Journal of Hydrology (2008) 353, 129– 142

available at www.sciencedirect.com

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

Fully conservative coupling of HEC-RAS with

MODFLOW to simulate stream–aquifer interactions
in a drainage basin
a,* b,1 a,2
Leticia B. Rodriguez , Pablo A. Cello , Carlos A. Vionnet ,
David Goodrich c

a
Department of Hydrology and Water Resources, University of Arizona, Tucson, AZ USA
b
Facultad de Ingenierı̀a y Ciencias Hı̀dricas, Universidad Nacional del Litoral, CC 217, 3000 Santa Fe, Argentina
c
Southwest Watershed Research, Agricultural Research Service, U.S. Department of Agriculture,
2000 East Allen Road, Tucson, AZ 85719, USA

Received 20 September 2007; received in revised form 10 January 2008; accepted 6 February 2008

KEYWORDS Summary This work describes the application of a methodology designed to improve the
Groundwater–surface representation of water surface profiles along open drain channels within the framework
water interaction; of regional groundwater modelling. The proposed methodology employs an iterative pro-
Hydrologic modelling; cedure that combines two public domain computational codes, MODFLOW and HEC-RAS. In
MODFLOW; spite of its known versatility, MODFLOW contains several limitations to reproduce eleva-
HEC-RAS tion profiles of the free surface along open drain channels. The Drain Module available
within MODFLOW simulates groundwater flow to open drain channels as a linear function
of the difference between the hydraulic head in the aquifer and the hydraulic head in the
drain, where it considers a static representation of water surface profiles along drains.
The proposed methodology developed herein uses HEC-RAS, a one-dimensional (1D) com-
puter code for open surface water calculations, to iteratively estimate hydraulic profiles
along drain channels in order to improve the aquifer/drain interaction process. The
approach is first validated with a simple closed analytical solution where it is shown that
a Piccard iteration is enough to produce a numerically convergent and mass preserving
solution. The methodology is then applied to the groundwater/surface water system of
the Choele Choel Island, in the Patagonian region of Argentina. Smooth and realistic

* Corresponding author. Permanent address: Facultad de Ingenierı̀a y Ciencias Hı̀dricas, Universidad Nacional del Litoral, CC 217, 3000
Santa Fe, Argentina. Tel.: +1 54 342 4575234x198; fax: +1 54 342 4575224.
E-mail address: leticia@fich1.unl.edu.ar (L.B. Rodriguez).
1
Present address: Department of Civil Engineering, University of Illinois, Urbana-Champaign, IL, USA.
2
Present address: Facultad de Ingenierı̀a y Ciencias Hı̀dricas, Universidad Nacional del Litoral, CC 217, 3000 Santa Fe, Argentina.

0022-1694/$ - see front matter ª 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhydrol.2008.02.002
130 L.B. Rodriguez et al.

hydraulic profiles along drains are obtained while backwater effects are clearly repre-
sented.
ª 2008 Elsevier B.V. All rights reserved.

Introduction difference between the hydraulic head within the drain

and the hydraulic head in the adjacent aquifer. The MOD-
To determine the amount of water that is being exchanged FLOW DRN is a one-way flux package, whereby only aqui-
at any time at any given location between open channel and fer-to-drain flows are allowed, and not the reverse. Input
its surrounding aquifer poses a problem that not only has at- parameters to the module include the water elevation in
tracted the interest of the scientific community but also has the drain, or drain head, the spatial location of the drain,
many environmental implications. On one hand, depletion and the drain conductance. The drain head must be exter-
of streamflows and wetlands due to groundwater withdraw- nally calculated by the modeller based on field data. When
als often affects transient surface water flows, which in field data are limited, interpolation and/or extrapolation is
turn, are critical to sustain protected flora and wildlife. needed to compute drain heads for each hydrologic condi-
On the other hand, poorly drained soils due to a malfunc- tion spanned by the simulation. This task may be quite cum-
tioning drainage system may result in a build up of the water bersome and entails many uncertainties when dealing with
table that can impact the productivity of irrigated land intricate drainage networks. In addition to the difficulty
(Skaggs et al., 1995; Johnson and Koenig, 2003). Groundwa- and uncertainty of extensive space and time interpolation,
ter discharge to the drain ceases when the water head in the neither surface flow backwater effects nor surface flow
aquifer drops at/or below the elevation of the channel drain propagation is handled by the module. In actuality, as sta-
bottom. In actuality, groundwater discharge to channel ted by McDonald and Harbaugh (1988), ‘‘with proper selec-
drains can be regarded as a one-way stream–aquifer inter- tion of coefficients, the RIVER Module could be used to
action problem. Therefore, the aquifer-to-drain flux and perform the functions of the DRAIN Module’’. However, this
stream–aquifer interactions can be studied with a similar alternative would still maintain the limitations previously
methodology, as long as the exchange flux can be character- itemized.
ized with a Newton’s type cooling law (Carslaw and Jaeger, Pohll and Guitjens (1994) employed MODFLOW to simu-
1959), driven by the hydraulic head difference between the late regional flow and flow to the drains. Batelaan and De
two systems. Part of the objective of the present work is to Smedt (2004) also used MODFLOW with its DRN Module to
show that the vast experience gained over the years on the analyze the protection and development of groundwater-
treatment of stream–aquifer interactions (Sophocleous, dependent wetlands. Their work pointed out conceptual
2002) carries over intact to the analysis of the drainage and practical problems in the calculation of groundwater
flow-groundwater discharge problem mentioned above. discharge by the DRN Module (e.g., calculating water tables
Specifically, this work was motivated by a particular situa- above the land surface, difficult conductance parameteriza-
tion (Rodrı́guez et al., 2006) encountered when high eleva- tion, and large water balance errors). To overcome these
tion free surface flows that surround a shallow aquifer problems, a new SEEPAGE package for MODFLOW was devel-
overlap with the irrigation season. In this situation, the high oped, which resulted in more accurate results in comparison
river flows interfere with groundwater discharge through with those obtained with the DRN Module. The DRT1 Pack-
the drainage system. age (Banta, 2000) allows to specify a certain fraction of
Drainage of water from the soil profile is essential for the the simulated drain flow to be returned to any cell in the
proper functioning of intensely irrigated agricultural areas system, as opposed to the DRN Module in which drain flow
in semiarid regions to remove excess water and evapocon- was removed from the system. However, channel drain flow
centrated salts from the root zone. Natural drainage from calculations follow those implemented on the DRN Package.
irrigated areas accounts for a portion of unsaturated and Thus, in spite of the considerable progress made in re-
saturated flow to streams and vertical seepage to underlying cent years, a freely available model that couples groundwa-
aquifers. Artificial drainage water is usually discharged by a ter discharge to a drainage network flow, where backwater
network of canals and ditches, a fraction of which stays in effects on the free surface along the channel drains are ta-
the system and eventually builds up the water table. The ken into account is not available. In this study, the open
drainage efficiency depends, among other factors, on the source, highly tested and widely known HEC-RAS computer
interaction with the groundwater flow, the hydraulic condi- program is linked to MODFLOW in an iterative way in order
tions at the drain discharging points, and the maintenance to improve the drain flow–aquifer discharge problem. The
of the drainage network (ILRI, 1994). Modelling provides a main objective of the approach is to obtain a more physi-
rapid analysis tool for obtaining a better understanding of cally realistic hydraulic profile in channel drains within a re-
the behaviour of these complex systems. For example, the gional groundwater flow system. With the aid of a simple
representation of a drainage network in MODFLOW is analytical solution, it is then established that the proposed
accomplished through the DRAIN Module (DRN) (McDonald scheme is numerically convergent and mass conserving.
and Harbaugh, 1988). MODFLOW is a computational code Therefore, the paper is organized as follows: a brief review
that numerically solves the 3D form of the groundwater flow of numerous relevant works accumulated on the subject of
governing equation. The groundwater flow toward drains, stream–aquifer interaction is presented first. A simple
known as drain flow, is assumed to be proportional to the approximate analytical solution for the coupling problem,
Fully conservative coupling of HEC-RAS with MODFLOW to simulate stream–aquifer interactions in a drainage basin 131

which provides the mathematical setting for the proposed et al., 2004). DAFLOW computes unsteady streamflows by
approach is discussed in the section ‘‘Simplified analytical means of diffusive wave routing, where the stream–aqui-
solution’’, following the mathematical model posed in the fer exchange is simulated as a streambed leakage. The
section ‘‘A fully coupled mathematical model’’. Numerical SFR1 Package replaces the former Prudic’s STR1 package
results to the analytical solution obtained with a Piccard (Prudic, 1989) to simulate stream–aquifer interaction with
iteration between HEC-RAS and MODFLOW are covered in MODFLOW-2000 (Harbaugh et al., 2000). Other improve-
the section ‘‘Numerical solution with HEC-RAS/MODFLOW’’. ments directed at increasing MODFLOW capabilities to deal
In the section ‘‘Application to the Choele Choel Island, Pat- with the coupled surface–subsurface water systems are
agonia, Argentina’’, the approach is used to analyze the presented in Panday and Huyakorn (2004), whose code
water table build up problem in the shallow aquifer of the MODHMS is also based upon the diffusive wave approxima-
Choele Choel Island, in the Patagonian region of Argentina, tion of the 1D Saint Venant equations. MODHMS was re-
where the case was first studied with the DRN Module, and cently applied by Werner et al. (2006) to study the
now is addressed with HEC-RAS instead. Conclusions are stream–aquifer interactions in a tropical catchment in
drawn in the section ‘‘Conclusion’’. north-eastern Australia. Sophocleous et al. (1999) linked
the surface water code SWATMOD with MODFLOW to study
Surface water–groundwater interactions the stream–aquifer interactions on a basin in south-cen-
tral Kansas, USA. A similar approach was later followed
It is worth noting that the current study is limited to the by Sophocleous and Perkins (2000) to link SWAT to MOD-
case of groundwater discharge to the surface water and FLOW. The approach taken herein is in tune with the
not the reverse. Nonetheless, the subject of stream/aquifer emphasis given by Sophocleous and Perkins (2000) on the
interactions is reviewed from a broad perspective as the ap- practicality and advantages of using conceptually simple
proach here introduced is general enough to handle both approaches to address integrated dynamic modelling.
cases.
Over the years, numerous approaches have been devel-
oped to tackle stream–aquifer interaction problems. An A fully coupled mathematical model
overview of the state-of-the-art can be found in the recent
work of Sophocleous (2002). Analytical studies of the line- Assuming invariant properties such as hydraulic conductivity
arized 1D Boussinesq equation to describe changes in bank in the saturated porous medium and boundary resistance in
storage caused by temporal variations in water elevation the channel bed, the coupling between the conservation law
of the adjacent channel have been pursued by Cooper of a groundwater flow and the conservation of mass and
and Rorabaugh (1963), Hogarth et al. (1997), Moench and momentum fluxes of an open channel flow, can be posed
Barlow (2000), and Hantush, 2005), whereas its nonlinear mathematically as follows:
counterpart problem has been analyzed by Serrano and
Workman (1998) and Parlange et al. (2000), among others. oU
Sy  r  T r rU ¼ R on X; ð1Þ
The analytical work of Theis (1947) on streamflow deple- oT
tion by pumping was revisited and expanded by Hunt
(1999), among many contributors to the subject. Recently, oA oQ
þ ¼ D on C; ð2Þ
stream–aquifer exchange was also assessed using inverse oT oS
modelling (Szilagyi et al., 2005), and parameter uncer- !
tainty was addressed with stochastic modelling by Srivast- oQ o Q2 oH
þ þ gA ¼ gAðSo  SF Þ on C: ð3Þ
ava et al. (2006). In a multidimensional setting, the oT oS A oS
numerical solution of a fully conservative, integrated
groundwater/surface water modelling was pioneered by Eq. (1) governs the depth averaged flow in the porous med-
Pinder and Sauer (1971), whose simplified solution was ium within the Dupuit–Forchheimer approximation (Bear,
presented years later by Hunt (1990) with the linearized 1972), while Eqs. (2) and (3) are known as the Saint Venant’s
form of the kinematic wave approximation. Numerous equations (Chow et al., 1988), or the long wave approxima-
modelling studies concentrated on capturing the regional tion which is valid whenever the pressure in the water col-
water balance, a trend mainly favoured by the widespread umn is distributed hydrostatically. For the problem
use of the code MODFLOW in combination with the considered here, the lateral interacting flux D per unit
Streamflow Routing (STR1) Package (Prudic, 1989). The length of channel [L2T1] in Eq. (2) is given by
STR1 Package solves a water budget along each stream
D ¼ K½U  ðH þ Zb Þ on C: ð4Þ
reach, where the surface water discharge is computed
with the aid of Manning’s boundary resistance relationship As it will be shown shortly, D can be related to R through a
based on a prevailing normal streamflow assumption that very simple mathematical expression. With reference to
is seldom attained in practice. However, its simplicity Fig. 1, the variables used above are defined as follows: Sy
and mass preserving properties made it the commonly cho- is the aquifer storativity coefficient, U is the aquifer head
sen tool to simulate stream–aquifer interactions. Further [L] above datum as function of the horizontal coordinates
attempts to improve the approximation were made with X = (X, Y) and time T, $ is the 2D horizontal gradient opera-
MODBRANCH (Swain and Wexler, 1996), which essentially tor [L1], Tr is the aquifer transmissivity [L2T1], which de-
recovered the mathematical model of Pinder and Sauer pends upon the aquifer head, R is the net recharge [LT1]
(1971), and with the two well-documented USGS releases from rainfall and/or irrigation applied over the area X
DAFLOW (Jobson and Harbaugh, 1999) and SFR1 (Prudic [L2], K is the reciprocal of the hydraulic impedance of the
132 L.B. Rodriguez et al.

∂Ω g
R
Ω
D
S a Q H A Φ
Y
Γ D
a
n
XΓ(S) Z
Zb
Za Datum
X
a a

Figure 1 Sketch of the conceptual model. (left) Plan view of a regional phreatic aquifer of planar area X discharging onto an open
drain channel. oX represents a portion of X’s external boundary. (right) Cross-section a–a.

streambed material [LT1], H is the water depth [L] in the Γ

stream C, whose location in the horizontal plane is param- n_ n+
eterized with X = XC(S), where S is the arc length [L] in
the streamwise direction, A is the wetted area [L2] of the D
flowing stream cross-section, Q is the volumetric stream dis-
charge [L3T1], So is the channel bed slope, =dZb/dS for a
⇑
fixed bed, where Zb is the streambed elevation [L] above da-
tum measured along the vertical coordinate Z, which is
aligned with the acceleration of gravity g [LT2]. Finally,
SF = sb/qgH is the friction slope for which an external clo-
sure relationship is required, i.e.,
Figure 2 Total flow crossing the control volume from left and
n2 U2
SF ¼ ; ð5Þ right adding to D.
H4=3
in the case where the Manning’s resistance formula is confused with the Manning’s roughness coefficient. If it is
adopted, where n is the roughness coefficient [TL1/3], or now further assumed that D and R are constant, Eq. (7) re-
duces to the elementary mass balance
CF U2
SF ¼ ; ð6Þ RX
gH D¼ ; ð8Þ
Lx
if the more physically based dimensionless bed resistance
coefficient CF = sb/qU2 is preferred, whose estimation can which states the trivial fact that the ultimate goal of an
be obtained with the aid of the Keulegan (1938) relation- open ditch of length Lx is to drain all infiltrated irrigation
ship. Here,sb is the boundary shear stress [ML1T2], q is water in excess of root uptake, evaporation, and soil pore
the water density [ML3], and U = Q/A is the mean cross- water to reach field capacity. This result is used next to
sectional flow velocity [LT1]. set up the simplified closed solution to the interacting
The set of Eqs. (1)–(3) must be supplemented with drain–aquifer flow problem.
appropriate initial and boundary conditions to obtain a un-
ique problem solution and explicitly introduce the interact- Simplified analytical solution
ing flux D in Eq. (1) through a source term. Alternatively,
the flux D can be brought directly into play by applying In order to give ground to the numerical test used to vali-
the 2D divergence theorem to Eq. (1). By doing so, date the iterative coupling between HEC-RAS and MOD-
under the assumptions of steady state flow and a no-flow FLOW, it is sufficient to consider the steady-state solution
boundary on the external portion of oX whenever its inter- to the groundwater discharge problem of an idealized,
nal counterpart C concentrates the flux exchange dynamics homogeneous aquifer defined on 0 < X < Lx, 0 < Y < Ly, bor-
(Fig. 1), the following simple integral form is obtained: dered on one side by a fully penetrating channel drain with
Z Z an uniform rectangular cross-section of width B, running in
 n  T r rU dS ¼ R dX: ð7Þ the X direction only and whose sloping bed obeys
C X
Zb ¼ Zbo  So X (Fig. 3). To that aim, it is convenient first
Here, n denotes the unit vector on C, which can have oppo- to introduce dimensionless variables as follows:
site directions after introducing a slit along C in order to
ðX; Y; Z; H; UÞ TV Tr U
make the boundary of X a simple closed curve (Fig. 2). ðx; y; z; h; /Þ ¼ ; t¼ ; tr ¼ ; u¼ ;
Then, and within the bounds of the Dupuit–Forchheimer B B KB V
hypothesis (Fig. 2), the ambiguity in defining the direction ð9Þ
of n is irrelevant as long as the integrand of the line integral where K is the hydraulic conductivity [LT1] of the aquifer
above is recognized as the exchange flux TroU/on = D, and V is a velocity scale [LT1] given by
with D computed with Eq. (4). Here, the letter n indicating
the directional derivative in normal direction should not be V ¼ ðgSo D=CF Þ1=3 : ð10Þ
Fully conservative coupling of HEC-RAS with MODFLOW to simulate stream–aquifer interactions in a drainage basin 133
 
Y rly rly y
water divide /ðy; xÞ ¼ ho ðxÞ þ zb ðxÞ þ þ y 2 ; ð21Þ
Ly k 2tr ly

no flow
no flow s
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2  ffi
Ω rly y
R /ðy; xÞ ¼ ho ðxÞ þ zb ðxÞ þ þ rly y 2  ; ð22Þ
X k ly
D
Lx
sloping open drain channel B ho ðxÞ ¼ rx 2=3 ; ð23Þ
Ho
So Uo uo ðxÞ ¼ x 1=3 ; ð24Þ
Zbo
1
where the parametric dependence of / with x comes
Figure 3 Layout of the test case. Only the upper part of the through the variation of the water surface elevation along
aquifer is sketched here, the other half extents from Y = 0 to the channel drain. Eqs. (23) and (24) were obtained by
Y =Ly. Woolhiser and Liggett (1967) with classical overland flow
theory, adapted here to the drain flow problem. Eqs. (16)
Then, the dimensionless form of the governing Eqs. (1)–(3) and (17) represent the time-independent kinematic wave
becomes approximation of Lighthill and Whitham (1955) to the full
     equations of motion.
o/ o o/ o o/
e1  tr þ tr ¼ r; ð11Þ The solution of Woolhiser and Liggett (1967), i.e., Eqs.
ot ox ox oy oy
(23) and (24), can be recast in dimensional form as follows:
oh oðuhÞ U2o ðXÞ So
þ ¼ r; ð12Þ F2o ¼ ¼ ; ð25Þ
ot ox gHo ðXÞ CF o

ou ou oh u2 u which represents a strict balance between bed resistance

þ u þ F2
B ¼ F2
B So  CF r ; ð13Þ and gravity, i.e., whereas an increase in So tends to acceler-
ot ox ox h h
ate the flow, an increase in CF o will have the opposite ef-
whereas the dimensionless parameters are given by
fect. Based on the normal flow solution, the Froude
K R D K V number Fo represents an invariant quantity of the motion.
e¼ ; r¼ ; r¼ ; k¼ ; FB ¼ pffiffiffiffiffiffi ; ð14Þ
Sy V K VB K gB The relation between FB and Fo is therefore
where FB could be associated with the Froude number, al- u2o ðxÞ F2B
beit without any dynamical significance. A discussion in F2o ¼ F2B ¼ : ð26Þ
ho ðxÞ r
terms of the actual Froude number is given after describing
the analytical solution to the posed problem. Now, if the Using Manning’s resistance relationship, i.e., Eq. (5) instead
variation in x direction in Eq. (11) under a steady constant of Eq. (6), the normal flow solution would have the form
drain flow discharging into the open channel, and the iner-  3=5 pffiffiffiffiffi3=5  2=5
nD X So DX
tia, pressure, and momentum deficit due to the lateral in- Ho;n ðXÞ ¼ pffiffiffiffiffi ; Uo;n ðXÞ ¼ ; ð27Þ
So B n B
flux on Eq. (13) are all assumed negligible, the reduced
problem termed normal flow, hereafter referred with an under the assumption that Ho/B  1, in which case the
‘‘o’’ subscript is written from Eqs. (11)–(13), and Eqs. (4) hydraulic radius is well approximated by the local water
and (7), as depth. Knowing the value of one resistance coefficient,
  the corresponding value of the other can be computed with
d d/o
tr ffi r; ð15Þ CF ¼ gB1=5 Q 1=5 9=5
ðSo =0:216Þ1=10 , where, according to Eq.
dy dy out n
(8), Qout = D Lx.
d
ðuo ho Þ ¼ r; ð16Þ
dx
Numerical solution with HEC-RAS/MODFLOW
u2o
F2
B So  CF o ffi0 ð17Þ The HEC-RAS code
ho
subject to HEC-RAS is a public domain code developed by the US Army
 
d/o Corp of Engineers (USACE, 2002). It performs 1D steady and
tr ¼ k ½/o ð0Þ  ðho ðxÞ þ zb ðxÞÞ; ð18Þ unsteady flow calculations on a network of natural or man-
dy y¼0
made open channels. Basic input data required by the model
  include the channel network connectivity, cross-section
d/
tr o ¼ 0; ð19Þ geometry, reach lengths, energy loss coefficients, stream
dy y¼ly
junctions information and hydraulic structures data. Cross-
sections are required at representative locations throughout
ðuo ho Þx¼0 ¼ 0: ð20Þ
a stream reach and at locations where changes in discharge,
Eq. (15) is readily solved for the linear case, tr = const., and slope, shape or roughness occur. Boundary conditions are
for the nonlinear case, tr = /o (Bear, 1972), as well, whereas necessary to define the starting water depth at the stream
Eqs. (16) and (17) are simply solved by direct integration system endpoints, i.e., upstream and downstream. Water
134 L.B. Rodriguez et al.

surface profile computations begin upstream for subcritical computed with MODFLOW to obtain a first estimate of drain
flow or downstream for supercritical flow. Discharge infor- flows, D(1) = D(S; U(1), H(0)). It follows that, in order to obtain
mation is required at each cross-section in order to compute an improved approximation to the water elevation distribu-
the water surface profile. If the momentum deficit due to tion on the channel drain, a first Picard iterate is computed
the lateral inflow D is ignored, the non-conservation form with HEC-RAS, namely H(1) = H(S; D(1)). If the procedure is
of the governing equations, under a steady flow assumption, then repeated, a new approximation to the aquifer head
i.e., Eqs. (2) and (3), can be integrated between an up- U(2) = U(X, H(1)) is obtained, which in turn, produces a sec-
stream and a downstream cross-section as ond Picard iterate for drain flows, D(2) = D(S; U(2), H(1)).
Z 2 Z 2 The expectation is that, under suitable conditions, the Pi-
dQ ¼ D dS; ð28Þ card iterates H(k) and U(k) converge to the exact solution
1 1 of the system (1)–(3) as (k) grows. In practical terms, one
Z ! Z or two iterates suffice to reach convergence within some
2
U2 2
prescribed tolerance, as illustrated next. The iterative
d Zb þ H þ a ¼ SF dS; ð29Þ
1 2g 1 scheme can be algorithmically posed as follows:

where a is a momentum correction factor usually set equal Step 1 : ðaÞ Set k ¼ 0; and fix HðkÞ on C;
to one. HEC-RAS solves the resulting integral expressions of
: ðbÞ Solve for Uðkþ1Þ ¼ UðX; HðkÞ Þ on X;
these simple equations by means of an iterative procedure
called the standard step method. The right-hand side of : ðcÞ Compute Dðkþ1Þ ¼ DðS; Uðkþ1Þ ; HðkÞ Þ on C;
Eq. (29) includes contraction or expansion losses as well Step 2 : ðaÞ Fix Dðkþ1Þ ¼ DðS; Uðkþ1Þ ; HðkÞ Þ on C;
as bed resistance losses through an average friction slope
between the two consecutive cross-sections, the later : ðbÞ Compute Hðkþ1Þ ¼ Hðkþ1Þ ðS; Dðkþ1Þ Þ on C;
   
based on the Manning roughness coefficient. Further details Step 3 : ðaÞ If Hðkþ1Þ  HðkÞ =Hðkþ1Þ  htol ! end;
may be found in the HEC-RAS documentation manual
(USACE, 2002). : ðbÞ Otherwise set k ¼ k þ 1; and return to ð1bÞ
ð32Þ
The MODFLOW code
Above, the norm kÆk adopted during the computations rep-
A full description of MODFLOW capabilities can be found in resents the sum of the absolute values of all water depths
McDonald and Harbaugh (1988) and Harbaugh et al. (2000). defined along C.
For the purpose of this work, it is pertinent to revise some
definitions introduced in MODFLOW to compute drain flows Numerical solution
or exchange flows, with the understanding that in the latter
case only groundwater discharge toward drains is consid- First, the solution to the hypothetical stream–aquifer inter-
ered in this work. Nonetheless, the approach is general en- action problem sketched in Fig. 3 and approximated by Eqs.
ough to be applied to exchange flows in both directions. (21)–(24), and (27), was computed with a channel drain
From Eq. (28), for D = const., groundwater discharge toward running from west to east along the aquifer centre. The
a channel drain contained within a MODFLOW cell of size DS aquifer was symmetric with respect to the drain, with
is just DQ = DDS, where DS can either represent ‘‘delr’’ or dimensions Lx = 5000 m and Ly = 1260 m. Channel drain
‘‘delc’’ according to MODFLOW cell size definition (McDon- parameters were B = 5 m, bed elevation at the channel
ald and Harbaugh, 1988). Invoking now the constitutive rela- headwater Zbo ¼ 22 m, So = 4 · 104, and n = 0.05. This va-
tionship that defines drain flow, i.e., Eq. (4), the following lue of Manning’s roughness coefficient can be associated
equivalences between parameters used in MODFLOW and with the resistance encountered in not-well maintained
those introduced in this work are obtained: channels colonized by weeds on their banks, a situation that
resembles the drainage system of a real case application
DQ ¼ Cs ½U  ðH þ Zb Þ; ð30Þ
discussed later. In addition, for a drain flow per unit length
of channel D = 2 · 104 m2 s1, and a net recharge to the
Cs ¼ KDS ¼ K s DSB=es ; ð31Þ
aquifer R = 7.936508 · 108 m s1, it follows from Eq. (8)
where Cs is the hydraulic conductance of the channel drain– that the total outflow at the channel drain endpoint is
aquifer interface [L2T1], Ks is the hydraulic conductivity of Qout = DLx (=RX) = 1 m3 s1. Other parameter values used
the streambed material [LT1], and es is the thickness of the for the test were Tr = 0.014 m2 s1, K = 7 · 104 m s1, and
streambed layer [L]. From these definitions, the hydraulic K = 0.2 m s1. The associated dimensionless parameters
impedance of the channel–aquifer interface is defined as and the velocity scale were: tr = 4, r = 1.1338 · 104,
K1 = es/KsB [TL1]. k = 285.71, r = 1.3887 · 103, FB = 0.0041, Fo = 0.11,
CF = 0.03284, and v = 0.03 m s1. A 3D view of the solution
Picard iterates to couple HEC-RAS with MODFLOW to the hypothetical stream–aquifer interaction problem de-
fined by Eqs. (22) and (23) for these parameter values is de-
The system of Eqs. (1)–(4) can be solved with a sequence of picted in Fig. 4, where the channel width was exaggerated
approximations called Picard iterates. The computation be- for illustration purposes. Now, if D = const., Eqs. (2) and
gins with a crude approximation to the water depth distribu- (3) become uncoupled from Eq. (1), and the governing
tion along the drain channel, namely an initial constant Eqs. (28) and (29) can be solved independently by HEC-
function H(0) = H(0)(S). Then, a solution U(1) = U(X, H(0)) is RAS. Thus, the sensitivity of the model output to variations
Fully conservative coupling of HEC-RAS with MODFLOW to simulate stream–aquifer interactions in a drainage basin 135

Z that location, which was gradually reduced as far as possible

in successive simulations to approach zero. The final value
X
Y
for the simulations was 0.0001 m3 s1. For smaller values,
φ = Φ/B HEC-RAS reported a zero-flow error message. For upstream
inflow rates of 0.0001 and 0.00025 m3 s1 the water depth
5.1
5.5 5.0 computed at the channel headwater was 0.06 and 0.07 m,
4.9 respectively. Similar results were also obtained with a coar-
0 4.8
5.0 4.7 ser grid of 21 cross-sections. After uniformly subtracting the
4.6 initial overshoot, the consistency between numerical results
4.5
500 and the theoretical water profile distribution is shown in
4.5 4.4
4.3 Fig. 5. As explained, the approximately constant overshoot
4.2
1000 4.0 on the computed free surface is in part triggered by the
0 100 200 inadequacy of HEC-RAS in handling a null water depth at
the inlet, though a fraction of it could be attributed to
Figure 4 3D view of the coupled stream–aquifer solution
the dynamic response of the full equation of motion embed-
given by Eqs. (22) and (23).
ded into the solution. The normal flow solution discussed
earlier is strictly based upon the kinematic wave approxima-
in some input parameters such as the grid size and the in- tion. Nevertheless, if the normal flow solution itself is used
flow rate at the upstream end of the channel can be tested to evaluate the relevance of the neglected terms into the
without resorting to the fully coupled problem. balance of forces, i.e., inertia, pressure, and momentum
Initially, the channel length was discretized into 51 rect- deficit, with respect to the retained terms, i.e., gravity
angular cross-sections evenly spaced every 100 m. A lateral and friction, it is rather straightforward to determine that
inflow DQ = DDS = 0.02 m3 s1 was added at the upstream the weight of the neglected terms decays very rapidly with
cross-section of each simulated channel reach. In agree- x1/3. In summary, HEC-RAS capabilities and accuracy to
ment with Eq. (17), the condition SF = So was imposed at reproduce the essential features of the normal flow solution
the downstream boundary. Analytical solutions and com- discussed up to here were deemed acceptable for the pur-
puted results are depicted in Fig. 5. It is seen that the nor- pose of this work, given the fact that the slight overshooting
mal flow solution predicted by Eq. (23) closely follows the of the computed water depth introduces a rather small var-
solution given by Eq. (27) which, in turn, was made dimen- iation into the computed flux exchange D, as explained
sionless against B. The value of CF used for the curve shown next.
in Fig. 5 was obtained with the expression given at the end The approximate analytical solutions given by Eqs. (21),
of Simplified analytical solution, for n = 0.05 and (22) and (27) were then compared with the numerical re-
Qout = 1 m3 s1. The excellent agreement between the solid sults obtained when HEC-RAS and MODFLOW were itera-
line (Manning’s n) and the dash–dot line (Keulegan’s CF) tively coupled following the algorithm (32). The MODFLOW
indicates that in-channel water depth computations are grid consisted of 20 columns 250 m wide (mx = 20) and 21
quite insensitive to the type of hydraulic resistance law. rows 120 m wide (my = 21). The drain channel was located
On the other hand, a slight departure or overshoot in the along the centre row (i = 11) and represented by 20 MOD-
computed free surface is clearly noticeable in Fig. 5. Part FLOW drain cells. HEC-RAS channel discretization amounted
of the problem is originated by the fact that the upstream for 21 rectangular cross-sections located at the boundary
boundary condition given by Eq. (20) cannot be exactly rep- between two contiguous MODFLOW drain cells, except for
licated by HEC-RAS. In other words, the integral form of Eq. the first and last ones. Therefore, they were staggered with
(28), Q2 = Q1 + DDS, requires a non-trivial inflow value Q1 at the centre of MODFLOW drain cells. The exact value of the
the uppermost stream cross-section. Consequently, a hypo- groundwater flow discharging from each MODFLOW cell into
thetical inflow rate equal to 0.00025 m3 s1 was imposed at the channel drain, computed from Eq. (30), was
DQ ¼ DDX ð¼ Q out =mx Þ ¼ 0:05 m3 s1 . The hydraulic conduc-
tance of the channel–aquifer interface, Cs, was equal to
50 m2 s1. From Eq. (31), it follows that this value of Cs cor-
4.5
responds to a hydraulic conductivity of the streambed mate-
4.4 rial of 0.04 m s1, for a streambed layer thickness of 1 m.
The aquifer bottom was made coincident with the datum,
z = Z/B

4.3 set equal to zero, while the aquifer top was set equal to
zb(s)
ho(s),
25 m. MODFLOW computations were first restricted to the
4.2 CFo= 0.03284
linear- or confined-aquifer case. Fig. 6 shows the Picard
ho(s) , n = 0.050
ho(s) iterates of the water surface profiles computed with HEC-
4.1 HEC-RAS
RAS, starting from an initial guess H(0) = 0.30 m. Each HEC-
HEC-RAS, corrected
4.0 RAS computation was obtained after passing onto each of
the channel reaches the cumulative drain flow computed
0 200 400 600 800 1000
s = S/B
by MODFLOW. The first and second Picard iterates are
practically indistinguishable to the naked eye, obtained
Figure 5 Comparison between computed ho(x) with HEC-RAS, after imposing upstream and downstream boundary condi-
for D = const., and analytical distributions obtained for both tions as previously discussed. The corresponding water
bed resistance relations discussed in this work. depth calculated at X = 0 m was 0.07 m. The overshoot is
136 L.B. Rodriguez et al.

4.5 Table 1 Evolution of the average drain flux computed by

MODFLOW (minus sign obeys convention) and the in-channel
4.4
water depth error calculated by HEC-RAS
Picard iterate 1st 2nd Corrected
z = Z/B

4.3 zb(x) P
ho(x), n = 0.050 DQ j
(m3 s1) 0.0500575 0.0500580 0.0500530
4.2 (0) mx
ho P ðkþ1Þ ðkÞ
(1)
ho ho(x) jH H j
Pj ðKþ1Þ j
4.1 0.4819561 0.0045554
h(2)
o
jHj j
(2)
4.0 ho - corr
0 200 400 600 800 1000
x = X/B puted by MODFLOW, are given in Table 1. Here, the error
Figure 6 Variation of the Picard iterates h obtained after (k) tolerance mentioned in the three-step algorithm (32) was
coupling HEC-RAS with MODFLOW HEC-RAS solutions were fixed at 0.01 and the overall mass balance error reported
obtained by accumulating the value of D(k+1)(X; U(k+1) ,H(k)). by MODFLOW was less than 0.1%.
along drain channel reaches. Aquifer head contours computed with MODFLOW were
compared with those obtained with Eq. (22) in Fig. 8 for a
nonlinear (unconfined) aquifer case. For the linear case,
systematically propagated in a more or less constant man- the results are similar and were omitted here for the sake
ner in the downstream direction. It can be seen in Fig. 6 that
once that overshoot is subtracted from the Picard second
300
iterate, the computed water surface profile closely resem-
bles the normal flow solution given by Eq. (27). For each Pi-

ANALYTICAL
200 5.0
card iteration, the steady-state MODFLOW simulation
started from a constant initial aquifer head equal to 25 m. 100 4.8
4.6
The groundwater flow simulation required, on average, 11
y = Y/B

0 4.4
iterations to converge using a head closure criterion of 4.4

MODFLOW
0.0001 m and the Strongly Implicit Procedure-SIP solver. 4.6
-100 4.8
The cell drain flux values computed with MODFLOW for all 5.0
Picard iterates as well as their relative error expressed in -200
percentage are plotted in Fig. 7. An end-effect is clearly
present at both ends of the drain channel. One way to elim- -300
inate this effect is by locally averaging the flux within neigh- 0 200 400 600 800 1000
bouring cells, since the average flux is quite close to the x=X/B
exact value (see Table 1). Both, cell values as computed
by MODFLOW (Fig. 7) and their average were used without Figure 8 Contour lines of the dimensionless aquifer head,
any noticeable difference in the final solution. The evolu- nonlinear case. (upper half) Approximate analytical solution
tion of the error closure criteria for the in-channel water given by Eq. (22). (lower half) Solution computed by MODFLOW.
depth results, along with the average exchange flux com-

-0.04 st 25
1 iterate 5.2
2nd iterate, and 2ndcorr iter
exact 20
error ( % ) = (ΔQnum - ΔQexa ) / ΔQexa

-0.05 5.0
15

4.8
10
ΔQ num [m 2/s]

-0.06
φ = Φ/B

Analytical
Linear

5
4.6 x = 575
-0.07 x = 775
0 x = 975
4.4
Non-linear

Analytical
(1) -5
error x = 575
-0.08
(2) x = 775
error 4.2 x = 975
-10
(2)
error corr
0 50 100 150 200 250
-0.09 -15
0 200 400 600 800 1000 y = Y/B
x = X/B
Figure 9 Comparison between the approximated analytical
Figure 7 Cell drain flux values as computed by MODFLOW – solutions and the results computed with MODFLOW of the
error distribution of the computed flux, in percentage. aquifer head profiles in the y direction for x = const.
Fully conservative coupling of HEC-RAS with MODFLOW to simulate stream–aquifer interactions in a drainage basin 137

of brevity. Actually, three transverse aquifer profiles at se-

lected x positions are shown in Fig. 9 for both, linear and 1.0
nonlinear cases. The matching degrades as x increases, pre-
serving a remarkable fit throughout the domain close to the
channel boundary, where drain fluxes are computed. Part of
0.8
the discrepancies were triggered by the conditions MOD-
FLOW must meet at the no-flow boundary at Y = Ly given
the coarseness of the grid used. A better agreement at
0.6
the boundary was obtained with a finer grid composed by
mx = 50 columns and my = 41 rows, matching the HEC-RAS κ
calculations depicted in Fig. 5. Nevertheless, exchange
0.4
fluxes obtained with the finer grid were essentially indistin-
guishable from fluxes obtained with the coarser grid plotted
in Fig. 7, therefore the results could be considered grid-size
independent from a flux perspective. Obtaining similar re- 0.2

sults with different grids and validating the calculations be-

tween HEC-RAS and MODFLOW against analytical solutions
with a grid size similar to the grid used later on a real case 0.0
application were considered an essential aspect of the
study. 10
-6
10
-5
10
-4
10
-3
10
-2
10
-1
10
0
10
1

2 -1
Cs [ m s ]

Model sensitivity Figure 10 Variation of the sensitivity coefficient for decreas-

ing values of the hydraulic conductance of the drain–aquifer
Unfortunately, the approximate analytical solution defined interface.
by Eqs. (21)–(24), and (27) is not very useful to explore
model sensitivity, as it will become clear in a moment.
to be sensitive if j > 1, neutral if j = 0, and robust if j < 1.
HEC-RAS is known to be sensitive to channel width B and,
to a lesser extent, to the roughness coefficient n, whereas The variation of the sensitivity coefficient as a function of
MODFLOW is not very sensitive to the hydraulic conductance Cs depicted in Fig. 10 diminishes in the upper half of the
that controls the exchange flux DQ = DDS (Chen and Chen, drain channel. The sensitivity coefficient was less than
2003). The stiffness of the groundwater model response to one for the whole range of values of Cs explored. The model
changes in the hydraulic conductance can be understood was quite insensitive until small values of the hydraulic con-
from Eq. (8). This equation shows that the exchange flux ductance were reached. For all practical purposes, and for
is indeed independent of any groundwater model parame- the two smallest values of Cs shown in Fig. 10, the model be-
ter, unless the hydraulic conductance Ris regionally variable, haves as if the upper half of the drain channel was clogged,
in which case the mass balance reads C DdS ¼ RX. The ex- discharging the whole recharge on the lower half of the
drain at twice the rate in comparison with the base rate.
change flux DQ given by Eq. (30) is considered the quantity
of interest for the proposed approach. Consequently, in or-
der to analyze changes in DQ with respect to the base state Application to the Choele Choel Island,
or normal flow condition studied before, caused by changes Patagonia, Argentina
in the hydraulic conductance, a set of simple numerical
experiments were run with a fixed in-channel water surface The approach was applied to analyze the drainage problem
(i.e., the corrected water depth depicted in Fig. 6). Then, of the shallow aquifer of the Choele Choel Island, located in
the hydraulic conductance on the upper half of the drain the Patagonia region of Argentina. The Negro river, in the
channel was reduced one order of magnitude on each simu- Argentinean Patagonia, originates at the confluence of the
lation, whereas the lower half always preserved the origi- Neuquén and Limay rivers (Fig. 11). After traversing the
nally assigned value of 50 m2 s1. Since the computed Upper Valley, the Negro river enters the gently sloping Mid-
solution developed an increasing dependence with X for dle valley to continue through the Lower valley toward its
decreasing values of Cs on the upper half of the drain, the outlet in the Atlantic Ocean. The Choele Choel Island lies
approximate analytical solution was not helpful. Neverthe- at the bifurcation of the Negro river into its North and South
less, from elementary differential calculus it is straightfor- branches. The island is approximately 40 km long with a
ward to establish that the net effect of a departure dCs from maximum width of 15 km, and encompasses around
the base state Cs on the model output would be dDQ = dCs 34,000 ha, including Chica Island and other small islets.
oDQ/oCs, expression that can be written in terms of the The Choele Choel Island longitudinal slope is 5.8 · 104. In
absolute value of the relative change as the transverse direction, the island slopes gently from the
South Branch toward the North Branch. Summer tempera-
jdDQ =DQ o j tures average 23 C during January, while winter tempera-
j¼ ; ð33Þ
jdCs =Cso j tures average 6.8 C. Maximum temperatures of 30 C are
common in summer months. Below freezing temperatures
which is known as the sensitivity coefficient or condition occur in June, July, and August. The average annual precip-
number in the jargon of numerical analysis. A model is said itation is about 300 mm. Rainfall is unevenly distributed in
138 L.B. Rodriguez et al.

Figure 11 Study area.

space and time, as is typical in semiarid climates. The Mid- work. Irrigation is by gravity, with an application fre-
dle Valley was carved in the Patagonian Plateau by glaciers. quency of about two to three weeks, on average.
Three geologic units form the Valley’s stratigraphic profile: Natural drainage is supplemented by about 100 km of
(1) Negro river Formation (Rionegrense): the lowermost por- drain canals of different sizes that remove excess water
tion of the profile, composed by alternate layers of sand, accumulated during the irrigation season. Drainage water
silt, and clay of moderate to low permeability. Despite its is then discharged at downstream locations at both river
thickness, it is of little hydrogeologic interest; (2) Rodados branches. Small canals are usually poorly maintained, sil-
Patagónicos: composed of conglomerates and boulders, usu- ted up and colonized by weeds, while major canals are
ally cemented with calcium carbonate. These beds vary in periodically dredged. Fig. 12 shows the configuration of
thickness from 3 to 5 m and extend over the terraces sur- the drainage network.
rounding the valley; (3) Relleno Moderno: overlies the Rio- During the irrigation season, which extends from late Au-
negrense in the valley centre. It is mainly composed of a gust to early April, seepage losses through unlined distribu-
thin layer of silty sand, underlain by gravel deposits with tion canals and in irrigated fields cause water table
variable quantities of sand, and is characterized by high per- mounding and/or soil water logging at some locations.
meability. This is the main water-bearing unit of the profile. Moreover, high stream levels caused by high water releases
Unconfined in most of the area, it is highly dynamic and for hydroelectric power generation at upstream dams during
interacts closely with all surface water sources and sinks. the peak of the irrigation season interfere with free ground-
Transmissivity values for this aquifer range from 200 to water drainage causing backwater effects at some drainage
2500 m2 day1, while reported values for specific yield vary canals discharge points.
between 0.01 and 0.2. Field studies have shown that there is
no hydraulic connection between the Relleno Moderno and MODFLOW – HEC-RAS set up
the underlying formation. The North Branch conducts more
than 90% of the upstream flow. Both branches act as natural MODFLOW was implemented to simulate steady-state
drainage canals, providing over 155 km of stream–aquifer groundwater flow in the island during the non-irrigation sea-
contact that allows free groundwater drainage. The South son, when only drainage canals and streams are active. This
Branch is a highly meandering stream approximately 87 km simulation provided the first set of drain flows later used as
long, while the North Branch is 68 km long. lateral inflows to HEC-RAS. The finite difference grid con-
The local economy is based on the production of fruits tained 134 rows and 64 columns, oriented along the regional
and vegetables, sustained by an irrigation/drainage sys- groundwater flow direction, with a regular cell size of
tem. Irrigation water enters the island through an unlined 300 m. A single model layer, 20 m thick, representing the
19 km long channel, starting at the western corner of the unconfined Relleno Moderno was simulated. More details
island, 3.5 km downstream from the bifurcation point in regarding the model set-up and calibration results can be
the Negro river branches. Eight secondary, unlined chan- found in Rodrı́guez et al. (2006)); only drain related param-
nels 89.8 km long, and minor channels that reach outlying eters and results are summarized here for the purpose of
irrigated fields (61.2 km long) complete the irrigation net- this work.
Fully conservative coupling of HEC-RAS with MODFLOW to simulate stream–aquifer interactions in a drainage basin 139

Figure 12 Drain channels network of the Choele Choel Island, location of field data points.

The drainage system shown in Fig. 12 was simulated with height readings from gauging stations located upstream and
324 drain cells according to the DRN Package input require- downstream of the discharge points.
ments. The so-called System I drains the Southeastern por- Calibrated values for drain conductance ranged from
tion of the island and discharges into the South Branch of 0.003 to 0.0068 m2 s1. The total simulated drain flow was
the Negro river, System II drains the North-western and cen- 0.726 m3 s1, of which 0.247 m3 s1 (34%) were drained by
tral portions of the island and discharges into the North System I and 0.479 m3 s1 (66%) by System II. Not much
Branch of the Negro river. Water depth data for drain chan- information existed to assess the model performance
nels were available only at a few locations, hence it was regarding the magnitude and distribution of drain flows.
necessary to estimate initial values in order to implement However, agricultural engineers working at the site have
the DRN Package. Based on previous knowledge of the site roughly estimated a total drainage flow around 0.80 m3 s1
and those few data points, water depth at drain channels for the non-irrigation season. Therefore, using this esti-
headwaters was assumed to be about 0.2–0.3 m. These mate, simulated drain flows were underestimated by
depths were progressively increased downstream until a roughly 9.25%.
water depth of 1.80, 0.9, and 1.14 m was reached at dis- A total of 101.25 km of drain canals were simulated with
charge points of Drain IV, Drain I and the Gran Zanjón, HEC-RAS. The network was discretized with 535 cross-sec-
respectively (see Fig. 12 for their location). Continuity of tions distributed according to the hierarchy of drain chan-
water levels between stream channels and discharging drain nels, bottom slope, observed changes in cross-sections as
channels was preserved at discharge points. The stage at well as HEC-RAS computational requirements to warrant
the receiving stream was estimated by interpolating stream the energy balance for the calculation of water depths
140 L.B. Rodriguez et al.

between two consecutive cross-sections. The bottom slope

was obtained from topographic data, whereas all cross-sec-
a
tions were assumed rectangular-shaped. Channel widths
0
were assigned based upon the analysis of photographs taken
at the site and field surveys at selected locations. Due to

Drain flux difference (m /s)

deficient maintenance, flow conveyance is significantly re-

3
duced by weeds colonies and other plants at some locations,
mainly in upstream reaches of drain channels. These condi- -0.001
tions prevail along some reaches of Drain II and the Gran
Zanjón. Following the description of channel characteristics
given by Chow (1959), a Manning’s roughness coefficient of
0.07 was adopted for those reaches, while a value of 0.055
was used for the rest of the network. -0.002

Picard iterates scheme 1st iterate - init.guess

2nd iterate - 1st iterate
As explained before, drain flows obtained with the DRN Pack-
-0.003
age at each MODFLOW drain cell were integrated according 14 12 10 8 6 4 2 0
to the HEC-RAS cross-sections distribution and converted Distance from discharge point (km)
to lateral flows to start up a HEC-RAS simulation, and hence,
the iterative process. HEC-RAS calculations were performed b
under a subcritical flow regime, imposing water surface ele-
vations at drain canals discharge points into both streams. 0
Based upon stream height records for both branches of the
Negro river, the water surface elevation at the last cross- Drain flux difference (m /s)
section of the Gran Zanjón, Drain I and Reach 2, Drain IV
3

was fixed at 122, 119.7, and 116.8 m, respectively, equiva-

lent to water depths of 1.14, 0.9, and 1.8 m. -0.001
Following the algorithm of Eq. (32), convergence was
reached at the end of the second Picard iterate for a pre-
scribed tolerance of 0.03 m. Stream stages and stream–
aquifer interaction fluxes along both river branches were
-0.002
unaffected by the implementation of the iterative process.
However, minor local adjustments to the aquifer hydraulic
1st iterate - init.guess
conductivity were introduced at the end of the process for
2nd iterate - 1st iterate
a fine-tuning calibration of aquifer heads adjacent to the
drainage network. The total drain flow at the end of the -0.003
18 16 14 12 10 8 6 4 2 0
iterative process was 0.8369 m3 s1, 34% were drained by
System I and 66% by System II. Simulated total drain flows Distance from discharge point (km)
overestimated previous calculations by agricultural engi- Figure 13 Drain flow differences between successive Picard
neers by 4.6%, compared to the 9.25% underestimation ob- iterates: (a) Drain V; (b) Gran Zanjón.

Table 2 Comparison between observed and simulated water depths and drain flows at selected points (see Fig. 12 for their
location)
Location Dist (m) Hobs (m) Hsim (m) ErrH % Dobs (m3 s1) Dsim (m3 s1) ErrD (%)
A 4858 0.22 0.16 27 0.024 0.03 25
R 1192 0.5 0.35 30
F 3968 0.8 0.57 28 0.337 0.33 2
D 11,362 0.3 0.21 30
D1 9751 0.3 0.25 17 0.05 0.04 20
G 3866 0.55 0.24 50 0.065 0.05 23
H 1211 0.65 0.27 54
B 9129 0.3 0.43 30
C 7854 0.35 0.42 20 0.095 0.12 26
I 6147 0.9 0.67 25
L 6611 0.4 0.2 50
M 310 0.6 0.6 0
N 1744 1 1.04 4
P 11,196 0.25 0.28 12
ErrH,D = [H,Dobs  H,Dsim]/H,Dobs. Dist, cumulative distance from the discharge point.
Fully conservative coupling of HEC-RAS with MODFLOW to simulate stream–aquifer interactions in a drainage basin 141

124 and several junctions from smaller drain channels adding

complexity to the simulation of the channel network.
Finally, Fig. 14 compares water surface profiles along
122 Drain V-Reach 2 and Drain IV until its discharge into the South
Branch for all Picard iterates. HEC-RAS gradually varying
Elevation (ma.m.s.l.)

flow calculations improve the curvature of the hydraulic pro-

120 file upstream and at the junction of Drain IV with Drain V.

Outlet into Conclusions

South Branch
118
Coupling public domain and standard models to extract the
best of each individual model components is a modelling ap-
116 MODFLOW proach successfully pursued by many researchers. In this
Bottom
Joint
work HEC-RAS and MODFLOW have been iteratively coupled
HEC-RAS Drain V - Drain IV in an effort to improve the representation of hydraulic pro-
114 files in drain channels within a regional groundwater flow
14 12 10 8 6 4 2 0 system. With the aid of a simple analytical solution, it was
Distance from discharge point (km) established that the proposed scheme is numerically con-
vergent and mass conserving. That was accomplished by
Figure 14 Simulated water surface profiles along Drain V– developing a simple approximate analytical solution for
Reach 2 and Drain IV. the coupling problem, which provided the mathematical
setting for the proposed approach. Numerical results to
tained prior to the implementation of the Picard iterate the analytical solution obtained with a Picard iteration be-
procedure. tween HEC-RAS and MODFLOW were extensively analyzed.
Table 2 contains water depth and channel discharge The iterative process works as follows: drain flows result-
measured at selected points during the non-irrigation sea- ing from a MODFLOW run with its DRAIN Package (DRN) dri-
son (see Fig. 12 for points location) and corresponding sim- ven by an initial guess of water depth on the drains
ulated values. The application of the iterative modelling become lateral flows input to HEC-RAS. Then HEC-RAS is
procedure was successful at reproducing drain flow patterns run with appropriate geometric data to obtain a set of water
and water depths. Relative errors between observed and depths in drain canals. This new water depths are introduced
simulated drain flows fell between 2% and 26%. These val- back into MODFLOW DRN Package for a new MODFLOW run,
ues are considered satisfactory given the amount and qual- from which a new set of drain flows is obtained. These drain
ity of the field data available. On the other hand, simulated flows are input back into HEC-RAS as lateral flows and a new
water depths with HEC-RAS were reasonably close to ob- simulation is performed. Successive iterates are repeated
served values, though in practical terms they were similar until convergence, measured in terms of water depths, is
to the first guess input to MODFLOW DRN Package at check achieved. The procedure was applied in the Choele Choel Is-
locations. In general, the errors between observed and sim- land, Argentina, where scarce information regarding the
ulated water depths were bracketed between 0% and 50%, drainage system obligated a tedious interpolation of few
with the best results obtained for Drains IV and V. water depths to implement the MODFLOW DRN package.
Fig. 13a illustrates the differences between drain flows The approach not only provides a more sound hydraulic pro-
along Drain V calculated at successive iterates. Cumulative file along drain canals for a wide range of downstream
distances are measured from the drain discharge point up- hydraulic conditions, but also could mean a considerable
stream. Even though changes between the 1st and 2nd iter- time saving in the burdensome task of specifying water
ates are very small, it is worth noting the differences depths along a large and complex drainage system with lim-
between drain flows obtained with the first guess of water ited field data. It is recognized that this particular study was
depths input to the DRN package and drain flows obtained limited to the case of groundwater discharge to the surface
after the 1st iterate. The greatest differences are mainly water ant not the reverse, though the iterative procedure
concentrated along a 5 km reach downstream from the can be equally implemented for a stream/aquifer interac-
drain headwaters, i.e., the reach where most of the uncer- tion scenario replacing the DRN package by the RIV package.
tainly regarding drain channel information used to define Finally, the approach was developed under steady-state con-
initial MODFLOW drain heads was located. As indicated by ditions, and therefore care should be exerted to extend the
the negative differences, drain flows obtained with the first methodology to transient scenarios, in particular to handle
guess were less than those obtained with water depths cal- different time scaling and potential HEC-RAS instabilities.
culated with HEC-RAS. A similar analysis was repeated for
the Gran Zanjón (Fig. 13b). In this case, the maximum dif-
ferences were concentrated on a 2 km reach near the chan-
nel headwaters and along a 7 km reach located Acknowledgements
approximately at mid-distance between the channel end-
points. Similarly to Drain V, the first few kilometres entail This material is based upon work supported in part by the
great uncertainties due to the lack of field data. As for Rio Negro State Water Authority, Universidad Nacional del
the mid-channel location along the Gran Zanjón, it is char- Litoral, and Research Grant PICT 07-14721 of ANPCyT, all
acterized by a non-straight reach, channel width changes from Argentina, and by SAHRA (Sustainability of semi-Arid
142 L.B. Rodriguez et al.

Hydrology and Riparian Areas) under the STC Program of the Panday, S., Huyakorn, P.S., 2004. A fully coupled physically-based
National Science Foundation, Agreement No. EAR-9876800. spatially distributed model for evaluating surface/subsurface
The research was also supported in part by the Department flow. Advances in Water Resources 27, 361–382.
of Hydrology and Water Resources of the University of Ari- Parlange, J.-Y., Hogarth, W.L., Govindaraju, R.S., Parlange, M.B.,
zona. Thanks are also given to two anonymous reviewers Lockington, D., 2000. On an exact analytical solution of the
Boussinesq equation. Transport in Porous Media 39, 339–
that helped to improve the manuscript.
345.
Pinder, G.F., Sauer, S.P., 1971. Numerical simulation of a flood
References wave modification due to bank storage effects. Water Resources
Research 7 (1), 63–70.
Banta, E.R., 2000. MODFLOW-2000, The US Geological Survey Pohll, G.M., Guitjens, J.C., 1994. Modelling regional flow and flow
modular ground-water model. Documentation of packages for to drains. Journal of Irrigation Drainage Engineering ASCE 120
simulating evapotranspiration with a segmented function (ETS1) (5), 71–82.
and drains with return flow (DRT1). Open-File Report 00-466. Prudic, D.E., 1989. Documentation of a computer program to
Batelaan, O., De Smedt, F., 2004. SEEPAGE, a new MODFLOW DRAIN simulate stream–aquifer relations using a modular, finite-
package. Groundwater 42 (4), 576–588. difference, ground-water model. US Geological Survey Open-
Bear, J., 1972. Dynamics of Fluids in Porous Media. Dover, New File Report 88-729, 113 pp.
York. Prudic, D.E., Konikow, L.E., Banta, E.R., 2004. A new streamflow-
Carslaw, H.S., Jaeger, J.C., 1959. Conduction of Heat in Solids, 2nd routing (SFR1) package to simulate stream-aquifer with MOD-
ed. Oxford Science, 510 pp. FLOW 2000. US Geological Survey Open-File Report 2004-1042,
Chen, X., Chen, X., 2003. Sensitivity analysis and determination of 104 pp.
streambed leakance and aquifer hydraulic properties. Journal of Rodrı́guez, L.B., Cello, P.A., Vionnet, C.A., 2006. Modelling
Hydrology 284, 270–284. stream–aquifer interactions in a shallow aquifer, Choele Choel
Chow, V.T., 1959. Open Channel Hydraulics. McGraw-Hill, New Island, Patagonia, Argentina. Hydrogeology Journal 14, 591–
York. 602.
Chow, V.T., Maidment, D., Mays, L.W., 1988. Applied Hydrology. Serrano, S.E., Workman, S.R., 1998. Modelling transient stream/
McGraw-Hill, 572 pp. aquifer interaction with the non-linear Boussinesq and its
Cooper, H.H., Rorabaugh, M.I., 1963. Ground-water movements and analytical solution. Journal of Hydrology 206, 245–255.
bank storage due to flood stages in surface streams. US Skaggs, R.W., Breve, M.A., Gilliam, J.W., 1995. Predicting effects
Geological Survey Water Supply Paper, 343–366. of water table management of loss of nitrogen from poorly
Hantush, M.M., 2005. Modelling stream–aquifer interactions with drained soils. European Journal of Agronomy 4 (4), 441–
linear response functions. Journal of Hydrology 311, 59–79. 451.
Harbaugh, A.W., Banta, E.R., Hill, M.C., McDonald, M.G., 2000. Sophocleous, M., 2002. Interactions between groundwater and
MODFLOW-2000, The US Geological Survey modular ground- surface water: the state of the science. Hydrogeology Journal 10
water model – User guide to modularization concepts and the (1), 52–67.
Ground-Water Flow Process: US Geological Survey Open-File Sophocleous, M., Perkins, S.P., 2000. Methodology and application
Report 00-92, 121 pp. of combined watershed and ground-water models in Kansas.
Hogarth, W.L., Govindaraju, R.S., Parlange, J.Y., Koelliker, J.K., Journal of Hydrology 236, 185–201.
1997. Linearized Boussinesq equation for modelling bank storage Sophocleous, M.A., Koelliker, J.K., Govindaraju, R.S., Birdie, T.,
– a correction. Journal of Hydrology 198, 377–385. Ramireddygari, S.R., Perkins, S.P., 1999. Integrated numerical
Hunt, B., 1990. An approximation for the bank storage effect. modelling for basin-wide water management: the case of the
Water Resources Research 26 (11), 2769–2775. Rattlesnake Creek basin in south-central Kansas. Journal of
Hunt, B., 1999. Unsteady stream depletion from groundwater Hydrology 1999, 179–196.
pumping. Ground Water 37, 98–102. Srivastava, K., Serrano, S.E., Workman, S.R., 2006. Stochastic
ILRI (International Institute for Land Reclamation and Improve- modelling of transient stream–aquifer interaction with the
ment), 1994. Drainage principles and applications. In: Ritzema, nonlinear Boussinesq equation. Journal of Hydrology 328, 538–
H.P. (editor-in-Chief), ILRI Publication 16, 2nd ed. (Completely 547.
Revised), Wageningen, The Netherlands. Swain, E.D., Wexler, E.J., 1996. A coupled surface-water and
Jobson, H.E., Harbaugh, A.W., 1999. Modification to the diffusion ground-water flow model for simulation of stream-aquifer
analogy surface-water flow model (DAFLOW) for coupling to the interaction. US Geological Survey Techniques of Water-
modular finite-difference ground-water flow model (MODFLOW), Resources Investigations, Book 6, 125 pp, Chapter A6.
US Geological Survey Open-File Report 99-217, 59 pp. Szilagyi, J., Parlange, M.B., Balint, G., 2005. Assessing stream–
Johnson, M., Koenig, R., 2003. Solutions to soil problems. III. aquifer interactions through inverse modelling of flow routing.
Drainage. AG/Soils/2003/03. Utah State University. Journal of Hydrology 327, 208–218.
Keulegan, G.H., 1938. Laws of turbulent flow in open channels. Theis, C.V., 1947. The effect of a well on the flow of a nearby
National Bureau of Standards Research Paper RP 1151, USA. stream. Transactions of the American Geophysical Union 22 (3),
Lighthill, M.J., Whitham, G.B., 1955. On kinematic waves. I. Flood 734–738.
movements in long rivers. Proceedings of the Royal Society of USACE, 2002. HEC-RAS, River Analysis System. US Army Corps of
London, Series A 229 (1178), 281–316. Engineers. Hydrologic Engineering Center, v3.1, Nov 2002, CPD-
McDonald, M.G., Harbaugh, A.W., 1988. A modular three-dimen- 68.
sional finite difference ground-water flow model. US Geological Werner, A.D., Gallagher, M.R., Weeks, S.W., 2006. Regional-scale,
Survey Techniques of Water-Resources Investigations, Book 6, fully coupled modelling stream–aquifer interaction in a tropical
Chapter A1, 586 pp. catchment. Journal of Hydrology 328, 497–510.
Moench, A.F., Barlow, P.M., 2000. Aquifer response to stream-stage Woolhiser, D.A., Liggett, J.A., 1967. Unsteady, one-dimensional
and recharge variations. I. Analytical step-response functions. flow over a plane – the rising hydrograph. Water Resources
Journal of Hydrology 230, 192–210. Research 3 (3), 753–771.