.

To ensure that the instantaneous velocity field is based on the air bubbles in the measurement plane, we limited the DOF and shortened the exposure time by carefully setting up the camera and lens. By limiting the DOF, the objects located in front of the DOF near the front wall appear to be blurred due to being out of focus. The image intensity of the bubbles in these blurred images is significantly weaker than those in the sharp, focused images, so they have little effect on the correlation process (i.e. much lower image intensity/peak and therefore much lower correlation peak in PIV analysis). Figure 5 in Ryu et al. (Reference Ryu, Chang and Lim2005) showed that the velocity measured from the correlation of the superimposed sharp and blurry images is very close to the velocity measured from the sharp images alone. Moreover, by shortening the exposure time as well as using a weak continuous laser, we minimized image saturation for bubbles illuminated by the laser inside the light sheet, thus providing insignificant light intensity (and thus insignificant correlation contribution) to the bubbles near the front wall outside the light sheet.

Figure 5.

Mean relative velocity fields ( $U-C$ , $W$ ), with only one-quarter of the velocity vectors plotted, superimposed on mean images for (a)  $t/T=-0.01$ (FOV1), (b) 0.13 (FOV3), (c) 0.28 (FOV5), (d) 0.40 (FOV7), (e) 0.67 (FOV9), (f) 0.84 (FOV11), (g) 0.99 (FOV13), (h) 1.17 (FOV15), (i) 1.40 (FOV17), (j) 1.66 (FOV19), (k) 1.91 (FOV21) and (l) 2.14 (FOV23). The yellow dots represent the locations of the local maximum horizontal velocity. (m) Temporal variation of the maximum horizontal velocity.

Figure 5 shows the measured mean relative velocity fields $(U-C,W)$ with one-quarter of the velocity vectors plotted (every other row and every other column). Note that the velocities were plotted on a moving frame (moving with $C$ ) so the vortical motion inside the aerated region is legible, and the free surface was identified based on high contrasts between the bright bubble cloud region and the dark air region above in the PIV images. The breaking onset was initiated without an explicit jet impinging onto the quiescent front. When the primary breaking wave front becomes vertical, the maximum horizontal velocity reaches $1.1C$ at this breaking point (figure 5 a,m). The same value was found in Duncan (Reference Duncan2001) under weak spilling breaking waves. In plunging breaking waves, higher values of $1.3C$ (Perlin, He & Bernal Reference Perlin, He and Bernal1996) and $1.4C$ (Lim et al. Reference Lim, Chang, Huang and Na2015) were reported at the breaking point. Spilling waves may be initiated from the bumpy surface or small jet impingement (Duncan Reference Duncan2001). It is difficult to confirm which of the two processes led to breaking based on figure 5(a–c) and the corresponding BIV images (not shown here). However, it is clear from the images that a small aerated region is formed after the wave passes the breaking point and that this region grows as the wave propagates, similar to what was reported in Duncan et al. (Reference Duncan, Philomin, Qiao and Kimmel1994) and Rojas & Loewen (Reference Rojas and Loewen2010). As the breaking progresses, the vertical location of the maximum horizontal velocity shifts lower (figure 5 a–c), and a shear zone between the high-velocity aerated region and the low-velocity non-aerated region persists, developing turbulence near the lower boundary of the aerated region. At the later stages of breaking (figure 5 e,f), the maximum velocities are found to locate near the middle of the bubble plume as the wave propagates. The magnitude of the maximum horizontal velocity during the entire breaking process is $1.5C$ , found near the shear zone at $t/T=0.35$ (figure 5 m) occurring in between figures 5(c) and 5(d).

Note that the darker, blurred region dividing the upper and lower foamy (white) regions in figure 5(d–i) is caused by the bubbly flow between the front wall of the wave flume and the light sheet plane (i.e. out-of-focus bubbles near the front wall). These bubbles close to the front wall may block a certain area in the image. Nevertheless, the camera used in the current study has a high light sensitivity and a high dynamic range. With the carefully controlled laser power and camera exposure time, the camera was able to capture even the low-light-intensity tracers within these blurred regions except for the infrequent cases when the entire PIV interrogation window was completely blocked by the bubbles. Note that the original images are much darker in those bright areas but enhanced in the plot for better visualization and identification of the aerated region and boundaries. The instantaneous PIV velocities (not shown here) in the present spilling breakers are similar to those in the plunging breakers reported in Lim et al. (Reference Lim, Chang, Huang and Na2015). It is noted that the number of empty velocity vectors is relatively low even in the highly aerated region.

The instantaneous horizontal and vertical fluctuating velocities, $u^{\prime }$ and $w^{\prime }$ , are computed as

(3.3) $$\begin{eqnarray}\displaystyle & \displaystyle u^{\prime }(x,z,t)=u(x,z,t)-U(x,z,t), & \displaystyle\end{eqnarray}$$

(3.4) $$\begin{eqnarray}\displaystyle & \displaystyle w^{\prime }(x,z,t)=w(x,z,t)-W(x,z,t). & \displaystyle\end{eqnarray}$$

Since the cross-tank velocity component was not measured, we define a two-dimensional turbulent intensity $I_{2D}$ as

(3.5a-c ) $$\begin{eqnarray}I_{2D}=\sqrt{\langle u^{\prime \,2}\rangle +\langle w^{\prime \,2}\rangle },\quad I_{x}=\sqrt{\langle u^{\prime \,2}\rangle },\quad I_{z}=\sqrt{\langle w^{\prime \,2}\rangle },\end{eqnarray}$$

where $\langle ~\rangle$ denotes the ensemble-average operator. The estimated uncertainty for turbulent intensity using bootstrapping is $0.018~\text{m}~\text{s}^{-1}$ , equivalent to $0.011C$ .

Figure 6.

Turbulent intensity fields ( $\text{m}~\text{s}^{-1}$ ) near the onset of breaking at (a,d,g)  $t/T=-0.09$ , (b,e,h) 0 and (c,f,i) 0.17. (a–c) Turbulent intensity $I_{2D}$ ; (d–f) horizontal turbulent intensity $I_{x}$ ; and (g–i) vertical turbulent intensity $I_{z}$ . (j) Mean velocity field ( $U-C,W$ ) at $t/T=-0.07$ . The estimated uncertainty for turbulent intensity using bootstrapping is $0.018~\text{m}~\text{s}^{-1}$ , equivalent to $0.011C$ .

Figure 6 shows the turbulent intensity fields near the onset of breaking. Before the wave front becomes vertical, turbulence has already appeared in the middle of the crest (figure 6 a,d). The onset of breaking in the present study was determined from the images by manually identifying the moment when the wave front becomes vertical. This is related to the kinematic breaking criteria, which often involve the horizontal crest particle velocity and the wave phase speed (Perlin et al. Reference Perlin, Choi and Tian2013). Perlin et al. (Reference Perlin, He and Bernal1996) and Chang & Liu (Reference Chang and Liu1998) supported these criteria by experimentally confirming that the horizontal crest particle velocity exceeds the wave phase speed when the wave becomes vertical. Nevertheless, determining the horizontal crest particle velocity is difficult, and defining the phase speed is somewhat ambiguous (Perlin et al. Reference Perlin, Choi and Tian2013). Hence, the onset of breaking may have occurred earlier and caused the turbulence in the middle of the crest (probably transferred from the front face of the crest). By examining the mean velocity before breaking, we observed the presence of a weak roller as depicted in figure 6(j). Although the cause is not clear, this roller may be a possible indication of turbulence in the middle of the crest. The horizontal turbulent intensity is more dominant in the inner region of the crest when compared to the vertical turbulent intensity, as shown in figure 6(d,g). A thin layer of turbulence formed following the wave front is initiated at the beginning of the process (figure 6 a,d,g). This thin layer of turbulence may be generated from the thin shear layer created by the initial impact of the jet (Tulin Reference Tulin, Grue, Gjevik and Weber1996). Nevertheless, we do not believe that this is similar to the jet impingement mechanism in plunging breakers. As the breaking wave progresses, the thin layer of turbulence becomes thicker and increases in intensity. Longuet-Higgins & Turner (Reference Longuet-Higgins and Turner1974) theoretically modelled spilling breakers, predicting that this layer of turbulence moves downslope with a constant acceleration, and its thickness increases linearly with distance from the crest. Similarly, Kimmoun & Branger (Reference Kimmoun and Branger2007) observed that spilling breakers are initiated from a small region of bubbles on the forward side of the crest; the region then becomes turbulent and grows by spreading downslope. Figure 6 and the corresponding BIV images (not shown) qualitatively show the bubble plume evolution.

Figure 7 shows the mean vorticity fields. As shown in the figure, a vortical region spreads along the surface located mainly below the toe immediately after breaking. As the breaking progresses, the vortical region stretches and extends to the crest above the toe (figure 7 f). Subsequently, negative vorticity appears near the toe, then moves downward along the wave front (figure 7 i) and fills the upper crest region above the shear zone between the fast-moving aerated region and the quiescent region below (figure 7 l). These processes are in agreement with the finding of Duncan et al. (Reference Duncan, Qiao, Philomin and Wenz1999) and Duncan (Reference Duncan2001). However, Duncan (Reference Duncan2001) extensively discussed the development of the ripples which are not discernible from the present BIV images. The ripples are not likely to be observed in the present study because the breaking wave has a wavelength of $L=1.84~\text{m}$ that is much longer than $0.77~\text{m}\leqslant L\leqslant 1.18~\text{m}$ generated in Duncan et al. (Reference Duncan, Qiao, Philomin and Wenz1999). Relatively longer-wavelength spillers, such as those in the present study, have weaker dependence on the surface tension forces. As a result, bubbles and spray associated with the intense splashing motions are produced (Duncan Reference Duncan2001), and the effects of the bubbles and spray on the flow become significant.

Figure 7.

Mean vorticity fields ( $\text{s}^{-1}$ ) at (a)  $t/T=0.06$ , (b) 0.08, (c) 0.10, (d) 0.12, (e) 0.14, (f) 0.16, (g) 0.17, (h) 0.19, (i) 0.21, (j) 0.23, (k) 0.25 and (l) 0.27. The estimated uncertainty for vorticity using bootstrapping is $0.30~\text{s}^{-1}$ , equivalent to $0.05C/H$ .

3.3 Wavelet-educed turbulence length scale

Whether a breaking wave has sufficient turbulent kinetic energy to overcome the stabilizing effects of gravity and surface tension depends on the characteristic turbulent length scale (Brocchini & Peregrine Reference Brocchini and Peregrine2001). If turbulent kinetic energy exceeds the stabilizing energy due to gravity and surface tension, the surface breaks up into bubbles and droplets. In the present study, we used a wavelet-based method to compute the turbulent length scale under the spilling breakers.

Coherent structures can be educed using a wavelet transform with the Morlet mother wavelet defined as

(3.6) $$\begin{eqnarray}\unicode[STIX]{x1D711}(z)=\text{e}^{iw_{0}z}\text{e}^{-z^{2}/2},\end{eqnarray}$$

where $z$ is the position of the signal for the (different) window of the mother wavelet, and $w_{0}=6$ is suggested to satisfy the admissibility condition (Farge Reference Farge1992). The wavelet coefficient of a velocity signal is then defined as the following using the continuous wavelet transform (Farge Reference Farge1992):

(3.7) $$\begin{eqnarray}W_{f}(s,z)=\frac{1}{\sqrt{s}}\int _{-\infty }^{\infty }u(\unicode[STIX]{x1D70F})\unicode[STIX]{x1D711}^{\ast }\left(\frac{\unicode[STIX]{x1D70F}-z}{s}\right)\,\text{d}\unicode[STIX]{x1D70F},\end{eqnarray}$$

where $u(\unicode[STIX]{x1D70F})$ is the horizontal velocity, $s$ is the scale dilation parameter, $\unicode[STIX]{x1D70F}$ is the translation parameter, $\sqrt{s}$ is for energy normalization across the different scales, $^{\ast }$  denotes the complex conjugate, and the integrand represents a convolution product between the dilated and the translated counterpart of the complex conjugate of the mother wavelet. A quantitative local intermittency measure $(LIM)$ introduced by Farge (Reference Farge1992) is defined as follows:

(3.8) $$\begin{eqnarray}LIM(s,z)=\frac{[W_{f}(s,z)]^{2}}{\langle [W_{f}(s,z)]^{2}\rangle _{z}},\end{eqnarray}$$

where $\langle [W_{f}(s,z)]^{2}\rangle _{z}$ is the average of the squared wavelet coefficient along the $z$ direction. The $LIM$ peak among different scales can be formulated as

(3.9) $$\begin{eqnarray}LIMM(z)=\max \{LIM(s,z)\}.\end{eqnarray}$$

The value of $LIMM$ represents the energy level of the most excited mode among the scale bands. Camussi & Felice (Reference Camussi and Felice2006) reported that inverting the scale dilation parameter of $LIMM$ directly gives the length scale ( $LS$ ) of the identified vortex. Following Na et al. (Reference Na, Chang, Huang and Lim2016), we used a conditional threshold, $LIM=0.01$ , assigned to the columns with $\langle [W_{f}(s,z)]^{2}\rangle _{z}<0.01$ , to separate the laminar flow region in which the $LIM$ approach is not applicable. Na et al. (Reference Na, Chang, Huang and Lim2016) reports that $LS$ not only represents the length scale of the most energetic eddies, but also is comparable to the integral length scale. In the present study, the obtained instantaneous $LIMM$ and $LS$ are ensemble-averaged to obtain the mean $LIMM$ and  $LS$ .

In figure 8, the ensemble-averaged length scale $LS$ , normalized by the wave height $H$ , is presented. The $H/LS$ plots were selected such that the wave front approximately locates at the middle of each FOV. The normalized length scale of eddies inside the aerated region ranges between $H/LS=5$ and 12 (i.e. length scale $LS=2{-}5~\text{cm}$ ) during the entire breaking processes. The length scale increases near the lower boundary of the aerated region. Govender et al. (Reference Govender, Mocke and Alport2004) showed that the length scale increases downwards from crest to trough, ranging from approximately $LS=0.1h$ to $0.4h$ (approximately $0.1h$ at the crest) under surf-zone spillers. This length scale is similar to the value found in the present results (approximately $0.07h$ at the crest). Previous studies (e.g. Govender et al. Reference Govender, Mocke and Alport2004; Longo Reference Longo2009; Huang et al. Reference Huang, Hwung and Chang2010) reported that the turbulence length scales are less than the wave height. The normalized turbulent length scales are in the range of $LS/H=0.05$ to 0.3 in Govender et al. and Huang, and $LS/H=0.02$ to 1.0 in Longo. Note that the laboratory spilling breaker heights vary from $H=3$ to 16 cm among these three studies. In the present study featuring a much greater wave height of $H=26.5~\text{cm}$ , the turbulence length scale is in the range of $LS/H=0.08{-}0.2$ under the spillers, comparable to the reported length scales above.

Figure 8.

Normalized length scale ( $H/LS$ ) at (a)  $t/T=-0.01$ (FOV1), (b) 0.14 (FOV3), (c) 0.28 (FOV5), (d) 0.39 (FOV7), (e) 0.66 (FOV9), (f) 0.81 (FOV11), (g) 0.95 (FOV13), (h) 1.17 (FOV15), (i) 1.46 (FOV17), (j) 1.63 (FOV19), (k) 1.89 (FOV21) and (l) 2.14 (FOV23).

When compared to the values of $H/LS=7{-}20$ in the highly aerated region of plunging breakers in Na et al. (Reference Na, Chang, Huang and Lim2016), the $LS$ values are comparable but confined to a narrower range in the present spilling breakers. Interestingly, figure 8(a) depicts that $H/LS$ just downstream of the toe increases to $H/LS=4{-}7$ near the onset of breaking. Duncan (Reference Duncan2001) discussed that for long-wavelength spilling breakers, like the ones in the current study, the spilling process starts with the appearance of a rough surface, which probably causes the increased turbulent length scales on the forward face of the crest. In the present study, the $LS$ value stays less than $0.2H$ above the shear zone between the fast-moving aerated region and the quiescent water region below (figure 8 b–j). By examining figures 5 and 8, one can see that the region above the lower boundary of the white foamy region exhibits a characteristic length scale of approximately $LS/H\sim 0.1{-}0.2$ (with some variations inside) in figure 8. However, in figure 8 the length scale immediately to the left of the toe is almost one order of magnitude greater $(LS/H\sim 1{-}2)$ . We used a conditional threshold, $LIM=0.01$ , assigned to the columns with $\langle [W_{f}(s,z)]^{2}\rangle _{z}<0.01$ , to separate the laminar flow region in which the $LIM$ approach is not applicable. The abrupt change of the length scale near the toe is due to $\langle [W_{f}(s,z)]^{2}\rangle _{z}$ being barely below 0.01 at this location. The discrete jump in the conditional threshold approach forces this abrupt and unrealistic change.

4.1 Void fraction and its effects on flow structures

The FOR technique was used to resolve phases (air and water in this study) with a high temporal resolution. With 20 repeated instantaneous void fraction ( $\unicode[STIX]{x1D6FC}$ ) measurements at each measurement point, the mean void fraction is computed by ensemble averaging, i.e.

(4.1) $$\begin{eqnarray}\langle \unicode[STIX]{x1D6FC}(x,z,t)\rangle =\frac{1}{N}\mathop{\sum }_{i=1}^{N}\unicode[STIX]{x1D6FC}_{i}(x,z,t),\end{eqnarray}$$

where $\langle \unicode[STIX]{x1D6FC}(x,z,t)\rangle$ is the mean void fraction, $\unicode[STIX]{x1D6FC}_{i}$ is the $i$ th local instantaneous void fraction and $N=20$ is the number of repeat runs in the experiment. To match the temporal resolution of the PIV velocity measurements, the original sampling rate of 100 kHz was averaged over every 1000 points, resulting in a final rate of 100 Hz for $\langle \unicode[STIX]{x1D6FC}\rangle$ . Furthermore, the wave-averaged (wet-period-averaged) void fraction $\unicode[STIX]{x1D6FC}_{wa}$ is defined as

(4.2) $$\begin{eqnarray}\unicode[STIX]{x1D6FC}_{wa}(x,z)=\frac{\displaystyle \int _{t_{tr}(x,z)}^{t_{tr}(x,z)+T}\unicode[STIX]{x1D6FF}(x,z,t)\langle \unicode[STIX]{x1D6FC}\rangle (x,z,t)\,\text{d}t}{\displaystyle \int _{t_{tr}(x,z)}^{t_{tr}(x,z)+T}\unicode[STIX]{x1D6FF}(x,z,t)\,\text{d}t},\end{eqnarray}$$

where $t_{tr}(x,z)$ is the time when the front trough reaches the specific FOR measurement cross-section and $T$ is the wave period. Here $\unicode[STIX]{x1D6FF}(x,z,t)=1$ when the point is in water and $\unicode[STIX]{x1D6FF}(x,z,t)=0$ otherwise.

After wave breaking, the entrained bubble cloud penetrates downwards. In the present spilling breakers, it is interesting to note that the maximum penetration depth (i.e. the maximum depth at which the bubble cloud was detected by the FOR probe) remained above the still-water level at all three FOR stations (station 1 at $x=0.5L$ , station 2 at $x=L$ and station 3 at $x=1.55L$ ) during the entire breaking process. The penetration depth at FOR stations 1, 2 and 3 reached $z=5~\text{cm}$ , $z=2~\text{cm}$ and $z=2~\text{cm}$ , respectively. Void fraction measurements below these depths were also conducted (but not shown here) to ensure that the void fraction was negligibly small. This is in accordance with the finding in Rojas & Loewen (Reference Rojas and Loewen2010), which reported a maximum penetration depth of $z=5~\text{cm}$ under mechanically generated spilling breakers. The current results are also not significantly different from the maximum penetration depth of $z=3~\text{cm}$ under wind-generated breaking waves reported by Leifer & de Leeuw (Reference Leifer and de Leeuw2006).

Figure 9.

Vertical profiles of (a) wave-averaged void fraction ( $\unicode[STIX]{x1D6FC}_{wa}$ ) and (b) corresponding normalized wave-averaged void fraction. The void fraction is normalized by the maximum void fraction and the vertical axis is shifted with the penetration depth $(z_{B})$ and then normalized by the cloud thickness $(d_{B})$ at each FOR station. The thick dashed line in (b) is a linear least-squares fit over all the data points, expressed as $\unicode[STIX]{x1D6FC}_{wa}/\text{max}(\unicode[STIX]{x1D6FC}_{wa})=0.82(z-z_{B})/d_{B}$ . The error bars are standard deviations of the averaged quantities.

Figure 10.

Temporal evolution of void fraction at (a) FOR station 1 ( $x/L=0.50$ ), (b) FOR station 2 ( $x/L=1.00$ ) and (c) FOR station 3 ( $x/L=1.55$ ). The dashed line is the contour boundary of $\langle \unicode[STIX]{x1D6FC}\rangle =0.2$ , and the solid line is the contour boundary of $\langle \unicode[STIX]{x1D6FC}\rangle =0.1$ . (d) Estimated bubble plume thickness at each FOR station.

Figure 9(a) shows the vertical profiles of the wave-averaged void fraction measurements at FOR stations 1 ( $x=0.5L$ ), 2 ( $x=L$ ) and 3 ( $x=1.55L$ ); the corresponding maximum wave-averaged void fraction at each FOR station is 0.25, 0.25 and 0.33, respectively. In addition, the bubble cloud thicknesses (i.e. the vertical distance from the highest to the lowest edges of the cloud) are 7–9 cm. The cloud thickness was found to depend on the wind speed (the thickness increases with the increase of wind speed in general) and vary from 5 cm to 17 cm (Anguelova & Huq Reference Anguelova and Huq2012). These measured bubble cloud thicknesses also agree with those reported in Koga (Reference Koga1982), Thorpe (Reference Thorpe1982) and Kalvoda et al. (Reference Kalvoda, Xu and Wu2003). Rojas & Loewen (Reference Rojas and Loewen2010) measured the time-averaged void fractions of 0.17, 0.29, 0.20 and 0.26 at $x/L=0.31$ , 0.57, 0.83 and 1.08. In their study, instantaneous void fractions were averaged over the time interval only when the void fraction was less than 0.5 and, at each horizontal location, void fraction measurements were taken at only one vertical location, determined by the occurrence of the largest number of bubbles. The present study ( $ka=0.45$ , $H=0.265~\text{m}$ ) may be directly comparable to that of Rojas & Loewen (Reference Rojas and Loewen2010) because of the similar wave properties used $(ka=0.38,H=0.200~\text{m})$ in their study. The measured maximum void fraction of 0.25 in the present study is similar to that measured by Rojas & Loewen (Reference Rojas and Loewen2010) at $x/L=0.5$ and 1.0, but reaches 0.33 at $x/L=1.5$ in the present study (no data from Rojas & Loewen at this location). However, void fraction estimates by Leifer & de Leeuw (Reference Leifer and de Leeuw2006) under wind-generated spillers are lower than our values by one to two orders of magnitude, partly because their void fraction estimates are not time-averaged values. It is expected that less air entrainment occurs during breaking for less energetic spillers. The wavelength and the wave height of the wind-generated spillers in Leifer & de Leeuw (Reference Leifer and de Leeuw2006) are 1.3 m and 0.12 m, respectively; these are shorter than those in the present study. The discrepancy is supported by the void fraction comparison in Rojas & Loewen (Reference Rojas and Loewen2010) and Loewen, O’Dor & Skafel (Reference Loewen, O’Dor and Skafel1996). The latter under very gentle spillers has void fraction approximately two orders of magnitude lower than that of the former under similar spillers as in the present study.

Figure 9(b) shows the normalized vertical profiles of the wave-averaged void fraction at the three FOR stations. At each station, the void fraction was normalized by the corresponding maximum void fraction, and the vertical axis was shifted with the penetration depth $z_{B}$ and then normalized by the corresponding cloud thickness $d_{B}$ . Figure 9(b) suggests that the normalized void fractions are self-similar, even though they are one wavelength apart at $x=0.5L$ , $1.0L$ and $1.55L$ . The normalized void fraction can be formulated as

(4.3) $$\begin{eqnarray}\unicode[STIX]{x1D6FC}_{wa}/\text{max}(\unicode[STIX]{x1D6FC}_{wa})=0.82(z-z_{B})/d_{B}.\end{eqnarray}$$

The self-similar behaviour was also reported in the first, second and third splash-ups for plunging breaking waves in deep water by Lim et al. (Reference Lim, Chang, Huang and Na2015).

Figure 10 shows the temporal evolution of void fraction at each FOR station. The contours were obtained with a temporal resolution of 0.01 s and a spatial resolution of 10 mm. At FOR station 1 $(x/L=0.5)$ , a void fraction higher than 0.6 is found in the lower half of the aerated region (at $t=0.43{-}0.47~\text{s}$ or $t/T=0.39{-}0.43$ in figure 10 a). In the upper half of the aerated region, the void fraction is relatively lower (around 0.4–0.5). Bubbles are concentrated near the shear zone and fragmented during this initial phase. Leifer & de Leeuw (Reference Leifer and de Leeuw2006) refer to this as the ‘injection phase’ when bubbles are advected downwards after the bubble plume is formed and before the bubbles rise. We believe that the high void fraction in the lower half of the plume is strongly related to the injection phase.

At FOR station 2 $(x/L=1.0)$ , the contour in figure 10(b) shows that the discrepancy in void fraction between the upper half and the lower half of the aerated region is less distinguishable as moderately high void fraction appears sporadically within the region. At this stage, the bubble plume is likely to be in transition from the ‘rise phase’ (Leifer & de Leeuw Reference Leifer and de Leeuw2006) to the ‘senescence phase’, when the aerated region is mixed with dispersed larger bubbles that are rising due to buoyancy and smaller bubbles that are left at deeper locations. During the rise phase, the dominant advective process is buoyant rise for the larger bubbles in the plume; while during the senescence phase, turbulence and wave motion are important (Leifer & de Leeuw Reference Leifer and de Leeuw2006). During this transition phase, buoyancy, turbulence and wave motion are likely to affect the trajectories of the bubbles, resulting in the unorganized aeration levels within the region as shown in figure 10(b).

At FOR station 3 $(x/L=1.55)$ , figure 10(c) may indicate that the majority of the larger bubbles have approached or risen and burst at the free surface, but the remaining smaller bubbles rise more slowly. The highly aerated layer (the warm colour region centred at around $z=0.08~\text{m}$ ) at the left face of figure 10(a) initially stretched vertically with high void fraction, then gradually deformed into two separate patches – centred at around $z=0.05~\text{m}$ and $z=0.10~\text{m}$ in figure 10(b) and then around $z=0.04~\text{m}$ and $z=0.09~\text{m}$ in figure 10(c) – with moderate void fraction divided by a relatively lower aerated area in between the patches. This feature was not observed in the plunging breakers (Lim et al. Reference Lim, Chang, Huang and Na2015; Na et al. Reference Na, Chang, Huang and Lim2016). Leifer & de Leeuw (Reference Leifer and de Leeuw2006) referred this to the ‘senescence phase’. In the present spillers, once the larger bubbles rise to the surface, the remaining smaller bubbles rise at their relatively slow velocities and persist within the cloud. In contrast, in plungers, the plunging jets continuously introduce large bubbles from the splash-ups and breaking-up of entrapped air cavities. The large bubbles rise faster and scavenge the smaller, slowly rising bubbles, leaving a relatively lower number of smaller bubbles (Anguelova & Huq Reference Anguelova and Huq2018).

Blenkinsopp & Chaplin (Reference Blenkinsopp and Chaplin2007) used a threshold void fraction of 0.5 under plunging breaking waves to separate the highly aerated region (splash-up) above the free surface and moderately aerated region (impinging) below the free surface. As the void fraction level is lower in the present spilling breakers, we used a threshold of $\langle \unicode[STIX]{x1D6FC}\rangle =0.2$ in figure 10 (shown as a dashed line) to approximately mask the weakly aerated region. For each vertical measurement location, the time elapsed from detecting the arrival of the wave front to this dashed line is considered as the duration when the probe is in an ‘active bubble plume’. Lamarre & Melville (Reference Lamarre and Melville1992), Stansell & MacFarlane (Reference Stansell and MacFarlane2002) and Anguelova & Huq (Reference Anguelova and Huq2012) reported that the horizontal velocity of the bubble plume normalized by the phase speed of the wave is between 0.5 and 1. The mid-value of 0.75 was used to approximate the bubble plume horizontal thickness $\unicode[STIX]{x1D6FF}$ . The time interval in which we assumed that the plume is moving with a speed of $0.75C$ is essentially the coloured region in figure 10 from the front surface (the left surface) to the yellow solid line (with a mean void fraction = 0.1) in each panel. The mean time interval used is $0.07T$ , $0.11T$ and $0.10T$ at FOR station 1, 2 and 3, respectively. Figure 10(d) shows the bubble plume thickness at the three FOR stations. At FOR station 1, the vertical profile of the thickness is ‘curving upwards’ (shaped like the right half of $U$ ) from the lower boundary to approximately $z/H=0.3$ , above which the bubble plume thickness is nearly constant. The plume thickness is in transition at FOR station 2, eventually transforming to ‘curving downwards’ (shaped like the left half of an upside-down U) at FOR station 3 after roughly one wavelength downstream from FOR station 1.

Rojas & Loewen (Reference Rojas and Loewen2010) reported that the minimum averaged void fraction ranges approximately from $\langle \unicode[STIX]{x1D6FC}\rangle =0.1$ to 0.2 at four measurement points under the spillers. We noticed that the contour lines of $\langle \unicode[STIX]{x1D6FC}\rangle =0.1$ and $\langle \unicode[STIX]{x1D6FC}\rangle =0.2$ show distinct evolution patterns, as shown specifically in figure 11(b,c) of Rojas & Loewen (Reference Rojas and Loewen2010). Accordingly, we used the lower bound of $\langle \unicode[STIX]{x1D6FC}\rangle =0.1$ (and below) as a criterion to define the weak bubble plume and the higher bound of $\langle \unicode[STIX]{x1D6FC}\rangle =0.2$ (and above) to define the active bubble plume. Thus, the volume of the active bubble plume was estimated by integrating the thickness over $z$ using the trapezoidal method based on the defined threshold of $\langle \unicode[STIX]{x1D6FC}\rangle =0.2$ . Similarly, the volume of the weak bubble plume was estimated using a threshold of $\langle \unicode[STIX]{x1D6FC}\rangle =0.1$ based on the same procedure as above. The estimated volumes of the active bubble plume per unit width are $0.0059~\text{m}^{2}$ , $0.0054~\text{m}^{2}$ and $0.0043~\text{m}^{2}$ at $x=0.5L$ , $x=L$ and $x=1.55L$ , respectively, while the estimated volumes of plume per unit width are $0.0078~\text{m}^{2}$ , $0.0078~\text{m}^{2}$ and $0.0062~\text{m}^{2}$ . Thus, the ratios of the weak bubble plume to the plume volume are 0.76, 0.69 and 0.69. The decreasing rate of these ratios is more significant at the earlier stage of breaking ( $x/L=0.5$ –1.0), but insignificant at the later stage of breaking ( $x/L=1.0$ –1.55). Lamarre & Melville (Reference Lamarre and Melville1992), Leifer & de Leeuw (Reference Leifer and de Leeuw2006) and Blenkinsopp & Chaplin (Reference Blenkinsopp and Chaplin2007) showed that the normalized volume of bubble plume follows an exponential decay. However, the number of measurement locations in the present study is too small to observe that.

Figure 11.

Vertical profiles of time-averaged quantities: (a,d,g) void fraction; (b,e,h) turbulent intensity; and (c,f,i) swirling strength. (a–c) FOR station 1 ( $x/L=0.50$ ); (d–f) FOR station 2 ( $x/L=1.00$ ); and (g–i) FOR station 3 ( $x/L=1.55$ ). Horizontal lines indicate standard deviations. The horizontal error bars are standard deviations of the time-averaged quantities.

Based on the local velocity gradient tensors, the swirling strength $\unicode[STIX]{x1D713}$ , defined as the imaginary eigenvalue of the local deformation matrix

(4.4) $$\begin{eqnarray}DM=\left[\begin{array}{@{}cc@{}}\displaystyle \frac{\unicode[STIX]{x2202}u}{\unicode[STIX]{x2202}x} & \displaystyle \frac{\unicode[STIX]{x2202}u}{\unicode[STIX]{x2202}z}\\ \displaystyle \frac{\unicode[STIX]{x2202}w}{\unicode[STIX]{x2202}x} & \displaystyle \frac{\unicode[STIX]{x2202}w}{\unicode[STIX]{x2202}z}\end{array}\right]\end{eqnarray}$$

(Zhou et al. Reference Zhou, Adrian, Balachandar and Kendall1999; Adrian, Christensen & Liu Reference Adrian, Christensen and Liu2000; Camussi Reference Camussi2002) can be computed. The swirling strength is proportional to the rotational speed of the vortical structures and has been proven useful in revealing the local vortical structures (Adrian et al. Reference Adrian, Christensen and Liu2000). Adrian et al. (Reference Adrian, Christensen and Liu2000) showed experimentally using PIV velocity maps that swirling strength is less noisy in identifying eddies and more effective in isolating eddies from local shear layers. On the contrary, vorticity identifies not only vortex cores but also any shearing motion present in the flow. Figure 11 shows the time-averaged quantities of void fraction, turbulent intensity and swirling strength in the active bubble plume region confined by the $\langle \unicode[STIX]{x1D6FC}\rangle =0.2$ threshold. A time-averaged quantity is defined as

(4.5) $$\begin{eqnarray}q_{ta}(x,z)=\frac{\displaystyle \int _{t_{tr}(x)}^{t_{tr}(x)+T}\unicode[STIX]{x1D709}(x,z,t)q(x,z,t)\,\text{d}t}{\displaystyle \int _{t_{tr}(x)}^{t_{tr}(x)+T}\unicode[STIX]{x1D709}(x,z,t)\,\text{d}t},\end{eqnarray}$$

where $q_{ta}$ is the time-averaged quantity (i.e. void fraction, turbulent intensity and swirling strength) and $\unicode[STIX]{x1D709}(x,z,t)=1$ when $\langle \unicode[STIX]{x1D6FC}\rangle \geqslant 0.2$ and $\unicode[STIX]{x1D709}(x,z,t)=0$ when $\langle \unicode[STIX]{x1D6FC}\rangle <0.2$ . As discussed in figure 11, the transition of the vertical profiles of void fraction from curving upwards to curving downwards can be seen in figure 11(a,d,g). The vertical profiles of void fraction, turbulent intensity and swirling strengths (especially the latter two) show depending trends and similar locations of the local peaks at all three FOR stations. In particular, the secondary local peaks located at $z=0.1~\text{m}$ at FOR station 2 are consistent among void fraction, turbulent intensity and swirling strengths in figure 11(d–f). These equivalent local peaks, caused by the weak jet impingement of the breaking waves, are barely visible at $z=0.13~\text{m}$ at FOR station 1. The apparent secondary local peak at FOR station 2 is probably caused by the mild overturning and subsequent jet impingement. Although this weak overturning was not evidently confirmed by the PIV and BIV images, it is supported by the local peak of $B$ near $t/T=0.9$ as shown in figure 3. At FOR station 3, the void fraction increases near the surface because of the risen bubbles due to buoyancy. This may in turn increase the magnitudes of turbulent intensity and swirling strength as shown in figure 11(g–i). The results imply that turbulent intensity is positively correlated with void fraction under spilling breaking waves. Lim et al. (Reference Lim, Chang, Huang and Na2015) also reported a strong correlation between turbulent intensity and void fraction in highly aerated plunging breaking waves. However, Deane & Stokes (Reference Deane and Stokes2002) argued that the coincidence of high void fraction with high turbulence in breaking waves may not be sufficient to conclude that the turbulent properties are strongly enhanced by void fraction. Note that the distribution of the swirling strength in figure 11 is very similar to that of the vorticity magnitude (not plotted in the figure).

Although the present study was performed in fresh water, its goal is to deepen our understanding of the wave breaking process and its induced air entrainment in open oceans. It is worth pointing out the potential effects of salinity on void fraction and bubble generation. Anguelova & Huq (Reference Anguelova and Huq2018) experimentally showed that salinity alters the bubble size and number by performing a single case of jet impingement to mimic a plunging breaker. The number of large bubbles and void fraction increase with salinity, reaching the maximum at salinity of $S=12{-}25~\text{g}~\text{kg}^{-1}$ . Both the number of large bubbles and void fraction decrease with further increase of salinity beyond $S=25$ . They referred this to a non-monotonic dependence of the number of bubbles and void fraction on salinity, which may explain the previously contradictory literature (Monahan & Zietlow Reference Monahan and Zietlow1969; Carey et al. Reference Carey, Fitzgerald, Monahan and Wang1993; Monahan et al. Reference Monahan, Wang, Wang and Wilson1994). Although Anguelova & Huq (Reference Anguelova and Huq2018) tested only one case, they further argued that the dependence of their results on various breaking strengths may not be significant based on Blenkinsopp & Chaplin’s (Reference Blenkinsopp and Chaplin2007, Reference Blenkinsopp and Chaplin2011) results.

4.2 Wave energy variation considering void fraction

Void fraction is essential for investigating the flow properties that involve fluid density variation, such as the mean kinetic energy, turbulent kinetic energy and potential energy in the present study. Lim et al. (Reference Lim, Chang, Huang and Na2015) showed that the discrepancy of estimated kinetic energy with and without considering void fraction reached 52 % under plunging breaking waves. They concluded that the mixture density of highly aerated flows such as breaking waves must be accounted for in estimating energy and dissipation. However, density variations have rarely been considered in experimental studies of wave breaking. This hurdle is probably caused by difficulties in measuring both velocities and void fraction (and hence the mixture density) simultaneously in experiments.

In the present study, the mean kinetic energy per unit water mass $(K)$ was obtained from the mean horizontal velocity $(U)$ and the mean vertical velocity $(W)$ with the assumption that the mean cross-tank velocity $(V)$ is zero, i.e.

(4.6) $$\begin{eqnarray}K={\textstyle \frac{1}{2}}(U^{2}+W^{2}).\end{eqnarray}$$

The corresponding turbulent kinetic energy $k$ was approximated as

(4.7) $$\begin{eqnarray}k=\frac{1.33}{2}(\langle u^{\prime \,2}\rangle +\langle w^{\prime \,2}\rangle )=\frac{1.33}{2}I_{2D}^{2},\end{eqnarray}$$

where the constant 1.33 accounts for the missing $y$ -direction velocity (e.g. Chang & Liu Reference Chang and Liu1999). Note that the factor of 1.33, suggested by Svendsen (Reference Svendsen1987) and used in many wave breaking studies, has been verified for surf-zone plunging breaking waves (Christensen Reference Christensen2006).

The period-averaged mean kinetic energy $(K_{pa})$ and turbulent kinetic energy $(k_{pa})$ considering void fraction are given as

(4.8) $$\begin{eqnarray}\left.\begin{array}{@{}c@{}}\displaystyle K_{pa}(x,z)=\frac{\displaystyle \int _{t_{tr}(x)}^{t_{tr}(x)+T}\unicode[STIX]{x1D6FF}(x,z,t)(1-\langle \unicode[STIX]{x1D6FC}\rangle )K(x,z,t)\,\text{d}t}{\displaystyle \int _{t_{tr}(x)}^{t_{tr}(x)+T}\,\text{d}t},\\ \displaystyle k_{pa}(x,z)=\frac{\displaystyle \int _{t_{tr}(x)}^{t_{tr}(x)+T}\unicode[STIX]{x1D6FF}(x,z,t)(1-\langle \unicode[STIX]{x1D6FC}\rangle )k(x,z,t)\,\text{d}t}{\displaystyle \int _{t_{tr}(x)}^{t_{tr}(x)+T}\,\text{d}t}.\end{array}\right\}\end{eqnarray}$$

Furthermore, the depth-integrated mean kinetic energy $(K_{di})$ and turbulent kinetic energy $(k_{di})$ were computed by integrating the corresponding period-averaged quantities from the bottom to the free water surface $(\unicode[STIX]{x1D702})$ as

(4.9) $$\begin{eqnarray}\left.\begin{array}{@{}c@{}}\displaystyle K_{di}(x)=\int _{-h}^{\unicode[STIX]{x1D702}(x)}K_{pa}(x,z)\,\text{d}z,\\ \displaystyle k_{di}(x)=\int _{-h}^{\unicode[STIX]{x1D702}(x)}k_{pa}(x,z)\,\text{d}z.\end{array}\right\}\end{eqnarray}$$

The total kinetic energy $(KE)$ can be obtained by summing the mean kinetic energy and turbulent kinetic energy, i.e.  $KE=K_{di}+k_{di}$ . Note that in the present study the PIV measurement covered only from $z=-0.64h$ (equivalent to $0.28L$ ) to the free surface. The missing values between $z=-0.64h$ and $z=-h$ can be interpolated in the integration but their contributions are insignificant and thus neglected.

In the present study, the potential energy $(PE)$ was determined using the free-surface elevation from the images with void fraction considered as follows:

(4.10) $$\begin{eqnarray}PE=\int _{-h}^{\unicode[STIX]{x1D702}}(1-\langle \unicode[STIX]{x1D6FC}\rangle )gz\,\text{d}z-\frac{1}{2}gh^{2}.\end{eqnarray}$$

Lim et al. (Reference Lim, Chang, Huang and Na2015) showed that the presence of aeration in plunging breakers led to an underestimated potential energy when using only the free-surface elevation measured by resistance wave gauges, whereas using only the free-surface elevation obtained from images led to an overestimated value.

Figure 12(a) compares the variation of period-averaged, depth-integrated mean kinetic energy ( $K_{di}$ ) and turbulent kinetic energy ( $k_{di}$ ) with and without considering void fraction. If void fraction is not considered, the mean kinetic energy would be overestimated by 23 %, 49 % and 43 % at the three FOR stations at $x=0.5L$ , $L$ and $1.55L$ , respectively. The corresponding ratios of $k_{di}/K_{di}$ are 3 %, 5 % and 7 % with void fraction considered. These ratios are significantly lower than the 25 % ratio at the first impingement in the plunging breaking waves reported by Lim et al. (Reference Lim, Chang, Huang and Na2015). Note that, without considering void fraction, the ratios of $k_{di}/K_{di}$ increase slightly to 4 %, 8 % and 10 % in the present study. In figure 12(a), if the mean kinetic energy at the pre-breaking region at $x/L=-0.14$ is used as a reference with void fraction considered, the mean kinetic energy decreases by 16 %, 45 % and 58 % at $x/L=0.5$ , 1.0 and 1.55, respectively.

Figure 12.

(a) Comparison of depth-integrated mean kinetic energy ( $K_{di}$ ) and turbulent kinetic energy ( $k_{di}$ ) with and without considering void fraction. (b) Turbulent kinetic energy ( $k_{di}$ ) with and without considering void fraction.

Figure 13.

(a) Depth-integrated total kinetic energy ( $KE=K_{di}+k_{di}$ ) and (b) potential energy $(PE)$ with and without considering void fraction. (c) Total wave energy ( $E_{\unicode[STIX]{x1D6FC}}=KE+PE$ ) accounting for void fraction normalized by the corresponding pre-breaking wave energy ( $E_{L}$ ). The dashed line for $E_{\unicode[STIX]{x1D6FC}}/E_{L}$ is from the empirical formula $E_{\unicode[STIX]{x1D6FC}}/E_{L}=1.06\text{e}^{-0.55x/L}$ for $x/L>0.1$ and $E_{\unicode[STIX]{x1D6FC}}/E_{L}=1.0$ for $x/L\leqslant 0.1$ .

Figure 12(b) shows the variation of turbulent kinetic energy ( $k_{di}$ ) with and without considering void fraction. If void fraction is not considered, the turbulent kinetic energy would be overestimated by 69 %, 163 % and 104 % at the three FOR stations at $x=0.5L$ , $L$ and $1.55L$ , respectively, and probably more than 300 % at some other locations based on figure 12(b). Note that at the pre-breaking location of $x/L=-0.14$ the value of $k_{di}$ is not zero (also shown in figure 6). The value of $k_{di}$ then increases 47 %, 39 % and 57 % from the pre-breaking region, as shown in figure 12(b). The flow seems to maintain a somewhat even level of turbulent kinetic energy as the spilling wave propagates from $x=0.5L$ to $1.55L$ . This is consistent with the relatively even aeration levels measured at the three FOR stations shown in figure 9(a) and the implication between void fraction and turbulence intensity. Nevertheless, that trend is very different from the observation in plunging breaking waves, where the aeration level decreases significantly near $x=L$ (see figure 13a,b in Lim et al. Reference Lim, Chang, Huang and Na2015).

Figure 13(a) shows the variation of depth-integrated total kinetic energy ( $KE=K_{di}+k_{di}$ ) with and without considering void fraction. If void fraction is not considered, the total kinetic energy would be overestimated by 25 %, 56 % and 47 % at $x/L=0.5$ , 1.0 and 1.55, respectively. With the density variation accounted for, the total kinetic energy drops 15 %, 44 % and 56 % at $x/L=0.5$ , 1.0 and 1.55, respectively, when compared to the total kinetic energy at the pre-breaking region at $x/L=-0.14$ . On the contrary, in plunging breaking waves the total kinetic energy increases by 18 % at $x/L=0.4$ because of the significant increase in the turbulent kinetic energy during the first impingement and splash-up processes (Lim et al. Reference Lim, Chang, Huang and Na2015). This comparison shows that the production of the turbulent kinetic energy in the spilling breakers is not as significant as that in the plunging breakers.

Figure 13(b) compares the spatial variation of potential energy $(PE)$ with and without considering void fraction. If void fraction is not considered, the potential energy, based on video images, would be overestimated by 32 %, 47 % and 39 % at $x/L=0.5$ , 1.0 and 1.55, respectively. With the density variation accounted for, the potential energy drops 22 %, 42 % and 47 % at $x/L=0.5$ , 1.0 and 1.55, respectively, when compared to the pre-breaking potential energy at $x/L=-0.14$ . Interestingly, the potential energy has a significant decrease between $x/L=0$ and $x/L=1.0$ but only a slight decrease between $x/L=1.0$ and $x/L=1.55$ . Similarly, in plunging breaking waves, the potential energy has a significant decrease of approximately 32 % between $x/L=0.4$ and $x/L=0.8$ and a slight decrease of 2 % between $x/L=0.8$ and $x/L=1.1$  (Lim et al. Reference Lim, Chang, Huang and Na2015).

Figure 13(c) shows the spatial variation of potential energy, total kinetic energy and total energy $(KE+PE)$ accounting for the mixture density. The total energy decreases relatively quickly before $x/L=1$ and then more slowly beyond that point. In particular, the total energy drops 19 %, 43 % and 52 % at $x/L=0.5$ , 1.0 and 1.55 when compared to the total energy at the pre-breaking region at $x/L=-0.14$ . The total energy is dissipated following an exponential decay similar to that under plunging breaking waves in Lim et al. (Reference Lim, Chang, Huang and Na2015). The wave energy variation considering void fraction under spilling breakers may be formulated as

(4.11) $$\begin{eqnarray}E_{\unicode[STIX]{x1D6FC}}/E_{L}=1.06\text{e}^{-0.55x/L}\quad \text{for }x/L>0.1,\end{eqnarray}$$

where $E_{\unicode[STIX]{x1D6FC}}$ is the breaking wave energy with void fraction considered and $E_{L}$ is the pre-breaking wave energy. For $x/L\leqslant 0.1$ , $E_{\unicode[STIX]{x1D6FC}}/E_{L}=1.0$ is assumed, based on the curve fitting. Note that this exponential decay rate of $-0.55$ in the equation is approximately one-half of the rate of $-1.17$ in the plunging breaking waves in Lim et al. (Reference Lim, Chang, Huang and Na2015).

The spatial variations of total kinetic energy and potential energy in figure 13(c) exhibit similarities which suggest that the equipartition assumption may be valid not only outside of the actively breaking zone, but also inside the active aerated region under the spilling breakers. In contrast, Lim et al. (Reference Lim, Chang, Huang and Na2015) showed that the equipartition assumption is not applicable during the highly aerated phase and is valid only beyond $x/L=2$ with a relatively low aeration level under the plunging breaking waves. Based on figure 13(c), approximately 65 % and 71 % of the total energy is dissipated at $x/L=2.0$ and 2.3, respectively.

Based on figure 13(c), the rate of loss of total energy (i.e. the non-dimensional slope $\text{d}(E_{\unicode[STIX]{x1D6FC}}/E_{L})/\text{d}(x/L)$ ) for the spiller is milder near the breaking point at $x=0.5L$ when compared to that for the plunger in Lim et al. (Reference Lim, Chang, Huang and Na2015). The difference in the slope between the spiller and the plunger becomes less prominent near $x=1.5L$ . Of particular interest is that the magnitude of the slope showed an increase during the active breaking stage around $0.5L$ , followed by a decrease around $1.0L$ in both the spiller and the plunger. The measured data confirm the simulated results in Derakhti & Kirby (Reference Derakhti and Kirby2014) in which the dissipation rate increased during active breaking and decreased at the end of active breaking.

The ratio of potential energy to total energy has been known to depend on the breaking strength. Derakhti & Kirby (Reference Derakhti and Kirby2016) found that the ratio of potential energy density to total energy density, $PE/E$ , varied from approximately 0.6 in the pre-breaking stage to approximately 0.5 near the breaking onset with presence of breaking-induced bubbles. Derakhti & Kirby (Reference Derakhti and Kirby2016) suggested that $PE/E$ varies but is centred around 0.6 in the spiller. However, figure 13(c) in the present study shows that this ratio stays close to 0.5 (i.e. following equipartition) throughout the breaking process in the spiller. In the plunger, Lim et al. (Reference Lim, Chang, Huang and Na2015) reported that $PE/E$ decreases from 0.5 in the pre-breaking stage to around 0.4 near the onset. These are inconsistent with the reported values in Derakhti & Kirby (Reference Derakhti and Kirby2016). Additionally, a steep increase of $PE/E$ near $t/T=1$ was not observed in the present study.