In October 2022 a prototype dike was constructed on the Falsterbo peninsula in the south of Sweden (lat: 55°24′46.8″N, lon: 12°52′2.4″E) to facilitate in situ testing of nature-enhanced vegetation erosion resistance. The Falsterbo peninsula is a low-lying coastal area with a population of approximately 20,000, with more than 4,600 houses in areas identified as flood-risk areas (Ahlström and Åberg, 2018). In response, the local municipality is planning to implement a coastal protection system comprising dikes, dunes, and seawalls to protect urban areas from coastal flooding. In Sweden, constructing such protective measures requires Swedish Land and Environmental Court approval. A condition of the permit for this coastal protection scheme mandated that the dike cover vegetation consists of native plant species to minimize the ecological impact on the adjacent Natura 2000 protected areas. Due to the limited understanding of the erosion resistance of nature-enhanced vegetation on dikes, a full-scale prototype dike was constructed to assess the technical properties of nature-enhanced vegetation as a protective cover on dikes. The prototype dike is approximately 50 m long, 2 m high above the ground, the slope is 18° (1:3), and the crest width is 3 m. It was constructed using a core of local material (clay till), a 30 cm thick clay layer, and covered with 35 cm of topsoil. Two different topsoils and four vegetation strategies were used in the construction, resulting in a total of eight different vegetation areas (see Table 1 and Figure 1). For areas denoted by 1 (Figure 1), the local topsoil underneath the prototype dike was reused, and for areas denoted by 2, a nutrient-poor soil was used. The testing of two different top-soils was motivated by an interest to compare the plant composition when using the local topsoil, which is preferred to utilize, with a nutrient poor soil that potentially should result in higher plant diversity (Roem and Berendse, 2000). Four different vegetation strategies were applied: (A) Native seeds from a hay cutting in a nearby pasture were spread on the dike slope to encourage germination. (B) Herb-rich seed mixture ¹ was sown. (C) Heather (Calluna vulgaris) was transplanted from a nearby heath. (D) Standardized grass seeds ² were planted to represent standard grass revetment on the dike. Unfortunately, the heather (C) did not establish well and was therefore excluded from the study. Maintenance of the vegetation cover on the dike included one cutting in September 2023, but no irrigation.

The dike body consisted of glacial till. A 30 cm thick clay layer was placed on top of the dike body. As described in Table 1, local top-soil is underneath the vegetation on the east side, nutrient-poor top soil underneath the vegetation on the west side of the dike (see Figure 2). The uppermost layer on the east side consists of the uppermost part of the vegetation that was stripped underneath the prototype dike (0–10 cm). This part was stripped first and put in one separate pile. Thereafter, the entire topsoil of 10–40 cm was stripped and placed as the second layer of the topsoil.

Figure 3 shows the grain size distribution for each vegetation area for local (Figure 3A) and nutrient-poor soil (Figure 3B). All soil samples from local soil show a similar profile, while the soil samples for the three test areas for the nutrient-poor soil differ especially in the distribution of higher grain sizes. This becomes clear when comparing the D₉₀ values, which indicate that 90% of the particles in the soil are smaller than this grain size. For local soil D₉₀ ranges between 2.7 and 4.6 mm. D₉₀ for test section A2 amounts to 8.4 mm; for B2, the value is 2.9 mm and 0.9 mm for D2, respectively. Larger D₉₀ values indicate a higher coarse fraction, while smaller D₉₀ values suggest a higher fine fraction. The fine soil fraction (c_f) in the soils, defined as the percentage of particles below 0.063 mm, ranges from 3.5% to 4.5% local soil and 5.7% to 8.4% for nutrient-poor soil. The local soil can be classified as sandy soil, while the nutrient-poor soil contains a higher proportion of fine particles and thus, is classified as humus-rich sand.

Vegetation in the six analysed areas was measured in terms of coverage and length at the time of the wave impact tests in March 2024. In addition, root analysis was performed in the first week of April 2024.

The vegetation coverage was quantified using MATLAB’s ColorThresholder tool, applied to photographs taken prior to the experiments. This method employed a color-based binary conversion process to distinguish vegetation from soil. Specifically, the RGB color ranges used for thresholding were: Red (110–250), Green (100–250), and Blue (50–250). This calibration effectively isolated the vegetation in the images. Figure 4 illustrates an example of this coverage estimation technique. The thresholding process converted vegetation to white pixels and soil to black pixels, allowing for a straightforward calculation of vegetation coverage as the percentage of white pixels in the total image area. To complement the coverage analysis, vegetation height was determined by measuring multiple individual stems and calculating their mean length. This dual approach of assessing both coverage and height provided a comprehensive characterization of the vegetation structure prior to testing.

The root density (RD) was sampled on 2024-04-05. RD is defined as the ratio of the dry root weight divided by the soil sample volume. Soil samples were collected at one point per plot in close proximity to one of the wave impact tests. At each point the above-ground vegetation was clipped at the surface and a cylinder with a diameter of 10 cm was used to extract samples for the depths 0–5 cm, 5–10 cm, 10–15 cm, and 15–30 cm from the surface. Soil was excavated around the cylinder and roots were cut using a steel wire to extract the soil sample in the cylinder. To separate the soil from the roots, the samples were soaked in water for 24 h. The dry root weight was obtained by drying the roots for 24 h at 105°C.

A species-based vegetation analysis was performed on 2023-08-16. Therefore, a 1 × 1 m frame was placed in the middle of each dike section. The percentage of each forb species was estimated in addition to the overall cover of the following parameters and functional groups: graminoids, forbs, bryophytes, bare soil, and litter. Graminoids are herbal plants with a grass-like morphology. Forbs are herbal species that are not graminoids and include flowering plants. Bryophytes are non-vascular plants, mainly mosses (Park and Allaby, 2017). The total cover can exceed 100% since layers may overlap. Species richness and diversity (Shannon-Weiner diversity index, H′) was calculated based on the forb species data.

The wave impacts were simulated using a wave impact simulator, constructed at the Institute of Hydraulic Engineering and Water Resources Management (IWW) at RWTH Aachen University according to the model used in Pachnio (2004), Stanczak (2008) and Scheres and Schüttrumpf (2020). It has been shown that the impact of a wave can be simulated with a falling water column (Pachnio, 2004) which is the principle of the simulator.

The simulator consists of a cylinder that is mounted to a stand. The cylinder can be filled with water and is closed at the bottom through a valve. A pneumatic system opens the valve rapidly, allowing the water to fall in one splash. The intensity of the impact can be adjusted by filling the cylinder to different water levels h_w and adjusting the cylinder height above the ground h_f. A schematic drawing is provided in Figure 5A. A picture of the simulator during testing on the prototype dike can be seen in Figure 5B.

The advantage of the simulator used in this study is that it can be easily moved and mounted. The wave impact generator used by van Steeg et al. (2014) is larger and thus, can create larger waves. However, it has to be moved and attached to a tractor (van Steeg et al., 2014). While the wave impact simulator used in the present study can adapt to different dike slope angles, its opening remains fixed in a vertical position. In contrast, the wave impact generator used by van Steeg et al. (2014) can adjust its opening to various angles, allowing it to align perpendicularly to the dike slope. The process of wave breaking is complex, involving various forces and stresses. However, the experiments conducted in this study concentrated exclusively on the gravitational aspect of wave breaking on a coastal structure, neglecting other processes such as wave run-up, run-down, and shear stresses. The wave impact simulator used in this study affects a smaller area compared to the real-life impact of a wave front, which would influence the entire length of the dike more extensively. Consequently, the analysis should emphasize the maximum erosion depth rather than the overall spatial extent of the erosion holes.

Five test series were conducted for each vegetation type. The specifications of the tests are given in Table 2. Each test consisted of ten splashes. The impact pressures p_max were calculated (Stanczak, 2008) for one individual splash. The individual p_max for each test was cumulated (p_cum) for the total pressure exerted on the test area. From the particular pressure of one splash per test configuration, significant wave heights were calculated based on the assumption of p_max = p_max,50% (Führböter and Sparboom, 1988; Schiereck, 2003).

The erosion depth was measured using a pin profiler developed and used by Scheres and Schüttrumpf (2020). It consists of a perforated plate (see Figure 6A) with a 0.01-m grid in the center area and a 0.02-m grid in the outer area. The distance between the plate and the ground is measured with a pin in each hole (see Figure 6B). The pin profiler allows measurements to be taken without interference from vegetation roots or surface residues covering the erosion hole. An area of 18 × 18 cm could be measured with this method.

The pin profiler was set up before testing so that the center area of the perforated plate aligns with the extended opening of the cylinder. Before the first test, the initial surface profile at the test location was measured. The measuring position on the ground was kept the same for every following measurement.

This study aimed to describe the evolution of erosion caused by wave impact. As previously mentioned, the wave impact simulator used in this study could focus only on a smaller area and, therefore, simplified the real-world process. Thus, the discussion of erosion patterns is centered on the maximum erosion depth, while the erosion’s overall extent is not considered in greater detail. It is important to note that larger wave fronts would influence a more extensive area along the length of the dike.

Figure 9 shows the erosion patterns after a cumulative impact pressure of 1,126 kPa for each vegetation and soil section. While the holes are visible for most sections, the erosion for section A1 is barely visible. This is due to vegetation and roots covering the hole. Test of B1 stands out in terms of a large erosion volume due to an unexpected burrow tunnel underneath the vegetation (Figure 10).

The development in erosion depths after each of the five test cases is shown in Figure 11 for the different test sections. A steady increase in erosion depth after each wave impact was found (Figure 11). Furthermore, for some tests, there was a decrease in erosion depth on the upstream end. That means, that the soil was piled up and pushed by the water of the impact simulator. However, the diagram also shows that with the measuring position fixed from the beginning, the erosion patterns could not be detected entirely for every test section. The water column was deflected in some cases on its way to the ground. Thus, the maximum erosion depth was close to the edge for some test sections (e.g., B1). Spikes in the profile indicate the presence of roots (e.g., A2). In general, D1 and D2 exhibit relatively similar profiles, while the other test sections show distinct erosion patterns for the two soil types.

Figure 12 shows the maximum erosion depth as a function of the cumulative impact pressure for the different test sections. In general, both repetitions (cross and circles) show similar results, except for B1 where a burrow tunnel interfered the tests (marked with cross). However, for test case B1 the first impact still showed similar erosion depths. Only after the hole was reached the erosion depth increases significantly. A slightly lower maximum erosion depth could be observed for local soil. The greatest difference between local and nutrient-poor soil can be seen for vegetation type A.

Figure 13 shows the comparison of the maximum erosion depth for the cumulative impact pressure (left) and the maximum erosion depth per impact pressure (right) as a mean of the two repetitions. Excluded from the figure was the second repetition in section B1, as there was an animal burrow near the test area, influencing the extent of the erosion depth. While A1 has the lowest root density in the first 10 cm of the soil within the sections for local soil, the erosion depth is the lowest. A1 also shows the lowest erosion depths per impact pressure. Still, test sections with nutrient-poor soil showed lower root densities compared to sections with local soil and also showed higher erosion depth over the impact pressures. Furthermore, the erosion depth after the first test case showed a moderate correlation with the root density in the top layer of the soil. However, the significant difference in root density between the soil types was not reflected in the erosion depth. While it is evident that nutrient-poor soil exhibited greater erosion depths with lower root density, the difference in erosion depth is not as pronounced. This suggests that while the influence of roots cannot be disregarded, the soil and vegetation type likely played a more decisive role. The latter is supported by the fact that the highest grass coverage of 80% was found in test areas A1 and B1, which showed the lowest erosion depth.

Scheres and Schüttrumpf (2020) observed erosion depths of up to 0.15 m (tests were stopped after reaching an erosion depth of 0.15 m). Present study’s section B can be compared to test vegetations (TV) 5 and 6 (both species-rich herb-dominated mixtures), and D can be compared to TV 1 (grass-dominated mixture).

Figure 14A shows the comparison of section B of the present data and TV5 and TV6 by Scheres and Schüttrumpf (2020). One trial of TV5 and TV6 was eroded more quickly, while the other repetition was more resistant than the present data. The differences between the repetitions in their investigations could be due to the fact that, particularly for TV6, the vegetation density was around 45%, resulting in bare spots within the test area. Consequently, a bare spot may have been affected during one repetition, while in another, a vegetated spot was hit, potentially leading to differences in the repetitions. In the present study, the vegetation cover was relatively uniform, which is why the observed differences were minimal. However, this highlights the importance of vegetation coverage when addressing the question of erodibility. The comparison between section D of the present data and TV1 by Scheres and Schüttrumpf (2020) is presented in Figure 14B. The test of TV1 aligns with the tests of the present study, which shows the robustness of the method and its applicability under laboratory conditions as well as for in situ tests.

Figure 15A shows the erosion volume after each test case for the different test sections. It can be observed that–similar to the erosion depth–a higher soil volume was eroded at the end of the tests on nutrient-poor soil. Figure 15B indicates that the erosion volume per impact pressure (L/kPa) during the initial splash is greater for local soil compared to nutrient-poor soil. Contrastingly, in tests 2 and 3, the L/kPa decreases for local soil, while it increases for nutrient-poor soil. Both soil types exhibit an increase in erosion volume in tests 4 and 5. Notably, the initial observation deviates from expectations based on root density. Despite higher root density in the upper layer of local soil, erosion volume was generally greater across most tests. However, nutrient-poor soil demonstrated a higher erosion volume in the final case than local soil. If vegetation height is considered, A1 had the highest height in soil section 1, which potentially helped to attenuate the first impact pressure. However, this connection was not true for nutrient-poor soil, indicating the interaction of roots, vegetation height, coverage and soil for erosion due to impact pressure. It should also be mentioned that the results of the erosion volumes can not directly be transferred to actual wave impacts as the extent of erosion due to actual wave fronts would impact a greater area along the dike length. However, it shows the differences between the soil types and emphasizes the interaction between soil and vegetation parameter.

In general, the root density seems not to be the only influencing factor when it comes to erosion depth. Thus, the correlation between soil types could be reasonable.

Furthermore, within one soil type section, no significant differences between the vegetation areas regarding the erosion resistance were observed, allowing the conclusion that biodiverse vegetation cover (vegetation type B) is equally resistant as grass cover (vegetation type D). Similar results were obtained by de Kroon et al. (2024) for the hydraulic load of overtopping. In detail, they found that erosion resistance levels for higher biodiverse vegetation were close to those of traditional grass cover.

Figure 16 shows the relationship between two dimensionless numbers. On the y-axis, maximum erosion depth, vegetation coverage, and vegetation height, on the x-axis, root density, soil parameters, and impact pressure are plotted. The fitted power function indicates that with higher RD, D₅₀ value and p_cum and lower c_f, the erosion depth d_e decreases. This suggests that soil with greater grain sizes is eroded slower than soil with a higher proportion fine fraction. This aligns with the differences noted between the previous study by Scheres and Schüttrumpf (2020) and the present one, as the soil used in the former study had a higher fine particle content. In contrast, experiments on wave run-up and overtopping revealed that the dike body eroded more quickly with larger grain sizes, likely due to the cohesion between finer particles (Evangelista, 2015). This indicates the different forces acting during run-up and impact pressures. Since wave breaking occurs along with wave run-up and run-down, the erosion around the impact pressure point may vary. Furthermore, in this study, the soils were very similar; a more systematic investigation could give further insights. In addition, higher root density results in smaller erosion depths, as indicated in Figure 16.

During the testing period, there was little to no rainfall, with an average of 0.04 mm/h. However, storm surges are usually accompanied by rainfall and infiltration, which increases the water content in the soil. In this study, we did not measure water content, but it is important to note that it still affects the soil’s erodibility. Higher soil water content typically reduces soil cohesion (Pezowicz and Choma-Moryl, 2015; Ravindran and Gratchev, 2022), making particles more susceptible to detachment and transportation by flowing water. Consequently, the likelihood of erosion due to wave impacts could be increased.