Journal of Geochemical Exploration 148 (2015) 196–205

Contents lists available at ScienceDirect

Journal of Geochemical Exploration

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

Determination of natural backgrounds and thresholds of nitrate in South

Korean groundwater using model-based statistical approaches
Kyoung-Ho Kim a, Seong-Taek Yun a,b,⁎, Hyun-Koo Kim c, Ji-Wook Kim d
a
Department of Earth and Environmental Sciences, Korea University, Seoul 136-701, South Korea
b
KU-KIST Green School, Korea University, Seoul 136-701, South Korea
c
National Institute of Environmental Research (NIER), Incheon 404-170, South Korea
d
Korea Water Resources Corporation, Daejeon 306-711, South Korea

a r t i c l e i n f o a b s t r a c t

Article history: Increased nitrate loading of groundwater has emerged as a major environmental problem in many countries, in-
Received 18 January 2014 cluding South Korea. This study aims to evaluate the nitrate levels of South Korean groundwater on a regional
Accepted 2 October 2014 (national) scale and specifically to demonstrate the procedure to better estimate the natural background level
Available online 12 October 2014
(NBL) and threshold of nitrate as the basis of groundwater management. For this work, nitrate data of groundwa-
ter (n = 8510) in two major hydrogeologic units (alluvium and bedrock) were collected from the National
Keywords:
Background level and threshold of nitrate
Groundwater Monitoring Network (NGMN) of South Korea. Four supplementary datasets (n = 1074) were
Anthropogenic polluted level also used to test the rationality of estimated thresholds by comparing them with NGMN datasets. Compared
Finite normal (Gaussian) mixture model with the data reported in many countries, the nitrate concentrations in NGMN groundwater in 2009 are high,
National Groundwater Monitoring Network of with median values of 12.2 and 8.7 mg/L, respectively, for alluvial groundwater and bedrock groundwater. The
South Korea nitrate levels of South Korean groundwater seem to have been historically steady at these high levels between
1997 and 2009, suggesting widespread diffusive contamination since the 1980s. The NBLs and anthropogenic
polluted levels (APLs) of nitrate on a regional (national) scale are statistically established by the model-based ap-
proach using a finite normal (Gaussian) two-component mixture model, because (1) the sample size (frequency)
of the natural background group is much smaller than that of the polluted group, as a result of widespread nitrate
contamination, and (2) nitrate concentrations are more or less affected by natural attenuation processes. Accord-
ingly, thresholds of nitrate (as the concentration level indicating groundwater pollution) are selected as the
lower limits (i.e., 10th percentile) of the polluted group, which are 3.0 and 5.5 mg/L NO− 3 , respectively, for bed-
rock groundwater and alluvial groundwater. This study provides a practical guideline for national groundwater
management, based on a heuristic procedure to statistically determine the NBLs and thresholds in the case of
groundwater systems with pervasive contamination. Compared with the other classical methods to estimate
NBLs, the model-based approach using a finite normal-mixture model can be more effective to reasonably sepa-
rate the polluted samples from a regional (or national) dataset.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction 1993). As a common constituent in groundwater, nitrate is the most im-

Approximately 20% of the global water use is groundwater usage, and
this share is rapidly increasing (WMO, 1997; Arnell, 1999). However,
groundwater quality in many aquifers worldwide has been deteriorating
by increased human impact over the past few decades. In particular,
groundwater pollution is highly associated with diffusive (nonpoint)
sources in relation to agricultural activities in which many types of inor-
ganic and organic fertilizers are used (Burkart and Stoner, 2008; Burow
et al., 2010; Scanlon et al., 2007; Sebilo et al., 2013). Thus, nitrate became
the most ubiquitous pollutant in groundwater, frequently threatening
public water supply sources and human health (Spalding and Exner,
⁎ Corresponding author. Tel.: +82 2 32903176; fax: +82 2 32903189.
E-mail address: styun@korea.ac.kr (S.-T. Yun).
portant indicator of groundwater quality status and, therefore, a practical
need exists to precisely assess the nitrate levels in aquifers for better man-
agement and regulation of groundwater quality.
In this context, it is essential to determine both the natural back-
ground level (NBL) and the threshold value (in general, the upper
limit or maximum of NBL) of nitrate to differentiate between natural
controls (i.e., geogenic, biological, and atmospheric processes) and an-
thropogenic impacts on groundwater quality (Corniello and Ducci,
2014; Limbrick, 2003; Panno et al., 2006; Preziosi et al., 2014). The
threshold is commonly used as a practical reference value for determin-
ing the “good” status of groundwater quality and also serves as the level
for an early warning of groundwater pollution, because of the limita-
tions of the existing drinking water standard (DWS) (Edmunds et al.,
2003; Langmuir, 1997; Reimann and Garrett, 2005). In recent years,

http://dx.doi.org/10.1016/j.gexplo.2014.10.001
0375-6742/© 2014 Elsevier B.V. All rights reserved.
 K.-H. Kim et al. / Journal of Geochemical Exploration 148 (2015) 196–205
EU member states have established threshold values (as the regulatory
levels) corresponding to the NBLs of a number of constituents in many
aquifers (Edmunds and Shand, 2008; Hinsby et al., 2008; Müller et al.,
2006; Preziosi et al., 2010), based on scientific principles established
through research projects such as BASELINE and BRIDGE. These studies
also used the NBL of nitrate as b10 mg/L NO3 for a simplified separation
approach called preselection.
Many previous studies have published the NBLs of groundwater ni-
trate based on a number of approaches that were reviewed by Panno
et al. (2006). The reported values are low (b 5 mg/L NO3) because natural
or geologic sources of nitrate are not abundant and common (Böhlke,
2002; Halberg and Keedny, 1993). Even so, the NBL of nitrate from
datasets is still difficult to accurately define because of its ubiquitous oc-
currence that has resulted from the widespread and/or long-term con-
tamination from diffusive sources (Panno et al., 2006; Reimann and
Garrett, 2005). Furthermore, the concentrations of nitrate in groundwater
can also depend on redox conditions because anthropogenic nitrate can
be naturally attenuated by biogeochemical transformations, such as deni-
trification (Appelo and Postma, 1999; Langmuir, 1997; Postma et al.,
1991). Therefore, a good understanding of biogeochemical processes in
the investigated groundwater system is needed to precisely assess the
NBL of nitrate. For these reasons, statistical methods, including a model-
based approach, were recently used to determine the NBL of nitrate.
Groundwater contamination in South Korea has been an important
environmental concern since the 1980s because of rapid industrialization
and urbanization. Especially, severe nitrate pollution of groundwater has
been reported in many localities with active agricultural practices
(e.g., Chae et al., 2004, 2013; Choi et al., 2007, 2014; Joo et al., 2009;
Koh et al., 2010). In South Korea, anthropogenic nitrogen inputs from N-
fertilizers and manure are much higher than in other countries. The use
of synthetic N–P–K-fertilizers in South Korea is the fifth highest among
the OECD countries (OECD, 2004). The average application rate of N-
fertilizers amounts to 224 kg/ha per year (MOAF, 2001), compared with
128 kg/ha per year in England (Petry et al., 2002) and 27.5 kg/ha per
year in the United States (Nolan, 2001). Moreover, the nitrogen efficiency
(i.e., balance between inputs and outputs) in the agricultural soils of
South Korea is the lowest among the OECD countries (OECD, 2008).
These situations indicate that groundwater of South Korea can be highly
vulnerable to nitrate pollution without efficient and sustainable measures
for nutrient management in agricultural practices. Accordingly, the Kore-
an government recently began to assess the status of nitrate pollution to
establish measures for effective groundwater quality management.
As an initial step for groundwater management in South Korea, the
current study aims to evaluate the regional status of groundwater quality
and to establish the NBL and threshold of groundwater nitrate as the
guidelines for management. For these purposes, large datasets collected
from the National Groundwater Monitoring Network (NGMN) of South
Korea are evaluated. In addition, nitrate datasets obtained from four dif-
ferent hydrochemical surveys are also used to compare with NGMN
data. To set the NBL and threshold, the statistical distribution of nitrate
concentrations is divided into two groups (natural and anthropogenic)
by the model-based clustering approach using a finite normal (Gaussian)
mixture model. This study shows the application of a heuristic statistical
procedure to determine the NBL and threshold of ubiquitous and wide-
spread aquatic contaminants, such as nitrate. The proposed method can
be successfully used for the groundwater systems in which the influence
of anthropogenic contamination is prevailing or overwhelming natural
processes.
2. Materials and methods
2.1. Usage and hydrogeology of groundwater in South Korea
In South Korea, groundwater is an important resource that accounts
for approximately 11% (about 3.7 billion tons per year) of the total
water use. The groundwater is largely used for domestic (49%) and
197
agricultural (45%) purposes and comes from two types of aquifers: shal-
low alluvial aquifer and relatively deep bedrock aquifer (MOCT, 2007).
The alluvial aquifers, consisting mainly of sand with subordinate silt or
gravel, have thicknesses of 10–50 m and a total areal extent of approx-
imately 27,000 km2, and commonly occur along rivers and streams. The
transmissivity (T) and storage coefficient (S) values of alluvial aquifers
are 50–2000 m2/day and 0.1–0.01, respectively. In rural areas, alluvial
aquifers are extensively used for domestic and agricultural water sup-
plies and have potential yields ranging from 30 to 800 m3/day/well.
Bedrock aquifers represent the typical form of groundwater occur-
rence in South Korea and are developed in or along weathered zones,
faults, fractures, joints, and lithologic boundaries of bedrocks, beneath
shallow alluvial aquifers. Groundwater yields from bedrock aquifers
vary significantly, from 10 to 5000 m3 per day, according to the rock
type, weathered zone thickness, and topography. For example, ground-
water wells in metamorphic rocks in highlands generally have low yields,
whereas wells in flat lowlands with sedimentary and volcanic rocks tend
to have high yields. Bedrock aquifers in South Korea can be categorized
into five major geologic groups (Chae et al., 2007): granitoids, metamor-
phic rocks, complex (intermixed) rocks, volcanic rocks, and sedimentary
rocks. Granitic and metamorphic rocks comprise about 70% of the surface
outcrops in South Korea (see Fig. 1B), forming the most important and
prevailing lithologic units of fractured bedrock aquifers. The hydraulic
conductivity of bedrock aquifers is generally low (average 0.076 m/day)
and varies significantly by more than 4 orders of magnitude (Jeon et al.,
2005). Even so, bedrock aquifers in South Korea are used for various pur-
poses (i.e., agricultural, industrial, and domestic). More importantly, bed-
rock aquifers provide the public drinking water supply in rural areas and
are also pumped for production of commercial bottled water.
2.2. Nitrate datasets used for this study
The major dataset used for this study is the nitrate concentration
data collected from the NGMN (Fig. 1A). The NGMN has been operating
since 1995 by the Korea Water Resources Corporation (KOWACO, cur-
rently called K-Water) to monitor the overall status of both quantity
(by the water level fluctuation) and quality of groundwater (Choi
et al., 2014; Lee et al., 2007). A total of 320 monitoring stations are cur-
rently assessing bedrock groundwater, at an average depth of 74 m.
Among these stations, 158 also have parallel shallow monitoring wells
(average depth = 12 m) for monitoring alluvial aquifers. The regular
monitoring of water quality at the NGMN stations has been performed
twice per year for 15 parameters, including nitrate, chloride, heavy
metals, and organic contaminants. In this study, the yearly trend of ni-
trate concentrations in groundwater is evaluated for the time period be-
tween 1997 and 2009 (thus, a total of 24 monitoring times) (Table 1).
However, the NGMN's water quality data may not be sufficient to give
precise information regarding hydrogeochemical processes and con-
tamination sources, because of the general absence of information on
sampling and analysis. Thus, we first investigated the hydrochemistry
of NGMN groundwater in the first half of 2009 (Choi et al., 2014). A
total of 19 parameters, including field measurements (temperature,
pH, Eh, DO, EC), alkalinity, total dissolved solid (TDS), and major cat-
ions/anions, were obtained following the standard procedures of sam-
pling and analysis (APHA, 1985). The calculated charge balance errors
(C.B.E.) for the analyses were mostly within ±5%, indicating the good
quality of the dataset (Hounslow, 1995). The nitrate concentration
data obtained from the hydrochemical survey in 2009 (Choi et al.,
2014) are also used as the main dataset in this study (Table 1).
Additional datasets are used in this study and they were initially col-
lected from four independent groundwater surveys conducted by the
current authors. The nitrate concentration data obtained from the sur-
veys are used as supplementary datasets for comparing and statistically
evaluating the nitrate levels according to the locality and scale of
groundwater sampling (Table 1). Fig. 1B and C show the sampling local-
ities for the supplementary datasets, together with geologic and
198 K.-H. Kim et al. / Journal of Geochemical Exploration 148 (2015) 196–205

Fig. 1. Topographic maps showing (A) the locations of the NGMN stations (n = 320) of South Korea, (B) the sampling localities for supplementary datasets (Cases 1, 2, 3, and 4), and
(C) regional geology and major river basins.

topographic information. These datasets are the (1) unpublished Case 1
dataset (n = 33; filled triangles in Fig. 1B and C) that was collected from
deep wells (average depth N 500 m) for production of commercial bot-
tled water in remote unpolluted areas over South Korea; (2) Case 2
dataset (n = 65; filled squares in Fig. 1C) that was collected by the de-
tailed groundwater survey in a skirt of Guri City, east of Seoul (KECO,
2008), and represents the significant pollution of an alluvial aquifer be-
cause of its high vulnerability to nitrate pollution caused by high perme-
ability of soils (weathered coarse-grained granite) and because of over-
fertilization in agricultural fields (mainly orchards); and (3) Case 3 and
Case 4 datasets that were collected during regional groundwater
surveys in the Han River and Geum River basins, respectively (see
Fig. 1C for localities) (KOWACO, 2002, 2006), and contain a few hundred
samples of alluvial and bedrock groundwater (n = 482 for Case 3; n =
494 for Case 4).
2.3. Methods of statistical evaluation
To determine the NBL of a contaminant in groundwater, a variety of
methodologies have been applied in previous studies that generally
used large-sized datasets (hundreds or more samples). Particularly,
the cumulative probability plot (i.e., cumulative frequency diagram on
a normal probability scale) has been considered to be most useful for
determining the inflection points that separate different subpopulations
in the large datasets (Matschullat et al., 2000; Preziosi et al., 2014;
Sinclair, 1974, 1991). From the inflection points, the threshold value

Table 1
General descriptions of nitrate datasets used for this study.

Dataset Aquifer type Sampling time Sampling size/scale Data adequacya Data source

National Groundwater Monitoring Network (NGMN)

NGMN at 2009 Bedrock/alluvium 2009 Large/regional (national) Adequate Choi et al. (2014)
and this study
NGMN at 1997 to 2009 Bedrock/alluvium 1997–2009 Large/regional (national) Cannot be estimated KOWACOb
(24 sampling campaigns)

Supplementary datasets
Case 1 (pristine wells) Bedrock 2009 Small/regional Adequate Unpublished
Case 2 (high risk area) Alluvium (sandy) 2008 Small/local Adequate KECO (2008)
Case 3 (Han River basin) Bedrock/alluvium 2005 Large/regional Adequate KOWACO (2006)
Case 4 (Geum River basin) Bedrock/alluvium 2001 Large/regional Adequate KOWACO (2002)
a
Charge balance errors (C.B.E.) for the analyses are mostly within ±5%.
b
Datasets are available from the National Groundwater Information Management and Service Center of KOWACO (K-water) of Korea at http://www.gims.go.kr.
 K.-H. Kim et al. / Journal of Geochemical Exploration 148 (2015) 196–205
Table 2
Descriptive statistics of nitrate levels (mg/L) in alluvial and bedrock groundwater of the National Groundwater Monitoring Network (NGMN) of South Korea.
Dataset Aquifer type N Mean
NGMN at 1997–2009 Alluvium 2788 19.2
Bedrock 5244 16.6
Total 8032 17.5
NGMN at 2009 Alluvium 158 17.1
Bedrock 320 15.9
Total 478 16.3
n.d. = not detected (assumed as 0.035 mg/L, corresponding to the detection limit of the 2009 dataset).
a
Significant value from nonparametric test (Kolmogorov–Smirnov test) for normality.
indicating anthropogenic pollution can be selected as an upper limit (or
maximum) of the “background (natural)” subpopulation with concen-
trations of NBL. However, graphical inspections of the cumulative prob-
ability plot may lead to an unreasonable threshold value because of the
dependency on “the eyes of the investigator” (Edmunds et al., 2003;
Reimann et al., 2005). As an alternative to the graphical method, a clas-
sical model-based (objective) method (i.e., iterative 2-sigma technique)
was adopted to cluster the background population based on a theoreti-
cal density function (Matschullat et al., 2000; Nakić et al., 2007). By use
of this method, natural background samples can be identified by fitting
the frequency diagram of a dataset into a normal (or log-normal) distri-
bution after iteratively removing outliers (anomalous samples)
(i.e., normality test). However, the effectiveness of this classical
model-based method also largely depends on the sample size of the
dataset and can be attained only when the number of removed outliers
is small relative to the total sample size (Lindley, 1957; Qian and Lyons,
2006). Especially for the nitrate dataset from some localities, such as
South Korea, this classical statistical method can be inadequate for the
satisfactory estimation of NBL (and threshold) of nitrate because
(1) the outliers among the samples can be very large relative to the sam-
ple size because of the ubiquitous and pervasive nitrate pollution, and
(2) the datasets may follow the bimodal (or polymodal) distribution be-
cause of the coexistence of two or more processes controlling water
quality.
Therefore, nitrate datasets from the NGMN in this study are exam-
ined by the model-based approach using a finite mixture model that is
very suitable for directly estimating the parameters of component dis-
tributions in the datasets without hypothetical normality tests
(Biernacki et al., 2006; Casella et al., 2002). Because nitrate in ground-
water on a regional scale has an origin from both natural and anthropo-
genic processes, the distribution of nitrate data can be theoretically
bimodal and thus can be divided into two components based on a finite
normal (Gaussian) mixture model (Molinari et al., 2012; Rodríguez
et al., 2006; Wendland et al., 2008). This approach is consistent with
the basic contamination model that has been widely used to describe
a mixture with the combination of statistical distributions (Aitkin and
Wilson, 1980; Joo et al., 2007, 2009; Titterington, 1993). If the random
variable x follows a finite mixture distribution with two normal densi-
ties of natural (background) and anthropogenic components, the prob-
ability density function p(x) of the model can be defined as
pðxÞ ¼ ð1−πÞf 1 ðxÞ þ π f 2 ðxÞ;
where f1(x) and f2(x) are normal forms, such that fi = Ø(x|μi, σi) with
mean (μi) and variance (σ2i ), and the parameter π denotes the mixing
ratio of anthropogenic distribution. Among a few different approaches
to estimate the parameters of a finite mixture distribution (McLachlan
and Peel, 2000), we use the expectation–maximization (EM) algorithm,
which converges to a maximum likelihood estimate of the mixture pa-
rameters. The estimations can be simply determined by the Mixture
Modeling (MIXMOD) program (Biernacki et al., 2006). Additionally,
nonparametric tests, such as the Mann–Whitney (M–W) and Kolmogo-
rov–Smirnov (K–S) tests, are used to compare different independent
datasets used in this study and to test the goodness-of-fit of the
199
SD Min. Median Max. CV (%) P-valuea
32.1 n.d. 11.1 797.4 167.0 b0.01
27.0 n.d. 8.9 494.4 162.6 b0.01
28.9 n.d. 9.7 797.4 165.0 b0.01
21.7 0.035 12.2 132 127.2 b0.01
24.3 0.035 8.7 188.9 153.2 b0.01
23.5 0.035 9.7 188.9 144.3 b0.01
empirical data on the assumed probability models (i.e., normal and
log-normal distributions).
3. Results and discussion
3.1. Nitrate levels of South Korean groundwater
Two types of NGMN datasets (i.e., datasets of 1997 to 2009 and
2009) are examined to assess the status of nitrate levels in both alluvial
and bedrock groundwater on a regional (national) scale. Table 2 sum-
marizes descriptive statistics for the two datasets. The mean and medi-
an values of nitrate concentrations of the 1997–2009 dataset (n =
8032) are 17.5 mg/L and 9.7 mg/L, respectively. According to the aquifer
type, the mean and median values are 19.2 mg/L and 11.1 mg/L in allu-
vial groundwater (n = 2788) and 17.5 mg/L and 9.7 mg/L in bedrock
groundwater (n = 5244). The maximum concentrations of nitrate are
very high (797.4 mg/L). Our NGMN data (n = 478) collected in a
hydrochemical survey in 2009 have the mean and median values of
17.1 mg/L and 12.2 mg/L in alluvial groundwater and 15.9 mg/L and
8.7 mg/L in bedrock groundwater (Table 2), which are in agreement
with the results from the NGMN 1997–2009 dataset. Compared with a
number of large datasets from other countries, the nitrate levels of
South Korean groundwater seem to be generally high. The median
values of nitrate analyses (n = 6000) in 22 counties of the State of
Iowa in the US was 5.8 mg/L (Halberg and Keedny, 1993). A large
dataset from the National Water Quality Assessment (NAWQA) Pro-
gram by the United States Geological Survey (USGS) showed the medi-
an values of 0.9 mg/L in the low-risk group (n = 779) and 20.6 mg/L in
the high-risk group (n = 1032) (Nolan et al., 1997).
The statistical distribution of the two NGMN datasets is examined by
the coefficient of variation (CV) and the nonparametric normality test
(i.e., Kolmogorov–Smirnov test). The percentages of the CV range
from 127.2% to 167.0%, and the P-value of the K–S test on the null hy-
pothesis that the distribution is normal is b 0.01 (Table 2), indicating
that the NGMN nitrate data are characterized by a substantially skewed
distribution with a heavy tail to the right (i.e., non-normality). The ma-
jority of environmental datasets, including groundwater constituents,
generally does not follow a normal distribution (Reimann and
Filzmoser, 2000). Panno et al. (2006) also suggested that the non-
normality for large datasets of groundwater nitrate is the result of re-
gional influences on aquifers by widespread nonpoint sources. Because
statistical distributions of NGMN nitrate datasets are closer to the log-
normality, log-transformed nitrate data are used in the following
analyses.
The box plots in Fig. 2 show the historical (time series) variation of
nitrate levels in NGMN groundwater during the past 12 years (1997
to 2009). Any distinct temporal change of nitrate concentrations does
not appear in both alluvial and bedrock groundwater, suggesting that
regional nitrate contamination has been steadily prevailing over South
Korea at least since 1997. Although no data on nitrate concentrations
before 1997 are available, nitrate pollution of groundwater in South
Korea has likely prevailed for approximately 4 decades, since the sub-
stantial production and consumption of chemical nitrogen fertilizers
(N150 kg/ha per year) began in the 1970s in South Korea (MOAF,
200 K.-H. Kim et al. / Journal of Geochemical Exploration 148 (2015) 196–205

Fig. 2. Historical changes of nitrate levels in the NGMN datasets collected from 24 sampling campaigns during 12 years (from 1997 to 2009). The lines above the box plots indicate the
percentage of samples exceeding the drinking water standard (DWS) of nitrate (44.3 mg/L) in Korea.

2001). Fig. 2 also shows a trend of an increase in the number of samples
exceeding the Korean drinking water standard (44.3 mg/L), which is re-
markable for bedrock groundwater. This trend can be a result of the in-
creased number of monitoring stations in recent years. Nevertheless, it
should be noted that NGMN datasets do not show any decreasing trend
of nitrate concentrations in either bedrock or alluvial groundwater.
The nitrate levels of four supplementary datasets are listed in Table 3
and are shown in Fig. 3. The decreasing order of median values are allu-
vial groundwater of Case 2 (high-risk area) (76.2 mg/L), alluvial
groundwater of Case 3 (Han River Basin) (21.2 mg/L), bedrock ground-
water of Case 4 (Geum River Basin) (18.2 mg/L), alluvial groundwater of
Case 4 (15.3 mg/L), bedrock groundwater of Case 3 (13.3 mg/L), and
bedrock groundwater of Case 1 (bottled water from pristine areas)
(3.0 mg/L). As expected, the Case 1 dataset shows the lowest nitrate
concentrations, whereas the Case 2 dataset shows the highest concen-
trations. The other two supplementary datasets (Case 3 and Case 4) ob-
tained from two large river basins show nitrate levels similar to the
NGMN 2009 hydrochemical dataset.
The statistical difference between neighboring datasets is tested by
the Mann–Whitney U test and the results are shown as P-values
above the box plots (Fig. 3). The datasets of Case 1 and Case 2 are signif-
icantly different (P-values b 0.01) from the other datasets, whereas the
datasets of Case 3 and Case 4 (i.e., data from regional surveys) are not
statistically different from each other (i.e., P-values N 0.05), except for
the difference between bedrock groundwater in the two cases
(Fig. 3B). The nitrate levels of the NGMN dataset tend to be higher for
alluvial groundwater than bedrock groundwater (Fig. 3A), which sug-
gests that shallow aquifers are more vulnerable to nitrate pollution
than deep aquifers (Böhlke, 2002; Spalding and Exner, 1993). In Fig. 3,
for the interquartile range (IQR) of datasets, Case 3 and Case 4 fall be-
tween the IQR values of the other two datasets (Case 1 and Case 2).
Also, the nitrate data of Case 3 and Case 4 show the non-normality as
observed in the NGMN datasets because the CV values of these datasets
are high (N100%) and the P-values of the K–S test are b 0.01 (Table 3).
However, the datasets of Case 1 and Case 2 accept the null hypothesis
(i.e., P-values of the K–S test are higher than 0.01), indicating the normal
distribution of these datasets. Such contrasting results for the two types
of datasets (i.e., Cases 1 and 2 versus Cases 3 and 4) clearly indicate that
groundwater nitrate data will show the normal distribution if the sur-
veyed groundwater system is dominantly influenced by a single source
process (i.e., the natural process as in Case 1 or the anthropogenic pollu-
tion as in Case 2), whereas regional datasets, such as NGMN, Case 3 and
Case 4, show the non-normal distribution because of intermixed source
processes. Therefore, the NBLs are not so simple and easy to estimate
from the nitrate distributions of the NGMN datasets, and the use of in-
depth statistical and graphical approaches is required.
3.2. Natural background levels of nitrate
To estimate the NBLs (and thresholds) of nitrate in South Korean
groundwater, several statistical approaches using different datasets
are tested. As described above, log-transformation of nitrate data and

Table 3
Descriptive statistics of nitrate levels (mg/L) in four cases of supplementary datasets of groundwater in South Korea.

Dataset N Mean SD Min. Median Max. CV (%) p-valuea

Case 1 (pristine wells) 33 4.2 2.8 0.1 3.0 10 66.5 N0.01

Case 2 (high risk area) 65 85.5 62.2 1.6 76.2 252 72.7 N0.01
Case 3 (bedrock GW; Han River basin) 191 18.8 22.6 0.01 13.3 112.7 120.3 b0.01
Case 3 (alluvial GW; Han River basin) 291 45.3 60.6 0.04 21.1 513.2 133.6 b0.01
Case 4 (bedrock GW; Geum River basin) 304 26.8 23.7 0.1 18.2 122.8 88.6 b0.01
Case 4 (alluvial GW; Geum River basin) 190 42.7 61.5 0.1 15.3 295.9 143.8 b0.01
a
Significant value from non-parametric test (Kolmogorov–Smirnov test) for normality.
 K.-H. Kim et al. / Journal of Geochemical Exploration 148 (2015) 196–205
Fig. 3. Box plots comparing the nitrate levels of the different datasets in which the outliers are defined as samples out of the upper and lower whiskers, and the P-values (b0.05) indicate a
significant difference between the neighboring datasets from the Mann–Whitney U test.
the exclusion of outliers (i.e., samples falling outside the whiskers on a
box plot as in Fig. 3; Reimann et al., 2005) are employed for the datasets
showing the negatively skewed distribution (e.g., NGMN datasets).
Then, the distribution of log-transformed nitrate concentrations, except
outliers, is separated into two subpopulations (sample groups) by the
nitrate concentrations of the NBL and anthropogenic polluted level
(APL). Each level is set as a credible interval (i.e., from the 10th to the
90th percentile in the distribution). From the limits of estimated levels,
the threshold directly indicating groundwater pollution can be defined
as a fixed value (e.g., an upper or outer limit of NBL).
3.2.1. An empirical approach using the hypothetical mixture model between
datasets of Cases 1 and 2
The datasets of Case 1 and Case 2 show the log-normality as indicat-
ed by the P-values of N0.01 (Table 3), because of their influence from a
single source process (i.e., natural for Case 1 or anthropogenic for Case
2). The histograms of log-transformed datasets, except outliers, show
that the nitrate distributions follow the normal probability density func-
tion (PDF), with distinct mean and variance (Fig. 4A and B). These two
independent populations may empirically represent the groups of NBL
and APL of nitrate in South Korea. Therefore, each level (from the 10th
to the 90th percentile) can be set as 1.6 to 8.0 mg/L for the empirical
NBL and 43.3 to 177.6 mg/L for the empirical APL. The upper limit of
the empirical NBL (i.e., NBL90 = 8.0 mg/L) can be regarded as a thresh-
old indicating anthropogenic pollution (Fig. 4A and B).
In advance, a hypothetical mixture model with two components
(natural and anthropogenic) can be established by combining the two
datasets (Case 1 and Case 2). Fig. 4C shows the mixture distribution as
a cumulative probability plot on which two normal components (blue
and red straight lines) are superimposed. The mixture distribution is
curved in the range between the NBL90 (8.0 mg/L) and APL10
(43.3 mg/L), suggesting a bimodality of the mixed distribution on log-
scale. Moreover, the plot shows the occurrence of two inflection points
at the upper and lower limits of the empirical normal components
(i.e., NBL90 and APL10); here, the first point (NBL90 = 8.0 mg/L) corre-
sponds to the empirically selected threshold. Therefore, we now suggest
that the relevant NBL (and threshold) in a mixed dataset can be
201
estimated by the classical graphical approach (i.e., detecting the first in-
flection point on the cumulative probability plot).
3.2.2. The use of a finite normal (Gaussian) mixture model on the NGMN
datasets
The hypothetical mixture model from the two supplementary
datasets (Case 1 and Case 2) also provides prior knowledge about the
statistical distribution of nitrate concentration data on a regional (na-
tional) scale. We can expect that the NGMN datasets should have the bi-
modal distribution with two normal components (natural and
anthropogenic) on a log-scale. Thus, a finite normal (Gaussian) mixture
model with two components was tested to fit the observed distributions
of nitrate concentrations in the NGMN data, which allows the separa-
tion of the datasets into two subpopulations, reflecting NBL and APL
by directly estimating model parameters such as the mixing ratio and
the mean, variance, and interval of each component.
The histograms in Fig. 5 present the empirical distribution of log-
transformed nitrate data, except the outliers, in alluvial and bedrock
groundwater of the NGMN 2009 dataset. As expected, the distribution
is strongly negatively skewed to the low level of nitrate concentrations
(i.e., polluted samples are predominate in the datasets) and is neither
normal nor log-normal (i.e., non-normality on log-scale). The nitrate
data show a type of the bimodal distribution (i.e., mixture between nat-
ural and anthropogenic components). Therefore, the empirical bimodal
distribution was fitted by the finite normal mixture model with two
components (Fig. 5). Table 4 lists the estimated parameters for the
two normal subpopulations and the confidence intervals indicating
NBL and APL in each groundwater unit. The estimated NBLs of nitrate
range from 0.1 to 1.0 mg/L for bedrock groundwater and from 0.7 to
1.1 mg/L for alluvial groundwater. In addition, the upper limits of NBL
(i.e., NBL90 = 1.0 and 1.1 mg/L) may represent the thresholds indicating
the anthropogenic influences on the respective groundwater units.
However, these estimated values (NBL90) are significantly lower than
the threshold (8 mg/L) estimated from the empirical NBL (Case 1).
Therefore, the reactivity of nitrate should be examined because
some geochemical processes (e.g., denitrification) can attenuate nitrate
concentrations, as have been reported for many aquifers in South Korea
(e.g., Choi et al., 2011, 2012; Kim et al., 2009). If the NGMN dataset is
202 K.-H. Kim et al. / Journal of Geochemical Exploration 148 (2015) 196–205

Fig. 5. Statistical distributions (histograms) of nitrate levels for (A) bedrock and
(B) alluvial groundwater units in the NGMN datasets, which are fitted by each normal
mixture model with two components and thus separated into two subpopulations
reflecting the estimated NBL and APL, respectively.

in Fig. 5) may represent the groundwater influenced by natural attenu-

ation (i.e., denitrification).
On the other hand, the polluted groundwater observed in high-risk
areas (Case 2) have high NO3/Cl ratios (average 2.6), and the NGMN
groundwater samples with high nitrate concentrations show widely
varying ratios (Fig. 6). This wide variation is mainly caused by diverse
diffusive sources with different composition, even though varying de-
Fig. 4. Statistical distributions (histograms) of nitrate levels for two supplementary grees of denitrification can also occur in some polluted groundwater.
datasets with normality on log-scale: (A) Case 1 (pristine wells) and (B) Case 2 (high- Nevertheless, it is likely that natural (or less polluted) groundwater
risk area), and (C) the cumulative probability plot of two combined datasets showing a hy-
should have the NO3/Cl ratios below or near 1.0 (i.e., the ratios of rain-
pothetical mixture distribution affected by natural and anthropogenic impacts,
simultaneously. water and Case 1 groundwater), regardless of the natural attenuation
of anthropogenic nitrate. In Fig. 6, NGMN groundwater with NO3/Cl ra-
tios below or near 1.0 has nitrate levels below the APL10 (3.0 and
5.5 mg/L). This observation suggests that the estimated subpopulation
affected by denitrification, the estimated NBLs and thresholds (NBL90) reflecting APLs involves no or very few samples originated from natural
may be too underestimated to directly indicate the anthropogenic sources; i.e., the upper limits of true NBLs approach the lower limit of
pollution. the APLs in the NGMN datasets. Thus, we consider that the acceptable
thresholds indicating nitrate pollution fall between the upper limit of
NBL (i.e., N NBL90) and the lower limit of APL (i.e., bAPL10) in Fig. 5.
3.3. Thresholds (as reference values) of nitrate Moreover, the subpopulation of APL dominates the nitrate distribution
of the NGMN datasets (i.e., the number of polluted samples are much
Through an inspection of the relation between nitrate and chloride
in the NGMN 2009 dataset, we tried to analogically assess the effect of Table 4
denitrification on nitrate levels. Chloride ion generally accompanies ni- The results of model-based approach using a finite normal two-component mixture
trate in groundwater affected by diffusive pollution but behaves conser- model.
vatively, unlike nitrate. Fig. 6 shows the relations between nitrate NGMN at 2009 Component Parameter Credible interval
concentrations and NO3/Cl mass ratios in Korean NGMN groundwater. (center)
The NO3/Cl ratios are very low in groundwater, with the nitrate concen-
Mixing ratio Mean SD P10 P50 P90
trations below the proposed NBL90 (1.0 and 1.1 mg/L), whereas the
Bedrock GW NBL 0.11 0.5 2.8 0.1 0.3 1.0
values are highly variable with increasing nitrate levels. The NO3/Cl ra-
APL 0.89 11.9 2.7 3.0 11.6 43.1
tios for the low nitrate groundwater are much lower than the average Alluvial GW NBL 0.09 1.1 1.5 0.7 1.0 2.0
ratio (1.0) of rain water in South Korea (Kim, 2007). The ratios are APL 0.91 15.6 2.2 5.5 16.6 37.1
also lower than the average ratio (1.3) for the pristine groundwater Shown are estimated parameters of two subpopulations: groups reflecting national back-
(Case 1), having very low nitrate levels (Fig. 6). Therefore, the subpop- ground level (NBL) and anthropogenic polluted level (APL) in the NGMN 2009 datasets.
ulation with very low nitrate concentrations (below the estimated NBLs The bolded numbers are selected as thresholds.
 K.-H. Kim et al. / Journal of Geochemical Exploration 148 (2015) 196–205
Fig. 6. The NO−
3 /Cl mass ratios versus the nitrate levels in the NGMN datasets; the samples
showing the significant low ratios (b1.0 of average rainwater of Korea) are mainly ob-
served in low nitrate levels below NBL90, which indicate that the natural attenuation pro-
cesses may control the estimated components of NBL in the datasets.
higher than that of the natural samples). Thus, the normal abundance can
be defined for the APL while the NBL distribution can be anomalous. This
relation induces the lower tail of APL to cover the upper limit of NBL (see
Fig. 5), and therefore the estimated NBLs from the mixture model may be
statistically not significant. This pattern is also shown in the cumulative
probability plot as follows.
As shown in the hypothetical mixture (Fig. 4C), inflection points are
observed on the cumulative probability plots of the NGMN 2009
datasets (Fig. 7). Blue and red straight lines in the figure indicate the
fitted normal components reflecting the estimated NBL and APL values,
respectively. Unlike the hypothetical mixture, the inflection points are
difficult to precisely determine because the distribution shape on the
cumulative probability plot can be affected by geometric features,
such as sizes (or mixing ratios), means, and variances, of subpopulations
consisting of the regional dataset (Qian and Lyons, 2006). The inflection
points can be clearly shown when the means and sizes of two subpop-
ulations are far apart and thus can be equivalent as the hypothetical
mixture (see Fig. 5). In the case of the NGMN 2009 datasets, a single in-
flection point revealed by intersecting the normal lines superimposed
on each plot presents the nitrate concentrations of 2.6 mg/L for bedrock
groundwater and 5.3 mg/L for alluvial groundwater (Fig. 7). The inflec-
tion points are approximately the same with the lower limits of APL
(i.e., APL10) because the upper limits of NBL (i.e., NBL90) are significantly
overlapped with the distribution of APL. Thus, our model-based ap-
proach, including graphical methods, indicates that the assumed proba-
bility density function is a robust fit onto a polluted subpopulation.
This study suggests that the lower limits of the polluted subpopula-
tion (APL10 = 3.0 mg/L for bedrock groundwater and 5.5 mg/L for allu-
vial groundwater; Figs. 5 and 7) in the NGMN 2009 datasets are most
appropriate as the nitrate thresholds for the following reasons:
(1) The estimated NBLs are largely affected by natural attenuation
(i.e., denitrification), and (2) the sampling size of APL is much larger
203
Fig. 7. Cumulative probability plots of the NGMN datasets on which two normal subpop-
ulations (the estimated NBL and APL) are superimposed from each mixture model. In each
plot, a single inflection point is observed at intersecting points of the partitioned compo-
nents (lines on log-scale).
than that of the natural (background) subpopulation, because of ubiqui-
tous and pervasive nitrate pollution.
Furthermore, our statistical investigation can provide a heuristic
procedure to statistically determine the NBL (and threshold) of any
widespread pollutants, including nitrate from a large dataset. The meth-
od consists of four steps:
(1) Log-transformation followed by exclusion of outlying samples to
achieve a symmetrical distribution.
(2) If the distribution shows a bimodality, the empirical dataset can
be successfully fitted using a finite normal model with two com-
ponents, and this approach can help estimate the two subpopu-
lations (NBL and APL) with the basic statistical parameters
(i.e., mixing ratios, means, and variances).
(3) Between the upper and lower limits of the two subpopulations,
we can select a threshold (as a reference value indicating the pol-
lution) considering geochemical processes.
(4) Assuming that the NBL subpopulation can be considered as
anomalous in the regional dataset, we can graphically inspect
the inflection points on a cumulative probability plot by assum-
ing the normal distribution of anthropogenic polluted subpopu-
lation on a log-scale. In this case, the estimated inflection point
can indicate the lower limit of the subpopulation of APL.
Using the thresholds (3.0 mg/L for bedrock groundwater and
5.5 mg/L for alluvial groundwater), we estimated the pollution status
of South Korean groundwater in two major river basins (i.e., regional
datasets of Cases 3 and 4) (Fig. 8). In Fig. 8, a P-value of N 0.05 from
the K–S test indicates that the polluted subpopulation has normality,
and thus the estimated threshold can be valid. The percentages of
groundwater samples with nitrate concentrations exceeding the
thresholds are significantly high: 73.3% and 84.2%, respectively, of
204 K.-H. Kim et al. / Journal of Geochemical Exploration 148 (2015) 196–205
Fig. 8. Nitrate distribution of the supplementary datasets from a regional hydrochemical survey in which groundwater pollution is assessed based on the selected thresholds (3.0 and
5.5 mg/L in bedrock and alluvial groundwater) from the NGMN datasets. In each plot, P-values (N0.05) from a K–S test indicate that an estimated polluted population of each dataset fol-
lows normality on log-scale.
bedrock groundwater and alluvial groundwater in the Han River Basin
and 91.8% of bedrock groundwater in the Geum River Basin. This esti-
mation indicates the very widespread influence on South Korean
groundwater from diffusive sources, even in bedrock groundwater.
Therefore, effective groundwater quality management plans are urgent-
ly needed in South Korea.
4. Summary and conclusions
In this study, we assessed the nitrate levels in groundwater from the
National Groundwater Monitoring Network of South Korea. The median
values of nitrate in 2009 are 8.7 and 12.2 mg/L in alluvial groundwater
and bedrock groundwater, respectively, which are higher than those re-
ported from a number of regional surveys in other countries. The avail-
able historical data from the NGMN show that high mean levels of
nitrate have been consistent during the period between 1997 and
2009, suggesting widespread nitrate pollution of regional aquifer sys-
tems of Korea at least since 1997.
The statistical distribution of regional nitrate datasets is characteris-
tically skewed substantially with a heavy tail to the right (i.e., non-
normality); moreover, it does not follow the normality on a log-scale.
Rather, the distribution can be defined as a bimodality with two normal
subpopulations. Thus, the NGMN 2009 datasets could be fitted with a fi-
nite normal (Gaussian) mixture model with two components. The
model fitting resulted in a good separation of the datasets into two sub-
populations, each reflecting the NBL and APL.
However, the estimated NBLs in the NGMN datasets are lower than
the expected level because natural attenuation processes, such as deni-
trification, largely affect the subpopulation of NBL. Moreover, anthropo-
genic polluted samples are predominant and show a log-normal
distribution, whereas natural samples form a small anomalous
population. Therefore, the estimated APLs should be more robust than
the NBL obtained from model-based approaches, including the cumula-
tive probability plot. This study suggests that the lower limits of the pol-
luted subpopulation (i.e., APL10) are more appropriate as the thresholds
in the NGMN datasets for two reasons: (1) The estimated NBLs are
largely affected by denitrification, and (2) the sampling sizes of APL
are more abundant than those of background samples because of the
widespread nitrate pollution. Our results provide the regional (nation-
al) thresholds of nitrate as 3.0 and 5.5 mg/L, respectively, for bedrock
groundwater and alluvial groundwater.
This study provides the useful thresholds of nitrate as a practical
guide to differentiate anthropogenic pollution on a regional (national)
scale, which can be used for groundwater quality management. Further-
more, this study shows the usefulness of a heuristic procedure to deter-
mine the NBL and threshold of widespread pollutants, such as nitrate in
groundwater systems. The statistical approach in this study can be suc-
cessfully applied to groundwater datasets with smaller portions of back-
ground samples than polluted samples.
Acknowledgments
This work was supported by the 2013 project (Survey of Groundwater
Contamination and Backgrounds in Livestock Farming Areas, Korea) funded
by the National Institute of Environmental Research and by the National Re-
search Foundation of Korea Grant (2013, University-Institute Cooperation
Program) funded by the Ministry of Science, ICT & Future Planning.
References
Aitkin, M., Wilson, G.T., 1980. Mixture models, outliers, and the EM algorithm.
Technometrics 22, 325–331.
 K.-H. Kim et al. / Journal of Geochemical Exploration 148 (2015) 196–205
APHA (American Public Health Association), 1985. Standard Methods for the Examination
of Water and Wastewater, 20th ed. APHA and AWWA, Washington DC.
Appelo, C.A.J., Postma, D., 1999. Geochemistry, Groundwater and Pollution. Balkema, Rot-
terdam (536 pp.).
Arnell, N.W., 1999. Climate change and global water resources. Glob. Environ. Chang. 9
(Supplement 1), S31–S49.
Biernacki, C., Celeux, G., Govaert, G., Langrognet, F., 2006. Model-based cluster and dis-
criminant analysis with the MIXMOD software. Comput. Stat. Data Anal. 51, 587–600.
Böhlke, J.K., 2002. Groundwater recharge and agricultural contamination. Hydrogeol. J. 10,
153–179.
Burkart, M.R., Stoner, J.D., 2008. Nitrogen in groundwater associated with agricultural sys-
tems, In: Hatfield, J.L., Follett, R.F. (Eds.), Nitrogen in the Environment, 2nd ed. Aca-
demic Press, San Diego, pp. 177–202.
Burow, K.R., Nolan, B.T., Rupert, M.G., Dubrovsky, N.M., 2010. Nitrate in groundwater of
the United States, 1991–2003. Environ. Sci. Technol. 44, 4988–4997.
Casella, G., Mengersen, K.L., Robert, C.P., Titterington, D.M., 2002. Perfect samplers for
mixtures of distributions. J. R. Stat. Soc. Ser. B (Stat. Methodol.) 64, 777–790.
Chae, G.T., Kim, K., Yun, S.T., Kim, K.H., Kim, S.O., Choi, B.Y., Kim, H.S., Rhee, C.W., 2004.
Hydrogeochemistry of alluvial groundwaters in an agricultural area: an implication
for groundwater contamination susceptibility. Chemosphere 55, 369–378.
Chae, G.T., Yun, S.T., Mayer, B., Kim, K.H., Kim, S.Y., Kwon, J.S., Kim, K., Koh, Y.K., 2007.
Fluorine geochemistry in bedrock groundwater of South Korea. Sci. Total Environ.
385, 272–283.
Chae, G.T., Yun, S.T., Mayer, B., Choi, B.Y., Kim, K.H., Kwon, J.S., Yu, S.Y., 2013.
Hydrochemical and stable isotopic assessment of nitrate contamination in an alluvial
aquifer underneath a riverside agricultural field. Agric. Water Manag. 96, 1819–1827.
Choi, W.J., Han, G.H., Lee, S.M., Lee, G.T., Yoon, K.S., Choi, S.M., Ro, H.M., 2007. Impact of
land-use types on nitrate concentration and δ15N in unconfined groundwater in
rural areas of Korea. Agric. Ecosyst. Environ. 120, 259–268.
Choi, B.Y., Yun, S.T., Mayer, B., Kim, K.H., 2011. Sources and biogeochemical behavior of ni-
trate and sulfate in an alluvial aquifer: hydrochemical and stable isotope approaches.
Appl. Geochem. 26, 1249–1260.
Choi, B.Y., Yun, S.T., Kim, K.H., Kim, K., Choh, S.J., 2012. Geologically controlled agricultural
contamination and water–rock interaction in an alluvial aquifer: results from a
hydrochemical study. Environ. Earth Sci. 1–15.
Choi, B.Y., Yun, S.T., Kim, K.H., Kim, J.W., Kim, H.M., Koh, Y.K., 2014. Hydrogeochemical in-
terpretation of South Korean groundwater monitoring data using Self-organizing
Maps. J. Geochem. Explor. 137, 73–84.
Corniello, A., Ducci, D., 2014. Hydrogeochemical characterization of the main aquifer of
the “Litorale Domizio-Agro Aversano NIPS” (Campania — southern Italy). J. Geochem.
Explor. 137, 1–10.
Edmunds, W.M., Shand, P., 2008. Natural Groundwater Quality. Wiley-Blackwell, London
(488 pp.).
Edmunds, W.M., Shand, P., Hart, P., Ward, R.S., 2003. The natural (baseline) quality of
groundwater: a UK pilot study. Sci. Total Environ. 310, 25–35.
Halberg, G.R., Keedny, D.R., 1993. Nitrate. In: Alley, W.M. (Ed.), Regional Groundwater
Quality. Van Norstrand Reinhold, New York, pp. 297–322.
Hinsby, K., Condesso, de Melo, M.T., Dahl, M., 2008. European case studies supporting the
derivation of natural background levels and groundwater threshold values for the
protection of dependent ecosystems and human health. Sci. Total Environ. 401, 1–20.
Hounslow, A., 1995. Water Quality Data: Analysis and Interpretation. CRC-Press, (416 pp.
).
Jeon, S.K., Koo, M.H., Kim, Y., Kang, I.O., 2005. Statistical analysis of aquifer characteristics
using pumping test data of national groundwater monitoring well of Korea. Korean
Soc. Soil Groundw. Environ. J. 10, 32–44 (in Korean).
Joo, Y., Lee, K., Min, J.H., Yun, S.T., Park, T., 2007. Logistic mixture of multivariate regressions
for analysis of water quality impacted by agrochemicals. Environmetrics 18, 499–514.
Joo, Y., Kim, D., Lee, K., Yun, S.T., Kim, K.H., Mercante, D., 2009. Estimation of anthropogen-
ic pollution using a Bayesian contamination model: an application to fractured bed-
rock groundwater from Han River Watershed, South Korea. Environmetrics 20,
221–234.
KECO (Korea Environment Corporation), 2008. Report of 2008 Detailed Survey of
Groundwater Contaminaion, 2002. Korea Environment Corporation, Incheon (in
Korean).
Kim, J.H., 2007. Acid Rain. Seoul National University Press, (ISBN 978-89-521-0784-8
93470, in Korean).
Kim, K.H., Yun, S.T., Choi, B.Y., Chae, G.T., Joo, Y., Kim, K., Kim, H.S., 2009. Hydrochemical
and multivariate statistical interpretations of spatial controls of nitrate concentra-
tions in a shallow alluvial aquifer around oxbow lakes (Osong area, central Korea).
J. Contam. Hydrol. 107, 114–127.
Koh, D.C., Mayer, B., Lee, K.S., Ko, K.S., 2010. Land-use controls on sources and fate of ni-
trate in shallow groundwater of an agricultural area revealed by multiple environ-
mental tracers. J. Contam. Hydrol. 118, 62–78.
KOWACO (Korea Water Resources Corporation), 2002. A Regional Groundwater Survey
Report in the Geum River Watershed, Korea. Water Resources Corporation, Taejon
(in Korean).
KOWACO (Korea Water Resources Corporation), 2006. A Regional Groundwater Survey Re-
port in the Han River Watershed, Korea. Water Resources Corporation, Taejon (in
Korean).
205
Langmuir, D., 1997. Aqueous Environmental Geochemistry. Prentice Hall, Upper Saddle
River, N.J. (600 pp.).
Lee, J.Y., Yi, M.J., Yoo, Y.K., Ahn, K.H., Kim, G.B., Won, J.H., 2007. A review of the National
Groundwater Monitoring Network in Korea. Hydrol. Process. 21, 907–919.
Limbrick, K.J., 2003. Baseline nitrate concentration in groundwater of the Chalk in south
Dorset, UK. Sci. Total Environ. 314–316, 89–98.
Lindley, D.V., 1957. A statistical paradox. Biometrika 44, 187–192.
Matschullat, J., Ottenstein, R., Reimann, C., 2000. Geochemical background — can we cal-
culate it? Environ. Geol. 39, 990–1000.
McLachlan, G.J., Peel, D., 2000. Finite mixture models. Applied Probability and Statistics
Section, Wiley Series in Probability and Statistics. Wiley, New York (419 pp.).
MOAF (Ministry of Agriculture and Forestry, Republic of Korea), 2001. Agricultural and
Forestry Statistical Yearbook1000-51023-26-01 , (in Korean).
MOCT (Ministry of Ocean, Construction and Transportation), 2007. Master Plan for
Groundwater Management of Korea: 2007–2011, (149 pp.).
Molinari, A., Guadagnini, L., Marcaccio, M., Guadagnini, A., 2012. Natural background
levels and threshold values of chemical species in three large-scale groundwater bod-
ies in Northern Italy. Sci. Total Environ. 425, 9–19.
Müller, D., Blum, A., Hart, A., Hookey, J., Kunkel, R., Scheidleder, A., 2006. Final Proposal of a
Methodology to Set Up Groundwater Threshold Values in Europe. www.wfd-bridge.net.
Nakić, Z., Posavec, K., Bačani, A., 2007. A visual basic spreadsheet Macro for geochemical
background analysis. Ground Water 45, 642–647.
Nolan, B.T., 2001. Relating nitrogen sources and aquifer susceptibility to nitrate in shallow
ground waters of the United States. Ground Water 39, 290–299.
Nolan, B.T., Ruddy, B.C., Hitt, K.J., Helsel, D.R., 1997. Risk of nitrate in groundwaters of the
United States. A national perspective. Environ. Sci. Technol. 31, 2229–2236.
OECD (Organisation for Economic Cooperation and Development), 2004. OECD Environ-
mental Data: Compendium 2004, Agriculture. Environmental Performance and Infor-
mation Division of OECD Environment Directorate, (www.oecd.org/dataoecd/30/7/
32613678).
OECD (Organisation for Economic Co-operation and Development), 2008. Environmental
Performance of Agriculture in OECD countries since 1990France, Paris www.oecd.
org/tad/env/indicators.
Panno, S.V., Kelly, W.R., Martinsek, A.T., Hackley, K.C., 2006. Estimating background and
threshold nitrate concentrations using probability graphs. Ground Water 44,
697–709.
Petry, J., Soulsby, C., Malcolm, I.A., Youngson, A.F., 2002. Hydrological controls on nutrient
concentrations and fluxes in agricultural catchments. Sci. Total Environ. 294, 95–110.
Postma, D., Boesen, C., Kristiansen, H., Larsen, F., 1991. Nitrate reduction in an unconfined
sandy aquifer: water chemistry, reduction processes, and geochemical modeling.
Water Resour. Res. 27, 2027–2045.
Preziosi, E., Giuliano, G., Vivona, R., 2010. Natural background levels and threshold values
derivation for naturally As, V and F rich groundwater bodies: a methodological case
study in central Italy. Environ. Earth Sci. 61, 885–897.
Preziosi, E., Parrone, D., del Bon, A., Ghergo, S., 2014. Natural background level assessment
in groundwaters: probability plot versus pre-selection method. J. Geochem. Explor.
143, 43–53.
Qian, S.S., Lyons, R.E., 2006. Characterization of background concentrations of contami-
nants using a mixture of normal distributions. Environ. Sci. Technol. 40, 6021–6025.
Reimann, C., Filzmoser, P., 2000. Normal and lognormal data distribution in geochemistry:
death of a myth. Consequences for the statistical treatment of geochemical and envi-
ronmental data. Environ. Geol. 39, 1001–1014.
Reimann, C., Garrett, R.G., 2005. Geochemical background — concept and reality. Sci. Total
Environ. 350, 12–27.
Reimann, C., Filzmoser, P., Garrett, R.G., 2005. Background and threshold: critical compar-
ison of methods of determination. Sci. Total Environ. 346, 1–16.
Rodríguez, J.G., Tueros, I., Borja, A., Belzunce, M.J., Franco, J., Solaun, O., Valencia, V., Zuazo,
A., 2006. Maximum likelihood mixture estimation to determine metal background
values in estuarine and coastal sediments within the European Water Framework Di-
rective. Sci. Total Environ. 370, 278–293.
Scanlon, B.R., Jolly, I., Sophocleous, M., Zhang, L., 2007. Global impacts of conversions from
natural to agricultural ecosystems on water resources: quantity versus quality. Water
Resour. Res. 43, W03437.
Sebilo, M., Mayer, B., Nicolardot, B., Pinay, G., Mariotti, A., 2013. Long-term fate of nitrate
fertilizer in agricultural soils. Proc. Natl. Acad. Sci. U. S. A. 110, 18185–18189.
Sinclair, A.J., 1974. Selection of threshold values in geochemical data using probability
graphs. J. Geochem. Explor. 3, 129–149.
Sinclair, A.J., 1991. A fundamental approach to threshold estimation in exploration geo-
chemistry: probability plots revisited. J. Geochem. Explor. 41, 1–22.
Spalding, R.F., Exner, M.E., 1993. Occurrence of nitrate in groundwater — a review. J. En-
viron. Qual. 22, 392–402.
Titterington, D.M., 1993. A contamination model and resistant estimation within the sam-
pling paradigm. J. R. Stat. Soc. Ser. B Methodol. 55, 817–827.
Wendland, F., Berthold, G., Blum, A., Elsass, P., Fritsche, J.G., Kunkel, R., Wolter, R., 2008.
Derivation of natural background levels and threshold values for groundwater bodies
in the Upper Rhine Valley (France, Switzerland and Germany). Desalination 226,
160–168.
WMO (World Meteorological Organisation), 1997. Comprehensive Assessment of the
Freshwater Resources of the World. WMO, Geneva (34 pp.).