Sie sind auf Seite 1von 52

4A

Chemical and physical changes due to water estimation of their distribution are briefly examined. Applica-
tions of this work to discontinuum and continuum modelling
931022 of jointed rock masses and seepage flow analysis are outlined.
Groundwater flow in a compressible unconfined aquifer with The geotomography technique is described.
uniform circular recharge
Zlotnik, V; Ledder, G 931026
Water Resour Res V28, N6, June 1992, P1619-1630 Evaluation of the effect of morphological features of flow
paths on solute transport by using fractal dimensions of
The distribution of the hydraulic head and its gradient (veloc-
methylene blue staining patterns
ity vector) for transient groundwater flow in an unconfined
Hatano, R; Kawamura, N; Ikeda, J; Sakuma, T
compressible aquifer under a circular source of recharge have
Geoderma V53, N1/2, May 1992, P31-44
been obtained from the linearised mathematical model by the
use of the integral transforms. The result generalises the solu- Fractai dimensions were obtained of the methylene blue stain-
tion of Dagan (1967) which neglected compressibility. Results ing patterns in 5 undisturbed soil columns. These values were
indicate that Dagan's formulation is accurate at large times, examined in relation to Brenner numbers for chloride break-
but gives large relative errors (small absolute errors) at small through measured under saturated steady state conditions.
times. Values of D of the internal structure of staining patterns
decreased with increasing depth, whilst fractal dimensions of
931023 the perimeter of stains remained relatively constant. An empir-
Finite layer method for groundwater flow models ical equation was developed relating Brenner number for chlo-
Smith, S S; Allen, M B ride breakthrough to depth averages of both fractal
Water Resour Res II28, N6, June 1992, P1715-1722 dimensions and macroporosity.
The finite layer method is an extension of the finite strip
method familiar in structural engineering. Discretization in 931027
two dimensions is by truncated Fourier series, variations in the Automatic orientation mapping of some types of soil fabric
third being approximated by finite elements. The method Tovey, N K; Smart, P; Hounslow, M W; Leng, X L
reduces 3D problems to sets of independent matrix equations, Geoderma V53, N3/4, June 1992, P179-200
solvable sequentially by PC or concurrently by parallel proces- A method is described which automatically delineates areas of
sor. It is thus suitable for computationally intensive problems micrographs of soils which have similarly oriented
such as optimization or the inverse problem. The F L M is microfabrics. It requires two passes over a digitized image. In
demonstrated applied to 4 groundwater flow models: single the first, an intensity gradient algorithm is used to specify the
fully-penetrating well; point source injection; single well in a orientation at each pixel. A second pass then examines the
leaky aquifer; and multiwell field. values attributed to each pixel, and if one general orientation
is dominant in the neighborhood of one pixel, the central pixel
is assigned an appropriate grey scale value. The method has
applications to study of fabric development due to mechanical
Properties of Rocks and Soils or environmental effects.

931024 931028
Rock mechanics contributions from defense programs Techniques to quantitatively study the microfabric of soils
Heuze, F Tovey, N K; Krinsley, D H; Dent, D L; Corbett, W M
Proc 33rd US Symposium on Rock Mechanics, Santa Fe, 3-5 Geoderma V53, N3/4, June 1992, P217-235
June 1992 P3-26. Publ Rotterdam." A A Balkema, 1992 A powerful new technique to examine the mineralogy and
Finance from defence programs has played an important role microstructure of soils has been developed by combining sev-
in advancement of rock mechanics over the past 3 decades. eral existing image processing and analysis methods. It begins
Major areas of work are summarised: dynamic instrumenta- with multispectral classification of SEM backscattered elec-
tion; measurement of rock properties; physical modelling; tron images and X-ray maps, followed by particle size routines
development of constitutive relations and numerical models; on the segregated mineral grains. These grains are then
dynamics of underground explosions and underground struc- masked to allow intensity gradient techniques to be used to
tures; cratering explosions; projectile penetration; and under- study orientation patterns within the matrix. Finally, segmen-
ground storage of radioactive waste. tation of the orientation image into similarly aligned domains
shows how the orientation of the matrix relates to that of the
skeleton mineral grains. The separate techniques may be run
Composition, structure texture and density automatically in batch mode.

See a~o: 931222, 931232 931029


Modifications to a method of rapid assessment of soil
931025 macropore structure by image analysis
Investigation and estimation on discontinuity (including McBratney, A B; Moran, C J; Stewart, J B; Cattle, S R;
geotomography) Koppi, A J
Kikuchi, K Geoderma V53, N3/4, June 1992, P255-274
In: Rock Mechanics in Japan, Volume VI 1'81-84. Publ Pore structure studies based on image analysis are generally
Tokyo: Japanese Committee for ISRM, 1991 time consuming, and three modifications are proposed to the
An overview is presented of recent work in Japan on study of original techniques in order to improve productivity. A tech-
the properties, distribution, and influence of discontinuities in nique is described to enable taking of a single monolithic soil
rock masses. Laboratory testing of single discontinuities, the sample up to 700ram in length for video digitizing. The use of
use of the crack tensor to represent a system of discontinuities, resins of different colours in the field and laboratory to allow
hydraulic properties of single and multiple discontinuities, and assessment of a larger number of parameters to distinguish

© 1993 Pergamon Press Ltd. Reproduction not permitted


Appropriate boundary condition for
Dupuit-Boussinesq theory on the steady groundwater
flow in an unconfined sloping aquifer with uniform
recharge
1 1 1
Ying-Hsin Wu , Takahiro Sayama , and Eiichi Nakakita

Ying-Hsin Wu, yhwu@hmd.dpri.kyoto-u.ac.jp

1
Disaster Prevention Research Institute,

Kyoto University, Gokasho, Uji, Kyoto

611-0011, Japan

This article has been accepted for publication and undergone full peer review but has not been through
the copyediting, typesetting, pagination and proofreading process, which may lead to differences be-
tween this version and the Version of Record. Please cite this article as doi: 10.1029/2018WR023070

2018
c American Geophysical Union. All Rights Reserved.
Abstract. The application of Dupuit-Boussinesq theory on the two-dimensional

steady groundwater flow in an unconfined sloping aquifer with uniform recharge

is theoretically investigated. The examination of conditions at the downstream

and upstream is performed. The method of phase plane is used to geomet-

rically interpret the characteristics of the non-linear theory by various ini-

tial conditions. For phase portrait analysis, three configurations for aquifers

are considered. In each type of aquifer three conditions of recharges as well

as downstream groundwater tables are also considered for comprehensive anal-

ysis. Based on the phase portrait analysis, one explicit formula in terms of

known parameters and downstream boundary condition is derived to eas-

ily and straightforwardly determine an appropriate upstream boundary con-

dition. The derived explicit formula can facilitate correct modelling using Boussi-

nesq theory.

Keypoints:

• The non-linear characteristics of groundwater flow in an unconfined Boussi-

nesq aquifer are analysed by using the method of phase plane.

• Appropriate boundary conditions at the downstream and upstream bound-

aries are examined.

• An explicit formula to determine the appropriate upstream boundary

condition is derived.

2018
c American Geophysical Union. All Rights Reserved.
1. Introduction

Modelling gravity-driven flow in a porous media has been intensely studied for several

decades in many aspects, e.g., coastal engineering, geotechnical engineering, industrial

process, and hydrology [e.g., Liu and Wen, 1997; Lambe and Whitman, 1979; Huppert,

2006; Brutsaert, 1994]. In the hydrology field, groundwater flow in an unconfined aquifer

is of great importance on water resource management. Particularly, the flow process in

a soil layer on hillslope can influence slope stability and further induce mass wasting

and landslide hazards. As more hazards were triggered by extreme rainfall on slopeland

recently, understanding the response of an unconfined sloping aquifer on intense rainfall

is certainly important for accurate assessment of hillslope stability.

To model groundwater flow in an unconfined sloping aquifer, one of conventional theo-

ries is the hydraulic groundwater theory [Brutsaert, 2005], or called Dupuit-Forchheimer

theory or Dupuit-Boussinesq theory in some literatures [e.g., Bear , 1972; Harr , 1990;

Guérin et al., 2014]. The main feature of this theory is to assume shallow flow motion of

groundwater flow in an aquifer. The shallow flow obeys hydrostatic pressure. Thus, the

theory can be simplified by reducing the dependence of vertical coordinate, and infiltra-

tion process is excluded for consideration. Although the theory itself becomes simplified,

the simplification alters the theory into a non-linear one. For problems in a horizontal

aquifer or without any source, the simplified theory can be easily solved without any an-

alytic difficulty [e.g., Polubarinova-Kochina, 1962; Bear , 1972; Harr , 1990; Lister , 1992;

Troch et al., 2003, 2004; Vella and Huppert, 2006; Brutsaert, 2005; Guérin et al., 2014].

However, for problems in a steeper hillslope with rainfall recharge, the gravitational ef-

2018
c American Geophysical Union. All Rights Reserved.
fect and the source term become important in the theory, and to our knowledge, only

one analytic solution in an implicit form has been proposed [Henderson and Wooding,

1964; Schmid and Luthin, 1964] and discussed in literature [e.g., Beven, 1981; Basha and

Maalouf , 2005; Chapman, 2005]. Hence, the method of linearization has been widely

used for finding solutions for decades [e.g., Brutsaert, 1994; Verhoest and Troch, 2000].

Recently, the transient groundwater flow in a steep Boussinesq aquifer were numerically

and analytically investigated using the full set of Boussinesq theory [Stagnitti et al., 2004;

Bogaart et al., 2013; Bartlett and Porporato, 2018], and then were applied to recession

analysis [Rupp and Selker , 2006; Bogaart et al., 2013; Hogarth et al., 2014; Dralle et al.,

2014]. This explains the importance of hydraulic groundwater theory on hillslope hydrol-

ogy. A comprehensive review of the application of Boussinesq theory has been summarized

in great detail [Troch et al., 2013].

As mentioned previously, only numerical methods are practical for solving problems in a

steep and unconfined sloping aquifer using the non-linear Boussinesq theory. When using

numerical methods, correct boundary conditions are certainly demanded for successful

iteration algorithms. Similar issues for overland flow have also been investigated numer-

ically and experimentally [e.g., Morris, 1979; de Lima and Torfs, 1990; Van der Molen

et al., 1995] for some applications [e.g., Hu et al., 2014]. In the overland flow modelling,

velocity and flow depth are two different physical variables, so zero volumetric discharge

at the upper boundary can be directly imposed as zero velocity without any attention on

flow depth. However, in the subsurface modelling, as seepage velocity utilizing Darcy’s law

depends on groundwater table, the zero volumetric discharge cannot be simply imposed as

zero velocity, as is briefly explained in the following. In a two-dimensional sloping aquifer

2018
c American Geophysical Union. All Rights Reserved.
having a flat bottom, with the Dupuit assumption and Darcy’s seepage, the volumetric

discharge q 0 in the slope coordinates reads [Brutsaert, 2005]

dη 0
!
0 0
q = −k0 η cos α 0 + sin α , (1)
dx

where η 0 is the groundwater table, k0 is the hydraulic conductivity, and α is the aquifer

inclination. With (1) the zero-volumetric discharge at the upstream boundary yields two

possible conditions that either groundwater table or seepage velocity is zero, i.e.,

dη 0
η 0 = 0 or = − tan α, (2)
dx0

as will be examined in detail in the following sections. For a correct numerical solution,

the two possible conditions cannot be simultaneously imposed at the same boundary. It

requires some way to distinguish the appropriate condition for a given problem. In lit-

erature, different upstream boundary conditions have been used in different problems.

Widely accepted by physical intuition, the constant gradient of groundwater table is usu-

ally imposed at the upstream boundary for analysis of long-time aquifer response [e.g.,

Rupp and Selker , 2006; Troch et al., 2013; Hogarth et al., 2014]. In some other litera-

tures, the full equation of (1) is directly imposed as the upstream boundary condition

with linearization [e.g., Brutsaert, 1994; Verhoest and Troch, 2000; Dralle et al., 2014].

Particularly, without any linearization of (1), Henderson and Wooding [1964] proposed a

steady-state solution as well as an implicit formula to judge the upstream boundary con-

dition. So far, an explicit formula to determine a proper upstream boundary condition

has not been discussed yet. Therefore, we aim to investigate this issue for finding out a

new explicit and efficient way to determine the appropriate boundary conditions for the

2018
c American Geophysical Union. All Rights Reserved.
Boussinesq theory. For the sake of simplicity, only the steady problem is considered in

this study.

Since the main purpose of the present study is to investigate the applicability of an

upstream boundary condition, any presumption cannot be made at the upstream bound-

ary. Under this restriction, the method of phase plane [Strogatz , 2015; Jordan and Smith,

1999], mainly for analysis of non-linear dynamical systems, is one of the practical tools

for analysis. The major advantage of this method is to be able to geometrically inter-

pret the characteristics of any given non-linear system regarding various parameters and

boundary conditions without actually solving the system. In our problem, starting from a

representative point imposed due to the downstream boundary condition, the method of

line integration can be applied to demonstrate the evolution of a trajectory in the phase

diagram, then the end location of the trajectory can reveal the characteristics of the

upstream boundary. Hence, we can validate upstream boundary conditions, and obtain

some information to judge the proper one. Up to now, the phase plane method has been

widely applied in other scientific, engineering, and financial fields, but not in groundwater

Boussinesq theory yet. We would also like to elucidate the applicability of the phase plane

method on analysing hydrological processes.

The content of this paper is as follows. The formulation is described in Section 2.

Downstream and upstream conditions are examined in Section 3. Phase portrait analysis is

explained in Section 4. Then, one explicit formula is derived to distinguish an appropriate

upstream boundary condition in Section 5. Finally, conclusions are summarized in Section

6.

2018
c American Geophysical Union. All Rights Reserved.
2. Formulation

2.1. Governing Equations

It is widely accepted that the soil layer mantling a steep and vegetated hillslope is

usually thin in comparison to the slope length. Groundwater table in the thin soil layer

can be reasonably regarded to be shallow and without large variation in the depth direc-

tion. Figure 1 illustrates the sketch and definition of our problem. We adopt the slope

coordinate system, of which the x0 -axis is aligned along and z 0 -axis is perpendicularly

and upwardly directed from the surface of slope bottom. We consider a general hillslope

having the characteristic length L = O(10) [m] and the characteristic depth H = O(1)

[m], where O(·) denotes the big O symbol [Strogatz , 2015]. Hence we assume

H
=   1.
L

The shallow flow assumption above gives a hydrostatic pressure distribution and the

reduction of z 0 -dependence in the governing equations. As groundwater response to a

rainfall event is rapid in a thin soil layer, infiltration process is assumed to be neglected.

Using Dracy’s seepage law and shallow flow assumption, the steady groundwater flow

with rainfall recharge can be approximately modelled by the Dupuit-Boussinesq theory

[Brutsaert, 2005],

0
dη 0 I0
!
d 0 dη
η + tan α = − , (3)
dx0 dx0 dx0 k0

where η 0 is the groundwater table [m], k0 is the hydraulic conductivity [m s−1 ] ranging

from 10−1 to 10−5 for general hillslopes, α is the slope inclination [rad], and I 0 is the rain

recharge rate [m/s] which generally ranges from 10 mm/hr to 200 mm/hr to represent

from a slight to an extremely intense rainfall. Equation (3) represents the balance between

2018
c American Geophysical Union. All Rights Reserved.
mass in/out-flux and rainfall recharge. The depth-averaging specific flux, or called Darcy’s

seepage velocity, reads

dη 0
!
0
u = −k0 cos α 0 + sin α , (4)
dx

where the minus sign ensures the rightward discharge is in the positive x0 -direction. With

(4) the volumetric discharge can be expressed as

dη 0
!
0 0 0 0
q = η u = −k0 η cos α 0 + sin α .
dx

Due to (3) two boundary conditions are required for a solution. A constant groundwater

table is imposed at the downstream boundary, as

η 0 = D = constant, at x0 = 0.

Zero volumetric discharge is imposed at the upstream boundary, as below

dη 0
!
0 0
q = −k0 η cos α 0 + sin α = 0, at x0 = L. (5)
dx

No simplification is made in the upstream boundary conditions. Besides, attention shall be

paid to the two possible conditions in (5). To our knowledge, as the governing equation (3)

is nonlinear, even if an implicit analytic solution exists, only numerical methods with an

iteration algorithm are available for finding solutions. However, the two conditions in (5)

constitute two different boundary value problems. To clarify, we would like to investigate

the appropriate boundary condition for solving the non-linear Dupuit-Boussinesq theory

without any simplification.

2.2. Normalization

All normalized variables, without primes, are assumed to be

x0 η0
2018
c x=
American and η = . Union. All Rights Reserved.
Geophysical (6)
L H
Using the normalized variables above, the governing equation (3) becomes
!
d dη dη
η +β = −γ, (7)
dx dx dx

where

L tan α I 0 L2
β= and γ = . (8)
H k0 H 2

Depending on the shallowness and inclination of an aquifer, β is called groundwater hills-

lope flow number. A higher β represents the flow in a shallower or steeper aquifer. On the

other hand, γ denotes the ratio of external rain rate to the hydraulic conductivity, and it

represents the storage capability of an aquifer under rainfall recharge. A higher γ means

a higher rainfall on a lower permeable aquifer. Hence, with (6) and (8) the normalized

Darcy’s velocity becomes


!

u = −λ +β ,
dx

where one more normalized parameter is

k0 H cos α
λ= ,
L

reflecting the relation among shallowness, inclination, and aquifer permeability. Then,

the normalized volumetric discharge reads

q0
!

q= 2 = −η +β .
H k0 cos α/L dx

Finally, with (6) the downstream boundary condition becomes

D
η= = η0 , at x = 0. (9)
H

Attention is paid here that the conventional boundary condition of zero-groundwater-table

at the downstream [e.g., Brutsaert, 2005], D = 0, is not considered as it gives the trivial
2018
c American Geophysical Union. All Rights Reserved.
solution of no discharge. This condition can also be explained by the phase portraits in

the following section. At the upstream boundary, the zero-volumetric-discharge boundary

condition reads
!

q = −η + β = 0, at x = 1. (10)
dx

2.3. Volumetric Discharge Distribution

By integrating (7) with respect to x and applying the zero discharge (10) at the upstream

boundary, the volumetric discharge can be obtained as


!

q = −η + β = γ(x − 1), (11)
dx

which is the same as the one in Basha and Maalouf [2005]. The discharge distribution is

linearly proportional to γ, and has a maximum leftward discharge, −γ, at the downstream

boundary. So far, no analytic solution in an explicit form is available for (11). Although

an implicit solution of the steady groundwater table profile, η, has been proposed using

(11) [Henderson and Wooding, 1964], numerical methods are still demanded for calculating

the implicit solution.

3. Examination of Two Boundary Conditions

Despite non-linearity and not having an explicitly analytic solution, (11) can still provide

some information at the upstream and downstream boundaries, as are discussed separately

in the following.

3.1. Downstream Condition

2018
c American Geophysical Union. All Rights Reserved.
At the downstream boundary, with the downstream boundary condition (9) and some

algebra, the volumetric discharge gives a gradient of groundwater table

dη γ
= −β at x = 0, (12)
dx η0

of which the minimum gradient is −β when γ = 0. Beyond the objective of the present

paper, the case of γ = 0 is excluded hereafter. The equation above simply gives a gra-

dient of groundwater table in terms of β, γ, and η0 . By comparing the two terms in the

right-hand-side of (12), in the dimensional form, if total rainfall amount on an aquifer is

greater than drainage discharge at the downstream, I 0 L > k0 D tan α, a convex ground-

water table exists near the downstream boundary. On the other hand, the groundwater

table near the downstream boundary is horizontal or concave if total rainfall amount is

less than or equal to the drainage discharge, I 0 L ≤ k0 D tan α. Besides, if η0 → 0, the

gradient of groundwater table approaches positive infinity. Therefore, no groundwater at

the downstream cannot give a reasonable solution.

3.2. Upstream Condition

At the upstream boundary, the zero-discharge condition (10) can only provide the am-

biguous conditions that either the groundwater table is zero, or the groundwater table

gradient is −β, or both are true, as below


η = 0 or = −β, at x = 1. (13)
dx

Herein, the governing equation (7) is used for analysing conditions at the upstream.

First, we examined the condition of a constant groundwater table gradient of −β at

x = 1 by substituting it into (7) to obtain

d2 η
η = −γ, at x = 1.
dx2 2018
c American Geophysical Union. All Rights Reserved.
For physical significance, the equation above indicates that a finite and convex ground-

water table must exist at the upstream boundary,

d2 η
η > 0 and < 0, at x = 1.
dx2

However, the resultant equation above obviously contradicts the zero-groundwater-table

condition, η = 0.

On the other hand, we instead examined the zero-groundwater-table condition, η = 0 at

x = 1, by applying it to (7) to obtain a quadratic equation of groundwater table gradient,

as below
!2
dη dη
+β = −γ, at x = 1.
dx dx

Hence, we obtained two negative groundwater table gradients at the upstream boundary,

1 1
 q   q 
−β + β 2 − 4γ and −β − β 2 − 4γ , (14)
2 2

where the discriminant for a real groundwater table gradient reads

∆ = β 2 − 4γ ≥ 0.

The validity of the two gradients in (14) is investigated in the following sections. For

physical significance in our problem, the parameter bounds read

β2
0<γ≤ . (15)
4

In (14) special attention is paid here that groundwater table gradients do not equal to

−β. This also testifies again that both conditions in (13) cannot be true simultaneously.

Then, recovering the dimensions of (15) with some manipulations, we obtained a relation

among rainfall intensity, permeability and inclination of the aquifer, for a real-valued

2018
c American Geophysical Union. All Rights Reserved.
groundwater table gradient, as below

k0 tan2 α
0 < I0 ≤ . (16)
4

Taking some examples for using (16), in an aquifer inclining at α = 45◦ and consisting

of well-sorted sand and gravel k0 ≈ 10−3 m/s, the maximum rain rate for a real-valued

groundwater table gradient at the upstream is 2.5×10−4 m/s, namely 900 mm/hr, which

represents an impossible extreme rainfall. If the same slope consists of fine sand k0 ≈ 10−5

m/s, the maximum rain rate becomes 9 mm/hr which is a moderate rainfall on average.

Equation (16) provides parameter bounds for a real solution using Boussinesq theory.

We discover that if it holds a constant groundwater table gradient of −β, there must

be a finite and a convex profile of groundwater table at the upstream. Otherwise, if the

zero-groundwater-table boundary condition is applied, the groundwater table gradient

must not equal −β. This result argues that only one condition in (13) is valid for a

given problem. However, at this point it is lacking of enough information to judge which

condition is appropriate at the upstream boundary. Only the groundwater table and its

gradient at the downstream boundary are known beforehand.

4. Phase Portrait Analysis

In our problem, the information of groundwater table and its gradient at the downstream

boundary are already known, but the appropriate upstream boundary condition is not.

Herein, instead of actually solving the non-linear differential equation of our problem, we

shall utilize the phase plane method to investigate the condition at the upstream boundary

by tracing a path in the phase plane starting from given downstream groundwater table

η0 , hillslope flow number β, and auqifer storage capacity γ. Consider a thin aquifer

2018
c American Geophysical Union. All Rights Reserved.
having H/L = 0.1 for a natural soil-mantled hillslope. Three configurations of inclination

and shallowness for aquifers are considered, including β = L tan α/H = 10, 1, and 0.1,

three different rainfall forcings γ’s satisfying the bounds of (16), and three downstream

groundwater table η0 ’s are considered.

4.1. Phase Plane System

The phase portrait analysis focuses on geometrically interpreting the relation between

groundwater table and its gradient. One new variable for groundwater table gradient is

defined as


ζ≡ = H(η, ζ), (17)
dx

and, by rearranging (7), the curvature of groundwater table, or ζ gradient, reads

dζ 1
= − (ζ 2 + βζ + γ) = Z(η, ζ), (18)
dx η


where a singularity exists when η = 0 as dx
approaches negative infinity. As being a

nontrivial solution, the case of zero groundwater table is excluded for discussion. Herein

we only consider the condition of η > 0 for physical significance, and investigate the

relation between η and ζ in the (η, ζ) phase plane. With (17) and (18), in the phase
dη dζ
plane, the vector field can be expressed as ( dx , dx ) = (H, Z), and the tangent function

reads

dζ Z(η, ζ) ζ 2 + βζ + γ
= =− . (19)
dη H(η, ζ) ζη

With β and γ satisfying (15), (19) is used for visually interpreting the relation between

groundwater table and its gradient in the phase plane.

No equilibrium point exists in this system because letting H = 0 and Z = 0 simultane-

ously cannot give any real-valued


2018
c pair of (η,
American ζ). By letting
Geophysical H = 0All
Union. or ZRights
= 0 we Reserved.
can obtain
three horizontal isoclines defined as below,

1 1
 q   q 
ζ1 = 0, ζ2 = −β + β 2 − 4γ , and ζ3 = −β − β 2 − 4γ , (20)
2 2

with the discriminant (15) for a real-valued ζ. All three isoclines are horizontal lines.

Upon (15) a maximum γ = β 2 /4 yields ζ2 = ζ3 = −β/2. The geometrical property of

each isocline is briefly explained in the following. On the isocline of ζ = ζ1 = 0, all



vector direction along the line is vertically downward as dx
= − γη < 0. So, zero-gradient

and negative curvature of groundwater table give a local maximum of groundwater table.

Then, on the other two isoclines of ζ = ζ2 or ζ = ζ3 , vector direction everywhere is



horizontally leftward as vertical component is always zero, dx
= 0. Both the two isoclines

give a negative gradient and a zero curvature of groundwater table. In particular, the

groundwater table gradient on ζ = ζ3 is steeper than on ζ = ζ2 . Any path on either

the two isoclines can only stay on the line and approach η = 0 by a constant gradient

of ζ2 or ζ3 . This indicates a linear groundwater table distribution if the the downstream

groundwater table gradient is ζ2 or ζ3 . Moreover, ζ1 is not a boundary to separate different

behaviours of our non-linear system, only ζ2 and ζ3 are separatrices in the phase plane.

In different zones separated by ζ2 or ζ3 in (20), trajectories can reflect different be-

haviours of groundwater table η and its gradient ζ. In the zone of ζ > ζ2 , the direction

of any phase path is downward as a negative vertical component dx
< 0 always holds.

The horizontal direction is rightward provided that ζ > 0, then turns into leftward as

ζ2 ≤ ζ < 0. In this zone, any path reflects a concave groundwater table having a maxi-

mum value. Then, in the zone of ζ3 < ζ < ζ2 , the horizontal and vertical components of

any vector are always negative and positive, respectively. So, all phase paths go leftward

and upward, and rapidly converge to ζ2 as η approaches 0. In this zone, any groundwater

2018
c American Geophysical Union. All Rights Reserved.
table corresponding to any phase path has a decreasing distribution. A special case exists

for the separatrix of ζ = ζ3 . Any phase path on ζ = ζ3 can only stay on it and go leftward

to approach (0, ζ3 ), and the corresponding groundwater table has a linear distribution

with a constant gradient of ζ3 in the whole aquifer. Finally, in the zone of ζ < ζ3 , horizon-

tal and vertical components of any vector are both negative. So, all paths diverge from

ζ = ζ3 , and rapidly approach negative infinity of ζ-axis as η approaches zero very closely.

To sum up, phase paths approach ζ2 in the zone of ζ > ζ3 , and negative infinity in the

zone of ζ < ζ3 . All the features can be recognized from the figures introduced in the next

three subsections.

For phase portrait analysis, one should assign some representative points of specific

interest in the phase plane. In our analysis all representative points p are connected with

the downstream conditions by letting p = (η0 , ζ0 ). Hence, one phase path starting from

a given representative point p can be obtained by integrating from 0 to 1 with respect to

x using (17) and (18). According to Section 3.1, once a downstream groundwater table

η0 is given, the downstream groundwater table gradient ζ0 can be determined by (12).

Then, the representative point for a true solution can be imposed under any given η0 ,

and is defined as p6 in the following analysis. However, for the sake of clear illustration

of phase plane features, we assigned five more representative points pi where i = 1 . . . 5,

covering possible parameter ranges of practical interest. On every trajectory, the end

location represents the upstream boundary. Therefore, the information of η and ζ at

all these trajectory ends can be used to validate the appropriate upstream boundary

condition. To emphasize again, none of presumption is made for upstream boundary

because representative points are only assigned due to downstream conditions.

2018
c American Geophysical Union. All Rights Reserved.
For representative points of interest, three downstream groundwater tables are given as

η0 = 1.0, 0.5, and 0.1. As the valid range of γ changes by different β’s, all representative

points for the phase portraits of β = 10, 1, and 0.1 are defined separately in the following

sections. Each case of β has nine phase portraits in total.

4.2. An Aquifer of β = 10

A relatively steeper and/or shallower aquifer is considered herein by assuming β = 10.

According to (15), the parameter of β = 10 gives the valid range of 0 < γ ≤ 25. Here we

consider three γ = 5, 10, and 25. The representative points are defined as
(
pi = (η0 , ζ0 = −18 + 6i) ,
(21)
p6 = (η0 , ζ0 = γ/η0 − β) ,

where i = 1 . . . 5; η0 =1.0, 0.5, and 0.1. Figure 2 illustrates all trajectories change

rapidly in the nine phase portraits. Only the trajectory of p6 in each portrait reaches

(0, η2 ). Trajectories with ζ0 < −β go to negative infinity, and the rest of trajectories

converge to but then diverge from (0, ζ2 ). As a result, all p6 ’s trajectory ends reveal that

the only upstream boundary condition is zero-groundwater table. Additionally, only the

trajectories of p6 in Figs. 2a and d show the concave groundwater tables as the rainfall

recharge is γ = 5.0, the other solutions of groundwater table have convex profiles as

ζ0 > 0. Particularly, no matter what groundwater table is imposed at the downstream,

all convex groundwater tables of γ = 15 and 25 exceed the top aquifer surface as their

maximums are greater than 1.0. All these cases mean that drainage occurs on the aquifer

surface, and the resultant profiles are mathematically valid in the phase plane, but out of

physical significance. To sum up, under any reasonable rainfall recharge, 0 < γ ≤ 25.0, a

groundwater table in an aquifer of a higher β can only be zero at the upstream boundary no

2018
c American Geophysical Union. All Rights Reserved.
matter what the downstream groundwater table is. The zero-groundwater-table condition

is the only choice for upstream boundary condition.

4.3. An Aquifer of β = 1

An aquifer having a moderate inclination and/or shallowness is considered herein by

assuming β = 1. Comparing to the steeper aquifer of β = 10, the milder aquifer can

be regarded as having either a milder inclination or a greater depth. According to (15),

β = 1 gives the valid range of rainfall recharge parameter as 0 < γ ≤ 0.25. Here we

consider three γ = 0.05, 0.15, and 0.25. The representative points are defined as
(
pi = (η0 , ζ0 = −3 + i) ,
(22)
p6 = (η0 , ζ0 = γ/η0 − β) ,

where i = 1 . . . 5; η0 =1.0, 0.5, and 0.1. Comparing with the case of β = 10 aquifer in

Fig. 2, all trajectories in Fig. 3 have the same geometrical features, but do not change

rapidly. As a result, both of the zero-groundwater-table and the constant groundwater

table gradient of −β appear at the upstream boundary under certain parameters. The

zero-groundwater-table upstream boundary condition appears in the cases of shallower

downstream groundwater tables of η0 = 0.1 and 0.5, but the constant gradient of −β

exists in the cases of deeper one of η0 = 1.0. It is intuitive that much groundwater exists

in an aquifer, much groundwater can be accumulated everywhere, and a finite groundwater

table can appear at the upstream boundary. This indicates each boundary condition in

(13) can hold independently under certain parameters.

4.4. An Aquifer of β = 0.1

Finally, a relatively mild and/or deeper aquifer is considered herein by assuming β = 0.1.

According to (15), β = 0.1 gives the valid range of 0 < γ ≤ 0.0025. Here we consider

2018
c American Geophysical Union. All Rights Reserved.
three γ = 0.0005, 0.0015, and 0.0025. The representative points are defined as
(
pi = (η0 , ζ0 = −0.24 + 0.08i) ,
(23)
p6 = (η0 , ζ0 = γ/η0 − β) ,

where i = 1 . . . 5; η0 = 1.0, 0.5, and 0.1. Figure 4 illustrates all trajectories in the nine

phase portraits have slow change. Only the constant groundwater table gradient of −β

appears at the upstream boundary. From all the nine cases together with the ones of

β = 1 in Fig. 3a,b, and c, it can be readily observed that representative points of p6 are

all located between ζ3 and −β when the trajectories approach ζ = −β. This implies the

existence of the constant groundwater table gradient at the upstream boundary, and will

be used in the next section.

5. Criterion for Appropriate Upstream Boundary Condition

Section 3.2 states the two upstream boundary conditions in (13) cannot coexist for a

true solution. Recalling from Section 4.1, all phase paths starting from any location of

ζ > ζ3 always converge to (0, ζ2 ). Only the paths from ζ ≤ ζ3 travel in the zone of ζ ≤ ζ3

and rapidly approach negative infinity of ζ. Additionally, it must hold that ζ3 ≥ −β

according to (20). To combine them, only the paths starting from −β < ζ < ζ3 depart

from ζ = ζ3 and go downward in the phase plane. This is testified by our analysis.

According to Figs. 3a,b,c and 4, the phase portraits reveal that the constant groundwater

table gradient of −β exists at the upstream provided that the relating representative points

are all located in the range of −β ≤ ζ0 < ζ3 . Therefore, we can reasonably conclude the

sufficient condition of a finite groundwater table existing at the upstream boundary shall

be that the downstream groundwater table gradient must be less than ζ3 . So, by applying

the sufficient condition above with some algebraic manipulations, an explicit criterion

2018
c American Geophysical Union. All Rights Reserved.
formula F can be expressed as

γ 1
 q 
F (β, γ, η0 ) = − β − β 2 − 4γ , (24)
η0 2

with (15) for real-valued solutions. When F < 0 the constant groundwater table gradient

shall be imposed at the upstream boundary, dx
= −β at x = 1; Otherwise, when F ≥ 0

the zero-groundwater-table condition, η = 0, must hold instead. An analytical verification

can be referred in the appendix.

Figure 5 illustrates the criteria of the cases under different β, γ, and η0 for whether

a groundwater table exists or not at the upstream boundary. All cases of β = 10 are

excluded as only the zero-groundwater-table condition exists at the upstream boundary.

All thick color lines are obtained letting F = 0 using (24). The cases of β = 1.0 and

0.1 are denoted by black solid circles and triangles, respectively. As all black circles are

located below the criterion of F1.0 = F(η0 = 1.0) in Fig. 5, this means an aquifer of

β = 1.0 has a constant groundwater table gradient of −β at the upstream boundary if the

downstream groundwater table is η0 = 1.0; however, the other has zero-groundwater-table

at the upstream if η0 ≤ 0.5, as are shown in Figs. 3d-i. Finally, as shown in the inlet

figure, the cases of β = 0.1 are all below the criteria of F0.1 , F0.5 , and F1.0 . As a result,

all cases of β = 0.1 have the constant groundwater table gradient of −β at the upstream

boundary, as are testified in Fig. 4.

6. Concluding Remarks

The appropriate boundary conditions for Dupuit-Boussinesq theory on the steady flow in

an unconfined sloping aquifer have been comprehensively investigated in this study. Using

the phase plane method, we have explored the non-linear features of the full Boussinesq

2018
c American Geophysical Union. All Rights Reserved.
theory without any simplification. Two parameters representing inclination, shallowness,

and rainfall recharge are defined for analysis. We have examined all conditions at the

downstream and upstream boundaries, and obtained a parameter bound for true solutions.

In the phase portrait analysis, three aquifer flow number β, aquifer storage capacity γ,

and downstream groundwater table η0 are considered. The phase portraits show that

the a steeper and/or shallower aquifer has a zero-groundwater-table if the downstream

groundwater table is shallower; otherwise, a finite groundwater table with a constant

gradient shall exist at the upstream boundary. Hence, an explicit criterion formula is

successfully derived to straightforwardly determine an appropriate upstream boundary

condition. Any given problems can be numerically easily solved by adopting a correct

upstream boundary condition. To conclude, the major merit of the present work is to

provide a simple and explicit formula in terms of known parameters and downstream

groundwater table for efficiently judging whether a groundwater table exists or not at the

upstream boundary in a thin and unconfined sloping aquifer.

Acknowledgments. E.N. and Y.H.W. would like to thank the financial support of

Japan Society for the Promotion of Science (JSPS) Grants-in-Aid for JSPS Research Fel-

low (Grant No. 16F16378). Y.H.W. appreciates JSPS FY2016 Postdoctoral Fellowship for

Overseas Researchers. The authors are indebted to reviewers for the valuable comments

greatly improving this paper, and to Adrean Webb for editing assistance. The paper is

theoretical and no data are used.

Appendix A: Revisit of conventional steady-state solution [Henderson and

Wooding , 1964]

2018
c American Geophysical Union. All Rights Reserved.
Here, we shall revisit the conventional analytic solution to derive an explicit criterion

formula to determine an appropriate upstream boundary condition. This criterion formula

shall be used to verify the one obtained by phase portrait analysis.

Rearranging the equation of volumetric discharge (11) results in


η = −βη − γ(x − 1). (A1)
dx

Equation (A1) can be regarded as an Abel equation of the second kind, and can be

implicitly expressed in a parametric form [Polyanin and Zaitsev , 2002]. Here we directly

revisit the conventional solution [Henderson and Wooding, 1964] instead of investigating

the parametric solutions above.

For integration of (A1), inspired by [Schmid and Luthin, 1964], two new variables are

defined, including the first variable to transform the x coordinate into a reverse one,

X = 1 − x, (A2)

and a new dependent variable,

η η
Q= = . (A3)
1−x X

With the two variables above the downstream boundary condition (9) gives

Q = η0 , at X = 1. (A4)

With the chain rule and (A3) and (A2), we have


!
dη dQ
=− X +Q . (A5)
dx dX

Hence, after some algebra (A1) results in

QdQ dX
=− , (A6)
Q2
− βQ + γ X
2018
c American Geophysical Union. All Rights Reserved.
with the discriminant

∆ = β 2 − 4γ ≥ 0

for ensuring a real-valued Q. Again, having been explained in Section 3.2, this discrim-

inant denotes the sufficient condition for finding a real-valued solution. In what follows

the solutions of ∆ = 0 and ∆ > 0 are discussed separately.

A1. Solution for ∆ = 0

With ∆ = β 2 − 4γ = 0 (A6) can be further manipulated to

dm β dX
− d(m−1 ) = − , (A7)
m 2 X

where m = Q − β/2. Integrating (A7) once and taking the exponential function to the

resultant equation gives


! !
C1 β β
= Q− exp ,
X 2 β − 2Q

where exp (·) denotes the exponential function; C1 is an integration constant


! !
β β
C1 = η0 − exp
2 β − 2η0

by applying (A4). Recovering Q and X back into the original variables with some algebra,

we obtain the solution in an implicit form of η,


! !
β β β β(1 − x)
η + (1 − x) = η0 − exp − . (A8)
2 2 β − 2η0 β(1 − x) − 2η

Substituting x = 1 into (A8) results in the groundwater table at the upstream boundary,
! !
β β
η(x = 1) = η0 − exp − E0 , (A9)
2 β − 2η0

where

β(1 − x) 2018
c American Geophysical Union. All Rights Reserved.
E0 = lim = 0,
x→1 β(1 − x) − 2η
provided that a finite groundwater table exists at the upstream boundary, η(x = 1) > 0.

For ensuring a finite groundwater table at the upstream, an inequality must hold

β
η0 > (A10)
2

as exp (β/(β − 2η0 )) in (A9) is always positive.

A2. Solution for ∆ > 0

Using partial fraction decomposition, (A6) becomes


√ ! √ !
∆−β dQ ∆+β dQ dX
√ √ + √ √ =− , (A11)
2 ∆ Q − (β − ∆)/2 2 ∆ Q − (β + ∆)/2 X

with ∆ = β 2 − 4γ > 0. Integrating (A11) once and taking the exponential function with

some algebraic rearrangements gives


√ √ √
X |Q − (β − ∆)/2|(β− ∆)/2 ∆
= √ √ √ , (A12)
C2 |Q − (β + ∆)/2|(β+ ∆)/2 ∆

where
√ √ √
(β+ ∆)/2 ∆
|η0 − (β + ∆)/2|
C2 = √ √ √ ,
(β− ∆)/2 ∆
(A13)
|η0 − (β − ∆)/2|

by applying (A4). To compare with the conventional solution [Henderson and Wooding,

1964], we replaced the normalized parameters analogous to their definition by β = 2 and

γ = λ, and rearranged the sign convention to obtain

X |1 − κ − η/X|(1−κ)/2κ
= , (A14)
C2 |1 + κ − η/X|(1+κ)/2κ
√ √
where κ = 1−λ = ∆/2. (A14) exactly equals the conventional solution. Henderson

and Wooding [1964] proposed that a constant gradient of groundwater table exists at the

upstream provided that 1 + κ < η/X < 2. However, as η/X is unknown before having

a solution, the inequality is of impractical use. Generally, numerical techniques utilizing


2018
c American Geophysical Union. All Rights Reserved.
an iterative algorithm is necessary to solve (A12) with an initial guess of solution. The

numerical solution is neither efficient to obtain, nor lacking of an appropriate way to

impose a reasonable initial guess. Hence, for easier and efficient use, it may demand an

explicit formula in terms of known parameters and the downstream boundary condition.

To find an explicit formula, another form of the solution is used here. With

1 √  1 √  √
β− ∆ = β+ ∆ − ∆
2 2

(A12) can be rewritten as


√ √
√ !(β+ ∆)/2 ∆

1 √  ∆


X Q − β − ∆ = C2 1 + , (A15)
2 |Q − (β + ∆)/2|

and replacing Q = η/X and X = (1 − x) by original variables with some manipulations

yields
√ √
√ !(β+ ∆)/2 ∆

1  √  2(1 − x) ∆


η − (1 − x) β − ∆ = C2 1 + , (A16)

2 |2η − (1 − x)(β + ∆)|

0 where C2 is expressed in (A13). Equation (A16) is an implicit solution of groundwater

table η. With unknown η we have no information about the profile of groundwater table.

However, we only have to investigate whether a finite groundwater table exists at the

upstream boundary, i.e., η(x = 1) > 0. Hence, substituting x = 1 into (A15) simply

yields
√ √ √
∆)/2|(β+ ∆)/2 ∆
|η0 − (β +
η(x = 1) = C2 = √ √ √ .
|η0 − (β − ∆)/2|(β− ∆)/2 ∆

Therefore, the two sufficient conditions for η(x = 1) > 0 read

1 1
q  q  
η0 − β − β 2 − 4γ > 0 and η0 − β + β 2 − 4γ > 0.
2 2

The intersection of the two conditions above reads

2018
c American
1 Geophysical
 q  Union. All Rights Reserved.
η0 > 2
β + β − 4γ . (A17)
2
Obviously, after taking reciprocals of both sides of (A17) with some manipulations, we

can obtain the inequality

γ 1
q 
− β − β 2 − 4γ < 0. (A18)
η0 2

for ensuring a finite groundwater table at the upstream boundary provided that β 2 − 4γ >

0. Moreover, applying β 2 − 4γ = 0 to (A18) can also yield the same inequality of (A10).

It is proven that (A18) is exactly equivalent to (24) obtained by phase portrait analysis.

References

Bartlett, M., and A. Porporato (2018), A class of exact solutions of the Boussinesq equa-

tion for horizontal and sloping aquifers, Water Resour. Res., 54, 767–778.

Basha, H. A., and S. F. Maalouf (2005), Theoretical and conceptual models of subsurface

hillslope flows, Water Resour. Res., 41 (7), doi:10.1029/2004WR003769, W07018.

Bear, J. (1972), Dynamics of Fluids in Porous Media, Dover.

Beven, K. (1981), Kinematic subsurface stormflow, Water Resour. Res., 17 (5), 1419–1424.

Bogaart, P. W., D. E. Rupp, J. S. Selker, and Y. van der Velde (2013), Late-time

drainage from a sloping Boussinesq aquifer, Water Resour. Res., 49 (11), 7498–7507,

doi:10.1002/2013WR013780.

Brutsaert, W. (1994), The unit response of groundwater outflow from a hillslope, Water

Resour. Res., 30 (10), 2759–2763.

Brutsaert, W. (2005), Hydrology: an introduction, Cambridge University Press.

Chapman, T. G. (2005), Recharge-induced groundwater flow over a plane sloping bed:

Solutions for steady and transient flow using physical and numerical models, Water

Resour. Res., 41 (7), doi:10.1029/2004WR003606, w07027.


2018
c American Geophysical Union. All Rights Reserved.
de Lima, J., and P. Torfs (1990), Upper boundary conditions for overland flow, J. Hydraul.

Eng.-ASCE, 116 (7), 951–957.

Dralle, D. N., G. F. Boisramé, and S. E. Thompson (2014), Spatially variable water

table recharge and the hillslope hydrologic response: Analytical solutions to the lin-

earized hillslope Boussinesq equation, Water Resour. Res., 50 (11), 8515–8530, doi:

10.1002/2013WR015144.

Guérin, A., O. Devauchelle, and E. Lajeunesse (2014), Response of a laboratory aquifer

to rainfall, J. Fluid Mech., 759, R1.

Harr, M. E. (1990), Groundwater and Seepage, Dover.

Henderson, F. M., and R. A. Wooding (1964), Overland flow and groundwater flow from

a steady rainfall of finite duration, J. Geophys. Res., 69 (8), 1531–1540.

Hogarth, W. L., L. Li, D. A. Lockington, F. Stagnitti, M. B. Parlange, D. A. Barry, T. S.

Steenhuis, and J.-Y. Parlange (2014), Analytical approximation for the recession of a

sloping aquifer, Water Resour. Res., 50 (11), 8564–8570, doi:10.1002/2014WR016084.

Hu, S.-Y., P.-T. Chiueh, and P.-C. Hsieh (2014), A novel semi-analytical approach for

non-uniform vegetated flows, Adv. Water Resour., 64, 1–8.

Huppert, H. E. (2006), Gravity currents: a personal perspective, J. Fluid Mech., 554,

299–322.

Jordan, D. W., and P. Smith (1999), Nonlinear Ordinary Differential Equations: An

Introduction to Dynamical Systems, vol. 2, Oxford University Press, USA.

Lambe, T. W., and R. V. Whitman (1979), Soil Mechanics SI Version, John Wiley &

Sons.

2018
c American Geophysical Union. All Rights Reserved.
Lister, J. R. (1992), Viscous flows down an inclined plane from point and line sources, J.

Fluid Mech., 242, 631–653.

Liu, P. L.-F., and J. Wen (1997), Nonlinear diffusive surface waves in porous media, J.

Fluid Mech., 347, 119–139.

Morris, E. (1979), The effect of the small-slope approximation and lower boundary con-

ditions on solutions of the saint-venant equations, J. Hydrol., 40 (1-2), 31–47.

Polubarinova-Kochina, P. (1962), Theory of ground water movement, Princeton University

Press.

Polyanin, A. D., and V. F. Zaitsev (2002), Handbook of exact solutions for ordinary

differential equations, 2 ed., CRC press.

Rupp, D. E., and J. S. Selker (2006), On the use of the Boussinesq equation for interpreting

recession hydrographs from sloping aquifers, Water Resour. Res., 42 (12), W12421.

Schmid, P., and J. Luthin (1964), The drainage of sloping lands, J. Geophys. Res., 69 (8),

1525–1529.

Stagnitti, F., L. Li, J.-Y. Parlange, W. Brutsaert, D. A. Lockington, T. S. Steenhuis, M. B.

Parlange, D. A. Barry, and W. L. Hogarth (2004), Drying front in a sloping aquifer:

Nonlinear effects, Water Resour. Res., 40 (4), doi:10.1029/2003WR002255, W04601.

Strogatz, S. H. (2015), Nonlinear Dynamics and Chaos: With Applications to Physics,

Biology, Chemistry, and Engineering, Westview Press.

Troch, P. A., C. Paniconi, and E. Emiel van Loon (2003), Hillslope-storage Boussi-

nesq model for subsurface flow and variable source areas along complex hill-

slopes: 1. formulation and characteristic response, Water Resour. Res., 39 (11), doi:

10.1029/2002WR001728.

2018
c American Geophysical Union. All Rights Reserved.
Troch, P. A., A. H. van Loon, and A. G. J. Hilberts (2004), Analytical solution of the lin-

earized hillslope-storage Boussinesq equation for exponential hillslope width functions,

Water Resour. Res., 40 (8), doi:10.1029/2003WR002850, W08601.

Troch, P. A., A. Berne, P. Bogaart, C. Harman, A. G. J. Hilberts, S. W. Lyon, C. Pan-

iconi, V. R. N. Pauwels, D. E. Rupp, J. S. Selker, A. J. Teuling, R. Uijlenhoet, and

N. E. C. Verhoest (2013), The importance of hydraulic groundwater theory in catchment

hydrology: The legacy of Wilfried Brutsaert and Jean-Yves Parlange, Water Resour.

Res., 49 (9), 5099–5116, doi:10.1002/wrcr.20407.

Van der Molen, W., P. Torfs, and J. de Lima (1995), Water depths at the upper boundary

for overland flow on small gradients, J. Hydrol., 171 (1-2), 93–102.

Vella, D., and H. E. Huppert (2006), Gravity currents in a porous medium at an inclined

plane, J. Fluid Mech., 555, 353–362.

Verhoest, N. E. C., and P. A. Troch (2000), Some analytical solutions of the linearized

Boussinesq equation with recharge for a sloping aquifer, Water Resour. Res., 36 (3),

793–800.

2018
c American Geophysical Union. All Rights Reserved.
Figure 1. Sketch of our problem and definition of all dimensional variables. (x0 , z 0 ) is the

slope coordinates, η 0 is the groundwater table, I 0 is the uniform rainfall recharge rate, α is the

inclination, and D is the groundwater table at the downstream. L and H are the characteristic

length and height of the aquifer. x0u = L and x0d = 0 denote the upstream and downstream

locations. Right panels (a) and (b) illustrate the two conditions in (13) due to zero volumetric

discharge.

2018
c American Geophysical Union. All Rights Reserved.
= 5.0 = 15.0 = 25.0
p6
15 15 15
a p5 b p5 c p5
10 10 10
p4 pp46 p4
5 5 5
0 = 1.0

0 p3 0 p3 0 p3
5 2 pp62 5 2
p2 5 2
p2
3 3 3
10 10 10
p1 p1 p1
15 15 15
0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0
15 15 15
d p5 e p5 f p5
10 10 10
p4 p4 p4
5 5 5
0 = 0.5

0 p63 0 p3 0 p3
2 2 2
5 p2 5 p2 5 p2
3 3 3
10 10 10
p1 p1 p1
15 15 15
0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0
15 15 15
g p5 h p5 i p5
10 10 10
p4 p4 p4
5 5 5
0 = 0.1

0 p3 0 p3 0 p3
2 2 2
5 p2 5 p2 5 p2
3 3 3
10 10 10
p1 p1 p1
15 15 15
0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0

Figure 2. Phase portraits of β = 10 under γ = 5, 15, and 25 and η0 = 1.0, 0.5, and 0.1. The black
dotted lines are η = 0 and the separatrices of ζ2 and ζ3 in (20). The red dotted lines denote ζ = −β.
Coloured solid circles and triangles denote the downstream and upstream locations, respectively. The
trajectories of ζ0 < ζ3 go to negative infinity, but the others first converge to and then diverge from
(0, ζ2 ). Each purple trajectory p6 denotes the solution under the given ζ0 , γ, and β = 10.

2018
c American Geophysical Union. All Rights Reserved.
= 0.05 = 0.15 = 0.25
2 a p5 2 b p5 2 c p5

1
p4 1
p4 1
p4
0 = 1.0

0
p3 0
p3 0
p3
2 2 2

p62 pp62 pp6


1 1 1 2
3 3 3

2 p1 2 p1 2 p1
0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0

2 d p5 2 e p5 2 f p5

1
p4 1
p4 1
p4
0 = 0.5

0
p3 0
p3 0
p3
2 2 2
p6 p6
1
p62 1 p2 1 p2
3 3 3

2 p1 2 p1 2 p1
0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0

2 g p5 2 h p5 2 i p5
p6
1 p4 1 p4 1 p4
p6
0 = 0.1

0 2
p3 0 2
p3 0 2
p3
p6
1 3
p2 1 3
p2 1 3
p2

2 p1 2 p1 2 p1
0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0

Figure 3. Phase portraits of β = 1 under γ = 0.05, 0.15, and 0.25 and η0 = 1.0, 0.5, and 0.1. The
black dotted lines are η = 0 and the separatrices of ζ2 and ζ3 in (20). The red dotted lines denote ζ =
−β. Coloured solid circles and triangles denote the downstream and upstream locations, respectively.
The trajectories of ζ0 < ζ3 go to negative infinity, the ones of ζ0 > ζ3 approach (0, ζ2 ), and the others
satisfying −β < ζ0 < ζ3 converge to (0, −β). Each purple trajectory p6 , which a linear curve with a
constant slope gradient of ζ2 = ζ3 , denotes the solution under the given ζ0 , γ, and β = 1.

2018
c American Geophysical Union. All Rights Reserved.
0.20
= 0.0005 0.20
= 0.0015 0.20
= 0.0025
0.15 a p5 0.15 b p5 0.15 c p5
0.10 p4 0.10 p4 0.10 p4
0.05 0.05 0.05
0 = 1.0

0.00 2
p3 0.00 2
p3 0.00 2
p3
0.05 0.05 0.05
pp2 pp2 pp2
0.10 3 6 0.10 3 6 0.10 3 6
0.15 p1 0.15 p1 0.15 p1
0.20 0.20 0.20
0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0
0.20 0.20 0.20
0.15 d p5 0.15 e p5 0.15 f p5
0.10 p4 0.10 p4 0.10 p4
0.05 0.05 0.05
0 = 0.5

0.00 2
p3 0.00 2
p3 0.00 2
p3
0.05 0.05 0.05
pp2 pp2 pp26
0.10 3 6 0.10 3 6 0.10 3

0.15 p1 0.15 p1 0.15 p1


0.20 0.20 0.20
0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0
0.20 0.20 0.20
0.15 g p5 0.15 h p5 0.15 i p5
0.10 p4 0.10 p4 0.10 p4
0.05 0.05 0.05
0 = 0.1

0.00 2
p3 0.00 2
p3 0.00 2
p3
0.05 0.05 0.05
pp26 p26 p62
0.10 3 0.10 3 0.10 3

0.15 p1 0.15 p1 0.15 p1


0.20 0.20 0.20
0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0 0.5 0.0 0.5 1.0 1.5 2.0

Figure 4. Phase portraits of β = 0.1 under γ = 0.0005, 0.0015, and 0.0025 and η0 = 1.0, 0.5, and
0.1. The black dotted lines are η = 0 and the separatrices of ζ2 and ζ3 in (20). The red dotted lines
denote ζ = −β. Coloured solid circles and triangles denote the downstream and upstream locations,
respectively. The trajectories of ζ0 < ζ3 go to negative infinity, the ones of ζ0 > ζ3 approach (0, ζ2 ), and
the others satisfying −β < ζ0 < ζ3 converge to (0, −β). Each purple trajectory p6 denotes the solution
under the given ζ0 , γ, and β = 0.1.

2018
c American Geophysical Union. All Rights Reserved.
1.50
1.25 F1.0
F0.5
1.00
0.15
= LtanH

0.75 F0.1
0.10
0.50
0.05
0.25
0.00
0.0005 0.0015 0.0025 0.0035
0.00
0.0 0.1 0.2 0.3 0.4
= kI0LH22
Figure 5. Criteria for appropriate upstream boundary condition. Solid black circles and

triangles denote the cases of β = 1.0 and 0.1. The blue, red, and green lines are the criteria

under different η0 = 1.0, 0.5, and 0.1 by using (24). The grey part is excluded for solutions as

the black dashed lines denote the lower bound for having real-valued solutions using (15).

2018
c American Geophysical Union. All Rights Reserved.
REPLY Reply to comment by Kong et al. on “Appropriate
10.1029/2019WR024872
Boundary Condition for Dupuit‐Boussinesq Theory
This article is a reply to a comment by
Kong et al. (2019), https://doi.org/
on the Steady Groundwater Flow in an Unconfined
10.1029/2018WR024665.
Sloping Aquifer With Uniform Recharge”
Key Points: Ying‐Hsin Wu1 , Takahiro Sayama1 , and Eiichi Nakakita1
• For the original Dupuit‐Boussinesq 1
theory, zero volumetric discharge at Disaster Prevention Research Institute, Kyoto University, Uji, Japan
the upstream yields zero
groundwater table or zero seepage
velocity Abstract The article aims to respond a comment made on our paper about appropriate boundary
• Groundwater table can be of condition for the original Dupuit‐Boussinesq theory for two‐dimensional steady groundwater flow in an
physical significance only upon a
nonnegative value unconfined sloping aquifer with uniform rainfall recharge. To respond to the comments arguing the
• A constant rainfall recharge existence of lateral groundwater flows and negative groundwater table, clarifications are made for our
determines a linear discharge analysis focusing on two‐dimensional groundwater flow without considering lateral effects by using the
distribution
original and classical approximate theory.

Correspondence to:
Y.‐H. Wu, The authors would like to thank Kong and other colleagues for their interest in and a comment article (Kong
yhwu@hmd.dpri.kyoto‐u.ac.jp et al., 2019) made on our research paper (Wu et al., 2018) investigating appropriate boundary conditions for
the original Dupuit‐Boussinesq theory for steady two‐dimensional groundwater flow in an unconfined slop-
Citation: ing aquifer with uniform rainfall recharge. We are delighted to read the comments (Kong et al., 2019) point-
Wu, Y.‐H., Sayama, T., & Nakakita, E. ing out shortcomings of the classical Dupuit‐Boussinesq theory for general sloping aquifers and providing
(2019). Reply to comment by Kong et al. possible directions for broadening the scope of analytical analysis on shallow groundwater flow in uncon-
on “Appropriate Boundary Condition
for Dupuit‐Boussinesq Theory on the fined aquifers. We do agree with Kong and other colleagues that an unsaturated zone is of importance for
Steady Groundwater Flow in an general and natural aquifers that are not the investigation focus of our paper (Wu et al., 2018). Instead,
Unconfined Sloping Aquifer With our paper's main purpose is to demonstrate one important analytical issue of determination of an appropri-
Uniform Recharge”. Water Resources
Research, 55, 3597–3598. https://doi. ate boundary condition, which has not drawn much attention and comprehensively discussed in the past.
org/10.1029/2019WR024872 Through this reply, we would like to further clarify our work to address the comments (Kong et al., 2019)
arguing the effects of lateral flow and negative groundwater table.
Received 28 JAN 2019
Accepted 8 MAR 2019 The original Dupuit‐Boussinesq theory (hereafter denoted as the original theory), or called hydraulic
Accepted article online 18 MAR 2019 groundwater theory, is widely used for modeling groundwater flow in an unconfined aquifer under the
Published online 3 APR 2019
assumptions that the capillary effect is insignificant and the aquifer is shallow (Brutsaert, 2005). These
assumptions successfully simplify analysis on relating problems but yield some constraints that pressure is
hydrostatic and seepage flow is independent of the vertical coordinate of the slope coordinate system.
These constraints do not always match flow conditions in real hillslope aquifers that are insufficiently shal-
low, unneglectable for capillary effect, and possess inhomogeneous porosity. However, the original theory
still can provide approximate solutions very close to ones obtained by a more complete formulation, so it
is the method of choice in many investigations (Brutsaert, 2005) or for further developments of improvement
(e.g., Hilberts & Troch, 2005, Kong et al., 2016, Luo et al., 2018, Troch et al., 2003). But, as is clearly pointed
out in Wu et al. (2018), it is still lacking of an efficient way to determine appropriate boundary conditions
when applying this original and classical theory. Based on its simplicity and relevance to hillslope hydrology
analysis, the original theory is still worthy of our investigation focus as a foundation for further
analytical development.
Here we would like to clarify again our analysis to possibly address the comment arguing that the seepage
velocity must be set to zero to satisfy the zero discharge boundary condition at the upstream boundary.
Without proposing any new formulation as well as source and/or sink in the aquifer, we were focusing on
revealing the characteristics of the dynamical system of the original theory on a two‐dimensional steady
flow. Within the scope of physical significance defined in problems of our specific interest, we have inter-
©2019. American Geophysical Union. preted dynamical behaviors of the value and the gradient of groundwater table and demonstrated parametric
All Rights Reserved. conditions on the phase planes when considering aquifers under three different groundwater hillslope flow

WU ET AL. 3597
Water Resources Research 10.1029/2019WR024872

numbers β = L tan α/H=10, 1.0, and 0.1, where α is the inclination and H and L are characteristic aquifer's
thickness and length, respectively. For precise discussion, all aquifers considered in our analysis were
categorized by βs not only by the inclination. As is selected for comprehensively demonstrating the
dynamical system, the phase plane method successfully reveals the original theory possesses the behavior
that either zero groundwater table or zero seepage velocity do separately exist in aquifers of certain
settings and under certain rainfall recharges. To emphasize again, in our analysis, another important
advantage of the phase plane analysis is that this method parametrically analyzes the original theory to
obtain analytical solutions of groundwater table and its gradient everywhere from the downstream
boundary without any presumption of upstream boundary beforehand. The analysis finally testifies the
existence of two possible boundary conditions at the upstream when using the original theory. Also, for
solutions of physical significance, the original theory always retains a nonnegative and real‐valued
groundwater table under a criterion,

k 0 tanα
0<I≤ ; (1)
4

where I is the rainfall recharge and k0 is the aquifer's hydraulic conductivity. Explanations refer to (16) and
following paragraph in Wu et al. (2018).
For the other comments regarding the discharge distribution, we would like to respond as follows. To fit the
steady state assumption, the original theory possesses a linear distribution of discharge relating to rainfall
recharge from the downstream outlet to the upstream with a nonnegative groundwater table. This condition
is also self‐evident in our analysis of the classical and original Dupuit‐Boussinesq theory for two‐dimen-
sional groundwater flow in unconfined sloping aquifers with constant recharge. To remind again, as no
source and/or sink is considered, the lateral flow may not be able to be correctly modeled using the
original theory.
The responses are summarized to possibly address the comments (Kong et al., 2019). Again, we thank the
comments (Kong et al., 2019) for allowing us to take this chance of reply to the comments (Kong et al.,
2019) to further clarify the applicability and validness of our analysis.

Acknowledgments References
The original research was financially
supported by Japan Society for the Brutsaert, W. (2005). Hydrology: An introduction. Cambridge: Cambridge press. https://doi.org/10.1017/CBO9780511808470
Promotion of Science (JSPS) through Hilberts, A. G. J., & Troch, P. A. (2005). Storage‐dependent drainable porosity for complex hillslopes. Water Resources Research, 41,
Grants‐in‐Aid for JSPS Research Fellow W06001. https://doi.org/10.1029/2004WR003725
(Grant No. 16F16378). No data were Kong, J., Shen, C., Luo, Z., Hua, G., & Zhao, H. (2016). Improvement of the hillslope‐storage Boussinesq model by considering lateral flow
used for this article. in the unsaturated zone. Water Resources Research, 52, 2965–2984. https://doi.org/10.1002/2015WR018054
Kong, J., Sun, J., Lu, C., Luo, C., Shen, C., & Hua, G. (2019). Comment on “Approximate Boundary Condition for Dupuit‐Boussinesq
Theory on the Steady Groundwater Flow in an Unconfined Sloping Aquifer with Uniform Recharge” by Wu et al. (2018). Water
Resources Research. https://doi.org/10.1029/2018WR024665
Luo, Z., Shen, C., Kong, J., Hua, G., Gao, X., Zhao, Z., et al. (2018). Effects of unsaturated flow on hillslope recession characteristics. Water
Resources Research, 54, 2037–2056. https://doi.org/10.1002/2017WR022257
Troch, P. A., Paniconi, C., & van Loon, E. E. (2003). Hillslope‐storage Boussinesq model for subsurface flow and variable source areas along
complex hillslopes: 1. Formulation and characteristic response. Water Resources Research, 39(11), 1316. https://doi.org/10.1029/
2002WR001728
Wu, Y.‐H., Sayama, T., & Nakakita, E. (2018). Appropriate boundary condition for Dupuit‐Boussinesq theory on the steady groundwater
flow in an unconfined sloping aquifer with uniform recharge. Water Resources Research, 54, 5933–5947. https://doi.org/10.1029/
2018WR023070

WU ET AL. 3598
WATER RESOURCES RESEARCH, VOL. 28, NO. 6, PAGES 1619-1630, JUNE 1992

Groundwater Flow in a Compressible Unconfined Aquifer


With Uniform Circular Recharge
VITALY ZLOTNIK

Department of Geology, University of Nebraska, Lincoln

GLENN LEDDER

Department of Mathematics, University of Nebraska, Lincoln

The distributions of the hydraulic head and velocity componentsof the transient groundwater flow
in an unconfinedcompressibleaquifer of finite thicknessunder constant uniform circular recharge are
obtained from the linearized mathematical model by the use of integral transforms. The result
generalizes Dagan's (1967) solution which was derived by neglectingthe compressibility. By treating
the compressibility parameter as a small value, the formula for the hydraulic head is analyzed by
asymptotic methods, resultingin approximationsto the exact solutionsfor the head and velocities on
small and large time scales.The hydraulic head and flow velocities can be accurately approximated by
Dagan's formula for large times; for small times, neglectingthe compressibilitygives a large relative
error but small absolute error.

1. INTRODUCTION the elevation of the groundwater mound under a source area


[Hantush, 1967], provided the rise of the water table relative
The interaction between groundwaterflow systems,water to the initial saturated thickness of the aquifer is as high as
supply wells, and natural or artificial groundwater recharge 50%.•h]s assumption
is frequently
usedin engineering
creates a complex velocity flow field in aquifers. This practice when the shape and volumetric characteristics of
velocity field can be represented as a result of the interaction the groundwater mound are of interest [Morel-Seytoux et al.,
of vertical line sinks (wells) and horizontal areal sources 1990]. However, this approach does not provide realistic
(recharge). The flow is three dimensional rather than two estimates for groundwater flow velocities. As far as we
dimensional and is transient rather than steady. Thus, to know, the only analytical three-dimensional solution avail-
study contaminant transport below the water table, a three-
able is that obtained for an unconfined incompressibleaqui-
dimensionalunsteady model is necessaryto delineate major
fer with areal circular recharge IDagan, 1967]. That analysis
contaminantpathways.
does not present the velocity distribution. The solution was
The role of groundwater flow modeling is to provide an
derived under the following assumptions:(1) that the ratio of
estimate of the flow velocities. Head predictionsare of little
recharge(I) to hydraulic conductivity(Kv) is a smallparam-
interest. Velocity estimates, however, are usually based on
eter and (2) that the specific yield of the aquifer is not
hydraulic head differences and therefore are much more
appreciably changed by the recharge. The problem was
sensitiveto numerical modelingerrors than are estimatesof
linearized using I/Kv as a small parameter. Green's func-
the hydraulic head alone. Satisfactory predictions of trans-
tions were then used to obtain the solution for the hydraulic
port often require that the velocity field be calculated on a
head.
fine spatial grid. Therefore analytical solutions have some
To analyze the influenceof the smallparameterI/Kv and
advantage over numerical procedures. Unfortunately, such
the applicability of the linearized solution for the full non-
solutions are not often available for cases of practical
linear problem, Lennon et al. [1979] applied the numerical
importance [National Research Council, 1990].
boundary integral method. The numerical simulation showed
In this paper we consider the flow induced in an uncon-
that the linearized, incompressible, constant-specific-yield
fined compressibleaquifer of finite depth by an areal source
solutionunderpredictsthe groundwater mound rise for large
(a source of finite horizontal extent), as illustrated in Figure
times in aquifers with thicknessesless than 2-10 times the
1. Such sources represent groundwater recharge for the
recharge radius if the small parameter approachesthe value
problems of transport of agricultural fertilizers [Dillon,
0.2. However, there was no analysis about the influence of
1989], oil spills from undergroundstoragetanks [Levy et al.,
compressibilityof the unconfined aquifer or the influence of
1990], and leachates from landfills [Ostendorfet al., 1989]. In
flow in the unsaturated zone on the development of water
the near field of the contamination source the velocity field is
mounds.
predominantly vertical, as contaminants are submergedinto
The effect of compressibility for flow caused by a well
deeper layers of an unconfined aquifer; farther from the
(vertical line sink) has been shown to be significantfor very
source, the velocity direction changessharplyfrom predom-
early stagesof pumping, causingdelayed responseof draw-
inantly vertical to predominantly horizontal [Hunt, 1971].
down in an unconfined aquifer INcurnan, 1974, 1979].
Experimental results confirm the Dupuit assumptionfor
"Dagan's method does not take into account the phenome-
Copyright 1992 by the American GeophysicalUnion. non of delayed gravity response,and therefore it is limited in
Paper number 92WR00462. its application to relatively large distancesfrom the pumping
0043-1397/92/92WR-00462505.00 well and to sutficiently large values of time, at which the
1619
1620 ZLOTNIK AND LEDDER.'FLOW IN COMPRESSIBLE
UNCONFINEDAQUIFERS

velocity vector can then be obtained as the gradient of the


hydraulic head. The model will be solved exactly, but the
solutionis somewhatcomplicated. We then derive approxi-
mate solutions on two different time scales from the exact
solutionand use the method of matched asymptoticexpan-
sions (see, for example, Murdock [1991]) to obtain an
,
approximatesolutionvalid for all time. This uniform approx-
imation can be usedin place of the exact solutionto provide
numerical estimateswithout sacrificingaccuracy.
z•'-" Kh
Problem for the Hydraulic Head

'11111111111IIIIIIIIIIIII11'11111'11;I IIIJJ///,'•--'- We consideran unconfinedaquifer of infinite lateral extent


o R • and finite thicknessthat rests on an impermeablehorizontal
Fig. 1. Schematic diagram of groundwaterrecharge in an uncon- layer (Figure 1). The aquifer material is uniform and aniso-
fined aquifer of finite thickness. tropic, the principal conductivitiesbeing orientedparallel to
the coordinate axes. The aquifer is recharged by a uniform
circular source that is turned on at time zero. We neglect
effect of elastic storage is very small" [Neuman, 1975, variations of specific yield. Within the limits of first-order
p. 329]. linearized theory, the problem may be written as
In this paper we consider the effect of compressibilityon
flow induced by areal recharge. We obtain the general
solution for the compressible case and so determine the
effect of compressibility by comparing the compressible
Kn -
r• 1O(O•) O--•
02•satO•
? + Kv = S '•_ (1)

solution to that obtained by Dagan. We are thus able to ?>0 O<Z<I t>0
discussthe error made by using the incompressiblesolution
to approximate the solution in the compressiblecase.
lim g(?,r, t• = 0
According to Kroszinskyand Dagan [1975] and Brutsaert r-• -- (0, r, t3 = 0 -- (?, 0, t• = 0
o• OF OZ
and EI-Kadi [1984], the unsaturated zone above the water
table has little quantitative effect on drawdown in the case of (2)
an aquifer with coarse soil structure. We consider only the
saturated zone in this work. A model which could address
(3)
the influence of the unsaturated
scope of the linearized theory.
zone would be outside the
(r,O,t-)+ O,t-)= r)
In the problem discussedin this paper there are two small
H(x) = 1 x -> 0
parameters that must be considered. The ratio of specific
H(x) = 0 x< 0
rechargeto conductivity will be taken to be small throughout
the following analysis, so that the effect of the other param-
eter can be studied within the framework of the linearized •(?, Z, O) = 0 (4)
theory. The ratio of storativity to specific yield (•r) will be
where g is the increase of hydraulic head over the initial
used as a small parameter to obtain useful approximations.
value; ?, Z, and t are the radius, height above the bottom,
Under the assumptionsof the linearized theory we obtain
and time; R is the radius of the circular source; b is the initial
an analytical three-dimensional solution for the hydraulic
saturated thickness of the aquifer; K n and Kv are the
head and spatial flow velocity componentsin an unconfined,
horizontal and vertical saturated hydraulic conductivities;
compressible,homogeneous,anisotropic aquifer under the
influence of low-intensity groundwater recharge by a con-
$s is the specific
(elastic)storage;Sy is the specific
yield;I
is the net specific recharge at the water table; and H is the
stant sourceuniformly distributedover a finite circular region.
unit step function. (See Dagan [ 1967] and Neuman [ 1974]for
The resultsappear in the form of definiteintegrals,which we
the derivation of the model.) The rechargerate is taken to be
evaluatenumerically.We also obtain an approximatesolution
steady for mathematical simplicity; the final results can be
for the compressiblecaseby asymptoticexpansionof the exact
generalized for time-dependent recharge using the convolu-
solutionfollowed by asymptoticmatching.The resultingap-
tion theorem [$neddon, 1972]. The bars are used to denote
proximationis virtually indistinguishablefrom the exact solu-
dimensionalvariablesand parametersthat will be replacedin
tion but has a simplerform and is thereforeof great value in
the model by their dimensionlesscounterparts.
obtainingan analytic comparisonof the incompressibleand
The equations are made dimensionless by introducing
compressiblesolutions.Further analysisyieldssimpleapprox-
dimensionlessvariables that are scaled so as to simplify the
imate quantitativestatementsof the long-termeffect of com-
equations and reduce the number of parameters in the
pressibility on the hydraulic head and spatial flow velocity
componentsfor all near-field locations. problemfromsix(R, b, K•, Kv, Ss, Sy) to two.Thesenew
parameters represent the ratio of recharge area radius to
aquifer depth and the compressibility (the ratio of specific
2. PROBLEM STATEMENT elastic storageto specificyield) and are defined by
We begin by writing down and solvingthe problem for the
increase of hydraulic head over the initial level in an uncon- (5)
fined aquifer under a circular source of recharge. The b Sy
ZLOTNIK AND LEDDER: FLOW IN COMPRESSIBLE
UNCONFINEDAQUIFERS 1621

To put the problem in dimensionlessform, we chooseb as section3 we present the exact solutionusingthe notation of
the reference quantity for Z. Then to balance the spatial the small time scale. In section4 we present approximations
derivatives in the partial differential equation, we choose valid on each time scale and a uniform approximationvalid
b(Kn/Kv)l/•-asthereference
quantity
for ?. Thereferencefor all time.
quantity for g is chosento be Ib/Kv to balance the source
term and the vertical gradient term of (3). There are two Velocity Field
choicesfor the time scale, dependingon whether we balance
the time derivative in the partial differential equation (1) or Once the increase of hydraulic head has been determined,
the time derivative in the recharge boundary condition (3). the velocity field on any time scale can be obtained from the
The first case gives the small time scale increase of hydraulic head by

ts= Ssb2/Kv (6)


Vz = IV z Vz = (18)
Oz
while the secondcase gives the large time scale

t! = Syb/Kv (7) Vr = I Vr Vr = (19)


Or
Note that thesescalesare relatedby the equationtr = ts/tt.
With these scales in mind, we define dimensionlessvariables Note that the vertical velocity at the surface is approxi-
by mately equal to the vertical velocity at the initial position of
the water table (z = 1) [Dagan, 1967]. It can be obtained
from the boundary condition (12) or (16) rather than by
z b r=b t=-tl r = ts - • (8) calculating the derivative directly from the solution s(r, z, t)
ors(r, z, r).
s(r, z, t)= (K•/Ib) s-It(r), g(z), •-(t)]
(9)
s(r, z, r)= (K•,/Ib) s--[?(r),
Z(z), F(r)] 3. EXACT SOLUTION

Depending on the choice of time scale, s may be considered Increase of Hydraulic Head
as a function of (r, z, t) or (r, z, r). The problem will look
different on the two time scales. Applying Hankel and Laplace transformsto the small time
In terms of the new variables the problem takes the form scale version of the problem (14)-(17) yields the solution

r Or
r +
0-•
= tr --
Ot
(10)
s=2trR••
Jo(yr)Jl(yR)
Z tOn(y,
z,r)dy n=0
(20)

r>0 0<z<l t>0


where
Os Os
lim s(r, z, t) = O
r-->•

Or '
(0 z, t) = O -- (r, O, t) = O
OZ
too(Y,
z, r)= •o(Y)Xo(Y,
z)(1-e -r(y2-Yø2))(21)

(11) ton(Y,
z, r) = •n(Y)Xn(Y,
Z)(1-- e-r(y2+
Yn2))(22)
Os Os n>0
-- (r, 1 t) + (r, 1 t) = H(R - r) (12)
0z ' •7 ' 2

•o(Y) =
s(r, z, 0) = 0 (13) (y2_ yo2)[yo2(1
+ tr) + y2_ tr-l(y2 _ yo2)2]
on the large time scale. On the small time scalethe problem
takes the form (23)
cosh (yoZ)
Xo(y, z)=
cosh ¾o

10( 0•_rSrS
r Or ) 02s
Osr q-
Oz2 Or
(14)

•n(Y) =
2

r>0 0<z<l r>0 (y2+ yn2)[.yn2(1


+ tr)- y2+ tr-l(y2 + yn2)2]
lim s(r, z, r) = 0
(24)
COS('YnZ)
(15) Xn(Y, Z) = n> O
Os Os COS ¾n
• (0 z, r) = O -- (r, O, r) = O
Or ' OZ and the quantitiesYn(Y) are given implicitly by
Os Os tr'yosinh'Yo- (y2_ ¾02)
cosh'Yo= 0 (25)
• (r, 1, r) + a-l m (r, 1 r) = H(R - r) (16)
OZ Or '
0< yo<y
s(r, z, 0) = 0 (17)
O"Yn
sinYnq-(y2+ yn
2)cosYn= 0 (26)
No approximationshave been used at this stage, so (10)-(13)
and (14)-(17) are different versionsof the exact problem. In (n - 1/2)•r < ¾n< n•r n> 0
1622 ZLOTNIK AND LEDDER: FLOW IN COMPRESSIBLE
UNCONFINEDAQUIFERS

Details of the computationappear in the appendix.The solu- 4. ASYMPTOTIC SOLUTION FOR o' << 1
tion presentedhere is valid for all time and for all valuesof
Under the assumptionrr << 1, asymptotic approximations
to the solution of the problem for the increase of hydraulic
Vertical and Radial Velocities headcan be obtainedas seriesin increasingpowersof rr. The
problems given by (10)-(13) and by (14)-(17) will give
Combining the exact solution for the hydraulic head with
different asymptotic approximations because of differences
(19) and (18) gives the exact solutions for the dimensionless
velocities: in the way rr appearsin the equations. An approximation
valid on the largetime scale(tt) is calledthe outer expansion
and can be obtainedfrom the problem (10)-(13) by assuming
a solutionas a power seriesin rr, or it can be obtainedfrom
Vz=-2trR
f•Jo(yr)Jl(yR)
•] Un(y,
z,r)dy n=0
(27)
the exact solutionby performing an asymptoticexpansionof
the solution for rr << 1 with t constant. An approximation
valid on the small time scale (ts) is called the inner expan-

Vr
=2trR
••yJl(yr)Jl(yR)
•] to,•(y,
z,r)dy n=0
(28) sion and can be obtained from the problem (14)-(17) by
assuminga solution as a power series in rr, or it can be
obtained from the exact solution by asymptotic expansion
where
with r constant.A uniformly valid approximate solutioncan
be obtainedfrom the inner and outer expansionsby asymp-
vo(y
z r)=•to(y
, , ) •/osinh
(•/oZ) _•(y2•
cosh 7 o
(1 - e - •o)) (29) totic matching [Murdock, 1991].
Because the inner and outer approximations are not valid
for all time, they are used to replace the exact solution for
vn(y
z,r) --•n(Y)
, 7nsin
----(7nZ) COS T n
(1-- e-•ty•+•'•) further analysis. The uniform approximation is generally
more complicatedthan either the outer or the inner approx-
n> 0 (30) imation but can be used to replace the exact solution for
numerical investigation.
The leading order approximation for the vertical velocity at
the surface can also be obtained from the boundary condi-
tion (16) as Approximate Solution on the Small Time Scale

The inner approximation up to O(rr) can be obtainedfrom


Vz(r,
1,r)=-H(R
- r)+2R
••Jo(yr)Jl(yR) (14)-(17) by looking for a solution of the form

st= rru(r, z, r) + O(• 2) u = O(1) (38)


Ot.O
n
with u(r, z, r) to be determined. The choice of s = O(rr) is
ßE Or(y'1,r)dy
n=0
(31)
necessary so that the nonhomogeneousterm in (16) remains
in the equationin the limit rr--> 0. The spatialderivative term
Vertically Averaged Increase of Hydraulic Head in the equation is thus small, with the consequencethat this
boundary condition can be integrated in time to yield
The increaseof hydraulic head that is actually measuredin
an observationwell perforated for elevationsin the interval u(r, 1, r) = rH(R - r) (39)
Zl < z < z2 (Figure 1) is the average of (20) over vertical From this analysis it is clear that at the beginning of the
distance and is thus given by the formula water recharge process, virtually all of the added water
serves to produce a mound, and very little creates a down-

(S)z,,z2--
(Z
2--Zl)
-1•z[2
s(r,
Z,r)dz (32) !
ward flow of water.
The function u(r, z, r) can be obtained directly from
By interchange of the limits of integration in the double (14)-(17), with (39) replacing (16); however, it is easier to
integral defined by substitution of (20) into (32), one can obtaintheinnerapproximation
st by asymptotic
expansion
obtain the average value as of (20) as rr--> 0. The result is

(s)z•,z
2=2rrR••
Jo(yr)J•(yR)
• (•On(y, s
r))dy (33)t =
rrRr• cosh
yz
Jø(Yr)J•(YR)
n=0
cosh
y dy

where
2
- trR Jo(yr)J•(yR)F(y, z) dy
(Wo(y,r)) = ½o(Y)(Xo(Y))(1
- e-r(y2-•o)) (34)
2

(O•n(y,
r)) = •bn(Y)(Xn(Y))(1
- e-r(y2+•'n)) (35)
sinh (3/oZ2) - sinh (3/oZl) - 2rrR• (--1)nCn
COS(CnZ)An(r
, r)
()Co(Y)) = (36) n=l
7o(Z2- z!) cosh 70
sin ('YnZ2)- sin ('YnZl) q-O(tr2) (40)
()Cn(Y)) = n>0 (37)
3/n(Z2- Zl) COS3/n where
ZLOTNIK AND LEDDER:FLOW IN COMPRESSIBLE
UNCONFINEDAQUIFERS 1623

tanh y cosh y z - z sinh y z needed. The uniform approximation is a combination of the


F( y, z)= (41) inner and outer approximations constructed so that it re-
2y cosh y
duces to the inner approximation for r = O(1) and to the
outer approximation for t -- O(1) and is also valid for

An(r,
r)=f•Jo(yr)Jl(yR)e
-•(c•
+y•)• +y•)•dy (Cn
intermediate times. The uniform approximation is obtained
using the method of matched asymptotic expansions, as
describedby Murdock [1991]. The result is
(42)
n>0
st/= sø- 20-R• (-1)ncn
cos(CnZ)An(r
, •')+ 0(0'2)
n=l

Details of the computationare available from the authors. (47)

withthe errorO(0-2)for all time.


Approximate Solution on the Large Time Scale
Since 0- is a small parameter, the outer approximationup 5. LONG-TIME(t --• oo)ANALYSISOF THE SOLUTION
to O(1) can be obtained by setting 0- = 0 in (10)-(13) and
solvingthe resultingproblem. Since the problem so obtained Typical parameter values for shallow sand and gravel
is the incompressiblecase, the result is that first demon- aquifersareb = 10 m, K v = 10 m/d,andSy = 0.1, giving
strated by Dagan [1967] using Green's functions: a large time scale t t of the order of 0.1 days.
Any study of the movement of groundwater over long
rechargeperiods (t -> 1 day) will benefit from knowledgeof
sl=0 • D(r, z, t) = R Jo(yr)Jl(yR)
the limiting behavior of the solutionto the linearized prob-
lem as t --> oo.This long-time behavior can be elucidated for
cosh yz both the incompressibleand compressible cases using La-
ß(1 - e-tytanh
y) dy (43) place's method for the asymptotic expansion of integrals
y sinh y with a large parameter [Bender and Orszag, 1978]. The
This solution can also be obtained by applying a Hankel principal computationalresult is that the long-time behavior
transform, solving the resulting boundary value problem, of an integral of the form
and inverting the Hankel transform.
For the compressiblecase the outer solution can be B(r, z, t, R) = •(y, r, z, R)e -ty tanh
y dy (48)
approximatedby an asymptoticseriesin powersof 0-,with D
as the leadingorder term. Up to O(0-), the outer approxima-
tion is of the form where • has a power series expansion of the form

sø= D(r, z, t) + 0-•(r, z, t) + 0(0-2) (44)


2n+ 1

where • is to be determined. The easiestway to obtain • is by


X(y, r, z, R)= E kn(r'z, R)y no _>O
n=no
expandingthe exact solution(20) in powersof 0-with t fixed.
The result is is given by

no!
• = -• Jo(yr)Jl(yR)&(y,
z, t) dy (45)
B= 2 t • nøkn
(r,z R)+O(t-2 o) (49)
where
provided r, R = O(1).

&(y, z, t) = e-tYtanh
y [(t + 1) tanhy Incompressible Case

cosh yz The solution for the hydraulic head when 0- = 0 is given by


+ ty sech2 y - z tanhyz] (46) Dagan's formula (43). Although it is not possibleto simplify
y cosh y this formula as t --> •, it is possibleto obtain a simple result
Note that & is always positive since the singlenegative term for the long-time behavior of s by examining the time
in the square brackets is always smaller than the first derivative of Dagan's formula. From (43) and (49) comes the
result
positive term, confirmingthat compressibilityhas the effect
of decreasingthe mound growth. Since this effect appearsin Os R2
the 0(0-) perturbation, it will be small on the large time -- = • + O(t-2) r, R = O(1) (50)
Ot 4t
scale. o'=0

Integration of this result yields


Uniform Approximation to 0(0-)
R2
The approximation(40) is a valid approximationwhen t - s = •4 In (t/to)+ O(t-•) r, R = O(1) (51)
O(ts), while the approximation(44) is a valid approximation
when t = O(tl). To show the continuousdependenceof the where to is an integrationconstantthat dependson r and z
solution on time, a uniform approximation to the solution is and can be approximated numerically.
1624 ZLOTNIK AND LEDDER: FLOW IN COMPRESSIBLEUNCONFINED AQUIFERS

From (43) and (49) the vertical and radial velocities are
approximated by ....Or + O(1) Jl(yr)Jl(yR)e-tytanhy

o= -R flJo(yr)Jl(yR)
Vzl•= sinh
yz
y dy
sinh (tanhy + y sech2y) coshyz
cosh y
R2z
+• + O(t-2) r, R = O(1) (52) with the result
4t
03 R2r
cosh yz ..... + O(t -2) r, R = O(1)
Vrl,,
=o=RfiJl(yr)J•(yR) sinh y
Or 8t

Combiningthis result with the incompressiblecase(53) gives


R2r the long-time approximation
-• + O(t-2) r, R = O(1) (53)
8t

Note that the vertical velocity has a particularly simple V r -- R Jl(yr)J•(yR)


form when z = 1. In this case the long-timeapproximation
is easily obtained from (18), (12), and (50):
coshyz R 2r
R2 •dy- (1 + tr) + O(t-2)
Vz(r,1, t)l• =0= -H(R- r)+ --4t+ O(t-2) (54)
sinh
y •- r,R=O(1)(57)
Dagan's solution thus slightly overestimates the radial ve-
r, R- O(1) locityby an amountthatis O(•rt-•) ast --•
Formulas(56) and (57) show the quantitativeeffect of
Effect of Compressibility compressibility
for vei-ylargetimes.Note that the effectof
compressibility on the hydraulic head does not disappear
Tounderstand
thelong-time
behavior
Ofthesolution
so in with time (as reported for the well problem [Neurnan, 1974]
(44) for the casewhere tr is smallbut not zero, it is sufficient and subsequently corrected [Garnbolati, 1976; Neurnan,
to examine the behavior of the function 3 that is given by 1979])but approaches a constantmultipleof •r. The effectof
(45)-(46). Any additional compressibility effects will be compressibility, on velocity does disappearwith time.
much smaller, accordingto (44), and need not be considered.
To obtain an estimate of the effect of compressibility on
the hydraulic head as t --• oo,we note that from (45), 6. NUMERICAL ANALYSIS AND DISCUSSION OF RESULTS

The previous sections provide exact and approximate


3= - -•-+O(1) jo(yr)J•(yR)e-ty
tanh
y formulas for the hydraulic head and flow velocities for
incompressibleand compressibleaquifers according to the
(tanhy + y sech2
y) coshyz linearized theory and neglectingthe effect of the unsaturated
zone above the water table.
y cosh y The examples and discussionthat follows are motivated
by two goals. The first is to give a description of the
Thus from (49) we obtain the result
evolutionof the groundwatermound and the velocity profile.
g2 The velocity profile is of particular importance since no
3= -•+ O(t -•) r, R=O(1) (55) results for the three-dimensional case have been published
4
previously.
and then (using (51)) To simplify computations, one would like to use Dagan's
formula to calculate vertical and radial velocities. Thus the
g2 secondgoal of the discussionis to indicate the effect of
s=•- [In(t/to)
- tr]+O(t-•)+O(tr
2) (56) compressibilityon the head and the velocities, since the use
of Dagan's formula is tantamount to neglectingthis effect.
To illustrate the solutions given by these formulas, the
r, R = O(1)
integrals in the formulas were evaluated numerically. The
For the effectof compressibility
on the verticalvelocityas procedure is not entirely routine, because the integration
t --• oowe may apply (49) to -03/Oz, where 3 is given by (45). interval is infinite and the integrandsinclude two oscillating
The result is factors and have removable singularitiesat y = 0. To avoid
these difficulties, the routine employed in these computa-
tions subdivided the y axis by placing a node at all points
---= O(t -2) r, R = O(1) where the integrand vanished owing to the vanishingof an
Oz
oscillatory factor, and each integral was evaluated on an
The long-time approximationfor the vertical velocity is thus interval [ e, L] rather than the interval [0, oo].The very small
given by (52) even for the compressiblecase. positive number e was chosento avoid removable singular-
For the effect of compressibility on the radial velocity as ities at y = 0 (if necessary) without introducing significant
t --• oowe apply (49) to error. The number L was chosen to be large enough so that
ZLOTNIK AND LEDDER: FLOW IN COMPRESSIBLE
UNCONFINEDAQUIFERS 1625

1.00

0.05
1
0.04 ............
-
Inner (.if=
Outer((t:. .
Uniform.
(•=0.1)
Exact
((t:O.1)
/
0.75

0.03
-
0.02-
$ 0.50

•• (t=O :0 1
0.01
0.25 Inner
(•=.__.•)..
1)
)
Outer (•:
............ Uniform i
0.00 -
Exact
(•:•t•1)
_

-0.01 • 0.00 I I I I I I I I
0.01 0.1

Fig. 2. Comparison of solutions for hydraulic head s for small Fig. 4. Comparison of solutions for hydraulic head s for large
times t, with r = 0.5 and z = 0.5. times t, with r = 0.5 and z = 0.5.

the neglected interval [L, •] did not make a noticeable effective value of •r, even though •r by definition is charac-
contributionto the integral. Infinite sumswere approximated teristic of the saturated zone under the water table.
by includingenoughterms to justify neglectof the remaining The range of •r values can be even wider in the case of
terms, and the implicitly defined functions 7n(Y) were areal recharge sourcesthat create vast zones of trapped air
approximated using Newton's method. Programs in FOR- and unsaturated flow. We will use •r = 0.1 as an example
TRAN are available from the authors. correspondingto the upper range of •r values. Note that
All numericalexamplesuse b - 10 m, Kv - 10 m/d, and predictive applicationsof formulas derived above must use
$y - 0.1; a valueof t -- 1 thencorresponds to thephysical field data only for estimation of •r.
time t - 0.1 days.
The ratio of source radius to saturated aquifer thickness
Hydraulic Head
(R) typically rangesfrom 0 to 10. For example, an isotropic
sand and gravel aquifer 10 m thick with a center pivot Figures 2-4 show a comparison of the various solutions
irrigation systemwith 100-mradiusgivesR - 10. Numerical and approximationsfor the hydraulic head at a typical point
examples presented here take R - 1 as a convenient value beneath the recharge area (r = 0.5, z = 0.5). Note that the
for illustration. incompressiblesolution may be considered as an approxi-
Analysis of many pumping tests in different formations mation to the exact solutionfor the compressiblecase.
reported by Neurnan [ 1979]yield values of (r that range from For small times (Figure 2) the uniform approximation is
0.001 to 0.01 and more. Larger values such as (r - 0.2 were just distinguishablefrom the exact solution. The absolute
obtained from sand tank experiments [Neurnan, 1981]. differenceis small becauseinsufficienttime has elapsedfor a
These values are much larger than values calculatedfrom the significantincreasein the hydraulic head, and both solutions
physicaldefinitionof the elastic storagecoefficient$s. An start with a head increase of zero. The inner approximation
explanation was given by Brutsaert and EI-Kadi [ 1984], who gives a good representation of the solution up to the time
analyzed the relative importance of compressibility and 0.05 (correspondingto v = 0.5) but begins to diverge after
partial saturation in unconfinedgroundwater flow and came that. Dagan's formula shows the correct trend in a qualita-
to a quantitative conclusion about the role of (r. In many tive sense, but without quantitative accuracy. The outer
situations,flow processesin the unsaturatedzone above the approximation is actually negative and therefore meaning-
water table can be accounted for by employing a high less. (Since the outer approximation is obtained under the
assumptionthat t = O(1), there is no reason to expect it to
be valid near t = 0.)
0.60 For intermediate times (Figure 3) the uniform and outer
-- -- o'=0 / approximations are indistinguishablefrom the exact solu-
tion. Note that the difference between the uniform and outer
Outer
(o'
.=0. //

$
o.4o
Inner
(?=0.!1•I
'
Exact
/////
Uniform.((t: 1)
approximationsis given by the exponentially decaying term
of the inner solution(40), which is virtually 0 even for v = 1.
Dagan's formula overestimates the hydraulic head by an
amount that is growing only slowly with time and appearsto
be approaching
a constantdeviation(-trR2/4 ,= 0.25), in
0.20 '"'...,• accordance with the analysis of (56).
For large times (Figure 4) the trends established in the
intermediate range are continued, with the deviation of
0.00 I i I I I I I I
Dagan's formula from the correct solution approximately
0.1 1 constant.

The vertically averaged hydraulic head is illustrated in


Fig. 3. Comparisonof solutionsfor hydraulic head s for moderate Figure 5 at various radii. The graph shows the approach of
times t, with r = 0.5 and z = 0.5. the deviation of Dagan's formula from the compressible
1626 ZLOTNIK AND LEDDER: FLOW IN COMPRESSIBLE
UNCONFINED AQUIFERS

1.00 0.03
:
- t=10.O

:
- t=3.0
0.02 -
: t=l .o
:
<S> 0.50 --o'• -
: t=0.1
:
O.Ol - _
t=o.o3
:
:
--

--

:
0.00 i i i i iii o.oo
O.Ol o,1 1 lO 0.00 0.20 0.40 0.60 0.80 1.00

Fig. 5. The vertically averagedhydraulic head (s) at various radii Fig. 7. Correction (due to compressibility)to the incompressible
r as a function of time t. hydraulic head as a function of depth z, with r = 0.

solution to a constant as t increases, and it shows the solution the term correspondingto Dagan's solution is just
approach to a logarithmic increase with time for t > 1. one of three leadingorder terms. There is no reasonto expect
Figure 6 shows how the variation of hydraulic head with the behavior of the compressiblesolutionto resemblethat of
depth at the center of the recharge area changeswith time. the incompressiblesolutionvery closelyfor smalltimes. How-
As the flow evolves, vertical gradients in head increase, ever, as time increases,the term correspondingto Dagan's
leading to larger downward velocity. solutionincreases,the secondterm in the inner approximation
The effect of compressibility increases with t and de- remainsO(rr), and the third term decaysrapidly to 0. In the
creaseswith r and z. The increase in the effect of compress- outerapproximation,Dagan'ssolutionis the only leadingorder
ibility with r is immediately obvious from the observation term, so that for rr sufficiently small we may regard Dagan's
that the integrand of the function • (45) achieves its maxi- solutionas being exactly correct on the large time scale.
mum magnitude at r = 0 for any value of y. The decreaseof Figure 8 showsthe variation of hydraulic head with radius
compressibility effect with z diminisheswith time, while the at the water table, with z = 1. Since the hydraulic head at
overall effect of compressibility reaches a limiting value as the water table represents the change in water table eleva-
time increases (Figure 7 and (56)). The reason for this tion due to recharge, this figure illustrates the evolution of
behavior is clear from the asymptotic results. The inner the groundwater mound causedby recharge. For small times
approximation (40) consistsof three terms. The first of these the groundwatermound is essentiallya step function, with a
correspondsto Dagan's solution on the small time scale. The uniform rise in head under the recharge area. This behavior
other two terms give the effect of compressibility. Of these, was predicted by the boundary condition (16) at the surface
the third term decaysrapidly, so that the principal small-time on the small time scale. Since the head is O(rr), the boundary
effect of compressibility is given by the second term. This condition says that in the initial stage of recharge, virtually
function vanishes when z = 1, so that the effect of com- all of the water added to the systemgoesinto the building of
pressibility is small near the water table. As time increasesto the mound (the time derivative term) rather than into the
large values, the difference between the hydraulic head in developmentof downward flow (the spatialderivative term).
compressible and incompressible aquifers is seen to ap- As time increases,the shapeof the mound becomessmooth.
proachthe constant
value- rrR2/4.
Dagan's solution clearly gives a better approximation to
Vertical Velocity
the compressiblesolutionfor large times than for smalltimes
under most circumstances, as seen from examination of the Figures 9-12 show the radial distribution of downward
inner and outer approximations (40) and (44). In the inner velocity for different times (t = 0.01, 0.1, 1.0, and 10.0,

1.00 O':0.1
o'=0 t=3.0_ • ,.oo
0.75
I N• o':0
$
0.75 _--
0.50-
.--••
_

--- .
$ 0.50
_
_

-____ t=0.3
0.25 -
_

-
_

_
-- -- t=0.1

0.00
_ -- . ¾T
0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.50 1.00 1.50 2.00 2.50 3.00

Fig. 6. Variation of hydraulic head s with depth z at various times Fig. 8. Variation of hydraulic head s with radius r at various
t, withr = 0. times, with z = 1.
ZLOTNIK AND LEDDER.'FLOW IN COMPRESSIBLE
UNCONFINED AQUIFERS 1627

0.05 0.75

0.=0.1
0.=0
• 0.=0.1
0.=0
0.50

0.03
z=0.75

/,z•0.75 /
0.25

-0.00 Z _

0.00
-

-
IF/
-

I/
-0.03 -0.25
0.00 0.50 1.00 1.50 2.00 2.50 3.00 0.00 0.50 1.00 1.50 2.00 2.50 3.00

Fig. 9. Variation of downward velocity -V z with radius r at Fig. 11. Variation of downwardve•ucity -V z with radius r at
various depths z, with t = 0.01. various depths z, with t = 1.0.

respectively). Each graph showsthe vertical velocity at the time. The third term of the inner solution (43), which is the
depths given by z = 0.95, z = 0.75, and z = 0.5. The largestterm for •near zero, becomessmall even for • = 1, so
vertical velocity at the bottom of the aquifer is zero. The that its effect is rapidly lost. For large times (t >> 1) the
vertical velocity is not a monotonic function of radius. error in Dagan's formula is essentially zero, as (52) repre-
Because there is no recharge outside of the circle r = R, sentsan excellent approximation to the vertical velocity for
water which enters the aquifer near the edge of the recharge the compressiblecase as well as the incompressiblecase.
area can move outward as well as downward; hence the Thus Dagan's formula may be safely used for t > 1.
downward velocity near the water table is greatest at the Moreover, it can be used to provide velocity estimates for
edge of the recharge area. The downward flow of water near particle tracking methods since the absolute error of time
r = R creates a regionjust outside the rechargearea where integrals of vertical velocity becomes stabilized after inter-
the head increaseswith increasing depth, creating the up- mediate times.
ward flow of water seen for r > R. This effect diminishes
with depth and diminishesalso with time. Eventually, for
Radial Velocity
t >> 1, the profile becomes a monotonic function of the
radius r at any depth. Figure 13 shows the downward Figures 14-16 show the evolution of the radial distribution
velocity at the water table, which approachesa stepfunctionof radial velocity at the depthsgiven by z = 0.95, 0.75, and
as time increases, in accordance with (54). 0, respectively. Note that at any depth the radial velocity
For small times (t << 1) the differencebetween compress- shows an absolute maximum at the edge of the recharge
ible and incompressibleaquifers is significant. Figure 9 area. The gradient of the radial velocity is large near the
showsthat compressibilitycausesan increasein downward water table but diminisheswith depth. Indeed, the radial
velocity in the earliest stage of recharge. The effect of velocity at the water table below the edge of the recharge
compressibilityis to retard the propagation of spatial hy- area cannot be computed because the integral for radial
draulic head differencesfrom the sourceof the perturbation velocity diverges at r = R. The graphs also show that the
down the aquifer. This gives a larger vertical gradient of effect of compressibility on the radial velocity is always
hydraulic head than in the incompressiblecase. Thus for small near the water table and is small at all depths for large
small times, Dagan's formula is adequate for the vertical times.
velocity near the bottom of the aquifer but gives a result The effect of compressibility on the radial velocity is
which is much too low for the downward velocity near the rather small, so that with the exception of the point r = R,
water table. z = 1, where Dagan's formula cannot be used, the use of
The error in Dagan's solution diminishes rapidly with Dagan'sformula for the compressiblecaseis justified. In the

0.30 1.00

0.20
0.=0.1
0.=0 0.75 -
_
• 0.'=0.1
0.=0
_

_
z=0.75
0.50 -
0.10 3,z=0.75 -

0.25 -
-0.00 X,•___•_•
_.•..__
__ _

-0.10 ••'"•
.•
_

0.00
-

-0.25
0.00 0.50 1.00 1.50 2.00 2.50 3.00 0.00 0.50 1.00 1.50 2.00 2.50 3.00

Fig. 10. Variation of downward velocity -V z with radius r at Fig. 12. Variation of downwardvelocity -V z with radius r at
various depths z, with t = 0.1. various depths z, with t = 10.0.
1628 ZLOTNIKANDLEDDER.'FLOWIN COMPRESSIBLE
UNCONFINED
AQUIFERS

1.00 0.60
- t=10.O
-

- t=l• o'=0,1
0'=0

0.50 0.40 -
_
t=10.O
_

0.00 _ t=O. 1 t= 10,0 0.20 -

-0.50
0.00
.....
-

0.50 1.00
I t=0.1

1.50 2.00 2.50 3.00


0.00
i 0.00 0.50 1.00 1.50 2.00 2.50 3.00

Fig. 13. Downwardvelocity - Vz at the water table (z = 1) as a Fig. 15. Variation ofradial velocity V r with radiusr at z = 0.75 at
function of radius r at various times t. various times t.

upper levels of the aquifer, Dagan's formula gives good Since R is the ratio of the recharge radius to the aquifer
results for the radial velocity even for small times. At the depth, we conclude that for a given recharge area the effect
bottom of the aquifer, Dagan's formula gives a larger radial of compressibilityis lessfor a deep aquifer, which will have
velocity for small times. This effect occursbecausechanges a small value of R, than for a shallow aquifer. Thus Dagan's
in the hydraulic head propagate toward the bottom of the formula makes a smaller absolute error for a deep aquifer
aquifer at a slowerrate for the compressiblecasethan for the than for a shallow aquifer.
incompressiblecase.
The error in using Dagan's formula for the radial velocity
will not be important in problems of contaminanttransport. 7. CONCLUSIONS
Contaminants that are present near the water table at the In contrast with most previous works on transient ground-
beginningof the recharge (at time r = 0) will not reach the water flow in an unconfinedaquifer under uniform recharge,
lower levels of the aquifer until the time is large enoughfor compressibilityof the aquifer has been taken into account
Dagan's formula to give an accurate estimate for the radial analytically. The solution obtained from this analysis gener-
velocity near the bottom. alizes Dagan's [1967] solutionfor the hydraulic head, which
The overall direction of groundwater flow is apparent by was derived by neglectingthe compressibility.
comparison of the illustrations of the vertical and radial By treating the compressibilityor,the ratio of storativity to
velocities. For the portion of the aquifer that is well within specific yield, as a small parameter, further analysis of the
the rechargearea (r < R, r % R), the radial velocity is small solutionfor the hydraulichead in a compressibleaquifer was
and the flow is primarily downward. For the portion of the performed by asymptotic methods, resulting in approxima-
aquifer that is well outside the recharge area the vertical tions to the exact solutions for head, vertical velocity, and
velocity is small and the flow is primarily outward. radial velocity on small and large time scalesand also in the
limit of very large time.
Importance of •r and R It was found that Dagan's solutionfor the hydraulic head
For all quantities of interest the error in Dagan's formula always overpredicts the growth of the groundwater mound.
as an estimate of the solution for the compressiblecase is For large times the error in Dagan's formula does not depend
proportional to or.Thus for aquifers where cris less than 0.1, on the location of the observationpoint or on the time and is
the value used for all illustrations, the error in Dagan's equalto the constantvalue-err 2/4.
formula is reduced accordingly. For small times (t < tt, where tt is of the order of 1-10
It is interesting to note that all estimates of the effect of hours) the profile of the vertical velocity at any depth z > 0
compressibility
for largetimesare proportional
to (err2). demonstratesthat beyond the edge of the recharge area the

1.25 0.30

_- 0'=0.1 0'=0.1
0'=0
_

1.00 0'=0
t=
0.20
0.75

0.50
0.10
- /
0.25

0.00 0.50 1.00 1.50 2.00 2.50 3.00


0.00 ii • • i • I i i i • I • • • • I •-i i i T i-•' i-•' I I I i I
0.00 0.50 1.00 1.50 2.00 2.50 3.00

Fig. 14. Variation of radial velocity Vr with radiusr at z = 0.95 at Fig. 16. Variation of radial velocity V r with radius r at z = 0 at
various times t. various times t.
ZLOTNIK AND LEDDER: FLOW IN COMPRESSIBLE
UNCONFINED AQUIFERS 1629

flow of water is directed upward rather than downward. cosh (r/z)


During this early period the compressibility of an aquifer
p(trr/ sinh r/ + p cosh r/)
r/= •/y2
+p
may have a significant influence on hydraulic heads, water
table shape, and velocities.
Inversion of the Laplace transform (59) yields
After F = tt the verticalvelocitygraduallybecomes
a
monotonic function of the radius r; in particular, the vertical trRJ•(yR)
velocity near the water table approaches a step function. s*(y, z, r)= #(y, z, r)
Nonsteady components of the vertical and radial velocities
show a decrease inversely proportional to time. The effect of
compressibilityon the velocities decreasesafter t = t t. For = • eP'•(y, z, p) dp
large times (t >> tt) the error in the radial velocity causedby # 2rci
.,y-i•
neglecting compressibility is inversely proportional to t,
while the error for vertical velocity is inversely proportional and inversion of the Hankel transform (58) then yields
to •.
Thus application of Dagan's formula to approximate flow
in a compressible aquifer provides a very reasonable approx-
imation for t > tt. For t < t t the relative error in Dagan's
s(r,
z,r)=trR
f•Jo(Yr)J•(yR)#(y,
z,r)dy (60)

formula can be significantfor an aquifer with tr • 0.1, but the To complete the solution, one needs only to compute the
absolute error is always small. This makes Dagan's formula function #( y, z, r). The initial condition (17) yields the result
an attractive choice for the calculation of three-dimensional
s*(y, z, 0) - 0, from which we see that #(y, z, 0) - 0.
velocity fields for various particle tracking techniques and Hence we may obtain # from the modified equation [Neu-
transport problems, where velocities are to be integrated man, 1974]
over times of at least tt.

APPENDIX: DERIVATION OF THE EXACT SOLUTION a(Y, 1 f7


z,r)= •-• +io•
•-i• G(p)dp= • Res{G(p),Pn)
(61)
For any functionf(r, z, r) we definef*(y, z, r) to be the
where
Hankel transform off, and for any functionf*(y, z, r) we
define f*(y, z, p) to be the Laplace transform of f*
[Sneddon, 1972; Neuman, 1974]' G(p)
=•b[p, r/(p)] r/ 2=y 2+p
½[r/(P)]

f*(y,
z,r)=f•Jo(yr)f(r,
z,r)rdr (58)
qb(p,r/)= (ep*- 1)p-1 cosh(r/z)

½(r/)= trr/ sinhr/ + (r/2_ y2) coshr/


f(r,
z,r)=f•Jo(yr)f*(y,
z,r)ydy and Res {G(p), pn} is the residueof G at the pole pn. The
poles of G are the roots of the equation

f-*(y,
z,p)=
f•e-P*f*(y,
z,r)dr (59)
•[r/(p)] =0 (62)

since the function •b has only a removable singularity. It is


=
1.,fy
f*(y,z,r) •-• ,/-+i•
io•eP*f*(y, z, p) dp known [Neuman, 1974] that (62) has one real root and an
infinite number of imaginary roots:

Applying Hankel and Laplace transforms to (14)-(17) yields r/o = Yo(Y), and r/n = iyn(Y) n = 1, 2,... (63)
the boundary value problem
where the real functions 'Ynare given implicitly by
02•,
(y2 + p)g-.= 0 try,o sinh•'o- (y2_ To
2)coshTo= 0 0 < •'o< Y
Oz2

2) COS
tYTnsinTn+ (y2 + Tn Tn= 0
(y, 0, p) =0
(n - 1/2)•r < 'Yn< nrr n> 0
Og* RJ•(yR)
(y, 1, p) + tr-lpg*(y, 1, p)= The calculation of residuescan be performed by [Churchill
Oz yp and Brown, 1984]

The solution of this problem is

g* =
trRJi(yR)
O(y, z, p)
qb[p,
r/(p)]
} qb[Pn,
r/(Pn)]
Res •[r/(p)], Pn= d•- 2r/nqb(P
d• r/n)
dp
[ r/ ( P n)]
dr/
( r/ n)

where Hence
1630 ZLOTNIK AND LEDDER: FLOW IN COMPRESSIBLEUNCONFINED AQUIFERS

r/n4>(Pn, r/n) REFERENCES


g(y,z, r)=2 Z ton(y,
z, r), ton= Bender, C. M., and S. A. Orszag, Advanced Mathematical Methods
n=0
('r]n) for Scientists and Engineers, McGraw-Hill, New York, 1978.
dr/ Brutsaert, W., and A. I. E1-Kadi, The relative importance of
(64) compressibilityand partial saturation in unconfinedgroundwater
flow, Water Resour. Res., 20(3), 400-408, 1984.
With Churchill, R. V., and J. W. Brown, Complex Variables and Appli-
cations, 4th ed., McGraw-Hill, New York, 1984.
e p"r-- 1 Dagan, G., Linearized solutions of free-surface groundwater flow
qb(Pn, tin)= cash (rt .z) with uniform recharge, J. Geophys. Res., 72(4), 1183-1193, 1967.
Pn Dillon, P. J., An analytical model of contaminant transport from
diffuse sources in saturated porous media, Water Resour. Res.,
25(6), 1208-1218, 1989.
2 2 cash (rt .z) Gambolati, G., Transient free surface flow to a well: An analysis of
Y -r/n theoretical solutions, Water Resour. Res., 12(1), 27-39, 1976.
Hantush, M. S., Growth and decay of groundwater-mounds in
responseto uniform percolation, Water Resour. Res., 3(1), 227-
(r/n)=r/•-l[r/n2(1
+o')+y2-o'-l(y2-r/n2)
2]cashr/n 234, 1967.
Hunt, B. W., Vertical recharge of unconfined aquifers, J. Hydraul.
Div. Am. Soc. Civ. Eng., 97(7), 1017-1030, 1971.
and r/• given by (63), the final result (20) is given by (60) and Kroszinksy, U. I., and G. Dagan, Well pumping in unconfined
(64). aquifers: The influence of the unsaturated zone, Water Resour.
Res., 11(3), 479-490, 1975.
Lennon, G. P., L.-F. Liu, and J. A. Ligget, Boundary integral
NOTATION
equation solution to axisymmetric potential flows, 2, Recharge
b initial saturated thickness of the aquifer. and well problems, Water Resour. Res., 15(5), 1107-1115, 1979.
D(r, z, t) Dagan's solution for s for an incompressible Levy, B. S., P. J. Riordan, and R. Schreiber, Estimation of leak
rates from underground storage tanks, Ground Water, 28(3),
aquifer.
378-384, 1990.
H unit step function. Morel-Seytoux, H. J., C. Mirzacapillo, and M. J. Abdulrazak, A
I net specific recharge at the water table. reductionist approach to a unsaturated aquifer recharge from a
J0 Bessel function of the first kind of order 0. circular spreading basin, Water Resour. Res., 26(4), 771-777,
J1 Bessel function of the first kind of order 1. 1990.

Kh horizontal hydraulic conductivity. Murdock, J. A., Perturbations: Theory and Methods, 509 pp., John
Wiley, New York, 1991.
Kv vertical hydraulic conductivity.
National Research Council, Ground Water Models: Scientific and
r, ? dimensionless and dimensional radius.
Regulatory Applications, 333 pp., Committee on Ground Water
R, R dimensionless and dimensional radius of Modeling Assessment, Water Science and Technology Board,
recharge area. Commissionon Physical Sciences, Mathematics, and Resources,
s, •- dimensionless and dimensional increase of National Academy Press, Washington, D.C., 1990.
hydraulic head over its initial value. Neuman, S. P., Effect of partial penetration on flow in unconfined
(s) vertical average of s from z = 0 to z = 1. aquifers considering delayed gravity response, Water Resour.
Res., 10(2), 303-312, 1974.
(S)zl,Z2verticalaverage
of s fromZl to z2. Neuman, S. P., Analysis of pumping test data from anisotropic
s• inner(smalltime)approximation to s for small unconfinedaquifers consideringdelayed gravity response, Water
0'. Resour. Res., 11(2), 329-342, 1975.
so outer(largetime)approximation
to s for small Neuman, S. P., Perspective on "delayed yield," Water Resour.
0'.
Res., 15(4), 899-908, 1979.
Neuman, S. P., Delayed drainage in a stream-aquifer system, J.
sv uniformapproximation
to s for small Irrig. Drain. Div. Am. Soc. Civ. Eng., 107(IR4), 407-410, 1981.
Ss specific (elastic) storage. Ostendorf, D. W., D. A. Reckhow, and D. J. Popielarczyk, Vertical
Sy specificyield. transport processesin unconfinedaquifers, J. Contam. Hydrol.,
t dimensionless time on the large time scale. 4, 93-107, 1989.
t dimensional time. Sneddon, I. H., The Use of Integral Transforms, McGraw-Hill,
New York, 1972.
t l large time scale.
t s small time scale.
G. Ledder, Department of Mathematics, University of Nebraska,
Vr dimensionlessradial velocity. Lincoln, N E 68588.
V z dimensionlessvertical velocity. V. Zlotnik, Department of Geology, University of Nebraska,
z, g dimensionless and dimensional distance from Lincoln, NE 68588.
the bottom of the aquifer.
tr compressibility parameter.
r dimensionless time on the small time scale.

Acknowledgment. The present work was partially supportedby (Received May 29, 1991'
the Water Center of the University of Nebraska--Lincoln and the revised February 6, 1992;
MSEA project. accepted February 20, 1992.)

Das könnte Ihnen auch gefallen