Faunal engineering stimulates landscape-scale accretion in southeastern US salt marshes

Regional context

To understand the variation in salt marsh geomorphology and mussel coverage across the South Atlantic Bight (SAB), we assessed density and areal coverage of¹ tidal creekheads and² mussel aggregations with a combination of published data and new field surveys across the region. First, to assess creekhead density, we selected 10 sites ranging from Cape Romain (SC) to Amelia Island (FL). Given that all of our experiments were conducted on Sapelo Island and the surrounding marsh islands, we selected three sites on Sapelo Island for comparison with four sites to the north and three to the south⁶¹. At each site, we scored the total number of tidal creekheads in a 1 km² contiguous marsh area using Google Earth. Assuming each tidal creekhead constitutes approximately 0.0025 km², we calculate the creekhead areal coverage to be:

Differences across northern, Sapelo Island, and southern sites were assessed with a one-way ANOVA with location as the main factor.

To next test the hypothesis that creekhead mussel coverage is similar at sites across the SAB, we conducted surveys of mussel aggregations at 12 sites across the region from Edisto Beach (SC) to Amelia Island (FL). Previous work¹⁴ has shown that mussel aggregations decrease in size and density with increasing distance from the tidal creekhead, so we focused our measurements at three distances from one tidal creekhead onto the marsh platform: 0 m, 20 m, and 40 m. We note that at all sites, mussel aggregations extended >40 m from the tidal creekhead. Sites were again distributed across the region, and included 3 sites to the north, 3 sites to the south, and 6 sites on Sapelo and its back barrier marsh islands. At each site, we selected one representative creek 100–175 m in length and ensured that the tidal creekhead did not overlap spatially with a tidal creekhead of an adjacent creek. At each distance from creekhead, we established one 50 m x 1 m transect. Walking the transect line, we scored each mussel aggregation, counting the total number of mussels and measuring the mound dimensions (L x W x H). We then calculated the areal coverage of mussels within each transect (50 m²) and took the mean value across the three distances as the measure for the site. All data was collected between May and August in 2016 and 2017. Differences across northern, Sapelo Island, and southern sites were assessed with a one-way ANOVA with location as the main factor. Finally, we calculated creekshed mussel areal coverage in the three sub-regions, as the product of the percent of creekshed occupied by creekheads (sub-region mean, %) and the proportion of creekhead area occupied by mussels at each site.

Landscape assays of sediment deposition over seasons and tidal phases

To quantify the relative rates of sediment deposition across marsh landscapes, we deployed 9-cm diameter filter papers (Whatman Quantitative Filter Paper, Grade 42 Circles, Ashless, 90 mm; 57) at 13 location types across 3 sites. Locations included: 1) outer marsh levee (‘outer levee’), 2) marsh platform 10 m inland from outer marsh levee (‘outer levee-adjacent’), 3) inner tidal creek levee (‘inner levee’), 4) marsh platform 10 m inland from inner tidal creek levee (‘inner levee-adjacent’), 5) non-mussel marsh platform (>50 m from mussel creekhead), 6,7) ridge/runnel area at tidal creekhead (‘ridge’ and ‘runnel’), 8,9) mussel aggregations and adjacent non-mussel marsh areas at the tidal creekhead (‘0 m ON mussel mound’ and ‘0 m OFF mound’), 10,11) mussel aggregations and adjacent marsh areas 10 m onto marsh platform from tidal creekhead (‘10 m ON mussel mound’ and ‘10 m OFF mound’), and 12,13) mussel aggregations and adjacent marsh areas 20 m onto marsh platform from tidal creekhead (‘20 m ON mussel mound’ and ‘20 m OFF mound’). At each location type, we used 15 replicate filters, spaced 1–2 m apart. Each pre-weighed and labeled filter paper was deployed attached to a Polystyrene Petri Dish (100 × 15 mm) using 1.5 mm steel wire. After 24 h in the field, all filters were harvested, dried in an oven at 60 °C, and reweighed. Filter papers were deployed at four tides: Summer Spring (August 2017, +2.5 m), Summer Neap (August 2017, +2.1 m), Winter Spring (February 2018, +2.5 m), and Winter Neap (February 2018, +2.0 m).

To quantify the total and percent inorganic and organic material that was deposited on the marsh surface over a 24 h period, we deployed 8 replicate 4.7 cm diameter filter papers (Whatman Glass Microfiber Filter Paper, Grade GF/F Circles, 47 mm) across five marsh locations at one site. Locations included: 1) outer marsh levee (‘outer levee’), 2) marsh platform 10 m inland from outer marsh levee (‘outer levee-adjacent’), 3–4) mussel aggregations and adjacent non-mussel marsh areas at the tidal creekhead (‘0 m ON mussel mound’ and ‘0 m OFF mound’), and 5) non-mussel marsh platform. Prior to deployment, filter papers were combusted in a 450 °C furnace for 4 h and stored in aluminum foil packets. Packeted filter papers were then labeled and pre-weighed. Once in the field, filter papers were removed from their packet with forceps, placed on a petri dish inserted into the marsh sediment during a Summer Spring low tide (+2.5 m), and secured with 1.5 mm steel wire.

After 24 h, the filter papers were collected with forceps and inserted back in their corresponding packet. Upon transport back to the lab, the packeted filter papers were dried in a 60 °C oven until constant mass was obtained and re-weighed. The change in weight between pre- and post-deployment was used to calculate total dry weight. Packeted filter papers were combusted again in a 450 °C furnace for 4 h and re-weighed. The total dry weight and the weight lost from the second combustion were then used to calculate total inorganic and organic dry weight and percent organic material for each filter paper.

To calculate the organic and inorganic material in persistent in marsh sediment layers, 5-cm cores were collected from the sediment layer using a 60 mL syringe with a 2.5 cm diameter. Cores were taken at same five location types: levee crest, levee-adjacent, on-mound, mound-adjacent, and non-mussel marsh platforms. Eight cores, 1–2 m apart, were collected from each location and placed into pre-weighed foil packets. Cores were dried at 60 °C in an oven until constant mass was obtained, weighed, and combusted in a 450 °C furnace for 4 h. The cores were then reweighed, and the weight loss after combustion was used to calculate the percent organic (and inorganic) material.

The mass of both organic and inorganic material deposited on each mussel aggregation filter was far greater (0.11 g and 0.50 g, organic and inorganic sediment, respectively, here and below) than that deposited on levee crests (0.02 g and 0.06 g), levee-adjacent (0.04 g and 0.15 g), and non-mussel marsh platforms (0.04 g and 0.19 g; F_4,38 = 9.5; p < 0.0001; Adj. R² = 0.47). The percent organic content of this deposited material was higher on levee crests (22.8%) than on mussel aggregations (18.7%) and non-mussel marsh platforms (18.8%; F_4,38 = 9.0; p < 0.0001; Adj. R² = 0.46; Fig. S3), with intermediate values in both levee-adjacent (19.9%) and aggregation-adjacent areas (20.1%; p > 0.05). Despite these differences in the composition of material deposited on filters, the collection and processing of cores spanning the top 5 cm of marsh sediment across the same five zones revealed that the organic content of this upper layer of the marsh was similar across the landscape (p > 0.20), with all locations exhibiting 13–14% organic content (Fig. S3).

Field experiment 1: fate of mussel biodeposits

To assess the distribution of sediment supplemented by mussels via local biodeposition and, in turn, their contribution to sediment supply across the broader marsh landscape, we measured the transport of previously settled biodeposits as well as those actively deposited over one tidal cycle. For each process, we selected 6 mussel mounds in two marsh zones where mussels commonly aggregate: 1) the creekhead and 2) 20 meters away from the creekhead on the marsh platform. All focal mounds were at least 5 meters apart to avoid mixing of biodeposits. We addressed the transport of previously settled biodeposits by first removing 2 cm of each mound’s biodeposit layer, homogenizing it with fluorescent chalk (Irwin Straight-Line Fluorescent Orange Marking Chalk) at a 2:1 ratio (biodepost:chalk), and evenly distributing the mixture back on the mounds. We then revisited the mounds at night after one tide had flooded over the mounds (max tidal height +2.2 m) and traced the distribution of fluorescent material through black light detection. We measured the maximum distance fluorescent material traveled in each direction to quantify transport of previously settled biodeposits across the marsh landscape.

To account for the distribution of biodeposits ejected by actively filter-feeding mussels, we collected 10 mussels from each mound, transported them back to University of Georgia Marine Institute’s wet lab, depurated them in saltwater (Instant Ocean, 28 ppt) for 24 h, and allowed them to feed on a mixture of seawater and fluorescent chalk for 2 h. We then rinsed the mussels to remove any loose fluorescent material from their shells before transplanting them back into the focal mounds at low tide. We then revisited the mounds at night after one tide had flooded over the mounds and traced the distribution of fluorescent material through black light detection. We measured the maximum distance fluorescent material traveled in each direction to quantify transport of actively ejected biodeposits across the marsh landscape.

Field experiment 2: local scale depositional effects of mussels and cordgrass

The second experimental study was conducted at Airport Marsh on Sapelo Island, Georgia, USA. At this site, the experiment was deployed at two zones: the marsh platform >85 m from the nearest tidal creek (31°25’25.3“N 81°17’29.8“W) and the creekhead, where the tidal creek enters onto the marsh platform and tidal water first floods the marsh (31°25’28.1“N 81°17’30.2“W). Within each zone, we deployed seven experimental treatments (n = 5 replicates per treatment per zone) in which we varied mussel (M) presence and density, as well as cordgrass (C) presence. The full set of seven treatments included: 1) no-mussel, no-cordgrass controls (0 M, 0 C); 2) cordgrass-only controls (0 M, C + ); 3) 1-mussel (1 M, 0 C) blocks; 4) small mussel aggregations (20 M, 0 C); 5) intermediate size mussel aggregations (50 M, 0 C); 6) intermediate size mussel aggregations plus cordgrass (50 M, C + ); and 7) large mussel aggregations (80 M, 0 C; Fig. S5).

In July 2017, we harvested 70 blocks of marsh peat (50 cm x 50 cm x 20 cm) from the experimental site using flat-edge shovels. We selected 30 blocks of standardized cordgrass density (48.9 ± 9.0 g dry biomass per block; mean ± SD) from non-mussel areas, 10 blocks containing small mussel aggregations (~20 mussels), 20 blocks of intermediate-size mussel aggregations (~50 mussels), and 10 blocks of large mussel aggregations (~80 mussels). All marsh blocks were transported back to the lab where they were washed completely clean of all surface sediment. With the exception of 10 non-mussel blocks and 10 intermediate-size mussel aggregation blocks, all cordgrass was clipped to the marsh surface. For the 1-mussel treatments, we harvested 10 mussels (6–8 cm in length) from the experimental site and individually inserted them in the center of the marsh block so that they were 40–50% below the marsh surface.

After cleaning and cordgrass removal, all blocks were cut to new dimensions (36 cm x 36 cm x 16 cm) and placed within plastic-encased bins of the same dimensions. Bins containing marsh blocks were then centrally placed and fitted within an additional larger bin (61 cm x 61 cm x 8 cm), with the top of each box flush to the same height. The outside bin was filled with 64, 5 cm diameter PVC poles and 32, 2.5 cm diameter PVC poles (both 8 cm in height) so that all bin edges were held upright and PVC was rigidly filling all space within the outer box (Fig. S4). PVC poles were oriented in this way to capture all deposited sediment and minimize resuspension by substantially decreasing the fetch within the catchment bins. These sediment catchment units were then transported back to the experimental site where recipient holes were dug to the exact dimensions, so that the top of the marsh block (along with the top of each PVC pole) was exactly flush with the marsh surface sediment. We stapled 1-cm hardware cloth mesh (66 cm x 66 cm, with central 36 cm x 36 cm cutout) above PVC and flush to the marsh surface to allow invertebrate access to and from mussel aggregations and to limit the amount of disturbance to and resuspension of the settled material. Finally, to minimize mussel mortality in the absence of cordgrass, we built shades using 2 layers of 5-cm Aquamesh, attached these shades to four bamboo stakes, and inserted them above each plot at a height of ~1 m. The experiment ran for one month, from July 18 to August 18, 2017.

After one month in the field, all experimental units and their contents were returned to the lab, rinsed into recipient aluminum tins, dried, and weighed. The contents of the central bins and sediment on plant tissue were dislodged and collected using spatulas, scraper tools, and a Waterpik Flosser device. After all sediment was collected, each mussel was removed from the aggregation, measured for length, and weighed for biomass. Finally, from treatments containing vegetation, all aboveground cordgrass biomass was harvested, dried, and weighed (Fig. S6).

Delft3D Model

To evaluate the contribution of mussel mounds to marsh accretion, we performed numerical simulations using the Delft3D-FLOW model^63,64. We first modified the source code by adding a bivalve module (Delft3D-BIVALVES) to simulate sediment filtration and deposition processes that lead to mussel mound formations. In building this module, we assumed that mussels remove sediments from the water column because of filtration, and expel them as very cohesive pseudofeces, which are attached to the mounds, increasing their elevation. These processes are simulated by adding, in the computational cells containing the mussel mounds, a depositional term due to mussel filtration that reads:

where ({rho }_{{MM}}) is the density of mussels in the mounds [mussel m⁻²], set equal to 177 mussel m⁻²¹⁴, and ({f}_{{MM}}) is the volume of water filtered by each mussel per unit of time [m³ s⁻¹ mussel⁻¹], set equal to 0.115 m³ s⁻¹ mussel⁻¹. ({C}_{{sed}}) is the sediment concentration in the water column above each mussel mound [kg m⁻³], ({dt}) is the simulation time step [s], set equal to 0.6 s, and ({rho }_{{sed},{dry}}) is the dry density of the sediments [kg m⁻³], set equal to 800 kg m⁻³⁷³. The volume of sediments correspondent to the mussel filtration depositional term obtained from Eq 2. is removed from the lower computational layer of the water column above the mussel aggregation by adding the following sink term in the advection-diffusion equation:

where ({A}_{{cell}}) is the area of the computational cell [m²]. Numerically, the term is implemented implicitly to prevent the appearance of negative concentrations. For settling velocity, we used a value of 0.1 mm s⁻¹. This value provides the best fit of the Total Suspended Sediment (TSS) concentration we surveyed in a creek, on the adjacent Little Sapelo Island, with an error of 0.022 ± 0.025 kg m⁻³ (Fig. S8, MAE + RMSE). The fit was obtained by using the exponential decay formulation that reads:

where ({w}_{s}) is the settling velocity in [m s⁻¹], (h) is the slow depth in [m], ({C}_{s0}) is the initial sediment concentration in [kg m⁻³], and (t) is the time in [s]. We set ({C}_{s0}) equal to 0.10 g m⁻³, which approximates the average value measured during flood tide, at the same location and tidal cycle. In addition, we set h equal to 0.30 m, which is the local mean annual high tide, calculated for 2018. To assess the sensitivity of the results to settling velocity, we ran a simulation in which we increased settling velocity by 50% (i.e., settling velocity equal to 0.15 mm s⁻¹), and extra deposition due to mussel mounds varied by only approximately 6.5% of the original value.

We next established a rectangular model domain to describe our study area in a simplified fashion (Fig. 5a). Within the model domain, the marsh platform is connected to the main channel by a tidal creek. The domain extends for 50 m and 207 m in the long-shore and landward directions, respectively. It is discretized using a rectangular grid constituted of 50 cm × 50 cm cells at the creek head and 50 cm × 100 cm cells elsewhere. In our model domain, mussel aggregations occupy only the creekhead, which is the 50 m × 50 m area between the creek and the upper part of the domain. We assign that each mussel mound has an area of 0.25 m², corresponding to a mound diameter of ~0.5 m. At our resolution, a mound occupies a single cell. A sensitivity analysis using cells of 0.25 m and 0.125 m showed negligible changes in the results. The main channel occupies the lower 20 m of the domain, and its depth goes from 0 m AMSL at the marsh edge to −6 m at the seaward boundary. The tidal creek is located in the middle of the marsh platform and stops 50 m from the landward boundary of the domain. It is 2 m wide, and its depth goes from 0.79 m AMSL at the creek head to −1 m where it connects to the main channel. The marsh system consists of four subareas: (i) the levees (0.94 m AMSL), which are 5 m wide cordons separating the marsh platform from the channel and the creek (except at the creek head) and are vegetated by tall-form cordgrass; (ii) the levee adjacent areas (0.79 m AMSL), which are 10 m wide and vegetated by intermediate size cordgrass, (iii) mussel aggregations, which occupy a set proportion of the creek head (0, 10, or 20%), are vegetated by short-form cordgrass, and form a regular array (0.79 m AMSL, a newly formed mound); and (iv) the marsh platform, all remaining area consisting of short-form cordgrass and located at a uniform elevation of 0.79 m AMSL (Table S2).

We used the Delft3D “trachytopes” functionality to impose vegetation resistance on flow propagating through the model domain. At every time step, a Chézy friction coefficient ((C)) is calculated for the vegetation, using a formulation developed by⁸³. The formula is based on the unvegetated bed roughness (({C}_{b})), the drag coefficient (({C}_{D})), the vegetation height (({h}_{v})), and the vegetation density ((n)), expressed as the number of stems per square meter ((m)) times the stem diameter (({D}_{S})). In our model, only cells with an elevation higher than 0 m above MSL are vegetated. We considered four vegetation zones, as described above (Table S2; see details for collection of cordgrass and mussel parameters below). For each vegetation type, we used the same ({C}_{b}) and ({C}_{D}), equal to 45 m^1/2s⁻¹ and 1.65, respectively⁸⁴. The vegetation properties, for each class, are based on local surveys and are reported in Table S1. For each of the three mussel scenarios analyzed, we considered two vegetation distributions. The first one sticks with the description of the vegetation zones we report above. In the second scenario, the vegetation is absent from the entire domain.

To compute the sediment deposition in our numerical model, we simulated deposition from October 6^th to October 22^nd, 2018. This period contains the most representative spring and neap tides of the year and was obtained using the following procedure. First, we reconstructed the astronomic signal for 2018 using the tidal components of the NOAA station “Daymark #156, Head of Mud River, GA” # 8674975”, which is the closest to our study area. We then calculated the tidal ranges in 2018 using consecutive low and high tide levels extrapolated from the astronomic tidal signal. Next, we classified the tidal ranges using the 25th and 75th quantiles of their distribution (i.e., Q25 and Q75): ranges lower than the 25^th quartile were neap tides and ranges greater than the 75th were spring tides. The 2018 astronomic tide was then divided into periods containing a spring and a consecutive neap tide. For each period, we identified the tidal ranges associated with spring and neap tides by using Q25 and Q75. Finally, for each period, we calculated the average tidal range for neap and spring tide, the difference between these average values and the yearly average, and the sum of these two differences. The period with the lowest value of this sum contains the most representative spring and neap tides of 2018. For this date range, we then ran our model under six scenarios: mussel cover at 0, 10, and 20%, but with and without vegetation present. We report both sediment deposition and annual accretion in the five location types (i.e., levee crest, levee-adjacent, mussel aggregation, aggregation-adjacent, and non-mussel marsh platform) at local (1 m²), creekhead (2500 m²) and entire domain scales (10,350-m²).

Field experiment 3: creekshed mussel manipulation

To assess the effects of mussel presence and population size on marsh accretion at the creekshed scale, we first selected a marsh creekshed with three adjacent tidal creeks of similar length, structure, associated mussel populations, and marsh platform characteristics (i.e., size, elevation, and cordgrass characteristics). For each of the three tidal creeks, we first delineated a 50 m by 50 m creekhead area, oriented perpendicular to the direction of the tidal creek entry into the marsh, and located with midpoint of the front edge positioned at the point of tidal creek entry into the marsh. We then delineated a larger creekshed area associated with each creek of ≥10,000 m² within which we would deploy our experimental treatments. To quantify initial mussel and cordgrass cover, we set up three 50 m² transects (50 m long, 1 m wide) within the creekhead area, located at 0 m, 20 m, and 40 m distance from the tidal creek point of entry (and oriented perpendicular to the direction of flow). Within each transect, we counted each mussel aggregation, scoring each individual mussel as well as the length, width, and height from marsh platform of each mussel aggregation structure.

For a subset of 20 mussel aggregations randomly selected within each transect (3 transects per creek, 180 aggregations total), we scored the total number of cordgrass tillers on each aggregation. For a subset of 5 randomly selected tillers on each aggregation, we measured both length and width. To assess the differences in cordgrass characteristics between mussel aggregations and aggregation-adjacent areas, we also measured cordgrass stem density, height, and diameter in non-aggregation areas (1 m²) located 1m away from each mussel aggregation.

After all initial data was collected, we removed and transplanted approximately 200,000 mussels from one tidal creekhead to another. To do so, we initially flagged approximately 4000 mussel aggregations within the creekshed area of the “Removal” creek, encompassing both the 2500 m² creekhead area as well as the surrounding ≥10,000 m² creekshed extent. Mussel individuals were removed by hand over the course of 16 weeks, with all field personnel taking care to leave all pseudofeces in place and cordgrass intact. Field crews were split between the mussel removal and mussel addition creek, such that mussels were re-transplanted within 24 h of removal to minimize mortality. Due to logistical and permitting constraints, it was not feasible to replicate the treatments across multiple sites; instead, the three plots occupied a single contiguous creekshed (Fig. 6a, b).

To assess changes in marsh elevation, we first quantified initial creekhead elevation (mean m AMSL in 2500-ft² area perpendicular to point of entry) using two metrics: 1) Real Time Kinematic (RTK) elevation datapoints (Trimble R6 GNSS System) distributed across the creekshed; and 2) measurements of mussel mound heights throughout each transect at set distances from the point of water entry. For the RTK datapoints, we collected 86 total points across the creekshed in June 2017. Elevation datapoints were randomly selected in each 2500 m² creekhead zone (minimum of 20 points per creekhead; Fig. S7). However, given the low number of RTK points across a large area, we additionally utilized mussel mound height calculations to provide a second estimate of initial elevation across the creekshed. Mussel aggregations and other bivalves, such as oysters, exhibit a height ceiling of growth, above which survivorship and growth are hypothesized to decrease. Previous work on Sapelo Island marshes reported the height ceiling to be +0.84 ± 0.004 m AMSL (mean ± SE). Therefore, assuming mature mussel aggregations (i.e., with tops at the aforementioned height ceiling), then mussel aggregation height (i.e., the distance between the marsh platform and the topmost point of the mussel aggregation mounded structure) will inform our knowledge of the marsh platform elevation by the following equation: Marsh Elevation (m AMSL) = Mussel Height Ceiling (+0.84 m AMSL) – Mussel Aggregation Height (m AMSL). For each distance from creekhead from which we conducted a 50 m² transect (0, 20, and 40 m), we estimated mean platform elevation using each of the measured mussel aggregation heights. We then took the mean value of marsh elevation across the three distances (0 m, 20 m, and 40 m) as a measure of creekhead elevation in 2017 for each of our experimental creeks (>60 mounds per creekhead; 250 total).

To assess elevation three years after treatment deployment, we compared creekhead elevation using a 2020 Digital Elevation Model (DEM) of the creekshed. To build the DEM, we flew a DJI Matrice 600 Pro drone carrying a custom build Lidar payload in August 2020. The payload consisted of a Velodyne Puck Lite VLP16, paired with a Novatel Stim300 Inertial Measurement Unit. The point clouds from the drone were orthorectified from GPS data continuously measured on the drone (see the procedure described in ^85,86). To remove the vegetation and any other surface perturbations (i.e., from digital surface model to digital elevation model), we used the CloudCompare software (https://github.com/cloudcompare/cloudcompare). The cloth Simulation Filter (CSF; ⁸⁷) was applied twice to the dataset, which successfully removed the vegetation data. The point cloud of the marsh surface was then exported to ArcGIS 10.7 where the DEM was generated by raster interpolation. Once completed, the mean elevation within each 2500 m² creekhead location was calculated using the Zonal Statistics tool in ArcGIS 10.7.

Statistical analyses

To quantify the effects of season, tidal phase, and location type on short-term deposition, we first square root transformed short-term sediment deposition (i.e., filter paper results) to meet the assumptions of parametric statistics. We then conducted a three-way fully factorial ANOVA, with main effects season, tidal phase, and location type. Post-hoc analyses were conducted with Tukey HSD test, with Bonferroni-corrected p-values (STATA v 15.1). We further analyzed the effects of site, season, tide, and marsh location on short-term sediment deposition using regression tree analysis (rpart, R version 3.1.0). Over-fitted trees were pruned using k fold cross-validation. To next assess the effects of marsh location type on total organic material deposited over 24 h (surface) and percent organic material (surface and to 5 cm depth), we ran three separate one-way ANOVAs. Post-hoc analyses were again conducted with Tukey HSD tests, with Bonferroni-corrected p-values (STATA v 15.1). For Experiment 1, we assessed the fate of mussel biodeposits, both previously settled and newly ejected, with a one-way ANOVA with location (creekhead versus platform) as the main effect. Finally, for Experiment 2, to assess whether cordgrass and mussel aggregations significantly affected sediment deposition over the one-month experimental deployment, we used multiple regression analysis with cordgrass biomass and mussel biomass as predictor variables for sediment biomass collected in each zone (STATA v 15.1; Table S1).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Source: Ecology - nature.com