Effect of vegetation cover and sediment type on 3D subsurface structure and shear strength in saltmarshes

The vulnerability of saltmarshes to lateral erosion at their margin depends on the local biogeomorphological properties of the substrate. In particular, the 3D architecture of pore and root systems is expected to influence shear strength, with repercussions for the wider‐scale stability of saltmarshes. We apply X‐ray computed microtomography (μCT) to visualize and quantify subsurface structures in two UK saltmarshes at Tillingham Farm, Essex (silt/clay rich substrate) and Warton Sands (sand‐rich substrate), with four types of ground cover: bare ground, Spartina spp, Salicornia spp and Puccinellia spp. We extracted μCT structural parameters that characterize pore and root morphologies at each station, and compared them with field measurements of shear strength using a principal component analysis and correlation tests. The 3D volumes show that species‐dependent variations in root structures, plant colonization events and bioturbation activity control the morphology of macropores, while sediment cohesivity determines the structural stability and persistence of these pore structures over time, even after the vegetation has died. Areas of high porosity and high mean pore thickness were correlated to lower values of shear strength, especially at Tillingham Farm, where well‐connected vertical systems of macropores were associated with current or previous colonization by Spartina spp. However, while well‐connected systems of macropores may lower the local deformation threshold of the sediment, they also encourage drainage, promote vegetation growth and reduce the marsh vulnerability to hydrodynamic forces. The highest values of shear strength at both sites were found under Puccinellia spp, and were associated with a high density of mesh‐like root structures that bind the sediment and resist deformation. Future studies of marsh stability should ideally consider time series of vegetation cover, especially in silt/clay‐dominated saltmarshes, in order to consider the potential effect of preserved buried networks of macropores on water circulation, marsh functioning and cliff‐face erosion.

, they have been shown to be vulnerable to lateral erosion at the margin (Bendoni et al., 2016). It has been argued that the destruction and rejuvenation of saltmarshes is a natural process occurring over an order of a few hundred or thousand years, dominated by sedimentological processes (Chauhan, 2009;Fagherazzi, 2013;Van de Koppel et al., 2005). However, in the context of anthropological pressure on coastal environments, sediment starvation and increased wave impact and storm frequency accompanying sea-level rise, trends of net saltmarsh loss have been observed around the world (Allen, 2000;Gedan et al., 2009;Gu et al., 2018;Schwimmer, 2001). While wind-waves play a primary role on longterm marsh edge erosion at the landscape scale, local marsh characteristics such as vegetation cover are also important (Finotello et al., 2020). Therefore, rates of saltmarsh erosion from wave action are variable from marsh to marsh (Ford et al., 2016;Wang et al., 2017), and even over small spatial scales within the same marsh (Bernik et al., 2018;Priestas et al., 2015;Van de Koppel et al., 2005;Wang et al., 2017). Since local vulnerabilities in the marsh structure can have broader implications for the whole marsh and lead to widespread erosion (Bendoni et al., 2016;, better understanding of what causes these local changes in stability or susceptibility to erosion is needed to more accurately project potential future losses and efficiently mitigate against these in the context of a changing climate. The intrinsic capacity of saltmarsh substrates to resist hydrodynamic erosive forces at the local scale is often measured as localized shear strength. While wave flume experiments can help us understand the specific effect of wave thrust on the erosion of 'transposed' marsh cliffs, in-situ measurements have the advantage of preventing disturbance during sampling, transport and storage (Grabowski, 2014).
At the local scale, this resistance to deformation depends on bulk sediment properties such as the grain size, cohesivity and water retention properties of the sediment (Crooks & Pye, 2000;Grabowski et al., 2011), but also on biogeomorphological factors such as the presence and morphology of vegetation, roots and pores (Brooks et al., 2020;Wang et al., 2017). While influences on surface shear strength, such as the presence of vegetation and biofilm, have been extensively studied (Feagin et al., 2009;Gedan et al., 2011), the impacts of subsurface structures and processes on shear strength remain challenging to observe and quantify (Brooks et al., 2020).
Structural pores or macropores, caused by cracks, burrows and decaying roots (as opposed to micropores or matrix pores which are formed by the space between sediment particles; Rabot et al., 2018), can create areas of structural vulnerability in the soil (Vu et al., 2017).
They can also promote vertical water movement in the subsurface environment, improve drainage (Tempest et al., 2015) and facilitate root growth . Roots are another important architectural component of the marsh substrate. The tensile strength provided by the roots complements the sediment, which is naturally strong in compression (Gyssels et al., 2005), and thus helps to prevent block failure (Brooks et al., 2020;Wang et al., 2017). However, the roots' exact role in substrate stability depends on species-and environmentspecific structural characteristics (Gyssels et al., 2005). A particularly understudied aspect is how the 3D architectures of roots and pore networks interact within different types of substrates, and influence the internal shear strength of a saltmarsh (Brooks et al., 2020). X-ray computed microtomography (μCT) combines the penetrating capacity of X-rays with 3D volume reconstruction to observe the internal 3D structure of objects in a non-destructive manner (Cnudde & Boone, 2013). μCT has been applied extensively to agricultural soils to investigate the impact of subsurface structures on crucial soil functions such as water infiltration (Jarvis et al., 2017;Katuwal et al., 2015;Müller et al., 2018;Pot et al., 2020;Tracy et al., 2015), root-pore interactions and patterns of plant growth (Hu et al., 2020;Lucas et al., 2019;Pulido-Moncada et al., 2020). In recent years, the technique has been extended to saltmarsh substrates (Dale et al., 2019;Spencer et al., 2017;Van Putte et al., 2019); however, distinguishing roots from pores is challenging because their greyscale values overlap due to the partial volume effect (Cnudde & Boone, 2013;Helliwell et al., 2013), especially in these complex heterogeneous substrates. Indeed, saltmarshes are transitional habitats formed by a constant interplay of sediment deposition and erosional processes, and where ground cover and other soil characteristics can vary rapidly both in space and time. Episodes of storm surges, colonization by burrowing organisms, or colonization and die-off of plants may be recorded as sedimentary features, and therefore subsurface features may be critical to interpreting surface information and marsh response. Recent studies have developed new approaches for root analysis (Chirol et al., 2021), which allow us to capture the complexity of heterogeneous saltmarsh substrates with unprecedented precision.
This study applies μCT to the 3D structural analysis of roots and pores in two UK saltmarshes of contrasting sediment type and under four contrasting ground covers (bare ground, Spartina spp, Salicornia spp, Puccinellia spp). We provide a detailed analysis, both visual and quantitative, of saltmarsh below-ground structures, and discuss the interplay of root and pore systems under different ground covers and sediment types. We select parameters that best capture the structural variability of roots and pores, and explore correlations between substrate morphology and internal shear strength using a principal component analysis. Taking into consideration other geochemical factors of erodibility such as the proportion of clay-sized particles, the organic matter content and the dispersibility of the clay, we then discuss the wider implications in terms of how sediment properties and morphology contribute to marsh stability at different spatial scales, and provide recommendations for further study.

| METHODS
We analysed below-ground structure and shear strength at two minerogenic saltmarshes in the UK. In order to compare the effect of vegetation and substrate on the subsurface structure, we considered two sediment types (sand-rich at Warton Sands and silt/clay-rich at Tillingham) and four types of ground cover (bare ground, Spartina spp, Salicornia spp, Puccinellia spp), for a total of eight stations (Figure 1).
The ground cover choices reflect the zonation of vegetation in the saltmarshes, with bare ground mudflats fronting the marsh, then pioneer species Spartina spp and Salicornia spp, and lower marsh species Puccinellia spp (Figure 2). This may inform us on how a saltmarsh inherits structural features as it accretes vertically from fronting mudflat to an inner marsh. The plant species were also chosen for their contrasting root structures: Spartina spp have long stems with internal voids as an adaptation strategy to anoxic conditions (Mitsch & Gosselink, 1986); Salicornia spp have a shallow and sparse root system and Puccinellia spp have a dense system of thin roots (Chapman, 1960).
Three replicate sediment cores (15 cm depth and 15 cm diameter) were collected at each station in January 2019. The replicates were spaced within 0.5-2 m from one another to ensure that all replicates are independent but have similar substrates. The sediment cores were collected to minimize disturbance of structural features, as summarized in Carr et al. (2020). After extraction, the cores were stored upright in a cooling box filled with bubble wrap to minimize disturbance during transport, and stored at 4 C until required.
The whole, intact cores were scanned using a Nikon Metrology XT H 225 μCT system at 205 kV and 46 μA (9.4 W). The exposure time was 500 ms at 36 dB gain. A Cu 1 mm copper filter was used to reduce beam hardening artefacts. 4486 projections were acquired with four frames per projection, for a scan time of 4.5 h. The effective voxel size is 61.79 μm, downscaled to 62.5 μm during volume reconstruction. The scanned volumes were cropped to an 8.75 Â 8.75 cm square base to reduce edge effects and remove any disturbance from sampling. All 24 scanned volumes were processed following the method detailed in Chirol et al. (2021) to segment the μCT data into three phases: pore space, organic matter elements (including roots and degraded organic matter) and finally the bulk inorganic mineral phase. All elements larger than 5000 voxels (1.22 mm 3 ) were removed as noise, and the minimal thickness of elements at any point is twice the resolution, so 125 μm. This method was developed to distinguish live and decayed (necromass) roots from pores in heterogeneous saltmarsh soils, which makes it highly relevant here.
Each phase was visualized in 3D using the volume-rendering software Drishti (Limaye, 2012), and a detailed morphological analysis was performed using the automated software plugin 'Particle Analysis' for ImageJ (Schindelin et al., 2012) to extract a number of shape parameters (Table 1). Out of these parameters we selected those that best represent the structural differences between vegetation and sediment types, basing ourselves on previous studies (Rabot et al., 2018;Spencer et al., 2017) and on our own observation of the dataset. Each selected variable was then normalized within the interval [0, 1] to visualize variations across stations for all variables, and plotted as spider plots.
While the main focus of the paper is to establish relationships between substrate structure and shear strength, other parameters commonly associated with soil stability or vulnerability to erosion were also considered. These include the proportion of clay-sized particles, which influences the cohesivity of the sediment, the proportion of organic matter as determined by loss on ignition and the sodium adsorption ratio of the sediment. The latter considers how high content of exchangeable Na + in the soil can lead to the formation of thick water films around the clay particles and to slow rates of sediment consolidation, thus making the marsh more prone to erosion (Crooks & Pye, 2000). Three replicate cores per station were taken for the analysis of the sodium adsorption ratio at two depths (0-1 and 7-8 cm from the surface): samples were freeze-dried, sieved at 2 mm, then mixed with a recorded mass of distilled water until the obtention of a saturation paste as outlined by Rowell (1994), and left overnight for the cations to equilibrate. The samples were then centrifuged to retrieve the extracts, and the exchanged cations were measured in the extract using inductively coupled plasma optical emission spectrometry (ICP-OES). The sodium adsorption ratio (SAR) was then calculated as SAR = [Na + ]/([Ca 2+ ] + [Mg 2+ ]) 0.5 , with [ ] the concentration in mmol À1 . Of the three replicate cores, one core per station was also processed for particle size analysis by laser granulometry and organic matter content every 1-2 cm from the surface to 15 cm deep (see Table 2 for details). Measurements of organic matter content, including both particulate organic carbon and roots, were obtained by loss on ignition following the method of Rowell (1994): soil samples were first air-dried, heated overnight at 105 C, then weighed and combusted at 500 C overnight.
Finally, shear vane data were collected in August and September 2019 from a distributed survey across a large area of the Tillingham Farm and Warton Sands saltmarshes. The shear vane measures pressure applied at failure point at a depth of 7.5 cm from the surface by rotating a handle against the vane head, and quantifies the undrained geotechnical shear strength of the sediment, that is to say its resistance to deformation and fracture at a very local scale (Grabowski, 2014). While there is a spatial and temporal mismatch between the shear strength measurements and the position of our sediment cores, the survey was designed to capture the characteristic shear strength for each station, with measurements taken at a frequency of 150 per sediment and vegetation type. A summary of the sampling procedure for each data type is provided in Table 2.
Due to the small number of measurements for all considered parameters except the shear strength, the normality hypothesis cannot be assumed to distinguish between groups using analysis of variance (ANOVA) tests. We used the non-parametric tests of Kolmogorov-Smirnov for normality and Bartlett-Levene for homoscedasticity. When the conditions of normal distribution and homogeneous variances were not met, we relied on the non-parametric Kruskal-Wallis test, which is less sensitive to outliers.
Finally, we performed a principal component analysis (PCA) to compare the μCT morphological characteristics to the shear strength in order to estimate which structural parameters are the main drivers of variability between locations. Since we want to focus on the role of soil structure on shear strength, sedimentological and geochemical properties were not included in the PCA; instead, their specific contributions to shear strength were studied using linear regression.
PCA transforms the variables in a dataset into a set of principal components in order to reduce the dimensionality while retaining as much of the variation as possible (Jolliffe, 2002). PCA assumes that all considered variables follow a normal distribution, and that the F I G U R E 2 Plant zonation in NW Europe saltmarshes (Redelstein et al., 2018) [Color figure can be viewed at wileyonlinelibrary.com] variables considered fully represent the statistical variation in the dataset; however, even if these conditions are not met, as is the case in our dataset according to the Kolmogorov-Smirnov test, PCA is a robust analytical tool that still provides a useful means to group intercorrelated parameters as a function of their contribution to the overall variability of the dataset (Chirol et al., 2018;Jolliffe, 2002;Reid & Spencer, 2009;Steel, 1996). It is therefore well suited to the analysis of novel parameters such as μCT structural indicators, because their relations with one another and with shear strength are still poorly understood. For this step, we subsampled the shear strength dataset to three or four data points per location while remaining representative of the mean and spread of the sample. We calculated the 10th, 50th and 90th percentiles for all stations with three replicates, and the 10th, 40th, 60th and 90th percentiles for the station TF PUC where four replicates had been selected. All percentiles were sorted randomly to not skew the dataset. All datasets presented in the paper can be found online in the Supporting Information.  Table 3. Three main types of macropores are observed in our T A B L E 1 List of variables considered when interpreting the μCT data, with their definition and corresponding structural parameters when applied to pores and organic matter elements Connectivity (%) Volume of the largest connected particle divided by the total volume of the studied phase (how much of the total phase belongs to a single connected system) Connectivity of the pore system Connectivity of the root system Emptiness (mm) Mean distance between voxels of the same phase Mean distance between pores Mean distance between organic matter elements

Euler-Poincaré characteristic
Topological invariant that describes the shape or structure of a topological space (Vogel, 1997), calculated using the 'Particle Analysis' plugin in ImageJ. A value of 0 means perfectly simple (i.e. one single pore/root); the further the value deviates from 0, the greater the topological complexity of the phase Complexity of the pore system Complexity of the root system Mean thickness (μm) Mean value of the local thickness ('Particle Analysis' plugin in ImageJ), measured at each point in a particle as the diameter of the greatest sphere that fits within the particle and which contains the point samples: (1) highly connected vertical pore systems; (2) sub-horizontal sheets of porosity corresponding to internal cracks (the internal cracks observed during sampling were surrounded by iron precipitates, confirming them as pre-existing structures rather than a product of disturbance during core sampling); and (3)

| Geotechnical, sedimentological and geochemical properties
Since the conditions of normal distribution and homogeneous variances are not met for the geotechnical, sedimentological and geochemical properties considered (see Table A1  T A B L E 6 Comparison of p-values for Kruskal-Wallis tests for the organic matter content, percentage of particles below 63 μm, sodium adsorption ratio and shear strength for each ground cover and sediment type (p < 0.05 means two groups are significantly different, with blue and yellow highlights denoting significant or non-significant differences)  Figure 6D

| Correlations between subsurface morphological properties and erosion resistance
We conduct a PCA to explore the controls on variability in shear strength and in the morphological characteristics of the organic matter elements and macropores under each ground cover and sediment type. The Kaiser-Meyer-Olkin (KMO) test yields a measure of sampling adequacy of 0.63, which corresponds to an acceptable but mediocre degree of common variance class. The low KMO reflects the small sample size available for the PCA (three or four cores per ground cover type, 25 data points in total). Another limitation of PCA is the assumption that the variables selected fully represent the statistical variation of the dataset (Jolliffe, 2002), which is unlikely in a complex saltmarsh substrate. Therefore, the interpretations should be treated with caution, but graphical observation of the principal components (PCs) using biplots offers an indication of the relative importance of each considered variable (Figure 7). In order to increase the interpretability of the PC loadings, we use a varimax rotation to rotate the orthogonal axis so that it aligns with the data points in a way that maximizes the degree of variance in the data (Steel, 1996).  Table 8.

F I G U R E 6
Boxplots showing the distribution of (A) organic matter content (%); (B) percentage of particles below 63 μm (%); (C) sodium adsorption ratio (no unit); (D) shear strength (kPa) for each ground cover and sediment type. Black circles represent the individual measurements for each boxplot (see Table 2  after the roots have fully decayed suggests that the above-ground plant breakage and removal occurred without causing widespread erosion of the bed or infill of the macropores, which signifies that the substrate around the pores has enough internal cohesion to retain its shape even under tidal inundation. Since no such pore system is found under the Warton Sands stations, it is probable that these complex, highly connected and vertical pore systems are less structurally stable in sandier, less-cohesive sediment types. This would explain why the distance between pores dominates the morphological differences between the Warton Sands and Tillingham Farm samples according to the first PC of the PCA. We do find evidence of pore structures being preserved at Warton Sands, but these are thin, horizontal bioturbation horizons, with characteristic I-and U-shaped burrow structures observable in 3D (Figure 3). Burrowing organisms tend to consolidate their burrow structures by coating the walls with secretions (Kristensen & Kostka, 2005;Pagenkemper et al., 2015), which might explain why these horizons have been so well preserved.
Among in accordance with previous observations on fibrous root systems (Vannoppen et al., 2015). The second PC also shows correlations between the complexity of the root system and the complexity and connectivity of the pore system. These observations indicate that vegetation cover (and burrowing organisms) control the type of macropores that form within the substrate, and that the sediment type controls how well these macropores will be preserved.

| Influence of substrate internal structure and geochemistry on shear strength
Higher values of shear strength are found at Warton Sands compared to Tillingham Farm, even though the sandier sediment at Warton Sands is more erosion-prone according to previous remote sensing and flume experiments (Ford et al., 2016;Pringle, 1995 promote drainage, which not only reduces the water's erosive capacity at the surface (Tempest et al., 2015), but also improves sediment aeration, biogeochemical cycling, plant growth and the overall productivity of the saltmarsh (Xin et al., 2009). Therefore, the instantaneous, localized weakening effect of macropores may be compensated by their indirect contribution to marsh stability. Our results highlight complex interactions between substrate structure, potential water flow and erosion vulnerability, which occur at different spatial and temporal scales. The 3D volumes of pore systems obtainable in μCT could provide a framework for water infiltration models in different types of saltmarsh substrates, and help us understand these feedback processes in future studies.
Links between shear strength and root system morphology are harder to decipher in the PCA. According to Brooks et al. (2020), in the upper 15 cm of the marsh, resistance to erosion should be controlled by both the root mat and the sediment properties. Because the binding action of fine root meshes is considered to have an impact on shear strength as measured by the shear vanes (Grabowski, 2014), we hypothesized that either the Euler-Poincaré characteristic or the mean distance to root elements could be used as a descriptor of the 3D mesh-like structure and to quantify this structure's contribution to marsh strength. Here, however, only the number of organic matter elements is grouped with the shear strength in the second PC's loadings. At present, while our method allows us to visualize this mesh-like structure in the 3D volumes, the resolution limit of μCT means that this mesh is too disconnected to be correctly described with quantitative parameters. The Puccinellia spp stations are characterized by both the highest number of root elements ( Figure 5) and the highest shear strength for both Warton Sands and Tillingham Farm (Figure 6), indicating that the mesh-like root structure does have an impact on bed/soil stability. The impact of vegetation type on shear strength appears greater in the sand-rich than in the silt/clay-rich substrate, in accordance with previous studies (De Battisti et al., 2019;Ford et al., 2016). This could be due to a facilitated root penetration in coarser sediments, which exacerbates the structural differences between tap root and mesh root traits: the observed effects of sediment type and root morphology on macropore fraction and shear strength are schematized in Figure 8.
The lack of a significant relationship between shear strength and organic matter content from loss on ignition also suggests that the binding action of roots has more impact on shear strength than their contribution to organic matter content in the substrate, at least within the root zone.
Finally, we found higher SAR values at Tillingham Farm compared to Warton Sands, despite higher SAR normally being associated with a more erosion-prone sediment. This occurs because the difference in soil properties between the two sites affects the relationships between geotechnical and sedimentological properties (see Figure A2 in the

| Future perspectives
μCT has the potential to capture the whole 3D structural complexity of the saltmarsh: future studies could also incorporate shell deposits, or refine the method for smaller scales to resolve fine roots. Whilst we focused on monospecific locations to describe the root structure of common saltmarsh species, the impact of species richness should also be explored: root structure depends not only on the growth strategy specific to each plant species, but can also change as a function of nutrient availability, redox potential and competition with other species (Bardgett et al., 2014;Bouma et al., 2001;De Battisti et al., 2019). Enhanced biodiversity has been found to exacerbate competition strategies between species and lead to greater root biomass and greater sediment cohesivity (Ford et al., 2016).
In order to better correlate these structures to marsh stability in future studies, further geotechnical tests and flume experiments are required to better understand the effect of different ground covers on substrate resistance to deformation and to hydrodynamic forcing, so that we may capture the different processes that contribute to marsh resistance at different scales. Indeed, the effect of local, centimetre-scale pore and root structures on erosion resistance depends not only on shear strength, but also on the position and orientation of these features relative to the dominant wind direction, water depth and tidal regime (Brooks et al., 2020;Schwimmer, 2001).
Consideration of the marsh topography and foreshore morphology will therefore be necessary to fully understand saltmarsh morphodynamics at the landscape scale. To that end, remote sens-

| CONCLUSION
In this study, we compared morphological parameters of macroporosity and root structure from μCT with shear strength data obtained in the field in saltmarshes of contrasting sediment types and in four contrasting ground cover types, in order to explore links between marsh subsurface structure and marsh strength. Our results show that a combination of ecological factors (different root structures create different porosity elements) and sedimentological factors (the soil cohesivity controls its capacity to preserve these pore structures even after the above-ground vegetation has died) play a significant role in determining the macropore structures in saltmarshes.
Large, vertically connected systems of macropores were found at Tillingham Farm under all ground covers except Puccinellia spp: these macropores reduce the internal shear strength, but may facilitate water infiltration and drainage and reduce erosive forces at the surface. The mesh-like root structure characteristic of Puccinellia spp contrasts with the tap root morphology at Salicornia spp and Spartina spp and was found to be the most efficient at increasing the shear strength due to its binding action, at least when looking at monospecific locations. At the scale considered, vegetation type was a better predictor of shear strength than sodium adsorption ratio, which did not change significantly from location to location at Tillingham Farm.
The subsurface structure and strength of saltmarshes results from a complex balance between the marsh history (succession of ground covers and species over time, storm events and other variations in sedimentation rates leading to different sedimentary horizons) and the capacity of the marsh substrate to preserve its internal structure, which depends on the cohesivity of the sediment but also on consolidation by living organisms and plants.
APPENDIX A.
F I G U R E A 1 3D visualization of pores and organic matter elements for all eight stations, showing the differences between replicate cores. Pore features are represented in grey and organic matter elements in green. Left to right: Pores + organic matter elements, pores, organic matter elements [Color figure can be viewed at wileyonlinelibrary.com] F I G U R E A 2 Visualization of the correlations between geotechnical and sedimentological properties at the two saltmarshes. Linear correlation tests were done over the whole dataset (dashed line) and for each study site (solid lines). Linear correlations were tested between (A) organic matter content from loss on ignition and the organic fraction obtained from the μCT data; (B) shear strength and organic matter content from loss on ignition; (C) shear strength and clay fraction; (D) shear strength and sodium adsorption ratio; (E) sodium adsorption ratio and clay fraction [Color figure can be viewed at wileyonlinelibrary.com] T A B L E A 1 p-Values for Kolmogorov-Smirnov test for normality, Bartlett-Levene for homoscedasticity for the clay fraction, organic matter content, sodium adsorption ratio and shear strength (p < 0.05 means two groups are significantly different, with blue and yellow highlights denoting significant or non-significant differences)