Deltares Individual Wave Overtopping Tests

Hydralab+, Recipe Task 8.2: Video imagery for estimating individual wave overtopping volumes over coastal structures


Coastal structures, like breakwaters and revetments, are designed to allow for a specified maximum overtopping rate during design conditions. Due to sea level rise, the difference between the level of the crest and the sea water level will decrease, while wave conditions increase due to greater water depth. These developments potentially threaten the stability of coastal structures worldwide. For risk and safety assessment or design studies, the roughness parameter of the front slope is a key parameter in determining the overtopping rates for coastal structures. In Capel (2015) it was shown that due to an increase in flow depth, i.e. the thickness of the run-up tongue, the effective roughness of the front slope reduces. The roughness of the front slope is not a constant factor, but reduces for larger overtopping rates due to an increase in flow depth. The present roughness values, as given for instance in the EurOtop (2016) manual, are therefore only valid (and sometimes conservative) for typically mild overtopping rates of 0.1 to 1 l/s/m. A correct assessment of the overtopping rates is a key parameter when it comes to risk and safety assessment. In many areas, governmental organizations are keen on allowing reduced maximum overtopping rates when the risk to casualties or damage to properties is limited. This could mean that crest levels will not be adjusted in case of sea level rise. However, in the decision-making process, correct values of overtopping should be available. Using typical roughness values, that are only valid up to overtopping rates of 1 l/s/m, there is likely to be an underestimation which may lead to incorrect decisions with negative consequences in the future.

In this study, the possibilities of deriving flow depths and bore speeds along a coastal structure from high speed video imagery is explored. Automated semantic segmentation techniques are used to detect the instantaneous flow depth, bore speed of the wave run-up of individual waves, which potentially provides all parameters to validate the individual overtopping quantities for each wave and the roughness parameter as a function of flow depth on complex structures. Semantic segmentation is a form of image classification that is widely used in computer vision science. It aims to subdivide images or videos in meaningful (semantic) regions.

Tests and experiments

A series of 27 tests in the Scheldt Flume at Deltares were used in this study. The tests were characterized by a breakwater with a smooth slope that was exposed to approximately 1000 waves. Between the different tests the water levels (3), wave height (3) and wave steepness (3) varied. In addition, the same series of 27 tests (and 6 repetitions) with the same wave forcing and similar breakwater shape, but with a rubble mount slope and crest wall, was used to broaden the applicability of the automated semantic segmentation techniques to more turbulent waves. These tests were used for model training, but model testing focused on the smooth slope tests since a better model/data comparison was possible. Data were acquired from 150 Hz high-speed video footage through the glass flume wall using a single camera mounted on a tripod approximately 1 m from the flume (Figure 1). A flow depth measurement at the breakwater crest is used to order the waves on impact. From the top 100 waves with largest flow depth in each test 50 waves are used in the analysis. Figure 1 shows an overtopping wave.

Figure 1 Example of high-speed video footage. Red dots are reference points for calibrating the images. Green line is generated by the deep image segmentation algorithm technique

Data Processing and Model training

The high-speed video footage of a single test is split into individual wave recordings. For each wave, the video footage from 1 s prior to 1 s after the recorded impact peak is taken, resulting in fifty 2-second (or 300 frame) videos of individual waves per test. The per-wave videos are subsequently converted to 300 timestacks with the vertical dimension and time preserved, but the horizontal dimension discarded (See Figure 2, left). By using timestacks rather than the full video footage, segmentation algorithms can be much simpler (2D vs. 3D), but relations between adjacent pixels in horizontal direction are effectively discarded in favour of relations in time. In each timestack, the passing wave is detected by the algorithm and thus the time-variation in local flow depth is determined. For each wave, the maximum local flow depth at the breakwater crest is then compared to the measured maximum local flow depth in the flume. Typical segmentation results are shown in Figure 2.

Figure 2 Examples of a 300x600 timestack. Vertical axis is in space, horizontal axis is in time. Left panel of a set: original timestack. Right pane of a set: resulting binary mask.


About 120 timestacks from 1356 waves in 27 experiments with a smooth slope were analysed. All of the analysed timestacks provided sensible results in the sense that a wave-like shape was detected. A spatiotemporal description for each wave can be obtained by stacking the water level positions detected in each timestack corresponding to a single wave. Figure 3 visualizes the largest waves from 4 tests with a smooth slope and maximum still water level, but varying wave steepness and significant wave height.

Figure 3 Examples of waves detected by the deep image segmentation algorithm. Each subplot depicts 6 frames from the total of 60 frames obtained in the 0.4 s before the maximum flow depth was recorded at the physical measurement location.

Figure 4 visualizes a per-wave comparison between video-based measurements and physical measurements. The figure depicts the maximum flow depth per wave per test as obtained from video-based measurements compared to the physical measurements. The figure shows that video-based measurements in general compare well to physical measurements, but overestimates the flow depth due to wave splashes.

Figure 4 Video-based vs. physically measured flow depths at breakwater crest for the tests with smooth slope.

The algorithm provides time series of flow depths for every position along the flume axis. If all flow depth time series of an individual wave are stacked, a spatiotemporal description of the flow depth for an individual wave is obtained (Figure 5). From this spatiotemporal description, physical parameters per wave, like the time-averaged bore speed (=red line, being the slope of the bore depth), overtopping duration (=green line), time-averaged flow depth (=bluish colour) and ultimately the total overtopping volume can be derived.

Figure 5 From time-stacks to spatiotemporal representation of a single wave overtopping event.

Figure 6 gives a visual comparison between physically measured overtopping volume time series and the video-based overtopping volumes, which are here drawn at 1 time position.

Figure 6 Comparison between physically and video-based measured overtopping volumes in tiem. Note that the physically measured overtopping time series is detrended for better comparison. Therefore, inclining sections correspond to relatively large overtopping events, horizontal sections correspond to average overtopping events and declining sections roughly correspond to no overtopping events.

The first application of an automated semantic segmentation algorithm in the Deltares facilities has given very promising results. The new technique will be beneficial in assessing future coastal risks.


Capel, A. (2015). Wave run-up and overtopping reduction by block revetments with enhanced roughness. Coastal Engineering 104, p.76-92, Elsevier.

EurOtop, (2016). Manual on wave overtopping of sea defences and related structures. An overtopping manual largely based on European research, but for worldwide application. Van der Meer, J.W., Allsop, N.W.H., Bruce, T., De Rouck, J., Kortenhaus, A., Pullen, T., Schüttrumpf, H., Troch, P. and Zanuttigh, B.,

B. Hoonhout/ A. Capel
Rev. 0 22 February 2018