Coastal wetland resilience to climate change: modelling ecosystem response to rising sea level and salinity in a variable climate

We investigate coastal wetland ecosystem resilience to sea level rise by modelling sea level rise trajectories and the impact on vegetation communities for a coastal wetland in South Africa. The rate of sediment accretion was modelled relative to IPCC sea level rise estimates for multiple RCP scenarios. For each scenario, inundation by neap and spring tide and the 2, 4, and 8 year recurrence interval water level was modelled over a period of 200 years. When tidal variation is considered, the rate of sediment accretion exceeds rising sea levels associated with climate change, resulting in no major changes in terms of inundation. When sea level rise scenarios were modelled in conjunction with recurrence interval water levels, flooding of the coastal wetland was much greater than current levels at 1 in 4 and 1 in 8 year events. In the long term, increases in salinity may cause a reduction in Phragmites australis cover. Very small increases in depth and frequency of inundation are likely to cause an expansion of samphire species at the expense of Juncus spp. The study suggests that for this wetland, variability in flow may be a key factor in balancing wetland resilience.


Introduction
The future of the world's coastal wetlands will be influenced by anthropogenic contributions to climate change and associated sea level rise. These contributions are driven by accelerating anthropogenic concentrations of greenhouse gas emissions that are largely a result of increasing economic activity and population growth (IPCC 2013). Despite global and regional efforts to reduce greenhouse gas emissions through policies, environmental management plans, and advances in technology, anthropogenic influences on the global climate continue (IPCC 1990(IPCC , 1996(IPCC , 2001(IPCC , 2007(IPCC , 2013. The IPCC 5th Assessment Report suggests that global mean sea level will rise into and beyond the 21st century. In addition, forecasts of extreme sea levels, which take into account changes to sea level, tides, wind waves, and storm surges, are projected to increase by between 34 and 76 cm globally by 2100 under a moderate emissions scenario (Vousdoukas et al. 2018), further increasing the vulnerability of the world's coastal wetlands (Church et al. 2013). Coastal wetlands are highly productive ecosystems, providing indispensable services and resources to society and the environment (Mitsch and Gosselink 2015). It is estimated that by the 2080s up to 22% of the world's coastal wetlands could be lost as a result of sea level rise (Nicholls et al. 1999). Neumann et al. (2000) documents a number of biophysical losses and degradation of coastal wetlands associated with sea level rise; namely inundation, erosion, total or partial loss and (or) displacement of wetland marsh and ecosystems, and saltwater intrusion of coastal aquifers and rivers. These cumulative impacts will strongly affect the future of our coasts, their natural ecosystems, and associated human livelihoods.
Our ability to predict changes to coastal wetlands is complicated by the variety of nonlinear interactions occurring between biota and the physical environment. In some environments, certain biological interactions are known to offer resistance and (or) resilience to perturbations or gradual change (Martinez et al. 2017). Holling (1973) originally defined resilience as the amount of disturbance that an ecosystem could withstand without selforganised processes and structures changing. Integral to ecosystem resilience is the understanding that an ecosystem may exist at multiple stable states. As such, ecological resilience may be considered as the tolerance of a system to perturbation before transition to an alternative stable state (Gunderson 2000). An example here would be the transition of a salt marsh to a mud flat due to continuous sea level rise. Martinez et al. (2017) suggests that some coastal ecosystems are not merely resilient; they may be resistant in that the system is able to hold a force without any modification.
For coastal wetlands, increasing levels of inundation, salinity, and occasionally disturbance are the primary drivers of potential change related to sea level rise. To maintain process and structure, sediment must accrete vertically at a rate faster or equivalent to that of rising sea levels (Day et al. 1999;Morris et al. 2002;Mitsch and Gosselink 2015). Sediment accretion rates in coastal wetlands are influenced by catchment sediment yields, which may vary due to anthropogenic activities, the construction of dams within catchments , and (or) as a result of natural rainfall variability (Aalto et al. 2003). Sediment accretion may also be restricted by topography and human construction along the coast (e.g., coastal infrastructure or barrier protection against flooding; Titus 1991;Nicholls 2004;Mitsch and Gosselink 2015), or there may simply be insufficient space for the ecosystem to migrate inland as sea level rises (i.e., "coastal squeeze"; Doody 2013). In other systems, despite significant rates of coastal accretion, subsidence may constrain wetland resilience (Jankowski et al. 2017).
Applying models to forecast changes in physical environments is essential to aid sustainable and comprehensive planning. Reed (2002) suggests that by measuring and quantifying important environmental processes, such as sediment accretion and the rate of sea level rise, it is possible to understand and predict the response of coastal wetlands to ongoing sea level rise. However, this is a complex task as processes within physical environments are influenced by many biophysical interactions (Rogers et al. 2012;Mogensen and Rogers 2018). For example, as inundation levels change, vegetation communities change, affecting flow attenuation and susceptibility to flooding (Rodríguez et al. 2017). Lentz et al. (2016) suggest that dynamic models that take into account vegetation feedbacks are essential for guiding coastal management decisions. However, where empirical evidence is lacking, changes to ecosystems may be understood by studying existing environmental gradients to determine which factors may have an impact on ecosystem dynamics (Grenfell et al. 2016). In contrast, Ge et al. (2015Ge et al. ( , 2016 adopt an experimental approach to understanding vegetation response by physically simulating the response of individual plant species to different flood regimes. Studies have shown that some coastal wetlands may be resilient to rising sea levels (see Patrick and DeLaune 1990;Morris et al. 2002;Reed 2002;Temmerman et al. 2004). In these environments, increasing or adequate sediment yields have a positive influence on wetland accretion rates as elevation is sustained above that of rising sea levels. However, in environments where sediments yields are low, coastal wetlands are unable to maintain elevation and are therefore unable to keep pace with rising sea levels, resulting in wetland loss (see Day et al. 1999;Van Goor et al. 2003).
In this study, we attempt to model and forecast changes associated with sea level rise to a coastal wetland on the Agulhas Plain, South Africa. In addition, we examine relationships between the physical environment and vegetation communities to gauge whether certain characteristics can provide a degree of resilience or resistance to change.

Study area
The Droё River wetland is located alongside the Heuningnes estuary and falls within the Agulhas Plain, a coastal plain nested in the Overberg region of South Africa (Fig. 1). This coastal plain is a remnant of an old wave-cut platform, stretching across 270 000 ha, bounded by the Cape Fold Mountain ranges in the north and the Atlantic and Indian Oceans in the south (Kraaij et al. 2008). The Agulhas Plain is characterised by strong westerly and easterly winds, prevailing all year round, and a Mediterranean climate with hot dry summers and cold wet winters. Annual rainfall ranges between 400 and 600 mm, with most rainfall occurring between the months of May and October (Bertsky 2013).
The Agulhas Plain is a unique and diverse ecologically sensitive region, serving as a storehouse for a variety of endemic, rare, and endangered species (i.e., terrestrial and wetland vegetation, bird, freshwater, estuarine, and marine species). The conservation of fynbos and wetland ecosystems are arguably the most significant components of the biota (Kraaij et al. 2008). As such, this region is renowned for its conservation value, housing the Cape Floristic Region, which is an internationally recognized Biodiversity Hotspot; the Heuningnes Estuary and De Hoop Vlei, which are Ramsar Wetlands of International Importance; and Soetendalsvlei, which is the second largest lacustrine wetland in South Africa.
Nestled in these coastal lowlands, towards the south-eastern region of the plain, are a variety of interlinked and isolated wetland ecosystems. These are mainly associated with the Nuwejaars River, of which the Kars River is a tributary. The combined Kars and Nuwejaars River is called the Heuningnes River and flows out to sea at an estuary that is known locally as De Mond.
The main study area is along the course of the Droё River wetland, which is an abandoned channel course of the Kars River that once flowed into the Heuningnes Estuary just 500 m from the sea. It is not known when the channel was abandoned but the old meandering course is well-preserved in the landscape.
The Agulhas region is used extensively for the cultivation of wheat and canola, as well as livestock farming. Agriculture is one of the major anthropogenic pressures threatening the sustainability of ecosystems within this region.

Determination of wetland accretion rates
Two methods were used to estimate sediment accretion rates within the wetland. To gauge sedimentation rates over a single year, 38 astroturf mats were installed for the duration of the wet winter season of 2016 (see Fig. 1). These allowed direct measurement of sedimentation rates and provided insight into potential spatial variability of deposition.
The mats were installed in three transects and were placed on individual squares of heavy duty plastic (0.13 m 2 in size). The mat and plastic sheet were secured using stainless steel nails. Twenty-seven mats and sheets were recovered the following February during the dry season. The remaining 11 mats could not be located.
The sediment that had accumulated on the astroturf mats was rinsed into pre-weighed beakers. The total sediment mass for each astroturf mat was determined once the sample had been oven dried overnight. Loss on ignition was used to calculate the organic content and involved burning ∼3 g of dry sediment in a muffle furnace for 4 h at 450°C.
The long-term sediment accretion rate was calculated from a single core taken during the dry season in February 2017. The core was taken with a stainless steel gouge corer at an undisturbed site within the wetland. It was then sub-sampled into 2 cm increments Fig. 1. The location of the study site near the southernmost tip of Africa is shown. The large map indicates the study site relative to the Heuningnes Estuary. The wetland is located on an abandoned channel of the Kars River, which now flows into the Heuningnes River toward the west. The inset map indicates the position of sample plots, astroturf mats, the sediment core, as well as the area of detailed DGPS survey. and transported to the University of Exeter (UK) for preparation and dating using Pb 210 (Appleby 2001).
Pb 210 activities were measured using Ortec alpha spectrometers following an aqua regia extraction of the samples on a hot plate and plating of the daughter of 210 Po on a silver planchet. The "constant rate of supply" dating model (Appleby and Oldfield 1978) was applied as event based dating was not necessary or possible given the available data.

Topographic survey
To model the current wetland surface, a Differential Global Positioning System (DGPS) survey was conducted using a Stonex base station and rover. This system allows collection of xyz coordinates to sub-centimetre accuracy corrected in real-time. The base station was setup at Lammerskop Trigonometric beacon 174 (identified on the 3420CA & CC Bredasdorp 1:50 000 topographic map; NGI 2007). To maximise efficiency while maintaining accuracy, fewer data points were surveyed across broad flat areas while more data points were surveyed in areas of more rapid changes in elevation.
A LiDAR (light detecting and ranging) dataset was obtained from the Department of Environmental Affairs and Development Planning. Ground hit points representing bare earth were extracted from the LiDAR point cloud. The DGPS and LiDAR data sets were merged to create a raster surface elevation model in ArcMap 10.3 using the Topo to Raster interpolation tool (1 m resolution).

Sea level rise forecasts
Forecasts from the Intergovernmental Panel on Climate Change (IPCC) global mean sea level rise projections were obtained from the 5th Assessment Report (Church et al. 2013). Sea level rise forecasts ranging between 0.28 and 0.98 m by 2100 were based on four representative concentration pathways (RCP) projections, RCP 2.6, 4.5, 6.0, and 8.5 (Fig. 2). These projections are based on the Process-Based Model outputs from 21 Coupled Model Intercomparision Project Phase 5 (CMIP5) Atmosphere-Ocean General Circulation Models (AOGCMs) (Fig. 2). The RCP scenarios are based on predicted anthropogenic greenhouse gas emissions and mitigations with RCP 2.6 being a low emission scenario, RCP 4.5 and 6.0 are considered stabilization and late mitigation scenarios, and RCP 8.5 is an extremely high greenhouse gas emissions scenario. Median sea level rise values were used as model inputs, percentiles were not run as the range of scenarios was considered to provide a good range of potential outcomes.

Vegetation and surface sediment sampling
Surface sediment and vegetation sampling was conducted during the late dry season of May 2017. A subjective assessment of the wetland suggested the occurrence of 3-4 plant communities, sampling was therefore stratified to represent these. Fifteen vegetation plots were sampled. Species within a 2 m 2 plot were identified and a coverage scale given using the Braun-Blanquet scale. Undisturbed surface sediment samples were taken at each sample site using an Eijelkamp bulk density sampler.
Bulk density samples were oven dried and weighed to calculate bulk density. Thereafter, organic content was measured using the loss on ignition method (as described in Sect. 3.1). Particle size was determined using the settling method following sample pre-treatment. Sediment conductivity was measured using an YSI conductivity meter after samples had been prepared using the water extract method (Dahnke and Whitney 1998).

Two-way indicator species analysis (TWINSPAN)
WinTWINS 2.3 (Hill and Šmilauer 2005), a windows-based version of the original TWINSPAN software, was used to develop a community-based classification of the wetland vegetation. In TWINSPAN, sample plots are classified first followed by a species classification, producing a succinct two-way table that may be interpreted to identify vegetation communities. Classification is conducted using a series of ordinations (correspondence analyses), that systematically divide the samples into two groups to create a dichotomous hierarchy. Species abundance is accounted for by "pseudospecies". Division continues until a user-defined criterion is reached. TWINSPAN may be used to identify environmental variables that have an impact on vegetation distribution.

Development of the sediment accretion and sea level rise model
To investigate coastal wetland response to rising sea levels, a simple GIS-based empirical model was developed in ArcMap 10.3. Two main processes were modelled over time. Firstly, change in elevation as a result of ongoing sediment accretion within the wetland, and secondly, the effect of rising sea level on frequency of inundation. The model starting point is based on the topographic survey conducted as part of this study. Predicted rising sea level conditions such as mean high water spring (MHWS) and mean high water neap (MHWN) tidal conditions were modelled for respective time steps (25 years for this study).

Model data inputs
The DGPS survey was considered more accurate as the LiDAR did not use ground reference points. However, to visualise more extensive changes that could not be physically surveyed, the LiDAR was appended to the DGPS survey.
MHWN and MHWS tide estimates were obtained from the Hydrographic office of the South African Navy. To incorporate these with the LiDAR and DGPS survey, the tidal data was corrected to the Land Levelling Datum (LLD) of South Africa, Hartebeesthoek94.
Daily tidal records for the Heuningnes G5T002 water level gauge at De Mond, referenced to the LLD, were provided by the Department of Water and Sanitation, South Africa, for 2001-2018 (see Fig. 1). As the gauge is located within the river inlet, it provides an indication of the combined impact of river floods and tidal or storm surges. Recurrence Intervals were calculated by ranking and calculating recurrence interval using RI = n + 1/m, where n is the number of events and m is the event rank. The 2, 4, and 8 year recurrence interval water levels were estimated for respective sea level rise RCP scenarios. Current research suggests that coastal water levels during extreme events are rising with an increase in wave height and sea level (e.g., Hemer et al. 2013;Camus et al. 2017;Mentaschi et al. 2017;Vousdoukas et al. 2018). Statistical tests of the tidal gauge data indicated no significant change in extreme sea levels over the period of record. As such, increasing sea levels associated with extreme events were not added to the model due to difficulties in selecting an appropriate level of increase.

1.
The spatial extent of sediment accretion is determined: The area of active wetland sediment accretion was mapped visually by considering topography and vegetation. As the accretion rate had been measured in a very particular geomorphic setting, only similar regions were included. The actively meandering stretch of the Heuningnes River was not modelled in terms of sediment accretion.

2.
The impact of sediment accretion on wetland elevation is calculated: The model assumes that sediment accretion rates are constant over time and space. This is reasonable considering the geomorphic setting selected. The area of active sediment accretion defined in the first step is used to predict new wetland elevation for the respective year. This output is added onto the wetland surface elevation as determined in the previous iteration.

3.
Inundation due to forecast sea level rise is determined: The predicted elevation determined in steps 1 and 2 is compared to respective predicted RCP sea level rise scenarios. The output is based on static inundation, and assumes inundation is instantaneous. Dynamic models are considered more accurate than static models, but are more computationally expensive, and are thus often applied to large areas at a low resolution (e.g., >100 m cells; Breilh et al. 2013;Ramirez et al. 2016;Vousdoukas et al. 2016). In this case, a static model was selected to reduce computational requirements considering the high resolution of the elevation data set (1 m), which was necessary to capture rapid changes in elevation over a small distance. This was considered acceptable as Breilh et al. (2013) showed that static model predictions were improved when wetlands were within 5 km of the landward boundary, in this case the wetland is <5 km from the tidal gauge. For each time step, areas inundated by neap and spring tides and at the 2, 4, and 8 year recurrence intervals were extracted.

Sediment accretion rates
Sediment accretion rates were estimated using two methods (Fig. 3). 210 Pb activity was measured on a 140 cm core taken within the wetland, the resultant long term average rate of accretion was 3.99 mm·a −1 . In contrast, the mean rate of accretion measured using the amount of sediment collected on the recovered astroturf mats (corrected for bulk density and organic content) after a single season was 0.75 mm·a −1 . There was considerable variation within measured rates on the astroturf mats, with a minimum value of 0.006 and a maximum value of 3.864 mm·a −1 . Bioturbation and post-depositional disturbance were considered to be a major factor as the wetland was used by a large flock of Lesser flamingoes (Phoeniconaias minor) and was also occasionally grazed by sheep and horses. The effect of relative elevation and landscape setting on accretion rate was considered but no relationship was found.
A visual inspection of the rate of 210 Pb sediment accretion curve suggests that the rate is slowing, a view that conforms to the geomorphic evolution of the system. The sediment accretion rate was likely higher immediately following abandonment, but would have slowed as less flood waters moved down the drainage line. The date at which the channel and floodplain were abandoned is not known, but predates the first aerial photography of the area (ca. 1939). It is likely that the 3.99 mm·a −1 as measured from the 210 Pb core indicates an upper maximum sediment accretion rate and that the real rate is slightly lower. The accretion rate of the wetland is not coupled to sea level, which has been gradually rising for the last century. Instead, radiometric dating suggests that variation in accretion rates is linked to fluvially derived sediment rather than marine sources.

Sediment accretion and sea level rise model outputs
The sediment accretion rate gauged from the 210 Pb core was used to forecast changes associated with rising sea level for respective RCP scenarios and extreme flood events over a period of 200 years (Figs. 4-7). Results for RCP scenarios 4.5 and 6.0 were similar, therefore only results for RCP scenarios 2.6, 6.0, and 8.5 are shown.
Inundation associated with the MHWN and MHWS tides increases across all RCP scenarios, but not in the areas where sediment accretion has been modelled at 3.99 mm·a −1 (Fig. 4). These model simulations suggest that the sediment accretion rate is locally faster than the rate of sea level rise under all RCP scenarios (Fig. 4). However, when the combined effect of river inputs and tides are considered using recurrence interval data from the De Mond water level gauge, the impact of rising sea level on inundation levels is markedly different (Figs. 5-7). Despite sediment accretion rates being relatively high, and allowing the maintenance of the current hydrology during normal tidal conditions, model simulations indicate that the frequency and depth of inundation will increase when water levels associated with river inputs and tides are considered.
Under current conditions, an event that occurs every 2 years typically results in inundation close to the estuary, with only small downstream portions of the modelled drainage line being flooded. Flooding increases incrementally in 125 years for the 2.6 and 6.0 RCP scenarios. In contrast, in RCP 8.5, a 1 in 2 year event would result in large-scale flooding of the entire drainage line in 100 years. Results for the 1 in 4 year event are similar, but flooding is slightly more extensive for RCP 2.6 and 6.0 scenarios in 125 years than for a 1 in 2 year event.     In the RCP 8.5 scenario, flooding during the 1 in 4 year event is extensive, with the entire modelled wetland flooded within 75 years.
Currently, an event that occurs every 8 years results in flooding that is generally confined to the Heuningnes floodplain and lower reaches of the modelled Droë River wetland. In RCP scenario 2.6, depressions all the way up the modelled wetland are linked by flood water. In RCP scenario 6.0, the entire system is flooded to an average depth of 2.6 m. In scenario 8.5, the depth of inundation is further increased to ∼3 m.

Vegetation communities
An interpretation of the TWINSPAN output identified three plant communities (Table 1). The first community comprised salt-tolerant species of samphire and grass, as well as some small creeping herbs. Common to every plot was either Sarcorcornia perennis or Sarcocornia littorea. This community occurs in the depressions of the wetland and is accompanied by a fair amount of bare ground. The second community was dominated by Phragmites australis reeds with more than 75% coverage in all plots. Some other species were also encountered, such as the grass Sporobolus virginicus and the rush Juncus spp. The third community was the Juncus spp. rush-dominated community with samphire (S. littorea only). The rush community occurred on the fringes of the salt-tolerant samphire community at a slightly higher elevation.
An analysis of the physical environment of each community provides insight into each vegetation type's tolerance to different variables (Fig. 8). The samphires, grass, and creeping herb community occurred in depressions, with the Juncus spp. rush with samphire occurring in a fringing ring as elevation increased. As relative elevation is an indicator of both depth and frequency of inundation, it appears that the samphires are able to tolerate this variability better than the rush community. The P. australis reedbed occurred at a variety of elevations, but was always located on the northern edge of the wetland where a line of outcropping rocks was visible. Reeds are usually associated with permanent water, and it seems likely that the northern edge of the wetland experiences some groundwater seepage. As this water is not supplied from surface runoff or accumulation in a depression, the reedbeds are not associated with any specific elevation.
Soil conductivity can be used as a proxy for soil salinity. The samphires, grass, and creeping herbs community occurred in soils that were variable in salinity. Cycles of flooding followed by dessication result in the accumulation of salts in the soil and in some locations soil salinity exceeded that of sea water. Variability in salinity was also high for the Juncus spp. rush with samphire community. In contrast, soil salinity values approached zero in all P. australis -dominated reedbed plots. This strengthens the suggestion that these reedbeds are groundwater fed as P. australis is known to be tolerant to saline conditions (Chambers et al. 2003).
The organic content of soils was lower in the salt-tolerant samphires, grass, and herbs community, with slightly higher but similar values found in the rush and reed communities. In terms of sediment type, there was no difference in terms of sand or clay content.

Coastal wetland resilience: sediment accretion versus sea level rise
Perennial rivers and tides dominate sediment supply to coastal wetlands. In the Droё River wetland, neither of these sources is constant as the wetland is located within an abandoned channel course that occasionally receives floodwater from a downstream river course. Despite the absence of a river within the valley, the long-term rate of sediment accretion is relatively high at 3.99 mm·a −1 . There are multiple potential sediment sources, such as wind-blown sand from coastal dunes, marine sediment delivered by extremely high tides, and lastly, terrestrial sediment from the Heuningnes River during extreme floods.
A wetland's ability to maintain elevation and survive under rising sea levels depends on the rate of sediment deposition (Mitsch and Gosselink 2015), prompting some authors to assert that accretion rates are the driving force behind coastal wetland resilience to sea level rise (Schuerch et al. 2012;Belliard et al. 2016). In the Droё River wetland, the long-term sediment accretion rate is faster than the projected rate of sea level rise in all RCP scenarios (Fig. 4). This suggests that the Droё River wetland has the capacity to maintain elevation for the next 200 years, providing ecosystem resilience to mean sea level rise. However, consideration of water level data that combines the impact of variation in river flows with tides provides a deeper understanding. While accretion rates may be sufficient to offer resilience to changes in mean sea level, MHWN and MHWS, the wetland is vulnerable to slight increases in water level associated with a combination of tides and river flow (i.e., river floods, see Figs. 5-7). In these scenarios, increases in inundation of the Droё River wetland occurred at all recurrence intervals modelled. Furthermore, the model outputs suggest that flooding of the Heuningnes River provides connectivity between the wetland and the estuary. The wetland is rarely flooded by increases in mean sea level alone, but requires heightened water levels in the Heuningnes River.
Some authors have indicated that in some coastal wetlands, there is a feedback between inundation and accretion rate (e.g., Friedrichs and Perry 2001;Temmerman et al. 2004). During flood events, sediment transported by the river and (or) tides is deposited on the wetland. As a result, an increase in sea level can result in an increase in accretion rate provided sediment concentrations are sufficiently high. Depending on the balance of sediment supply and elevation, it is possible that an initial increase in frequency of inundation can result in an increase in accretion rate that results in maintenance of the wetland elevation, and therefore wetland resilience. In a scenario where sediment accretion remains lower than the rate of rising sea level, elevation is not maintained and the wetland becomes vulnerable to flooding.
The Droё River wetland's current exposure to seasonal floods may be a predisposing factor in facilitating its resilience to rising sea levels. The deposition of suspended sediments associated with seasonal floods provides a surplus concentration of sediment to the wetland, enabling sediment accretion after a flooding event (see Friedrichs and Perry 2001).
Modelling results indicate that the depth of inundation and frequency of such flood events may increase. Climate change studies forecast an increase in extreme sea level and a marginal increase in local wave height, indicating that these model results may well be conservative estimates (Vousdoukas et al. 2018). However, this increase may favour an increase in the sediment accretion rate, as it is likely to provide additional sediment to the wetland.

Coastal wetland resilience in balance: the impact of flow variability
The maintenance of elevation is dependent on the rate of sea level rise relative to sediment accretion within the wetland; this balance is what underlies wetland resilience (Fig. 9a). However, the relationship is dynamic rather than simplistic in that change in the frequency of inundation, as occurs when sea levels rise, also causes a change to sediment supply and therefore accretion rate (e.g., Temmerman et al. 2004). Some wetlands may therefore be able to dynamically respond to an increase in sea level by increasing the rate of sediment accretion.
The modelling outputs suggest that in the Droë River wetland, the current sediment accretion rate exceeds potential sea level rise in all scenarios, resulting in minor changes to frequency of inundation when only the effects of tides were considered. However, this understanding changes significantly when water levels affected by river flows and tides are modelled. In coastal wetlands characterised by variable climates, wet and dry phases may have a major impact on wetland resilience. A wet phase is a series of years where rainfall is above the mean, resulting in the wetland being flooded for an extended period of time. In contrast, a dry phase is when rainfall is below the mean. In extreme dry phases, the wetland may not be flooded at all during the year.
The effect of flow variability on wetland resilience is conceptualised in Fig. 9. Wetland resilience is a balance between sediment accretion and sea level rise. If sea level rise accelerates, provided sediment accretion accelerates at a similar pace, the wetland maintains elevation and processes remain unchanged. The wetland is resilient (Fig. 9a).
However, resilience (or balance) may also be maintained through feedbacks or cycles of climatic phases. During a wet climatic phase, seasonal river flows are higher and longer, increasing the frequency and depth of wetland inundation. In combination with a slow increase in sea level, flooding can be extensive. This results in an increase in sediment supply to the wetland. In Fig. 9b, sediment supply is conceptualised as a ball that moves and can adjust the balance between sea level rise and sediment accretion rate. As sediment supply increases, sediment accretion rates are enhanced. This increases the elevation of the wetland and reduces the impact of sea level rise. The system is momentarily out of balance as the sediment accretion rate is higher than the rate of sea level rise. However, as the elevation of the wetland is increased, it is subjected to fewer flood events, which reduces sediment supply and accretion on the wetland. The system therefore rebalances through a feedback between sediment supply and inundation frequency.
During a dry phase, the wetland experiences fewer periods of inundation as the river is less frequently in flood. This reduces the rate of sediment accretion on the wetland. In theory, a reduction in sediment accretion rate should lead to an increase in frequency of inundation if the rate of sea level rise remains constant (Fig. 9c). However, modelling results indicate that for the Droë River wetland, this effect is minimal (Fig. 4), with inundation of the modelled wetland negligible for spring flood tides as sea level rises. In this wetland, inundation only occurs when the river floods, regardless of sea level rise. During the dry phase, the counter feedback between accretion rate and frequency of inundation does not operate; the sediment accretion rate is merely reduced. When the wet phase begins, flooding would initially be more severe, reducing over time as the accretion rate increased. Longterm wetland resilience is therefore dependent on the occurrence of wet phases of seasonal flooding that bring sediment to the wetland.

Wetland vegetation response to rising sea level
Soil conductivity data suggests that the samphire, grass, and herb and the Juncus spp. rush community with samphire are both accustomed to a wide range of salinity conditions. These two vegetation communities would therefore tolerate increased levels of salinity associated with rising sea level. In contrast, P. australis reedbeds were characterised by low salinity levels. The location of these reedbeds is likely a response to permanence of water supply and their occurrence probably indicates areas of groundwater discharge. It has been shown that P. australis can tolerate extremely saline conditions although the range is limited to salinity levels lower than 45‰ (Chambers et al. 2003), which is higher than the average salinity of sea water. Due to evaporation, salts tend to accumulate in these coastal soils. Thus, evidence from this study and the literature suggests that all of the current vegetation communities offer some resilience to future increases in salinity as a result of sea level rise.
The coverage of P. australis is currently limited by the availability of permanent water. None of the RCP scenarios suggest that flooding will become permanent, and we can therefore expect that the small patches of P. australis will not expand in the lower portion of the Droë River wetland.
While the degree of water permanency is the likely control on the occurrence and expansion of the P. australis -dominated reedbed, separation between the samphire, grass, and herb community and the Juncus spp. rush community with samphire is related to the depth and frequency of inundation (expressed as relative elevation in Fig. 8). The differences between the two are very small, the average relative elevation for the samphire, grass, and herb community is 0.032 m, compared to 0.053 m for the Juncus spp. rush community with samphire. This small variation is expected as S. littorea occurs in both communities. However, the degree of overlap in terms of elevation is marginal, with the maximum relative elevation for the samphire, grass, and herb community at 0.048 m, while the minimum relative elevation recorded for the Juncus spp. rush community with samphire was 0.042 m. The two communities therefore occur at very distinct ranges in inundation depth, suggesting that a small change in frequency of inundation or inundation depth could result in a dramatic change in vegetation community coverage. The data suggest that an increase in mean water depth of just 0.006 m could cause a switch from the Juncus spp. rush community with samphire to the samphire, grass, and herb community. While this is not positive in terms of ecosystem biodiversity, there is currently space for upstream migration of vegetation communities, and there is thus potential for both vegetation communities to be conserved.

Conclusion
In this study, the long-term accretion rate exceeded sea level rise in all RCP scenarios, resulting in very little difference in terms of flooding at spring or neap tides. However, the inclusion of water levels associated with river floods indicated the potential for massive changes in inundation extent and frequency with sea level rise. Furthermore, results suggest that it may be overly simplistic to consider wetland resilience as a relative comparison of the rate of sea level rise and sediment accretion. Feedback between inundation, sediment supply, and therefore sediment accretion rate require further analysis and morphological modelling to further unpack likely changes. Nevertheless, the study provided insight into wetland vegetation communities and the relative importance of salinity and depth and frequency of inundation. In this case, one vegetation community may not be viable under conditions of increased salinity (P. australis dominated reedbed), while another (Juncus spp. rush community with samphire) may be forced to migrate upstream due to an expansion of the samphire, grass, and herb community, which is better adapted to wetter conditions.