w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Hydrograph-based approach to modeling bacterial

fate and transport in rivers

Bhuban Ghimire, Zhiqiang Deng*

Dept. of Civil and Environmental Engineering, Louisiana State University, Baton Rouge, LA 70803-6405, USA

article info abstract

Article history: A new approach, called hydrograph-based approach, is proposed for predicting bacterial
Received 19 March 2012 concentrations in rivers. The new approach is relatively simple and efficient in terms of
Received in revised form data requirements. It uses widely available hydrographs as the main input data for esti-
25 July 2012 mating flow and sediment transport parameters responsible for bacterial transport under
Accepted 28 November 2012 varying flow conditions. The major component of the hydrograph-based approach is a new
Available online 8 December 2012 model, called VARTBacT model which is an extension of the Variable Residence Time
(VART) model by including effects of unsteady flow, sediment transport, and bacterial
Keywords: decay/growth processes on bacterial transport and fate in rivers. The applicability of the
Fecal bacteria new hydrograph-based approach is demonstrated through three case studies, each with
Hydrograph distinct sediment and flow conditions: (1) steady low flow without sediment transport, (2)
Modeling flood events with significant sediment transport due to watershed inputs, and (3) sediment
Flood resuspension from the streambed. While the sediment resuspension from streambed may
Sediment resuspension be an important process for bacterial transport during high flows, results from this study
Hyporheic exchange indicate that the most important mechanism responsible for bacterial transport in streams
is watershed loading during flood events and hyporheic exchange during low flow periods.
ª 2012 Elsevier Ltd. All rights reserved.

1. Introduction coli levels during rising hydrographs (Muirhead et al., 2004).

Bacterial concentrations in rivers have been observed to be
often much higher during storm events than during low flows.
In fact, storm events export major part of the annual load of
Escherichia coli (E. coli, fecal indicator bacteria) reaching as high
as 98% (Chu et al., 2011; McKergow and Davies-Colley, 2010) in
some rivers. Wilkinson et al. (1995) observed an increase in
fecal coliform concentrations by 25 times during an artificial
flood. Peak concentrations of bacteria have been found to
occur usually during rising limb of a storm hydrograph
(Davies-Colley et al., 1994; Jamieson et al., 2005b) well ahead of
the discharge peak and close to the line of maximum flow
acceleration (McKergow and Davies-Colley, 2010; Nagels et al.,
2002). Artificial flood experiments without any watershed
input of bacterial loads also showed a significant increase in E.
The bacterial concentration was often increased after the bed
stress reached a certain critical value (Jamieson et al., 2005b)
suggesting a close relationship between entrainment of
riverbed sediment and bacterial concentration in the water
column. As fecal coliforms are often concentrated near the
sedimentewater interface (SWI) and are mostly associated
with fine particulates of low settling velocity (Wilkinson et al.,
1995), accurate assessment of the entrainment of fine sedi-
ment from the channel bed is important to modeling bacterial
transport, especially during high flow events.
Various numerical models have been developed to simu-
late bacterial transport and fate in rivers by considering the
sediment and water column interaction. Jamieson et al.
(2005a) studied controlling processes for fate and transport
of enteric bacteria in alluvial streams by combining field

* Corresponding author. Tel.: þ1 225 578 6850; fax: þ1 225 578 8652.
E-mail address: zdeng@lsu.edu (Z. Deng).
0043-1354/$ e see front matter ª 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.watres.2012.11.051
1330 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3
experiments and mathematical modeling. A strain of E. coli
was mixed with stream water and bed sediment, and loaded
in streams to monitor the transport of sediment and E. coli at
downstream locations. As the experiment was carried out
during low and steady flow, no entrainment was included in
their model. Bai and Lung (2005) added fecal bacteria transport
component to the Environmental Fluid Dynamics Code model
to study the impact of sediment transport process on fecal
transport in rivers. The flux of fecal bacteria was linked with
sediment dynamics across SWI. Hipsey et al. (2006) developed
a model within an aquatic ecology model Computational
Aquatic Ecosystem Dynamics Model (CAEDYM) to include
sedimentation and resuspension processes in addition to
other processes such as growth, mortality and predation.
Although more generic, this model is more suitable for
simulating microbial pollution in slowly flowing water bodies
such as reservoirs and estuaries than for rivers during short
storm events. Firstly, it has a relatively high input require-
ment. Secondly, the storm events in the rivers are often so
short that some of the processes associated with the microbial
fate are not important for overall simulation results. More-
over, the model ignores the dispersion process, which is one of
the important processes affecting the pollutant transport in
rivers. Rehmann and Soupir (2009) quantified the effect of
interaction between sediment and water column for microbial
concentration using one dimensional steady state model of
transport in a river. Transport equations were derived for
depth averaged microbial concentrations in the water column
and sediment separately, and solved. The longitudinal
dispersion process was ignored in their model. Cho et al.
(2010a,b) followed the approach by Steets and Holden (2003)
for bacterial transport by incorporating the resuspension
and sedimentation terms into a net resuspension term in their
models. Both used a simple formula for bed shear stress
calculated based on flow velocity using a constant friction
coefficient. Unlike the other two, Cho et al. (2010b) did not use
sediment storage model but determined the bacterial
concentration in bed sediment from model calibration.
Recently, Gao et al. (2011) developed a numerical model based
on DIVAST (Depth Integrated Velocities and Solute Transport
Model) with a focus on predicting the effect of sediment fluxes
on fecal bacteria levels in water column. The model was
applied to several idealized case studies and also to an artifi-
cial flood study. Finally, Wilkinson et al. (2011) modeled E. coli
pulses in Motueka River, New Zealand, using records of E. coli
concentration during several storm events in 2003e2004.
Their model domain consists of main river reach and sub-
catchments with three layers: riparian land, river reach
water column and river reach channel storage. The model
includes sediment resuspension and deposition processes
along with a bacterial die-off term but does not use advec-
tionedispersion equation and is very much site specific.
Despite efforts to include all processes in bacterial trans-
port modeling, the transient storage effect was mostly
ignored. It is well observed that natural streams possess
permeable banks and bed sediment which create transient
storage zones and thereby generate significant mass exchange
between surface and subsurface waters due to the hyporheic
exchange (Deng and Jung, 2009). Grant et al. (2011) measured
the flux of fecal bacteria across the SWI in a small effluent
stream with a turbulent flow and found that the hyporheic
exchange controls the transport of bacteria across the SWI in
turbulent streams. By combining dual tracer test results and
the transient storage model (Runkel, 1998), Shen et al. (2008)
showed that a bacteriophage P22 can be successfully used as
a tracer in complex surface water environments. When the
concentration of free E. coli is high in the water column during
low flows, the mass exchange between storage zones and the
main channel is substantial. Thereby, the transient storage is
an important mechanism controlling bacterial transport and
should be included in models for description of bacterial
transport and fate in rivers.
The primary objective of this study is to present a simple
yet effective approach to modeling bacterial fate and trans-
port in natural streams. The new modeling approach should
be applicable to both low flow and high flow (especially flood
flow) conditions. To that end, the Variable Residence Time
based (VART) model (Deng and Jung, 2009) is extended in this
study to simulate bacterial fate and transport by taking into
account: i) unsteady flow using a hydrograph-based approach,
ii) effect of sediment transport on bacterial concentrations,
and iii) bacterial decay/growth processes in addition to
advection, dispersion, and hyporheic exchange processes
included in the original VART model. The extended VART
model is applied to simulate bacterial transport in natural
streams under different sediment and flow conditions,
ranging from steady low flow without sediment transport
to flood events with significant sediment transport due
to watershed inputs and sediment resuspension from
streambed.
2. Model development
2.1. Conceptual model
Major processes controlling the fate and transport of bacteria
in streams include advection, dispersion, transient storage
(including hyporheic exchange), decay/growth, and resus-
pension/settling of attached fraction. While the advection and
dispersion processes are generally included in mass transport
models, other processes are selectively included. Two con-
trasting flow conditions controlling bacterial fate and trans-
port are often encountered in natural streams: 1) low flow e
when a stream has low flow discharge, shallow depth, clearer
water, higher residence time, and clear weather with
sunshine, and 2) high flow e when a stream has high flow
discharge, deep and turbid water, lower residence time and
generally cloudy weather with less sunshine. Accordingly,
during the low flow there is a likelihood of higher inactivation
rates due to longer residence time, clearer water and more
sunlight. The exchange due to transient storage may also play
an important role during low flow as the flow in the main
channel is relatively small and slow. On the other hand,
resuspension of sediment associated bacteria from the
streambed, particularly during rising flows, and subsequent
deposition during receding flows may play a dominant role
during storm events (Wilkinson et al., 2011). Due to shorter
residence time and favorable environment for survival of
bacteria in water column, the solar inactivation plays less
 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3 1331

important role during such storm events. The transient (1973) formula gives the depth averaged equilibrium sedi-
storage effect is insignificant during storm events due to the ment concentration Se (kg/m3). Therefore, the sediment
sediment resuspension-induced destruction of the storage concentration at the reference level (z ¼ 0.05 h) was deter-
zones in the upper riverbed and banks. mined later using Abad and Garcı́a (2006) formulation for
The simulation of bacterial transport during both low flow Einstein Integral INT1.
and high flow events requires the incorporation of sediment The depth averaged net sediment flux Fr can be calculated
resuspension/deposition processes and bacterial decay/ following the sediment carrying capacity approach (Gao et al.,
growth into the VART model. The processes represented in 2011; RuiJie et al., 2009):
this model are conceptually depicted in Fig. 1.
ws ðSe  qSÞ
Fr ¼ E  D ¼ (2)
INT1
2.2. Hydrograph-based approach to modeling sediment
resuspension where E ¼ sediment erosion rate (kg/m2s); D ¼ deposition rate
(kg/m2s); ws ¼ particle settling velocity (m/s); S ¼ depth-
Flow parameters, including friction slope, flow velocity, and averaged sediment concentration (kg/m3); Se ¼ equilibrium
bed shear stress, are determined using a hydrograph-based sediment concentration (kg/m3); and q ¼ Se/S is a sediment
method (Ghimire and Deng, 2011). The method requires the concentration profile factor (Gao et al., 2011), q ¼ 1 at the
use of flow hydrograph as basic input data for computation of equilibrium sediment transport state and qs1 in a non-
the discharge gradient ðvQ=vtÞ term which is again used to equilibrium state. The net vertical displacement of SWI, vSWI
determine bed shear stress for unsteady flow. More specifi- is then computed following the approach by Motta et al. (2010)
cally, the bed shear stress, which is considered to be the most as
important driving factor for sediment entrainment, is deter-
Fr
mined as vSWI ¼ (3)
ð1  pÞrs


vQ
sb ¼ rgRh So þ a (1) where vSWI ¼ velocity of SWI (m/s), p ¼ porosity of bed sedi-
vt
ment (), and rs ¼ specific density of sediment grains. The
where sb ¼ shear stress exerted by flow on the bed (N/m2), settling velocity may be obtained using a relation from Wu
So ¼ channel bed slope (), Rh ¼ hydraulic radius (m) which and Wang (2006) for naturally worn particles with a Corey
may be replaced with water depth h (m) in natural rivers, shape factor (CSF) of 0.7,
Q ¼ flow discharge (m3/s), t ¼ time (s), and a is a parameter sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

1=n0 !n0
relating the slope of the flood hydrograph with sb (m3s2). In M0 n 1 4N0 3 1
ws ¼ 0 þ D  (4)
case of natural flood events a ¼ 1=ðBC0 Þ, where B ¼ channel Nd 4 3M02  2
width (m) and C0 ¼ average wave celerity (m/s). The parameter
C0 is normally determined using friction law: C0 ¼ bU, where where the coefficients M0 ¼ 33.94, N0 ¼ 0.98, and n0 ¼ 1.33. The
b is kinematic wave parameter. For a wide rectangular kinematic viscosity n was calculated as 1.79  106/
channel (B >> h), Manning’s law gives the maximum value of (1 þ 0.3368T þ 0.00021T2) m2/s, where T is in degree Celsius
b ¼ 1.67. Flow parameters, such as flow velocity U (m/s) and (Garcı́a, 2008). The porosity of the sediment was also esti-
friction slope Sf (), were determined following Ghimire and mated as per Wu and Wang (2006).
Deng (2011). In case of cohesive sediment transport, the net upward
The sediment transport during unsteady flows can be resuspension flux may be written as
calculated using the hydrograph-based flow parameters 8

>
> sb
(velocity U and friction slope Sf). For non-cohesive sediment, >
> E¼M 1 ; sb  scr;e
>
< scr;e
Ackers and White (1973) formula is used. Unlike formulas
Fr ¼ 0;
 scr;e > sb > scr;d (5)
giving entrainment rates near sedimentewater interface >
> sb
>
>
directly (e.g., Smith and McLean, 1977), Ackers and White : D ¼ ws S 1  s
> ; sb  scr;d
cr;d

where M ¼ empirical erosion rate constant (kg/m2s),

scr,e ¼ critical bed shear stress for erosion (Pa), and
scr,d ¼ critical bed shear stress for deposition (Pa). Settling
velocity ws was determined using Eq. (4) for CSF ¼ 0.4. The net
vertical displacement of SWI can be obtained by dividing Fr by
the dry density of the bed sediment, as shown in Eq. (3). Herein
commonly used AriathuriePartheniad relation for surface
erosion of cohesive sediment (Ariathurai, 1974) and the sedi-
ment deposition formula by Krone (1962) were used. A linear
erosion formulation with a constant critical shear stress was
adopted for simplicity. A more realistic formulation for
natural rivers would have a depth-varying critical stress either
determined experimentally or indirectly using measured dry
Fig. 1 e Conceptual diagram of key processes in VARTBacT density of bed sediment (Sanford and Maa, 2001). In rivers
model. with fine-grained sediment bed, for example, the exponential
1332 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3
function given by Parchure and Mehta (1985) with depth-
varying critical stress may be used to model the erosion of
a soft top layer. No matter which formula is chosen for
modeling, accurate determination of site-specific sediment
parameters, such as critical shear stress, is very important
(Lick, 2008; Sanford and Maa, 2001).
2.3. Variable residence time based bacterial transport
(VARTBacT) model
The Variable Residence Time (VART) model (Deng and Jung,
2009) was originally developed for longitudinal dispersion
and transport of solute in natural streams. The model is based
on the concept of varying residence time and is able to
produce various types of breakthrough curves commonly
observed in natural rivers due to the mass flux across the
sedimentewater interface (Deng and Jung, 2009; Deng et al.,
2010).
In order to incorporate the effect of sediment on the
bacterial fate and transport, the original VART model is
modified according to the conceptual model in Fig. 1 by adding
appropriate terms for sediment resuspension, deposition, and
bacterial die-off/growth in water column and storage zones.
The modified VART model for bacterial fate and transport,
herein after called VARTBacT model, is given by the following
equations:
vC vC v2 C As 1 E D fp C
þU ¼ Ks 2 þ ðCs  CÞ þ Cb   kC
vt vx vx A Tv h h S
vCs 1
¼ ðC  Cs Þ  ks Cs
vt Tv
 studies with differing flow and sediment conditions were
Tmin for t  Tmin
Tv ¼ ðTmin > 0Þ
t for t  Tmin
 riophage and a conservative tracer. No sediment resuspension
0 for t  Tmin
ts ¼
t  Tmin for t  Tmin
where C, Cs ¼ bacterial concentration in water column and
transient storage zones (CFU/100 mL), respectively;
Cb ¼ bacterial concentration in the sediment bed (CFU/0.1 g);
U ¼ cross-sectionally averaged flow velocity (m/s); h ¼ depth of
water column (m); A, As ¼ cross-sectional flow areas of main
channel and transient storage zones (m2), respectively;
t ¼ time (s); Tv ¼ actual varying residence time of bacteria in
storage zones (s); Ks ¼ longitudinal Fickian dispersion coeffi-
cient excluding the transient storage effect (m2/s); ts ¼ time
since the release of bacteria from storage zones to the main
stream (s); Tmin ¼ minimum mean residence time for bacteria
to travel through the advection-dominated storage zone (s); k,
ks ¼ bacterial decay/growth rates in the water column and
storage zones (1/day), respectively; fp ¼ fraction of bacteria
attached to suspended sediments (); and fp  C ¼ volume-
specific concentration on the particles (CFU/100 mL). As is
determined as: As ¼ Aadv þ Adif; where Aadv, Adif ¼ areas of
advection-dominated transient storage zone and effective
diffusion-dominated transient storage zone (m2), respectively,
and Adif ¼ 4pDsts, where Ds ¼ effective diffusion coefficient in
storage zone (m2/s). The estimated value of fp in the literature
is highly variable, ranging from 0.1 (Liu et al., 2006) for Lake
Michigan to 0.9 (Steets and Holden, 2003) in a coastal lagoon.
In this study the approach by Gao et al. (2011) was followed by
using a linear partition coefficient assuming instantaneous
equilibrium:
Kd S
fp ¼ (10)
1 þ Kd S
where Kd ¼ linear partition coefficient (m3/kg). This method
allows the attached fraction of bacteria to vary depending
upon S. The third and fourth terms on the right side of Eq. (6)
represent bacterial source/sink terms due to sediment resus-
pension and deposition. The parameters E and D represent
erosion and deposition rates (kg/m2s) which are calculated
according to Eqs. (2) and (5) for non-cohesive and cohesive
sediments, respectively. In contrast with other bacterial
transport models which account for sedimentewater inter-
action (e.g., Bai and Lung, 2005; Gao et al., 2011; Rehmann and
Soupir, 2009), the new VARTBacT model considers the inter-
action with the transient storage zone, reflecting the exchange
between surface stream water and subsurface sediment pore
water. As there is no separate equation for the concentration
of bacteria in the sediment bed, the concentration Cb has to be
estimated for computing the resuspension flux of bacteria into
the water column. A split-operator method used for the
numerical solution of the VARTBacT model is provided in the
Appendix.
(6)
3. Model applications
To test and demonstrate the use of the VARTBacT model for
(7)
bacterial transport and fate in natural streams, three case
(8) considered. First, the model was applied to a river with low
flow, where a dual tracer study was conducted using bacte-
(9) and deposition processes were included in the first case as the
tracer test was carried out during a low and relatively steady
flow. The second case involved field experiments on resus-
pension of the bed sediment seeded with bacteria from the
upstream boundary of the study reach during artificial storm
events. The third case study focused on simulating bacterial
transport in a river during natural storm/flood events, char-
acterized with significant sediment transport from the
watershed and sediment resuspension from streambed.
Although actual flow and concentration data were used, some
important assumptions were made to test the performance of
the model during high flows.
3.1. Bacterial transport in low river flow
Shen et al. (2008) carried out a dual tracer study to evaluate the
performance of bacteriophage P22 relative to a conservative
tracer Rhodamine WT (RWT) in a 40 km reach of Grand River,
MI, USA. The test was done on May 8, 2006 during a low flow
period near the end of the recession limb of hydrograph. The
sampling was conducted for both tracers simultaneously at
locations including 4.56, 13.69, and 28.38 km downstream of
the tracer release station. The water samples were collected
from just below the surface, assuming complete vertical
 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3 1333

mixing of tracers. To accommodate the lateral variability in

Table 2 e Estimated tracer specific parameters, RMSE,
tracer concentration, sampling and flow measurement were and r2.
done at multiple locations (left, right, and center) on each site
Tracer Reach k, ks M* RMSE r2
and flow-weighted concentrations were computed. The study
type (1/day)
used the conventional transient storage model (Runkel, 1998)
to simulate the solute transport. RWT 1 0.0 1.14 0.46 0.83
2 0.0 1.08 0.19 0.98
The data from the above-mentioned study were first used
3 0.0 0.64 0.13 0.98
to test the performance of the VART model, which is the basic
P22 1 0.0 0.31 1.06 0.72
model of VARTBacT, for simulating P22 as well as RWT tracer 2 1.72 0.27 0.42 0.94
breakthrough curves. With lateral inflow and reaction terms, 3 0.86 0.18 0.34 0.94
the VARTBacT model Eq. (6) can be expressed as

vC vC v 2 C As 1 qL
þU ¼ Ks 2 þ ðCs  CÞ þ ðCL  CÞ  kC
vt vx vx A Tv A
where CL ¼ solute concentration in lateral inflow, qL ¼ surface
lateral inflow rate per unit channel length. All other symbols
have been defined earlier.
The primary flow and channel geometry parameters were
taken from Shen et al. (2008) and others were either calculated
or estimated. The flow velocity at each site was determined
using the distance from the injection site and the time to peak
concentration at each sampling site. Lateral flows were
included for reaches 2 and 3.
For RWT tracer, Eqs. (7)e(9) and (11) were solved on a reach
basis. Decay rates k and ks were set to zero, assuming RWT to
be a conservative tracer. Four VART model parameters Ks,
Aadv/A, Ds/A, and Tmin (Table 1) were determined based on best
fit with the observed data using a trial and error approach for
parameter estimation. Goodness of fit of each breakthrough
curve was first assessed using visual inspection of the plot on
logarithmic scale. Then the root mean squared error (RMSE)
(Bard, 1974) was used for determining the best fit and model
parameters, following the approach by Deng and Jung (2009).
In addition, the coefficient of determination r2, which is the
square of the Pearson correlation coefficient r, was also
computed to evaluate the strength and direction of relation-
ship between the modeled and observed values. The same
parameter values listed in Table 1 were then used to simulate
P22 tracer breakthrough curve. Following the approach by
Shen et al. (2008), the inactivation rate at main channel (k) and
storage zones (ks) were kept same, despite the fact that the
inactivation rate in the storage zone is generally low. Bacterial
loss rate due to sedimentation was expected to be relatively
small and, therefore, assumed to be lumped into the single
decay rate. The estimated inactivation rates (k, ks), recovery
ratio M*, which is the ratio of tracer mass recovered to the
mass injected, RMSE, and r are given in Table 2. Comparisons
between observed and simulated concentrations of RWT and
P22 are given in Fig. 2 and Fig. 3, respectively.
(11) Sensitivity analysis was done to examine the relative effect
of change in input parameters on the model results. The
sensitivity was assessed with respect to RMSE e a primary
performance measure for evaluating the best fit in this study.
The model input parameters were systematically changed by
50% of initial value, one parameter at a time, and the model
was run to determine the RMSE values. The magnitude of
change in RMSE values of simulated concentrations relative to
calibrated values shows the sensitivity of that particular
parameter. Normalized sensitivity coefficient given by the
following relation (Zheng and Bennett, 1995) was computed
for each input parameters:
 
Saj þDaj  Saj
Xj ¼ (12)
Daj =aj
where Xj ¼ normalized sensitivity coefficient for jth param-
eter; and Saj , Saj þDaj ¼ calibration statistics (RMSE) for the base
case and the case where the base parameter aj has been
changed by Daj, respectively. Calculated normalized sensi-
tivity coefficients of model parameters along with percent
changes in RMSE values for reach 2 are given in Table 3. The
sensitivity coefficients indicate that the model is most sensi-
tive to the parameter As/A (2.86), followed by Ds/A (0.18), Tmin
(0.10), k or ks (0.08), and Ks (0.06). The simulation results show
that the VARTBacT model reproduces breakthrough curves for
the bacteriophage tracer reasonably well. The RMSE values
from this model are comparable to those from Shen et al.
(2008) even though the same parameter values for both the
conservative tracer RWT and reactive tracer P22 are used in
the VARTBacT model. Shen et al. (2008) used different
parameter values for two tracers and assumed that the
different processes might have contributed to transient
storage within the same reach for two tracers, demonstrating
the performance of the VARTBacT model.
3.2. Sediment resuspension-induced bacterial transport
during high flows

The numerical model VARTBacT was applied to simulate the

Table 1 e Estimated common parameters for RWT and resuspension of E. coli in Swan Creek, Ontario, Canada using
P22 in the VART model. field observations. Jamieson et al. (2005b) seeded a part of
Reach U Ks Aadv/A Ds/A qL Tmin streambed with a strain of nalidixic acid resistant E. coli (E. coli
(m/s) (m2/s) (1/s) (m2/s) (hours) NAR), which is rarely found in natural environment, and
monitored the survival and transport of the tracer bacteria for
1 0.43 0.02 0.03 1.5  107 0.00 0.23
a two-month period covering three storm events. The flow,
2 0.59 1.00 0.27 7.8  107 1.3  104 0.75
suspended sediment concentration and bacterial concentra-
3 0.55 10.00 0.16 9.3  108 2.2  104 1.10
tion in water column at 10, 100, 500, and 1700 m downstream
1334 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3
Fig. 2 e Comparison between observed and simulated
Rodamine WT tracer concentrations at three sampling
stations in the Grand River: (a) 4.56, (b) 13.69, and (c)
28.38 km downstream of tracer release. Right, center and
left are the sampling locations and Q-wt refers to the flow-
weighted concentration at each sampling site.
of the source cell, along with bacterial concentration in
bed sediment in the source cell were measured and reported.
Four depth-integrated water samples were collected at each
sampling location and composited for analysis. Bed sediment
samples at source cell were taken from 2 cm depth. The
median grain size was 0.11 mm and 32% of the sediment
particles were reported to be finer than 0.075 mm. The river
Fig. 3 e Comparison between observed and simulated
bacteriophage P22 concentrations at three sampling
stations in the Grand River: (a) 4.56, (b) 13.69, and (c)
28.38 km downstream of tracer release site. Right, center
and left are the sampling locations and Q-wt refers to the
flow-weighted concentration at each sampling site.
width ranged from 6 to 10 m and the water surface slope was
reported to vary from 0.0007 to 0.008 m m1. Manning’s
roughness coefficient n was estimated as 0.045.
To understand the effect of sedimentewater interaction on
microbial concentration in the stream, two scenarios were
analyzed. The first scenario without sedimentewater inter-
action was simulated using the VART model with a bacterial
 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3 1335

Table 3 e Summary of sensitivity analysis for all three cases showing: i) calibrated parameter values, ii) percentage
changes in RMSE values in response to 50% change in calibrated parameter values, and iii) normalized sensitivity
coefficients (Xj).
Parameters Case I: Grand River Case II: Swan River Case III: Motueka River

Calibrated % Change Xj Calibrated % Change Xj Calibrated % Change Xj

parameter in RMSE parameter in RMSE parameter in RMSE
values values values

Ks (m2/s) 1.00 7.1 0.06 4.50 0.0 0.00 135.00 9.8 0.08
As/A () 0.27 328.6 2.86 0.10 0.0 0.00 0.10 0.0 0.00
Ds/A (1/s) 7.8  107 21.4 0.18 7.0  107 0.0 0.00 5.4  108 0.0 0.00
k (1/d) 1.73 9.5 0.08 8.60 3.1 0.04 1.18 4.9 0.40
ks (1/d) 1.73 9.5 0.10 0.12 0.0 0.00 0.00 e e
Tmin (hrs) 0.75 11.9 0.10 10.00 0.0 0.00 24.00 0.0 0.00
Cb (CFU 101/g) e e e 10.00 7.7 0.10 500.00 12.2 0.10
scr,e (Pa) e e e 1.50 21.5 0.28 e e e
scr,d (Pa) e e e 1.00 7.7 0.10 e e e
M (kg/m2s) e e e 0.12 3.1 0.04 e e e
fp () e e e 0.20 13.8 0.18 e e e
seq (Pa) e e e e e e 5.00 9.8 0.08
Kd (m3/kg) e e e e e e 4.0  103 17.1 0.16
x (104 CFU/g) e e e e e e 4.00 80.5 0.66

decay/growth term while the second scenario with sed-
imentewater interaction was modeled using the VARTBacT
model given by Eqs. (6)e(9). Values of model input parameters
related to flow, river geometry, sediment and bacterial
concentration were taken from Jamieson et al. (2005b). The
average river width of 8 m was used throughout the study
reach. The water surface slope during the base flow was set to
0.004 m m1, an average value of the water surface slopes
reported by Jamieson et al. (2005b). The hydrograph-based
method (Ghimire and Deng, 2011) was used to compute flow
parameters such as s (Fig. 4), h and U. Minimum, mean and
maximum values of s during the simulation period were 0.41,
1.49, and 3.05 N/m2, respectively. Hourly flow data were used
in the calculation. The concentration of bacteria at 10 m
downstream from the source cell was used as an upstream
boundary condition for the model.
The first scenario assumes no sediment resuspension/
deposition in the selected reach. Therefore, the third and
fourth terms on the right-hand side of Eq. (6), which are
Fig. 4 e Observed flow (Q) and computed bed shear stress
(s) at 100 m downstream of source cell.
relevant to sediment resuspension and deposition, were
dropped. The net decay rate ks in storage zones is taken as
0.12 d1, as measured by Jamieson et al. (2005b), and the net
decay rate k in water column was adopted as 8.60 d1, an
average value for the same three flow events in Swan River
estimated by Rehmann and Soupir (2009). The dispersion
coefficient Ks was calculated based on the formula presented
by Deng et al. (2001) for straight rivers:

5=3
2
Ks 0:01j B U
¼ (13)
hU 8εt0 h U
where, U* ¼ bed shear velocity (m/s) whose minimum, mean
and maximum values were 0.02, 0.038, and 0.055 m/s,
respectively; j ¼ 4.0 (Deng and Jung, 2008); and εt0 ¼ transverse
mixing coefficient () given by


1:38
1 U B
εt0 ¼ 0:145 þ (14)
3520 U h
Other parameters were either estimated or calibrated as
recommended by Deng and Jung (2009) for VART model. It was
assumed that no lateral flow or hyporheic exchange occurred
during the simulation period. The estimated parameters for
all three reaches are as follows: Ks ¼ 4.5 m2/s, Aadv/A ¼ 0.1, Ds/
A ¼ 7.5  107 (1/s), Tmin ¼ 10 h. The results are shown in
Fig. 5. The RMSE values of the fit for reaches 1, 2 and 3 are 1.33,
0.86, and 0.78, respectively, and corresponding r2 are 0.61, 0.83
and 0.96 respectively.
The second scenario involving sediment resuspension
and deposition was simulated with all terms in Eq. (6). The
flow parameters and the decay rates were kept same as in
the first scenario. The measured median particle size d50 was
set to 0.1 mm (Jamieson et al., 2005b) and scr,e was set to
1.5 N/m2 for sediment resuspension computations. Sedi-
ment was considered to be cohesive with scr,d ¼ 0.08 N/m2,
which falls within the commonly observed range of 0.07 and
1.1 N/m2 (Lick, 2008). The empirical erosion rate constant M
was estimated as 0.12 g/m2s. The settling velocity ws
1336 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3
Fig. 5 e Simulated concentrations of E. coli in water column
at (a) 100, (b) 500, and (c) 1700 m downstream from the
source cell without sediment resuspension and deposition.
(¼0.0032 m/s) was determined using Eq. (4) and porosity p
(¼0.46) was obtained from a formula by Wu and Wang (2006).
The fraction of bacteria attached to suspended sediments fp
was kept 0.20, which is in the lower range of 0.20 to 0. 55
reported by Characklis et al. (2005) for surface waters and
storm waters. The concentration of bacteria in bed sediment
Cb was estimated as 1.5 CFU/0.1 g based on model calibra-
tions, which was carried out primarily based on RMSE
as explained for Grand River. As resuspension and deposi-
tion processes were involved, the calibration process also
involved additional parameters such as M, scr,d, and fp. The
RMSE of the fit for reaches 1, 2, and 3 are 1.15,0.65, and 0.61,
respectively, and corresponding r2 are 0.62, 0.83, and 0.37,
respectively. The sensitivity of the parameters was assessed
using normalized sensitivity coefficient Xj and percentage
change in RMSE values for reach 2. The result of analysis
(Table 3) shows that the most sensitive parameters based on
Xj was scr,e (0.28) followed by fp (0.18), Cb (0.10), scr,d (0.1), k
(0.04), and M (0.04). Other parameters including As/A, Tmin,
Ds/A, ks, and Ks have either very small or zero sensitivity
within the selected range.
A comparison between Figs. 5 and 6 shows that the
consideration of sedimentewater interaction in the second
scenario did not result in substantially improved predictions
of bacterial concentration. While simulated concentrations
were slightly overpredicted in the first scenario without
sediment interaction, the second scenario with sedimente-
water interaction slightly underestimated the concentration
especially in the downstream reaches. Considering the
uncertainties associated with the additional parameters such
as settling velocity, sediment erosion and attached fraction,
a slightly lower RMSE in the second scenario does not provide
compelling evidence of sedimentewater interaction.
Rehmann and Soupir (2009) reported that their model always
overpredicted the concentration without including sed-
imentewater interaction. When the sedimentewater inter-
action was included, their model results were much better
although slightly underpredicted. This slight discrepancy in
the result might be due to the use of different approaches in
computation of sediment resuspension or due to the consid-
eration of full time series of flow using hydrograph-based
method in this study. Rehmann and Soupir (2009) simulated
three events with average flows estimated using a steady state
model. A comparison of both results is therefore not
straightforward, although both studies demonstrated that the
impact of sedimentewater column interaction on the model
may be substantial. It should be noted that Jamieson et al.
(2005b) recovered no bacterium in any bed sediment
samples collected downstream of the source cell. Moreover,
the E. coli concentration did not increase in the downstream
direction, leading them to believe that the bacteria, which
were resuspended from the source cell during flood events,
traveled through the study reach without deposition. This
finding is consistent with the comparison results from Figs. 5
and 6. It means that the first scenario simulated with the
simplified VARTBacT model best describes the experiments by
Jamieson et al. (2005b).
3.3. E. coli transport during high flow events with
contribution from watersheds
The first two cases discussed in Sections 3.1 and 3.2 focus on
effects of individual processes, such as transient storage (case
1) and sediment resuspension (case 2), on bacterial transport
and fate under controlled experimental conditions. In addi-
tion to the individual processes simulated in previous two
sections, bacterial transport and fate in streams under natural
conditions like flood events are also highly affected by
bacterial inputs from watersheds which are rarely taken into
account in previous studies. This section is intended to model
bacterial fate and transport in streams under natural condi-
tions which may include effects of all individual processes.
 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3
Fig. 6 e Simulated concentrations of E. coli in water column
at (a) 100, (b) 500, and (c) 1700 m downstream from the
source cell considering sedimentewater interactions.
E. coli concentration during several natural high flow
events in Motueka River, New Zealand were monitored over
a period of 13 months from June 2003 (Wilkinson, 2008). The
Motueka River has a drainage area of 2180 km2 and a median
discharge of 47 m3 at about 8 km upstream of the mouth of
river at Woodmans Bend. The discharge and turbidity were
also measured continuously at this location. The E. coli
concentration during storm events at Woodmans Bend were
found to be an order of magnitude or more higher than during
the base flow (McKergow and Davies-Colley, 2010). As routine
sampling efforts normally do not cover such events, the
monitoring efforts provided excellent data for investigating
1337
the controlling factors and sources of E. coli in a river under
natural conditions. Since the storm events export a major part
of the annual load of E. coli up to 98% (Chu et al., 2011;
McKergow and Davies-Colley, 2010), timing and magnitude of
E. coli concentration are highly important.
In this case study, the VARTBacT model tested in the
previous two cases was used to simulate the E. coli dynamics
in Woodmans Bend of Motueka River using monitored data for
twelve storm events (SE 1 to SE 12) between 2003 and 2004.
Observed E. coli and flow data from Wilkinson et al. (2011) and
suspended sediment concentration data from Wild et al.
(2006) were used in VARTBacT simulations. The width of the
river is about 70 m and the water surface slope during the base
flow was about 1.0  104 m m1. Manning’s roughness
coefficient was estimated as 0.045 based on Chow (1959). Flow
parameters and sediment transport were estimated using the
hydrograph-based method. Hourly flow data were used as an
input in Eq. (1).
As the streambed sediment was mostly sand mixed with
gravel (Basher et al., 2011), a non-cohesive formula was more
appropriate for sediment transport. Median grain diameter d50
of 0.125 mm was used in Ackers and White (1973) formula for
sediment transport in the hydrograph-based method. As the
sediment size distribution was not available at the study site,
a single median grain size was used. The simulated and
observed suspended sediment concentrations during two
periods are shown in Fig. 7.
To understand relative contributions of the streambed and
watersheds to E. coli transport during storm events and to
demonstrate the applicability of the VARTBacT model, a 5-
km-long stream reach with its downstream boundary at
Woodmans bend was selected. The reach was assumed to
have a constant channel geometry and sediment properties.
Since there were no measured E. coli concentration data
available for the flood events upstream of Woodmans bend,
the concentration at the upstream boundary was estimated by
assuming a direct correlation of E. coli concentration with the
simulated depth averaged sediment concentration S through
a parameter x. The parameter x, which represents the contri-
bution of the watershed as well as channel source upstream of
the reach, was determined through calibrations. The depth
averaged net sediment flux Fr was calculated using Eq. (2).
Actual average sediment concentration S was set to increase
linearly from 0.7  Se for the highest bed shear stress
(smax ¼ 12.0 N/m2) to 1.3  Se for lowest bed shear stress
(smin ¼ 0.5 N/m2) to ensure net erosion during high flows and
net deposition during low flows. As the study reach was not
undergoing significant erosion or deposition over time,
minimal net erosion was ensured in this way. The equilibrium
transport capacity was estimated to occur at bed shear stress
seq ¼ 5.0 N/m2. Net vertical displacement was computed using
Eq. (3), whereas the settling velocity was computed using
Eq. (4).
All terms in Eqs. (6)e(9) were included and solved. The
same trial and error approach of parameter estimation, as
described for Swan River earlier, was used for model calibra-
tion. The unknown parameters were varied within a possible
range to obtain the lowest RMSE and r2. The VART parameters
thus obtained are as follows: Ks ¼ 135 m2/s, Aadv/Aav ¼ 0.1,
Ds ¼ 1.0  105, and Tmin ¼ 24 h. The inactivation coefficient
1338 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3

for all events. The parameter x and the concentration of E. coli

Fig. 7 e Observed and simulated sediment concentrations
in the Motueka River at Woodman’s Bend during (a) 9/18/
03e9/22/03, and (b) 6/14/04e6/20/04.
of bacteria in water column k and in storage zones ks were set
as 1.1 d1 and zero, respectively. Although some die-off is
expected to present in the storage zones, it was not expected
to play an important role in simulation results. Significant die-
off of bacteria during storm events is unlikely due to increased
turbidity, higher water depth and faster flows (Wilkinson,
2008). The linear partition coefficient Kd was set to 4 L/g,
which is less than the value (Kd ¼ 10 L/g) used by Bai and Lung
(2005). This value of Kd gives the fraction of bacteria fp ¼ 0.45 at
sediment concentration of 200 g/m3, which is within the range
of 0.20 to 0. 55 reported by Characklis et al. (2005) for surface
waters and storm waters.
Those parameters which are not expected to vary signifi-
cantly from event to event such as seq, Kd, and k were kept
constant for all events. The VART parameters (Ks, Aadv/Aav, Ds,
Tmin), which were found to have a small or negligible sensi-
tivity for the event based simulations, were also kept constant
in bed sediment Cb were, however, varied for all storm events
and determined through calibrations. The calibration result
showed that Cb values in all storm events except SE 4 were
within 4.5  102 and 5.5  102 CFU/0.1 g, which are within the
range commonly observed in river sediments: 1e104 CFU/0.1 g
(Cho et al., 2010b), 102e9  103 CFU/0.1 g (Smith et al., 2008). In
SE 4, however, Cb value of 5.0  104 CFU/0.1 g provided the best
fit in terms of RMSE as well as r2. This concentration is higher
than those in other storm events by two orders of magnitude
and is higher than that normally observed in river sediments.
Interestingly, the E. coli concentration in the water column
during this event is also significantly higher than other events
of similar magnitude. Possible reasons for such an increased
concentration level, may include effect of rainfall distribution,
timing between events (McKergow and Davies-Colley, 2010),
and suitable environment in bed sediment for survival and
growth of E. coli bacteria. Occurrence of some other events
such as sewer leaks or overflows causing increased concen-
tration is least expected due to the nature and size of the
sparsely populated Motueka catchment.
Based on the average normalized sensitivity coefficient Xj,
computed using Eq. (12) for Motueka River SE 3, the most
sensitive parameter is x (0.66) followed by Kd (0.16), Cb (0.10),
seq (0.08), Ks (0.08), and k (0.04) (Table 3). Other parameters
including As/A, Tmin, Ds/A have either very small or zero
sensitivity within the selected range (Daj/Da ¼ 0:0.5) and ks
was not considered. The values of parameters x, Cb, RMSE, and
r2 for storm events are given in Table 4. A comparison between
simulated and observed E. coli concentrations for flood events
shows that the model generally captured observed concen-
tration patterns well, as shown in Fig. 8. The simulated peak
concentration mostly occurred during the rising limb of the
hydrograph, matching the trend of the observed data well. A
slight under-estimation during low flows might be due to the
overestimation of settling velocity caused by: i) under-
estimation of bed shear stress during falling limb by the
hydrograph-based method (Ghimire and Deng, 2011), and ii)
adoption of a single median particle size, which does not
reflect the effect of presence of various size fractions on the
sediment transport (Molinas and Wu, 1998).
The best fit for storm events was achieved primarily with
the two parameters x and Cb. Although the model is also
sensitive to other parameters such as Kd and seq, these
parameters were kept constant for all flood events. Inactiva-
tion rate and most of the VART model parameters were not
very important during these short high flow events although
their roles can be important during moderate to low flows for

Table 4 e Parameters x, RMSE, Cb and r2 values for storm events.

Parameters Storm events at Woodmans Bend

1 2 3 4 5 6 7 8 9 10, 11 12
4
x (10 CFU/g) 1.50 2.0 4.0 2.0 1.0 3.0 4.0 7.0 4.0 2.0 1.0
Cb  102 (CFU/0.1 g) 4.5 5.0 5.0 5.0  102 5.0 5.5 5.5 4.5 5.0 5.0 4.5
RMSE 0.55 0.54 0.41 0.61 0.45 0.46 0.46 0.46 0.43 0.56 1.56
r2 0.42 0.79 0.90 0.85 0.46 0.36 0.46 0.72 0.90 0.55 0.42
 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3 1339

Fig. 8 e Observed and simulated E. coli concentrations in the Motueka River at Woodman’s Bend during storm events SE 1 (a)
to SE 12 (k) which occurred between June 2003 and June 2004.

an extended period. While resuspension of E. coli, represented
by Cb, was important to simulating peak concentrations
correctly, the watershed input parameter x has the greatest
impact on the model output (x  S ¼ assumed upper boundary
concentration). The remobilization of channel sources of E.
coli can be an important part of the total E. coli load during peak
flows, especially under suitable bed sediment environment for
E. coli preceding the flow event.
One of the major limitations in this model application is the
assumption of sediment concentration dependent upstream
boundary. Although this assumption helped us to understand
the dynamics of bacterial concentration during storm events e
especially as it concerns with the relative contribution of the
watershed and riverbed sediment e it inherently assumes that
a good correlation exist between suspended sediment
concentration and E. coli levels in flow. Such assumption might
be useful only when the E. coli concentration is closely corre-
lated with turbidity (e.g., Muirhead et al., 2004) and exhibit
hysteretic relationships (Nagels et al., 2002). When this rela-
tion is poor, this assumption of boundary condition is not
appropriate. This limitation, however, will not be an issue
when the upper boundary data are available. The term x will no
longer be required in that case.
4. General discussion
One of the distinctive features of the hydrograph-based
modeling approach in comparison with other models for
bacterial transport is that the new approach incorporates
the transient storage effect and the effect of sediment resus-
pension and deposition into the VART model as external
1340 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3
sources or sinks, respectively. In contrast, many other models
(e.g., Gao et al., 2011; Rehmann and Soupir, 2009; Steets and
Holden, 2003) considered the mass balance between water
column and sediment bed, and neglected the transient
storage effect. Therefore, the present model is suitable to use
when i) the transient storage exchange is important for
bacterial transport such as when the hyporheic exchange is
the main cause for bacterial flux (e.g., Grant et al., 2011), and
there is no significant sediment resuspension, and ii) when
the sediment resuspension is significant but the duration is
relatively short such as during flood events. In the first case,
the resuspension term in Eq. (6) is irrelevant, whereas in the
second case, the transient storage effect is often negligible and
becomes irrelevant due to longer time scale and destruction of
storage zones by high flow-induced sediment resuspension.
However, if the resuspension and deposition occur for longer
time or in succession, this model may not be efficient as it
does not consider the change in concentration in bed sedi-
ment Cb in response to resuspension/deposition to and from
the water column directly. When it is applied for individual
flow events where the settling does not significantly affect the
concentration, this model should work well.
With the resuspension term in the VARTBacT model, Cb
has to be estimated together with the rate of resuspension E.
This was done while modeling for Swan Creek and Motueka
storm events in this study. The resuspension was first esti-
mated and then Cb was determined through calibrations. As
uncertainties related to each term is usually high, one of the
options is to use resuspension flux (E  Cb) as a single
parameter to determine through calibrations as proposed by
Cho et al. (2010b). This approach is helpful when estimating E
is difficult due to lack of knowledge on sediment properties or
sediment transport data. This is usually the case when the
sediment is cohesive and no measured erosion rate is avail-
able. In view of the possibility of decreasing values of both E
and Cb vertically in the streambed, the calibration appears to
be the only viable option.
In addition to resuspension, two processes related to the
sedimentewater interaction are important: settling and
attachment of bacteria to particles. Proper understanding of
uncertainties associated with these processes is important for
bacterial modeling. For example, the particle size of bottom
sediment vary over 2e3 orders of magnitude and the settling
velocity varies over 4e6 orders of magnitude. Yet a single
representative d50 size was adopted in this study for simplicity
of computation. Similarly, the effect of flocculation on settling
velocity for cohesive sediment was ignored, despite the fact
that the concentration of bacteria is highest in flocs domi-
nating in the suspended sediment (Droppo et al., 2009). Most
of the existing models for cohesive sediment transport use
Stoke’s law although the law is only applicable to a steady and
laminar flow (Lick, 2008) and gives higher settling velocity for
cohesive sediments (Camenen and Larson, 2009). Another
uncertainty is due to assumed fraction of attached bacteria fp
or partition coefficient Kd in the suspended sediment.
Attachment ratio greatly influences the settling rate of the
bacteria from the water column. The attached fractions vary
very widely from about 0.10 (Liu et al., 2006) to more than 0.80
(Hipsey et al., 2006; Steets and Holden, 2003) in natural waters.
Characklis et al. (2005) found that 20e55% of E. coli was
associated with settleable solids in surface waters and storm
waters. Reported values of fp in rivers and estuaries are mostly
in the range from 0.22 to 0.44 (Jamieson et al., 2005a), 0.30
(Dorner et al., 2006), 0.38 (Fries et al., 2006), and 0.50 (Cho et al.,
2010b; Wu et al., 2009). In this study, Kd was set to 4 L/g for
Motueka river floods, which gives fp ¼ 0.45 at a concentration
of about 200 g/m3. In case of Swan Creek, the value of 1.0 was
used in Swan Creek as assumed by Jamieson et al. (2005b) in
their study. Unattached bacteria were assumed to have zero
settling velocity following Hipsey et al. (2006).
Although less important to simulating short storm events,
the bacterial decay/growth terms k, ks may be important to
overall model results. The factors that influence the decay of
E. coli in aqueous environment include the exposure to
sunlight, temperature, salinity, pH and nutrients. Excellent
literature is available on the fate of bacteria in aquatic (Hipsey
et al., 2008; John and Rose, 2005) and sediment (Davies et al.,
1995; Garzio-Hadzick et al., 2010) environments. In general
the survival rate of E. coli bacteria in sediment is much longer
than in surface waters. Jamieson et al. (2004) found that E. coli
may survive in streambed sediments for more than 6 weeks
where they are protected against predation and sunlight
exposure. Evidence of growth in the sediment has also been
reported (Hipsey et al., 2008).
Finally, the developed model for high flow events uses the
hydrograph-based method, which provides better estimates of
bed shear stress and other flow variables as compared with the
conventional steady uniform flow approach (Ghimire and Deng,
2011). The hydrograph-based method takes into account the
unsteadiness of flow by including the discharge gradient (vQ=vt)
for correcting friction slope during unsteady flows. This might
be one of the reasons why the VARTBacT model has improved
predictions of bacterial concentration particularly in timing and
magnitude of the peak for flood events.
5. Conclusion
A new modeling approach to prediction of the fate and
transport processes of bacteria in streams is presented. The
model extends VART model by including resuspension and
deposition processes during high flow events. It is developed
using hydrograph-based method for flow, sediment transport
and sedimentewater interface displacement. Sediment fluxes
are computed using sediment carrying capacity approach for
non-cohesive sediment and critical bed shear stress for
cohesive sediment. The model is applied to three different
rivers with distinct characteristics, ranging from steady flow
with significant transient storage exchange to highly unsteady
storm flows where the sediment resuspension plays an
important role for bacterial fate and transport in streams. The
main findings from this study are as follows:
 The VART model is able to simulate bacteriophage P22
tracer breakthrough curves in a low flow river without
significant sediment resuspension/deposition.
 The VARTBacT model is able to predict the field measure-
ments reasonably well using time series of hourly flow data
generated from the hydrograph-based method. This model
is appropriate for studying effect of sedimentewater
 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3 1341

 
column interaction on the bacterial concentration during nþ1=2
ð1  εDt=4  ks εDt=4ÞCsi þ ðεDt=4Þ Cinþ1 þ Cinþ1=2
flood events in rivers. Cnþ1 ¼
si
1 þ εDt=4 þ ks Dt=4
 While transient storage and settling may be important
(A5)
during low and relatively steady flows, the transient storage
effect is negligible when resuspension process becomes Substituting Eq. (A5) into Eq. (A3) and rearranging terms to
dominant during high storm flows. have all known terms on the right side gives
 The model can be applied effectively for simulating indi-
vidual flood events. The demonstration application for the aCnþ1
iþ1 þ ð1 þ 2a þ g þ bÞCi
nþ1
 aCnþ1
i1
nþ1=2 nþ1=2
Motueka River in New Zealand shows that while the ¼ ð1  g  bÞCi þ 2b Csi þh (A6)
resuspension from bed sediment can be a significant source
where the following definitions are used:
of E. coli bacteria, watershed inputs are by far the most
!
important contributor to instream bacterial transport Ks Dt D fp Dt ðεRDtÞð1 þ ks Dt=4Þ
during flood events. More efforts are needed in estimating a¼ ;g ¼ b þk ;b ¼ ;h
2ðDxÞ 2
hS 4 4ð1 þ εDt=4 þ ks Dt=4Þ
watershed loading of bacteria.
EDt
¼ Cb
2h
(A7)

Acknowledgments The parameters a, g, b and h are either known or calculable.

The concentration Cinþ1=2 is computed from the solution of Eq.
nþ1=2
Support for this research by the USGS/Louisiana Water (A1) whereas the concentration Csi is assumed to be equal
Resources Research Institute and the LaSPACE NASA grant is to Cnsi . The Eq. (A7) then may be grouped as follows
gratefully acknowledged. The authors wish also to thank two
anonymous reviewers for their valuable comments and iþ1 þ PCi
UCnþ1 nþ1
þ UCnþ1
i1 ¼ W
nþ1=2
(A8)
suggestions which helped improve the manuscript. where, U ¼ a, P ¼ 1 þ 2a þ g þ b, and
Appendix. Numerical solution of VARTBacT nþ1=2 nþ1=2
Wnþ1=2 ¼ ð1  g  bÞCi þ 2b Csi þh (A9)
model
The left-hand side of Eq. (A8) may be assembled into a tri-
A split-operator method is employed to split the Eq. (6) into diagonal matrix and solved to determine the concentration at
a pure advection and a dispersion equation with the transient time level n þ 1.
storage, erosion/deposition and decay terms:

references
vC vC 
þU ¼ 0; t˛ tn ; tnþ1=2 (A1)
vt vx

Abad, J.D., Garcı́a, M.H., 2006. Discussion of “Efficient Algorithm

vC v 2 C As 1 E D fp C 
¼ Ks 2 þ ðCs  CÞ þ Cb   kC; t˛ tnþ1=2 ; tnþ1 for Computing Einstein Integrals” by Junke Guo and Pierre Y.
vt vx A Tv h h S Julien. Journal of Hydraulic Engineering 132 (3), 332e334.
(A2) Ackers, P., White, W.R., 1973. Sediment transport: new approach
and analysis. Journal of the Hydraulics Division 99 (11),
where n stands for time step. Eq. (A1) is solved using semi-
2041e2060.
Lagrangian approach (Deng et al., 2006), whereas Eq. (A2) in Ariathurai, C.R., 1974. A Finite Element Model for Sediment
conjunction with Eqs. (7)e(9) is solved using forward time Transport in Estuaries. Ph.D. thesis. University of California,
scheme and fully implicit F.3 central finite-difference scheme Davis.
presented by Deng et al. (2004). The discretized form of the Eq. Bai, S., Lung, W.-S., 2005. Modeling sediment impact on the
(A2) may be written as transport of fecal bacteria. Water Research 39 (20), 5232e5240.
Bard, Y., 1974. Nonlinear Parameter Estimation. Academic Press,
New York.
Ks  nþ1 nþ1=2
Cnþ1  Cinþ1=2 si þ Csi
Cnþ1 Basher, L.R., Hicks, D.M., Clapp, B., Hewitt, T., 2011. Sediment
i
¼ Ciþ1  2Cnþ1 þ Cnþ1
i1 þ εR
Dt=2 ðDxÞ2 i
2 yield response to large storm events and forest harvesting,
! ! ! Motueka River, New Zealand. New Zealand Journal of Marine
nþ1=2
C þ Ci
nþ1
D fp Ci þ Cinþ1=2
nþ1
E
 i  þk þ Cb and Freshwater Research 45 (3), 333e356.
2 hS 2 h Camenen, B., Larson, M., 2009. A general formula for noncohesive
(A3) suspended sediment transport. Journal of Coastal Research 24
(3), 615e627.
where ε ¼ 1/Tv, R ¼ As/A. The Eq. (7) is discretized as Characklis, G.W., Dilts, M.J., Simmons III, O.D., Likirdopulos, C.A.,
Krometis, L.-A.H., Sobsey, M.D., 2005. Microbial partitioning to
nþ1=2 nþ1=2
! settleable particles in stormwater. Water Research 39 (9),
Cnþ1  Csi Cnþ1 þ Cinþ1=2 Cnþ1 þ Csi 1773e1782.
si
¼ε i  si
Dt=2 2 2 Cho, K.H., Cha, S.M., Kang, J.-H., Lee, S.W., Park, Y., Kim, J.-W.,
nþ1=2
! Kim, J.H., 2010a. Meteorological effects on the levels of fecal
C nþ1
þ Csi
 ks si (A4) indicator bacteria in an urban stream: a modeling approach.
2 Water Research 44 (7), 2189e2202.
Cho, K.H., Pachepsky, Y.A., Kim, J.H., Guber, A.K., Shelton, D.R.,
Rearranging terms in Eq. (A4) yields Rowland, R., 2010b. Release of Escherichia coli from the bottom
1342 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3
sediment in a first-order creek: experiment and reach-specific
modeling. Journal of Hydrology 391 (3e4), 322e332.
Chow, V.T., 1959. Open-Channel Hydraulics. McGraw-Hill, New
York, USA.
Chu, Y., Salles, C., Tournoud, M.-G., Got, P., Troussellier, M.,
Rodier, C., Caro, A., 2011. Faecal bacterial loads during flood
events in Northwestern Mediterranean coastal rivers. Journal
of Hydrology 405 (3e4), 501e511.
Davies-Colley, R.J., Bell, R.G., Donnison, A.M., 1994. Sunlight
inactivation of enterococci and fecal coliforms in sewage
effluent diluted in seawater. Applied and Environment
Microbiology 60 (6), 2049e2058.
Davies, C.M., Long, J.A., Donald, M., Ashbolt, N.J., 1995. Survival of
fecal microorganisms in marine and freshwater sediments.
Applied and Environment Microbiology 61 (5), 1888e1896.
Deng, Z.-Q., Bengtsson, L., Singh, V., 2006. Parameter estimation
for fractional dispersion model for rivers. Environmental Fluid
Mechanics 6 (5), 451e475.
Deng, Z.-Q., Jung, H.-S., 2008. Scaling dispersion model for
pollutant transport in rivers. Environmental Modelling and
Software 24 (5), 627e631.
Deng, Z.-Q., Jung, H.-S., 2009. Variable residence time based
model for pollutant transport in streams. Water Resources
Research 45, W03415. http://dx.doi.org/10.1029/
2008WR007000.
Deng, Z.-Q., Jung, H.-S., Ghimire, B., 2010. Effect of channel size on
solute residence time distributions in rivers. Advances in
Water Resources 33 (9), 1118e1127.
Deng, Z.-Q., Singh, V.P., Bengtsson, L., 2004. Numerical solution of
fractional advection-dispersion equation. Journal of Hydraulic
Engineering 130 (5), 422e431.
Deng, Z.-Q., Singh, V.P., Bengtsson, L., 2001. Longitudinal
dispersion coefficient in straight rivers. Journal of Hydraulic
Engineering 127 (11), 919e927.
Dorner, S.M., Anderson, W.B., Slawson, R.M., Kouwen, N.,
Huck, P.M., 2006. Hydrologic modeling of pathogen fate and
transport. Environmental Science and Technology 40 (15),
4746e4753.
Droppo, I.G., Liss, S.N., Williams, D., Nelson, T., Jaskot, C.,
Trapp, B., 2009. Dynamic existence of waterborne pathogens
within river sediment compartments. Implications for water
quality regulatory affairs. Environmental Science and
Technology 43 (6), 1737e1743.
Fries, J.S., Characklis, G.W., Noble, R.T., 2006. Attachment of fecal
indicator bacteria to particles in the Neuse River Estuary, N.C.
Journal of Environmental Engineering 132 (10), 1338e1345.
Gao, G., Falconer, R.A., Lin, B., 2011. Numerical modelling of
sediment-bacteria interaction processes in surface waters.
Water Research 45 (5), 1951e1960.
Garcı́a, M.H., 2008. Sediment transport and morphodynamics. In:
Garcia, M.H. (Ed.), Sedimentation Engineering: Processes,
Measurements, Modeling, and Practice, ASCE Manuals and
Reports on Engineering Practice No. 110. ASCE, Reston, VA,
pp. 21e163.
Garzio-Hadzick, A., Shelton, D.R., Hill, R.L., Pachepsky, Y.A.,
Guber, A.K., Rowland, R., 2010. Survival of manure-borne E. coli
in streambed sediment: effects of temperature and sediment
properties. Water Research 44 (9), 2753e2762.
Ghimire, B., Deng, Z.-Q., 2011. Event flow hydrograph-based
method for shear velocity estimation. Journal of Hydraulic
Research 49 (2), 272e275.
Grant, S.B., Litton-Mueller, R.M., Ahn, J.H., 2011. Measuring and
modeling the flux of fecal bacteria across the sediment-water
interface in a turbulent stream. Water Resources Research 47
(5), W05517.
Hipsey, M., Brookes, J., Regel, R., Antenucci, J., Burch, M., 2006. In
situ evidence for the association of total coliforms and
Escherichia coli with suspended inorganic particles in an
Australian reservoir. Water, Air, & Soil Pollution 170 (1),
191e209.
Hipsey, M.R., Antenucci, J.P., Brookes, J.D., 2008. A generic,
process-based model of microbial pollution in aquatic
systems. Water Resources Research 44 (7), W07408.
Jamieson, R., Joy, D.M., Lee, H., Kostaschuk, R., Gordon, R., 2005a.
Transport and deposition of sediment-associated Escherichia
coli in natural streams. Water Research 39 (12), 2665e2675.
Jamieson, R.C., Joy, D.M., Lee, H., Kostaschuk, R., Gordon, R.J.,
2004. Persistence of enteric bacteria in alluvial streams.
Journal of Environmental Engineering and Science 3 (3),
203e212.
Jamieson, R.C., Joy, D.M., Lee, H., Kostaschuk, R., Gordon, R.J.,
2005b. Resuspension of sediment-associated Escherichia coli in
a natural stream. Journal of Environmental Quality 34 (2),
581e589.
John, D.E., Rose, J.B., 2005. Review of factors affecting microbial
survival in groundwater. Environmental Science and
Technology 39 (19), 7345e7356.
Krone, R.B., 1962. Flume Studies of the Transport of Sediment in
Estuarial Shoaling Processes. Hydraulic Engineering
Laboratory and Sanitary Engineering Research Laboratory,
University of California, Berkeley, CA.
Lick, W.J., 2008. Sediment and Contaminant Transport in Surface
Waters. CRC Press, Taylor & Francis Publishing, Boca Raton, FL.
Liu, L., Phanikumar, M.S., Molloy, S.L., Whitman, R.L.,
Shively, D.A., Nevers, M.B., Schwab, D.J., Rose, J.B., 2006.
Modeling the transport and inactivation of E. coli and
Enterococci in the near-shore region of Lake Michigan.
Environmental Science and Technology 40 (16), 5022e5028.
McKergow, L.A., Davies-Colley, R.J., 2010. Stormflow dynamics
and loads of Escherichia coli in a large mixed land use
catchment. Hydrological Processes 24 (3), 276e289.
Molinas, A., Wu, B., 1998. Effect of size gradation on transport of
sediment mixtures. Journal of Hydraulic Engineering 124 (8),
786e793.
Motta, D., Abad, J.D., Garcia, M.H., 2010. Modeling framework for
organic sediment resuspension and oxygen demand: case of
Bubbly Creek in Chicago. Journal of Environmental
Engineering 136 (9), 952e964.
Muirhead, R.W., Davies-Colley, R.J., Donnison, A.M., Nagels, J.W.,
2004. Faecal bacteria yields in artificial flood events: quantifying
in-stream stores. Water Research 38 (5), 1215e1224.
Nagels, J., Davies-Colley, R., Donnison, A., Muirhead, R., 2002.
Faecal contamination over flood events in a pastoral
agricultural stream in New Zealand. Water Science and
Technology 45 (12), 45e52.
Parchure, T.M., Mehta, A.J., 1985. Erosion of soft cohesive sediment
deposits. Journal of Hydraulic Engineering 111 (10), 1308e1326.
Rehmann, C.R., Soupir, M.L., 2009. Importance of interactions
between the water column and the sediment for microbial
concentrations in streams. Water Research 43 (18), 4579e4589.
RuiJie, L., Feng, L., WenJin, Z., 2009. The suspended sediment
transport equation and its near-bed sediment flux. Science in
China Series E: Technological Sciences 52 (2), 387e391.
Runkel, R.L., 1998. One-dimensional Transport with Inflow and
Storage (OTIS): a Solute Transport Model for Streams and
Rivers, 73 pp., Denver.
Sanford, L.P., Maa, J.P.Y., 2001. A unified erosion formulation for
fine sediments. Marine Geology 179 (1e2), 9e23.
Shen, C., Phanikumar, M.S., Fong, T.T., Aslam, I., McElmurry, S.P.,
Molloy, S.L., Rose, J.B., 2008. Evaluating Bacteriophage P22 as
a tracer in a complex surface water system: the Grand River,
Michigan. Environmental Science and Technology 42 (7),
2426e2431.
Smith, J., Edwards, J., Hilger, H., Steck, T.R., 2008. Sediment can be
a reservoir for coliform bacteria released into streams. Journal
of General and Applied Microbiology 54 (3), 173e179.
 w a t e r r e s e a r c h 4 7 ( 2 0 1 3 ) 1 3 2 9 e1 3 4 3
Smith, J.D., McLean, S.R., 1977. Spatially averaged flow over
a wavy surface. Journal of Geophysical Research 82 (12),
1735e1746.
Steets, B.M., Holden, P.A., 2003. A mechanistic model of runoff-
associated fecal coliform fate and transport through a coastal
lagoon. Water Research 37 (3), 589e608.
Wild, M., Hicks, D.M., Merrilees, R., 2006. Suspended Sediment
Monitoring in the Motueka Catchment: Data Report to 1 May
2006, NIWA Client Report CHC2006-087. NIWA, Christchurch.
Wilkinson, J., 2008. Faecal Indicator Organism Modelling (FIO):
Application to Motueka River. Cawthron, 52 pp., Prepared for
ICM.
Wilkinson, J., Jenkins, A., Wyer, M., Kay, D., 1995. Modelling faecal
coliform dynamics in streams and rivers. Water Research 29
(3), 847e855.
1343
Wilkinson, R.J., McKergow, L.A., Davies-Colley, R.J.,
Ballantine, D.J., Young, R.G., 2011. Modelling storm-event E.
coli pulses from the Motueka and Sherry Rivers in the South
Island, New Zealand. New Zealand Journal of Marine and
Freshwater Research 45 (3), 369e393.
Wu, J., Rees, P., Storrer, S., Alderisio, K., Dorner, S., 2009. Fate and
transport modeling of potential pathogens: the contribution
from sediments. Journal of the American Water Resources
Association 45 (1), 35e44.
Wu, W., Wang, S.S.Y., 2006. Formulas for sediment porosity and
settling velocity. Journal of Hydraulic Engineering 132 (8),
858e862.
Zheng, C., Bennett, G.D., 1995. Applied Contaminant Transport
Modeling: Theory and Practice. Van Nostrand Reinhold, New
York.