Incorporating flowpaths as an explicit measure of river-floodplain connectivity to improve estimates of floodplain sediment deposition

Variation in floodplain topography can lead to gradual flooding and increase river-floodplain connectivity. We show that incorporating flowpaths as an explicit measure of river-floodplain connectivity can improve estimates of floodplain sediment deposition. We focus on the floodplain of the South River, down-stream of Waynesboro, Virginia, where measurements of mercury accumulation have been used to estimate decadal-scale sedimentation rates. We developed a two-dimensional Hydrologic Engineering Center’s River Analysis System (2D HEC-RAS) hydrodynamic model and used simulated model results with sediment deposition data to create regression models describing sedimentation across the floodplain. All of our statistical models incorporated a flowpath length from the location on the floodplain downstream to the riverbank as an explicit measure of river-floodplain connectivity that improved our estimates of floodplain sediment deposition ( r 2 = 0 . 514 ). We applied our best regression model to our hydrodynamic model results to create a map of floodplain sedimentation rate and discuss differences of three separate sections of floodplain. We found that floodplains with variable topography had wider, bimodal probability distribution functions (PDFs) of sedimentation rate (aggregated spatially) than floodplains without this topographic relief (with narrower log-normal PDFs). Our work highlights how floodplain topography and river-floodplain connectivity affect sedimentation rates and can help inform the development of floodplain sediment budgets

A significant portion of the eroded sediment from a drainage basin is deposited within river channel belts and floodplains (Lambert and Walling, 1987;Walling, 1983;Allison et al., 1998;Costa, 1975;Goodbred Jr and Kuehl, 1998).Sediment transported and deposited in the floodplain during overbank flow can be stored in the floodplain for decades or much longer (Pizzuto, 2014;Pizzuto et al., 2008).Estimating the deposited sediment volume in floodplains is an integral part of calculating a sediment budget (Goodbred Jr and Kuehl, 1998;Belmont et al., 2011;Dunne et al., 1998;Fryirs and Brierley, 2001;Malmon et al., 2002;Nicholas et al., 2006;Park, 2020;Trimble, 1999).The floodplain component of a sediment budget quantifies the inputs, outputs, and storage within the floodplain control volume over a defined period of time (Allmendinger et al., 2007;Wohl, 2021).
There are a number of methods that have been used to quantify one or more aspects of the floodplain component in a watershed or reach-scale sediment budget.Regression models have been developed to explain the spatial variability of sediment and nutrients (Hopkins et al., 2018) and to estimate the deposited sediment volume on a floodplain (Chen et al., 2019;Noe et al., 2022).Seasonal and annual patterns of floodplain sediment storage have been calculated from the measurement of water level and flow discharge in combination with surface suspended sediment maps from remote sensing (Park, 2020;Park and Latrubesse, 2019).Naipal et al. (2016) developed a floodplain sediment budget by simulating spatial patterns and long-term trends of soil erosion and redistribution in floodplains from a sediment mass-balance model.Schenk et al. (2013) expressed floodplain storage as net and gross floodplain trapping factors.Wasson (2003) used a simple approach to estimate sediment storage by measuring the floodplain dimensions and estimating the sediment deposition rate from sediment dating.Moody (2019) used long term measurements of the cross-sectional topography of active floodplains and point bar to develop dynamic re-lations between peak-flood discharge and annual sediment deposition.
Over the years many different methods have been developed to estimate sedimentation rates on floodplains over different timescales.At an event timescale, these methods include post-flood topographic surveying, measuring depth of deposits from probing and digging shallow pits, measuring high-water marks on trees and buildings, aerial image analysis (Brinke et al., 1998;Gomez et al., 1997;Kesel et al., 1974), sediment traps (Middelkoop and Asselman, 1998;Walling and Bradley, 1989), and artificial clay pads markers (Hupp et al., 2015;Shenk and Linker, 2013).At > 10 year timescales, these methods include fallout radionuclides 137 Cs and 210 Pb (Du and Walling, 2012;He and Walling, 1996a,b;Walling and He, 1997a,b;Allison et al., 1998;Goodbred Jr and Kuehl, 1998;Remor et al., 2022;Terry et al., 2002Terry et al., , 2006)), tree-ring (dendrogeomorphic) techniques (Hupp et al., 2015;Mizugaki et al., 2006;Remo et al., 2018;Tichavský et al., 2018), repeat surveying of monumented cross sections (Trimble, 1999;Leopold, 1973), and pollen analyses (Hupp et al., 2015).All of these methods directly or indirectly measure the sedimentation rate at a limited spatial extent (point scale) and many such measurements are required to improve the accuracy of estimates of the spatial distribution of sedimentation rates across the floodplain.
Sediment deposition pattern, rate, and amount on the floodplain is governed by variable hydraulic and geomorphic parameters such as duration and frequency of inundation, flow depths and velocities, sediment load, suspended sediment concentration, flow patterns over the floodplain, vegetation density, and sediment transport mechanism (Middelkoop and van der Perk, 1998;Nicholas andWalling, 1997, 1998;Pizzuto et al., 2008;Nicholas and Mitchell, 2003;Park and Latrubesse, 2019;Maaß and Schüttrumpf, 2019;Zwoliński, 1992).Studies on overbank flows demonstrated strong control of floodplain topographic features on floodplain inundation frequency, flow depth, and velocities (Nicholas and Walling, 1997;Nicholas and McLelland, 2004;Bates et al., 1992;Czuba et al., 2019;Lewin and Hughes, 1980;Lindroth et al., 2020).Furthermore, floodplain topography has been shown to be highly correlated with overbank sedimentation rates (Middelkoop and Asselman, 1998;Nicholas and Walling, 1998).Floodplain topographic surfaces consist of both positive and negative relief from features such as surface depressions, channel margin slackwater zones, bar-shelter backwaters, tie channels, internal drainage channel networks, drainage ditches, and abandoned channels (Lewin and Ashworth, 2014).Proper representation of these floodplain features is a prerequisite in improving the accuracy of estimates of sedimentation pattern, rate, and amount (Middelkoop and van der Perk, 1998;Nicholas and Walling, 1997).
The purpose of this work is to show that incorporating flowpaths as an explicit measure of river-floodplain connectivity can improve estimates of floodplain sediment deposition.We focus on the floodplain of the South River, Virginia, where historical mercury contamination has accumulated in the floodplain.Measurements of mercury accumulation have been used to estimate decadal-scale sedimentation rates across hundreds of points on this floodplain (Pizzuto et al., 2016).We developed a two-dimensional Hydrologic Engineering Center's River Analysis System (2D HEC-RAS) hydrodynamic model for the South River downstream of Waynesboro, Virginia, and used simulated model results to investigate floodplain sedimentation.The results of our work inform how floodplain topography and riverfloodplain connectivity affects sedimentation rates and can help inform the development of floodplain sediment budgets.

Study Area
The South River flows through Waynesboro, Virginia (drainage area of 330 km 2 at Waynesboro), and flows for another 48 km downstream until it joins the North River at Port Republic, Virginia (drainage area of 608 km 2 at Port Republic), to form the South Fork Shenandoah River (Fig. 1).This reach of the river is a fifth-order, gravel-bed, bedrock river with frequent pools and riffles.The channel has a bankfull width and depth of 40 m and 1.5 m, respectively (Rhoades et al., 2009), and a channel slope of 0.0013 (Skalak and Pizzuto, 2010).Frequent bedrock exposures along its bed, banks, and forested riparian zone limit channel migration rates to only a few centimeters per year (Rhoades et al., 2009).The South River floodplain consists of a mixture of pasture, forest, and developed areas and is confined in places by artificial levees, fluvial terraces, quaternary alluvial fans, bedrock valley walls, and a railroad bed.Floodplain sediment accumulation rates along the South River are low, having previously been estimated at rates of only a few centimeters per 100 years (Pizzuto et al., 2016).More recent regional scale analysis (including data collected from the South River; Pizzuto et al., 2023) has revised previous estimates and has reported that modern sediments (1950-present) accumulated at a median rate of 0.25 cm/year, which is one order of magnitude higher than reported earlier by Pizzuto et al. (2016).
Mercury contamination of the South River was first reported in 1976 (Carter, 1977) but initially introduced from a manufacturing facility in Waynesboro, during 1929Waynesboro, during -1950. .During high overbank flows, mercurycontaminated sediments were transported and deposited on the floodplain (Pizzuto, 2012;Rhoades et al., 2009).Mercury concentrations on suspended sediment reached a peak at about 1, 000 mg/kg during 1940-1960, then decreased rapidly to 10 mg/kg by 1970 (Pizzuto, 2012(Pizzuto, , 2014) ) and were still observed at 10 mg/kg around 2006-2009 (Flanders et al., 2010).Mercurycontaminated sediments that cycle between the river and floodplain are an important source of mercury in the South River (Flanders et al., 2010;Rhoades et al., 2009).Pizzuto (2014) found that suspended sediment in transport enters floodplain storage after a relatively short distance (ca. 10 km) and that mercury deposition on near-channel floodplains declines exponentially downstream.Downstream of Waynesboro, there were 13 colonial mill dams, many established in the 1800s.All of these dams except one were breached by 1957 and the last one was breached in 1976 (Pizzuto and O'Neal, 2009).Overbank sedimentation rates could have been locally higher upstream of mill dams within the mill ponds (Walter and Merritts, 2008).The approximate upstream extent of the backwater influence due to mill dams on the South River was estimated to be within 3 km (Pizzuto and O'Neal, 2009).Studies by Pizzuto et al. (2014Pizzuto et al. ( , 2016) ) did not find any strong correlations between mercury deposition on the South River floodplain and the presence of historical mill dams.
There are two USGS streamflow gaging stations within our model domain: USGS 01627500 South River at Harriston (daily data 1925-present) and USGS 01626850 South River at Dooms (daily data 1975-present) (Fig. 1b).Over the simulated 48 km reach of the South River, the drainage area nearly doubles, with increasing discharge along the reach.However, we only investigate static flows of a specific discharge uniformly across the entire reach, which captures how a specific location would be inundated by that specific flow, but does not capture a realistic picture of the entirety of the South River during a single flood event.When selecting flow discharges to perform hydraulic calculations, we chose a range of flows based on the Harriston gage because of its central location in the reach, and for reference, we report recurrence intervals of these flows relative to the Harriston gage because it has the longest data record.We focus on describing our results within an area of interest (AOI, Fig. 1c), which is an 8.8km reach of the South River just downstream of Waynesboro, Virginia, and includes three distinct floodplains A, B, and C (Fig. 1c).The AOI was based on the simulated inundation extent of the largest flow simulated (peak of record at Harriston, 824 m 3 /s).We selected this reach because it is close to Waynesboro, has high concentrations of mercury in its floodplain (Pizzuto et al., 2014), and there are many field measurements of mercury that were used by Pizzuto et al. (2016) to estimate sedimentation rates.

Floodplain sediment and mercury data
Within our AOI, mercury accumulation was measured at 107 sampling locations (orange circles, Fig. 1c) in the South River channel and floodplain by multiple teams and these data were collectively described by Pizzuto et al. (2016).In 2008, five samples were collected adjacent to the channel and five from cut banks by Pizzuto et al. (2016).The total depth sampled ranged from 1.2 to 2.4 m.At each sample location, the topmost 15 cm was analyzed, deeper deposits were analyzed at intervals of 30 cm.Another 80 floodplain cores were collected by URS Corp. in 2008 from the surface down to 75cm depth (URS Corp, 2012).For mercury content, the uppermost 15cm and the remaining 60cm cores were analyzed separately (Pizzuto et al., 2016).The Environmental Protection Agency obtained the remaining 17 samples, all were from the surface down to 75 cm deep and sampled at 15 cm intervals (Pizzuto et al., 2016;Bruzzesi et al., 2007).These collected samples were analyzed at commercial laboratories to estimate mercury concentration along with the sand, silt, clay, and organic carbon.The total amount of mercury deposited (kg/m 2 ) in each location was used by Pizzuto et al. (2016) to estimate a sediment deposition rate (cm/year).Mercury deposition was estimated from the collected samples by applying the equation (Pizzuto et al., 2016): (1) where, I Hg represents mercury deposition (kg/m 2 ) per unit surface area of the floodplain between 1930 (when mercury was first introduced into the South River) and 2007 when the samples were collected, ρ b is the floodplain deposit bulk density of 1200 kg sediment/m 3 (Pizzuto et al., 2016), C soilHgi is the mercury concentration (units of kg Hg/kg sediment) measured in each soil core, T i is the sampling interval (m) in each core, and n is the number of sampling intervals in each core.
Mercury deposition I Hg (kg/m 2 ; 1930-2007) was then converted into a sedimentation rate , T t (m/year; afterwards converted to cm/year) by applying the equation (Pizzuto et al., 2016): (2) where T is the duration of Hg accumulation (1930-2007; 77 years) and C Hgs is the concentration of mercury on suspended sediment, which was initially estimated at 0.000239 kg Hg/kg sediment by Pizzuto et al. (2016).However, after comparing mercury deposition I Hg to sedimentation rates measured by Pizzuto et al. (2023) where both measurements overlapped (at four points), we back-calculated an average C Hgs of 0.00087 kg Hg/kg sediment that we used in our analysis.This parameter, C Hgs , is the key parameter in converting measured mercury to an estimate of deposited sediment because it assumes the depositing sediment mass had this average concentration of mercury.The reconstruction of mercury concentrations from collected data and the underlying assumptions are described in detail in previous publications Pizzuto (2012Pizzuto ( , 2014)); Skalak andPizzuto (2010, 2014); Pizzuto et al. (2023).It is important to note that we are considering median values of sedimentation rate as our "true" measured values, whereas Pizzuto et al. (2016) acknowledge a range and uncertainty in these estimates.Additionally, the full process of data collection, laboratory analysis, and conversion of mercury accumulation into sedimentation rate were described by Pizzuto et al. (2016).From here onward, we only focus on the resulting sedimentation rate data.

Model construction
A two-dimensional unsteady hydrodynamic model was constructed for the South River and its floodplain using 2-D HEC-RAS version 5.0.6 developed by the U.S. Army Corps of Engineers (Brunner, 2016).The model domain included a 48 km reach of the South River and its floodplain from Waynesboro to Port Republic, Virginia (at the confluence of South and North Rivers; Fig. 1b).The model terrain surface was constructed from 0.76 m lidar data (VGIN, 2017).The elevation of the channel bed was approximated by lowering the water surface at the time of lidar acquisition uniformly by about 1 m, equal to the difference between the lidar water-surface elevation and the zero-discharge stage at the USGS gage (01627500 South River at Harriston, Virginia).This approximation has been shown to provide reasonable results for floodplain-inundating flows when compared to surveyed bathymetry (see supporting information of Czuba et al., 2019).The floodplain topography of the terrain surface was additionally smoothed using a 5 m smoothing window so that smoother streamlines could be generated from the resulting depth-averaged velocity data.The computational mesh was generated with a 23 m nominal orthogonal cell spacing and refined to 7.6 m along break lines positioned along the river channel centerline, banks, major floodplain channels, and roadways.By generating a coarser grid across flat topography and a finer grid in the channel and along structures, we were able to balance model resolution and computational efficiency with 188 thousand cells overall and an average cell area of 371 m 2 .The upstream flux boundary condition was specified as a series of 24 flow discharge values that spanned the range of observed flows at the Dooms and Harriston USGS gages.Each flow was held constant for a few days of model simulation to achieve steady state before incrementing to the next higher flow.The downstream boundary condition was set as normal depth with a friction slope of 0.01.Roughness values were initially assigned based on land cover (Wickham et al., 2021) using the default Manning's n values from HEC-RAS (Brunner, 2016) and were later adjusted during model calibration.

Model calibration and validation
The 2D hydrodynamic HEC-RAS model was calibrated to the USGS streamflow gaging station 01627500 at Har-riston (Fig. 2a-b) and validated against the USGS gaging station 01626850 at Dooms (Fig. 2c-d).We calibrated to the Harriston gage because of its longer-term data availability and the floodplain land cover in its vicinity better represented the conditions along the South River.We then validated to the Dooms gage because it was in our AOI.Land cover-based (2016 National Land Cover Database; Wickham et al., 2021) Manning's roughness values were used as calibration parameters (Brunner, 2016).The roughness values that we changed from the default values (Brunner, 2016) included an open water roughness value that was calibrated to 0.0375 (12% of inundated area), low-intensity developed area to 0.06 (5% of inundated area), high-and medium-intensity developed area to 0.05 (6% and 4% of inundated area, respectively), barren to 0.035 (0.13% of inundated area), deciduous, evergreen, and mixed forest to 0.0325 (14%, 9%, and 1.2% of inundated area, respectively), developed open space to 0.03 (8.7% of inundated area), pasture/hay and row-crop agriculture to 0.015 (34.6% and 5.3% of inundated area, respectively), and woody wetlands to 0.08 (0.15% of inundated area).
Model calibration at Harriston (USGS 01627500) involved comparing the model simulated water surface elevation (WSEL) with the USGS rating curve (Fig. 2a) and also comparing the model simulated depth-averaged velocity to the USGS measurements of channel velocity (Fig. 2b).The reported USGS velocity measurements are cross-sectional average velocities measured at the gage.To appropriately compare to these values, we extracted depth-average velocities along the wetted channel-floodplain cross section at the gage and averaged these values at each flow.Both gage locations were just downstream of a bridge (from where high flow measurements were likely taken) and the bridge abutments here confined the flow primarily to the channel for most of the flow conditions considered.The root mean square error (RMSE) of the simulated water surface elevation (WSEL) to the USGS rating curve at Harriston (USGS 01627500) was 0.12 m and mean absolute error (MAE) was 0.10 m (n=18).The RMSE and MAE of the USGS measurements to the USGS rating curve at Harriston was 0.25 m and 0.21 m, respectively (n=22; ≥ 28 m 3 /s, RI: 1-year) (Fig. 2a).Model error compared to the USGS rating curve was close to the variability between the USGS measurements and the rating curve (Czuba et al., 2019;Sumaiya et al., 2021).The RMSE and MAE of the model simulated depth-averaged velocity to the USGS measurements of channel velocity was 0.25 m/s and 0.21 m/s, respectively (n=21; ≥ 28 m 3 /s, RI: 1-year) (Fig. 2b).
Model validation at Dooms (USGS 01626850) involved comparing the model simulated water surface elevation with the USGS rating curve (Fig. 2c) and also comparing the model simulated depth-averaged velocity to the USGS measurements of channel velocity (Fig. 2d).The RMSE of the simulated WSEL to the USGS rating curve (Fig. 2c) was 0.13 m and MAE was 0.11 m (n=15).The RMSE of the USGS measurements to the USGS rating curve at Dooms was 0.10 m and 0.07 m, respectively (n=25) (Fig. 2d).The RMSE and MAE for simulated depth-average velocities with USGS measure-ments was 0.20 m/s and 0.13 m/s, respectively (n=24).These velocity errors were comparable to the hydrodynamic model validation errors (RMSE of 0.27 m/s, MAE of 0.21m/s) reported by Czuba et al. (2019) and Sumaiya et al. (2021).Our reported calibration and validation statistics only include the flows that inundate the floodplain (≥ 28 m 3 /s, RI: 1-year) because our analysis is focused on only floodplain-inundating flows.

Model outputs
Simulated water depth and depth-averaged velocity (as both x and y components and velocity magnitude) data were the fundamental results used from the 2D HEC-RAS model.The HEC-RAS model computed depth at cell centers and depth-averaged velocity at cell face points and stored the results in a Hierarchical Data Format (HDF) file (Brunner, 2016;Worley et al., 2022).We loaded the HDF file in MATLAB and extracted only the data within the AOI and created a 3 × 3 m uniformly spaced mesh grid and interpolated the depth and depthaveraged velocity data to this common grid.We calculated bed shear stress at the common grid points from the resistance equation (Garcia, 2008): where τ b is the estimated bed shear stress at each common grid cell (P a), ρ is the water density (kg/m 3 ), ν is the depth-averaged velocity magnitude (m/s), and C f is the roughness coefficient calculated from the Manning-Strickler form of the velocity profile (Garcia, 2008): where H is the spatially-averaged flow depth at each grid cell (m) and k s is the effective roughness height (m) proportional to the mean sediment diameter in the absence of bedforms such as dunes (Brownlie, 1981).The Manning-Strickler form of Manning's roughness coefficient (n) is (Brownlie, 1983): where g is the gravitational acceleration (m/s 2 ).Combining Equations 3, 4, and 5, the simplified form of these equations for calculating bed shear stress is: Calibrated roughness values based on land cover were used in this equation for calculating bed shear stress at the common grid points (with both depth and depth-averaged velocity data).We generated streamlines in MATLAB from our gridded x and y depthaveraged velocity component data at each flow to aid in visualizing the velocity field using the MATLAB streamline and streamslice function.

Flowpath length
We defined a flowpath as the path followed by a water particle from a point on the floodplain downstream or upstream to the river channel.A downstream flowpath was generated from each sampling point (Fig. 1c) at each of the 18 different flows, when inundated, downstream to the river channel using the gridded x and y depth-averaged velocity components.We also calculated the upstream flowpath, but mention why this was not used in section 3.3.4.Out of the 107 sampling points, only 76 points had complete flowpaths at least at the highest flow.The points at which flowpaths were not generated were either collected in the river bank or outside of the inundation boundary of the highest flow simulated and were not included in the model development.There were multiple flowpaths at each sampling point, one for every flow that inundated the point, that connected each point downstream to the river channel.We calculated the length of all the individual flowpaths starting from each sampling point and then probabilistically integrated the lengths for each point based on flow recurrence as follows, where, PFIFL is the Probabilistic Flow Integrated Flow-path Length, L i is the flowpath length at the ith flow, P i is the probability of occurrence of the ith flow, and n is the number of flowpaths (up to 18 because of the 18 simulated flows that inundate the floodplain).Essentially, Equation 7weights each flowpath length by the probability of that flow occurring when averaging to a single probabilistic flow integrated flowpath length for each sampling point.Values (here flowpath length) calculated at lower floodplain-inundating flows are weighted more than those at higher floodplaininundating flows when calculating an average length because they have a higher probability of occurrence.This method is similar to the Probabilistic Flow Integrated Grain Size (PFIGS) method described in Sumaiya et al. (2021), which probabilistically integrated a grain size by flow recurrence, and is similar to calculating an effective discharge (Wolman and Miller, 1960), which probabilistically integrates a streamflow by flow recurrence.

Exceedance probability of flow
We calculated the exceedance probability (EP ) of the flow that first inundates each sampling point using the annual peak flow series (USGS 01627500 South River at Harriston, Virginia, 1925Virginia, -2021)).We also calculated a corresponding recurrence interval (RI) as RI = 1 EP .

Suspended sediment concentration
Suspended sediment concentration (SSC) in the river was estimated for the 18 simulated flows using the sediment rating curve established by Gray (2018) for the South River (USGS stations 01627500 and 01626000) as: where SSC is the suspended sediment concentration (mg/l) and Q is the flow discharge in f t 3 /s.We determined a single integrated value of SSC for each mercury sampling point by probabilistically integrating the SSC values at and above which the sampling point was first inundated.This integration followed the same approach as we described for integrating flowpath length as described in section 3.3.1.(Equation 7), but replacing L i with SSC i (suspended sediment concentration at ith flow).It is important to note that this suspended sediment concentration is for the river channel without any modification for likely settling that would occur with distance across the floodplain.

Other parameters considered
We compared many other hydraulic and geomorphic parameters to the measured sedimentation rate data from the South River floodplain.Among all the parameters considered, downstream flowpath length, exceedance probability of flow, and suspended sediment concentration had the highest correlations, which we describe later.Other parameters that we compared included upstream flowpath length (i.e., the distance from each sampling point upstream to the river channel; r 2 = 0.02, p-value = 0.19); shortest distance from the sample point to the riverbank (r 2 = 0.26, p-value = 3.2e−6); shear stress at each specific flow (highest: r 2 = 0.17, p-value = 1.7e−4; flow: 368 m 3 /s, RI = 8.9years); shear stress and (equivalently in a correlation) shear velocity at the flow first inundating the sampling point (r 2 = 1.6e−3, p-value = 0.69), probabilistically integrating shear stress by flow recurrence (r 2 = 0.20, p-value = 3.9e−5), Probabilistic Flow Integrated Grain Size (PFIGS, see Sumaiya et al., 2021) in suspension (r 2 = 0.11, p-value = 2.7e−3) and as bedload (r 2 = 0.14, p-value = 8.3e−4).Additionally, the measured sedimentation rate data were not significantly correlated with the spatial gradient of these hydraulic and geomorphic parameters (such as change in shear stress per cell distance) and the grain size distribution of the floodplain at the sampling point.

Statistical model development
We developed three multiple linear regression models to estimate the sedimentation rate along the South River floodplain.The probabilistically integrated downstream flowpath length (section 3.3.1),exceedance probability of the first inundating flow (section 3.3.2),and probabilistically integrated suspended sediment concentration (section 3.3.3)were used as predictor variables.We considered multiple combinations of parameters, but we report out on the best three different multiple linear regression models: Model I, with the only predictor variable of downstream flowpath length; Model II, with predictor variables of downstream flowpath length and exceedance probability of the first inundating flow; and Model III, with predictor variables of downstream flowpath length and suspended sediment concentration.To assess the multiple linear regression models, we computed the p-value (of the model and coefficients), RMSE, coefficient of determination (r 2 ), and adjusted r 2 of the measured and estimated sedimentation rates (76 samples).

Spatial distribution of floodplain sedimentation rate
By applying each of the three multiple linear regression models to the 2D hydrodynamic model results, we created three maps of sedimentation rate within the AOI.From each of these maps we computed the total amount of sediment deposition in the AOI.Model II and III showed very similar results, and therefore we only show the result maps of Model I and Model III.
From the spatial maps of sedimentation rate, we also reported the empirical probability distribution function (PDF) and cumulative probability distribution function (CDF) of sedimentation rates spatially within each of three different floodplains (A, B, and C; Fig. 1c).We expected to learn how floodplain topography affects the spatial pattern and characteristics of floodplain sedimentation.We determined a best-fit distribution to the empirical sedimentation rate distributions in each of the three floodplains, primarily for visualization purposes.The fitting process assessed the goodness-of-fit of the distribution by computing a log likelihood value and standard error (SE).We selected the best fit based on the log likelihood value.The distributions that were best fits included Weibull, Nakagami, Beta, and lognormal.

Hydrodynamic model simulation results
Simulated water depth (Fig. 3a-e) and streamlines (Fig. 3f-j) at five representative flow conditions are highlighted here.At the 1.3-year RI flow (83 m 3 /s), a few floodplain channels in floodplains A and B (Fig. 3c) become inundated and connected (Fig. 3a).In floodplain C, inundation occurred mostly from backwater flow.There is a roadway at the upstream end of floodplain C (Fig. 3b) which limits the amount of water conveyed across the floodplain from upstream.
With increasing flow, inundation depths increase in the floodplain with maximum depths in floodplain channels.At a 1.7-year RI flow (116 m 3 /s) more floodplain channels become inundated.The streamlines are sparsely distributed and follow the underlying topography at this flow.A larger portion of the floodplain becomes inundated at the 3.5-year RI flow (280 m 3 /s) and streamlines are beginning to organize into a collective floodplain flow direction.Floodplain C is still largely affected by backwater flow while floodplain A more directly conveys floodwaters (Fig. 3h).Floodplain B has low lying regions similar to floodplain A and higher elevations similar to floodplain C. At the next higher flow (11.5-yearRI, 477 m 3 /s), most of all three floodplains become inundated except a small high point in floodplain B (Fig. 3d).Additionally, streamlines are organized into a collective floodplain flow direction across all three floodplains (Fig. 3i).At the peak of record (824 m 3 /s), nearly the entire South River floodplain was inundated with streamlines following the general floodplain flow direction (Fig. 3e, 3j).
Sediment deposition at any location on the floodplain was affected by how frequently these locations became inundated (Fig. 4a) and the connectivity of that floodplain location with the river (Fig. 4b).Generally, locations that were inundated more frequently had higher sedimentation rates (Fig. 4a).Locations proximal to the main channel and floodplain channels with shorter downstream flowpath lengths and that were inundated at 1 to 2-year RI flows had the highest sedimentation rates.Most of the floodplain with lower sedimentation rates were inundated at flows higher than the 2-year RI flow and had longer downstream flowpath lengths.

Multiple linear regression models to estimate sedimentation rate
Our best three multiple linear regression models (Table 1) included: Model I, with the only predictor vari-able of downstream flowpath length (m); Model II, with predictor variables of downstream flowpath length (m) and exceedance probability of the first inundating flow (1/year); and Model III, with predictor variables of downstream flowpath length (m) and suspended sediment concentration (mg/l).The flowpath length coefficient was in a negative exponential form in all three models (Table 1) indicating that sedimentation rate decreased exponentially as the flowpath length increased.
Model I, with only flowpath length as the predictor variable, was the simplest of the three multiple linear regression models that we reported.It had a substantially higher r 2 value of 0.45 (Table 1, Fig. 5a) compared to the other single parameters that were considered, all with r 2 values ≤ 0.26 (see section 3.3.4).Both Model II and III were improvements over Model I, with r 2 values of 0.51 and 0.514 by incorporating either the exceedance probability (EP ) of the first inundating flow or the probabilistic flow integrated suspended sediment concentration (SSC), respectively (Table 1; Fig. 5b-c).The coefficient p-values were also significant for Models II and III, which indicated that the included parameters were significant (Table 1).Model II and III were marginally different, but Model III was the best performing overall with the highest r 2 value and lowest pvalue (Table 1).The coefficient in Models II and III both decreased by nearly the same amount from Model I, First flow that inundates the floodplain and corresponding flow recurrence interval (RI).Dark blue is the start of floodplain inundation (vertical dash line at Fig. 2; 28 m 3 /s, RI: 1-year).Each color gradation represents one of the 18 simulated flows that inundate the floodplain.Sizes of the orange circles represent the sedimentation rate with larger circles indicating higher rates (Pizzuto et al., 2016(Pizzuto et al., , 2023)).Gray lines denote the channel-floodplain boundary following the high points along the channel banks.(b) Circles represent the sampling points with generated flowpaths (dashed gray lines) at the highest flow (824 m 3 /s; peak of 96-year record).Colors of the circles correspond to the first flow that inundates this location, same as in (a).
suggesting that the exceedance probability of the first inundating flow and the probabilistic flow integrated suspended sediment concentration had similar quantitative effects.This is not surprising because both incorporate flood frequency explicitly or implicitly in the calculation, as described in sections 3.3.1-3.3.3.

Spatial pattern of sedimentation on the floodplain
To visualize how the multiple linear regression models map to spatial patterns of sedimentation rate, we applied the multiple linear regression model equations to the 2D HEC-RAS results within the AOI (Fig. 6).Model I estimated that the highest deposition would occur closest to the channel and then decrease exponentially upstream (Fig. 6a) because the model was only based on the flowpath length from the sampling point downstream to the channel (or equivalently, the flowpath length from the river channel upstream to the sampling point).Model II and III additionally included inundation frequency and suspended sediment concentration, respectively, and show high deposition in frequently inundated floodplain channels (Fig. 6b).We show the results of Model III (Fig. 6b) because the spatial pattern shown by Model II was very similar to Model III and Model III performed slightly better than Model II (Table 1).The combination of using the 0.76 m lidar data (that was smoothed) with a 2D HEC-RAS model and applying Model III, we were able to describe the fine scale differences in deposition, primarily in and between floodplain channels (Fig. 6b).Even though sedimentation was estimated to be highest in the frequently inundated floodplain channels, these channels still persist.Understanding the temporal dynamics of floodplain sedimentation is beyond the scope of the present work.However, over the 77-year duration of mercury accumulation, a sedimentation rate of 0.15 cm/year (highest in Fig. 6b), would only result in an accumulation of 11.6 cm of sediment.This amount, with 2/3 of that amount occurring outside of the floodplain channels (green colors with 0.10cm/year sedimentation rate; Fig. 6b), would not be enough to fully fill these floodplain channels over this time period.The upstream-most section of our AOI is in Waynesboro, which is highly developed, with parking lots, many buildings, a wastewater treatment plant, and other modifications.Furthermore, there is an abandoned meander channel at the first significant strong bend, where the floodplain shows a large "bulge" off to the east.These locations are outside of the extent of the measurements we used and without further analysis, we cannot be certain of the accuracy of these estimates.Finally, there is an area of the AOI near North Park that has already been remediated (Church, 2019) with a long section of the bank redesigned and hardened, however, our lidar data predates this remediation effort.
From these spatial results (Fig. 6), we estimated the average sedimentation rate in the 1.63 km 2 area of the AOI.The spatial average of sedimentation rate between 1930 and 2007 throughout the AOI was 0.08, 0.08, and 0.09 cm/year from Model I, II, and III, respectively.The map of sedimentation rate (Fig. 6b) shows that most of the deposited sediment is concentrated near the river channel and in low lying areas of the floodplain that are frequently inundated.From this map (Fig. 6b), we have quantified the percentage of the total amount of sediment that has deposited if we first start by targeting the areas with the highest amounts of sedimentation and then expanding to areas with lower amounts (Fig. 7).We estimated that 56%, 82%, and 95% of the sediment deposited in only 25%, 50%, and 75% of the floodplain area, respectively (Fig. 7).Conversely, 9%, 21%, and 41% of the floodplain area contained 25%, 50%, and 75% of the deposited sediment, respectively.
From the sedimentation rate maps (Fig. 6), we analyzed the spatial distributions of sedimentation rate within three physically distinct floodplains.Specifically, we show the empirical probability distribution function (PDF) and cumulative distribution function (CDF) for each of three floodplains, A, B, and C (Fig. 1c) and for results from Models I and III (Fig. 8).We also Table 1 Three multiple linear regression model equations, coefficient values, p-values, RMSE, and r 2 for estimating the sedimentation rate on the floodplain of the South River, Virginia, in the vicinity of the USGS gage at Dooms.Variable definitions: MLR: multiple linear regression; SR, sedimentation rate (cm/year); x, probabilistic flow integrated flowpath length (m); EP , exceedance probability of the first inundating flow (1/year); and SSC, probabilistic flow integrated suspended sediment concentration (mg/l).fitted statistical distributions to the empirical data for visualization purposes.The PDFs of Model I (downstream flowpath length only) in floodplains A and B are similar to a Gaussian distribution (Weibull and Beta as the best fits; Fig. 8ab).However, the PDFs of Model III for these same floodplains are bimodal with more skew.Here the best fit is the Nakagami distribution (a special form of the Gamma distribution) for both floodplains although these are not very good fits because of the bimodal nature of the data.We applied the statistical fits to help guide the eye rather than definitively state that the data follow a specific distribution.The first mode of these bimodal distributions is narrow with rates less than 0.03 cm/year (Fig. 8ab) and represents the floodplain areas away from the river channel that are less frequently inundated (dark blue areas in Fig. 6b).The second mode is wide with sedimentation rates between 0.03 and 0.1 cm/year with the peak of this mode around 0.07 cm/year (Fig. 8a-b).This second mode represents deposition near the river channel and in low-lying areas of the floodplain that are frequently inundated.The presence of topographic relief in the floodplain seems necessary to strongly realize this second mode.For instance, the PDFs of Models I and III in floodplain C both show heavy tailed right skewed empirical distributions that are best fit with a log-normal distribution (Fig. 8c).Because floodplain C is generally flatter with less distinct floodplain channels (Fig. 1c) and becomes inundated almost uniformly (Fig. 4) compared to floodplains A and B, the high deposition rates that characterize the second mode do not occur.

Discussion
We constructed a 2D hydrodynamic model of the South River, Virginia, and combined simulation results with the sedimentation rate data from Pizzuto et al. (2016) into multiple linear regression models describing sedimentation across the floodplain.We developed three multiple linear regression models (Table 1) using variables with the highest predictive ability that included the probabilistically integrated downstream flowpath length (section 3.3.1),exceedance probability of the first inundating flow (section 3.3.2),and the probabilistically integrated suspended sediment concentration (section 3.3.3).The resulting form of the regression equations and coefficients with flowpath distance raised to a negative exponential (Table 1) generally indicates that the sedimentation rate decreased exponentially away from the river channel.This finding was consistent with previous literature describing exponential decay of floodplain deposition rates progressing away from the river channel by turbulent diffusion (Pizzuto, 1987;Allen, 1985;James, 1985), advection (Lauer and Parker, 2008;Pizzuto et al., 2008), and from field observations of finegrained sediment deposition on relatively flat floodplain topography (Middelkoop and Asselman, 1998;Walling and He, 1998;Allison et al., 1998).Additionally, the positive exponent values on exceedance probability and suspended sediment concentration (Table 1) indicate that the sedimentation rate was higher in areas that are inundated more frequently (Liu et al., 2019;Saint-Laurent et al., 2010).
Regression Model I used downstream flowpath length (distance along a streamline starting at a point in the floodplain, proceeding in the downstream direction, and ending at the river) as the only predictor variable (Table 1, Fig. 5a).Downstream flowpath length was the most predictive distance variable (r 2 = 0.45), more so than the upstream flowpath length (distance along a streamline starting at a point in the floodplain, proceeding in the upstream direction, and ending at the river; r 2 = 0.02), maximum/minimum of both distances, and the shortest distance to the riverbank (r 2 = 0.26).This means that the distance over the floodplain upstream of the riverbank (which corresponds to downstream flowpath length) is more predictive of sediment deposition than the distance over the floodplain downstream of the riverbank (which corresponds to upstream flowpath length) for the South River.This suggests that sediment along our study reach of the South River may have been deposited primarily via backwater from the river channel (water pushing upstream) rather than from flow spilling out of the river onto the floodplain (water moving downstream across the floodplain).However, in other floodplains, upstream flowpath length (describing connectivity) may be a more important factor due to other possibly dominant floodplain sedimentation processes.
While downstream flowpath length was the variable with the single greatest predictive ability, we know that other hydro-morphodynamic parameters should also influence sedimentation and further improve the regression model.For instance, sediment deposition on floodplains increases with flood magnitude and duration (Middelkoop and Asselman, 1998;Lambert and Walling, 1987;Heitmuller et al., 2017;Liu et al., 2019;Toonen et al., 2015).After additionally including flood frequency (via the exceedance probability of the first flow that inundates each point on the floodplain; Model II, Table 1, Fig. 5b), we improved our regression model (r 2 = 0.51).Furthermore, during these floods the sediment suspended in the river is carried across the floodplain and settles out of suspension (Sumaiya et al., 2021).Quantifying the portion of the suspended sediment load in a river that is conveyed overbank and de-posited on floodplains has been used to estimate floodplain sedimentation rates (Lambert and Walling, 1987;Nicholas et al., 2006;Narinesingh et al., 1999).Therefore, by further incorporating suspended sediment concentration (with flood frequency implicit in the calculation, see section 3.3.3)as a predictor variable, we found our best-performing regression model (Model III, r 2 = 0.514; Table 1, Fig. 5c).In all of these models, we incorporated hydrologic flowpaths as an explicit measure of river-floodplain connectivity that improved our estimates of floodplain sediment deposition compared to the other parameters considered in section 3.3.4.
We created a map with our best estimate for sedimentation rates (Fig. 6b) along our study reach by applying our best multiple linear regression model (Model III, Table 1, Fig. 5c) to our 2D hydrodynamic model results (Fig. 4).Over our study area (AOI) of the South River floodplain in Virginia, we estimated a spatial average sedimentation rate of 0.08 cm/year, with maximum values of 0.15 cm/year.In analyzing the sedimentation rates at sample points directly, Pizzuto et al. (2016) reported a spatial average sedimentation rate of 0.025 cm/year with median sedimentation rates of 0.038 cm/year, 0.0137 cm/year, 0.004 cm/year, and 0.001 cm/year for areas in the <0.3-year floodplain, 0.3to 2-year floodplain, 2-to 5-year floodplain, and 5-to 62-year floodplain, respectively.More recently, Pizzuto et al. ( 2023) compiled data from rivers in the Mid-Atlantic (including from the South River) and reported that modern sediments (1950-present) accumulated at a median rate of 0.25 cm/year.Our model estimated floodplain sedimentation rates similar to the findings of Pizzuto et al. (2023) because we effectively calibrated the Pizzuto et al. (2016) rates to the Pizzuto et al. ( 2023) rates (see section 3.1).However, our model estimates were underestimating the largest sedimentation rates (see the cloud of points with the largest rates relative to the 1:1 line in Fig. 5), which is why our estimates were low relative to Pizzuto et al. (2023).
Our results show that floodplain topography is a strong controlling factor in establishing the spatial distribution of sedimentation rates (Fig. 1c, 4, 6b).Low-lying areas along the South River, such as floodplain channels, are inundated frequently, affect flow hydraulics, and experience high sediment deposition rates (Büttner et al., 2006).However, some welldeveloped floodplain channels have been shown to be likely net erosional (Sumaiya et al., 2021).Within our study area, floodplains A and B have well connected floodplain channels which were inundated more frequently with greater inundation depths than floodplain C (Fig. 4).Floodplain C is relatively flat with less topographic variation than the other floodplains and was affected more strongly by backwater flow.As a result, floodplains A and B have wider PDFs of sedimentation rate (best fit by a Nakagami distribution, special form of the Gamma distribution) that are bimodal with the one mode occurring at values less than 0.03cm/year and another mode occurring between 0.03 and 0.1 cm/year with a peak of this mode around 0.07 cm/year (Fig. 8ab).This wider second mode represents deposition near the river channel and in low-lying areas of the flood- plain that are frequently inundated.Floodplain C has a log-normal PDF of sedimentation rate with most values less than 0.03 cm/year (Fig. 8c).The absence of major topographic relief in floodplain C seems to preclude a wider distribution of sedimentation rates compared with floodplains A and B.
In our approach, we generated flowpaths based on depth-averaged velocity vectors simulated by a 2D hydraulic model that incorporated the lidar terrain and the hydraulics of the water moving over that terrain at different flow rates or water levels.We contend that incorporating flowpaths generated in this way can be a better approach to quantifying channel-floodplain connectively, than simply water level or inundation (e.g., Park, 2020;Park and Latrubesse, 2019) for topographically variable floodplains.This is because flowpaths better represent the pathways water and sediment travel on the floodplain.For the South River in Virginia, specifically, incorporating flowpaths has improved estimates of floodplain sediment deposition.Furthermore, for the South River, simply using depth or exceedance probability of first inundation alone were not as predictive as flowpath length in predicting sedimentation.This result may not be universal, as there are other factors and processes affecting sedimentation in other floodplains that may be different from our study area.However, we have laid out an approach to explicitly incorporate connectivity into sedimentation estimates that should be applicable on similar floodplains.
Another important reason for considering flow-field derived flowpaths is that channel-floodplain hydraulics can steer the flowpaths in ways that do not follow the terrain.While at low floodplain-inundating flows, the flowpaths are most likely to follow the topography of the terrain.However, at high floodplain-inundating flows, the flowpaths more or less ignore the underlying terrain, where they effectively become roughness elements and follow a general floodplain flow direction (Fig. 3) (Czuba et al., 2019).This has implications for modeling channel-floodplain connectivity based on a terrain-only approach in a Geographic Information System, for instance, where the connectivity map derived from the terrain would not necessarily match that derived from a hydraulic model.An advantage of this approach is that it provides a way to extrapolate point measurements to a broader spatial coverage without further expensive and time-consuming field efforts (e.g., Fig. 6).However, this approach requires high-resolution floodplain topography data and at least a few measurements or estimates of floodplain deposition with which to build even a limited statistical model.Ideally, we would like to determine what specific characteristics of the floodplain topography best explain the PDFs of sedimentation rate (Middelkoop and Asselman, 1998;Nicholas andWalling, 1997, 1998;Nicholas and Mitchell, 2003).However, for our study, we have only investigated three distinct floodplains and would need maps of sedimentation rates for many more floodplains.This is possible with our current 2D hydrodynamic model results that extend from Waynesboro to Port Republic, Virginia (48 km reach; Fig. 1b), but was beyond the scope of the present study.Knowing more about how the subtle floodplain topographic relief affects the spatial pattern of sedimentation is important in determining the floodplain component of sediment budget models (Goodbred Jr and Kuehl, 1998;Dunne et al., 1998;Fryirs and Brierley, 2001;Malmon et al., 2002;Nicholas et al., 2006;Park, 2020).
Our map of sedimentation rate shows that most of the deposited sediments are concentrated in hotspots near the river channel and in low lying areas of the flood-  plain that are frequently inundated (Fig. 6b).We estimated that 75% of the sediment has deposited on only 41% of the South River floodplain (Fig. 7).The results of our work better inform how floodplain topography and river-floodplain connectivity affects sedimentation rates and can help inform the development of floodplain sediment budgets.

Summary
We combined results from a 2D hydrodynamic model of the South River, Virginia, with sedimentation rate data from Pizzuto et al. (2016) into multiple linear regression models describing sedimentation across the floodplain.Downstream flowpath length, exceedance probability of the first inundating flow, and suspended sediment concentration were identified as the dominant predictor variables describing sedimentation rate on the floodplain of the South River downstream of Waynesboro, Virginia.Sedimentation rate decreased expo-nentially with increasing flowpath distance, measured upstream of the river channel, and increased with inundation frequency and suspended sediment concentration.Our best multiple linear regression model to estimate the sedimentation rate incorporated downstream flowpath length and suspended sediment concentration (with flow exceedance implicit in our calculation) with an r 2 goodness-of-fit statistic of 0.514.In all of these models, we incorporated hydrologic flowpaths as an explicit measure of river-floodplain connectivity that improved our estimates of floodplain sediment deposition compared to the other parameters considered.
Next, we created a spatial map describing sedimentation rates along our study reach by applying our best multiple linear regression model to our 2D hydrodynamic model results.Floodplains with variable topography leading to low-lying areas, such as floodplain channels, that were inundated frequently had wider PDFs of sedimentation rate that were bimodal with the one mode occurring at values less than 0.03 cm/year and another mode occurring between 0.03 and 0.1 cm/year with a peak of this mode around 0.07 cm/year.This wider second mode represents deposition near the river channel and in low-lying areas of the floodplain that are frequently inundated.In the absence of major topographic relief, another floodplain had a log-normal PDF of sedimentation rate with most values less than 0.03 cm/year.Our estimated spatial average sedimentation rate (0.08 cm/year) between 1930 and 2007 was strongly dominated by this large flat relief floodplain.
Most of the deposited sediment is concentrated near the river channel and in low lying areas of the floodplain that are frequently inundated.We have estimated that 75% of the sediment deposited on 41% of the floodplain area just downstream of Waynesboro, Virginia.In the future, the approach we applied in three floodplains can be extended to the rest of the South River to compile more data relating floodplain topographic variability to spatial distributions of sedimentation rate to better inform the development of floodplain sediment budgets.Furthermore, future work should apply these methods to other floodplains outside of our study area to test the broader applicability of our approach and advance our understanding of floodplain sedimentation processes.

Figure 1
Figure 1 Study area of the South River near Waynesboro, Virginia.(a) Red star represents the location of the South River in Virginia, USA.(b) Local setting of the South River downstream of Waynesboro.Dark blue line represents the center line of the South River model domain (HEC-RAS).Red triangles show the location of the USGS gages (01627500 and 01626850) used for model calibration and validation.Light blue lines are other rivers in the area.Light gray line with transverse crosses represents the railroad and the dark gray lines are highways (I-64 and I-81).Black line demarcates the Waynesboro city boundary and the black rectangle shows the areal extent of (c).White line within the black rectangle is our area of interest (AOI) where we have focused our analysis.Background image is a 10 m digital elevation model (DEM) from USGS (2022).(c) Detrended elevation of the floodplain (using a Gaussian filter) within the AOI to highlight floodplain features by removing the downstream trending slope from the floodplain topography.White line shows the boundary of the AOI and the black line shows the boundary of three separate sections of floodplain, labeled A, B, and C. Orange circles are the mercury accumulation (kg/m 2 ) data from Pizzuto et al. (2016); larger sizes indicate higher mercury accumulation.The low relative elevations at the edges of the AOI, particularly next to steep hillslopes shown in the hillshade, are artifacts of the filtering and detrending process.Background image is a 0.76 m lidar DEM with a hillshade from VGIN (2017).

Figure 2
Figure 2 Hydrodynamic model calibration and validation at USGS streamflow-gaging stations 01627500 (Harriston) and 01626850 (Dooms) on the South River, Virginia (location shown in Fig. 1b, red triangle).Water surface elevations (WSEL) relative to flow discharge at Harriston (a) and Dooms (c).Depthaveraged velocities relative to flow discharge at Harriston (b) and Dooms (d).Vertical dash line denotes the first flow inundating the floodplain (28 m 3 /s; RI: 1-year).

Figure 3
Figure 3 Simulated depth (a-e) and streamlines (f-j) for 1.3-year RI flow (a, f; 83 m 3 /s), 1.7-year RI flow (b, g; 116 m 3 /s), 3.5-year RI flow (c, h; 280 m 3 /s), 11.5-year RI flow (d, i; 477 m 3 /s), and peak of 96-year record (e, j; 824 m 3 /s).The river channel can be seen as the darkest blue line in (a).Gray line is the inundation boundary for the 96-year peak flow.Light gray box in (a) is the extent of streamlines shown in the bottom panels.

Figure 4
Figure 4Frequency of inundation and river-floodplain connectivity.(a) First flow that inundates the floodplain and corresponding flow recurrence interval (RI).Dark blue is the start of floodplain inundation (vertical dash line at Fig.2; 28 m 3 /s, RI: 1-year).Each color gradation represents one of the 18 simulated flows that inundate the floodplain.Sizes of the orange circles represent the sedimentation rate with larger circles indicating higher rates(Pizzuto et al., 2016(Pizzuto et al., , 2023)).Gray lines denote the channel-floodplain boundary following the high points along the channel banks.(b) Circles represent the sampling points with generated flowpaths (dashed gray lines) at the highest flow (824 m 3 /s; peak of 96-year record).Colors of the circles correspond to the first flow that inundates this location, same as in (a).

Figure 5
Figure 5 Three multiple linear regression models used to estimate the sedimentation rate on the South River floodplain near the USGS gage at Dooms.(a) Model I: downstream flowpath length is the only predictor variable (b) Model II: downstream flowpath length and exceedance probability of flow are the predictor variables.(c) Model III: downstream flowpath length and suspended sediment concentration (SSC) are the predictor variables.Measured sedimentation rates are from Pizzuto et al. (2016, 2023).

Figure 6
Figure 6 Estimated sedimentation rate (cm/year) applying multiple linear regression Model I (a) and Model III (b).For Model I (a), downstream flowpath length was the only predictor variable and for Model III (b), downstream flowpath length and suspended sediment concentration were used as predictor variables.Gray lines denote the boundary of the river and floodplain along the highest points on the channel banks.Results from Model II are not shown because they are similar to those from Model III (b).

Figure 7
Figure 7 Percentage of the total amount of sedimentation (from Model III) relative to the percentage of area targeted if first starting in the areas with the highest amounts of sedimentation and then expanding to areas with lower amounts.

Figure 8
Figure 8 Probability distributions of sedimentation rates taken from values spatially across three floodplains along the South River (extent of each floodplain is shown in Fig. 1c as a black line and marked as floodplain A, B, C).Probability density function (PDF) and cumulative distribution function (CDF) for floodplain A (a, d), B (b, e), and C (c, f).Gray shaded areas and solid lines indicate the empirical data from Model I and light blue shaded areas and solid lines indicate the empirical data from Model III.Dark blue line is the Nakagami fit for floodplain A (a, d) and B (b, e) for Model III.Black dashed line (a, d) is the Weibull fit and orange dashed line (b, e) is the Beta fit for Model I. Red solid and dashed line is the log-normal fit for Model I and Model III, respectively, for floodplain C.