Multiple discriminant analysis of the Drava River alluvial plain

advertisement
Multiple discriminant analysis of the Drava River alluvial plain
sediments
Zoran Peh1 (corresponding author), Robert Šajn2, Josip Halamić1 and Lidija Galović1
Abstract:
Three discriminant function models are raised and cross-compared in order to distinguish
geochemical patterns characteristic for the Drava River floodplain sediments. Based on data
representing total element concentrations in samples collected from alluvium (A), terrace (T),
and unconsolidated bedrock (B) at the border of a floodplain, four element clusters emerged
accounting for discrimination between the referred groups of sediments. The most prominent is
contaminant/carbonate cluster characteristic for alluvium. The other two are: silicate cluster
typical for unconsolidated geological substrate (Neogene sedimentary rocks); and naturally
dispersed heavy metal cluster separating terrace from the former two groups. Models introducing
depth intervals and single profiles as grouping criteria reveal identical sediment-heavy metal
matrices. The second important issue of this paper is possibility of reclassification of samples
originally assigned to one of the a priori defined groups of sediments, based on established
geochemical pattern. The mapped geological units can be reconsidered by the post hoc
assignments to a different group if geological border between alluvium and terrace or between
terrace and bedrock can not be established geologically with absolute certainty.
Key words: floodplain sediments, heavy metals, multiple discriminant analysis, Drava River,
Slovenia, Croatia
1
1
Z. Peh ()  J. Halamić  L. Galović
Croatian Geological Institute,
Sachsova 2, P.O. Box 268, HR-10000 Zagreb, Croatia
E-mail: zpeh@hgi-cgs.hr
Tel.:+385 (1) 6160-753
Fax: +385 (1) 6144-718
2
R. Šajn
Geological Survey of Slovenia,
Dimičeva 14, 1000 Ljubljana, Slovenia
E-mail: robert.sajn@geo-zs.si
1.
INTRODUCTION
The modern floodplains are potential repositories for a wide range of contaminants among
which heavy metals represent by far the greatest threat to human health and environment. The
release of metal-rich waste into the fluvial environment often combines various anthropogenic
sources such as mining operations, industrial and urban wastes, transport, agricultural activity,
deforestation, and others. This impact is recorded in the vertical overbank profiles of river
alluvium, sometimes with dramatic increase of heavy metal concentration, mainly in the upper
part (e.g. Macklin et al. 1994, 2006; Swennen et al. 1994; 1998; Swennen and Van Der Sluys
2002; Langedall and Ottesen 1998; Ciszewski 2002; Bølviken et al. 2004; Pavlovic et al. 2004;
Cappuyns et al. 2006), or, generally, in depth-integrated surficial overbank sediment samples or
topsoil developed on floodplains (e.g. Ottesen et al. 1989; McConnel et al. 1993; Ódor et al.
1997; Rognerud et al. 2000; Bidovec et al. 1999; Halamić et al. 2003; Romic and Romic 2003).
2
However, in longer terms, heavy metals may be also introduced into the soil and hydrosphere by
natural processes – weathering, erosion and transport – easily blending anthropogenic signal with
a natural one and causing the pattern of metal distribution in alluvial plain sediments more
difficult to assess. The problem is magnified by the fact that floodplains join together various
distinct geomorphic and hydrologic features which directly influence their geochemical
signature. It has long been recognized that dispersion patterns, storage and remobilization of
heavy metals in the fluvial system are directly associated with processes of sediment transport
(e.g. Foster and Charlesworth 1996; Macklin et al. 2006). Floods regularly rejuvenate the lowest
extensive land surfaces by lateral and vertical accretion (overbank deposition) of mixed recent
material. Conversely, relatively flat surfaces, or terraces, lying above the modern floodplain may
escape inundation and thus (largely) preserve the geochemical signature from the time of their
origin. Since these are no longer in direct relation to recent hydrological conditions (especially
flood events), distinguishing between alluvium and terrace may be focused on difference
between “current” and “historic” geochemical patterns which may be useful in mapping young
terraces close in altitude to the modern floodplain but showing no visible scarp. This is
particularly important since various studies show that, in general, variations in the heavy metal
concentration across terraced floodplains are more strongly associated with terrace height than
with terrace age (Lambert and Walling 1987: Brewer and Taylor 1997). By the same token
terraces can be contrasted with bedrock if boundary between floodplain and upland is difficult to
trace due to unconsolidated material and poorly dissected relief which indicate the latter. In this
case different geochemical signals can be expected on account of diverse nature of the geological
substrate: overbank sediment filling the alluvial plain versus soil developed on unconsolidated
bedrock. It must be also borne in mind that during dry times floodplain sediments, especially on
higher terrace levels, are subject to various external influences including soil formation processes
which can substantially alter element concentrations (Volden et al. 1997; Rognerud et al. 2000).
3
The general aim of this research was to find geochemical contrast between the “recent” and
“historic” sedimentary material (alluvium and terrace) deposited on alluvial plain of a great river
during its long fluvial history, and both against the older, Upper Miocene/Pliocene weakly
consolidated sedimentary rocks (bedrock) underlying the floodplain. In this light the presented
study can be viewed as a multivariate statistics variant of the geomorphological-geochemical
approach to provenance analysis, based on the chemical composition of Holocene-Pleistocene
fluvial sediments as the main carriers of the heavy metal load. The study draws on the premise
that chemical (and physical) properties of the sediment can be used to infer its provenance
although the applied method is based on indirect observation rather than direct comparison of
floodplain sediments with individual potential sources as in the case of classical fingerprinting
technique using multiple discriminant functions (e.g. Molinaroli and Basu 1993: Collins et al.
1997; 1998; Owens et al. 1999; Bottrill et al. 1999). This approach is deemed appropriate due to
different age of floodplain sediments and the lack of precise terrace chronology. Again, it differs
from the line using direct comparison of a number of well defined predictors (diagnostic
properties) with defined hierarchical relationship (e.g. lithology-mineralogy-geochemistry) such
as in canonical correlation analysis where each element, mineral and lithologic unit represents a
family of fingerprint properties (Čović 2003; unpublished MSc thesis). A multivariate statistical
assessment of primary geochemical pattern distinguishing between a priori defined groups of
lithological units (“lithology”) – alluvium, terrace and bedrock – was performed preferring
multiple discriminant analysis (MDA) as the most helpful mathematical tool. Additional
discriminant models were constructed in order to examine the influence of depth (“depthlithology”) and local variations (“profile”) on primary geochemical signature defined by
distinctive variable clusters, especially the families of heavy metals.
2.
DESCRIPTION OF THE STUDY AREA
4
The study area lies between 14˚55’ and 16˚55’ E longitude and 46˚13 and 46˚38’N latitude.
It occupies the large portion of the Drava River drainage area spreading from lead-zinc mining
and smelting region in northeastern Slovenia (Mežica) to the mostly agricultural lowlands
(Drava alluvial plain) of northern Croatia (Fig. 1). The upstream, Slovenian, part is built
predominantly of igneous and metamorphic rocks (Palaeozoic) with rare occurrences of
sedimentary rocks (mostly Mesozoic) (e.g Mioč and Žnidarčič 1977) while downstream, towards
Croatia, the terrain is geologically substantially different, being composed almost solely of
sedimentary rocks of the Tertiary and Quaternary age (e.g Mioč and Marković 1998). The
youngest, alluvial, sediments along the Drava River are notorious for their increased heavy metal
composition which may be either naturally dispersed by Drava tributaries draining the Pb-Zn
(Cu, As, Cd) mineral occurrences and deposits hosted in Triassic sedimentary rocks (mostly
limestones and dolomites) and Palaeozoic igneous-metamorphic (Palaeozoic) complex, or may
draw its origin from anthropogenic input. The latter case is of special concern for the study area
because heavy metals are released into the Drava fluvial system for a long time being by erosion
of the mine spoils, tailings and contaminated topsoil (Mežica) or in the forms of contaminated
aqueous effluents from the metal processing facilities situated in the upstream reaches of the
Drava River. Diffuse inputs of airborne metal particulates from the furnace stacks (e.g., Prevalje,
Ravne, Kidričevo) can also contribute to overall pollution (Šajn 2002; 2003; Vreča et al. 2001;
Šajn and Gosar, 2004).
3. FIELD AND ANALYTICAL PROCEDURES
3.1 Sampling
During earlier investigations carried out for the purpose of the Basic Geochemical Maps of
Slovenia and Croatia, anomalous values for Pb, Zn, As, and Cd were obtained from the chemical
5
analysis of soil samples (topsoil A0-25) collected in the area encompassing Drava River valley
and its tributary valleys (Andjelov 1993; Šajn et al. 2005; Miko et al. 2001; Halamić et al. 2003;
2005). In order to carry the research further for origin and distribution of anomalous elements six
profiles lines in Slovenia (Dravograd – DR; Mežica – ME, Radlje – RA; Maribor – MB; Ptuj –
PT and Zlatoličje – ZL) and three in Croatia (Sračinec – SR; Belica – BE and Legrad – LE) in a
downstream sequence were sampled more or less perpendicularly to the river course before the
confluence with the Mura River. The tenth profile line Ormož (OR), placed between ZL and SR
was common for both Slovenian and Croatian part and was sampled separately by each research
team (Fig. 1). Sampling was carried out according to recommendations of the IGCP and
FOREGS task group for the geochemical mapping (Darnley et al. 1995; Salminen et al. 1998).
The total of 47 boreholes (individual profile points) of various depths was drilled along the
selected profile lines providing as much as 333 samples for chemical analysis. Boreholes were
located in a pre-designed scheme set across the river valley to include each terrace level on the
both sides of the river bed and alluvial sediments in the middle. The peripheral boreholes were
located on the bedrock. As the hand-operated drilling device could not go through gravel drilling
was halted as soon as the first thick gravel layer was reached. The samples were taken at 20 cm
intervals as a composite, while the maximum depth varied between 60 cm (PT) and 260 cm
(DR).
3.2 Sample preparation and analysis
Samples were air-dried at temperature <400C for approximately three months to prevent the
loss of volatile components and then sieved to the <0.125 mm fraction and finally homogenized
in agate mortar. The <0.125 mm fraction was used following some earlier investigations which
have shown that high content of most elements, particularly the trace elements, is present in the
fine-grained fraction (e.g. Whitney 1975; Gibbs 1977; Förstner and Whitmann 1981; Foster and
6
Charlesworth 1996; Sutherland 1998; Singh et al. 1999; Rhoads and Cahill 1999). Also, this
fraction usually contributes to more than 95% of the particles in most samples (Swennen et al.
1998).
All analytical work was carried out at the ACME Analytical Laboratories in Vancouver
(Canada). Samples were analyzed by inductively-coupled plasma mass spectrometry (ICP-MS)
after total hot 4-acid (HCl-HNO3-HF-HClO4) digestion at temperature of 200C on the set of 41
elements. The digestion was only partial for some Cr and Ba minerals and some Al, Hf, Mn, Sn,
Ta and Zr oxides. The volatilization during fuming may have also resulted in some loss of As, Sb
and Au. Mercury was determined with cold vapour atomic absorption spectrometry (CV-AAS)
after aqua regia digestion.
3.3 Accuracy and precision
Accuracy was controlled by certified geological reference materials (DST6/DS6 – ACME
Laboratory). For most elements analyzed in reference soil materials accuracy was found in the
range of ±10% of the certified values, except for Ag (±25%). Precision was determined by
repeated analysis of both certified reference samples and randomly selected soil samples (16
duplicated samples). The resulting coefficient of variation was, on average, approximately 5%.
4.
DATA PROCESSING
4.1 Univariate statistics and data transformation
A suite of 27 elements including 8 major and 19 trace elements were selected as predictor
variables in computing discriminant functions. Summary statistical results for the whole dataset
prior to the multivariate statistical procedure are given in Table 1 (minimum, maximum, median,
mean, standard deviation and skewness). Taking into account that most variables are
7
characterized by non-normal, mostly positively skewed frequency distributions (except Al and
Na which are negatively skewed), appropriate transformations were found necessary to get more
symmetric distribution for most elements. As a general rule, most geochemical and
environmental data do not follow normal (neither lognormal) distribution (e.g. Matschullat et al.
2000; Reimann and Filzmoser 2000; Reimann et al. 2005), which is the outcome of complex
non-linear dynamics, feedbacks and thresholds resulting in outliers within investigated system
(sample media such as, for example, soils, stream- or overbank sediments) represented by the
particular set of variables (e.g. Hugget 1998; Phillips 1999). To stabilize their variance the usual
remedy in environmental and geochemical exploration was applied in this work represented by
conventional normalization procedure using log-, ln-, and sqrt-transformations, which included
all elements except Al and K. Distributional characteristics of the input data were examined by
the test of normality (Table 1). According to the applied Shapiro-Wilk’s (S-W) statistical test
only two original variables exhibit normal distribution (p>0.05, marked by the asterisk). Even
after the different transformation methods were used to approach the normal distribution,
normalization procedure failed to improve normality of a number of elements. In the case where
applied transformation produced even greater skew than original (such as with V) the data were
rather left untransformed.
4.2 The Strategy of Analysis
Multiple (multi-group) discrimination analysis (MDA) can be particularly useful
multivariate method when applied to the data scattered within and across various spatial
boundaries such as geological (lithological) units, depth intervals, or profile lines, being
composed of the same suite of observed attributes. Its primary purpose is to establish the major
sources of between-group differences (Dillon and Goldstein 1984; Rock, 1988) often
represented, in environmental studies, by accumulation of heavy metals in various geologic
8
(sampling) media distributed horizontally and vertically from the focal point, or line of
dispersion.
Data collected from a number of profile lines set across the investigated part of the Drava
valley can be organized in several ways as the grouping criteria can follow different conceptions.
The most obvious approach to grouping is based on the underlying (mostly Quaternary) geology
as well as on the depth intervals from which the samples were collected, after which the two
types of divisions (models) can be formed, namely LITHOLOGY and DEPTH. The first of such
“a priori” groupings can be accepted as most “genuine”, so that basic differences between the
groups can be easily inspected provided that the data variability is great enough to disclose them.
It must be noted, however, that sampling procedure skipping the natural boundaries between the
soil horizons may decrease the between- against within-group variability rendering the
geochemical depth differences more difficult to understand. The “noise” can be partly reduced
by combining the two divisions in such a way that newly formed groups more clearly indicate
differences between the top and the deeper intervals, for example DEPTH-LITHOLOGY. This
approach is introduced into analysis to dispel ambiguity between upper intervals sampled in
areas further from the most possible source of pollution on the one side and deeper intervals
nearer to water course exposed to recurrent flooding on the other.
However, with groups thus formed their mutual relationship can not be represented
meaningfully on relevant field maps since there is not enough mapping points to lay a grid. Also,
the profile data are arranged in intervals rendering their representation, unless portrayed in a
cross-section, largely abstract. To avoid this problem and, possibly, to focus on a narrower stand
such as the single profile, the groups can be reshaped by combining single profiles (sampling
points, or boreholes) with LITHOLOGY or DEPTH, or both. Thus, each profile point can be
portrayed as a group of its own, its intervals representing objects (samples). By this
rearrangement, the new groups can be created (PROFILE) following the adjusted grouping
9
criterion and represented on a map via the relevant profile locations. However, being arbitrarily
created, these groups can introduce a sort of (additional) artificiality into original data. In such a
case interpretation must be exercised with great care since differences between larger groups,
following the criteria of LITHOLOGY or DEPTH, can be easily offset by the variations within
the smaller ones, localized in individual profiles.
More than one MDA can be independently performed following the modes of data grouping
discussed above. First and foremost, the data are arranged in such a way that sources of variation
between the geochemistry of underlying lithological units (LITHOLOGY) – alluvium (A),
terrace (T) and bedrock (B) – can be examined (3 groups). The second grouping criterion deals
with differences between combination of lithological units and three depth intervals (top, middle
and bottom) at which the samples were collected (DEPTH-LITHOLOGY) – including 9 groups
altogether which are composed of 20 cm intervals (0-20, 20-40, and 40 to the bottom). The third
model is a depth-profile combination which is extended to all three lithological units and
contains 47 groups of profiles (PROFILE). This last approach was designed to highlight the
possible variations within the single profiles. The accuracy of the group affiliation can be later
easily examined by comparison of the “a priori” (observed) and mathematically computed (post
hoc) classification for each object. In all three cases the same set of objects (N=333) and
variables (p=27) is utilized.
5. RESULTS AND DISCUSSION
The results of the MDA are briefly summarized in the joint table (STATISTICA, Version 6)
(Table 3) comprising all three of the a priori built exploratory models (cases). Before that the
overall significance of their discrimination is tested by the appropriate multivariate tests (Table
2) divulging the vanishingly low associated probabilities (p<0.000), a prerequisite to safely
10
proceed with computing discriminant functions (DF). According to different number of groups
(K) in various models the total number of DFs is K-1 or, as with the PROFILE model where the
number of groups (K=47) is greater than number variables (p=27), it can not exceed the latter (p1). Irrespective of statistical significance of the variation between the observed groups (p-level in
Table 2) which defines dimensionality of the discriminant space, the smaller number of DFs is
used to explain the natural variation between groups. The natural variation hidden behind the
original data actually served as an effective parsimony criterion reducing the number of
discriminant axes to only two or three. Note, however, that in the LITHOLOGY model only two
DFs exist (K-1) which completely explains the difference between the three observed groups.
5.1 Labelling discriminant space
Labelling DFs is of paramount significance in MDA as it opens the door to unraveling the
hidden relationships between the grouped data. Thus the natural processes can be deduced
underlying the mathematical structure of geochemical observations. In this work discriminant
loadings (structure coefficients) representing the simple correlation of a variable with particular
DF are used to assess individual contribution of each descriptor variable (chemical element) to
the overall discrimination between the groups. On that premise, a number of variables that little
contribute to discrimination can be omitted from interpretation in all three models – those with
small discriminant loadings such as, for example, As, Ba, Co, Cr, Cu, Fe, Hg, La, P, Sc, Th, Ti,
V – being clustered around the intersection of discriminant axes, which is particularly evident in
the PROFILE model (Figs. 2a-c and 3a-b). Variable diagrams provide the quickest and most
informative insight into the structure of discriminant space suggesting which variables
(descriptors) would be retained to explain the meaning of discriminant axes as well as which
subset of variables most clearly separates the a priori defined groups. Further, association among
descriptor variables and related groups is best represented geometrically, viewing the DFs as
11
axes in the reduced discriminant space. However, the variable and group diagrams can not be
compared directly as different scales are used in both cases. In the scatterplot of variable
loadings (Figs. 2a-c; 3a-b) discriminant axes are drawn as normalized vectors, while in the
scatterplot of canonical means (group centroids) (Figs. 2d-f; 3c-d) and individual objects
(samples) (Fig. 4) these are defined as discriminant score vectors. Thus, interdependence
between variables and groups should always be considered using their shared position along the
appropriate axis. The points placed close to axes intersection (main centroid) play a little or no
part whatsoever in discrimination.
Scatterplots of variable loadings and group centroids are constructed for all three investigated
models applying only those DFs that add most to the total explanation of the between-group
differences (Table 3). Three DFs satisfactorily separate among the predefined groups while the
remainder is discounted, excluding the first model (LITHOLOGY) where only two exist. In the
second model (DEPTH-LITHOLOGY) the first three explain almost 89% of the variation
between the groups, while in the third model (PROFILE) explanation is reduced to the still high
71% of the total variability. Models are compared using the multiple scatter diagrams of the pairs
of axes DF1 and DF2 (Fig. 2) and DF1 and DF3 (Fig. 3).
LITHOLOGY model
Evaluating DF1-DF2 scatterplots on Figure 2, and again DF1-DF3 scatterplots on Figure 3,
essentially the same pattern emerges in all three models, based on the polar (mutually exclusive)
relationship between the variable clusters. To begin with the LITHOLOGY model, the first
discriminant axis is bipolar and can be interpreted as reflecting inverse relationship between the
suite of elements representing the carbonate composition (Ca, Mg, Sr) combined with strong
anthropogenic input (Pb, Zn, Cd, Mo) against another suite of elements representing the silicaterelated composition (Zr, Al, K) of the investigated groups. In this model a close inspection into
12
the related variable and group scatterplots (Figs. 2a against 2d) reveals that this pattern closely
separates alluvium (A) from bedrock (B), while the terrace (T), on the other side, is fixed in the
middle position conveying information of the average geochemical composition in comparison
with the former two groups. Geologically, bedrock as the oldest sampled material widely ranging
from Helvetian (Miocene) to Quaternary and consisting mostly of sands, sandy marls and marls
(Miocene-Pliocene), aeolian sands (Pleistocene), loess and diluvium (Holocene), is typically
alkaline in composition with addition of zirconium (clays and sands). This material definitely
reveals the natural signal from older (felsic) rocks such as gneisses and tonalites outcropping in
the Pohorje Mt. in Slovenia. It stands in sharp contrast to alluvium (A) which is deposited during
comparatively recent geological time indicating a quite different source of parent material
(mostly the Miocene and Pleistocene marls) as well as strong pollution from metal mining,
processing facilities and industry along the Drava river and its tributaries (Mežica situated on the
Meža River, Dravograd on the Drava River, and others) (Fig. 1). Thus the high positive
discriminant scores for some bedrock samples (Fig. 4) indicate the sedimentary material, in this
case Upper Pontian sands, and deluvium, respectively, highly enriched in (Zr, Al, K) suite of
elements, while in the same time containing the lowest concentrations of carbonate material and
accompanying heavy metal suite. On the other side of the scale are the samples with the high
negative discriminant scores, allocated to alluvium and indicating the inverse variable-group
relationship with respect to the former case. These pertain to the parts of alluvial plain more
affected with modern pollution although not necessarily in the uppermost depth interval.
The relative abundance of carbonate fraction in alluvial sediments can be sought chiefly in the
recurrent physical erosion and transport preventing pedogenesis due to effective mixing and
homogenization of the recently deposited material (derived mostly from Tertiary carbonate-rich
sedimentary rocks traversed by numerous Drava tributaries downstream of Maribor). By
complementary positions of the carbonate-related/contaminant and silicate-related clusters it is
13
highlighted both that soil forming processes are complete on much older and consolidated
sedimentary material, and that heavy metals, usually associated with smaller grain-size particles,
conversely, bind to carbonate phases indicating their present-day, anthropogenic ingress into
alluvial sediment (Foster and Charlesworth 1996). Various studies show that occurrence of
heavy metals in easily soluble phases such as carbonate fraction would indicate definite
anthropogenic influence (Fernandes 1997; Abd El-Azim and El-Moselhy 2005; Cappuyns et al.
2006), particularly in cases when this fraction comprises greater part of the non-geogenic heavy
metal content in the contaminated sediment (Singh et al. 1999). Possible mechanism of pollutant
trapping in the alluvial sediment can be either forming of the various carbonate complexes, or
adsorption on the calcite surface, while Mg and Sr are usually incorporated into the calcite
structure (Zachara et al. 1991; Ettler et al. 2006). Recent investigations in the Drava valley near
Varaždin (Marković 2007; unpublished PhD thesis) (Fig. 1) show that underground waters of the
Varaždin aquifer are highly saturated with calcite and oxygen during the most part of the
hydrological year causing that heavy metals with affinity for carbonate phase (Pb, Zn, Cd)
precipitate together with calcite in the form of coatings. In process terms, the bipolar character of
DF1 axis reflects opposed mobile/immobile nature of element clusters discriminating between
the investigated groups. In contrast to alluvium (A), bedrock (B) is composed primarily of
silicate and oxide minerals being more resistant to weathering which is why their elemental
constituents are far less mobile than those bound to carbonate phases (McMartin et al. 2002).
The second discriminant axis DF2 accounts for barely 21 % of the difference between groups. It
is also bipolar and is largely concerned with an association between the heavy metals such that
Pb, Zn, Cd and Mo form one set of variables operating against Ni, Cr, Fe, Cu, Al in the other set.
The contribution of the carbonate-related cluster to overall discrimination in this case is
negligible. This relationship represents a functional model further clarifying the position of the
terrace (T) group in the discriminant space in that it is separated from both alluvium (A) and
14
bedrock (B) based on the elevated contents of heavy metals such as Ni and Cr accompanied with
Al. This may be caused by complex interactions, among which provenance, hydrodynamic
conditions and depositional environment are the most important factors. Assuming the pristine
environment in the pre-industrial, terrace-forming period, the discriminating trace element
content in the terrace samples must have originated in (Ni-Cr-Cu-V)-bearing mafic rocks
traversed by the Drava system in Slovenia and Austria. Accordingly, DF2 can be provisionally
labelled as the heavy metal axis discriminating between recent-pollutant and historic-natural
(lithogenic) heavy metal content in examined profiles. Characteristically, the samples with high
negative discriminant scores, assigned to T are clustered furthest downstram (focused on LE
profile line) indicating the area occupied by widest terrace expanses in the investigated area.
Parts of the terrace material was probably eroded and re-deposited, and following the changes in
depositional environment during the Quaternary (terrace uplift and cutting with periodical water
table changes) eventually assumed geochemical signature which is more difficult to trace. Much
lower explanatory potential of the second axis DF2 as well as a general failure of the
LITHOLOGY model to clearly (or even fairly) separate between carbonate-related and pollutant
variable clusters purports the assumption.
Anthropogenic/carbonate versus lithogenic heavy metal signature of DF2 established in this
work is indirectly corroborated by the recent investigations in the Alpine-provenance
neighborhood (northern Italy) (Amorosi and Summartino 2007). It is found that mutually
exclusive (inverse) relationship between Ca-Sr (carbonate-related anthropogenic heavy metal
input not evaluated) and Ni-Cr (naturally dispersed) sets of elements differentiates the modern
alluvial deposits of Apennine provenance from formerly active but now abandoned delta lobe of
the Po River. These results establish the Ni-Cr combination as effective provenance indicator for
Alpine mafic source rocks.
15
DEPTH-LITHOLOGY model
The second model considered (DEPTH-LITHOLOGY) reflects essentially the same functional
relationship as in the former case (Figs. 2b, e). Interpretation of the first two DFs is, however,
much clearer, being reduced to discrimination between lithologic units (DF1) and depth (DF2).
As the nine groups (K=9) combining lithology types and depth intervals (from a-A to c-B) are
included into analysis, eight discriminant functions in total are available for interpretation. The
first discriminant axis DF1 accounts for more than 52 percent of the variability between the
groups (Table 3). It is distinguished by the same variable relationship as in the previous model
weighing Ca-Mg-Sr and Pb-Zn-Cd-Mo variable clusters against the single Zr-Al-K cluster.
Visual inspection of the discriminant space shows that DF1 is the primary source of difference
between the sediment groups (A-T-B) and can be interpreted in the same way as DF1 in the
LITHOLOGY model. The second axis DF2, accounting for additional 22.4% of the variability,
further clarifies the relationship between the intervals a, b and c. Visual comparison between the
scatterplots of variables and group centroids is straightforward, reducing almost altogether the
possibility of misinterpretation because the groups a priori defined as combination of
“lithologies” and “depths” can be easily delineated as compact and distinct units in discriminant
space (Figs. 2b, e). In contradistinction to the LITHOLOGY model where DF2 is concerned with
separation between lithology types, in this case it allows each sediment group to be clearly
structured according to the depth. Typically, individual intervals do not mix together but retain
their “normal” position in a functional model (a over b over c) in relation to decreasing influence
of pollutant Pb-Cd-Zn-Mo cluster through carbonate-related Ca-Mg-Sr cluster to Ni-(Fe)-(Th)
cluster. Also, by the group cross-comparison (Fig. 2e) it is obvious that groups are self-similar in
the sense that a-interval in alluvium (a-A) is most similar to a-interval in terrace (a-T) which in
turn is most similar to a-interval in bedrock (a-B). The same is also valid for b- and c-intervals.
Nevertheless, however informative the diagrams can be on the mutual group relationships, the
16
best way to assess how much the group centroids actually differ between themselves is to take a
closer look into the distance matrix (matrix of squared Mahalanobis distances) (Table 4).
Distances between a- and b-intervals increase from alluvium (most similar) through terrace to
bedrock (most differing). It may reflect (cf. Figs. 2b and e) reducing impact of the pollutant
cluster (Cd-Pb-Zn-Mo) on b-intervals (20-40 cm) in terrace and bedrock samples.
Simultaneously, the variation between the uppermost a-interval (0-20 cm) and deepest cintervals (deeper than 40 cm) is the greatest in bedrock (Table 4) indicating that contamination,
if present, does not affect the lower portions of bedrock profiles. The lowest, c-intervals, in both
bedrock and terrace samples show higher contents of Ni. Interestingly, only alluvial c-intervals
are enriched in carbonate-related cluster (Ca-Mg-Sr) indicating that alluvium is strongly
influenced by the process of leaching. This may be important because of potential release and
remobilization of heavy metals bound to carbonate particles in the upper part of the sediment
column where anoxic or low pH conditions occur (Foster and Charlesworth 1996). In general, aintervals tend to be enriched in pollutants reducing their impact from alluvium to bedrock, while
c-intervals (alluvial excluding) are typical for their Ni-(Fe)-(Th) content with increasing impact
of siliceous Zr-(Al-K) composition from terrace to bedrock (clay component). Besides, a slight
tilt of the three combined “lithology-depth” groups towards the upper right quarter of the group
scatterplot (perpendicular to the line of carbonate/contaminant enrichment) (Fig. 2e) can be
interpreted as an effect of general decrease in pollution from a-A to c-B.
PROFILE model
The third analyzed model (PROFILE) deals with differences between 47 individual profiles in
the study area independently of their direct allocation to one of the a priori defined lithology or
depth groups. In this case the first two discriminant functions, DF1 and DF2, explain a large
portion of the total variance (45% and 15%, respectively). In contrast to the former two cases
17
(LITHOLOGY and DEPTH-LITHOLOGY) with DF1 reflecting basically the same functional
model, DF2 is here primarily concerned with differences within the Ca-Mg-Sr carbonate-related
cluster. An examination of diagram 2c reveals negative association between Ca and Mg on the
one side and Sr on the other. However, this relationship is geochemically not straightforward
because Sr is usually strongly (positively) associated with Ca which is primarily indicative of
calcareous rocks and its weathering products particularly in combination with Mg (DF1). It must
be taken into account that investigated profiles are sampled in various sedimentary material
(alluvium, terrace and bedrock), and in various depths. Thus it is clear that possible mechanism
of this relationship can be elucidated only after the closer look is taken into the scatterplot of
group centroids (Fig. 2f) revealing the fact already observed in relation with differences between
the lithology groups (Figs. 2a and d). The profile PT-0 sampled in the bedrock material (sandy
marls) is characterized by the highest concentration of Sr (up to 387 mg kg-1) accompanied with
the lowest concentrations of Ca and Mg (0.18 g kg-1 and 0.25 g kg-1, respectively) (Table 3). In
sedimentary processes, distribution of Sr is affected by strong adsorption on clay minerals but Sr
may be also contained in lithic fragments and detrital feldspars, particularly in lower parts of the
examined bedrock profiles (c-intervals) (De Vos et al. 2006). Ca and Mg, on the other side, are
typically enriched in alluvium (ME-1) due to the constant accumulation and removal of the bedload and suspended material preventing the soil forming processes such as eluviation and
illuviation to fully develop (Fig. 2f).
Reverting to the much simpler scheme conveyed by DF1 it is obvious that quite a few profiles
are distinctly separated by the carbonate (contaminant)-silicate axis (Fig. 2f). Hardly a single
group can be clearly recognized in the central cloud of profiles due to the very weak discriminant
loadings characterizing the siliceous variable cluster. This picture should be interpreted rather as
negative image (scarcity) of the contents characterizing the opposite (negative) pole of DF1 axis,
18
that is, abundance of pollutants bound to carbonate material. A number of profiles such as, for
example, PT-1, OR4A, SR-6 and OR-4, fit neatly into the picture (Fig. 2f).
The two latter models can be further explained introducing additional discriminant functions
into consideration whose explanatory potential, however, is decreasing rapidly. It is often the
case in MDA that statistical significance seems a lesser issue in evaluation of the validity of
results since geological interpretability of successive roots always questions the purpose of
carrying on the experiment further. In the DEPTH-LITHOLOGY and PROFILE models it is the
third discriminant axis that still carries a considerable portion of the total variability (14% and
11%, respectively) (Table 3) but it may not be easy to interpret because, in spite of the noticeable
division of variable clusters, the associated discriminant loadings are rather low to indicate
unambiguously a geological process involved. In the DEPTH-LITHOLOGY model DF3 is
monopolar (Fig. 3a), resembling almost exactly DF2 (Fig. 2b) with positive pole removed. Both
Ca-Mg-Sr and Cd-Zn-Pb-Mo variable clusters do not really contribute to discrimination so that
overall knowledge of DF3 depends entirely on Ni-Al-(Cr-Cu-As-Fe) variable cluster. This can be
interpreted as natural distribution of elements derived from mafic rocks as their principal source.
Cross-comparison with group centroid scatterplot (Fig. 3c) reveals terraces as their main carrier
but only in their topmost interval (a-T). There is decreasing succession from deeper intervals (c)
to the top, which pattern is continuing into alluvium. In bedrock, it is almost absent. By contrast
with the foregoing, in the PROFILE model DF3 is functionally entirely different involving
another two variable clusters in discrimination between groups (profiles) (Figs. 3b and d).
Although separation of the variable clusters is rather vague due to the weak discriminant
loadings DF3 introduces a worthwhile component into analysis which failed to manifest in
previous two models, namely distinction between the pollutant and carbonate-related element
clusters. The profile MB-7 sampled on bedrock is the single group separated in this case. Its
solitary position in the group scatterplot (Fig. 3d) is, however, more due to very low content of
19
carbonate material (probably leached) than to enrichment in pollutants. MB-7 is farthest from the
profile groups both characterized by elevated content of contaminants such as PT-1 or OR-4A,
and particularly those enriched in carbonates such as OR-5, OR-4 or ZL-1.
5.2 Classification issues
Discrimination of groups based on the a priori knowledge of their geochemical
(environmental) attributes provides the possibility of their consequent assignment to the most
probable group. In all investigated models these assignments are founded on geochemical
composition of samples irrespective of the lithology type, depth of sampling, or selected
location. Integrity of previously defined “natural” groups can be further examined weighing the
mathematically predicted (computed) assignments against the original (observed) classifications
while efficiency of classification is often used as a very helpful tool in later geological
considerations because it can say a lot about the data structure describing the media sampled,
particularly if criteria for the a priori classifications were based on similar features. If all three
models are compared (Tables 5 to 7) it is apparent that the PROFILE model containing 47
groups is arranged with the highest efficiency – more than 95.5 % of all profiles are correctly
classified. Statistically, profiles differ between themselves significantly in chemical composition,
allowing most samples to fit unequivocally in their pre-assigned groups (Table 7). Geologically,
however, it is hard to find a rationale for discrimination between the profiles due to rapidly
decreasing explanatory potential of successive discriminant functions (Table 2) and relevant
variable contributions (discriminant loadings). Quite a few groups are separated by each
subsequent DF from the compact cloud around the main centroid (Figs. 2f and 3d), strongly
suggesting a number of local factors or unidentified sources of variability operating at the
sampling area and affecting particular groups at the spot.
20
In general, only one interval per group is misclassified (mostly a) except for LE-4 (a, b). Among
15 misclassified samples most of them are confined locally to the Legrad (LE) profile line or
super-group (6 altogether) implying their similarity to some other group: LE-1 to BE-6; LE-2 to
SR-5; LE-4 to BE-3 (two intervals); LE-5 to LE-6 and LE-5A to OR-2 (Table 8). Since Legrad
profile line was sampled furthest downstream in the investigated area this may indicate that
sediment derived from upper reaches of the Drava valley is simply reworked and mixed
downstream, or its heavy metal constituents are chemically remobilized due to changed pH and
redox conditions stripping LE-groups (on their related sampling levels – alluvium, terrace or
bedrock) of its geochemical “identity” (Fig. 5). Examination of the misclassified groups further
reveals that only a minor part of the observed groups changed their affiliation to originally
defined lithology type.
As contrasted with the PROFILE model the former two models, although more directly
related to geological aspects as the grouping criteria, are less efficiently classified. Inspection
into classification matrices (Tables 5 and 6) reveals that 78.1% and 70.3% samples, respectively,
are correctly classified. One must bear in mind that rows in classification matrix always relate to
the actual group membership whereas columns give the predicted group membership leaving the
correct classifications in the main diagonal. However, of necessity, classification matrix for the
PROFILE model (Table 7) must have been simplified due to a great number of groups.
Classification results are important, providing the final test of discriminant analysis (Rock 1988)
because mathematically computed (predicted) sample and group assignments assessed from
geochemical data are compared with original a priori classifications. Results show that even for
the simplest model consisting of only three groups (LITHOLOGY) the overlapping is large (Fig.
4), while introducing a new criterion (depth) that would hopefully remove the confusion only
adds more to it. Bedrock (B) is the group with the highest classification efficiency (93.4%)
losing the least of its samples to other two lithology groups. Geochemically, bedrock is the most
21
compact group with only 4 misclassified samples (lost to terrace). No bedrock sample is
confused with alluvium, however. Similarly, alluvium which is farthest from the bedrock (Table
4) overlaps considerably only with terrace decreasing its assignment efficiency down to 77.7%,
which is still relatively high (Fig. 4). Geologically, misclassifications in the LITHOLOGY model
may raise some doubts in the accurate position of some youngest (Quaternary) units mapped as
alluvium or terrace although geochemistry was not the criterion included in the geological
mapping. For example, several alluvium-allocated profiles are completely (all depth intervals) or
mostly misclassified and look more akin to terrace after their geochemical composition. These
include DR-6, DR-4, RA-1 and MB-3 (Fig. 5). On the other side, there are a priori classified
terrace-allocated profiles which should be properly assigned to alluvium according to their
geochemistry (enrichment by heavy metals), such as OR-5 (entirely) and LE-3 (for its most part).
Finally, there is a question of terrace-bedrock samples affiliation. Two of the terrace profiles are
also completely misclassified, namely SR-2 and SR-3 with all the samples lost to bedrock. Now,
if all samples pertaining to individual profile exchange their group “membership” the idea of
misclassification is clear enough because all samples are taken into analysis individually (each
depth interval per se). However, the problems arise if a smaller number of samples from the
same profile is wrongly classified, a case such as PT-5 where the three deepest bedrock intervals
(of seven altogether) are reclassified as terrace while upper intervals remain in the original class.
Obviously, the converse is more logical in the geological sense, though it is an isolated example
keeping in mind that classification efficiency for bedrock samples exceeds 93% while terrace is
the most “indistinct” group (Table 5) whose classification rate is only 73% (which, nevertheless,
fares quite well). Partly to avoid possible indeterminacy of this kind and better relate the mapped
units to chemical and other processes in the vertical profile of each sampled profile the DEPTHLITHOLOGY model was introduced into analysis. The primary idea was to reduce the “noise”
created by mixing of same intervals with various sediment groups (units). This, however, helped
22
as much as hindered better classification efficacy, most probably because certain artificial noise
was introduced by the fact that depth intervals (a, b and c) were not based on natural boundaries
(as horizons in a pedologic profile). Nonetheless, it is easy to see that bedrock depth intervals
maintain the highest classification accuracy (Table 6), with overall 75.4% success rate, while
correct assignment of c-intervals (c-B) is even greater, surpassing 82%. The c-intervals are in
general classified most correctly indicating that deeper below the surface (>40 cm) lithology
units (A-T-B) are contrasted considerably between themselves in chemical composition
(between group differences are much greater than within group differences). Typically, incorrect
c-intervals assignment appears between alluvium and terrace, while bedrock c-intervals are
distinctly different. Closer to the surface, predominantly in b-interval (20-40 cm), confusion is
the greatest. For example, only 40% of b-intervals (Table 6) in alluvium are assigned to their
original groups, others being allocated to other alluvium depth intervals, or even to terrace. It is
clear from the table of classification efficiency (Table 9) that introduction of depth into analysis
lays high amount of probability on inaccurate a priori classification of SR-2 and SR-3 profiles
into the original terrace groups. In both LITHOLOGY and DEPTH-LITHOLOGY models these
were unequivocally post hoc assigned to the bedrock group (Upper Pontian sands and marls). On
the other side, DR-6 and OR-5 profiles originally (miss)classified as alluvium and reclassified
100% as terrace in the LITHOLOGY model regain their original group membership as alluvium
in the DEPTH-LITHOLOGY model (Fig. 5). The latter is drastic example of how bringing
additional grouping criterion into the analysis can alter the previously established “correct”
classification. As the classification results should always be scrutinized geologically this doublecheck must speak in favor of the “depth-sediment” criterion despite the lack of the precisely
defined (soil) horizons with associated variety of soil processes wherever these affected the
floodplain sediments. The profiles such as DR-4 and LE-3 are also in both models greatly a
priori misclassified as alluvium (DR-4) and terrace (LE-3), respectively, and can easily switch
23
their original alluvium-terrace assignments (Table 9). A few other profiles such as SR-5 (terrace
to alluvium), RA-1 (alluvium to terrace) MB-3 (alluvium to terrace), PT-5 (bedrock to terrace)
are possible suspects as well.
6.
CONCLUSIONS
The three MDA models were constructed in order to distinguish geochemically between adjacent
sedimentary facies (“lithologies”) laterally spreading from the river course outward into the
floodplain. Considerable waterborne pollution in alluvial sediments caused by heavy metals from
the mining/smelting areas in the upper reaches of the Drava River had been already known as a
result of recurrent inundations during recent sedimentary history. Pollution was suspected, to
some extent, also on the river terraces (diffuse airborne contamination from metal processing
facilities) further away from the river course but it was, generally, not expected in the older,
Neogene sediments (bedrock) outcropping at the perimeter of the Drava valley away from
anthropogenic point sources. The highest contamination was also expected in the surficial depth
intervals of all sampled lithologies. In all cases (including the single profiles) this variation
should have been reflected in good separation according to a priori defined grouping criteria –
LITHOLOGY, DEPTH-LITHOLOGY and PROFILE.
Based on the above premises the designed models were actually aimed to differentiate between
natural and anthropogenic patterns (based on total element concentrations) counting on the
maximum contrast between alluvium (recent) and bedrock (Neogene) as the most conspicuous
one. The terraces, on the other side, as the intermediate (Pleistocene – Holocene) member
deposited in apparently pristine setting should reflect the natural signal controlled primarily by
geological substrate. Their geochemical signature, however, must draw its origin, at least partly,
from areas drained by the old Drava system during the Quaternary after abandonment of the old
24
(palaeo Drava) channel during Pliocene and thus it must also stand in a visible contrast to the
older sedimentary rocks at the periphery of the sampled profile lines. However, in the former
case the difference is caused by anthropogenic, while in the latter by the natural causes.
Basically, the same geochemical pattern emerges among all three investigated models except that
the third one is more concerned with local variations confined to individual profiles sampled
along the pre-designed sets of profile lines. This is confirmed by the great number of statistically
significant DFs with quite obscure or unfathomable geological meaning. However, despite its
relatively weak geological explanatory potential the PROFILE model succeeded in
distinguishing between the carbonate-related and contaminant variable clusters which were
otherwise entirely inseparable in the models dealing with more general geological picture in the
investigated area – discrimination between lithological units and depth intervals. The greatest
discriminant potential in all models is contained in the first the first two DFs whose geological
interpretation is unambiguous. They separate the a priori established groups on the ground of
geological processes underlying a relationship between a few distinctly outlined variable
clusters. The first discriminant function DF1 is bipolar and represents carbonatecontaminant/silicate axis separating alluvium (A) from bedrock (B) in the LITHOLOGY and
DEPTH-LITHOLOGY model. The second discriminant function DF2 is also bipolar and can be
described as heavy metal axis which separates the terrace (T) from both alluvium (A) and
bedrock (B) in the LITHOLOGY model but, more distinctly, distinguishes between the
individual sampling intervals (a, b and c) in the DEPTH-LITHOLOGY model. However, as
opposed to DF1, this separation is achieved by opposing the two heavy metal clusters of which
one is primarily anthropogenic (Pb, Zn, Cd, Mo) while the other reflects the natural origin (Ni,
Cr, Fe, Cu).
The next important issue concerning the investigated models is the possibility to reassign the
previously classified objects (samples) to another group if their calculated geochemical affinity
25
with that group has proved greater than to the observed one. It is therefore possible to use the
average geochemical composition (particularly the total heavy metal content) of a particular
group as an identification card in mapping the youngest, unconsolidated sediments developed in
a variety of quaternary facies on the Drava floodplain which sometimes may be incorrectly
mapped due to the lack of fossil content or other characteristic geological observations.
Investigated models disclose probability of wrong classification of neighboring lithologic units –
alluvium can be mistaken for terrace, or (more rarely) bedrock for terrace, and vice versa.
Confusion may also appear with regard to particular intervals sampled in any of the three
lithologic groups. However, the deepest c-intervals in bedrock (B) are, generally, not confounded
with others. This is natural considering the bedrock as an “old”, consolidated, sedimentary
material while various degrees of mixing is present in younger, Quaternary, sediments,
particularly in the case of close-to-surface and topmost horizons being most heavily exposed to
contamination. The greatest confusion occurs in alluvium b-horizons where more than half of all
samples are incorrectly classified and assigned to other alluvium horizons or even to the terrace.
7.
ACKNOWLEDGMENTS
This study was funded jointly by The Ministry of Science, Education and Sports, Republic
of Croatia, and The Ministry of Higher Education, Science and Technology of the Republic of
Slovenia – Bilateral Project Slo-Hr: “Heavy metals in alluvial sediments of the Drava River”
(BI-HR/07-08-002) – as well as the Ministry of Science, Education and Sports, Republic of
Croatia – Scientific Project: “The Basic Geochemical Map of Croatia” (181-1811096-1181).
Their support is greatly appreciated. The authors would like to express their gratitude to all who
participated to the project. They are also indebted to Tamara Marković for making constructive
26
comments on behavior of heavy metals in the hydrological cycle in the Croatian part of the
investigated terrain (Varaždin aquifer).
8.
REFERENCES
Abd El-Azim H, El-Moselhy KhM (2005) Determination and partitioning of metals in sediments
along the Suez Canal by sequential extraction. Journal of Marine Systems 56(3-4):363-374
Amorosi A, Summartino I (2007) Influence of sediment provenance on background values of
potentially toxic metals from near-surface sediments of Po coastal plain (Italy). International
Journal of Earth Sciences 96:389-396
Andjelov M. (1993) Results of radiometric and geochemical measurements for the natural
radioactivity map of Slovenia. Geologija 36:223-248. (In Slovenian; English abstract)
Bidovec M, Šajn R, Gosar M. (1998) The use of recent overbank sediments in geochemical
mapping of Slovenia (in Slovenian; English abstract). Geologija 41:275-317
Bølviken B, Bogen J, Jartun M, Langedal M, Ottesen RT, Volden T (2004) Overbank sediments:
a natural bed blending sampling medium for large-scale geochemical mapping. Chemometrics
and Intelligent Laboratory Systems 74:183-199
Bottrill LJ, Walling DE, Leeks GJ (1999) Geochemical characteristics of overbank deposits and
their potential for determining suspended sediment provenance; an example from the River
27
Severn, UK. In: Marriot SB, Alexander J (eds) Floodplains: Interdisciplinary Approaches.
Geological Society, London, Special Publications 163:241-257
Brewer PA, Taylor MP (1997) The spatial distribution of heavy metal contaminated sediment
across terraced floodplains. Catena 30:229-249
Cappuyns V, Swennen R, Vandamme A, Niclaes M (2006) Environmental impact of the former
Pb-Zn minings and smelting in East Belgium. Journal of Geochemical Exploration 88:6-9
Ciszewski D (2002) Heavy metals in vertical profiles of the middle Odra River overbank
sediments: evidence for pollution changes. Water, Air, and Soil Pollution 143:81-98
Collins AL, Walling DE, Leeks JL (1998) Use of composite fingerprints to determine the
provenance of the contemporary suspended sediment load transported by rivers. Earth Surface
Processes and Landforms 23:31-52
Collins AL, Walling DE, Leeks JL. (1997) Use of the geochemical record preserved in
floodplain deposits to reconstruct recent changes in river basin sediment sources.
Geomorpohlogy 19:151-167
Čović M (2003) Geochemical and mineralogical characteristics of overbank sediments of
Žumberak and Samobor Mts., Croatia. (unpublished Master of Science Thesis; in Croatian).
University of Zagreb, Faculty of Mining, Geology and Petroleum Engineering, 139 p
28
Darnley AG, Björklund A, Bölviken B, Gustavsson N, Koval PV, Plant JA, Steenfeld A,
Tauchid M, Xie Xuejing (1995) A global geochemical database for environmental and resource
management. Recommendations for international geochemical mapping. Final report of IGCP
project 259, UNESCO Publishing, 122 p
De Vos W, Demetriades A, Marsina K, Ottesen RT, Reeder S, Pirc S, Salminen R, Tarvainen T
(2006) Comparison of elements in all sample media, gemeral comments and conclusions. In: De
Vos W, Tarvainen T (eds) Geochemical Atlas of Europe, Part 2: Interpretation of geochemical
maps, additional figures, maps and related publications; Sr-Strontium, pp 347-352
Dillon WR, Goldstein M (1984) Multivariate analysis: methods and applications. John Wiley &
Sons, New York, 575 p
Ettler V, Zelena O, Mihaljevič M, Šebek O, Strnad L, Coufal P, Bezdička P (2006) Removal of
trace elements form landfill leachate by calcite precipitation. Journal of Geochemical
Exploration 88:28-31
Fernandes HM (1997) Heavy metal distribution in sediments and ecological risk assessment: The
role of diagenetic processes in reducing metal toxicity in bottom sediments. Environmental
Pollution 97(3):317-325
Förstner U, Whitmann GTW (1981) Metal pollution in the aquatic environment. SpringerVerlag, Berlin-Heidelberg-New York, 481 p
29
Foster IDL, Charlesworth SM (1996) Heavy metals in the hydrological cycle: trends and
explanation. Hydrological Processes 10:227-261
Gibbs RJ (1977) Transport phases of transition in the Amazon and Yukon Rivers. Geological
Society of America Bulletin 88:829-943
Halamić J, Galović L, Šparica M (2003) Heavy Metal (As, Cd, Cu, Hg, Pb, and Zn) Distribution
in Topsoil Developed on Alluvial Sediments of the Drava and Sava Rivers in NW Croatia.
Geologia Croatica 56/2:215-232
Halamić J, Šajn R, Peh Z, Galović L (2005) Heavy Metals in the Alluvial Sediments of the River
Drava (Croatia, Slovenia). Third Croatian Geological Congress, Opatija, September, 29th –
October, 1st, 2005. Abstracts Book, pp. 193-194, Zagreb
Hugget RJ (1998) Soil chronosequences, soil development and soil evolution: a critical review.
Catena, 32:155-172
Lambert CP, Walling DE (1987) Floodplain sedimentation: a preliminary investigation of the
contemporary deposition within the lower reaches of the river Culm, Devon, UK. Geografiska
Annaler, 69A(3-4):393-404
Langedal M, Ottesen RT (1998) airborne pollution in five drainage basins in Eastern Finmark,
Norway: an evaluation of overbank sediments as sampling medium for environmental studies
and geochemical mapping. Water, Air and Soil Pollution 101:377-398
30
Macklin MG, Ridgway J, Passmore DG, Rumsby BT (1994) The use of overbank sediment
geochemical mapping and contamination assessment: results from selected English and Welsh
floodplains. Applied Geochemistry 9:689-700
Macklin MG, Brewer PA, Hudson-Edwards KA, Bird G, Coulthard TJ, Dennis IA, Lechler PJ,
Miller JR, Turner JN (2006) A geomorphological approach to the management of rivers
contaminated by metal mining. Geomorphology 79.423-447
Marković T (2007) Vulnerability assessment of the unsaturated zone by geochemical modeling.
(unpublished PhD Thesis; in Croatian). University of Zagreb, Faculty of Mining, Geology and
Petroleum Engineering, 155 p
Matschullat J, Ottenstein R, Reimann C (2000) Geochemical background – can we calculate it?
Environmental Geology, 39(9), 990-1000
McConnel JW, Finch C, Hall GEM, Davenport PH (1993) Geochemical mapping employing
active and overbank stream-sediment, lake sediment and lake water in two areas of New
Foundland. Journal of Geochemical Exploration 49:123-143
McMartin I, Henderson PJ, Plouffe A, Knight RD (2002) Comparison of Cu-Hg-Ni-Pb
concentrations in soils adjacent to anthropogenic point sources: example from four Canadian
sites. Geochemistry, Exploration, Environment, Analysis 2:57-74
Miko S, Halamić J, Peh Z, Galović L (2001) Geochemical Baseline Mapping of Soils Developed
on Diverse Bedrock from Two Regions in Croatia. Geologia Croatica 54/1:53-118
31
Mioč P, Žnidarčič M (1977) Geology of Slovenj Gradec sheet. Basic geological map of
Yugoslavia, 1:100000. Geološki zavod Ljubljana, Beograd (Federal Institute of Geology) (In
Slovenian; English summary)
Mioč P, & Marković S (1998) Geology of Čakovec sheet. Basic Geological map of Croatia and
Slovenia, 1:100000. Inštitut za geologijo, geotehniko in geofiziko, Ljubljana and Institut za
geološka istraživanja, Zagreb (In Croatian; English summary)
Molinaroli E, Basu A (1993) Toward quantitative provenance analysis: A brief review and case
study. In: Johnsson MJ, Basu A (eds) Processes controlling the composition of clastic sediments.
Geological Society of America Special Paper 284:323-333
Ódor L, Horváth I, Fügedi U (1997): Low-density geochemical mapping in Hungary. Journal of
Geochemical Exploration 60:55-66
Ottesen RT, Bogen J, Bølviken B, Volden T (1989) Overbank sediment as arepresentative
medium for regional geochemical mapping. Journal of Geochemical Exploration 32:257-277
Owens PN, Walling DE, Leeks GJL (1999) Use of floodplain sediment cores to investigate
recent historical changes in overbank sedimentation rates and sediment sources in the catchment
of the River Ouse, Yorkshire, UK. Catena 36:21-47
Pavlovic G, Prohic E, Tibljas D (2004) Statistical assessment of geochemical pattern in overbank
sediments of the river Sava, Croatia. Environmental Geology 46:132-143
32
Phillips JD (1999) Earth surface systems: complexity, order, and scale. Blackwell, Malden, 180
p.
Reimann C, Filzmoser P (2000) Normal and lognormal data distribution in geochemistry: death
of a myth. Consequences for the statistical treatment of geochemical and environmental data.
Environmental Geology 39(9):1001-1014
Reimann C, Filzmoser P, Garret RG (2005) Background and threshold: critical comparison of
methods of determination. Science of the Total Environment 346:1-16
Rhoads BL, Cahill RA (1999) Geomorphological assessment of sediment contamination in an
urban stream system. Applied Geochemistry 14:459-483
Rock NMS (1988) Lecture Notes in Earth Sciences, 18: Numerical Geology. Springer-Verlag,
Berlin-Heidelberg-New York-London-Paris-Tokyo, 427 p.
Romic M, Romic D (2003) Heavy metals distribution in agricultural topsoils in urban area.
Environmental Geology 43:795-805
Rognerud S, Hongve D, Fjeld E, Ottesen RT (2000) Trace metal concentrations in lake and
overbank sediments in southern Norway. Environmental Geology 39(7):723-732
Salminen R, Tarvainen T, Demetriades A, Duris M, Fordyce FM, Gregorauskiene V, Kahelin H,
Kivisilla J, Klaver G, Klin H, Larson JO, Lis J, Locutura J, Marsina K, Mjartanova H, Mouvet C,
33
O’Connor P, Ódor L, Ottonello G, Paukola T, Plant JA, Reimann C, Schermann O, Siewers U,
Steenfelt A, van der Sluys J, de Vivo B, Wiliams L (1998) FOREGS Geochemical mapping field
manual, Guide 47, 35 p. Geological Survey of Finland, Espoo.
Singh AK, Hasnain SI, Banerjee DK (1999) Grain size and geochemical partitioning of heavy
metals in sediments of the Damodar River – a tributary of the lower Ganga. Environmental
Geology 39(1):90-98
Sutherland RA (1998) A comparison of geochemical information obtained from two fluvial bed
sediment fractions. Environmental Geology 39(3-4):330-341
Swennen R, Van Keer I, De Voos W (1994) Heavy metal concentration in overbank sediments
of the Gaul river (East Belgium): Its relation to former Pb-Zn mining activities. Environmental
Geology 24:12-21
Swennen R, Van Der Sluys J, Hindel R, Brusselmans A (1998) Geochemistry of overbank and
high-order stream sediments in Belgium and Luxemburg: a way to assess environmental
pollution. Journal of Geochemical Exploration 62:67-79
Swennen R, Van Der Sluys J (2002) Anthropogenic impact on sediment composition and
geochemistry in vertical overbank profiles of river alluvium from Belgium and Luxemburg.
Journal of Geochemical Exploration 75:93-105
Šajn R (2002) Influence of mining and metallurgy on chemical composition of soil and attic dust
in Meža valley, Slovenia. Geologija 45/2:547-552. (In Slovenian; English abstract)
34
Šajn R. (2003) Distribution of chemical elements in attic dust and soil as reflection of lithology
and anthropogenic influence in Slovenia. Journal de Physique 107:1173-1176
Šajn R, Gosar M (2004) An overview of some localities in Slovenia that become polluted due to
past mining and metallurgic activities. Geologija 47/2:249-258. (In Slovenian; English abstract)
Šajn R, Peh Z, Halamić J, Miko S, Galović L (2005) Experimental geochemical Map of Croatia
and Slovenia. Third Croatian Geological Congress, Opatija, September, 29th – October, 1st,
2005. Abstracts Book, pp 243-244, Zagreb
Volden T, Reimann C, Pavlov VA, de Caritat P, Äyräs (1997) Overbank sediments from the
surroundings of the Russian nickel mining and smelting industry on the Kola Peninsula.
Environmental Geology 32(3):175-185
Vreča P, Šajn R, Pirc S (2001) Natural and anthropogenic influences on geochemistry of soils in
terrains of barren and mineralized carbonate rocks in the Pb-Zn mining district of Mežica,
Slovenia. Journal of Geochemical Exploration 74:99-108
Whitney PR (1975) Relationship of manganese – iron oxides associated heavy metals to grain
size in stream sediments. Journal of Geochemical Exploration 4:251-263
Zachara JM, Cowan CE, Resch CT (1991) Sorption of divalent metals on calcite. Geochimica
and Cosmochimica Acta, 55:1549-1562
35
36
Download