Estimation of relative recharge sequence to groundwater with minimum entropy deconvolution
-
Upload
taehee-kim -
Category
Documents
-
view
212 -
download
0
Transcript of Estimation of relative recharge sequence to groundwater with minimum entropy deconvolution
Estimation of relative recharge sequence to groundwater
with minimum entropy deconvolution
Taehee Kima, Kang-Keun Leeb,*
aKorea Institute of Geosciences and Mineral Resources, South KoreabSchool of Earth & Environmental Sciences, Seoul National University, Kwanak Gu San 56-1, Seoul 151 742, South Korea
Received 30 December 2003; revised 25 November 2004; accepted 14 December 2004
Abstract
This study applies the minimum entropy deconvolution (MED) for the estimation of the sequential groundwater recharge
rate. Groundwater recharge rates have conventionally been estimated as an average value over some period of time, e.g. annual
or seasonal recharge rates. Such estimates however, are not suitable for the analysis of the dynamic sequential behavior of the
groundwater head fluctuation. If the sequential groundwater recharge rate is obtained, numerical groundwater models can be
effectively applied to evaluate the effect of any change in the groundwater system following the change in natural or artificial
components. This study successfully applies MED to estimate the sequential recharge rates. As recharge rates are obtained by
relative values, a series of timed observations are necessary. The validity of the estimated sequences of relative recharge rates
can be checked by cross-referencing. Cross-correlations between the applied recharge sequence and the estimated results are
above 0.985 in all study cases. Through the numerical test, the suitability of MED in the estimation of the recharge sequence to
groundwater is investigated.
q 2005 Elsevier B.V. All rights reserved.
Keywords: Groundwater; Recharge estimation; Sequential recharge; Minimum entropy deconvolution; Numerical test
1. Introduction
The estimation of groundwater recharge is a
classical topic in hydrogeology. A large number of
methods have been introduced for the estimation of
groundwater recharge. Scanlon et al. (2002) reviewed
various kinds of techniques for quantifying
0022-1694/$ - see front matter q 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhydrol.2004.12.005
* Corresponding author. Tel.: C82 2 880 8161; fax: C82 2 871
3269.
E-mail addresses: [email protected] (T. Kim), kklee@
snu.ac.kr (K.-K. Lee).
groundwater recharge. They subdivided techniques
for estimation recharge into four types, water budget,
techniques based on surface-water studies, techniques
based on unsaturated studies and techniques based on
saturated studies.
The former two categories are related to indirect
approaches to groundwater systems. And the latter
two categories are directly related to physical
groundwater systems. However, from the methodo-
logical viewpoint, it is hard to distinguish the latter
two cases from each other. Therefore, we tried to re-
categorize these methods to estimate groundwater
Journal of Hydrology 311 (2005) 8–19
www.elsevier.com/locate/jhydrol
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–19 9
recharge into four types with the viewpoint of
monitoring data types and resulting types of recharge
estimate focused on whether the resulting estimate is
given by dynamic time-series or a representative
recharge value for some period of time. That is, the
re-categorization criteria are closely related to the
ranges of time period the resulting recharge estimates
can represent. The four types are: (1) the water
balance method, (2) baseflow/springflow recession
analyses, (3) analyses of chemical or physical tracers,
and (4) dynamic recharge estimation from time
series data.
The first type is related to the water balance (WB)
technique. WB is the most conventional method to
estimate the recharge. Bradbury and Rushton (1998)
estimated a runoff-recharge model using the water
balance method, Finch (1998) investigated the
sensitivity of direct groundwater recharge to land
surface parameters with a water balance model, and
Bekesi and McConchie (1999) applied the Monte
Carlo technique to the water balance model to
estimate the groundwater recharge.
The estimation of the groundwater recharge from
baseflow/springflow recession is another type of
method which has been widely employed. Many
studies related to this method have been carried out
(Rutledge and Daniel, 1994; Avery et al., 1999;
Ketchum et al., 2000). But methods related to the
baseflow/springflow recession are directly correlated
with the discharge, rather than the recharge.
The third type is the method of groundwater
recharge estimation by measuring the groundwater
age with environmental tracers or temperature
profiles (Solomon et al., 1993; Leduc et al., 1997;
Bromley et al., 1997; Taniguchi, 1993; Taniguchi
et al., 1999a,b). The applications included in the
above-mentioned three groups are mostly concerned
with the estimation of groundwater recharge as forms
of seasonal average or annual recharge rates. If there
are some studies related to the estimation of a daily
or dynamic recharge rate, those studies are somewhat
or highly related to the following fourth type.
To estimate more dynamic fluctuations of ground-
water recharge, a method for sequential estimation of
recharge is necessary. Unfortunately, it is not easy to
find studies on sequential approaches to estimate the
recharge rate to groundwater. We categorized the
fourth type as methods related to the sequential
estimation of groundwater recharge rate. Sometimes,
the methods related to the fourth group are called
‘saturated-volume fluctuation (SVF) analysis
methods’. Ketchum et al. (2000) proposed the storage
accumulation method calibrated with the spring
discharge records. Healy and Cook (2002) tried to
extend the discussion of SVF to the estimation of
specific yield in fractured media. But in these studies,
the local discharge, described with the recession of the
hydraulic head, was not considered. The local
discharge for the studied area is merely considered
as the area average by dividing the spring flow
discharge by the catchment area.
On the other hand, there are many studies on the
estimation of the source sequence with time series
data in geophysics (Broadhead, 1993; Sacchi et al.,
1994; Kaaresen and Taxt, 1998; Sibul et al., 2002).
But there have been only rare applications of
geophysical methods to the estimation of the ground-
water recharge. Even though predictive deconvolution
techniques are one of the most prevalent methods for
the estimation of source wavelets in signal processing,
they have rarely been applied to the estimation of
recharge to groundwater.
In general, the classical predictive deconvolution
technique is based on the minimum phase condition.
However, the minimum phase condition can hardly be
satisfied in nature. On the contrary, the input signal in
a natural system can be considered as a random signal.
To apply the classical predictive deconvolution, some
appropriate sequence satisfying the minimum phase
condition must be picked up. To avoid the strong
restriction of the minimum phase condition, Wiggins
(1978) proposed another deconvolution technique,
called the minimum entropy deconvolution (MED).
Wiggins (1978) insisted that the MED process seeks
the smallest number of large spikes that is consistent
with the data. In other words, MED finds the simplest
signals consistent with the observed data. Wiggins
(1978) represented the simplicity with the varimax
norm. However, the solution process with the varimax
norm is non-linear. To acquire the desired digital
filters for the estimation of source wavelets, some
iteration is needed in the MED proposed by Wiggins.
For the linearization of the MED problem, Carbrelli
(1984) suggested another criterion, the D norm,
instead of the varimax norm. Carbrelli (1984) showed
that the behavior of the D norm is consistent with
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–1910
the varimax norm. Using the D norm as a criterion, the
solution process in the MED is linear, and iteration is
not needed. Thus MED was applied and developed in
various geophysical studies (Broadhead, 1993; Sacchi
et al., 1994).
In this study, the MED with the D norm is applied
to the estimation of the sequence of relative recharge
rates. The applicability of MED to evaluation of the
recharge sequence is to be investigated with
synthesized data obtained from numerical modeling.
2. Mathematical framework
In linear systems, the total response to the external
stresses can be described with a convolution equation
hi Z r � ii Cni (1)
where hi is the total response, r is the system response,
ii represents the external stresses, ni is the noise
component, and * denotes the convolution. hi is based
on N observed data. Eq. (1) can be described in matrix
form
H Z RI C n̂ (2)
where H, R, I are the matrices of h, r, and i,
respectively. Now, another matrix, f, is assumed to
operate on both sides of Eq. (2). Eq. (2) is transformed
into
f � hi Z f � ðr � iiÞC f � ni Z yi (3)
The operation of f in Eq. (2) is the convolution.
When f operates on the above equation, f*(r*ii) must
approximate to the original stress, ii and f*n should be
eliminated. The MED is based on the assumption that
the Y obtained from applying the filter F on H has a
simple structure (Carbrelli, 1984). The problem of
composing the matrix with the greatest simplicity
from an infinite set is that of the maximization or the
minimization of a norm that can measure the degree of
simplicity. Wiggins (1978) proposed the varimax
norm as a criterion for the measurement of the degree
of simplicity. The definition of the varimax norm is
VðAÞ ZX
ViðAiÞ (4)
ViðAiÞ Z
Pj a4
ijPj a2
ij
� �2
In statistical sense, Vi(Ai) represents the variance
of the squares of the normalized rows of matrix A.
The method proposed by Wiggins (1978) composes
the applied filter with a finite length to maximize the
varimax norm in Eq. (4). However, this process
achieves a solution that maximizes the varimax norm
through a non-linear process. To eliminate nonlinear-
ity in arriving at a solution, the D norm was proposed
by Carbrelli (1984). The simplicity proposed by
Carbrelli (1984) was a function of the distance from
y to the points e1,.,e2m where
ei Z ð0;.; 1i;.; 0Þ; i Z 1;.;m
emCi Z ð0;.;K1i;.; 0Þ; i Z 1;.;m(5)
Using 2m unit vectors, a new criterion for the
simplicity of the output y was defined (Carbrelli,
1984)
D1ðyÞ Z min0%i%2m
kðy=kykÞKeik2 (6)
The term, k(y/kyk)Keik2 is rewritten as follows:
kðy=kykÞKeijj2 Z2K2ðyi=kykÞ; i Z1;.;m
kðy=kykÞKeijj2 Z2C2ðyi=kykÞ; i ZmC1;.;2m
(7)
Hence,
min0%i%2m
kðy=kykÞKeijj2 Z 2 1 K max
0%i%mðjyij=kykÞ
� �(8)
Therefore, to minimize the D1(y), the following
term must be maximized (Carbrelli, 1984)
DðyÞ Z max0%i%m
ðjyij=kykÞ (9)
Consequently, D1(y) and D(y) are equivalent to
each other. The newly defined D norm is consistent
with the varimax norm (Fig. 1).
Now, to perform the predictive deconvolution with
MED, a digital filter with finite length l maximizing
D(y) should be calculated. From the mathematical
viewpoint, the following two processes are equivalent,
as both of the two processes are linear (Carbrelli, 1984):
Fig. 1. Plot of norms D(y) and V(y) for output vector with (cos a,
sin a) filter ([a2[0,2p]) and xZ(1,1) (a), (K1,K2) and (100,1)
(Carbrelli, 1984).
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–19 11
(i)
to explore the D norm from vector y and to find theappropriate filter to maximize the D norm;
(ii)
to find the appropriate filters for each component ofyi/y, compare the yi/y to each other and define the
filter of the maximum yi/y.
The equivalence between the above two statements
can be described as:
supf2R
DðyÞ Z supf2R
½maxiðjyij=jjyjjÞ�
Z maxi
supf2R
ðjyij=jjyjjÞ
" #(10)
It remains to determine the appropriate filter in order
for each of the normalized components to be
maximized. This filter can be found by differentiating
with respect to the coefficients fk of the filter and
equating to zero
v
vfk
ðjyij=kykÞ Z 0; k Z 1;.; l (11)
The differentiated components are defined as
kyk ZX
j
y2j
!1=2
(12)
yi ZX
s
fshiKsC1 (13)
and the derivative of each component is
vyi
vfkZ hiKkC1 (14)
vkyk
vfk
Z kyjjK1X
j
yj
vyj
vfk
Z kyjjK1X
j
yjhjKkC1 (15)
Therefore, the left-hand side of Eq. (11) can be
rewritten as:
v
vfk
ðyi=jjyjjÞ
Z jjyjjK2 hiKkC1jjyjjKjjyjjK1X
j
yjhjKkC1
!yi
( )
Z jjyjjK1 hiKkC1 KyijjyjjK2
!
Xj
Xs
fshjKsC1hjKkC1
!)
(16)
Equating to zero gives
yi=kyjj2X
j
fsX
s
hjKsC1hjKkC1
!Z hiKkC1 (17)
Let
rsKk ZX
s
hjKsC1hjKkC1 (18)
Fig. 2. Conceptual model used in numerical modeling to synthesize
the hydraulic data.
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–1912
Eq. (18) then substitutes into Eq. (17)
yi=kyjj2X
j
fsrsKk
!Z hiKkC1 (19)
Eq. (19) can be converted into matrix form
yi=kyjj2R$f Z hi (20)
where
R Z
r0 r1 / rlK1
r1 r0 / «
« « r1
rlK1 / r1 r0
26664
37775
hi Z½hi hiK1 / hiKlC1 �T; if iKk!0; hiKlC1 Z0
R here is the autocorrelation matrix.
Using equations from (11) to (20) (Carbrelli, 1984),
some appropriate filters and the D norm for each filter
are calculated, and the maximum for each of the D
norms is identified. Through this process, the desired
filter can be obtained.
3. Numerical test for the applicability of MED
For the applicability of MED in the estimation of the
recharge sequence to groundwater, a numerical test
was performed in a two-dimensional domain. For an
unconfined aquifer, the Boussinesq equation was
adapted as the governing equation, as follows
v
vxhKx
vh
vx
� �C
v
vyhKy
vh
vy
� �Cq Z Sy
vh
vt(21)
where h denotes hydraulic head (saturation thickness),
K denotes hydraulic conductivity, Sy is specific yield, q
is recharge, x, y denote coordinates, and t is time.
Eq. (21) for an unconfined aquifer is nonlinear, and this
equation cannot be solved directly with a numerical
model. A linearized Boussinesq equation was therefore
applied for the numerical test. In the equation for an
unconfined aquifer, vZh2 is substituted
v
vxKx
vv
vx
� �C
v
vyKy
vv
vy
� �C2q Z
Syffiffiffi�v
pvv
vt(22)
3.1. Numerical tests with simple aquifer systems
Basically, a deconvolution technique can be applied to
the linear system However, the Boussinesq equation is
nonlinear, even though the equation would be linearized.
The tested domain is 200 m!200 m. The boundary
conditions and data sampling points are identified in
Fig. 2. The boundary conditions are set with specified
hydraulic heads and saturation thickness. Head variation
along with recharge sequence was obtained at three
points: 218, 221, 224 nodal points (Fig. 2). To confirm the
applicability of MED to the unconfined aquifer, two
numerical tests were performed with two simple systems:
(a) homogeneous aquifers, (b) simple heterogeneous
aquifers with 40, 50, 60, 70, 80, 90, 100 m saturation
thickness (Fig. 3). The applied sequence of recharge in
these cases is presented in Fig. 4(a) and hydraulic
parameters are assigned as below:
KZ1.0 m/day, SyZ1.0!10K1;
KZ9.0 m/day, SyZ1.0!10K1 in the dark part.
3.2. Numerical test with random field of hydraulic
conductivity
A conceptual model based on the geostatistical
approach, approximating the real aquifer, was applied.
Fig. 3. Distribution of hydraulic parameters in the vertically layered
systems: (a) case 1, (b) case 2.
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–19 13
The spatial distribution of hydraulic conductivity was
constructed through random generation with the turning
band method (Tompson et al., 1989). The statistical
model for random generation is:
†
assumed distribution: lognormal distribution;†
expectation of hydraulic conductivity: 3.0 m/day;†
assigned variance (log): 0.4;†
correlation length (l): 10 m;†
applied covariance structure: spherical modelCðrÞ Z 1 K3r
2lK
r3
2l3; 0%r% l
0; rR l
8<: (23)
The assigned expectation of hydraulic conductivity is
within the conventional ranges in weathered, fractured
rocks. The generated hydraulic conductivity field is
shown in Fig. 5. In this case, the value of the specific
yield is also set as 1.0!10K1. Also the initial saturation
thickness is 30 m. The sequences of hydraulic head
through the numerical simulation are in Fig. 6. For the
application of the boundary condition, the recharge
sequence was randomly generated. The sequence of the
random recharge is set out in Fig. 4(b).
3.3. Numerical test with induced noise
To make the synthesized sequences more similar to
the real monitoring data set, three sequences of white
noise were imposed on the synthesized results through
the random field of hydraulic conductivities The
standard deviations (sn) of the applied noise sequences
are 0.025, 0.05, 0.075 m. When we monitor the
hydraulic sequences with commercial pressure transdu-
cers—30 m of maximum pressure head—the maximum
error can be up to 0.03 m from the sensor itself.
Synthesized sequences with white noise are shown in
Fig. 8(a).
4. Results of the application of MED to synthesized
data sets
MED with D norm was applied to estimate the
sequence of relative recharge with synthesized data sets.
In actual application of the MED, it is the most sensitive
problem how the length of the applied filter can be
determined. The statistical concepts of MED come from
factor analysis in multivariate statistics. In factor
analysis, the number of factors is determined by
eigenvalue analysis. In general, the number of eigen-
values above 1 is applied as the number of factors in
factor analysis. In this study, to determine the length of
the filter, a householder transform was applied to
evaluate the number of eigenvalues above 1 from the
constructed autocovariance matrix, firstly. In most
cases for the models tested the filter length is above 20.
Fig. 4. Applied recharge sequence to numerical test: (a) for cases 1 and 2, (b) for random field of hydraulic conductivity.
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–1914
With a very simple sequence of assumed recharge, the
above method using eigenvalues may be used to estimate
the sequence of recharge. However, when synthesized
sequences from numerical modeling were considered,
the results showed some time shift and dissipation. Then
trial and error were used in this study to determine the
length of the filter. In most cases, it was revealed that 2 is
appropriate for the length of the filter to estimate the
recharge sequence in groundwater phenomena in applied
models. All the applied lengths of filters in this study
were 2. The physical meaning of this length will be
discussed later. The results of the estimations are
presented in Figs. 7 and 8, and Tables 1–4.
Fig. 5. Spatial distribution of hydraulic conductivity
To evaluate the applicability of this method, cross-
correlations between the applied recharge sequence and
the estimated results were made, because the results of the
methods proposed in this study are presented as relative,
not absolute values. The similarity in the signal shape is
the measure of the fit in cross-correlation between the
plots of differently timed series of observations. The more
similar any two compared sequences, the closer their
cross-correlation fit is to 1. When the cross-correlation is
1, the sequences are identical in shape. In this study, all
cross-correlations that were made achieved a conformity
higher than 0.985 (simple aquifer systems with various
saturation thicknesses, Tables 1 and 2) and higher
in arithmetic scale (a), and in log scale (b).
Fig. 6. Results from numerical modeling with random hydraulic conductivity field.
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–19 15
than 0.992 (random field of hydraulic conductivity,
Table 3, Fig. 7). In cases with simple aquifer systems,
cross-correlations are slightly increased along with the
increase of the saturation thickness. This means that the
unconfined aquifer could approximate to the linear
system as the saturation thickness gets thicker and
thicker. And, from the results of the tested cases, it can be
concluded that MED can be applied to estimate the
recharge sequences in the unconfined case. Based on
the results from the simple aquifer cases, we applied the
MED to the results synthesized with the random field
of hydraulic conductivity without white noise and with
three cases of white noise sequences. In the case without
white noise, the estimated results show very good
correlations with the assigned sequence of recharge
(Table 3, Fig. 7). Even in the cases with white noise
sequences, the cross-correlations show values higher
than 0.83 (Table 4, Fig. 8).
5. Discussion
In this study, the full procedure using the MED to
estimate the sequence of relative recharge with the
sequence of hydraulic head was discussed with a
numerical test. In numerical tests, the applicability of
MED to the estimation of the relative recharge
sequences was investigated. For the investigation, a
linearized Boussinesq equation was employed. In
addition, the conceptual model based on the geostatis-
tics, approximating the real aquifer, was applied. The
results of these numerical tests demonstrated the
applicability of the MED. All the values of cross-
correlations in the numerical tests are above 0.993
(Table 1), and the resultant wavelets of recharge are
almost the same as the original sequence of recharge.
We also tried to justify the application of MED when
some random noise is introduced into the monitored data
Fig. 7. Estimated sequence of relative recharge with MED in numerical test.
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–1916
set. Even though the cross-correlations decreased as the
standard deviations of induced white noise increase in
tested cases (Table 4), cross-correlations with daily
fluctuating data sets are higher than 0.98 (snZ0.025 m),
0.92 (snZ0.05 m), and 0.83 (snZ0.075 m). In general,
a cross-correlation higher than 0.8 means that two
sequence are highly similar in the statistical sense.
Therefore, MED can be applied under the conditions of
white noise due to the accuracy of the sensor and/or the
measuring error. In this study, we did not consider the
artificially stressed conditions such as pumping, artificial
drain, and injection because those kinds of disturbances
can be removed with a simple superposition model.
It is noticeable that the length of filter in every case
is 2. The length 2 has an important mathematical and
physical meaning in that the system response is an
exponential decay. In all results through the MED,
small negative spikes follow the estimated recharge
(a positive spike). This is due to the length of filter, 2.
The length, 2, means that the system response shows
an exponential behavior. It depends on the system
design whether the system response is exponential or
not. The apparent response with a discrete time
domain can be described in terms of an exponential
decay. The assumption of the exponential decay
introduces an error into the results of estimation with
Fig. 8. Results from numerical modeling with induced noise (a), and estimated sequence of relative recharge with MED (b) at node 221.
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–19 17
the MED, the negative spikes following the estimated
impulses. For simple comparison, the other sequence
of hydraulic head was synthesized through the time
domain convolution with a response function that
Table 1
Correlation between applied recharge and estimated results with the simp
Saturation
thickness
40 50 60 7
218 9.85!10K1 9.87!10K1 9.89!10K1 9
221 9.85!10K1 9.87!10K1 9.89!10K1 9
224 9.85!10K1 9.87!10K1 9.89!10K1 9
Table 2
Correlation between applied recharge and estimated results with simple h
Saturation
thickness
40 50 60 7
218 9.90!10K1 9.92!10K1 9.94!10K1 9
221 9.91!10K1 9.93!10K1 9.94!10K1 9
224 9.90!10K1 9.92!10K1 9.94!10K1 9
decays exponentially with time. The same sequence of
recharge in Fig. 4(b) was assumed to work on the
assumed exponential system. Using a synthesized head
sequence, the sequence of relative recharge was
le homogeneous aquifer
0 80 90 100
.90!10K1 9.91!10K1 9.92!10K1 9.93!10K1
.90!10K1 9.91!10K1 9.92!10K1 9.93!10K1
.90!10K1 9.91!10K1 9.92!10K1 9.93!10K1
eterogeneous aquifer
0 80 90 100
.95!10K1 9.92!10K1 9.96!10K1 9.97!10K1
.96!10K1 9.93!10K1 9.97!10K1 9.98!10K1
.95!10K1 9.96!10K1 9.96!10K1 93.97!10K1
Table 3
Correlation between applied recharge and estimated results in the
case with random field of hydraulic conductivity
Node 218 221 224
Correlation 9.92!10K1 9.92!10K1 9.92!10K1
Hydraulic
conductivity
(m/day) around
the node
4.502 1.871 7.782!10K1
2.372 1.336 2.417
3.239!10K1 1.630 1.346
2.836 2.204 3.393
Table 4
Correlation between applied recharge and estimated results with
induced noise
Standard
deviation (m)
218 221 224
2.5!10K2 9.81!10K1 9.82!10K1 9.82!10K1
5.0!10K1 9.23!10K1 9.27!10K1 9.23!10K1
7.5!10K2 8.37!10K1 8.43!10K1 8.43!10K1
Fig. 9. Comparison of results from numerical modeling a
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–1918
estimated with the MED (Fig. 9(d)). Cross-correlation
conformity is over 0.994. When the negative spikes
are removed from the estimations, cross-correlation
conformity in the study increases greatly, up to 0.999.
These results indicate that two sequences, the applied
sequence of recharge in numerical modeling and the
estimated sequences of relative recharge with the
MED, are almost identical in their shape.
The MED has a significant limitation in that the result
of estimation is provided in the form of a relative
sequence, not absolute values. Therefore, for the exact
estimation of the recharge sequence, the level of
recharge should be recorded at least one time or as the
total amount of recharge over a period of time. It is
difficult to measure the recharge rate for the ground-
water, and long-term monitoring of the recharge
sequence is very expensive. From this viewpoint,
utilization of the MED can be a powerful and convenient
method in the estimation of the recharge sequence for
groundwater.
nd convolution with exponential response function.
T. Kim, K.-K. Lee / Journal of Hydrology 311 (2005) 8–19 19
Acknowledgement
This study was supported by grants from Sustainable
Water Resources Research Center of the Frontier
Research Program (#3-2-2 and #3-4-2).
References
Avery, W.H., Donovan, J.J., Ketchum, J.N., 1999. Recharge estimation
by stage-discharge interpolation of springflows from cross-
correlated well measurements. Ground Water 37 (3), 332–337.
Bekesi, G., McConchie, J., 1999. Groundwater recharge modelling
using the Monte Carlo technique, Manawatu region, New Zealand.
Journal of Hydrology 224, 137–148.
Bradbury, C.G., Rushton, K.R., 1998. Estimation runoff-recharge in
the Southern Lincolnshire Limestone catchment, UK. Journal of
Hydrology 211, 86–99.
Broadhead, M.K., 1993. Minimum entropy deconvolution of under-
water acoustic transients. ASA 126th Meeting, Denver.
Bromley, J., Edmund, W.M., Fellman, E., Brouwer, J., Gaze, S.R.,
Sudlow, J., Taupin, J.-D., 1997. Estimation of rainfall inputs and
direct recharge to the deep unsaturated zone of southern Niger
using the chloride profile method. Journal of Hydrology 188–189,
139–154.
Carbrelli, A.C., 1984. Minimum entropy deconvolution and simplicity:
a non iterative algorithm. Geophysics 50 (3), 394–413.
Finch, J.W., 1998. Estimating direct groundwater recharge using a
simple water balance model—sensitivity to land surface par-
ameters. Journal of Hydrology 211, 112–125.
Healy, R.W., Cook, P.G., 2002. Using groundwater levels to estimate
recharge. Hydrogeology Journal 10 (1), 91–109.
Kaaresen, K.F., Taxt, T., 1998. Multichannel blind deconvolution of
seismic signals. Geophysics 63 (6), 2093–2107.
Ketchum, J.N., Donovan, J.J., Avery, W.H., 2000. Recharge
characteristics of a phreatic aquifer as determined by storage
accumulation. Hydrogeology Journal 8 (6), 579–593.
Leduc, C., Bromley, J., Schroeter, P., 1997. Water table fluctuation and
recharge in semi-arid climate: some results of the HAPEX-Shale
hydrodynamic survey (Niger). Journal of Hydrology 188/189,
123–138.
Rutledge, A.T., Daniel III., C.C., 1994. Testing an automated method
to estimate ground-water recharge from streamflow records.
Ground Water 32 (2), 180–189.
Sacchi, M.D., Velis, D.R., Cominguez, A.H., 1994. Minimum entropy
deconvolution with frequency-domain constraints. Geophysics 59
(6), 938–945.
Scanlon, B.R., Healy, R.W., Cook, P.G., 2002. Choosing appropriate
techniques for quantifying groundwater recharge. Hydrogeology
Journal 10 (1), 18–39.
Sibul, L.H., Roan, M.J., Erling, J., 2002. Deconvolution and signal
extraction in geophysics and acoustics. The Journal of the
Acoustical Society of America 112 (5), 2389.
Solomon, D.K., Schiff, S.L., Poreda, R.J., Clarke, B.W., 1993. A
validation of the 3H/3He method for determining groundwater
recharge. Water Resources Research 29 (9), 2951–2962.
Taniguchi, M., 1993. Evaluation of vertical groundwater fluxes and
thermal properties of aquifers based on transient temperature-depth
profiles. Water Resources Research 29 (7), 2021–2026.
Taniguchi, M., Shimada, J., Tanaka, T., Kayane, I., Sakura, Y.,
Shimano, Y., Dapaah-Siakwan, D., Kawashima, S., 1999a.
Distribution of temperature-depth profiles due to surface climate
change and subsurface water flow. 1. An effect of linear increase in
surface temperature caused by global warming and urbanization in
the Tokyo metropolitan area, Japan. Water Resources Research 35
(5), 1507–1517.
Taniguchi, M., Williamson, D.R., Peck, A.J., 1999b. Distribution of
temperature-depth profiles due to surface climate change and
subsurface water flow. 2. An effect of step increase in surface
temperature caused by forest clearing southwest Western Australia.
Water Resources Research 35 (5), 1519–1529.
Tompson, A.F.B., Ababou, R., Gelhar, L.W., 1989. Implementation of
three dimensional turning bands random field generation. Water
Resources Research 25 (10), 2227–2243.
Wiggins, R.A., 1978. Minimum entropy deconvolution. Geoexporation
16, 21–35.