Modeling Impacts of Submersed Aquatic Vegetation on Sediment Dynamics Under Storm Conditions in Upper Chesapeake Bay

Abstract

Submersed aquatic vegetation is an important modulator of sediment delivery from the Susquehanna River through the Susquehanna Flats into the Chesapeake Bay. However, the dynamic interactions between vegetation and the physical drivers of sediment transport through the region are not well understood. This study used a new vegetation component in a coupled flow-wave-sediment transport modeling system (COAWST) to simulate summer through fall 2011, when the region experienced a sequence of events including Hurricane Irene and Tropical Storm Lee. Fine sediment was exported under normal flows and high-wind forcing but accumulated under high flows, with accumulation doubling in the presence of vegetation. The relative effects of vegetation under normal flows and high-wind forcing depended on previous sediment dynamics. The presence of vegetation facilitated deposition of fine sediments within the patch under normal flows and subsequent resuspension during a wind-driven event. However, during significant discharge events, the system was completely dominated by riverine dynamics with vegetation enhancing net deposition as well as channeling the flow. While further refinement of the bed model may be needed to capture some nuances, the COAWST modeling system provides new insights into detailed sediment dynamics in complex vegetated deltaic systems.

Introduction

Suspended sediments play an important role in maintaining estuarine ecosystems; for example, they provide nutrients to support productivity (Kemp et al. 2005) and facilitate marsh development through deposition (Redfield 1972; Donatelli et al. 2018). However, excess suspended sediments can degrade water quality, limiting light availability to benthic organisms like submersed aquatic vegetation (SAV) (Davis 1985; Cheng et al. 2013), prevalent in many estuaries and coastal embayments. SAV provides nursing grounds for juvenile fish (Orth et al. 2010) and enhances water quality by reducing water current velocities, thus facilitating settling of smaller suspended particles (Sand-jensen 1998), creating a positive feedback loop that enhances their resilience to storm events (Gurbisz and Kemp 2014). Understanding feedbacks between sediment transport and SAV is integral to determining the ability of SAV to modulate water quality (Hirsch 2012), especially by trapping terrestrial material otherwise bound for estuarine environments (Kemp et al. 2005).

In the upper Chesapeake Bay (CB), excess fine sediment is one of the main contributors to the degradation of water quality and adversely impacts habitats of living resources (Langland and Cronin 2003). Anthropogenic activities have greatly influenced the supply of watershed sediments to the Susquehanna River (SR), the main tributary of CB, and subsequent delivery to upper CB (e.g., milldams, Conowingo Dam) (Walter and Merritts 2008; Zhang et al. 2016). However, the fate of sediment is affected by biophysical processes within the SAV beds of the Susquehanna Flats (Flats), which is the subaqueous delta of the SR at the head of CB. Extensive SAV beds historically occupied the Flats but effectively disappeared following Hurricane Agnes in 1972. However, they made a dramatic resurgence in the early 2000s (Gurbisz and Kemp 2014), motivating several studies to determine their impact on sediment transport in the region (Bayley et al. 1978; Dennison et al. 2012; Palinkas et al. 2014; Gurbisz et al. 2016; Russ and Palinkas 2018). These studies show that SAV does modulate suspended sediment transfer from fluvial to estuarine environments by facilitating seasonal storage of material (Bayley et al. 1978; Kemp et al. 2005; Gurbisz et al. 2016; Russ and Palinkas 2018), but the specific physical mechanisms active during storm events lack robust evaluation.

From mid-summer 2011 through late-fall 2011, a series of events offered the opportunity to examine these mechanisms. In particular, an extreme wind event, Hurricane Irene, occurred from 27 to 30 August 2011, providing insight into the impacts of wind events during times of low river flow. Then, from 7 to 16 September 2011, the remnants of Tropical Storm Lee (TS Lee) produced significant precipitation over the SR watershed, leading to the second highest recorded SR discharge (20,000 m³ s⁻¹) and massive associated sediment discharge over the SF (Gurbisz et al. 2016) but relatively low winds. Approximately 1 month after TS Lee, a high-wind event enhanced water turbidity over the Flats, likely via resuspension of fine particles deposited during TS Lee (Gurbisz et al. 2016; Russ and Palinkas 2018).

In this study, we use the Coupled-Ocean-Atmosphere-Wave-Sediment Transport (COAWST) modeling system (Warner et al. 2010) with a recent flow-vegetation interaction module (Beudin et al. 2017) to evaluate the impact of SAV on the transport of suspended sediment in the region. A necessary first step in using this relatively novel mechanistic model is to establish its ability to reproduce this dynamic environment at a high resolution. Then, we investigate 4 scenarios to assess responses to different types of events with and without vegetation, as well as the impact of event timing: (1) a typical flow and wind pattern before Hurricane Irene, (2) a high-wind event associated with Hurricane Irene, (3) a high-flow event associated with the remnants of TS Lee, and (4) a high-wind event after TS Lee. Our overarching hypotheses are that sediment dynamics during typical and event conditions are significantly different and that the sequence of events is crucial to those impacts. Additionally, we hypothesize that the existence of plants over the Flats modulates sediment transport to the rest of CB by reducing current speeds and waves over the Flats, therefore enhancing sediment deposition and acting as a temporary storage area for fine particles (Gurbisz et al. 2016).

Calibrating and evaluating the COAWST system provides an open-source approach to investigating the complex interactions within this highly dynamic fluvial-estuarine interface. By evaluating the various mechanisms controlling sediment retention and erosion within the region during storm events, potential impacts to downstream ecosystems can be better predicted to aid in management efforts.

Methods

Site Description

The Susquehanna Flats region is the subaqueous delta of the Susquehanna River (SR) located at the tidal headwaters of the Chesapeake Bay (CB) (Fig. 1). The depositional basin of the Flats has a bowl-shaped geometry covering roughly 89 km² (Davis 1985) with a primary channel along the western side approximately 3–7 m deep (Gurbisz and Kemp 2014). Physical processes in this region are influenced primarily by the flow of the SR, which has an average discharge of 1100 m³ s⁻¹ (Schubel and Pritchard 1986) and is the largest source of sediment to the upper CB (Fig. 1). The SR typically has its highest discharge in spring from snow melt and spring rains followed by low-to-moderate flow for most of the year (Gross et al. 1978). Tidal currents ranging from 0.5 m s⁻¹ in the channel to 0.3 m s⁻¹ over SF are associated with a mean tidal range of 0.6 m (Browne and Fisher 1988; Bayley et al. 1978). Wave action with significant wave heights < 1 m and mean periods of < 2.5 s (https://buoybay.noaa.gov/), and seasonal vegetation (Bayley et al. 1978; Gurbisz and Kemp 2014) also play significant roles in sediment transport in upper CB. In particular, vegetation reduces currents and wave activity, facilitating sediment deposition (Gambi et al. 1990; Granata et al. 2001; Peterson et al. 2004). Historically, the Flats hosted a robust vegetation community; however, this population was decimated by Hurricane Agnes in 1972. There was a resurgence of SAV in the mid-2000s (Gurbisz and Kemp 2014). Recent observations have highlighted the ability of SAV to trap sediments, modulating sediment input into CB (Russ and Palinkas 2018).

Fig. 1

Map of study region with points denoting the locations of USGS Conowingo Dam discharge observations (C), turbidity observations from Maryland Department of Natural Resources Chesapeake Bay-Segment 1 Susquehanna Flats (FLT) and Susquehanna River—Havre de Grace (SUS), wind and current observations (CBIBS), wave observations from 2013 (Tripod), surface elevation prediction (SHAD), and surface elevation measurement (TOL). Lines T1 and T2 are the transects used in the flux calculations at the entrance and exit of the system, respectively. The colors indicate the MLLW bathymetry, in meters, scaling from shallow water (yellow) to deep water (green), used for model initialization. The vegetation patch supplied to the model is the dark yellow region. The dotted line represents the 3-m isobath

The SR contributes 50–75% of the annual sediment load to the CB. The total load is composed of approximately 10% sand, 50% silt, and 40% clay, transported during peak stream discharge, with fine material clearly dominating the sediment loading (Williams and Reed 1972; Chesapeake Research Consortium 1976). On the Flats, surficial sediments are sandy, with median diameters ranging from 113.1 to 405.3 μm and generally decreasing with distance from the river mouth (Russ and Palinkas 2018). However, fine sediment accumulation has been observed in the middle of the vegetation in regions with high sedimentation rates, ~ 1 g cm⁻² year⁻¹ (Russ and Palinkas 2018).

The variety of wind and river discharge magnitudes from late summer and through the fall of 2011 provide insight into potential controlling mechanisms for sediment transport and deposition (Table 1 and Fig. 2). From mid-July to late-August, wind velocities and river discharges were both relatively low (“before Irene”), representing typical summer conditions with pulses of river discharge due to the peak production facility at Conowingo Dam (Palinkas et al. 2019). Then, Hurricane Irene passed through the region at the end of August, with sustained maximum winds of 14 m s⁻¹ but low river discharge (“Irene”). This was followed by the passage of the remnants of Tropical Storm (TS) Lee in early September (“Lee”), which brought much rainfall but little wind to the region. SR discharge peaked at 20,000 m³ s⁻¹ during this event, the second highest discharge on record. Finally, a wind event occurred in mid- to late-October that had sustained winds of 10.5 m s⁻¹ and moderate river discharge (“Post-Lee”). In this study, we take advantage of the different environmental conditions associated with each of these periods to investigate potential controlling mechanisms on sediment transport and deposition.

Table 1 Description of the wind and river discharge conditions during each event encompassing the time period from start date to end date. High indicates the dominant forcing during the event

Fig. 2

Stacked plot of the model forcing used throughout the simulation with the events from Table 1 highlighted in grey. The top panel represents a stick vector plot of the wind velocities at the NOAA-NOS CBIBS Susquehanna station (U₁₀). N and S are included in the panel to indicate the direction the winds are moving towards. The middle panel represents the river discharge observations from the USGS sensor at Conowingo Dam (Q). The bottom panel represents the phase-adjusted Tolchester Beach water surface elevation observations at MLLW (d)

COAWST Description and Initialization

This study uses the Coupled-Ocean-Atmosphere-Wave-Sediment Transport (COAWST) modeling system (Warner et al. 2010) to evaluate the impacts of vegetation on sediment transport and deposition during the 2011 events on the Flats. The COAWST modeling system includes several community-developed model components, including the Regional Ocean Modeling System (ROMS; Shchepetkin and McWilliams 2005), Simulating Waves Nearshore (SWAN; Booij et al. 1999), Weather Research and Forecasting Model (WRF; Skamarock et al. 2005), and the Community Sediment Transport Modeling System (CSTMS; Warner et al. 2008). For this study, the ROMS-SWAN-CSTMS components were used to investigate interactions among currents, waves, sediment, and vegetation. New vegetative components recently implemented in ROMS, namely plant posture-dependent three-dimensional drag, vertical mixing, and wave-induced streaming (Beudin et al. 2017), were activated as well.

As detailed in Chen et al. (2007), various models have been developed to simulate flow through vegetation, but each has its limitations. The implementation by Beudin et al. (2017) attempts to resolve these limitations through a coupled approach. This is accomplished by sending the wave energy dissipation due to vegetation, calculated in SWAN, into the ROMS momentum balance as an additional wave-averaged forcing (Beudin et al. 2017). This formulation allows for the flow through vegetation to be calculated based on the flexibility of the vegetation, instead of an enhanced bottom roughness used previously (Morin et al. 2000). The flexible vegetation formulation better accounts for the observed wave energy dissipation from the drag force on the vegetation, which modifies the wave characteristics (Beudin et al. 2017).

COAWST was initialized similar to the scenario described in Beudin et al. (2017) across 18 parallel nodes. The model grid, used in both ROMS and SWAN, was established as a rectangular grid, with a longitudinal distance of 16.49 km and a latitudinal distance of 23.76 km, evenly divided into 100 × 100 cells and 5 layers in topography following sigma coordinates (Fig. 1). The simulation was executed from 19 July 2011 through 1 November 2011 with time steps of 30 s for both ROMS and SWAN with coupling every 30 s. The precise start date and time were selected to match low tide exactly to mean lower low water, which is the reference for the bathymetric grid; the end date was selected to encompass the “post-Lee” wind event.

Bathymetry was obtained from the National Oceanic and Atmospheric Administration’s National Ocean Service, 30-meter resolution Digital Elevation Model (https://ngdc.noaa.gov/mgg/bathymetry/estuarine/index.html) (National Centers for Environmental Information 2017) (Fig. 1). Using GridBuilder v0.99, the land cells were established as masks using the Global Self-consistent Hierarchical High-resolution Geography (GSHHG) coastline in combination with manual manipulation of grid cells for accurate depiction of the coastline (Wessel and Smith 1996).

Submersed aquatic vegetation (SAV) distributions over the study area were obtained from the 2010 annual SAV survey by the Virginia Institute of Marine Sciences (VIMS), collected in early October 2010 over the upper Bay. Data were downloaded from the VIMS Chesapeake Bay Submersed Aquatic Vegetation data site (https://www.vims.edu/research/units/programs/sav/access/index.php). Areas with 70–100% SAV coverage and water depths less than 2 m (Kemp et al. 2004; Kreiling et al. 2007) were included as vegetated cells in the model (Fig. 1). Vegetation characteristics were assigned based on vegetation parameters that worked well in previous wave-current-vegetation-sediment models (e.g., Chen et al. 2007. Beudin et al. 2017). Specifically, average vegetation height, diameter, thickness, and density were 0.305 m, 0.001 m, 0.001 m, and 100 stems m⁻², respectively. With the implementation of flow through vegetation as established by Beudin et al. (2017), a singular vegetation type with stem flexibility (elastic modulus) of 1.0 GPa was used, as estimated in Luhar et al. (2008) for similar submersed vegetation. Vegetation drag coefficients for ROMS and SWAN were set to 1.0, which is typical for low values of the areal density parameter ad such as those modeled here (Chen et al. 2007). Vegetation density was established as 700 kg m⁻³ (Beudin et al. 2017), and the additional horizontal viscosity coefficient at the edge of the vegetation patch was set to 0.10 m² s⁻¹ (Beudin et al. 2017). The vegetation parameters were chosen conservatively so as not to overestimate vegetation drag, buoyancy, and flexibility.

ROMS sediment characteristics were established using two sediment classes, cohesive mud (referred to as “fine sediment”; D50 = 4 μm and ws50 = 0.02 mm s⁻¹) and non-cohesive sand (referred to as “coarse sediment”; D50 = 375 μm and ws50 = 0.85 mm s⁻¹) with specific properties established based on average values from field observations (see below). Sand grain size is only important in the model for the active layer thickness of the sediment bed and the grain roughness, so coarse sediment size was parameterized using the range of values observed for surficial sediment (upper 1 cm) on the Flats by Russ and Palinkas (2018). The sand grain size intentionally does not correspond to the sand settling velocity. Settling velocities were determined separately as the average of the two finest and the two coarsest settling velocities of Palinkas et al. (2019). The simulation neglected the effects of flocculation due to the limited flocculation of particles in this high-energy freshwater system (Palinkas et al. 2019). Additionally, bedload transport was not included since the focus of this effort was to evaluate suspended sediment transport. This approximation is justified by the very low Rouse number characteristic of the modeled particles under moderately energetic conditions (0.005 and 0.2 for the fine and coarse particles respectively at a shear velocity of 0.01 m s⁻¹). Due to complex near-bed wave-orbital characteristics during the simulation period, the bottom boundary layer was simulated using a wave-current bottom boundary layer model which internally calculates the bottom roughness using the supplied grain sizes (Madsen 1995; Beudin et al. 2017).

The sediment bed was initialized with three sediment layers: a 1-cm-thick active-mixed layer composed of 13.24% fine and 86.76% coarse, overlying a 1-cm-thick coarse layer to act as a transition to a 99-cm-thick coarse layer, as observed by Russ and Palinkas (2018). All suspended sediment during the simulation was from either SR suspended sediment loads or resuspended sediment from the bed layers. Suspended sediment concentration (SSC) data for the SR were not available for the entirety of the model simulation. Instead, the empirical relationship between SSC and SR discharge (Q) during high-flow by Cheng et al. (2013) was implemented; SSC was separated into the two size classes following Palinkas et al. (2019) and using Eqs. (1) and (2) below.

$$ {f}_{\mathrm{fine}}= SSC\times \left(-0.941\times {10}^{-5}Q+0.983\right) $$

(1)

$$ {f}_{\mathrm{coarse}}= SSC\times \left(0.941\times {10}^{-5}Q+0.017\right) $$

(2)

The ROMS boundary conditions for the northern and western edges of the domain were established as gradient conditions for the free surface and depth-averaged velocities, which sets no variation across the boundary and simplifies computational complexity. The southern boundary was established as a Chapman Explicit condition for the free surface, which allows the free surface to vary based on input data (Hedström 2018), and a Flather condition for the depth-averaged velocities as recommended for river outflow (Hedström 2018). The velocity, temperature, salinity, and sediment were established as gradient conditions across all boundaries. Since the main interest of this study is the flow of SR water and sediment, the North East River and Elk River were masked to reduce possible sources of error.

Tidal forcing was established as a boundary condition along the southern grid using water elevation data from the NOAA CO-OPS station Tolchester Beach (station id: 8573364 https://tidesandcurrents.noaa.gov/stationhome.html?id=8573364) phase shifted to the model grid. The phase shift was accomplished by using the NOAA Chesapeake Bay Operational Forecast System (CBOFS) forecasted water elevations for Tolchester Beach and Shad Battery (a point near the model boundary) (Fig. 1). After comparing various temporal adjustments, a positive shift of 78 min applied to the Tolchester Beach forecast elevation data resulted in a matching of the high and low tides to the Shad Battery forecasted elevations. The amplitude of the surface elevation was adjusted according to Eq. (3).

$$ {z}_2=c\left({z}_1-{z}_{1\ avg}\right)+{z}_{1\ avg} $$

(3)

where z₁ is the elevation at the initial point (Tolchester Beach), z_1 avg is the average elevation at the initial point during the time series, c is the adjustment factor, and z₂ is the elevation at the point of interest. For this study, a value of 1.1 (a 10% increase) for the adjustment factor c leads to agreement between the adjusted Tolchester Beach predictions (z₁) and the model boundary Shad Battery predictions (z₂) (Lee et al. 2017). Tides were forced relative to mean lower low water, varying the surface strictly above mean lower low water. The historical data from CBOFS dates back to 2014; therefore, the September 2014 predictions were used to establish this relationship since seasonality and river discharge has a large impact on water level variations in the region on decadal scales (Barbosa and Silva 2009).

SWAN was established in non-stationary mode using GEN3 physics, vegetation, and forced by wind input from observations, as described below. For this study, only the wind observations from in situ stations were used to perturb the system. Due to fetch limitation (Sanford 1994) and typical wave characteristics of the region, the Komen method for wave dissipation (Komen et al. 1984), with a frequency range from 0.2 to 2 Hz with 24 meshes in frequency space, was established to capture small wave periods and small wave heights, which is vital for this region (Fisher et al. 2015).

Wind forcing for the SWAN portion of the simulation was established from the CBIBS Susquehanna station wind observations, which were adjusted from an observation height of 3 m to the required observation height of 10 m for SWAN by using the power law profile (Fig. 2) (Hsu et al. 1994; Kilbourne 2017). The 10-m velocities were applied to the model grid as spatially invariant constant wind speeds every 10 min. For river discharge, in situ observations from the USGS gauge at the Conowingo Dam outlet (https://waterdata.usgs.gov/usa/nwis/uv?01578310) were adjusted according to Eq. (4):

$$ Q=\left({Q}_{obs}/{n}_{rivers}\right)\ast 0.8 $$

(4)

where Q_obs is the observed river discharge from the Conowingo station in m³ s⁻¹, n_rivers is the number of river grid cells over which the flow is distributed (Harris et al. 2008; Xue et al. 2012; Liu and Wang 2014), and 0.8 represents a 20% reduction in flow. This was forced into ROMS as a horizontal momentum transport point source. The 20% reduction was needed to align the simulated current velocities with observed values at the CBIBS Susquehanna station (Fig. 3), as determined during development of the discharge model and similar to the adjustments made for the Hudson River in Warner et al. (2005). The timing of river discharge was phase shifted by 2 h to account for the 11.3-km distance downriver from the Conowingo Dam to the model’s northern boundary (Fig. 1 and Fig. 3).

Fig. 3

Time series (top panel) of the magnitude of the water velocity (m/s) from CBIBS bin 2 ADCP observations (dots) and the respective location within the model grid (black line). The bottom two panels are comparisons between the observations (y-axis) and the model predictions (x-axis) (m/s), for the North (left panel) and East (right panel) components of the water velocity (m/s). The red dots indicate the velocities between 6 September and 20 September and the grey dots are the observations outside that time interval. The solid line is the linear regression of the observed and predicted velocity components for the entire simulation (grey and red dots combined) and the dashed line is the 1:1 line

The model was configured to simulate conditions from August to November 2011 listed in Table 1. The conditions include typical wind and river discharges, a significant wind event (end-August, Hurricane Irene) followed by a large river discharge event (early-September, Tropical Storm Lee), and another significant wind event with moderate river discharge (mid-October) (Fig. 2). Two scenarios were run for each condition: one with vegetation (“veg”), and one without vegetation (“no veg”). Both of these scenarios used all of the sediment, river discharge, and wind forcings. However, only the vegetated scenario used the vegetation patch and vegetation module during model initialization.

Model performance was evaluated at the CBIBS and Tripod sites (see Results); the impact of vegetation was assessed at site FLT on the Flats (Fig. 1). Additionally, sediment budgets were evaluated by calculating imports/exports and geospatial differences between two transects: the mouth of the Susquehanna River (T1 in Fig. 1), and a longitudinal transect between Turkey Point and Sandy Point (T2 in Fig. 1). These two transects represent the entrance and exit from the Flats system, respectively. For each event, the bed elevation and mass change were calculated as the difference between the final and initial values at each cell during the selected time period. The sediment budget was determined by computing the cumulative southward flux of sediment, by class, and integrating over the specified time interval resulting in total metric tons crossing southward at each transect.

Results

Validation of Model-Predicted Currents and Waves

To validate model-predicted currents, depth-averaged velocities from the full model simulation and in situ velocity observations from the Chesapeake Bay Interpretive Buoy System (CBIBS) at station Susquehanna were compared (Fig. 3). Highest velocities in both time series occurred during TS Lee in September 2011, due to widespread, enhanced precipitation over the SR drainage basin and associated extreme river flow (Fig. 2). Large, flood-related increases in surface elevation were also simulated in the lower Susquehanna River and in Susquehanna Flats (not shown). However, the CBIBS current velocity observations during the event were flagged as questionable, either by an automated check using the Quality Assurance of Real-Time Oceanographic Data algorithms or a human expert reviewer, according to the quality control flags in the original data set (Kilbourne 2017; https://accession.nodc.noaa.gov/0159578). The buoy itself was likely tilted over and dragged by the flood waters, and had to be recovered, serviced, and repositioned after the storm (based on buoy position data archived at https://buoybay.noaa.gov). Therefore, CBIBS velocity observations during this event are included in Fig. 3 only to indicate that large, potentially destructive currents were observed, and not for model verification. The north and east velocities from the model before 6 September and after 20 September aligned well with the north and east velocities from the CBIBS platform, falling along the 1:1 component velocity lines in Fig. 3.

To validate the performance of the wave model (SWAN), an additional SWAN-only simulation was performed for July 2013, when direct observations of the wave environment were available (Sanford, unpub. data). The wave model was configured as described in the methods with the following adjustments to account for the specific characteristics of the region in 2013. Since tidal forcing was not included in this simulation, the water depth was uniformly increased by 0.4 m to adjust the reference water level from mean lower low water to mean tide level to better simulate wave dynamics over the Flats. Additionally, vegetation was not included in the wave configuration since it was absent during the 2013 observational period (http://web.vims.edu/bio/sav/sav13/quads/ss009th.html). The 2013 wave model predictions agreed well with the in situ observations (Fig. 4), even in the absence of tidally varying surface elevations. Specifically, agreement with the rapid increase and decrease in significant wave heights and peak wave periods associated with changes in wind direction and speed indicated the model’s responsiveness and accuracy in this dynamic wind environment. Although there is some under-prediction of peak wave period during small significant wave heights (< 10cm), this is consistent with other model-data comparisons (Lin et al. 2002) where it is attributed to low signal-to-noise problems in the data under small wave conditions.

Fig. 4

Observed winds a as exported from SWAN every 30 min from the tripod platform location (U₁₀), b comparison of significant wave height (H_s), and c comparison of peak wave period (T_p) from the tripod platform (grey dots) and SWAN model predictions for the same location (black line) from July 05 to July 15 2013. N and S are included in panel a to indicate the direction the winds are moving towards

Event Dynamics

To explore sediment transport dynamics during the events, several different aspects of the model predictions were used. These included (1) time series at the central Susquehanna Flats site (FLT); (2) geospatial changes of the sediment bed layer, depth-averaged water velocity, fine and coarse SSC, and significant wave height; and (3) sediment fluxes entering and exiting the system. These metrics were used to evaluate the impacts of widely varying forcing, presence or absence of vegetation, and timing during the four events. The results presented here are selected illustrations of the full set of results which are available for viewing in the Appendix.

Typical Conditions

Under typical conditions (“before Irene” in Table 1), wave and current magnitudes were low and river flow occurred as low, widely spaced pulses (Fig. 5). Vegetation on the Flats generated a nearly 30% reduction of current speeds within the SAV bed, resulting in bottom stress reduction of ~ 50% (Fig. 5). Tidal bottom stresses exceeded the critical value for erosion of both sediment classes under non-vegetative conditions, resulting in increased suspended sediments in both classes (Fig. 5). Fine sediment concentrations reached 0.05 kg m⁻³ under non-vegetative conditions compared to almost 0 kg m⁻³ with vegetation (not shown). On the other hand, peak current velocities in the main channel (during ebb tide and a discharge event on 1 Aug 2011 at 21:00) reached a maximum of 0.49 m s⁻¹ under vegetative conditions and 0.42 m s⁻¹ under non-vegetative conditions due to flow focusing into the channel by vegetation (not shown). There was net export of sediment from the system in both scenarios, primarily in the fine sediment class (Table 3). Export was enhanced under non-vegetative conditions, and included coarse sediments as well as fine sediments. Under vegetative conditions, there was net accumulation of coarse material within the system (Table 3). Changes in fine sediment bed layer mass showed removal from the main channel under vegetative conditions, and slight removal over the shallow Flats under non-vegetative conditions consistent with the local resuspension indicated in Fig. 5.

Fig. 5

Model prediction summary for FLT during typical conditions with vegetation (solid) and without vegetation (dashed). From top to bottom, the parameters are Q—river discharge (m³ s⁻¹); $S{S}^{-}{C}_f\_$—depth-averaged fine sediment concentration (kg m⁻³); $S{S}^{-}{C}_c$—depth-averaged coarse sediment concentration (kg m⁻³); $\mid {v}^{-}\mid$—depth-averaged current magnitude (m s⁻¹); and |τ_b|—maximum wave and current bottom stress magnitude (N m⁻²) with a dotted line indicating the critical shear stress of 0.049 N m⁻²

Hurricane Irene

During Hurricane Irene, river discharge was low at 780 ± 722 m³ s⁻¹ (values are reported as mean ± 1 standard deviation) and winds were high, peaking at 14.4 m s⁻¹ on 28 Aug at 13:00. Wave heights in deep water exceeded 0.5 m under both scenarios (Fig. 6). Wave heights with vegetation were strongly damped in the center of the bed (average 0.154 ± 0.066 m). Wave heights were still damped over the flats without vegetation, but much less strongly (average 0.215 ± 0.066 m). On 28 Aug at 08:00, the water depth at the FLT site approached 0 m, causing some questionable model responses for a short period of time. However, responses before and after the extreme low tide quickly rebounded to realistic current speeds and wave heights. Wave-induced bottom stresses were enhanced by an order of magnitude in the non-vegetated scenario, greatly increasing suspended material. Bed elevation increased within the vegetation patch when vegetation was present but decreased in the same region when vegetation was absent. Fine sediment bed mass over the flats increased with vegetation and decreased without vegetation (Fig. 6). Overall, there was net loss of fine sediments from the region and retention of coarse sediments within the region in both scenarios (Table 3). However, vegetation reduced the export of fine material and enhanced the retention of coarse material (Table 3).

Fig. 6

Top two panels are the spatial distribution of significant wave heights (m), smaller (blue) to larger (yellow), during peak winds in Irene, on 29 August 2011 at 13:00, under vegetative (left) and non-vegetative (right) conditions. On each panel, an arrow is included to indicate wind direction and speed at the specified time. The scale bar for these panels ranges from 0 (purple) to 0.5 m (yellow). Bottom two panels are the spatial distribution of the mud mass difference (Δm_f kg m⁻²) between the final and initial mass sum in the bed layer over the time period associated with the Irene event, under vegetative (left) and non-vegetative (right) conditions. Coloring indicates removal of mass (blues and greens) and addition of mass (reds) over the time period. The scale bar for these panels ranges from − 1 (blue) to 0.6 (red) kg m⁻². A dotted line is included in all panels for reference to the 3-m isobath

Tropical Storm Lee

During Tropical Storm (TS) Lee, the system was heavily influenced by the high river discharge (Fig. 7), which resulted in maximum current velocities of 4 m s⁻¹ at the mouth of the SR (Fig. 8). Under both scenarios, there was a significant bottom scour at the mouth of the SR (Fig. 8). The impact of vegetation was most obvious in the reduction of coarse suspended sediment within the vegetation patch (Fig. 7). The contrast between channel erosion and Flats deposition was greatly increased with vegetation (Fig. 8). The amount of sediment that entered the domain was extremely large and resulted in much deposition (Table 3 and Fig. 8). Under vegetative conditions, more fine sediment was trapped in the system, mainly within the vegetated region (Fig. 9). Note that the fine sediment deposition near the north-east boundary of the system was most likely an artificial response to the model boundary conditions in that region. The coarse material mass difference showed that vegetation impacted geospatial patterns of sand deposition mostly by sharpening the contrast between the flats and the channel and by limiting the extent of sand deposition over the flats (Fig. 9). Overall, vegetation increased the retention of fine sediment and decreased the retention of coarse sediment, relative to no vegetation (Table 3).

Fig. 7

Model prediction summary for FLT during Lee with vegetation (solid) and without vegetation (dashed). From top to bottom, the parameters are Q—river discharge (m³ s⁻¹); $S{S}^{-}{C}_f$—depth-averaged fine sediment concentration (kg m⁻³); $S{S}^{-}{C}_c$—depth-averaged coarse sediment concentration (kg m⁻³); h_b—bed thickness (m); $\mid {v}^{-}\mid$—depth-averaged current magnitude (m s⁻¹); and |τ_b|—maximum wave and current bottom stress magnitude (N m⁻²) with a dotted line indicating the critical shear stress of 0.049 N m⁻². The vertical dashed line represents the peak discharge event from Lee, on 9 September 2011 at 04:12 as depicted in Fig. 8

Fig. 8

Top two panels: spatial distribution of depth-averaged current speeds (m/s), from low (green) to high (pink/white), with velocity vectors, to indicate direction and magnitude, during the peak discharge event from Lee, on 9 September 2011 at 04:12, under vegetative (left panel) and non-vegetative (right panel) conditions. Bottom two panels: Spatial distribution of the elevation difference (cm) between the final and initial bed thickness over the Lee time period, under vegetative (left) and non-vegetative (right) conditions (Δh_b). Coloring indicates removal of elevation (blues) and addition of elevation (reds) over the time period. A black dot is included in the map to indicate the location of station FLT, where Fig. 7 is obtained from. A dotted line is included in all panels for reference to the 3-m isobath

Fig. 9

Spatial distribution of the mud mass difference (Δm_f; top two panels) and sand mass difference (kg/m²) (Δm_c; bottom two panels) between the final and initial mass in the bed layer sum over the time period associated with the Lee event, under vegetative (left) and non-vegetative (right) conditions. Coloring indicates removal of mass (blues and greens) and addition of mass (reds) over the time period. A black dot is included in the map to indicate the location of station FLT, where Fig. 7 is obtained from. A dotted line is included in all panels for reference to the 3-m isobath

Post-Lee

During the post-Lee period, river discharge averaged 1914 ± 669 m³ s⁻¹. There was a wind event that peaked on 20 Oct 2011 with 10.5 m s⁻¹ wind velocity. Significant wave heights increased due to higher wind speeds, but they were smaller within the vegetated patch (veg average 0.25 ± 0.04 m versus no veg average 0.29 ± 0.04 m (Fig. 10). Like the Irene event, vegetation reduced the wave and current bottom stress magnitude (veg average 0.104 ± 0.030 N m⁻² versus no veg average 0.176 ± 0.068 N m⁻²). However, the peak wind velocity was 5 m s⁻¹ lower than the Irene event, resulting in lower overall bottom stresses. Suspended coarse sediment concentrations increased with the pulses of river discharge and enhanced wave activity and were higher without vegetation. The increase in fine suspended sediment aligned well with the wind events in both vegetation scenarios. Both erosional and depositional patterns occurred near the SR mouth with little difference between the vegetation scenarios. Fine sediments were removed from the southeast edge of the vegetation patch; removal was enhanced under vegetative conditions (Fig. 10). The sediment budget indicated a net export of fine sediment that was accentuated under vegetative conditions. However, in both vegetation scenarios, coarse material was trapped (Table 3). Depth-averaged current velocities were higher in the main channel when vegetation was present, while velocities were more widely distributed when vegetation was absent. Significant wave heights during peak winds reached a maximum of 0.48 m with a clear wave height reduction when vegetation was present.

Fig. 10

Top two panels are the spatial distribution of significant wave heights (H_s; m), smaller (blue) to larger (yellow), during peak winds during post-Lee, on 20 October 2011 at 19:00, under vegetative (left) and non-vegetative (right) conditions. On each panel, an arrow is included to indicate wind direction and speed at the specified time. The scale bar for these panels ranges from 0 (purple) to 0.5 m (yellow). Bottom two panels are the spatial distribution of the mud mass difference (Δm_f kg m⁻²) between the final and initial mass sum in the bed layer over the time period associated with the Irene event, under vegetative (left) and non-vegetative (right) conditions. Coloring indicates removal of mass (blues and greens) and addition of mass (reds) over the time period. The scale bar for these panels ranges from − 1 (blue) to 0.6 (red) kg m⁻². A dotted line is included in all panels for reference to the 3-m isobath

Discussion

The impacts of vegetation on sediment distribution in the Chesapeake Bay are widely cited as a critical piece of Bay ecosystem dynamics (Kemp et al. 2005; Gurbisz and Kemp 2014; Russ and Palinkas 2018). In this study, the COAWST modeling system effectively recreated the physical environment (i.e., currents and wave heights) of an extremely dynamic bay-head delta system and its response to storm forcing at a high resolution (Fig. 3 and Fig. 4). In this section, the detailed descriptions of event responses presented above are compared and general tendencies are explored, focusing on the influences of vegetation over the shallow flats, channel flat contrasts, and sequential timing of events.

One of the most important influences of vegetation over the shallow flats is on sediment dynamics in the channel. Waves in this system were too small and short to influence the bottom in the channel, such that channel sediment dynamics were dominated by currents due to the combination of river flow and tides. The highest channel currents were always during ebb tide, when the river and tidal currents were in the same direction. This generated a net export of fine sediment (Table 3) primarily due to the greatest resuspension from the bed layer coinciding with downstream-directed currents. Vegetation over the flats enhanced channel export by focusing tidal and river flows into the channel (Fig. 8). This resulted in either resuspending more fine material or preventing its deposition (Fisk et al. 1954; Wright 1977; Cotton et al. 2006; Bouma et al. 2007; Russ and Palinkas 2018). In contrast, coarse material was deposited because the balance between erosion and deposition was biased towards deposition by the higher settling velocities and critical stresses for erosion of coarse material (Tables 2 and 3).

Table 2 Initialization parameters for the fine (clay-cohesive) and coarse (sand-non-cohesive) sediment for ROMS. The parameters indicated with a superscript plus symbol (⁺) are from Russ and Palinkas (2018) and an asterisk (*) are from Palinkas et al. (2014)

Table 3 Southward time integrated flux, in metric tons, for each sediment class during vegetative and non-vegetative conditions. T1 and T2 indicate which transect the flux is computed across and the event is the selected time period of calculation (see Table 1). T1–T2 is the difference, in metric tons, between T1 and T2 under the specified conditions (bold values)

The two periods with enhanced wind velocities (“Irene” and “Post-Lee”) had high wave heights, which were dampened by vegetation (Figs. 6 and 10). Under high wave conditions, fine material was exported but coarse material was deposited, most likely due to differences in erodibility (Table 3; Gurbisz et al. 2016). However, there were other physical factors that resulted in differences between these events. The Irene wind event was longer-lived with stronger peak wind velocities, which resulted in larger wave heights (Fig. 6), especially in deep water. However, the wave height contrast between deep water and the center of the bed was much greater in Irene, with essentially no waves in the center of the bed at the height of the storm (Fig. 6) as opposed to moderate waves in the center of the bed during the Post-Lee event (Fig. 10). This was most likely due to the extremely low tidal levels at the peak of the Irene event (Fig. 2), which severely limited wave heights in the shallowest regions of the bed. Comparing Figs. 6 and 10, it is apparent that the amount of fine sediment deposition over the bed is inversely related to the wave height in the center of the bed at the peak of the storm. This suggests that resuspension by wave action is a key component of fine sediment removal, such that the absence of waves leads to greater net accumulation. This agrees with previous studies during Irene and other similar events (Ward et al. 1984; Sanford 1994; Sanford 2008; Palinkas et al. 2014). It is also important to point out that river discharge was higher in the Post-Lee event, so more suspended sediment was delivered than during Irene (Table 3). This resulted in a nearly seven-fold increase in deposition in deeper waters near the mouth of the SR during Post-Lee compared to Irene (Fig. 10), even though deposition over the bed was less during Post-Lee.

During large discharge events, the system was completely dominated by riverine dynamics and associated high current velocities (Fig. 8), which far exceeded those during other time periods (e.g., 4 m s⁻¹ versus 0.5 m s⁻¹ under typical conditions). This massive river discharge led to massive sediment delivery to the entire region (Table 3), in agreement with previous estimates of 6 × 10⁶ tons (Hirsch 2012; Cheng et al. 2013; Palinkas et al. 2014). Interestingly, this was the only event with more import than export of both classes of sediment, consistent with sediment deposition into the bed layer across a wide swath of the region (Fig. 8 and Table 3). This process drives geomorphologic development of bay-head delta systems (Dalrymple et al. 1992) and likely also drives the evolution of the Flats region. Significant net deposition in SAV beds during high discharge events was also reported by López and García (2001), Russ and Palinkas (2018), and others.

Geospatial patterns of deposition and erosion were clearly influenced by the presence of vegetation (Fig. 9), which deflected currents around the vegetation and into the main channel and increased channel current speeds (Fig. 8), as well as enhanced deposition of fine material over the flats (Fig. 9). This limited the deposition of coarse material in the channel while enhancing deposition along the channel edges in vegetated regions (Fig. 8 and Fig. 9). Due to extremely high current velocities at the SR mouth, a significant amount of the bed layer was eroded there (Fig. 8) further enhancing channelization. Significant channel erosion also occurred in the channel to the southwest of the flats, where currents were accelerated due to constriction of the available channel cross-section.

We hypothesized that the existence of vegetation over the Flats plays a critical role in modulating sediment transport to the rest of CB by reducing current speeds and waves over the Flats and thus enhancing deposition. This is clearly true for the “before Irene” typical conditions and the “Irene” conditions (Table 3). Both fine and coarse sediments show greater retention or less loss in the presence of vegetation during these periods. However, this pattern is not as clear for the “Lee” and “Post-Lee” periods. During Lee, more sand was retained without vegetation, while during Post-Lee, more fine sediment was exported with vegetation.

There are several factors that help to explain the complexity of the later responses. One of these was already explored above. In all events with vegetation, ebb tide flows and bottom stresses in the channel increased due to the higher drag of the SAV beds (Fig. 8). This enhanced erosion and downstream transport within the channel and provided a pathway for suspended sediment to be shunted around the Flats and into CB (see also Luhar et al. 2008). Depending on the balance between channel and over-Flats transport, this difference might help to explain some of the unexpected responses in the later events. For example, it is likely that higher channel velocities and stresses with vegetation during Lee resulted in greater downstream transport of coarse sediment in the channels, leading to less overall retention of coarse sediment than without vegetation (Table 3). This implies that the channel is the primary conduit for sand transport in this system, consistent with the fact that it requires periodic dredging to maintain navigable depths (Nichols et al. 1990; Orth et al. 2010).

Another important factor is the sequence of previous events that influences depositional history and availability of sediment for resuspension. The Irene and Post-Lee events provide the opportunity to compare similar wind-driven events and investigate the potential role of sediment dynamics proceeding each event. Specifically, there was a significant amount of deposition of fine material during TS Lee, especially in a large zone along the southeast edge of the Flats in the vegetated scenario Lee (Fig. 9). This deposit contained up to 4 kg m⁻² of fine material, in the surface layer and thus available for potential resuspension later. This region also remained relatively protected from resuspension during the elevated river discharge and northerly winds between Lee and Post-Lee. The recent deposit was then easily eroded during the Post-Lee wind event (Fig. 10), accounting for the greater overall loss of fine sediment during Post-Lee with vegetation (Table 3). In comparison, Irene essentially only had the surface layer of fine material as initialized in the sediment bed model available for removal. Coupled with the very low wave heights over the Flats at the peak of the storm, this led to much greater net deposition over the Flats with vegetation during Irene.

Holistically, under vegetative conditions, there was less fine material removed during typical and Irene periods and a significant amount deposited during Lee (Table 3), making more fine material available for transport out of the system during Post-Lee. Under non-vegetative conditions, there was increased removal of fine sediment during typical and Irene periods and less deposition during Lee (Table 3), reducing the amount of sediment available for transport out during Post-Lee. This leads to the conclusion that sediment dynamics in the system are highly dependent on the composition of the sediment bed layer leading up to an event with indications that SAV acts as temporary storage of fine sediments.

A final complicating factor is the balance between applied bottom stress under different scenarios and the critical stress for erosion. This balance is most important under typical low-energy conditions, for example, the Before Irene period (Fig. 5). During this period, a 40% increase in tidal currents without vegetation translated into a doubling of applied bottom stress. More importantly, the doubling of bottom stress led to the critical stress for erosion of fines being exceeded on a regular basis without vegetation, while it was almost never exceeded with vegetation. The regular resuspension and lack of deposition of fines without vegetation allowed fine sediment concentrations to reach 0.05 kg m⁻³, which exceeds the threshold for successful SAV growth (0.02 kg m⁻³; Kemp et al. 2004) (Fig. 5). In comparison, suspended sediment in the vegetative conditions approached 0 kg m⁻³ (Fig. 5), highlighting the ability of SAV to reduce SSC to further facilitate its growth.

In a broader sense, SAV over the Flats act as temporary storage of fine suspended sediment entering CB from the SR. Excess fine suspended sediments impact CB water quality (Langland and Cronin 2003) by reducing water clarity and transporting attached nutrients and pollutants. These factors could lead to reduced SAV population downstream. Seasonal cycles in this sediment storage also alter the timing of nutrient delivery. For example, if a significant discharge event occurs in summer or early fall (e.g., TS Lee), this storage can reduce downstream transport of particulate nutrients at a time of enhanced biological productivity. The subsequent release of this material during the winter, when there is less biological activity, from SAV dieback and/or resuspension from winter storms would have much less impact to CB water quality (Bayley et al. 1978; Gurbisz and Kemp 2014).

Future Work

As shown in Fig. 11, the model predictions matched the rapid increase in suspended material during Lee and even during Irene (Gurbisz et al. 2016) relatively well, but it did not capture the magnitude of the Post-Lee event within the vegetated region. While there are many possibilities for this failure, mostly likely the bed model needs further refinement to make recently deposited particles available for erosion rather than sequestering them into the bed. The model has three bed layers, with deposited fine sediment immediately mixed through the upper 1 cm and rapidly sequestered into the coarse bed layers below. The mud portion in the 1-cm surface layer is not as readily available for resuspension as it likely is in nature.

Fig. 11

Time series for a wind velocities at the NOAA-NOS CBIBS Susquehanna station, b river discharge observations from the USGS sensor at Conowingo Dam, c in situ turbidity (NTU) observations from inside (grey) and outside (black) the plant bed, and d predicted sum of the depth average sand and mud suspended sediments (kg/m³) inside (grey) and outside (black) the plant bed, and N and S are included in panel a to indicate the direction the winds are moving towards

Further sensitivity experiments on some of the initialization parameters would provide insight into other important conditions for sediment trapping on the Flats. For example, the SAV population was established in this simulation as one species (V. americana). However, a wide range of SAV species exist over the region including Z. marina, M. spicatum, H. verticillata, and H. dubia (Gurbisz and Kemp 2014), with each species having its own characteristic plant density, height, thickness, and flexibility. For example, Z. marina has a typical shoot density of 988 shoots m⁻² and can reach up to 2 m in length (Short et al. 2010), which are both much higher than V. americana used in this study. The increase in shoot density and shoot length could increase the amount of sediment deposited during large discharge events by reducing energy more than V. americana. Conversely, a species with less density could result in increased shear stresses propagating further into the vegetation patch, increasing erosion of fine sediments. Although that shoot density did not affect flow speeds in a SAV bed exposed to ambient currents in previous work (Fonseca et al. 2019), those current speeds were lower than the peak predicted velocities presented herein.

SAV coverage on SF varies annually (Gurbisz and Kemp 2014). SAV area was reduced significantly following TS Lee and it has still not recovered completely (Gurbisz et al. 2016), though the remaining bed has been remarkably resistant to further degradation. This reduction in SAV coverage was not simulated in our model. One possibility for capturing the removal of SAV cover within the model would be to include a parameter for critical shear stress for uprooting vegetation (Preen et al. 1995; Cabaço et al. 2008). This would change the distribution of vegetation and thus also sediment dynamics during the simulation. Note this parameter is likely to vary during the seasonal vegetation cycle, such that a lower shear stress is required to uproot vegetation at the beginning of the growing season before plants have established full root systems (Moore et al. 1997; Gallegos and Bergstrom 2005). Thus, a large discharge event occurring at the beginning of the typical growing season from March to October (Patrick and Weller 2015) similar to Hurricane Agnes in late June 1972 might uproot more SAV than an event at the end of the typical growing season. Modeling the removal of SAV during TS Lee was beyond the scope of this study, however.

This study has highlighted the role of SAV in modulating physical processes in the upper CB by, for example, forcing greater flow through unvegetated channels and increasing sediment erosion and transport therein. We have not addressed the longer-term feedbacks between river flow, waves, sediments, and SAV, however. The current speeds required for channel erosion might exceed critical shear stresses for uprooting of vegetation (Preen et al. 1995; Cabaço et al. 2008) and thus distribute the flow and sediment to other regions within the bed, changing the prospects for future SAV growth. SAV is an ecosystem engineer (Jones et al. 1996; Koch 2001), enhancing fine sediment deposition to facilitate its survival. For example, during the “before Irene” event, fine SSC exceeded the threshold for water clarity needed to sustain SAV when vegetation was absent, but SSC was lower than the water clarity threshold when vegetation was present to reduce current speeds. Very large event flows might negatively affect SAV distributions and growing conditions, leading to a reversal of positive feedbacks and collapse of the SAV bed; this may be what happened during Tropical Storm Agnes, when the SF SAV bed disappeared for decades. It would be interesting to use the model developed here coupled to a SAV model to explore these longer-term feedbacks between planks, flow, waves, and sediments.

Conclusion

Submersed aquatic vegetation (SAV) has a significant impact on sediment transport in the upper Chesapeake Bay (CB). Due to the location of vegetated beds between the mouth of the Susquehanna River and the rest of the CB, their capability to modulate sediment transport may have significant impacts on downstream CB water quality. During typical flow conditions, SAV facilitated fine and coarse sediment retention. However, under wave-dominated conditions, SAV damped wave heights which facilitated export of fine sediment. Finally, under current dominant conditions, SAV enhanced channel export by focusing tidal and river flows into the channel.

The main objective of this study was to recreate, at a high resolution, the existing environment during storm events and evaluate the potential impacts of vegetation and event timing on sediment transport. The simulations showed how well COAWST can uniquely handle dynamic vegetation and sediment transport on an estuarine scale, indicating that system dynamics are temporally variable and geospatially diverse. This work emphasizes the capability of SAV to temporarily store suspended fine sediments after large discharge events, with vegetation nearly doubling the storage capacity of unvegetated flats. SAV act as ecosystem engineers by channeling flow and increasing water clarity, supporting their survival. The dependence of wind-driven resuspension on the pre-existing sedimentary conditions shows that event timing is critical to understanding and predicting the sedimentary response of the system. While some further refinements are needed to the bed model, this study establishes this modeling approach as a useful tool for evaluating sediment dynamics in shallow vegetated systems under a range of environmental conditions.