Vous êtes sur la page 1sur 14

Journal of Hydrology 305 (2005) 1–14

www.elsevier.com/locate/jhydrol

Estimating groundwater mixing ratios and their uncertainties using


a statistical multi parameter approach
Joerg Rueedia,b,*, Roland Purtschertb, Urs Beyerleb,c, Carmen Alberichd, Rolf Kipferc
a
Robens Centre for Public and Environmental Health, EIHMS, University of Surrey, Guildford, Surrey GU2 9WA, UK
b
Institute of Climate and Environmental Physics, University of Bern, Sidlerstrasse 5, 3012 Bern, Switzerland
c
Water Resources and Drinking Water, EAWAG, Ueberlandstrasse 133, 8600 Duebendorf, Switzerland
d
Institute of Hydromechanics and Water Resources Management, ETH, 8093 Zuerich Hoenggerberg, Switzerland
Received 29 January 2003; revised 11 June 2004; accepted 15 June 2004

Abstract
The estimation of groundwater mixing ratios is a crucial task in groundwater research and management. The quantity as well
as the quality of groundwater resources are sensible to, e.g. the admixture river water, the interaction between different
groundwater layers and the spatial distribution of such mixing zones. In this paper, we present a procedure to calculate mixing
ratios and their uncertainties using a set of chemical and isotopic parameters. The advantages compared to estimations based on
only few parameters are better-constrained mixing ratios and the limitation of systematic errors due to the parameter choice.
The statistical consistence of the resulting mixing ratios is checked with a c2-test.
In particular in larger aquifers, where sampling resolution is often coarse, the composition of end members may vary
considerably from one borehole to another due to rock water interaction or varying recharge conditions. After selecting
sampling sites where no mixing occurs, the spatial distribution of the end member composition was calculated using Kriging.
This provides the best estimates of end member compositions and their uncertainties at any location of the area of investigation.
The method is, in principle, applicable for any number of end members. In this paper, the spatial distribution of the
interaction of two layers of the Continental Terminal aquifer (Niger, Africa) was investigated. Mixing between the deep CT2
and the shallow CT3 aquifer occurs where the separating layers are thinning out or due to improper sealing of some boreholes.
Radiogenic 4He concentrations, which differ significantly in the two aquifers, were used as independent parameter to
crosscheck the mixing calculations.
q 2004 Elsevier B.V. All rights reserved.

Keywords: Groundwater; Mixing ratio; Uncertainty; End-member; Tracer; Kriging; Niger

1. Introduction

The estimation of groundwater mixing ratios is an


important task in groundwater research and manage-
* Corresponding author. Tel.: C44 1483 682948; fax: C44 1483
689971. ment, e.g. river infiltration (Beyerle et al., 1999;
E-mail address: j.rueedi@surrey.ac.uk (J. Rueedi). Edmunds et al., 1992; Farah et al., 2000) or intrusion
0022-1694/$ - see front matter q 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhydrol.2004.06.044
2 J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14

Nomenclature
a correlation distance (m) w weight of parameter
C concentration x co-ordinate
h distance (m) g half of the covariance (semi variogram)
m mixing ratio of a parameter i h nugget effect
n total number of parameters s standard deviation
M weighed sum of deviations from the mean u difference between maximal variance and
P groundwater parameter nugget effect h
S sample

from leaky aquifers or brines (Balderer et al., 1996; all parameters and can therefore eliminate the possible
Cook and Herczeg, 2000). Mixing ratios of different bias due to the observer’s choice. To determine the
groundwater components can be calculated if the uncertainty of estimated mixing ratios—even in
mixed waters have distinct signatures, e.g. because of simple cases—the uncertainties of the concentrations
different chemical and/or isotopic evolution. In the of the two end-members and those of the measure-
above studies mixing ratios were calculated mainly ments have to be included into the final calculation.
based on estimated end-member concentrations of one In the first part of the paper a combination of
or two parameters only. This approach is based on a statistical methods will be presented which involves
subjective choice of the usefulness of a parameter. all measured parameters in the investigated area to
When the a priori information about the groundwater calculate mixing ratios.
system is reliable and/or when the chemical and In the second part of the paper these methods will
isotopic signatures of the two end-members are very be applied on a data set obtained from the multi-layer
well distinguishable this approach leads to reliable aquifer system of the Iullemmeden Basin (Niger,
results. However, the consistency with other ground- Africa) where the main objective was to correct
water parameters is not guaranteed and uncertainties observed signatures of environmental tracers (e.g.
dating tracers, noble gases) with respect to mixing and
are rarely mentioned even though both points are
to detect regions where the groundwaters from two
essential for a further use of the results.
neighbouring aquifers (CT3 and underlying CT2) mix
Under the following conditions the above approach
together. In our study area, all conditions preventing a
leads to a low accuracy of estimated mixing ratios:
successful application of the conventional approach to
† Sparse information about the groundwater system estimate mixing ratios (see above) are fulfilled.
(e.g. drilling profiles, bore hole equipment,
groundwater interaction)
† Heterogeneous database of chemical and isotopic 2. Methods
data (e.g. different parameters measured, different
precision of measurement) The approach to estimate mixing ratios and their
† Small differences between the signatures of the two uncertainties is based on the following procedure:
end-members
† High spatial variations of the end-members used to 1. Analysis of available data in terms of data quality,
estimate mixing ratios measurement techniques and precisions
2. Exclusion of non-conservative parameters
Under these conditions, it would be favourable to 3. Selection of samples representing the end-mem-
base the calculation of mixing ratios on all available bers based on all available information (geology,
parameters measured because the error of estimation hydrogeology and spatial distribution of
can be reduced and the result becomes consistent with hydrochemistry)
J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14 3

4. Estimation of end-member parameters and their arbitrary location (x0) of the area based on the spatial
uncertainties at each location correlation of the data set. Best-fitting curves are
5. Correction for other sources influencing the usually applied to approximate the spatial evolution of
chemical and/or isotopic groundwater signature the data co-variances, where different fitting functions
(e.g. local nitrate pollution) can be found in literature (e.g. de Marsily, 1986). The
6. Calculation of the overall mixing ratio of a sample spatial behaviour of the co-variances of our data was
using the calculated mixing ratios of each found to be closest to a spherical evolution with
groundwater parameter increasing distance (Eq. (1)):

     
3 h 1 h 3
2.1. Selection of end-members gðhÞ Z h C u K h! a (1)
2 a 2 a
In the first step, one has to select all unmixed
samples that will be used as the best estimate of the
end-members. This step always involves arbitrariness, gðhÞ Z h C u hR a
which is why it is the least reproducible step of the
presented approach. Therefore, the selection has to
where g(h) is the value of the semi-variogram in a
rely on a well-defined decision basis.
distance h apart of the origin x0Zx(hZ0), h is the
Since the presented method only accounts for
nugget effect, hCu is the maximum variance in far
linear mixing of parameters one has to select all
distance and a is the correlation distance (Fig. 1B).
groundwater parameters fulfilling this condition.
Note that hCu is equal to half of the overall variance
When mixing of two groundwaters leads to mineral
of the parameter.
dissolution or precipitation the parameters involved in
these reactions are not conservative anymore and
therefore have to be excluded from the following 2.3. Correction of contamination
steps.
When samples S with concentrations CS,P (of
2.2. Estimation of spatial end-member concentration parameter P) are punctually contaminated with
CP and uncertainty sP ‘unnatural’ substances (e.g. nitrate or organic sub-
stances), which affect the major chemistry or isotopic
When it is considered that the end-member groundwater signature, unnatural parameters (e.g.
concentrations of a parameter P are not known at NOK 3 ) have to be excluded and we applied an
the sample location x0 where mixing occurs, they have adequate correction for the naturally occurring
to be interpolated using unmixed samples. One parameters affected by contamination (e.g. Ca2C,
possibility to estimate an end-member concentration NaC) have to be applied. When contamination occurs
at location x0 is the calculation of the mean CP(x0) and over large areas (e.g. nitrate only by precipitation) the
the corresponding standard deviation sP(x0) of all parameter can and should be used to estimate mixing
samples in the study area known to be end-members. ratios. However, it does not necessarily improve the
This approach is recommended for small numbers of final result.
end-member samples and leads to the most conserva- Correcting contaminated samples can be done by
tive error estimation as spatial heterogeneities and correlating the concentrations of the contaminating
trends result in higher standard deviations of CP(x0). substance using other groundwater parameters. The
For large amounts of samples Kriging can be used affected parameters can then be corrected based on
to optimally interpolate the unknown end-member the linear regression. The standard deviation of the
concentration at location x0. As it is described in the residuals (compared to the regression line) has to be
literature (Cressie, 1991; Matheron, 1963;1965; added to the measurement errors sS,P to include the
1970), Kriging enables an estimation of the concen- uncertainties of the regression analysis in the
tration C(x0) and its standard deviation s(x0) at each subsequent calculations.
4 J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14

Fig. 1. (A) Principle of point estimation problem. x1,.,x5 denote the co-ordinates where the parameter C is sampled, and x0 is the location where
C has to be estimated. The distance interval from h to hCDh indicates the area in which the mean of all co-variances gi,j between all locations xi
and xj is calculated. (B) Example of semi-variogram using the d18O ratios of CT3 aquifer as database (see Fig. 4 for comparison). The squares
indicate calculated mean variances assuming an averaging distance Dh of 2 km. The best fitting spherical function, indicated with a solid line, is
plotted as a function of distance x.

 s with
2.4. Calculation of the total mixing ratio mG hypothesis) and due to the fact that all mP are
two end-members normally distributed, m can be calculated by aver-
aging the ratios of all n parameters.
Based on the measured parameter P of the sample S Pn
(CS,P GsS,P) taken at the location x 0 and the PZ1 mP wP
m Z P n (4)
estimations of both end-members C1,PGs1,P and PZ1 wP
C2,PGs2,P (at the location x0) we can calculate the
mixing ratios mP and the standard deviation sP 1
wP Z (5)
obtained from each parameter as follows: s2P
CS;P K C2;P As a measure of the usefulness of a parameter the
mP Z (2)
C1;P K C2;P weights wP are chosen to be inversely proportional to
the variance of each single mixing ratio (Eq. (5)). The

1 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sP Z s2S;P ðC1;P K C2;P Þ2 C s21;P ðCS;P K C2;P Þ2 C s22;P ðCS;P K C1;P Þ2 (3)
ðC1;P K C2;P Þ2
Please note that hydrochemical parameters are inverse of the sum of all weights leads to the
sometimes log-normally, but usually normally dis- uncertainty of the estimated mixing ratio m (e.g.
tributed. With log-normally distributed end-members Brandt, 1992):
the above calculation leads to log-normally distrib-
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uted results (mp) and the following procedure to
1 1
calculate average mixing ratios had to be adapted. s Z Pn Z Pn (6)
PZ1 sP PZ1 wP
K2
When only few data are available a normal distri-
bution is usually the best assumption anyway. For
larger amounts of data, C and s are obtained from To validate whether each parameter represents the
Kriging and therefore are assumed to be normally  s a c2-test can be applied. It
same mixing ratiomG
distributed (Cressie, 1991; Matheron, 1963; 1965; can be shown that for normally distributed mP
1970). X
n
Since each conservative parameter must, by MZ  2
wP ðmP K mÞ (7)
definition, reflect the same mixing ratio m (null PZ1
J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14 5

follows a c2-distribution with nK1 degrees of free- 3.1. The study area
dom (e.g. Brandt, 1992). Note that the degree of
freedom will be reduced when two (or more) The Iullemmeden Basin is situated in the South-
parameters are well correlated and therefore, not West of Niger (Africa) (Fig. 2). The sediment layers
independent anymore (e.g. often observed good can be divided into the upper continental sediments of
correlation of NaC and ClK). When the hypothesis the Continental Terminal (CT) and the lower
is fulfilled, s in Eq. (6) is the final uncertainty of the continental sediments of the Continental Intercalaire
mixing ratio. Otherwise, Rosenfeld et al. (1967) (CI) separated by massive layers composed of silts
proposed that all standard deviations sP can be up- and marls from the last marine transgression (Dodo,
scaled linearly in order to fulfil the condition of the 1992; Greigert, 1966). The CT formation can be
c2-test. This up-scaling leaves m unchanged but leads divided into three water baring layers, namely the
to a larger uncertainty of the total mixing ratio CT3, CT2 and CT1 aquifer (from top to bottom)
resulting in a sufficient c2: separated by clay and silt layers (Dodo, 1992).
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi According to Greigert (1966) and Greigert and
M Pougnet (1967) the aquifers can be interpreted well
s Z P
ðn K 1Þ nPZ1 wP within the spatial context but they are difficult to
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi distinguish in detail. In the western part (west of 38E)
Pn
PZ1 wP ðmP K mÞ  2 of the study area, the CT2 and CT1 aquifers contain
Z P (8) water from the up-welling CI aquifer with a different
ðn K 1Þ nPZ1 wP
chemical and isotopic signature (Ousseini et al., 1992;
It has to be mentioned that this approach is not Le Gal La Salle, 1995). We therefore selected the
founded on a clear mathematical basis (Rosenfeld eastern part to test our method where no indications
et al., 1967). Therefore, it has to be used with care were found of up-welling of CI groundwater (Fig. 2).
when systematic errors are possible (e.g. non-linear We only used the top two water bodies of the CT
mixing) because this would mean that the null formation (CT3, CT2) to show the applicability of the
hypothesis is violated. statistical method because they provide the largest
The described method leads to an estimation of the database. Since the chemical and isotopic signatures
mixing ratio mG  s of each sample based on all of these two waters are only slightly different and vary
conservative parameters, where m and s only depend considerably in space no clearly preferable parameter
on the available data set and not on a subjective choice exists for the calculation of mixing ratios. Therefore, a
of groundwater parameters. combined use of all measured parameters is clearly
preferable to obtain the lowest possible uncertainty of
the final estimation of mixing ratios.
3. Results and discussion
3.2. Selection of end-members
During the first steps of data interpretation and
their comparison with new data measured during our In former studies within the investigated area the
research project two major problems were encoun- groundwater samples were interpreted based on the
tered. Firstly, large heterogeneities were observed structural information obtained from the drilling
when chemical and isotopic signatures were inter- profiles only (Dodo, 1992; Le Gal La Salle et al.,
preted based on available borehole information. 1995; Ousseini et al., 1992). The resulting picture of
Further, intermediate values were observed in both the distribution of chemical and isotopic signatures
chemical and isotopic signatures that did not make was therefore, characterised by very large differences,
sense in an overall context. However, no information sometimes over short distances, and the studies
was available whether or not these samples possibly concluded that the different aquifers could not be
contain a mixture of two groundwaters. These two distinguished in terms of hydrochemistry. Since
points lead to the development and application of the drilling profiles are difficult to interpret in detail it is
method explained above. possible that some drilling profiles were wrongly
6 J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14

Fig. 2. Map of the study area in the south-west of Niger (Africa). Locations of all CT3 (squares) and CT2 (triangles) wells in the project area
used to the calculate end-members are shown in the right plot. Main cities are indicated with stars.

18 2
interpreted. Therefore, we choose a new approach, and NOK3 ) and the stable isotope ratios d O and d H.
assuming that major differences only exist between NO3 is only used to correct for human contamination.
K

the aquifers whereas each aquifer is more or less The determination of end-members, which is
homogeneous in terms of hydrochemical and isotopic principally based on the drilling profiles, was
signature. This assumption is justified by the fact that performed in the following four steps:
the basin shows similar geologic features over large
distances. The only geological heterogeneities are 1. As we do not know the accuracy of most older data
documented in the southwest of the CT2 Aquifer (in we first checked the ion balance and dismissed all
south-western corner of the project area) where samples with a disagreement bigger that 10%.
limestone lenses are observed (Dodo, 1992). Since the statistic method presented takes analyti-
The samples used in this paper were partly cal errors into account we assumed a fixed relative
measured during this study but older samples from measurement error of 5% because we do not know
former projects were introduced to provide a broader the analytical error. For the stable isotope
basis. These data are held by the Ministry of Water measurements we dismissed measurements that
Resources (Niamey, Niger) and they are kindly were far off the local meteoric water line. A
provided by Abdelkader Dodo. The combined data- comparison of earlier and newly measured stable
base used in this paper, consisting of more than 300 isotope data showed no significant difference
measurements, is listed in Rueedi (2002). The indicating that no bias is introduced due to the
estimation of end-members and the subsequent measurement technique or laboratory. A big
calculation of mixing ratios are based on pH and advantage of the statistical approach presented is
conductivity values measured in the field, the major the fact that different uncertainties of measure-
ions (NaC, KC, Ca2C, Mg2C, ClK, HCOK 3 , SO4
2K
ments from different years, sources and labs can be
J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14 7

introduced into the statistical procedure and profiles, that they do not contain 3H or CFCs
propagated to the uncertainty of the mixing ratio. (Rueedi, 2002). For CT3 end-members the cation–
2. We had to remove samples that do not show either cation and the stable isotope plots were found to be
CT2 or CT3 groundwater signatures (e.g. mixture the most useful to distinguish the CT3 groundwater
with CI groundwater). Sulphur concentrations signature from the other aquifers (Fig. 3). The plots
showed to separate CI from CT groundwaters clearly show a systematic behaviour of the data but
because increased sulphur contents only appear in the overlapping of the circles in Fig. 3 indicates that
samples from the CI aquifer (at least in the study the distinction is not unique. However, when the
area). samples are analysed in an overall context (using
3. Contaminated samples containing more than hydrogeology and spatial distribution of hydro-
20 mg/l nitrate were not used as end-member chemistry) it was found that some of the intermedi-
because they do not represent natural conditions ate samples are mixtures or originate from another
(57 samples). The choice of the cut-off level is aquifer than proposed by the drilling profiles.
based on observations made in the western region
where the natural background level of nitrate is This entire selection and elimination procedure lead to
supposed to be about 20 mg/l (Elbaz-Poulichet 111 CT3 end-members from the originally 218
et al., 2002). supposed CT3 samples. Fifty-seven samples were
4. Since the method is used to determine mixing ratios eliminated due to a nitrate concentration above
between CT2 and CT3 groundwater we had to 20 mg/l, six samples contained increased sulphur
temporarily remove all supposed mixtures from the amounts and all others (44 samples) were temporarily
database because they will be analysed using eliminated and analysed as mixtures. For the CT2
unmixed samples. The primary selection condition aquifer, 10 from originally 13 samples were selected
for the CT2 end-member was, besides the drilling as end-members (Fig. 2).

Fig. 3. (A) Mass ratios of single and double charged cations of all available samples from the four aquifers displayed in log-scale. Squares
indicate samples from CT3 aquifer, triangles those from the CT2 aquifer, diamonds those from the CT1 aquifer and circles show samples from
the CI aquifer. The classification of the samples originates from the official documents. Areas where the samples of each of the four aquifers are
preferably found are indicated with ellipses. Open symbols show samples from older projects and full symbols are samples taken during this
project. (B) Stable isotope plot of all samples within the Basin. The straight line indicates the global meteoric water (GMWL) line from (Craig,
1961)).
8 J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14

3.3. Estimation of spatial end-member concentration Krigging showed no significant anisotropy. Prefer-
and uncertainty ential directions would be expected when chemical
evolution would be present. Therefore, the lack of
The concentrations of the CT3 end-members was preferential directions is an indication for well-
calculated based on Kriging because there are 111 distributed groundwater recharge and confirms the
samples available. The advantage of using Kriging is assumption of homogeneous geologic features (Sec-
that the uncertainties of estimation are usually smaller tion 3.2) in the study area.
than the standard deviation of all end-member The three fitting parameters of the spherical
samples. The semi-variograms of the CT3 end- function (Eq. (1)) needed for the interpolation
members are displayed in Fig. 4. We used ‘normal’ procedure are listed in Table 1. Fig. 4 shows that the
Kriging because different preferential directions of stable isotopes and the pH values underlie a clear

Fig. 4. Semi-variograms of all 11 parameters of CT3 samples selected as end-members. Squares show half of the mean variances g(h) (Eq. (7))
averaged within distance intervals Dh of 2 km. The solid line shows the best-fitting spherical function (Eq. (8)).
J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14 9

Table 1
Estimated parameters for spherical fit to variance analysis of CT3 end-members

g0 d18O d2H cond. NaC KC (mg/l) Mg2C Ca2C ClK (mg/l) SO2-
4 (mg/l) alc. (mg/l)
(‰) (‰) (mS/cm) (mg/l) (mg/l) (mg/l)
h 0.05 3.2 800 2 2.4 0.3 11 2.5 1.2 270
u 0.2 6 – 3.5 – 1.3 – 12 – –
A (km) 90 30 – 10 – 10 – 10 – –

spatial correlation. For these parameters Kriging leads It has to be mentioned that, in the western part of
to an improvement of the estimated uncertainty sP at a the basin, considerable annual fluctuations of con-
given location because Kriging favours samples in the ductivity and groundwater chemistry were found
vicinity of the wanted location x0. The spatial (Elbaz-Poulichet et al., 2002). However, in our
correlation of d18O and d2H could originate in the research area, the few repeated measurements taken
characteristically small-scale rain patterns in this area in January and May do not show significant
(Le Barbe et al., 1997; Lebel et al., 1997). However, differences in hydrochemistry but there is not enough
no reason can be named why the spatial behaviour of information available to dismiss fluctuations due to
the fitted correlation functions are different for these quick reaction-times on recharge and evaporation.
two parameters even though the original data clouds However, we expect smaller fluctuations in our area
look fairly similar. It is possible that only a few points due to the considerably deeper water tables.
negatively influence the automated fitting process.
The correlation distances of the chemical par- 3.4. Correction of contamination
ameters are small (NaC, Mg2C and ClK) or even zero
(all others). A negligible correlation distance indicates Since nitrate was only detected in young tritiogenic
that these parameters are spatially independent or (TritiumO0) CT3 groundwaters nitrate is a reliable
randomly distributed. parameter to detect samples containing CT3 ground-
In case of the CT2 end-members the calculation of water (Rueedi, 2002). However, at some sampling
variograms would be rather questionable, both sites (10 samples bigger than 100 mg/l) the nitrate
because of the small numbers of CT2 end-members concentrations are considerably higher than the
(10 end-members) and because they are all located in observed natural background of maximally 20 mg/l
a E–W line (Fig. 2), whereas Kriging is not supposed (Elbaz-Poulichet et al., 2002). Such local contami-
to be used for extrapolation. In fact, the small number nation sources most probably originate from sewage
of CT2 end-member sites would lead to Kriging outflows near bigger cities and/or from open watering
uncertainties that are equal to the overall statistical places. The signature introduced by these sources can
variance of all CT2 end-member samples. Therefore, be seen as a third end-member and therefore should be
we assumed a normal distribution and used the excluded from the parameters used to calculate binary
statistical means and standard deviations for the mixtures (see Section 2.3).
subsequent mixture calculation. The assumption of a All chemical parameters were checked for a
distribution different to the normal distribution could correlation with nitrate but only sodium, magnesium,
not be justified with only 10 samples (Section 2.4). calcium and consequently the conductivity were
The result are shown in Table 2. found to be significantly correlated (Fig. 5). A similar

Table 2
Statistical mean and standard deviations for CT2 end-members

d18O (‰) d2H (‰) Cond. NaC KC (mg/l) Mg2C Ca2C ClK (mg/l) SO2K
4 alc. (mg/l)
(mS/cm) (mg/l) (mg/l) (mg/l) (mg/l)
K7.1G0.5 K49.9G4.2 151G14 9.3G4.2 5.3G0.8 8.1G1.8 9.8G3.1 1.3G0.4 3.4G0.9 92G12
10 J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14

Fig. 5. Correlation of conductivity, sodium (NaC), magnesium (Mg2C) and calcium (Ca2C) concentrations with nitrate (NOK 3 ) contents in
groundwater samples of the CT3 aquifer. The straight lines indicate the regression lines including the sample containing more than 200 mg/l
nitrate (upper formula in top left of each plot). The dotted straight lines indicate the regressions without this sample (lower formula in top left of
each plot). The dotted lines show the standard deviations of the measurements compared to the regression line of all samples (note that it is not
the standard deviation of the regression line).

observation has been made in the western part of the 3.5. Calculation of the total mixing ratio of a sample
basin (Elbaz-Poulichet et al., 2002).
When all samples are included into the linear All chemical parameters measured in CT2 and CT3
regression one point with more than 200 mg/l of nitrate groundwaters are under-saturated with respect to any
dominates the result. Therefore, the dotted regression mineral phase except quartz. This guarantees the ideal
line in Fig. 5 shows the outcome without this point. The mixing behaviour of all 9 chemical species and 2
observed correlation most probably represents the isotopic parameters used to calculate mixing ratios. It
average chemical composition of the contamination was found that, in this project, the 11 parameters are
source. However, the sewage composition might vary quite poorly correlated (low R2 in Table 3). The poor
considerably leading to the observed scattering of the correlation of the stable isotopes is due to an
data. As it was proposed in Section 2.3, the correlation evaporation signature. Table 3 shows as well the
will be used to correct the influenced parameters results of the t-test applied on the correlations where
leading to a new sample concentration Cs and its (C) indicates a significant correlation and (K)
statistical uncertainty ss used in Eqs. (2) and (3). indicates insignificant correlation. Therefore,
J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14 11

Table 3
Correlation coefficients R2 between used parameters and significance of correlation resulting from t-test

Cond. ClK SO2K


4 HCOK
3 KC NaC Mg2C Ca2C d18O d2H
Cond. 1 0.109 0.001 0.185 0.062 0.355 0.303 0.483 0.087 0.023
ClK C 1 0.033 0.004 0.127 0.121 0.028 0.038 0.000 0.006
SO2K
4 – – 1 0.007 0.118 0.001 0.033 0.002 0.059 0.050
HCOK 3 C K K 1 0.079 0.015 0.370 0.619 0.001 0.000
K C
C K K K 1 0.004 0.063 0.094 0.012 0.018
NaC C K K K K 1 0.058 0.083 0.320 0.145
Mg2C C K K C K K 1 0.424 0.002 0.013
Ca2C C K K C K K C 1 0.002 0.000
d18O K K K K K K K K 1 0.343
d2H K K K K K K K K C 1

(C) indicates significant correlation and (K) indicates that there is no correlation between the two parameters.

the degree of freedom for the c2-test is equal to 5


when all parameters were measured.
The statistical procedure was applied on all
samples, which were not selected as end-members
(107 CT3 samples and 3 CT2 samples with and without
nitrate). Fig. 6 shows, as an example, the mixing ratios
of two samples calculated with each parameter
compared with the calculated total mixing ratio
including all parameters. The top plot shows a case
where the c2-test was fulfilled. The much smaller error
of the total mixing ratio clearly supports the argument
that the applied procedure leads to a smaller uncer-
tainty of estimation than the use of one parameter only.
The lower plot, on the other hand, shows the total ratio
(and uncertainty) before and after the c2-test failed.
It can be seen that the final uncertainty is actually

Fig. 6. Two examples of mixed samples and the corresponding


mixing ratios. Calculated mixing ratios mP and uncertainty sP of
each parameter are indicated with squares and error bars. The plots
show the final mixing ratios ðmG sÞ indicated with the hatched areas
(calculated based on all single ratios). The top plot shows a case Fig. 7. Amounts of radiogenic 4He (4Herad) in groundwater samples
where the c2 was fulfilled. The bottom plot shows a case where the plotted against the calculated mixing ratios of CT2 in CT3
test failed and the initially calculated ratio (inner hatched area) had groundwater. Typical amounts of 4Herad measured in the two end-
to be corrected to obtain the final mixing ratio (outer hatched area). members are indicated with hatched areas.
12 J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14

bigger than the ones of the stable isotopes but it is distinguishable amounts of 4Herad (see grey bars in
based on all parameters and therefore, independent of Fig. 7). Although the estimated mixing ratio seems to
the users choice of a parameter. be too large in case of one sample, the overall good
To show the applicability of the method on the data agreement between 4Herad and the calculated mixing
presented, Fig. 7 displays the comparison of the ratio shows that the presented method is applicable
calculated ratios with the radiogenic 4He (4Herad) even for an unfavourable database.
contents of the few samples measured on pumped The relative errors of mixing ratio estimations are
boreholes. The 4Herad is defined as the difference between 8 and 20% when all 11 parameters are used.
between the measured 4He concentration (4Hemeas) However, at sites where only few parameters were
and the atmospheric 4He component given by the measured the relative errors can reach up to 100%.
equilibrium concentration (4Heeq) plus the excess air Depending on the location of mixing ratio estimation
component (4Heexc). The calculation of 4Herad was the results mainly depend on stable isotope ratios,
done using the approach of Aeschbach-Hertig et al. ClK, KC, Mg2C and HCOK concentrations, whereas
(2000). Data can be found in Beyerle et al. (2003) and the other parameters (Na C, Ca 2C , SO 2K 4 and
Rueedi (2002). 4Herad was chosen to check the conductivity) do hardly constrain the calculations
results because the two groundwaters contain clearly because the variations of their end-members are large

Fig. 8. The map shows a contour plot of calculated mixing ratios (CT2 in CT3 groundwater) of all samples within the project area which were
taken from open wells. Diamonds indicate CT3 end-members and dots indicate CT2 end-members. The solid lines delineate the boundaries of
the CT2 and CT3 aquifers after Greigert and Pougnet (1967).
J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14 13

(Fig. 6). However, which parameter dominates the of the ‘best’ parameter but the result is independent of
mixing ratio calculation is determined by the the choice of the ‘best’ parameter. It is to be noted
statistical method applied and not by the researcher. that, even though the statistical procedure is clearly
Fig. 8 shows all calculated mixing ratios of wells in defined and consistent with all data, the results still
the project area. The reason is that wells only capture depend on the initial choice of the end-members. This
the top few meters of the CT3 aquifer whereas first step always involves arbitrariness and therefore it
boreholes may reach down to the CT2 layer. There is clearly the limiting factor for the reproducibility of
were a few supposed CT3 boreholes detected between the entire approach. However, if the geologic
Dogon Doutchi and Nassarawa where mixing ratios information is scarce, as in our study area, the
were between 0.5 and 0.9. These boreholes most presented approach clearly helps to understand and
likely withdraw water from both aquifers. The depths interpret the measurements better.
of these holes support this observation. The calculated The interpretation of calculated mixing ratios has
mixing ratios of these boreholes can and will be used to be undertaken with great care. For example,
to correct other measurements such as dating mixtures possibly originating from improper sealing
tracers. The map in Fig. 8 shows that mixtures of the borehole do not allow conclusions about the
can be observed mainly near the northern boundary groundwater system but they are essential to interpret
of the CT3 aquifer (thick full lines), where the CT3 is additional parameters (e.g. groundwater dating tracer)
only thin, and near the southern boundary of the area that are possibly only measured at a few selected
where CT3 and CT2 are supposed to be well separated
wells. For the interpretation of non-linear parameters
(thick dashed line) (Greigert and Pougnet, 1967).
(e.g. radioactive tracer, chemical reactions) it is
However, mixtures are detected further up-stream
essential to know if the groundwater sample is
(NNE) of this area indicating that interaction between
unmixed or a mixture of two groundwaters with
the two aquifers occurs further up-stream than
different signatures and it is important to know the
expected. Whether this boundary is the real limit of
uncertainties of estimation.
the confining layer or developed due to the increasing
Furthermore, the results have shown to provide
use of CT2 or CT3 groundwater for drinking water
supply remains open. Additional consultation of the more information about the groundwater system. The
hydraulic records of both aquifers unfortunately did spatial distribution of calculated mixing ratios indi-
not help because in the observed regions the gradients cates that the two groundwaters from the CT2 and the
between the two aquifers are of the same order of CT3 aquifer are hydraulically connected slightly
magnitude as the uncertainties of the levels (Dodo, further up-stream than assumed by former studies.
1992; Ousseini et al., 1992). However, the results This information is essential for hydrodynamic
enable the selection of regions where further, more groundwater models and it reveals regions where
detailed investigations should focus on to determine further investigations should focus on.
anthropogenic consequences on the local groundwater In general, the knowledge of the uncertainty of a
resources. mixture is essential for any further use of the mixing
ratios, e.g. correction of other measurements like
hydraulic heads and conclusions on water resources.
4. Conclusions

The presented approach to estimate groundwater


mixing ratios proved to be useful and allowed to Acknowledgements
calculate mixing ratios and their uncertainties.
The major improvements to former methods are that This study was supported by the Swiss National
the results are consistent with all groundwater Science Foundation (project no. 20-53629.98).
parameters and that the errors of estimation can be We would like to thank Abdelkader Dodo of the
drastically reduced in some cases. In the less optimal Ministry of Groundwater Resources for his support
cases uncertainties are slightly higher than the ones with additional groundwater data of the study area.
14 J. Rueedi et al. / Journal of Hydrology 305 (2005) 1–14

References Farah, E.A., Mustafa, E.M.A., Kumai, H., 2000. Sources of


groundwater recharge at the confluence of the Niles, Sudan.
Aeschbach-Hertig, W., et al., 2000. Palaeotemperature reconstruc- Environmental Geology 39, 667–672.
tion from noble gases in ground water taking into account Greigert, J., 1966. Description des formations cretacées et
equilibration with entrapped air. Nature 405, 1040–1044. tertiaires du Bassin de Jullemeden (Afrique occidentale).
Balderer, W., Synal H.A., Jull, A.J.T., 1996, Use of chlorine-36 as tracer BGRM, Paris.
for the evolution of waters in geothermal and tectonic active areas in Greigert, J., Pougnet, R., 1967. Notice explicative sur la carte
western Turkey. In: Abstracts of the Seventh international conference geologique de la République de Niger à l’échelle 1:2,000,000.
on Accelerator mass spectrometry. Radiocarbon, 38: 4–5. BGRM, Paris.
Beyerle, U., et al., 1999. Infiltration of river water to a shallow Le Gal La Salle, C., et al., 1995. Old Groundwater Circulation in the
aquifer investigated with 3H/3He, noble gases and CFCs. Iullemeden Basin (Niger); Preliminary Results of an Isotopic
Journal of Hydrology 220, 169–185. Study, Application of Tracers in Arid Zone Hydrology. IAEA,
Beyerle, U., et al., 2003. Evidence for periods of wetter and cooler Vienna pp. 129–139.
climate in the Sahel between 6 and 40 kyr BP derived from Le Barbe, L., et al., 1997. Rainfall climatology of the HAPEX-Sahel
groundwater. Geophysical Research Letters 30, 1173–1177. region during the years 1950–1990. Journal of Hydrology
Brandt, S., 1992. Datenanalyse mit statistischen Methoden und 188–189, 43–73.
Computerprogrammen. (Chapter 9). Wissenschaftsverlag, Man- Lebel, T., et al., 1997. Rainfall monitoring during HAPEX-Sahel; 2,
nheim p. 651. Point and areal estimation at the event and seasonal scales.
Cook, P.G., Herczeg, A.L., 2000. Environmental Tracers in Subsur- Journal of Hydrology 188-189, 97–122.
face Hydrology. Kluwer Academic Publishers, Boston p. 529. Matheron, G., 1963. Principles of geostatistcs. Economic Geology
Craig, H., 1961. Isotopic variations in meteoric waters. Science 133, 58, 1246–1266.
1702–1703. Matheron, G., 1965. Les variables regionalisees et leur estimation.
Cressie, N.A.C., 1991. Statistics for Spatial Data. Wiley, New York Masson, Paris.
p. 900.
Matheron, G., 1970. La théorie des variables régionalisées et ses
de Marsily, G., 1986. Quantitative Hydrogeology Groundwater
applications. Centre de Morphologie Mathématique, Fontaine-
Hydrology for Engineers. Academic Press, San Diego pp. 440.
bleau p. 212.
Dodo, A., 1992. Etude des Circulations Profondes dans le Grand
Ousseini, B., Bouzelboudjen, M., Dodo, A., Zwahlen, F., 1992.
Bassin Sédimentaire du Niger. Université de Neuchatel,
Circulations profondes des eaux du continental terminal (Niger),
Neuchatel p. 101.
vol. 11. CHYN, Neuchatel.
Edmunds, W.M., Darling, W.G., Kinniburgh, D.G., Kotoub, S.,
Mahgoub, S., 1992. Sources of recharge at Abu Delaig, Sudan. Rosenfeld, A.H., et al., 1967. Data on particles and resonant states.
Journal of Hydrology 131, 1–24. Reviews of Modern Physics 39, 1–51.
Elbaz-Poulichet, F., Favreau, G., Leduc, C., Eidel, J.L., 2002. Major Rueedi, J., 2002. Estimating Groundwater Recharge and
ion chemistry of groundwaters in the Continental Terminal Dynamics in Arid Regions using Environmental Tracers:
water table of southwestern Niger (Africa). Applied Geochem- Application on Jullemeden Basin (Niger). Universität Bern,
istry 17, 1343–1349. Bern p. 173.

Vous aimerez peut-être aussi