Sedimentary chemofacies characterization by means of multivariate analysis

Jean Carlos Montero-Serrano

a,b,
, Javier Palarea-Albaladejo
c
, Josep A. Martn-Fernndez
d
,
Manuel Martnez-Santana
a
, Jos Vicente Gutirrez-Martn
a
a
Instituto de Ciencias de la Tierra, Universidad Central de Venezuela, Caracas 1010-A, Apartado Postal 3895, Venezuela
b
Universit Lille 1, Laboratoire Gosystmes, FRE 3298 CNRS, btiment SN5, 59655 Villeneuve d'Ascq cedex, France
c
Biomathematics and Statistics Scotland, James Clerk Maxwell Building, The King's Buildings, Edinburgh, EH9 3JZ, UK
d
Departament d'Informtica i Matemtica Aplicada, Universitat de Girona, Campus Montilivi, 17071 Girona, Spain
a b s t r a c t a r t i c l e i n f o
Article history:
Received 28 September 2009
Received in revised form 4 April 2010
Accepted 27 April 2010
Available online 6 May 2010
Editor: G.J. Weltje
Keywords:
Geochemistry
Chemostratigraphy
Compositional data
Log-ratio transformation
Robust statistics
Constrained cluster
Multivariate statistical analysis is applied to geochemical data from three sections forming part of the
stratigraphic record of the Cerro Pelado Formation (OligoceneMiocene), in the central region of the Falcn
Basin, northwestern Venezuela. Our main goal is introducing and testing a statistical protocol for the
identication of chemofacies in the studied sections. The rst step involves data preparation and cleaning:
selection of relevant components, convenient replacement of values below the detection limit and
determination of outliers. Second, a biplot analysis allows us to infer geochemical processes that can be
interpreted from a paleoenvironmental point of view: detrital association, redox-organic matter association
and carbonatic association. Considering such geochemical associations, a constrained cluster analysis is then
carried out to determine the chemofacies for each section. According to the compositional nature of
geochemical data, all statistical analysis is conducted within a log-ratio analysis framework. In addition,
robust statistical methods are considered for outlier detection and biplot representation in order to smooth
the inuence of potential outliers on the estimates.
2010 Elsevier B.V. All rights reserved.
1. Introduction
Geochemical signatures of clastic sedimentary rocks provide
important sources of information that record different aspects of
provenance, as well as tectonic, environmental and ecological
evolution. Because hydraulic sorting, weathering and diagenesis can
alter the geochemical composition of basin sediments, much
emphasis has been placed on variations in the relatively immobile
elements such as Cr, Co, Th, Y, Th, Zr, Ce, La, Sc and Ti. The lowmobility
of these elements during sedimentary processes enables better
characterization of source rock compositions, paleo-climatic condi-
tions and tectonic setting (Taylor and McLennan, 1985; Condie, 1993;
Pearce et al., 1999; Spalletti et al., 2008).
Recent technological advances now enable the fast determination
of major and trace element concentrations in a large number of
samples from a sedimentary sequence. When a suite of elemental
concentrations is evaluated in the context of the stratigraphic log, a
chemostratigraphic study is accomplished, involving the application
of major and trace elements for the characterization of the sedimen-
tary sequence into geochemically distinct units (Winchester and Max,
1996; Das, 1997; Jarvis et al., 1998; Pearce et al., 1999; Retegui et al.,
2005); this tool can be very useful when applied to sequences with
poor biostratigraphic control (Pearce et al., 1999). Other aspects
revealed by chemostratigraphic studies are paleoproductivity, climat-
ic changes and chemical cyclicity in processes involving basin sedi-
mentation (Bellanca et al., 1996; Villamil, 1996; Dayong et al., 1999).
Temporal variations in elemental abundances allow the characteriza-
tion of sedimentary sequences by subdivision into chemical facies or
chemofacies; these stratigraphic units exhibit a chemical signature,
related to their environmental conditions of formation (Pearce et al.,
1999; Retegui et al., 2005).
Application of chemostratigraphic tools requires several compo-
nents or attributes (e.g., Al, Si, Ca, Fe, Co, Ni, Cu and Zn) in a large set of
rock samples collected in a stratigraphic order. As a result, multi-
variate data sets are obtained, so multivariate statistical methods are
needed to investigate the orientation or trends of the underlying
geochemical processes.
In this study, the geochemical data obtained are of a compositional
nature, that is, they are vectors of non-negative values subjected to a
constant-sumconstraint. Namely, they are measured in units like wt.%
(weight percent) or g/g (micrograms per gram). This fact implies that
Sedimentary Geology 228 (2010) 218228
Corresponding author. Universit Lille 1, Laboratoire Gosystmes, FRE 3298 CNRS,
btiment SN5, 59655 Villeneuve d'Ascq cedex, France. Tel.: +33 169823537; fax: +33
169823568.
E-mail addresses: jeanmontero@yahoo.es (J.C. Montero-Serrano),
Javi.Palarea@gmail.com (J. Palarea-Albaladejo), josepantoni.martin@udg.edu
(J.A. Martn-Fernndez), manmarti@gmail.com (M. Martnez-Santana),
jgutierr@ciens.ucv.ve (J.V. Gutirrez-Martn).
0037-0738/$ see front matter 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.sedgeo.2010.04.013
Contents lists available at ScienceDirect
Sedimentary Geology
j our nal homepage: www. el sevi er. com/ l ocat e/ sedgeo
relevant information is contained in the relative magnitudes, so
statistical analysis must focus on the ratios between components
(Aitchison, 1986). The suitability of the log-ratio approach for the
analysis of data representing parts of a whole is broadly supported, as
much from a theoretical viewpoint as from a practical one (see e.g.,
Weltje, 2002; von Eynatten, 2004; Pawlowsky-Glahn, 2005; Daunis-i-
Estadella and Martn-Fernndez, 2008; Barbera et al., 2009; Tolosana-
Delgado and von Eynatten, 2009).
The aim of this work is to introduce a suitable statistical protocol
for identifying chemofacies in sedimentary sequences, taking into
account the compositional nature of the data. We evaluate this
statistical approach on chemostratigraphic data of the Cerro Pelado
Formation.
2. Study sequence geological setting
The Cerro Pelado Formation lies in the Falcn Basin, northwestern
Venezuela (Fig. 1), with a thickness greater than 1000 m. The Cerro
Pelado Formation does not crop out in a continuous section. The two
main sections sampled were at the El Troncn (bottom of the unit)
and La Paloma (middle-upper part). A shorter section in the La Cuesta
coal mine was also sampled (Fig. 1). This formation was deposited in
an extensional basin associated with major strike-slip faults related to
the passage of the Caribbean plate (Macellari, 1995; Audemard, 2001;
Bezada et al., 2008). During its early history the basin (Oligocene and
Early Miocene) subsided rapidly, and marine deep-water sedimenta-
tion eventually extended over most of it (Daz de Gamero, 1977).
The whole basin was subjected to intense deformation, resulting in
numerous faulting and folding structures. The sedimentation of the
Cerro Pelado Formation took place in the Urumaco Trough, in the
western part of the basin. The unit was deposited during the Lower
Miocene, in a regressive cycle (Hambalek et al., 1994), within a
prograding deltaic system and its associated facies, including delta
front, delta plains with interdistributary channels, swamps, and
marshes. (Daz de Gamero, 1989). In summary, the formation rep-
resents a transition from the marine environment of the underlying
Agua Clara Formation to the very near-shore to coastal plain
environment that prevailed when the overlying Querales Formation
was deposited. The Cerro Pelado Formation consists mainly of lithic
sandstones in beds up to 20 m thick, intercalated with shale and
siltstone beds. Presumably the individual shale and siltstone beds
have thicknesses of up to 2.7 m (Daz de Gamero, 1977). The basal
part of the unit is composed of grey to reddish-brown ferruginous
sandstones with cross-bedding. The upper part of the unit is
composed of bedded calcareous sandstones and calcareous shales
containing coal beds up to 2 m thick (Daz de Gamero and Linares,
1989). Shales are sometimes carbonaceous and contain grayish-blue
limestone nodules; in addition, they contain fossils primarily typical
of brackish waters (Hambalek et al., 1994). Sub-bituminous to
bituminous coal beds are found in the middle and upper parts of
the Cerro Pelado Formation; these were deposited in a marine
sedimentary environment with lagoons and marshlands that devel-
oped due to delta advancement (Gonzlez et al., 1985). The Cerro
Pelado Formation is considered to be a probable reservoir rock in the
petroleum system of Falcn Basin; several coal mines (Fig. 1) are
present in its outcrop area (Ghosh et al., 1997).
Fig. 1. Location map showing the three studied sections of the Cerro Pelado Formation (shaded area) in the Falcn Basin, northwestern Venezuela.
219 J.C. Montero-Serrano et al. / Sedimentary Geology 228 (2010) 218228
3. Samples and analytical methods
3.1. Sampling
Outcrop samples from the Cerro Pelado Formation were collected
systematically, with an average frequency of one sample every 2 m
independently of the lithology (Fig. 2). The total gathered amount was
226 samples: 90 rock samples in 440 m of the stratigraphic log raised
in the El Troncn stream (lower section, bottom), 40 rock samples in
95 m of the stratigraphic log raised in the La Cuesta coal mine (central
section) and 96 rock samples in 600 m of the log raised in the La
Paloma stream (top section).
3.2. Petrography and X-ray diffraction analysis
Thin sections were prepared in the standard manner, using
selected sandstones from different points of the stratigraphic sections
(Fig. 2). Optically identiable components of the sandstones were
mono- and poly-crystalline quartz (7080%), rock fragments (520%;
metamorphic and sedimentary lithoclasts being the most common
type), feldspars (510%; Orthoclase was the most common type) and
accessory minerals: muscovite, biotite, chlorite, Fe-oxides and glau-
conite. Heavy minerals were rare, and dominated by zircon. Sand-
stones of the Cerro Pelado Formation are ne grained and moderately
sorted and contain sub angular to sub rounded clasts. Interstitial
components include varying percentages of nely crystalline clay
cement and blocky calcite spar with high iron oxide content.
Sandstones are generally matrix-free to matrix-poor (310%) and
generally contain fossil plant debris such as leaves. Sandstones studied
were classied on the range fromlitharenite to sublitharenite, making
use of the Pettijohn triangle (Pettijohn et al., 1987). Based on their
texture and mineralogy, the sandstones are immature to moderately
mature. Predominant contacts between grains were longitudinal and
concaveconvex, suggesting a moderate stage of diagenesis. Petro-
graphic interpretation of the sandstones suggests a contribution of
felsic metamorphic and plutonic source rocks (e.g., The Andes moun-
tain range and Paraguan Peninsula).
Major minerals in the raw shale samples, from different points of
the stratigraphic sections (Fig. 2), were identied by X-ray diffrac-
tometry (XRD). Minerals identied from XRD patterns include quartz,
k-feldspar, kaolinite, illite, and muscovite. However, minor amounts
of mixed-layer mica-montmorillonite, gypsum and jarosite are still
present in many of the samples, and trace amounts of calcite and
dolomite are also present in many samples.
3.3. Chemical determinations
Samples were ground in a tungsten carbide ball-mill. Dissolution
of samples was made by sinterizing with sodium peroxide, as de-
scribed by Borsier (1991). Geochemical data (Cr, Zn, B, P, Mn, V, Ti, Y,
Sr, La, Ce, Ba, Li, K, Rb, Fe, Mg, Ca, Cu, Ni, expressed as g/g) were
obtained by means of inductively coupled plasma optical emission
spectrometry (ICP-OES) using a Jobin Yvon, model JY-24. The
analytical accuracy and precision were found to be better than 2%
for major and minor elements, and 612% for trace elements, as
veried by international standards (granite USGS-G2 and andesite
USGS AGV-1) and analysis of replicate samples. Total sulfur was
determined in a LECO SC-432 analyzer. Total carbon (TC) and total
Fig. 2. Lithostratigraphic logs of the three studied sections of the Cerro Pelado Formation.
220 J.C. Montero-Serrano et al. / Sedimentary Geology 228 (2010) 218228
inorganic carbon (TIC) were determined in a coulometric carbon
analyzer (Coulometrics). The contents of total organic carbon (TOC),
as weight percent, were calculated as the difference between TC and
TIC (%TOC=%TC%TIC). Calcium carbonate content (CaCO
3
), as
weight percent, was quantitatively estimated from the inorganic
carbon (IC) content (%CaCO
3
=%TIC8.33).
4. Statistical methods
4.1. Compositional data and log-ratio analysis
Formally, compositional or closed data refer to D-dimensional
vectors x=[x
1
, x
2
,, x
D
] of positive components summing up to a
constant k, hence dened on the simplex sample space:
S
D
= x= x
1
; x
2
; ; x
D
j x
i
N 0;
D
i =1
x
i
=k
_ _
:
The characterization of S
D
as Euclidean space and related topics are
summarized in Aitchison and Egozcue (2005) or Pawlowsky-Glahn
et al. (2007).
Following Aitchison (1986), the statistical analysis of compositions
should be focused on the ratios between components. In this way,
data are moved to real space wherein standard methods can be
applied. Aitchison (1986) suggested two main transformations: rst,
the additive log-ratio (alr) transformation from S
D
to R
D1
is dened
as
y = alrx = ln
x
1
x
D
_ _
; ln
x
2
x
D
_ _
; ; ln
x
D1
x
D
_ _ _ _
: 1
From the Euclidean structure of S
D
, a distance, known as the
Aitchison distance (Aitchison et al., 2000), is induced:
d
a
x; x
0
=
1
D

ibj
ln
x
i
x
j
ln
x
0
i
x
0
j
_ _
2
_ _
1=2
: 2
Second, the centered log-ratio (clr) from S
D
to R
D
is dened as
z = clrx = ln
x
1
gx
_ _
; ln
x
2
gx
_ _
; ; ln
x
D
gx
_ _ _ _
: 3
where g(x)=[x
1
x
2
... x
D
]
1/D
is the geometric mean of the com-
position x.
More recently, another transformation was proposed by Egozcue
et al. (2003): the isometric log-ratio (ilr) transformation. The idea is to
work with the real coordinates of a composition with respect to an
orthonormal basis in the simplex. A strategy for determining such a
basis is suggested in Egozcue and Pawlowsky-Glahn (2005); however,
whenever necessary, we will use the basis ilr(x)=v=[v
1
,,v
D1
]
originally introduced in Egozcue et al. (2003), where
v
i
=
1

ii + 1
_ ln

i
j =1
x
j
x
i + 1

i
_ _
: 4
4.2. Data preliminaries: values belowthe detection limit and multivariate
outliers
4.2.1. Values below the detection limit
When collecting and recording geochemical sample data, some
values frequently fall below the minimum concentration at which
measuring devices can detect the presence of a chemical component.
The presence of values below the detection limit (VBDL) prevents us
from applying the log-ratio approach. Hence, a suitable imputation
procedure is required before data analysis can be conducted.
Following Palarea-Albaladejo et al. (2007) and Palarea-Albaladejo
and Martn-Fernndez (2008), a modied version of the well-known
ExpectationMaximization (EM) algorithm based on the alr transfor-
mation is applied. It is important to emphasize that the results from
the modied EM algorithm will be invariant with respect to
permutations of the elements of the chemical composition, conse-
quently, independent of the selected divisor in the alr transformation.
4.2.2. Multivariate outlier detection
Given the high-dimensional nature of our problem, we adopt a
multivariate approach for outliers. Following Filzmoser and Hron
(2008), the Mahalanobis distance (MD), based on robust location and
covariance estimators, namely, the Minimum Covariance Determinant
estimator (MCD) in the alr space, is usedto detect potential outliers. The
robust approach is justied here by the heterogeneity of the samples. In
this case, classical estimators can lead to questionable results. Assuming
a chi-square
D1
2
approximation for the squared MD, the 0.975
quartile of

2
D1
_
is used as the outlier cut-off value for extremeness.
4.3. Exploratory data analysis
To catch an initial glimpse of the data, ordinary univariate statistics
are computed (Table 1). Based on the part-of-a-whole nature of the
values, the geometric mean g
i
=
n
j =1
x
ij
_ _
1
n
; i = 1; ; D, is the mea-
sure that best represents the center of a component. Hence, the compo-
sitional measure of the central position is obtained by closing the vector
of univariate geometric means as g=C(g
1
,,g
D
), where C is the closure
operator dened as C(x)=[x
1
/ x
i
,,x
D
/ x
i
] (Pawlowsky-Glahn
and Egozcue, 2002).
Additionally, we studied the relative weight of every geochemical
component through the sections. For this, the data are centered by
calculating the compositional difference between each sample and
the center g as x g=C(x
1
/g
1
,,x
D
/g
D
). Then, taking into account the
multiplicative nature of movements in the simplex, the sections'
centers are respectively compared by ratio with the global center.
These ratios reect the relative concentration of a component in a
section with respect to the global geochemical pattern, thereby
helping us to characterize the three sections considered.
4.3.1. Compositional biplot
This graphical tool helps to nd subsets of components with sim-
ilar relative variation patterns. Since those components are essentially
redundant, they are clustered together into a single part, thus
reducing the dimension of the data. Biplot analysis is usually applied
to compositional data using the clr transformation (3) (Aitchison,
1990; Aitchison and Greenacre, 2002). The axes of the biplot cor-
respond to the rst two principal components (PCs) of the log-centered
data, and both the samples and the components are simultaneously
represented. Note that the high percentage of potential outliers
detected in our data (see Section 5) may ruin the estimation of the PCs
based on the classical covariance matrix, as was, in fact, observed in
our preliminary testing. However, a robust covariance estimator is not
directly applicable on clr-data since it requires a full rank data matrix.
For overcoming this problem, we follow the strategy suggested by
Filzmoser et al. (2009) to obtain robust PCs when dealing with
compositional data. First, the data are clr-transformed, then, they are
moved to the ilr-space by means of the orthonormal basis v in Eq. (4)
as y=z.v. Next, PCs are obtained from the ilr-coordinates y, based on
the MCD robust covariance estimator. Finally, the scores s
y
and
loadings l
y
from the PC analysis are back-transformed into the clr-
space by s
z
=s
y
.v' and l
z
=v.l
y
for obtaining the compositional biplot
representation, as usual. Details on the interpretation of a composi-
tional biplot are discussed in von Eynatten et al. (2003) and Martn-
Fernndez et al. (2005).
221 J.C. Montero-Serrano et al. / Sedimentary Geology 228 (2010) 218228
4.4. Constrained cluster analysis
One of our main objectives is partitioning every section into geo-
chemically meaningful zones, i.e., into distinctive chemofacies. These
zones are determined by forced grouping or constrained cluster
analysis (Gill et al., 1993). In this way, the special requirements of the
stratigraphic context of the zonation are taken into account by adding
an adjacency constraint. Then, each sample is grouped by comparison
with the adjacent ones (lower and upper samples in the stratigraphic
log). In the dendrogram, the samples are arranged according to the
depth-stratigraphic (in meters). Based on experimentation with
different clustering methods, Templ et al. (2008) suggest that the
well-known Ward's method is the best performing hierarchical
clustering when dealing with geochemical data. Following this
suggestion, the Ward's method is applied using the Aitchison distance
(2) as the measure of dissimilarity.
5. Results and discussion
5.1. Data pre-treatment and exploration
Table 1 summarizes the descriptive statistics of the raw data
obtained from the Cerro Pelado Formation. Simple examination
suggests great differences in the distribution of the concentrations
of the various chemical components, an inference supported by the
differences of the geometric means and also by the minimum and
maximumvalues. These differences are mainly due to the fact that the
major elements (e.g., K, Fe, Ti, Mg, Ca) occur in a much greater
concentration than the trace elements (e.g., Cr, Zn, La, Ce, Li, Y). On the
other hand, the quantiles reect the usual asymmetric distribution of
raw geochemical data. Geochemical variations observed in each
section are compatible with the lithological and paleoenvironmental
changes documented for the Cerro Pelado Formation. For instance, the
greater CaCO
3
and Ca concentrations in the La Paloma section are
related to a greater predominance of calcareous sandstones in the
upsection of the unit, which likely represent the main channel
deposits (Daz de Gamero, 1989; Hambalek et al., 1994). In addition,
the higher TOC and S concentrations in the La Cuesta and La Paloma
sections compared to the El Troncn are due to the fact that coal beds
predominate in the middle and upper parts of the formation, which
was previously interpreted as meaning that they were deposited in
marshlands or lagoons adjacent to the coastline (Gonzlez et al.,
1985). Finally, the greater Fe concentrations in the El Troncn section
relative to the La Cuesta and La Paloma are due to the presence of
ferruginous sandstones which likely reects intensive postdeposi-
tional processes in the basal part of the unit (Daz de Gamero and
Linares, 1989).
Additionally, note that the presence of VBDLs prevents us from
obtaining the true minimum value for some components; therefore,
the threshold value is reported instead (indicated by the less-than
operator in Table 1). Those threshold values will be used next as the
reference for the VBDL replacement method.
Originally, 23 chemical components were registered in each
section, but Cu and Ni were omitted in the subsequent analysis
because of the large number of VBDLs (N20% for Cu and N50% for Ni;
Fig. 3). As can be seen, the other components present a low-moderate
number of VBDLs (b10%; Fig. 3), so we proceed to replace them with
estimated values obtained by the modied EM algorithm. Hence, the
VBDLs for each section are conveniently replaced by small expected
values.
Next, we checked the data set for multivariate outliers by
computing the robust MD for every sample, distinguishing by
stratigraphic section (Fig. 4). The percentage of potential outliers
provided by that approach does not vary greatly from one section to
another (cut-off value

2
20;0:975
_
= 5:84; solid line in Fig. 4; 35% for
Table 1
Univariate descriptive statistics for the Cerro Pelado samples. Chemical components
expressed in micrograms per gram (g/g). n: number of samples; Min.: minimum; Q
1
:
rst quartile (25th percentile); Geo. Mean: geometric mean; Q
3
: third quartile (75th
percentile); Max.: maximum.
Section Comp. Min. Q
1
Geo. Mean Q
3
Max.
La Paloma (n=96) Cr b2 13 21 32 86
Zn b4 20 33 60 123
B b1 12 17 25 60
P 37 124 201 305 9549
Mn b1 54 170 438 5230
V 8.6 26 42 65 192
Cu b2 1.5 4.9 5.5 19
Ti 549 2520 3341 4890 11,275
Ni b6 4.2 14 13 44
Y b1 11 15 22 102
Sr b0.6 23 33 48 123
La b1 6.5 11 18 59
Ce b1 97 158 240 916
Ba 10 49 74 122 1116
Li 22 41 69 107 258
K b632 11,621 15,605 27,677 60,719
Rb b10 40 69 112 260
Fe 1941 7811 13,718 25,436 63,039
Mg b1 1250 2411 4758 27,900
Ca 57 1058 3518 12,202 116,682
S b100 1213 2705 6545 45,700
TOC b1 1005 1960 4861 15,324
CaCO
3
122 1240 6817 34480 231,373
La Cuesta (n=40) Cr b2 8.7 18 38 85
Zn b4 9.2 26 45 121
B b1 9.6 15 28 88
P 26 115 223 445 2149
Mn b1 62 148 387 1370
V 5.8 22 53 142 197
Cu b2 1.5 6.4 8.4 22
Ti 429 2010 3214 6405 10,175
Ni b6 4.2 15 11 54
Y 1.6 6.8 14 25 58
Sr 6.9 16 26 54 134
La b1 3.7 10 27 69
Ce 2.7 63 114 283 783
Ba 3.1 39 60 136 287
Li 13 42 63 115 222
K 2184 16,963 24,700 43,648 78,406
Rb 15 43 81 195 320
Fe 5687 7815 15,428 24,039 135,436
Mg b1 1290 2083 3888 9610
Ca b2 626 2007 6012 25,574
S b100 1543 3389 7573 42,700
TOC 84 2106 3185 6528 15,324
CaCO
3
122 1570 6173 27,719 77,900
El Troncn (n=90) Cr b2 8.1 15 25 56
Zn b4 15 26 46 94
B 3.0 10 16 28 58
P 28 160 231 350 659
Mn 15 71 171 343 4390
V 9.1 32 47 75 154
Cu b2 2.5 6.3 8.9 23
Ti 794 2337 3499 5472 8835
Ni b6 4.2 12 11 57
Y 3.9 12 16 23 43
Sr 7.6 17 27 38 90
La b1 7.1 12 20 43
Ce 7.5 122 178 303 598
Ba 9.0 33 58 97 1266
Li 24 56 82 132 306
K 5895 13,535 20,439 33,350 51,725
Rb b10 45 80 149 289
Fe 3247 14,253 20,335 30,482 102,431
Mg 19 2078 2530 4135 6490
Ca 538 1655 2837 4188 36,875
S 112 537 1357 2935 31,100
TOC 16 1255 1756 3734 13,112
CaCO
3
289 2072 4007 5218 112,445
222 J.C. Montero-Serrano et al. / Sedimentary Geology 228 (2010) 218228
the La Paloma, 25% for the La Cuesta and 37% for the El Troncn) and
reects high variability in the data set. A detailed examination of the
samples reveals that the potential outliers mostly correspond to
sandstones, although a clear pattern is not evident. On the one hand, it
is well-known that outliers may seriously inuence statistical results,
particularly when based on calculations involving covariance (Pison
et al., 2003; Filzmoser and Hron, 2008); on the other hand, removing
those samples from the data set would excessively distort the sample
information available from the unit. Thus, we decided to only remove
the more extreme samples, those isolated points which visually do not
cluster together with the main cloud of points in the distance plot
(grey symbols in Fig. 4), and adopted a robust approach for sub-
sequent analysis. This implies the overall discard rate is about 9.3% of
all samples, which fulll the requirements suggested by Filzmoser et
al. (2009) for the robustness and efciency of the estimators. Note
that since a clear gap is observed between points over and under the
cut-off value for the La Cuesta section (Fig. 4), all the points over the
line are considered to be extreme samples in this case. Certainly, this
visual criterion may be debatable, but the choice of discarding or
accepting some other additional point does not signicantly affect the
nal conclusions, as we observed in preliminary tests (not reported
here), given the robustness of the approach. Those preliminary tests
comprised a series of tests in which either all potential outliers or only
extreme potential outliers were removed. This study was combined
with only considering either sandstones or shales, or both. In addition,
the data analysis was also run without excluding potential outliers.
But nal results got clearly worst in this latter case, especially re-
garding to the biplot explained variability.
Next, the data were centered to account for the relative weights of
the geochemical components by section (Fig. 5). From the centered
data set, the compositional geometric mean of every section is
compared by ratio to the global center [1/21,,1/21] of the simplex
S
21
. Note that in Fig. 5, the values have been expressed as logarithms
for a better visualization. A bar over the line for a certain component
indicates that the samples are characterized by a concentration over
the mean in that component. On the contrary, bars below the line
indicate lowrelative concentrations. Meaningful differences are found
between the geochemical signature typical carbonatic versus
typical siliciclastic chemical elements of the El Troncn (lower
Cerro Pelado Fm.) and La Paloma (upper Cerro Pelado Fm.) sections.
Also, in general, the relative geochemical pattern of the La Cuesta
section slightly resembles in typical siliciclastic chemical elements
(e.g., Cr, Mn, Ti, Y, La, Ce, Li, Fe) that of the La Paloma section. From
these results, the following aspects are inferred: (1) the conditions of
sedimentation (e.g., terrigenous inux, redox conditions) of the Cerro
Pelado Formation vary frombase to top. It seems that these conditions
were rather similar during the deposition of the sediments in the La
Cuesta and La Paloma sections; and (2) the variations in the relative
weight of the chemical components provide some evidence of strati-
graphic correlation between the La Cuesta and La Paloma sections.
Fig. 3. Percentages of values below the detection limit (VBDL) for the samples of the three sections studied.
Fig. 4. Multivariate outlier detection based on the robust Mahalanobis distance. The cut-
off value of 6 (squared robust MD) corresponding to 0.975 quartile (solid line). Grey
symbols indicate samples considered as extreme outliers, lled circles refer to
sandstones, and empty circles refer to shales.
223 J.C. Montero-Serrano et al. / Sedimentary Geology 228 (2010) 218228
Also from Fig. 5, it is clear that the La Cuesta section exhibits the most
heterogeneous chemical composition. However, the El Troncn
section presents more relatively homogeneous levels; that is, the
bars are not very long and they are generally of similar length for all
the components, with the exception of low levels of S and CaCO
3
. This
could be due to fact that, during the deposition of sediment into the El
Troncn section, the conditions of sedimentation remained relatively
constant until the deposition of sediment into the La Cuesta section
(middle or upper part of the Cerro Pelado Fm.), when these conditions
became much more variable (see discussion below).
Lastly, the relative variation pattern of the data is analyzed by
means of a global compositional biplot (Fig. 6A). Importantly, this
biplot give us an initial idea of the underlying geochemical associa-
tions within the data. Concerning the samples, a trend specic to any
section was not observed. Only the samples fromthe La Paloma and La
Cuesta show a slightly higher relative variability than those obtained
from the El Troncn. Regarding the chemical components, the biplot
shows a large set of log-centered components with low relative
variability which clearly clusters together (Cr, B, V, Ti, Y, La, Ce, Ba, Li,
K, Rb and Sr); this reects the fact that the variances of the cor-
responding log-ratios are relatively small. The elements of this group
are concentrated in the detrital fraction of the sediments (argillaceous
and siliciclastic). On the other hand, components mainly related to the
carbonate minerals present in the rock samples (CaCO
3
, Ca and Mg)
also lie close together, implying that the ratios between them are
moderately constant. Finally, certain one-dimensional variability
patterns are observed; such is the case for the component Mn,
which also shows the highest relative variability. It must be stressed
that the biplot based on the rst two principal components explains
only 56% of the total variability, so one must be cautious when
drawing conclusions from it.
5.2. Geochemical associations
The compositional biplot was used to explore meaningful
geochemical associations within each stratigraphic section. The biplot
analysis allows for the identication of at least three geochemical
associations (Fig. 6BD, Table 2):
(a) Detrital association (Cr, B, P, V, Ti, Y, La, Ce, Ba, Li, K, Rb, Sr). This
chemical group consists of elements associated with clay, silt or
sand particles, mainly as chemical constituents of the detrital
mineral grains (e.g., K-feldspar, kaolinite, illite, and muscovite;
see petrographic interpretations) or adsorbed on their surface.
The hydraulic partitioning of the sediments controls distribu-
tion and accumulation of these elements.
(b) Redox-organic matter association (TOC, S, Fe). These species
are directly related to the organic matter composition and
represent their accumulation, controlled by the oxicanoxic
conditions of the sedimentary environment. Generally in a
deltaic system, which was proposed for the Cerro Pelado
Formation, higher concentrations of these elements are present
in those stratigraphic intervals that settled under low detrital
inuence (e.g., marsh, oodplain) and resulted in a high
accumulation and preservation of organic matter.
(c) Carbonatic association (Ca, CaCO
3
, Mg): this chemical group
consists of elements related with the calciumcarbonate (calcite
and Mg-calcite) associated with the calcareous sandstones and
shales in the La Paloma section and with calcareous cement in
the sandstones, which agrees with the petrographic inter-
pretations. This association could be a proxy of the marine
inuence in deltaic environments.
The general knowledge about a geochemical system allows us to
establish a denite number of processes governing the sedimentary
system (redox, detrital input, source change, climate, among others).
In our case, given pre-existing knowledge of the sedimentary depo-
sitional environment, the information from the biplots (Fig. 6BD)
and the complementary petrographic interpretations, three is the
number of processes that can be inferred from a paleoenvironmental
point of view. Geochemically and statistically, the detrital association,
which appears in all the biplots, is the most important, grouping 12
13 components.
Fig. 5. Geochemical signature of the three studied sections of the Cerro Pelado Formation: relative weights of the 21 chemical components. A bar over the line for a certain component
indicates that the samples are characterized by a concentration over the geometric mean in that component. On the contrary, bars belowthe line indicate lowrelative concentrations.
224 J.C. Montero-Serrano et al. / Sedimentary Geology 228 (2010) 218228
The redox-organic matter association only appears in the La
Cuesta. According to geological evidence that indicates both a major
frequency and an increase in the thickness of the coal layers in this
stratigraphic level, this association is connected with more restricted
conditions (lack of oxygen) in the sedimentary environment. For the
La Paloma and El Troncn sections, such geochemical processes are
less dened; total organic carbon (TOC) appears in the corresponding
biplots within the detrital group, which suggests a main detrital origin
for the organic matter in these sections (e.g., debris of fossil plants; see
petrographic interpretations). An explanation for this fact lies in the
absence of well-dened redox conditions present in these sections.
Unclear associations are observed in all the sections (Table 2). A
paleoenvironmental interpretation for them is not simple, but petro-
graphic interpretations suggest a notable presence of products of
chemical weatheringanddigenesis inthe studiedsamples; e.g., gypsum,
jarosite, FeMn-oxyhydroxides, among others. These products may
have a major contribution in the statistical variability of the El Troncn
section, whichexplains andis reectedinthe lowvariability inits biplot.
Fig. 6. Robust compositional clr-biplots: (A) for the whole data set, and (BD) for each stratigraphic section.
225 J.C. Montero-Serrano et al. / Sedimentary Geology 228 (2010) 218228
Additionally, it is necessary to take into account that the chemical
behavior of the elements can be inuenced by more than one geo-
chemical and geological process, e.g., the K variation may be due to
changes in provenance, or both sorting (grain size) and diagenesis
(illitization).
Lastly, note that interpretation of the obtained biplots is not
straightforward; in the end, geochemical associations are not evident
at all. This may be due to the following factors. First, there could have
been partial overlapping of the sedimentary processes in time, which
is very common within the prograding delta environments proposed
for the Cerro Pelado Formation. Second, the total variability explained
by the biplots is not high; it ranges between 59% for the El Troncn
and 71% for the La Paloma. For these reasons, we only consider in this
work the clearest geochemical associations.
5.3. Chemofacies
Once the main geochemical associations for every section were
established, a constrained cluster analysis was carried out with the
purpose of dening a number of distinctive chemofacies. The dendro-
grams obtained allowed us to identify the detrital, the redox-organic
matter and the carbonatic chemofacies, respectively (Figs. 68). Some
inverted or distorted hierarchical levels are observed, due to the
constrained nature of the clustering process: only stratigraphically
adjacent samples can be clustered during the agglomerative procedure.
The number of nal chemofacies derived from the dendrograms
depends ona selectedcut-off distance (Gill et al., 1993). This value canbe
visually determined, taking into account that, on one hand, too few
chemofacies may not reveal clear regional patterns and, on the other
hand, considering too many chemofacies may cause regional patterns to
disappear. It is thus preferable to provide a good estimate of the optimal
number(s) of chemofacies, via comparison of their regional distribution.
The rather subjective selection of the cut-off point can be evaluated
having a general knowledge of the geological system involved in the
studied area (Retegui et al., 2005; Templ et al., 2008). In our case, we
consideredacut-off point of about 3.5toidentifydetrital chemofacies and
of 6 to identify both redox-organic matter and carbonatic chemofacies.
Since the El Troncn section does not show the redox and
carbonatic associations (Fig. 6B), only the chemofacies for the detrital
association are displayed (Fig. 7). Four detrital chemofacies are
identied (ETD1ETD4) from the bottom to the top of the log. In the
La Cuesta coal mine section (Fig. 8), four detrital chemofacies (LCD1
LCD4) are dened, as well as two redox chemofacies (LCR1, LCR2).
This section does not show the carbonatic association (Fig. 6C).
Finally, in the upper section (La Paloma), ten detrital (LPD1LPD10)
and two carbonatic (LPC1, LPC2) chemofacies are identied (Fig. 9).
Some remarks can be made about these results:
(a) Thebounds of thechemofacies varyinfunctionbasedontheextent
in time of the different physical or chemical processes responsible
for changes in detritic composition or redox conditions.
(b) In each log, the boundaries between detrital chemofacies do
not coincide in stratigraphic height with the corresponding
ones between redox or carbonatic chemofacies. The coinci-
dence of simultaneous changes in chemical attributes in a
single point of the log (e.g., provenance, redox conditions, and
environment) can be a marker indicative of an interruption of
the sedimentary record. In an uninterrupted sedimentary
succession, chemical changes are present before lithological
ones, because of the higher sensitivity to trace element com-
position towards these processes (redox, climate, and prove-
nance) than macroscopic lithological attributes (e.g., Retegui
et al., 2005). In general, in the three sections of the Cerro Pelado
Formation, nearly all the boundaries between detrital chemo-
facies appear at least one or two meters before a lithological
change (e.g., ETD1, ETD3, LCD1, LCD3, LPD2, LPD4, LPD8, LPD10,
and LPC2).
(c) The pronounced variation in detrital chemofacies in the La
Paloma section (ten chemofacies in 600 m stratigraphic height
versus four chemofacies in 440m stratigraphic height for the El
Troncn section) and the absence of a carbonatic association in
the El Troncn section illustrate a clear difference between the
lower and upper sections of the unit. The reasons for such a
change in the sedimentary style throughout the same unit obey
the pattern derived from sequential stratigraphy.
(d) The chemofacies ETD1 (Fig. 7B), located in the lowermost part
of the Cerro Pelado Formation, may indicate a facies transition
from the marine environment of the underlying Agua Clara
Formation to the very near-shore to coastal plain environment
(deltaic environments). Similarly, the chemofacies LPD10 and
LPC2, located in the top of the unit, may both suggest a facies
transition frommoderately shallowwater to deeper water, also
indicating the stratigraphic limits between the Cerro Pelado's
sandstones and the Querales' shales.
Table 2
Summary of the geochemical associations identied from the compositional biplot
analysis.
Section Detrital
association
Redoxorganic
matter association
Carbonatic
association
Unclear
associations
La Paloma Cr, B, P, V, Ti, Sr, Y, La,
Ce, Ba, Li, K and Rb
Ca, CaCO
3
and Mg
Fe, Zn
Mn
S
La Cuesta Cr, B, P, V, Ti, Sr, Y, La,
Ce, Ba, Li, K and Rb
TOC, S and Fe CaCO
3
Zn, Mg, Mn
Ca
El Troncn Cr, B, V, Ti, Sr, La, Ce,
Ba, Li, Rb, K and Mg
S
Y, P, Zn, Fe,
CaCO
3
Ca, Mn
Fig. 7. Dendrogramfor chemofacies identication in the El Troncn streamsection: cut-
off value is enclosed by a circle. Shaded areas refer to the different chemofacies
identied.
226 J.C. Montero-Serrano et al. / Sedimentary Geology 228 (2010) 218228
6. Concluding remarks
The compositional data methodology used in this paper reveals a
coherent and convenient standard procedure in the chemostrati-
graphic analysis of sedimentary sequences. It involves a series of
stages: (1) pre-treatment of data: missing values, values below the
detection limit and outliers; (2) biplot analysis for studying the
relative variation pattern of the components, as well as establishing
the number and type of geochemical processes involved during
sedimentation; and (3) constrained cluster analysis for determining
chemofacies on the basis of the components characterizing each
process. Furthermore, the presence of outliers in our data leads us to
adopt a robust approach to statistical analysis.
This analysis sheds light on the quantitative characterization and
historical evolution of the sedimentary sequence of the Cerro Pelado
Formation in the Falcn Basin, northwestern Venezuela. Therefore,
three main geochemical associations were established: (1) detrital
association (argillaceous and siliciclastic), (2) redox-organic matter
association and (3) carbonatic association. These geochemical asso-
ciations are consistent with both the depositional processes in deltaic
sedimentary environments and petrographic interpretations.
Changes from the bottom (El Troncn section) to the top (La
Paloma section) of the Cerro Pelado Formation were found by
identifying detrital, redoxorganic matter or carbonatic chemofacies
for each section. We observed that the extent and bounds of detrital
chemofacies are independent of those of the redox and carbonatic
Fig. 8. Dendrograms for chemofacies identication in the La Cuesta coal mine section: cut-off value is enclosed by a circle. Shaded areas refer to the different chemofacies identied.
Fig. 9. Dendrograms for chemofacies identication in the La Paloma stream section: cut-off value is enclosed by a circle. Shaded areas refer to the different chemofacies identied.
227 J.C. Montero-Serrano et al. / Sedimentary Geology 228 (2010) 218228
chemofacies in nearly all cases. Finally, exploratory analysis reveals
a certain chemostratigraphic similarity between the La Cuesta and
La Paloma sections on the basis of the relative weights of the geo-
chemical components.
Acknowledgements
This work was supported by: (1) the Consejo de Desarrollo
Cientco y Humanstico (CDCH) de la Universidad Central de
Venezuela, through the projects PI-03.30.4702.2000 and PG-
03.32.4412.1999; (2) the Laboratoire Gosystmes (FRE 3298 CNRS)
of the Universit Lille 1 (France); (3) the Programme Alban, the
European Union Programme of High Level Scholarships for Latin
America, scholarship No. E06D100913VE; and (4) the Spanish
Ministry of Education and Science under two projects: Ingenio
Mathematica (i-MATH), Ref. No. CSD2006-00032, and CODA-RSS,
Ref. MTM2009-13272. Comments and helpful suggestions made by
the editor Gert Jan Weltje and the reviewers Dr. Barry Roser and Dr.
Hilmar von Eynatten greatly enhanced the quality of this manuscript.
References
Aitchison, J., 1986. The statistical analysis of compositional data. Monographs on
statistics and applied probability. Chapman & Hall Ltd., London; reprinted in 2003
by the Blackburn Press, New Jersey, 416 p.
Aitchison, J., 1990. Relative variation diagrams for describing patterns of compositional
variability. Math. Geol. 22 (2), 487511.
Aitchison, J., Egozcue, J.J., 2005. Compositional data analysis: where are we and where
should we be heading? Math. Geol. 37 (7), 829850.
Aitchison, J., Greenacre, M., 2002. Biplots of compositional data. Appl. Stat. 51, 375392.
Aitchison, J., Barcel-Vidal, C., Martn-Fernndez, J.A., Pawlowsky-Glahn, V., 2000. Log-
ratio analysis and compositional distance. Math. Geol. 32 (3), 271275.
Audemard, F.A., 2001. Quaternary tectonics and present stress tensor of the inverted
northern Falcon Basin, northwestern Venezuela. J. Struct. Geol. 23 (23), 431453.
Barbera, G., Lo Giudice, A., Mazzoleni, P., Pappalardo, A., 2009. Combined statistical and
petrological analysis of provenance and diagenetic history of mudrocks: applica-
tion to Alpine Tethydes shales (Sicily, Italy). Sed. Geol. 213, 2740.
Bellanca, A., Claps, M., Erba, E., Masetti, D., Neri, R., Premoli, I., Venezia, F., 1996.
Orbitally induced limestone/marlstone rhythms in the Albian-Cenomanian Cismon
section (Venetian region, northern Italy): sedimentology, calcareous and siliceous
plankton distribution, elemental and isotope geochemistry. Palaeogeogr. Palaeo-
climatol. Palaeoecol. 126, 227260.
Bezada, M.J., Schmitz, M., Jcome, M.I., Rodrguez, J., Audemard, F., Izarra, C., the
BOLIVAR Active Seismic Working Group, 2008. Crustal structure in the Falcn
Basin area, northwestern Venezuela, from seismic and gravimetric evidence.
J. Geodynamics 45 (4-5), 191200.
Borsier, M., 1991. Automated multielement analysis of geological materials-sample
decomposition and inductively coupled plasma (ICP-AES) procedures. Spectro-
chim. Acta Rev. 14 (1), 7994.
Condie, K.C., 1993. Chemical composition and evolution of upper continental crust:
contrasting results from surface samples and shales. Chem. Geol. 104, 137.
Das, N., 1997. Chemostratigraphy of sedimentary sequences: a reviewof the state of the
art. J. Geol. Soc. India 49, 621628.
Third compositional data analysis Workshop-CoDaWork'08. In: Daunis-i-Estadella, J.,
Martn-Fernndez, J.A. (Eds.), Proceedings, University of Girona, Girona (Spain),
CD-ROM. ISBN: 978-84-8458-272-4. available at http://ima.udg.es/Activitats/
CoDaWork08.
Dayong, J., Weicheng, H., Shunliang, B., 1999. Relationship between Milankovitch
eccentricitycyclicityandchemo-cycles fromtheupper Givetian(Devonian) inGuangxi,
South China. Chin. Sci. Bull. 44, 16971700.
Daz de Gamero, M.L., 1977. Estratigrafa y Micropaleontologa del Oligoceno y Mioceno
Inferior del centro de la Cuenca de Falcn, Venezuela. Geos 22, 360.
Daz de Gamero, M.L., 1989. El Mioceno temprano y medio en Falcn septentrional.
Geos 29, 2535.
Daz de Gamero, M.L., Linares, O.J., 1989. Stratigraphy and paleontology of the Urumaco
Formation from the upper Miocene in northwestern Falcon State. VII Congreso
Geolgico Venezolano, Mem. I, Maracaibo, 419-438.
Egozcue, J.J., Pawlowsky-Glahn, V., 2005. Group of parts and their balances in
compositional data analysis. Math. Geol. 37 (7), 795828.
Egozcue, J.J., Pawlowsky-Glahn, V., Mateu-Figueras, G., Barcel-Vidal, C., 2003.
Isometric logratio transformations for compositional data analysis. Math. Geol.
35 (3), 279300.
Filzmoser, P., Hron, K., 2008. Outlier detection for compositional data using robust
methods. Math. Geosci. 40 (3), 233248.
Filzmoser, P., Hron, K., Reimann, C., 2009. Principal component analysis for com-
positional data with outliers. Environmetrics 20, 621632.
Ghosh, S., Pestman, P., Melndez, L., Truskowski, I., Zambrano, E., 1997. Evolucin
tectonoestratigrca y sistemas petrolferos de la Cuenca de Falcn, Venezuela
Noroccidental. Proceedings VIII Congreso Geolgico Venezolano, vol. 1, pp. 317329.
Gill, D., Shomrony, A., Fligelman, H., 1993. Numerical zonation of Log suites and Log
facies recognition by multivariate clustering. AAPG Bull. 77, 17811791.
Gonzlez, E., Nieves, G., Torrealba, N., 1985. Geological characteristics of the Cerro
Pelado carboniferous basin. VI Congr. Geol. Venez., Mem. 6, Caracas, pp. 39663998.
Hambalek, N., Rull, V., Digiacomo, E., Daz de Gamero, M., 1994. Evolucin
paleoecolgica y paleoambiental de la secuencia del Negeno en el Surco de
Urumaco, Estado Falcn. Estudio palinolgico y litolgico. Boletn de la Sociedad
Venezolana de Gelogos 19, 719.
Jarvis, I., Moreton, J., Grard, M., 1998. Chemostratigraphy of Madeira Abyssal Plain
Miocene-Pleistocene turbidites, site 950. Proceedings Ocean Drilling Program,
Scientic Results 157, 535558.
Macellari, C., 1995. Cenozoic sedimentation and tectonics of the southwestern
Caribbean pull-apart basin, Venezuela and Colombia. AAPG Bull. 62, 757780.
Martn-Fernndez, J.A., Barcel-Vidal, C., Pawlowsky-Glahn, V., .Kovcs, L., Kovcs, G.P.,
2005. Subcompositional patterns in Cenozoic volcanic rocks of Hungary. Math. Geol.
37 (7), 729752.
Palarea-Albaladejo, J., Martn-Fernndez, J.A., 2008. A modied EM alr-algorithm for
replacing rounded zeros in compositional data sets. Comput. Geosci. 34, 902917.
Palarea-Albaladejo, J., Martn-Fernndez, J.A., Gmez-Garca, J., 2007. A parametric
approach for dealing with compositional rounded zeros. Math. Geol. 39 (7), 625645.
Pawlowsky-Glahn, V. (guest ed), 2005. Special issue: Advances in compositional data.
Math. Geol. 37(7), 671-850.
Pawlowsky-Glahn, V., Egozcue, J.J., 2002. BLU estimators and compositional data. Math.
Geol. 34 (3), 259274.
Pawlowsky-Glahn, V., Egozcue, J.J., Tolosana-Delgado, R., 2007. Lecture notes on
compositional data analysis. Universitat de Girona, Girona (Spain. 87 pp.; available
at http://dugi-doc.udg.edu/bitstream/10256/321/1/CoDa-book.pdf.
Pearce, T.J., Besly, B.M., Wray, D.S., Wright, D.K., 1999. Chemostratigraphy: a method to
improve interwell correlation in barren sequences a case study using onshore
Duckmantian/Stephanian sequences (West Midlands, U.K.). Sediment. Geol. 124,
197220.
Pettijohn, F.J., Potter, P.E., Siever, R., 1987. Sand and Sandstone, 2nd ed. Springer-Verlag,
Berlin. 553 pp.
Pison, G., Rousseeuw, P., Filzmoser, P., Crouxc, C., 2003. Robust factor analysis. J. Multivariate.
Anal. 84, 145172.
Retegui, K., Martnez, M., Esteves, I., Gutirrez, J.V., Martnez, A., Melndez, W., Urbani,
F., 2005. Geochemistry of the Mirador Formation (Late Eocene-Early Oligocene),
south western Venezuela: chemostratigraphic constraints on provenance and the
inuence of the sea level. Geochem. J. 39 (3), 213226.
Spalletti, L.A., Queralt, I., Matheos, S.D., Colombo, F., Maggi, J., 2008. Sedimentary
petrology and geochemistry of siliciclastic rocks from the upper Jurassic Tordillo
Formation (Neuqun Basin, western Argentina): implications for provenance and
tectonic setting. J. South Am. Earth Sci. 25 (4), 440463.
Taylor, S.R., McLennan, S.M., 1985. The continental crust: its composition and evolution.
Blackwell, London. 312 pp.
Templ, M., Filzmoser, P., Reimann, C., 2008. Cluster analysis applied to regional
geochemical data: problems and possibilities. Appl. Geochem. 23 (8), 21982213.
Tolosana-Delgado, R., von Eynatten, H., 2009. Grain-size control on petrographic
compositionof sediments: compositional regression androunded zeros. Math. Geosci.
41, 869886.
Villamil, T., 1996. Depositional and geochemical cyclicity in the cretaceous ne-grained
stratus of Colombia. A model for organic matter content: Ciencia, Tecnologa y Futuro,
vol. 1, pp. 523.
von Eynatten, H., 2004. Statistical modelling of compositional trends in sediments. Sed.
Geol. 171, 7989.
von Eynatten, H., Barcel-Vidal, C., Pawlowsky-Glahn, V., 2003. Sandstone composition
anddiscrimination: a statistical evaluationof different analytical methods. J. Sediment.
Res. 73 (1), 4757.
Weltje, G.J., 2002. Quantitative analysis of detrital modes: statistically rigorous
condence regions in ternary diagrams and their use in sedimentary petrology.
Earth Sci. Rev. 57, 211253.
Winchester, J., Max, M., 1996. Chemostratigraphic correlation, structure and sedimen-
tary environments in the Dalradian of the NW Co. Mayo inlier, NW Ireland. J. Geol.
Soc. Lond. 153, 779801.
228 J.C. Montero-Serrano et al. / Sedimentary Geology 228 (2010) 218228