Journal of Hydrology 450–451 (2012) 230–243

Contents lists available at SciVerse ScienceDirect

Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol

The impact of different sediment concentrations and sediment transport

formulas on the simulated floodplain processes
Rohan Benjankar ⇑, Elowyn M. Yager
Center for Ecohydraulics Research, University of Idaho, 322 E. Front Street, Boise, ID 83702, United States

a r t i c l e i n f o s u m m a r y

Article history: Overbank sedimentation is an important process in river floodplain ecosystems and is a component of the
Received 16 November 2011 floodplain geomorphologic evolution. The impact of suspended sediment supply on floodplain processes
Received in revised form 27 April 2012 is still unclear because sediment deposition can be influenced by many factors. We quantified the effect
Accepted 4 May 2012
of sediment supply (suspended sediment) and transport formulas on simulated floodplain processes
Available online 16 May 2012
This manuscript was handled by Laurent
using a coupled two-dimensional hydrodynamic and sediment transport model (MIKE21C). Erosion
Charlet, Editor-in-Chief, with the assistance and deposition depths, net sedimentation depth and total volume were quantified based on the last time
of Eddy Y. Zeng, Associate Editor step of the simulation period.
The MIKE21C model was validated by comparing simulated water surface elevations to those from a
Keywords: one-dimensional hydrodynamic model. We compared the sediment transport model simulated
Suspended sediment concentration suspended sediment concentrations (SSCs) to measured concentrations at a gage station. Erosion and
Floodplain processes deposition processes were simulated using five hydrograph scenarios as a function of high and low
Hydrodynamic model SSC and two sediment transport equations, Van Rijn (1984) and Engelund and Hansen (1967).
Sediment transport model A specific location could be an erosional or depositional zone at different time steps of the simulation.
Sediment transport formula Thus, floodplain deposition is a discontinuous function of river discharge and varies spatially and tempo-
Sediment deposition
rally over the floodplain. Large flows with high SSC were more effective for floodplain deposition than
lower discharges, which dominantly caused sediment scour. Coupled hydrodynamic and sediment trans-
port models that account for feedback processes between topography and hydraulics should be given first
preference for future floodplain restoration projects. From a restoration perspective, larger flows are
required for greater floodplain deposition rates and maintenance of dynamic processes. The Engelund
and Hansen (1967) equation simulated higher transport rates than the Van Rijn equation (1984). For
future studies, transport equations should be selected based on the study objectives and field character-
istics. The current model might be used to analyze the impact on floodplain processes from altered SSC
due to Libby Dam operation.
Ó 2012 Elsevier B.V. All rights reserved.

1. Introduction (Noe and Hupp, 2009). Floodplain processes can also create barren
floodplain surfaces that riparian vegetation may colonize
Information on floodplain processes is essential to understand (Benjankar et al., 2011; Mahoney and Rood, 1998).
the geomorphic evolution of floodplains and to restore riparian Little attention has been given to fine sediment deposition on
ecosystems. Overbank sedimentation influences downstream the floodplain, although it can have a significant impact on flood-
sediment loads as well as sediment associated nutrients and con- plain formation (Walling et al., 1998). The spatial extent and
taminants (Walling and Owens, 2003). Fine sediment accumulation depths of sediment accumulation on floodplains have been studied
on floodplains increases floodplain elevations and modifies inunda- over decades or longer timescales (e.g., Buttner et al., 2006; Dunne
tion frequency and channel conveyance capacity. Furthermore, et al., 1998; He and Walling, 1996; Heimann, 2001), but it may not
sediment deposition creates important riparian habitat such as cot- be possible to distinguish distinct events on annual timescales
tonwood stands and replenishes sediment associated nutrients (Aalto et al., 2003).
Overbank sedimentation rates are governed by floodplain
topography, local hydraulics, vegetation density, flood frequency,
⇑ Corresponding author. Address: Center for Ecohydraulics Research, University
flood duration, and suspended sediment concentration (SSC) and
of Idaho, 322 E. Front Street, Suite 340, Boise, ID 83702, United States. Tel.: +1 208
284 8472; fax: +1 208 332 4425.
depositional processes (Nicholas and Walling, 1997; Nicholas and
E-mail address: [email protected] (R. Benjankar). Walling, 1998). High deposition rates occur at low elevations that

0022-1694/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jhydrol.2012.05.009
 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243 231

are frequently fluvially inundated for long time periods and depo- formulas on floodplain processes. Erosion and deposition depths,
sition rates decrease with distance from the river (Aalto et al., net sedimentation depths and total volumes were calculated for
2003; Nicholas and Walling, 1998). a range of flood hydrographs, SSC rating curves and sediment
Suspended sediment transport is a complex phenomenon, and transport formulas.
suspended sediment concentration (SSC) and suspension duration
depend mainly on flow hydraulics but are also controlled by the
2. Methodology
particle settling velocity and upstream sediment supply. Sus-
pended sediment transport can be supply limited, and suspended
2.1. Study area
sediment fluxes need to be simulated using an unsteady flow mod-
el that considers advection and diffusion processes, and the up-
The study area (Fig. 1) is located within the Kootenai Basin,
stream sediment influx (Hardy et al., 2000).
which is an international watershed between the USA and Canada.
The impact of suspended sediment supply on floodplain
The natural flow regime of the Kootenai River is snowmelt domi-
processes is still unclear because sediment deposition can be influ-
nated (Hoffman et al., 2002), with a sustained peak in the late
enced by many factors such as grain size, flood duration, discharge,
spring, followed by a gradual recession to base flow by September,
floodplain topography and flow frequency (Middelkoop and Van
and low winter flows. The Kootenai Basin has been intensively
der Perk, 1998). Additionally, floodplain deposition rates may be
managed since the late 1800s. Historical basin management in-
different during the rising and falling limbs of the hydrograph
cluded floodplain diking (levees), wetland draining and conversion
(Buttner et al., 2006). The effective discharge for floodplain sedi-
of the floodplain into agricultural land (Tetratech, 2004). Dikes and
mentation is also not well understood although rapidly rising large
levee construction disconnected the floodplain from the main
flows are more effective for floodplain evolution than slowly rising
channel and eliminated floodplain processes. In addition, the flood-
medium flows (Aalto et al., 2003).
plain has been impacted by backwater effects from Corra Linn Dam
In addition to the uncertainty about processes that control
at Kootenay Lake. Libby Dam construction 350 km upstream of
floodplain sedimentation, field measurements (e.g., flood depth,
Corra Linn Dam began in 1966 and was completed in 1973. Com-
shear stress, sediment transport rates) are very difficult to make
pletion of Libby Dam altered peak flow, sediment transport, aqua-
because of the unpredictable timing of floods and the dangerous
tic habitat, and riparian vegetation. The post-dam peak discharge,
conditions during these flows. Therefore, a 2D hydrodynamic
inundation extent (Benjankar, 2009) and SSC decreased signifi-
(HD) model can be a very effective tool to better understand the
cantly from pre-dam condition. There are a total of five gage sta-
influence of floods of range of magnitude on floodplain processes.
tions in the main stem Kootenai River from Libby Dam to
There are many numerical models available to simulate sedi-
Boundary Creek (Fig. 1). The Copeland gage station was used to
ment transport and floodplain processes. Few previous studies
measure SSC from 1966 to 1983, but was discontinued recently.
have considered re-suspension of sediment from the floodplain
This study focuses on the meander reach of the Kootenai River,
and most studies are based on a one-dimensional (1D) model with
located between the town of Bonners Ferry and the Copeland gage
steady discharges (e.g., Carrolla et al., 2004). Floodplain processes
station (Fig. 1). The study area is 133 km downstream from Libby
are often simulated with a single or effective grain size distribution
Dam. The meander reach consists of a wide floodplain that is
(Buttner et al., 2006), which may have significant impacts on the
highly confined between levees. Our specific study area is the
calculated sediment transport rates (James et al., 2006). Further-
floodplain (16 km2) at Ball and Trout Creeks (Fig. 1), which is cur-
more, simulated floodplain processes may be significantly different
rently being used for agriculture, pasture and grassland. The flood-
depending on the applied sediment transport equations because
plain is constructed primarily of lacustrine deposits, whereas the
the suitability of given formula varies.
river bed and banks are constructed of fine-grained silt and sand
Many different formulas are available to estimate sediment
materials. The channel sinuosity is 1.7 and the gradient is
transport such as Meyer-Peter and Muller (1948), Ackers and
0.0006% (cf., Burke, 2006). We focused on this portion of the
White (1973), Yang (1996), Van Rijn (1984), and Engelund and
floodplain because of its potential use as a restoration site.
Hansen (1967). These formulas are developed for specific systems
and are mainly compared to a limited set of field or experimental
data (Camenen and Larroude, 2003). There is no universal formula 2.2. Hydro dynamic and sediment transport model development
that is valid for all cases because many parameters influence sed-
iment transport. 2.2.1. Hydrodynamic model
In addition to uncertainty in transport formulas, as the flood- A two-dimensional model (MIKE21C) with a curvilinear orthog-
plain topography evolves (through erosion and deposition), new onal grid (15 m by 15 m) was developed to simulate floodplain
topography will affect the hydraulics and subsequent sedimenta- processes. A major disadvantage of a large grid size is the loss of
tion. Therefore, a model should include the feedback mechanism important small-scale features (e.g., ditch, depression, etc.) that af-
between floodplain topography, hydraulics and re-suspension to fect flow hydraulics and eventually sediment transport. This might
simulate reasonable overbank sediment transport phenomena not have significant effects on flow hydraulics and sediment trans-
(Hardy et al., 2000). Most models used to simulate floodplain pro- port because the study area is currently being used as agricultural
cesses lack re-suspension of sediment (e.g., Middelkoop and Van land and is relatively flat.
der Perk, 1998) and feedback processes between the topography MIKE21C simulates hydrodynamic and morphological changes
and floodplain hydraulics. in rivers and estuaries using a finite volume-finite difference algo-
We used a spatially distributed two-dimensional (2D) hydrody- rithm. MIKE 21C calculates the total sediment flux (bed load and
namic model coupled with a sediment transport model on the suspended load) and sedimentation in each grid cell. We used MI-
Kootenai River floodplain, Northern Idaho, USA (Fig. 1). The Danish KE21C because it can simulate multiple size classes of cohesive and
Hydraulic Institute (DHI) program MIKE21C (DHI, 2008) was used non-cohesive suspended sediment, and bed load transport, includ-
to simulate floodplain processes as a function of different unsteady ing the deposition and re-entrainment of sediment (DHI, 2008).
hydrologic scenarios, sediment concentrations and sediment trans- Post-dam era topography, surveyed using LiDAR (Light Detection
port formulas. Our model includes re-suspension of sediment and and Ranging) in 2005, was used to represent the floodplain topog-
the feedback between hydraulics and topography. The model pro- raphy. The river bathymetry was developed by linear interpolation
vides information about the effect of SSC and sediment transport of eleven cross-sections. Levees were breached at eight different
232 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243

Fig. 1. Location of the study area. Floodplain topography, levee breach locations, and river cross-sections (XS) utilized to create river bathymetry. The H and are the
locations where erosion or deposition (E or D) and suspended sediment concentration (SSC) time series are extracted.

locations in the simulation (Fig. 1) and the elevations of the brea- coefficient in the channel (Table 1) until the simulated WSEs from
ched levee locations were set to 537.00 m, which is similar to the MIKE21C closely matched those from MIKE11, for the 1976
floodplain elevation near the levees. We breached levees because hydrograph.
none of the discharges considered in this study could inundate For the floodplain, a range of manning’s roughness values (Table
the floodplain with the existing levee configuration. The levee 1) were assumed based on land cover data from Benjankar (2009)
breach also provided an estimate of an effective discharge required and other studies (Acrement and Schneider, 1989; Alkema and
to restore natural floodplain processes in the future. Middelkoop, 2005; Ayres Associates, 2002; Chow, 1959; Hesselink
The MIKE21C model was developed using the hydraulic bound- et al., 2003). Other studies have shown that roughness parameters
ary conditions and other information, such as floodplain rough- did not significantly influence floodplain inundation simulations
ness, from a previously developed 60 m grid MIKEFLOOD HD (e.g., Benjankar, 2009; Paquier and Mignot, 2003), but this conclu-
model (Benjankar, 2009). MIKEFLOOD integrates river (1D) and sion is not universally applicable (Werner et al., 2005). Further-
floodplain (2D) models into a coupled system. Discharge and water more, the study reach is within a backwater affected zone
surface elevation (WSE) time series simulated from the one- (Berenbrock, 2006).
dimensional (1D) MIKE11 model (Benjankar, 2009) were used for Absolute mean error (AME) and root mean square error (RMSE)
the upstream and downstream boundary conditions, respectively, were calculated to estimate errors of the simulated WSE from MI-
in the MIKE21C model used in this study because there were no KE21C, assuming MIKE11 closely approximated the measured
gage stations within the study area. The root mean squared error WSE. Percent (%) AME and RMSE in flow depth were calculated
(RMSE) of the MIKE11 predicted WSE from the measured values as the ratio of mean error to the average MIKE11 water depths.
was 0.15 m and 0.16 m, during calibration and validation pro- Model validation used the 1982 discharge (Table 1) and the man-
cesses, respectively. For the MIKE21C calibration and validation, ning’s roughness coefficients that were optimized during the mod-
post-dam discharge hydrographs (represented by years 1976 and el calibration. As with the calibration process, the MIKE11 and
1982) were used (Table 1). MIKE21C was calibrated and validated MIKE21C WSE were compared to estimate RMSE and AME.
by comparing the simulated WSEs from MIKE21C to those from For the actual model runs, five different representative hydro-
MIKE11 at seven cross-sections (Fig. 1). The calibration process graphs were considered, in which there were three flows prior to
was performed by iteratively modifying manning’s roughness Libby dam operation (1945, 1965, and 1950) and two following
 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243 233

Table 1
Simulated peak discharges (Q*) at Libby Dam gage station with corresponding pre-dam recurrence interval (RI**) and the peak discharge (Q***) at the study area including the
recurrence interval (RIa) based on the Leonia gage station after Libby dam regulation.

Simulated year Peak Q* (m3/s) RI** (Year) Peak Q*** (m3/s) RIa (Year) Roughness/
Land cover Manning’s n
1976x 288 <1 1132 5 Forest 0.135
1982w 532 <1 1217 5 Shrub dominated riparian 0.053
1996 799 1.05 1492 20 Agriculture land 0.046
1974 1031 1.15 1452 20 Grassland 0.044
1945 1375 1.3 1708 50 Wetlands 0.042
1965 1798 2 2093 200 Gravel and sand bar 0.04
1950 2103 5 2569 >200 Open water (river) 0.025b
*
Peak discharge at Libby Dam gage station.
**
Pre-dam recurrence interval based on Libby dam gage station without dam regulation.
***
Peak discharge at the study area based on 1D model used in MIKE21C.
x
Discharge used for MIKE21C calibration.
w
Discharge used for MIKE21C validation.
a
Post-dam recurrence interval of peak discharge at the study area based on Leonia gage station.
b
The roughness in channel calibration of MIKE21C.
/
Assigned manning’s roughness coefficient on the floodplain is based on land cover data.

operation (1974, 1996). The peak discharge in 1974 was the max- (pre-dam) and (3) (post-dam) were used to calculate potential high
imum discharge measured at the study area after dam operation and low sediment concentrations for the upstream boundary con-
began (1973), until 2005. The pre-dam discharges (1945, 1965, ditions, respectively. The pre-dam SSCs were higher than the post-
and 1950) had pre-dam recurrence intervals (RI) of 1.3, 2, and dam values. The main reason for adopting these equations was that
5 years (Table 1), which have equivalent post-dam RI of 50, 200, they provided the maximum possible differences in SSC for a given
and more than 200 years, respectively, based on the Leonia gage discharge. These two sediment concentrations were used to simu-
station. Each event was simulated for five days (two days before late differences in floodplain process as a function of sediment sup-
and after the peak flood at the study area) because of the extensive ply using the Engelund and Hansen (1967) sediment transport
computational time need for a given flow. equation.

SSC ¼ 0:0034Q 1:4212 ð1Þ

2.2.2. Sediment transport model R2 ¼ 0:8136
SSC rating curves (Eqs. (1)–(4)) were developed using discharge
and SSC measurements from 1966 to 1983 at the Copeland gage SSC ¼ 0:0084Q 1:346 ð2Þ
station (Fig. 2, left). The United States Geological Survey (USGS) R2 ¼ 0:9361
measured SSC (near the bed) within a cross-section (Edwards
and Glysson, 1999). Eqs. (1) and (4) were developed by fitting SSC ¼ 0:0204Q 0:9102 ð3Þ
SSC data and corresponding discharges during pre- (pre-1968) R2 ¼ 0:5635
and post-dam (post-1975) eras, respectively (Fig. 2, left). The coef-
ficient of correlation (R2) between discharges and SSC is relatively
SSC ¼ 0:0545Q 0:829 ð4Þ
low (0.3024) for Eq. (4), probably due to the impact of Libby Dam
R2 ¼ 0:3024
operation on downstream sediment transport and supply.
The SSC data were classified into values greater and less than where SSC = suspended sediment concentration (mg/l); Q = dis-
these linear regressions to develop Eqs. (2) and (3). Therefore, Eq. charge (m3/s); R2 = correlation coefficient.
(2) was fit to data greater than the regression of SSC (Eq. (1)) in Two long duration (LD) simulations (6 months), with high sed-
the pre-dam era. Similarly, Eq. (3) represents the data located iment concentrations (Eq. (2)) and discharge hydrographs with
below the linear regression from the post-dam era. Eqs. (2) peaks of 799 and 1798 m3/s were also used to study differences

Fig. 2. Relationship between discharge and SSC at the Copeland gage station before and after Libby Dam operation (left). Grain size distribution on the floodplain and river
channel (right).
234 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243

in floodplain processes from the normal simulation duration transport is a function of the bed load concentration, saltation
(5 days). Higher erosion or deposition volumes were expected with height of the particles and particle velocity. Suspended sediment
a longer duration, because generally, the sediment erosion or depo- transport is computed by integrating the product of the SSC and
sition volume on the floodplain is function of flood duration (Mid- the depth-averaged flow velocity. The Van Rijn formula is given
delkoop and Van der Perk, 1998). by the following two equations.
Two transport equations, Van Rijn (1984) and Engelund and qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Hansen (1967), were chosen to simulate floodplain processes using T 2:1 3
Sbl ¼ 0:053 ðs  1Þgd50 ð10Þ
pre-dam SSC inputs (Eq. (1)). These equations represent the two D0:3

main types of models available for sediment transport based on
either bed-shear stress (Van Rijn) or energy (Engelund–Hansen). Ssl ¼ fca uh ð11Þ
Simplified forms of Engelund and Hansen (1967) and Van Rijn

(1984) formulas are used in the MIKE21C morphodynamic model where T = non-dimensional transport stage parameter; D = non-
(DHI, 2008). Refer to the DHI (2008) user’s manual for all the param- dimensional particle parameter; f = concentration factor of sus-
eters and variables used in these sediment transport formulas. pended load; ca = volumetric bed concentration; h = flow depth.
The Engelund and Hansen (1967) equation calculates the total Sediment proportions in six different grain size bins (with a
load and divides this into bed and suspended load based on stream maximum grain diameter of 0.5 mm) were used for both the chan-
power (product of bed shear stress and flow velocity). This equa- nel and floodplain, based on previous studies (Barton, 2004; Barton
tion is applicable for sediments ranging from 0.062 to 32 mm in et al., 2005). We used the same grain size distribution on both the
diameter. The equation uses energy slope, velocity, bed shear channel and floodplain, which likely over-estimates of floodplain
stress, median particle diameter, specific weight of sediment and grain sizes (Fig. 2, right). The D84 of the suspended load and stream
water, and gravitational acceleration to calculate sediment trans- bed sediments in post-dam conditions were 0.1 mm (Barton,
port. It is given by the following equations. 2004) and 0.34 mm (Barton et al., 2005), respectively but no surface
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi material size distribution on the floodplain is available. Our study
C 2 5=2 3 does not analyze sediment transport and changes in bathymetry
Stl ¼ 0:05 h ðs  1Þgd50 ð5Þ
g in the channel and only focused on the floodplain. We assumed
the upstream supply of bed load only occurred within the channel
s and not on the floodplain because the floodplain was approximately
h¼ ð6Þ
qgðs  1Þ 18 m (Table 2) above the channel. Thus, because we were only
interested in the floodplain, we neglected the upstream bed load
u2 supply entirely. Suspended sediment represents 80% of total load
s ¼ qg ð7Þ near the study area (Berenbrock and Bennett, 2005) and neglecting
C2
upstream bed load inputs is therefore reasonable.
where Stl = total load; C = Chezy number; h = Shields parameter; We assumed grain size distribution, dimensionless critical shear
g = acceleration of gravity; s = relative density of the sediment; stress, manning’s roughness, porosity, cohesion less sediment and
d50 = median grain size; s = flow shear stress; q = density of water; bare floodplain without vegetation. The dimensionless critical
u = flow velocity. shear stress was assumed 0.056 (Barton et al., 2005) in the model.
Sbl ¼ kb Stl ð8Þ Calculations of sediment transport are sensitive to the value of
dimensionless critical shear stress, which may vary by an order
Ssl ¼ ks Stl ð9Þ of magnitude (Buffington and Montgomery, 1997). Critical shear
stresses for incipient motion may also vary significantly within
where Sbl = bed load; Ssl = suspended load; kb = fraction of bed load; mixed-sized and poorly sorted sediments (Wilcock and McArdell,
ks = fraction of suspended load. 1993). Although the value of critical shear stress can impact the
The Van Rijn (1984) formula uses grain shear stress calculated overall magnitude of erosion and deposition, it will not change
from flow velocity and grain roughness. Bed load sediment our overall pattern of results on the difference between upstream

Table 2
Difference between measured and simulated water surface elevations at seven different cross-section locations and average errors (AME and RMSE) for model calibration and
validation processes.

Date Absolute error (m)

XS1 XS2 XS3 XS4 XS5 XS6 XS7
Calibration (1976 hydrograph)
Day 1 0.05 0.03 0.08 0.10 0.01 0.02 0.06 Mean (d) 18.23 m
Day 2 0.19 0.06 0.05 0.09 0.07 0.04 0.06 AME 0.12 m
Day 3 0.20 0.02 0.13 0.21 0.04 0.01 0.06 RMSE 0.15 m
Day 4 0.26 0.04 0.15 0.25 0.07 0.01 0.07
Day 5 0.36 0.11 0.11 0.24 0.04 0.04 0.08 AME (%) 1
Day 6 0.42 0.13 0.11 0.24 0.04 0.06 0.09 RMSE (%) 1
Day 7 0.45 0.15 0.12 0.26 0.05 0.07 0.10
Validation (1982 hydrograph)
Day 1 0.02 0.04 0.10 0.15 0.03 0.04 0.04 Mean (d) 18.73 m
Day 2 0.10 0.01 0.13 0.18 0.04 0.02 0.04 AME 0.13 m
Day 3 0.11 0.00 0.20 0.27 0.12 0.05 0.04 RMSE 0.17 m
Day 4 0.13 0.01 0.26 0.34 0.17 0.07 0.03
Day 5 0.13 0.04 0.32 0.41 0.23 0.10 0.04 AME (%) 1
Day 6 0.25 0.06 0.26 0.34 0.16 0.02 0.04 RMSE (%) 1
Day 7 0.44 0.24 0.10 0.18 0.00 0.15 0.04

The abbreviations refer to the following: d: measured mean water depth during the simulation period at the specific cross-section (XS5 in Fig. 1).
 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243 235

sediment concentrations, and sediment transport equations. fractions and reran the model until the predicted SSC closely
Porosity value of 0.3 was assumed in the model. The porosity of matched the measured values (i.e., low AME and RMSE). Further-
sediment generally varies from 0.3 to 0.7 (DHI, 2008). Porosity typ- more, we validated the model by comparing the measured and
ically varies from 0.282 to 0.376 for uniformly distributed grain simulated concentrations (Table 3A) during the 1982 discharge
sizes in the river bed (Sulaiman et al., 2007). Sediment transport hydrograph. The post-dam SSC rating curve at the Copeland gage
is relatively insensitive to porosity but is very sensitive to man- (Eq. (4)) was used as the upstream boundary condition of total sed-
ning’s roughness (Habersack, 1998). iment concentration for both calibration and validation of the
The minimum grain size (0.007 mm) considered in this study model.
was greater than maximum size (0.002 mm) of cohesive soil (Yang, The sediment transport and hydrodynamic models were com-
2006). Modeling cohesive sediment dynamics may be difficult be- bined to simulate floodplain processes (over 5 days) for five differ-
cause the physical properties of cohesive sediments are complex ent unsteady discharge scenarios (Table 1). Area weighted mean
(Lumborg and Windelin, 2003). Furthermore, vegetation on the erosion and deposition depths, net sedimentation depths and total
floodplain can substantially change floodplain stability and flow volumes (only for floodplain) were calculated at the end of each
resistance for different vegetation densities (Gran and Paola, simulation period to determine the change in floodplain topogra-
2001). Vegetation increases the local resistance to flow, reduces phy as an indicator of floodplain processes. Only a cell that had
flow velocities and decreases bed shear stresses. Concurrently, this an erosion or deposition depth of more than 1 mm was considered
will decrease soil erosion and sediment transport capacity (Carollo in the calculation of area weighted mean depths. Net sedimenta-
et al., 2002). Additionally, vegetation reduces soil erosion by tion depth was calculated based on the total quantity (volume)
increasing the soil strength from root binding (Gyssels et al., in an area and an area weight, assigning positive and negative signs
2005). We based our estimate of flow resistance from vegetation for deposition and erosion, respectively. Total volume was calcu-
on floodplain and therefore we have captured the bulk roughness lated by the summation of the product of erosion or deposition
effects of vegetation and the spatial variability of this roughness depths and the associated area in each grid cell. The floodplain
on the floodplain. We do not account for the increase in soil parameters were calculated as a function of suspended sediment
strength caused by vegetation, which would likely increase our concentration (high and low) and sediment transport formulas
net sedimentation depths under conditions of high scour. (Engelund and Hansen, 1967; Van Rijn, 1984).
To verify the sediment transport model results, we compared A one-tailed t-test (a = 0.05) was used to determine if the sim-
the SSCs from the MIKE21C simulation to the field measured values ulated erosion, deposition or net sedimentation depths were signif-
at the Copeland gage station (20 km downstream from the study icantly different between various sediment concentrations and
area). There is not a stream flow gage station located in the study transport formulas. Generally, the t-test is based on the assump-
reach. Therefore, the sediment concentration from the Copeland tion that two groups of data are normally distributed around their
gage station was used as a surrogate because conditions at this respective means, and that they have the same variance. However,
location were similar to those in the study reach (e.g., both are lo- if the two groups have unequal variances, the t-test statistics and
cated in the meander reach and backwater zone from Corra Linn calculations of degree of freedom (df) should be modified accord-
Dam). We assumed that there were not significant differences in ing to Satterthwaite’s approximation (Helsel and Hirsch, 2002).
sediment concentration between the study reach and Copeland The one-tailed t-test was performed assuming that the two groups
gage station because there are no major tributary inputs between have unequal variances using the equations by Helsel and Hirsch
them. The calibration and validation of the sediment transport (2002). The t-test was based on the entire floodplain with the num-
model includes additional potential error sources when coupled ber of grids as the sample size and the calculated means and vari-
with the HD model, because uncertainties also arise from the HD ances from all grid cells.
model.
The simulated SSC (using the 1976 discharge hydrograph) was
3. Results
extracted from fourteen different random locations in the study
area (Fig. 1) and averaged to compare with the measured concen-
The errors associated with calibration and validation of both the
trations (Table 3A) at the Copeland gage station. MIKE21C requires
hydrodynamic (HD) and sediment transport models are shown in
specifying the proportions of each grain size that are transported as
Tables 2 and 3A. The AME and RMSE of the WSE were 0.12 (1% er-
either bed load or suspended load to be used in the sediment trans-
ror) and 0.15 m (1% error), respectively for the calibration process.
port equations, which predict total load (DHI, 2008). Therefore, we
The measured SSC at the Copeland gage station ranged from 14 to
had to assume the fraction of each grain size that was transported
19 mg/l during the simulated period, resulting in an average con-
as either bed load or suspended load. We adjusted each of these
centration of 16 mg/l, whereas the simulated concentration ranged

Table 3
Measured and simulated suspended sediment concentrations for model calibration and validation processes (3A, left) and area weighted mean depths and velocities for different
discharges considering results from last time step of the simulation.

A B
Calibration (1976 hydrograph) Validation (1982 hydrograph) Year Q Ext Depth (m) Velocity (m/s)
Date Md Si Md Si Av sd Av sd
Day 1 15 19 Mean, C 16 19 22 Mean, C 17 1996 799 6.13 1.59 1.10 0.05 0.110
Day 2 17 22 AME 2 19 22 AME 2 1974 1031 6.61 1.76 1.11 0.07 0.100
Day 3 19 21 RMSE 3 20 21 RMSE 2 1945 1375 8.70 2.32 1.28 0.10 0.110
Day 4 17 18 18 18 1965 1798 10.88 2.67 1.46 0.11 0.100
Day 5 16 15 AME% 13 17 15 AME% 11 1950 2103 12.49 3.97 1.61 0.15 0.120
Day 6 15 14 RMSE% 16 13 11 RMSE% 12

The abbreviations refer to the followings: C: measured mean concentration (mg/l) at Copeland gage station during simulation period; Md: measured concentration (mg/l); Si:
simulated concentration (mg/l); Ext: inundation extent (km2); Av: Area weighted mean water depth (m) or velocity (m/s); Sd: standard deviation of depth (m) or velocity (m/s).
Q, peak discharge at Libby Dam gage station.
236 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243

from 12 to 22 mg/l. The AME and RMSE were 2 and 3 mg/l during locations (Fig. 1) for each of the different upstream sediment con-
calibration, respectively. Therefore, the percentage (%) AME and centrations and sediment transport formulas. These results reveal
RMSE were between 13 and 16%, respectively. floodplain processes at a specific location throughout the entire
The AME and RMSE of the WSE were 0.13 and 0.17 m, respec- simulation period. Second, the area weighted mean erosion and
tively during the validation process. Therefore, the percentage (%) deposition depths, net sedimentation depths, total volumes of ero-
of AME and RMSE were about 1%. The average measured SSC was sion and deposition and t-test results (Tables 4 and 5), based on the
17 mg/l during the hydrograph used for the validation processes. last simulation time step are presented.
Both AME and RMSE of concentration were 2 mg/l, resulting in
percentage (%) errors of 11% and 12%, respectively. 3.1.1. Sediment concentration
We used the last time step of the simulation to calculate flow The local erosion and deposition patterns at random locations
hydraulics (i.e., floodplain inundation extent, water depth and through time for high and low upstream sediment concentrations
velocity) for different flow scenarios (Table 3B). Floodplain inunda- are shown in Fig. 5. A specific location can experience erosion or
tion extent ranged from 6.13 to 12.49 km2 for low (799 m3/s) to deposition at different time steps of the simulation (Fig. 5A). Fur-
high (2103 m3/s) discharges, respectively. Area weighted mean thermore, a location could be an erosion zone during high sedi-
water depths were 1.59 ± 1.10 m (mean ± standard deviation) and ment concentrations and a deposition zone during low sediment
3.97 ± 1.61 m for low and high discharges, respectively. Similarly, concentrations (Fig. 5B), or the reverse may be true. A given loca-
the area weighted mean velocity increased from 0.05 ± .11 m/s tion can also consistently experience erosion (Fig. 5C) or deposition
for low discharge to 0.15 ± .12 m/s for high discharges. (Fig. 5D), regardless of the relative sediment concentration.
Water depth and velocity distributions varied spatially across For the floodplain averaged calculations, erosion depths were
the floodplain (Figs. 3 and 4). The variability in the velocity distri- larger with high sediment concentrations than with low sediment
butions were greater (Fig. 4 and Table 3B) for the low discharge concentrations (Fig. 6, left), but the reverse was true for deposition
(799 m3/s) than the high discharge (2103 m3/s). Relatively con- depths. Total erosion volumes were larger with low sediment con-
stant velocities across the floodplain occurred for the high dis- centrations than high sediment concentrations (Fig. 7, left) and the
charge except at few locations (Fig. 4). The flow paths were opposite was true for deposition volumes. As a result, net sedimen-
visually different for low and high discharges, therefore, the flow tation depths were larger with higher sediment concentrations.
channelized in the low discharges, which caused significant local The t-test for the area weighted mean depths revealed that the ero-
erosion. sion depths were significantly different between high and low sed-
iment concentrations (Table 4) for all discharge scenarios except
3.1. Floodplain processes 1031 m3/s. Similarly, the deposition depths were significantly dif-
ferent for all the discharges except for discharges 799 and
The floodplain processes simulated with the model are pre- 1798 m3/s. Meanwhile, net sedimentations depths were signifi-
sented using two different approaches. First, the local erosion cantly different between high and low sediment concentrations
and deposition patterns were extracted from a 100 random for all discharges.

Fig. 3. Spatially distributed water depths for discharges 1375 (left) and 2103 m3/s (right).
 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243 237

Fig. 4. Spatially distributed velocity magnitudes and directions for discharges 1375 (left) and 2103 m3/s (right).

Table 4
Area weighted mean depth and total volume of erosion or deposition, net sedimentation from the high and low sediment concentrations and sediment transport formulas.

Year Q Erosion Deposition Net sedimentation

High Low Testd High Low Testd High Low Testd
T V T V T V T V T V T V
1996 799 0.147 115 0.134 118 Yes 0.019 34 0.022 32 No 0.031 81 0.037 86 Yes
1974 1031 0.123 119 0.116 122 No 0.016 48 0.025 46 Yes 0.018 71 0.026 76 Yes
1945 1375 0.112 126 0.103 128 Yes 0.017 77 0.034 70 Yes 0.009 49 0.018 58 Yes
1965 1798 0.085 259 0.075 278 Yes 0.069 498 0.067 381 No 0.023 240 0.011 102 Yes
1950 2103 0.101 408 0.094 428 Yes 0.107 828 0.115 802 Yes 0.036 420 0.032 374 Yes

Long duration
Erosion Deposition Net sedimentation
T Testh %/ V T Testh %/ V T Testh %/ V
1996 799 0.195 Yes 25 185 0.017 Yes 12 63 0.026 Yes 18 122
1965 1798 0.107 Yes 21 298 0.070 No 1 535 0.023 No 3 237

The symbol and abbreviations refer to followings: T: area weighted mean depth of erosion or deposition in meter (m); V: Volume of sediment in 1000 m3; High or Low: high
or low sediment concentration; Yes or No: corresponding test is significantly different or not based on t-test; Testd : Statistical test of erosion and deposition mean depth
between high and low SSC; Testh: Statistical test result between normal and long duration simulation; %/: Ratio of difference between mean erosion or deposition depth from
long duration and normal simulation duration to the mean erosion or deposition depth from long duration simulation in percentage.

For given sediment concentration, mean erosion depths de- discharge 1375 m3/s and positive above this discharge, for both
creased with higher discharges except for 2103 m3/s, whereas, high and low sediment concentrations (Fig. 6, right).
deposition depths increased with higher discharges except for The erosion depths with the long duration (6 months) simula-
1031 m3/s and 1375 m3/s (Table 4). The total volumes of erosion tions were higher (Figs. 6 and 7 and Table 4) than normal simula-
or deposition increased with higher discharges for both high and tions (5 days) for both discharges 799 (25% higher) and 1798 m3/s
low sediment concentrations (Fig. 7, left and Table 4). The erosion (21% higher). Conversely, the deposition depths were lower than in
volumes were higher than deposition volumes for discharges 799, the normal simulation for the discharge 799 m3/s (12%), but
1031 and 1375 m3/s for both high and low sediment concentra- slightly higher for the 1798 m3/s flow (1%). The net sedimentation
tions, but the opposite was true for discharges 1798 and depth for a discharge of 799 m3/s, with a long duration simulation,
2103 m3/s. As a result, net sedimentation was negative up to the was negative and was less than that from the normal simulation
238 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243

Table 5
Area weighted mean depth and total volume of erosion or deposition, net sedimentation from the Engelund and Hansen (1967) or Van Rijn (1984) sediment transport formulas.

Year Q Erosion Deposition Net sedimentation

l l
EH VR Test EH VR Tes EH VR Testl
T V T V T V T V T V T V
1996 799 0.146 113 0.140 52 No 0.018 39 0.013 15 Yes 0.025 74 0.024 37 No
1974 1031 0.119 104 0.099 50 Yes 0.017 60 0.011 25 Yes 0.010 45 0.009 25 No
1945 1375 0.112 123 0.103 55 Yes 0.017 83 0.012 47 Yes 0.007 39 0.002 8 Yes
1965 1798 0.084 261 0.037 116 Yes 0.071 490 0.039 250 Yes 0.023 229 0.014 134 Yes
1950 2103 0.099 410 0.047 159 Yes 0.108 817 0.061 455 Yes 0.035 406 0.027 296 Yes

The symbol and abbreviations refer to the followings: T: area weighted mean depth of erosion or deposition in meter (m); V: Volume of sediment in 1000 m3; EH or VR:
simulated with Engelund and Hansen (1967) or Van Rijn (1984) sediment transport formula; Yes or No: corresponding values are significantly different or not based on
statistical test; Testl: Statistical test result between with Engelund and Hansen (1967) or Van Rijn (1984) sediment transport formula.

Fig. 5. Erosion or deposition patterns with high and low sediment concentrations (A–D), Engelund and Hansen (1967) and Van Rijn (1984) sediment transport formulas (E
and F). Simulated velocities are shown in (G and H) during entire simulated period for different discharges at random locations (Fig. 1). Vertical axis for Fig. 5, A–F and G–H are
erosion or deposition depth (m) and velocity (m/s), respectively. The abbreviations a, b, c, d, and e refer to discharges 799, 1031, 1375, 1798, and 2103 m3/s, respectively. High
or Low: high or low sediment concentration; EH or VR: simulated with Engelund and Hansen (1967) or Van Rijn (1984) sediment transport formula.

Fig. 6. Area weighted mean erosion or deposition (left) and net sedimentation depths (right) with high and low sediment concentration for different discharge conditions.
The abbreviations refer as follows: er: erosion; dep: deposition; LD: long duration simulated for 6 months. Refer to Fig. 5 for other abbreviations.

duration (18%). Similarly, the net sedimentation depth was positive significantly different for discharge 799 m3/s, but not for discharge
with 1798 m3/s, but was slightly less than that from the normal 1798 m3/s.
simulation (3%). The erosion and deposition volumes (Table 4)
were greater for a long duration simulation than a normal simula- 3.1.2. Sediment transport formulas
tion for both discharges (799 and 1798 m3/s). The mean erosion There were not any generalizable patterns of erosion or deposi-
depths were significantly different between normal and long dura- tion with different sediment transport equations (Engelund and
tion simulations (Table 4) for both discharges (799 and 1798 m3/s). Hansen (EH) and Van Rijn (VR)) during for any of the discharge
However, the deposition and net sedimentation depths were conditions. The EH transport formulation simulated higher erosion
 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243 239

Fig. 7. Erosion or deposition volumes for high and low sediment concentration (left) and with Engelund and Hansen (1967) and Van Rijn (1984) sediment transport formulas
(right) for different discharge conditions. Refer to Figs. 5 and 6 for abbreviations.

or deposition rates than the VR equation in most of the cases 4. Discussion

(Fig. 5E and F). Specific locations could change between erosion
and deposition at different time steps of the simulation period 4.1. Floodplain processes
(Fig. 5F). A given location could be a deposition zone with EH
and erosion zone with VR, or the opposite could occur. We used suspended sediment rating curves to estimate the sus-
The erosion and deposition mean depths, volumes and net sed- pended sediment load over time. SSC can often be higher in the ris-
imentation depths were higher with EH than VR for all discharges ing limb of hydrograph than in falling limb due to hysteresis effects
(Fig. 7, right and Table 5). The t-test for the area weighted mean (Araujo et al., 2012). The effect of the hysteresis on calculated SSC
depths showed that the erosion depths were significantly different would likely cause higher deposition rates in the rising limb and
with EH and VR for all discharges except 799 m3/s, and the mean lower deposition in the falling limb. Depending on the local
deposition depths were different for all discharges (Table 5). In hydraulics, these differences could balance out to produce the
addition, the t-test revealed that the net sedimentation depths net sedimentation rates that we predicted with our rating curves.
were significantly different between EH and VR for all the dis- Hysteresis will change the exact pattern of deposition and erosion
charges except 799 and 1031 m3/s. through time at a given location (Figs. 9 and 10). Given that the
In general, the erosion depths decreased as the discharge in- hysteresis direction often varies between flood events in a river
creased with both EH and VR, and the opposite trend occurred for (Araujo et al., 2012), we did not consider hysteresis effects because
deposition depths. Most noticeably, both equations (EH and VR) pre- these should be highly dependent on the highly variable upstream
dicted a change from negative to positive net sedimentation at the sediment supply conditions.
same discharge (Fig. 8, right). Total erosion and deposition volumes The comparisons of area weighted mean depths, net sedimenta-
increased as discharge became higher with both EH and VR (Fig. 7, tion depths, and volumes of erosion and deposition for different
right and Table 5). Furthermore, the net sedimentation depths were SSCs and sediment transport formulas were based on the last time
negative up to the discharge 1375 m3/s and positive above this dis- step of the simulation. Our results clearly demonstrate that a spe-
charge for both EH and VR (Fig. 8, right). The erosion depth decreased cific location could change between erosion and deposition
with higher flows (Fig. 8, left) with both EH and VR, although the ero- through time (Fig. 5). Therefore, analysis of the last time step is
sion volume increased with higher discharges (Table 5). The lower more useful to estimate net sedimentation at the end of floods or
erosion depth was not caused by decreased sediment transport after a specific time period. However, to understand floodplain pro-
capacity with higher flows, but was because the erosion volume cesses as a function of discharge and SSC, it is equally important to
was spread over a progressively larger area as flow increased. analyze the results at different intermediate time steps.

Fig. 8. Area weighted mean Erosion or deposition (left) and net sedimentation depths (right) with Engelund and Hansen and Van Rijn sediment transport formulas for
different discharge conditions. The abbreviations refer as follows: EH or VR: simulated with Engelund and Hansen (1967) or Van Rijn (1984) sediment transport formula.
240 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243

Fig. 9. Spatially distributed erosion or deposition areas from high (left) and low (right) SSC in 1798 m3/s flow scenario. The color scale bar shows net erosion/deposition. (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 10. Spatially distributed erosion or deposition areas from EH (left) and VR (right) sediment transport formulation in 2103 m3/s flow scenario. The color scale bar shows
net erosion/deposition. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243 241

Furthermore, the differences in mean depths, net sedimentation confirmed that the floodplain was an erosional and depositional
depths and volumes were attributed to variations in sediment con- zone during low and high flows, respectively. Uncertainties in
centration and sediment transport formulas, and we ignore simu- the model (e.g., critical shear stress, cohesion) should not signifi-
lated changes in the flow hydraulics. However, feedback cantly change the general pattern of simulated differences in ero-
processes between hydraulics and floodplain topography occurs, sion and deposition parameters between different SSC or
especially where the floodplain topography was altered signifi- sediment transport equations (i.e., mean depth, net sedimentation,
cantly (e.g., Fig. 5). Thus, the differences in the simulated floodplain volume, etc.).
processes are likely a result of both changes in SSC and floodplain Our results may indicate that floodplain sedimentation is a dis-
hydraulics. Nevertheless, this demonstrates the importance of a continuous process and large flows with high sediment concentra-
sediment transport model that has the capability to simulate feed- tion are more effective in floodplain development than small flows
back processes between hydraulics and the topography as in our with low sediment concentration. Nevertheless, this conclusion
study. In addition, some deposition zones (Fig. 5) changed into ero- might not be applicable to other systems because the current study
sion zones in the later time steps, which illustrate the effect of re- site is backwater affected, and bordered by high levees, some of
suspension and changing hydraulics on sediment transport. This which were breached intentionally at few locations. During low
result might have crucial implications for vegetation and river res- flows, the flow velocities were high and caused erosion and sedi-
toration. For example, an area might be a depositional zone at the ment transport over the floodplain without settling, which resulted
last time step, but it could have previously experienced erosion in inefficient deposition on the low level floodplain (see, Middelko-
that would scour out existing vegetation. Therefore, while using op and Van der Perk, 1998). However, during high discharges the
sediment transport models for restoration purposes or to study backwater effect was likely more pronounced, which slowed the
vegetation dynamics, it is necessary to analyze the results through flow velocities near the bed and resulted in net deposition. Thus,
different time steps instead of the last time step alone. not all overbank flows contribute to sediment deposition on the
The area weighted average erosion and deposition depths were floodplain, but they (especially low flows) may erode the flood-
significantly different with high and low SSC and different sedi- plain, depending on the flow magnitude, SSC and backwater ef-
ment transport formulas, except in few discharge scenarios. Higher fects. The general pattern of increasing deposition with discharge
SSC caused higher deposition volumes than low sediment concen- will be independent of many of our assumed model parameters be-
trations (Table 4) and the Engelund and Hansen (1967) equation cause it was driven primarily by a backwater effect and the up-
simulated higher deposition volumes than the Van Rijn (1984) stream sediment supply. Other floodplains with significant tree
equation for all discharges (Table 5).The simulated mean erosion cover, a lack of levees and backwater effects, would likely experi-
(0.075–0.147 m) and deposition depths (0.016–0.115 m) were rel- ence a different trend in net sedimentation with discharge. A study
atively high. Generally, the rate of suspended sediment deposition in the Amazon basin further confirms that rapidly rising large flows
depends on the sediment supply from upstream and could be as are more effective than slowly rising medium flows for sediment
low as 0.5–4 mm/year (Middelkoop and Van der Perk, 1998) and deposition (Aalto et al., 2003).
0–10 mm/year (cf., Hardy et al., 2000). From a restoration perspective, larger flows than those cur-
The large changes in sedimentation with upstream sediment rently available are required for greater floodplain deposition and
supply demonstrate that accurate estimates of sediment supply maintenance of dynamic processes at our study site, given that
are needed to model long-term changes in floodplains. Without the current SSC and discharges are very low after Libby dam oper-
such information, estimates of vegetation stability, filling of wet- ation (Fig. 2, left). If levees are breached and the flow is small and
land areas and overall potential restoration success, will be inaccu- has low SSC, it has the potential to scour the floodplain instead of
rate. The higher deposition depths in our study might be caused by deposit sediment, at least for our simulated conditions. Further-
high sediment concentrations from upstream, but might also be re- more, our study shows (Figs. 9 and 10) that there is large spatial
sult of floodplain erosion (Figs. 9 and 10 with red circles) at the variability in erosion and deposition over the floodplain, hence
flood entry locations (at levee breaches) and subsequent down- mean values of erosion, deposition, and net sedimentation do not
stream deposition on the floodplain. Figs. 9 and 10 (marked with allow one to envision the complex floodplain processes that act lo-
red circles) illustrate clearly that substantial erosion occurred at cally. Diversity in riparian ecosystems largely depend on the spa-
the flood entry locations where simulated velocities were large tial and temporal diversity of physical processes such as flood
(Fig. 4). In addition, areas adjacent to the river channel were ero- frequency, magnitude, duration, and erosion and deposition (Bor-
sional zones. Low elevation areas (where simulated depths were nette and Amoros, 1996).
relatively deep) just downstream of erosional zones were usually The long duration simulation in this study indicated that the
depositional zones. Local sediment deposition may be driven by quantity of sediment erosion and deposition is governed by the
change in local topographic features and flow hydraulics that cause number of days the floodplain is inundated. Six months of simula-
upstream scour and then flow deceleration. One of the main rea- tion resulted in higher erosion and deposition volumes than those
sons for the considerably high erosion depths (0.147 m in Table caused by the normal simulation duration (5 days). However, the
4) in this study may be from the fine sediment (<0.25 mm) distri- area weighted mean deposition depth was less during this long
bution on the floodplain (half of the grain sizes are less than run than in the normal simulation. The differences in mean depo-
0.28 mm) and the assumption of cohesionless sediment. The use sition and net sedimentation depths between simulations were
of spatially varied grain size distributions may help to predict ero- statistically significant for the low discharge (799 m3/s) but not
sion and deposition zones more accurately. In addition, the channel for the high discharge (1798 m3/s). This result demonstrates that
bed and floodplain sediment distributions were represented by six the duration of flood simulation may be important for inferred
grain size classes (Fig. 2, right) that might have considerable influ- changes in floodplain processes and therefore careful consideration
ence on the simulated floodplain processes (see James et al., 2006). is needed when modeling hydrographs. Simulations based on only
On average, floodplain processes were dominated by erosion peak flow will be highly inaccurate.
during low flows (799, 1031, 1375 m3/s) and deposition at high In comparing sediment transport rates from different formulas,
flows (1798 m3/s). This likely occurred because, sediment supplies it is important to note that the Engelund and Hansen equation
were low during low flows and more sediment was supplied to the (1967) is based on total load and the Van Rijn equation (1984) is
floodplain during high flows. The comparison between the Engel- based on bed and suspended load. The Engelund and Hansen equa-
und and Hansen (1967) and Van Rijn (1984) equations further tion (1967) simulated higher transport rates than the Van Rijn
242 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243

equation (1984) because the former considers overall shear stress Acknowledgements
instead of just the grain shear stress considered in the latter. This
result is consistent with Abdel-Fattah et al. (2004), who found that The study is part of a Kootenai River project. The project is
the Van Rijn (1984) and Engelund and Hansen (1967) equations sponsored by the Kootenai Tribe of Idaho (KTOI) and is funded
under- and over-predicted the measured transport rates, respec- by the Bonneville Power Administration (BPA). We would like to
tively. Therefore, the choice sediment transport equations can be thank and appreciate Scott Soults and Norm Merz for their project
very important in determining the magnitude of erosion or deposi- management and co-operation in the study. We also acknowledge
tion. A number of different transport equations should be used for DHI (Danish Hydraulic Institute) for providing MIKE21C software
given study, to obtain a range of possible sedimentation rates. license to the University of Idaho and providing technical assis-
tance. We would like to thank editor and two anonymous review-
ers for constructive comments, which helped to improve the
5. Conclusions and recommendations manuscript significantly.

Our study was focused on quantifying the effect of sediment

supply and transport formulas on the simulation of floodplain pro-
References
cesses using a coupled spatially distributed 2D HD and sediment
transport model. Erosion or deposition depth, net sedimentation Aalto, R. et al., 2003. Episodic sediment accumulation on Amazonian flood plains
depth and total volume were quantified based on the last time step influenced by El Nino/Southern Oscillation. Nature 425, 493–497.
of the simulation period as an indicator of floodplain processes. Abdel-Fattah, S., Amin, A., Van Rijn, L.C., 2004. Sand transport in Nile River, Egypt. J.
Hydraul. Eng. 130 (6), 488–500.
The HD model was calibrated based on previously developed HD Ackers, P., White, W.R., 1973. Sediment transport: new approach and analysis. J.
model results, while the sediment transport model was validated Hydraul. Eng. 99 (11), 2041–2060.
by comparing simulated SSC with measured concentrations at a Acrement, G.J., Schneider, V.R., 1989. Guide for selecting Manning’s roughness
coefficients for natural channels and floodplains. USGS, Water-Supply Paper
gage station. Floodplain processes were simulated as a function 2339, 1–44.
of high and low SSC and two sediment transport equations, Van Alkema, D., Middelkoop, H., 2005. The influence of floodplain compartmentalization
Rijn (1984) and Engelund and Hansen (1967), using five discharge on flood risk within the Rhine–Meuse Delta. Nat. Hazard. 36, 125–145.
Araujo, H.A., Cooper, A.B., Hassan, M.A., Venditti, J., 2012. Estimating suspended
scenarios. A t-test was used to analyze if the simulated floodplain sediment concentrations in areas with limited hydrological data using a mixed-
processes using various SSCs and transport formulations were sig- effects model. Hydrol. Process., pp. 1–11.
nificantly different. Ayres Associates, 2002. Two-Dimensional Hydraulic Modeling of the Upper
Sacramento River, rm 194.0 to rm 202.0 Including Riparian Restoration,
The RMSE associated with calibration and validation of the HD
Setback Levee, and East Levee Removal. Glenn and Butee Counties, California,
and sediment transport model were relatively low and acceptable. The Nature Conservancy, Chico, California.
The study showed that a specific location could be an erosional or Barton, G.J., 2004. Characterization of Channel Substrate, and Changes in
depositional zone at different time steps of the simulation. The net Suspended-Sediment Transport and Channel Geometry in White Sturgeon
Spawning Habitat in the Kootenai River near Bonners Ferry, Idaho, Following
sedimentation depths between high and low SSC were significantly the Closure of Libby Dam U.S. Geological Survey, Reston, Virginia.
different for all the discharges. The long duration simulation re- Barton, G.J., McDonald, R.R., Nelson, J.M., Dinehart, R.L., 2005. Simulation of Flow
sulted in higher erosion and deposition volumes compared to the and Sediment Mobility Using a Multidimensional Flow Model for the White
Sturgeon Critical-Habitat Reach, Kootenai River near Bonners Ferry, Idaho, U.S.
normal simulation duration. The deposition and net sedimentation Geological Survey, Reston, Virginia.
depths were significantly different between normal and long dura- Benjankar, R., 2009. Quantification of Reservoir Operation-based Losses to
tion simulations for low discharges but not for high discharges. The Floodplain Physical Processes and Impact on the Floodplain Vegetation at the
Kootenai River, USA. Unpublished Ph.D. Thesis, University of Idaho, Moscow,
erosion and deposition depths and volumes were higher with EH Idaho, 308pp.
than VR for all discharges. The weighted mean depths with EH Benjankar, R., Egger, G., Jorde, K., Goodwin, P., Glenn, N., 2011. Dynamic floodplain
and VR were statistically different for all discharges. vegetation model development for the Kootenai River, USA. J. Environ. Manage.
92, 3058–3070.
From a restoration perspective, it would be important to ana- Berenbrock, C., 2006. Simulation of Hydraulic Characteristics in the White Sturgeon
lyze the results at intermediate time steps rather than just the last Spawning Habitat of the Kootenai River Near Bonners Ferry, Idaho – A
time step. The simulated rate of erosion and deposition were com- Supplement to Scientific Investigations Report 2005–5110. U.S. Geological
Survey, Reston, Virginia.
paratively high compared to other studies (e.g., Middelkoop and
Berenbrock, C., Bennett, J.P., 2005. Simulation of Flow and Sediment Transport in the
Van der Perk, 1998). This result may be attributed to significant White Sturgeon Spawning Habitat of the Kootenai River near Bonners Ferry,
floodplain erosion at the flood entry locations on the floodplain. Idaho: U.S. Geological Survey Scientific Investigations: Report 2005-5173.
The study showed that large flows with high SSC were more effec- Bornette, G., Amoros, C., 1996. Disturbance regimes and vegetation dynamics: role
of floods in riverine wetlands. Vegetation Sci. 7, 615–622.
tive for floodplain deposition at our study site and this is consistent Buffington, J.M., Montgomery, D.R., 1997. A systematic analysis of eight decades of
with other studies (e.g., Aalto et al., 2003). incipient motion studies, with special reference to gravel-bedded rivers. Water
A coupled HD and sediment transport model is able to simulate Resour. Res. 33 (8), 1993–2029.
Burke, M., 2006. Linking Hydropower Operation to Modified Fluvial Processes
floodplain processes as a function of grain size distribution, SSC, Downstream of Libby Dam, Kootenai River, USA and Canada. Unpublished
feedback between hydraulics and topography, and re-suspension. Master Thesis, University of Idaho, Moscow, Idaho.
Floodplain deposition is a discontinuous function of river discharge Buttner, O., Otte-Witte, K., Kruger, F., Meon, G., Rode, M., 2006. Numerical modelling
of floodplain hydraulics and suspended sediment transport and deposition at
and varies spatially and temporally over the floodplain. Therefore, the event scale in the middle river Elbe, Germany. Acta Hydrochim. Hydrobiol.
a coupled HD and sediment transport model that accounts for 34, 265–278.
feedback processes between topography and hydraulics should Camenen, B., Larroude, P., 2003. Comparison of sediment transport formulae for the
coastal environment. Coast. Eng. 48, 111–132.
be given first preference for future floodplain restoration projects. Carollo, F.G., Ferro, V., Termini, D., 2002. Flow velocity measurements in vegetated
Furthermore, the results from this study provide valuable informa- channels. J. Hydraul. Eng. 128 (7), 664–673.
tion for the restoration of natural floodplain processes and riparian Carrolla, R.W.H., Warwick, J.J., James, A.I., Miller, J.R., 2004. Modeling erosion and
overbank deposition during extreme flood conditions on the Carson River,
ecosystems on the Kootenai River floodplain. Larger discharges
Nevada. J. Hydrol. 297 (1–21).
with higher SSC resulted in the most efficient floodplain deposi- Chow, V.T., 1959. Open Channel Hydraulics. McGraw Hill, New York.
tion, which may be used as a restoration measure if further test Danish Hydraulics Institute (DHI), 2008. MIKE21C Users Manual: Curvelinear Model
of model sensitivity are performed. The current model might be for River, Morphology, p. 147.
Dunne, T., Mertes, L.A.K., Meade, R.H., Richey, J.E., Forsberg, B.R., 1998. Exchanges of
used to analyze the impact on floodplain processes from altered sediment between the flood plain and channel of the Amazon River in Brazil.
SSC due to Libby Dam operation. Geol. Soc. Am. Bull. 110 (4), 450–467.
 R. Benjankar, E.M. Yager / Journal of Hydrology 450–451 (2012) 230–243 243

Edwards, T.K., Glysson, G.D., 1999. Field Methods for Measurement of Fluvial Meyer-Peter, E., Muller, R., 1948. Formulas for bed-load transport. In: Proceedings of
Sediment: U.S. Geological Survey Techniques of Water-Resources the International Association for Hydraulic Structures Research, Second
Investigations. Book 3, USGS, Reston, Virginia (Chapter C2). Congress, pp. 39–64.
Engelund, F., Hansen, E., 1967. A Monograph on Sediment Transport in Alluvial Middelkoop, H., Van der Perk, M., 1998. Modelling spatial patterns of overbank
Streams. Teknisk Forlag, Copenhagen, Denmark. sedimentation on embanked floodplains. Geogr. Ann. 80 (2), 95–109.
Gran, K., Paola, C., 2001. Riparian vegetation controls on braided stream dynamics. Nicholas, A.P., Walling, D.E., 1997. Investigating spatial patterns of medium-term
Water Resour. Res. 37 (12), 32375–32830. overbank sedimentation on floodplains: a combined numerical modelling and
Gyssels, G., Poesen, J., Bochet, E., Li, Y., 2005. Impact of plant roots on the resistance radiocaesium-based approach. Geomorphology 19, 133–150.
of soils to erosion by water: a review. Prog. Phys. Geogr. 29 (2), 189–217. Nicholas, A.P., Walling, D.E., 1998. Numerical modelling of floodplain hydraulics and
Habersack, H.M., 1998. Numerical sediment transport models-theoretical and suspended sediment transport and deposition. Hydrol. Process. 12, 1339–1355.
practical aspects. In: Modelling Soil Erosion, Sediment Transport and Closely Noe, G.B., Hupp, C.R., 2009. Retention of riverine sediment and nutrient loads by
Related Hydrological Processes, Vienna, Austria, pp. 299–308. coastal plain floodplains. Ecosystems 12, 728–746.
Hardy, R.J., Bates, P.D., Anderson, M.G., 2000. Modelling suspended sediment Paquier, A., Mignot, E., 2003. Use of 2-D models to calculate flood water levels:
deposition on a fluvial floodplain using a two-dimensional dynamic finite calibration and sensitivity analysis. In: Ganoulis, J., Prinos, P. (Eds.), XXX IAHR
element model. J. Hydrol. 229, 202–218. Congress, Thessaloniki, Greece.
He, Q., Walling, D.E., 1996. Use of fallout Pb-210 measurements to investigate Sulaiman, M., Tsutsumi, Daizo, Fujita, M., Hayashi, K., 2007. Classification of grain
longer-term rates and patterns of overbank sediment deposition on the size distribution curves of bed material and the porosity. Ann. Disas. Prev. Res.
floodplain of lowland rivers. Earth Surf. Proc. Land. 21, 141–154. Instit. Kyoto Univ. 50B, 615–622.
Heimann, D.C., 2001. Numerical Simulation of Streamflow Distribution, Sediment Tetratech, 2004. Kootenai River Geomorphic Assessment – Final Report. U.S. Army
Transport, and Sediment Deposition along Long Branch Creek in Northeast Corps of Engineers, Seattle District, Seattle, WA.
Missouri. USGS, Report No. WRIRO1-4269. Van Rijn, L.C., 1984. Sediment transport, part I: bed load transport. J. Hydraul. Eng.
Helsel, D.R. and Hirsch, R.M., 2002. Statistical Methods in Water Resources 110 (10), 1431–1456.
Techniques of Water-Resources Investigations, Book 4, Chapter A3, U.S. Walling, D.E., Owens, P.N., 2003. The role of overbank floodplain sedimentation in
Geological Survey. catchment contaminant budgets. Hydrobiologia 494, 83–91.
Hesselink, A., Stelling, Guss.S., Kwadijk, J.C.J., Middelkoop, H., 2003. Inundation of a Walling, D.E., Owens, P.N., Leeks, G.J.L., 1998. The role of channel and floodplain
Dutch River polder, sensitivity analysis of a physically based inundation model storage in the suspended sediment budget of the River Ouse, Yorkshire, UK.
using historic data. Water Resour. Res. 39 (9), 3–14. Geomorphology 22 (225–242).
Hoffman, G. et al., 2002. Instream Flows Incremental Method for Kootenai River. Werner, M.G.F., Hunter, N.M., Bates, P.D., 2005. Identifiability of distributed
Project No. 1994–00500. Bonneville Power Administration. floodplain roughness values in flood extent estimation. J. Hydrol. 314, 139–157.
James, S.C., Shrestha, P.L., Roberts, J.D., 2006. Modeling noncohesive sediment Wilcock, P.R., McArdell, B.W., 1993. Surface-based fractional transport rates:
transport using multiple sediment size classes. J. Coastal Res. 22 (5), 1125– mobilization thresholds and partial transport of a sand–gravel sediment.
1132. Water Resour. Res. 29 (4), 1297–1312.
Lumborg, U., Windelin, A., 2003. Hydrography and cohesive sediment modelling: Yang, C.T., 1996. Sediment Transport Theory and Practice. McGraw-Hill, USA, 396pp.
application to the Romo Dyb tidal area. J. Mar. Syst. 38, 287–303. Yang, C.T., 2006. Erosion and Sedimentation Manual: Chapter 4—Cohesive Sediment
Mahoney, J.M., Rood, S.B., 1998. Stream flow requirements for cottonwood seedling Transport. U.S. Department of the Interior Bureau of Reclamation, Denver,
recruitment – an integrated model. Wetlands 18 (4), 634–645. Colorado.