Marram grass establishes through seed germination or shooting of rhizome fragments detached from tussocks by coastal erosion. Konlechner and Hilton (2009) showed the potential for marine dispersal of such rhizome fragments over hundreds of kilometers, depending on regional sea currents. Seeds are mainly dispersed by wind, although the species only shows week morphological adaptations to wind dispersal (Huiskes, 1979). Dispersal experiments by Pope (2006) and McLachlan (2014) suggest that a large majority of the seeds end up within a distance of less than 1 m from the parent plant and wind dispersal abilities are probably restricted to several tens of meters. The potential for seed establishment of C. arenaria is very high. First, this is due to a substantial seed production. Salisbury (1952) estimated over 20,000 caryopses are formed yearly per plant tussock. Second, marram grass has a long-lived seedbank. Viable seeds of up to 21 years old were recovered (Hilton et al., 2019). Third, the germination potential is high. Experiments under optimal laboratory conditions yielded germination percentages between 82 and 94% (Huiskes, 1979; van der Putten, 1990; Bendimered et al., 2007; Lim, 2011). In the field, however, the establishment success of C. arenaria from seed is reputed to be very low on average (Huiskes, 1977), although locally frequent germination was observed in coastal dunes in Netherlands (van der Putten, 1990), New Zealand (Esler, 1974), and North America (Wiedemann, 1987). Frequent establishment of marram is observed in embryonic foredunes and damp dune slacks (Authors’ personal observations). Seed germination strongly decreases with sand burial (van der Putten, 1990; Lim, 2011; McLachlan, 2014). Seedling emergence decreases linearly with burial depth, with a 3 cm burial already resulting in a germination reduction of 60% and no more seedlings emerge when seeds are buried under 9 cm of sand (Figure 2). The results we obtained from a burial experiment (see Supplementary Material 1) are very similar to the findings of Lim (2011).

Huiskes (1977) and van der Putten (1990) showed that the highest germination rates are obtained with a fluctuating (day/night) temperature regime and a day temperature exceeding about 20°C. These results are enhanced by stratification (cold pre-treatment). Optimal germination was obtained with a 20/30°C night/day temperature, with germination inhibited at lower temperatures of 10/20°C night/day. These germination requirements retrieved in the lab correspond well with observations of seeds germination in spring, when the temperature has risen sufficiently. Germination occurs only under moist conditions (Huiskes, 1979) and is inhibited when salinity exceeds 9 g/L (Chergui et al., 2013).

Once established, marram grass growth and survival depend largely on the exposure of its local environment to the physical forces of wind and water, that can directly dislodge plants or indirectly affect growth and survival by transporting sediment. Partial burial of seedlings resulted in a 50–60% increase of shoots length and root dry mass, but this vertical growth increases at the expense of lateral growth and overall shoot biomass (number of tillers, which was maximal at 0–40% burial of the shoot height) (Maun, 1998; Ievinsh and Andersone-Ozola, 2020). C. arenaria biomass increase showed a parabolic response to burial with optimal growing performance at burial rates of 31 cm of sand per growing season (Nolet et al., 2018). The tolerance for burying was estimated to 78–96 cm burial/year. Reijers et al. (2021) found more mature tussocks (clonal fragments containing ±8 shoots) to perform equally well under 0 or 2 cm burial every 2 weeks, but high mortality when burial reached 4 cm. Sediment burial also indirectly influence marram grass growth by protecting the plants against the detrimental effects of coastal flooding. Higher and larger embryo dunes are less susceptible to erosion during the winter storm season, which positively influences marram grass growth during summer (van Puijenbroek et al., 2017b).

Besides exposure to physical forces, soil nutrient levels can have a large influence on marram grass performance as well. In general, sandy coastal systems are nutrient-limited and C. arenaria can cope with these nutrient-poor conditions through symbiosis with arbuscular mycorrhizal fungi and by recycling its own plant material through slow decomposition (Kowalchuk et al., 2002; Reijers et al., 2020).

Despite its occurrence in nutrient-limited conditions, C. arenaria requires substantial levels of nitrogen, phosphorus, and potassium for good growth (Willis, 1965). A higher availability of N and P in lime- and iron-poor dunes, due to atmospheric deposition, has been proposed as a mechanism of the species’ local expansion in coastal dunes (Kooijman et al., 1998; Kooijman and Besse, 2002). Increases in temperature, nutrients, and precipitation stimulate vegetation growth and lead to a global greening of coastal dunes (Jackson et al., 2019). This global greening affects the natural sediment-sharing capacity of coastal dunes, by hampering sediment transport to the hinterland (Gao et al., 2020). Reduced sediment mobility and dune stabilization are thought to threaten several ecological functions, while it can increase the protective function of coastal dunes by lowering erosion susceptibility (Delgado-Fernandez et al., 2019; Gao et al., 2020; Pye and Blott, 2020).

Marram grass was found to perform worse in its own rhizosphere soil than in either sand from the sea floor or in sterilized soil from its own rhizosphere (van der Putten et al., 1988, 1993), demonstrating that a biotic factor in the soil causes a decline in marram grass performance. The exact cause of this biotic control is to date unclear. The first studies attempting to pinpoint the soil organisms causing the decline of marram grass implicated root-feeding nematodes as well as pathogenic fungi (van der Putten et al., 1990; De Rooij van der Goes, 1995; van der Putten and van der Stoel, 1998; van der Stoel et al., 2002; vandegehuchte et al., 2010b; Brinkman et al., 2015). However, the exact species causing a performance reduction could not be identified across these studies. Competitive and facilitating interactions among these co-infecting belowground parasites (Brinkman et al., 2005a, b, c) but also more complex trophic interactions, including those with microbes within the rhizosphere (Piśkiewicz et al., 2008; Piskiewicz et al., 2009; Costa et al., 2012) were found to be mediators of marram performance under experimental conditions. Furthermore, it has been shown that the negative effect of certain nematode species can be mitigated by the positive effect of mycorrhizal and endophytic fungi (Little and Maun, 1996; de la Peña et al., 2006; Hol et al., 2007). Overall, the net effect on marram grass performance of all naturally occurring members of the soil community is generally negative. Although the exact mechanism is difficult to identify, evidence for the “escape hypothesis” remains strong, i.e., marram grass needs regular burial by wind-blown sand free of soil organisms so that it can grow new roots into an – at least temporarily – enemy-free space. Plant–soil feedbacks caused by other plant species also play a role. Conditioning of soils by Carpobrotus edulis (L.) N.E.Br., a species originating from South Africa and one of the most invasive plant species in the Mediterranean, suppresses marram grass biomass and in some cases survival rate (de la Peña et al., 2010). The increase of Carpobrotus in the dunes of central Italy (Sperandii et al., 2018) has therefore been linked to large-scale decreases in marram grass. Marram grass also shows a reduced germination on soil invaded by Acacia longifolia (Andrews) Willd., yet it performed better on invaded than native soil after 12 weeks of growth (Morais et al., 2019).

The aboveground organisms associated with marram grass are in general well known (i.e., Huiskes, 1979; Heie, 1982, 1986; Holman, 2009; vandegehuchte et al., 2010a), but very little is known about their effects on plant performance. The associated herbivore species have the potential to induce serious reductions in aboveground performance in a controlled environment (Balachowsky and Mesnil, 1935; Nye, 1958; Heie, 1986; vandegehuchte et al., 2010a), but so far no experiments were conducted in nature. Marram grass does not seem to be controlled to any significant extent by mammalian grazers either (Bhadresa, 1977; Huiskes, 1979), except for some feeding on young shoots (Rowan, 1913). Seed predation has been observed (Huiskes, 1979) but its magnitude and/or impact on marram grass demography is unknown.

Intraspecific variation among marram grass populations can have strong effects on the abundance and community composition of both above- and belowground invertebrate species (vandegehuchte et al., 2011). This variation is linked to genetic variation in plant growth, which likely explains higher abundances of aboveground invertebrates on local than on non-local marram grass populations. Contrasting effects were found for root herbivores as their abundance and species richness negatively covaried with the aboveground ones (vandegehuchte et al., 2011, 2012). Additionally, it has to be noted that a full soil biota community can have stronger effects on marram grass performance than local abiotic soil properties (vandegehuchte et al., 2010c), although performance can differ significantly among soils differing substantially in abiotic properties. The relationships between marram grass and its aboveground invertebrates can therefore not be understood independently of its belowground invertebrates and the abiotic conditions of the soil.

Explanations for success of marram grass in its novel range have been sought in the popular “enemy release hypothesis” (Keane and Crawley, 2002), mainly focusing on belowground enemies. Growth of marram grass was significantly less reduced on soils from South African sites than on soils from the Netherlands, indicating a weakened negative plant–soil feedback and thus potential role for enemy release in South African soils (Knevel et al., 2004). However, this contrasts with findings from coastal dunes of California, where soil sterilization experiments have shown that the performance of marram grass is reduced to similar extents as in Europe when grown on non-sterilized soil (Beckstead and Parker, 2003), suggesting there is no enemy release. Furthermore, soil biota from three native South African plant species did not suppress marram grass growth, but biota from soils beneath the tropical cosmopolitan dropseed Sporobolus virginicus (L.) Kunth did, suggesting that this plant species may confer biotic resistance against invasion by marram grass (Knevel et al., 2004). A large sampling campaign of soil and roots from Tasmania, New Zealand, South Africa, and the west coast of the United States revealed that marram grass did not have fewer root-feeding nematode taxa in these regions than in its native range. However, native plants in the novel range had more specialist root-feeding nematode taxa than marram grass, while specialists such as cyst and root-knot nematodes, which are common in the native range of marram grass, were not found in the southern hemisphere (van der Putten et al., 2005). Invasiveness of marram grass thus seems correlated with an escape from specialized root-feeding nematodes.

From section “Dynamic Feedbacks Between Aeolian Fluxes and Vegetation Development,” it is clear that feedbacks between the environment and the spatial distribution of marram grass impact dune development. We mapped marram cover and spatial contingency in 20 m × 20 m grid cells along the coastlines of northern France, Belgium, Netherlands, and South-England (see Supplementary Material 2), to identify realistic ranges in nature. Marram grass is – as predicted from the species’ biology – predominantly showing a clustered distribution with JC (join-count; an established method that quantitatively determines the degree of clustering or dispersion, see Supplementary Material 2) values between 20 and 80, so ranging from random (values close to zero) to highly clustered patterns (Figure 3). A mean clustering pattern with JC values around 50 is stable over the four studies regions. No underdispersed (so regular) patterns were observed. Interestingly, marram grass spatial cover at these spatial scales is strongly country-specific with United Kingdom and France being represented by well-vegetated dunes. Dunes in Belgium and the Netherlands appear to be in more dynamics states with quite a substantial presence of areas with a low cover (see section “Discussion and Outlook”).

We developed a grid-based dune simulation model that computes aeolian transport processes and changes in vegetation growth and dune morphology based on their dynamic feedbacks and marram spatial organization. The landscape is grid-based with cells having dimensions of 0.20 m × 0.20 m. The 100 × 100 cell matrix therefore corresponds with a dune area of 20 × 20 m². We refer to Supplementary Material 2 for a detailed process overview, references to the code. We simulated changes in aeolian processes and wind dynamics at a day-resolution. We used averages over 7 years (2010–2017) received from the Royal Meteorological Institute at Koksijde, and scaled them to the four main different wind directions as used in the model by Nolet et al. (2018), each corresponding to a side of the landscape. Wind speed and direction is drawn daily from a normal distribution, based on monthly average wind speed and its standard deviation. Sand input, the material blown into the system from the beach (N-direction here), is expressed as a relative percentage of the maximum sand saturation flux, i.e., the maximal amount of sand that can be carried by the wind. Lateral winds (E and W-directions in the grid) have an influx which corresponds with the most recent outflux of a lateral wind (thus, we represent the landscape as tube to avoid edge artifacts). This amount is constantly updated during a simulation.

Sand deposition is directly dependent on shear velocity which is a function of wind velocity (Durán et al., 2010; Hoonhout and de Vries, 2016), vegetation density (Durán et al., 2010) and its roughness (Durán and Moore, 2013). Increases in shear stress due to funnel effects are included, as are gravity and shelter effects. Maximum angles of repose are set to 34° (Durán et al., 2010) when vegetation is absent (Durán et al., 2010). These angles increase with vegetation density. As such, avalanches are less prevalent when plant density is high. Moreover, erosion is inhibited in locations sheltered by lee slopes at an angle of maximal 14° (Kroy et al., 2002). Marram grass dynamics are seasonal (only growth in spring and summer), with local growth modeled as outlined in this review. Vertical as well as lateral growth during the growing season is modeled as a logistic function up to maximal heights, and directly depending on sand deposition (Nolet et al., 2018), leading to positive growth under intermediate sand accretion and complete burial leading to marram grass die-off. Lateral growth follows Lévy-patterns as determined by Reijers et al. (2019), and are here modeled by simplified neighbor expansion processes. No growth occurs during autumn and winter but sand accumulation continuous. The net height after winter burial determines the starting conditions for vertical growth in the next season. No germination events were modeled as these are to date not (or only rarely) witnessed in foredunes the last decade.

To validate the model predictions, we compared outcomes from the model with those from a statistical model linking changes in topography over 5 years as derived from LIDAR in relation to the initial marram spatial configuration as determined from aerial photographs in 2015 in Belgium (Figure 4; detailed methods in Supplementary Material 3, 4). Our model simulations (Figure 5, upper panels) were run for initial marram cover in the range 0.1–0.9 (as without marram cover, only erosion of the initialized sand volume is occurring without establishment and join counts between 20 and 70).

The modeled height changes agree in general terms with those observed. The observed larger effects in the field suggest slightly larger sand input, either due to sand availability or changes in wind strengths, from the beach as sand input initiated in the model based on rough estimates from the Belgian coast (Rauwoens and Strypsteen, unpublished data). Alternatively, the small scale of the mechanistic model may underestimate wind saturation and therefore sand displacements (see section “Discussion and Outlook”). Dune height increases at intermediate cover of marram grass, so P ∼0.5. However, the simulation model predicts increases to be maximal under low cover and more random (i.e., less clustered) distributions of the marram grass tussocks (low P and low JC). According to the simulation model, local changes in dune topography, estimated as the coefficient of variation (CV) of grid cell-level differences in height, show most changes occurring in dunes with clustered marram patches or patches with a low amount of vegetation, but more random patterns. Predicted changes from LIDAR follow the same pattern as the ones generated by the simulation model. Only under low cover and intermediate clumping, a more homogenous increase of the dune is predicted by the computer model.

Analysis of the LIDAR data also showed decreasing dynamics with increasing distance from the sea (see Supplementary Material 4). The obtained effect sizes (Supplementary Material 4) and accompanying visualizations of the modeled effects (Figure 5) indicate that the observed changes in integrated dune height and form differ from those of the simulation in the sign and strength of the cover × spatial clustering (P:JC and the interaction between P² and JC). The most prominent difference lies in the predicted erosion dynamics under low marram cover and a strong clustering. This suggests that the sand accretion capacities of marram grass under these conditions needs to be re-evaluated.