Research Article
Spatial variability of sedimentological properties
in a large Siberian lake
Thomas Kumke1,
*,Annemarieke Schoonderwaldt2
and Ulrike Kienel3
1
Alfred-Wegener-Institute for Polar and Marine Research, Research Department Potsdam, Telegrafenberg A43,
14473 Potsdam, Germany
2
University of Leipzig, Department of Geology, Talstrasse 35, 04103 Leipzig, Germany
3
GeoForschungsZentrum Potsdam, Telegrafenberg C327, 14473 Potsdam, Germany
Received: 15 January 2004; revised manuscript accepted: 19 August 2004
Abstract. The analysis of spatial variability of lake sedi-
ment properties in Lake Lama, Central Siberia is pre-
sented. The aims were to characterize the spatial structure
in lake sediment composition, to determine major spatial
patterns of sediment properties by calculating estimates
at locations where no samples were available, and to assess
the main processes determining these patterns. Sediment
properties were measured at 71 spatially distributed loca-
tions in the lake, comprising particle size distribution,
biogeochemical properties such as total organic carbon,
total nitrogen or stable carbon isotopes, as well as geo-
chemical distribution of major and trace elements. Spatial
analysis consisted of a principal components analysis
and, subsequently, calculation and modelling of vario-
grams to describe the spatial structure in the data as well
as determination of spatial estimates using block-kriging
to map spatial structure. The results showed that particle
Aquat. Sci. 67 (2005) 86­96
1015-1621/05/010086-11
DOI 10.1007/s00027-004-0734-5
 EAWAG, Dübendorf, 2005
Aquatic Sciences
size, biogenic silica, CaO, d13
Corg, Zr and TOC/TN ratio
were a major source of variability in sediment properties
and, thus, dominate the sediment structure in Lake Lama.
Major anomalies occurred near river inlets, depending on
river size and its position as well as slope of the river
bed. Other anomalies were associated with water depths,
morphology of the lake basins, and wind-induced cur-
rents and re-suspension in the shallow part of the western
basin. The spatial structure in sediment properties indi-
cate that several processes act at different spatial scales.
Moreover, there was a considerable amount of small-
scale variability that could not be quantified due to
sampling design. The results showed that heterogeneity
in lake sediment composition is a main characteristic
of large lake systems, and must be taken into account,
especially in paleolimnological and environmental appli-
cations.
Key words. Geochemistry; heterogeneity; lake sediments; Siberia; spatial structure; variogram.
Introduction
Spatial heterogeneity in sedimentological, geochemical
and biogeochemical properties is one main characteristic
of lake sediments. The spatial distribution in these proper-
ties depends on processes such as current, turbulence,
chemical precipitation and turbidity, on physical factors
like lake morphology, and on catchment characteristics
like underlying geology, the number and size of inflows,
meltwater input and land use (Hkanson and Jansson,
1983; Ashley, 1995). In general, the spatial variability in
sediment composition is thought to increase with lake
size and the number and size of the inflows, or in other
words, with the increasing number of processes affecting
the lake (Pinel-Alloul, 1995; Kienel and Kumke, 2002).
Most often, the variability in lake systems is structured,
meaning that spatially distributed data are autocorrelated
(i.e., a spatially ordered variable is correlated with itself).
* Corresponding author phone: 00493312882165;
e-mail: tkumke@awi-potsdam.de
Published on Web: March 2, 2005
Aquat. Sci. Vol. 67, 2005 Research Article 87
Spatial autocorrelation is often strong for data points
closely spaced and weak or even non-significant for data
points far apart. For example, Kienel and Kumke (2002)
showed that the distribution in fossil diatom composition
is spatially autocorrelated up to 55 km in Lake Lama,
Central Siberia, depending on water depth and river in-
put. Lebo and Reuter (1995) suggested a dependency of
organic C, N, P and particle size fractions on water depth
of a large lake using data from 32 surficial sediment
samples. While the studies of Lebo and Reuter (1995)
and Kienel and Kumke (2002) were carried out on large,
deep lakes, the work of Anderson (1990; 1998) implies
that the spatial variability in sedimentary diatom com-
position and accumulation rate also occurs in small lakes
with simple morphology. Korsman et al. (1999) discussed
the variability in surface sediment composition of a small
Swedish lake shown by near-infrared spectroscopy (NIR)
and loss-on-ignition (LOI), revealing spatial patterns in-
fluenced by wastewater input, clear-cut forest and trans-
port of allochthonous material.
The methodology of analyzing spatial variation in
sediment properties in each study differed. For example,
Korsman et al. (1999) used principal component scores
of NIR spectra to draw contour maps for interpretation.
Kienel and Kumke (2002) used a more sophisticated
analysis of spatial variability in lake sediments, based on
geostatistical analysis of sample scores from correspon-
dence analysis (CA). Geostatistics is an effective method
to analyze spatial variation and has been widely used in
geology, hydrology, soil science and ecology (Russo, 1984;
Webster, 1985; Legendre and Fortin, 1989; Ellsworth and
Boast, 1996; Kumke et al., 1999; 2002). The aim of geo-
statistical analysis is twofold: (i) calculation and model-
ling of spatial structure in data by means of variograms,
and (ii) calculation of estimates at locations where no
data are available using the kriging technique, an optimal-
weighted interpolation technique (for a review of geosta-
tistical methods see Isaaks and Srivastava, 1989; Webster
and Oliver, 2000). The main advantage of geostatistical
analysis is that scale and direction of autocorrelation in
data are accounted for.
Analysis of sedimentological measurements is com-
plex because data consist of a multivariate ensemble of
variables that are often interrelated. Ordination techni-
ques provide an effective tool to reduce the dimensional-
ity in such data (Legendre and Legendre, 1998). If data
are quantitative and their relationships are presumed to
be linear, principal components analysis (PCA) is an ap-
propriate ordination technique (Ter Braak and Prentice,
1988). Principal components analysis is an eigenvector
technique that reduces the multidimensional data set to
a few principal axes, thereby decomposing the total va-
riance in the data. The first few axes represent most of
the variance in the data. In PCA, eigenvectors and prin-
cipal component scores are calculated and the latter can
be used for further spatial analysis as carried out, for ex-
ample, in Korsman et al. (1999) and Kienel and Kumke
(2002).
Generally, the sediment composition of lakes pro-
vides useful information about regional changes in en-
vironmental history, such as land use changes in the catch-
ment, eutrophication or pollution history. Examination of
spatial variation allows one to assess the different pro-
cesses affecting sediment variability and their impact on
the lake and its habitats. In terms of paleolimnological
considerations, quantifying the spatial variability in sur-
face sediment composition is important to estimate the
representativity of paleolimnological studies.
Here, the surface sediment composition of a large lake
in Central Siberia was studied using measurements of bio-
geochemistry, particle size distribution and geochemical
element distribution at 71 locations spatially distributed
in the lake. The aim of this paper is threefold: (i) to deter-
mine and characterize the spatial structure in lake sedi-
ment composition, (ii) to estimate the sediment composi-
tion at locations where no measurements were available,
thereby obtaining major spatial patterns in sediment pro-
perties, and (iii) to assess primary physical and hydro-
logical processes that determine spatial patterns. A sound
knowledge about these processes is required for further
studies on the lake to solve paleolimnological, paleocli-
matological and ecological problems. Lake Lama is an
excellent example on the complexity of features in large
lakes and their influence on lake sediment characteristics.
This paper shows how to analyze such features using
geostatistical techniques new to the field of lake sedi-
mentology.
Study area
Lake Lama (69°33N, 90°13E) is located in the north-
west foothills of the Putoran Plateau near the city of
Norilsk, Central Siberia. The lake is of tectonic origin,
about 80 km  13 km in size and up to 254 m deep. It con-
sists of an eastern (II) and a western basin (IV) connected
by a north-south trending subbasin (III), and an eastern
tip (I; Fig. 1). It is surrounded by steeply-sloped moun-
tains to the north and south that reach altitudes of
1234 m a.s.l.
The largest part of the catchment is on Late Permian
to Late Triassic Continental Flood Basalts (CFBs) and
associated tuffs of the Siberian Trap. The CFBs have
been subdivided into 11 lava types, suites or formations
(Lightfoot et al., 1990). The subdivisions were based on
differences in geochemical characteristics among the
suites, in CaO and MgO content. The geochemistry is
generally independent of textural properties. A range of
mafic to ultra-mafic intrusions have been formed in the
region, and are generally rich in nickel, copper, sulfide
and platinum.The geology of the western part of the catch-
88 T. Kumke, A. Schoonderwaldt and U. Kienel Spatial variability of sediment composition
ment is dominated by Quaternary lacustrine, glacifluvial,
glacial and alluvial deposits. In the western catchment
occurs some outcrops of Lower Palaeozoic to Late Per-
mian sedimentary rocks formed in a structural depression
called the Tunguska Syncline. The sequence contains
marine dolomites, limestones and agillites, calcareous and
dolomitic marls, sulfate-rich evaporites, shallow water
limestones, and lagoonal and continental sediments (e.g.,
siltstone and sandstone; Lightfoot et al., 1990). Structur-
ally, the area is characterized by several pre- and post-trap
positive and negative structures (uplift forms and basins)
and fault systems of different orientation (Naldrett et al.,
1992).
The regional climate around Lake Lama is continen-
tal. Annual temperatures fluctuate considerably. Average
annual air temperature is ­9.8°C with mean January air
temperatures between ­25 and ­35°C and mean July air
temperatures ranging from 10 to 15°C. Annual precipi-
tation is high compared to other areas in Central Siberia
(Harwart, 1999).The main wind direction is east to north-
east (Hagedorn et al., 1999). Lake Lama is ice-covered
from October to May. Thawing starts in May and takes
about one month. Lake Lama can be classified as a sub-
polar dimictic to polar monomictic lake type based on
water column temperature profiles (Kienel and Kumke,
2002). The catchment is about 6210 km2
. More than
10 large inflows enter Lake Lama, the largest being the
Mikchanda River with a delta on its north-west shore
(Fig. 1). The lake drains west via Lake Melkoye into Lake
Pyasino. The inundated shore line reaches up to 10 m in
width and is covered with pebbles, gravel and sand. At
high lake levels, vegetation near the shoreline is sub-
merged. Sparsely distributed macrophytes (Chara sp.)
were observed in well-illuminated places.
Vegetation of the study area is forest tundra, and is
dense and dominated by Taiga trees and shrubs. The taiga
forest consists of spruce (Picea obovata), larch (Larix
czekanovskii and L. sibirica) and birch (Betula pubes-
cens). The upper forest boundary is located between
200­400 m a.s.l. (Galaziy and Parmuzin, 1981), where
larch and shrubs form an open forest (Andreev et al.,
2004). Shrubs include dwarf birch (Betula nana), shrub
alder (Alnus fruticosa), juniper (Juniperus communis)
and willow (Salix spp.).
Methods
Sediment sampling and measurements
A total of 71 surface sediment samples (Fig. 1) were col-
lected during summer 1997. Sediment coring took place
from a swimming platform using a light gravity corer
(Kajak et al., 1965) that allowed undisturbed recovery of
soft upper sediments. The core samples containes 2 cm
Figure 1. Map of Lake Lama and the coring locations. The values given on the isolines are water depths [m]. Roman numbers denote
subregions of the lake as used in the text: I ­ eastern tip; II ­ eastern basin; III ­ north-south trending subbasin; IV ­ western basin. The
vertical lines indicate the borders between the subregions.
Aquat. Sci. Vol. 67, 2005 Research Article 89
of surface sediments. Sampling positions were estimated
using a Geographic Positioning System (Trimble Scout/
Scoutmaster GPSTM
accurate to 100 m).
All analytical procedures were carried out on freeze-
dried samples. Samples were oxidized and disaggregated
by means of 3­10% H2O2 solution for estimating the
particle size distribution. The sand fraction was separated
by wet sieving (63 mm mesh). Clay and silt fractions were
separated by an Atterberg cylinder, which uses differen-
ces in settling velocity between the fractions.
Total carbon (TC), total organic carbon (TOC), total
nitrogen (TN) and total sulphur (TS) fractions as well as
stable isotope composition of organic carbon (d13
Corg)
were analyzed. The TC, TN and TS fractions were mea-
sured on homogenized subsamples (10 mg) by a CNS-
932 micro-analyzer (LECO) that combusts samples at
1100 °C. The TOC fraction was determined by combust-
ing samples at 1400°C using a Metalyt-CS analyzer
(ELTRA). Before measurement, samples were treated with
hydrochloric acid to remove carbonates. The carbonate
fraction was estimated by multiplying the difference
between TC and TOC by a constant c = 8.33. The d13
Corg
contents were measured using an elemental analyzer
(HERAEUS) and MAT Delta S mass-spectrometer
(FINNIGAN).
Major and trace elements of sediment were mea-
sured by X-ray fluorescence spectrometry (SRS 3000,
SIEMENS). Measurements were carried out on glass
tablets produced in a fused mass method using 0.6 g
crushed, pre-dried and glowed material. Major elements
included Si, Ti, Al, Fe, Mn, Mg, Ca, Na, K and P, whereas
trace elements included Ce, Co, Cr, Cu, Nb, Ni, Pb, Rb,
Sr, V, Y, Zn and Zr. The biogenic silica content was esti-
mated using spectrophotometry. Since all elements were
measured as fractions of sediment, they were normalized
to the Al fraction to obtain absolute variations (Harwart,
1999).
Statistical methods
Prior to ordination, all variables were examined for ex-
treme values and statistical distributions. Extreme values
were removed when they deviated by more than four stan-
dard deviations (SD) from the mean. Since variables are
assumed to be normally distributed in ordination tech-
niques, variables with skewed distributions were trans-
formed using an appropriate transformation (Webster and
Oliver, 2000). Principal components analysis (PCA) was
performed to reduce the dimensionality and to simplify
the variation structure in the data. For PCA, the matrix Y
of the measured variables was converted into a correla-
tion matrix R because of the different dimensions of the
data (Legendre and Legendre, 1998). The principle axes
of variation were found by solving the characteristic
equation for R and calculating the eigenvalue l for each
principle axis k as well as the eigenvector uk associated
to lk. The position of sites (i.e., the principal component
scores) are a linear combination of the standardized val-
ues for each variable and the eigenvectors, that is:
yij ­ y­
j
F = 0 U (1)
SDj
where yi is the measured value for each variable j, SDj is
the standard deviation and U is the matrix of eigenvectors.
The principal component scores F are weighted averages
of all variables in the system. Principal component scores
of the first axis are most representative of the data by ac-
counting for most of the variance. The results of PCA
were analyzed using a graphical biplot of U and F of two
principle axes.
Analysis of spatial variability was carried out by vario-
gram modeling of the principal component scores fik
for axis k derived from PCA. The omnidirectional semi-
variances were estimated using:
1 N(h)
g(|h|) = 01  {z(xi) ­ z(xi + h)}
2
(2)
2N(h) i=1
where g is the semi-variance, h is the lag distance (m),
N (h) is the number of pairs at each lag distance, and z is
the value at the sampling point xi (i.e., the principal com-
ponent scores fik for axis k). The lag distance is the sepa-
ration distance between two sampling points from which
the squared differences were calculated. The experimental
variogram was plotted using the calculated semi-varian-
ces g versus each lag distance h (Webster and Oliver,
2000). Theoretical variogram functions were fitted to
experimental variograms using a least-squares approach.
Variogram models were examined using cross-validation
techniques (Russo, 1984). Cross-validation was perform-
ed by excluding a sampling point, estimating a value for
this point, and calculating the residual. The mean square
of residuals is the mean squared prediction error (MSPE)
and serves as a measure of accuracy for the variogram
model (Kumke, 1999).
Subsequent mapping of principal component scores
of axis k was achieved by obtaining estimates of principal
component scores using block kriging. Kriging is an
unbiased interpolation method in which estimates of a
variable z* at unknown positions are weighted linear
combinations of z (xk):
n
z*(x0) =  wiz(xi) (3)
i=1
where wi denotes the weight given to each principal com-
ponent score. To derive unbiased estimates, the weights
sum to unity. Weights depend on the shape and magnitude
of the variogram model (Isaaks and Srivastava, 1989).
For block kriging, the spatial field was decomposed into
90 T. Kumke, A. Schoonderwaldt and U. Kienel Spatial variability of sediment composition
blocks of 700 m  350 m to derive average block estim-
ates of z (xk).
All numerical analyses were carried out using
CANOCO 4.0 (Ter Braak and Smilauer, 1998) for ordi-
nation, and GEOPACK 1.0 (Yates and Yates, 1990) and
GEOEAS 1.2.1 (Englund and Sparks, 1991) for vario-
gram analysis and kriging, respectively.
Results and discussion
Summary of the sedimentological measurements
Summary statistics of surface sediment properties of Lake
Lama are given in Table 1. According to the average par-
ticle size distribution, the sediment can be characterized
as a silty loam. However, coefficients of variation (CVs)
are, especially for the sand fraction, quite large and show
a remarkable local variability. A high local variability of
the sand fraction is indicated by the large ratio between
the mean and median, which is typical if there are some
extremely large values outside the population distribu-
tion. These large values of the sand fraction were observ-
ed near river mouths in the steep sides along the eastern
part of the lake and caused by deposition of inserted
coarse material.
In general, biogenic components were characterized
by relatively low fractions of TC, TN and TS. Their
CVs were quite similar (ca. 40 to 50%) and common for
Table 1. Summary statistics for (a) biogeochemical and sedimentological, and (b) geochemical measurements. All percentages are related
to dry mass.
Variable Unit (a)
Mean Median SD CV Skewness
Siam % 6.01 6.07 2.62 43.54 0.26
TC % 1.13 1.05 0.59 51.95 2.52
TN % 0.10 0.10 0.04 37.51 ­0.31
TOC % 1.14 1.08 0.59 51.79 2.29
CaCO3 % 0.17 0.00 0.32 190.81 2.21
TS % 0.01 0.01 0.01 52.22 1.65
d13Corg  V-PDB ­26.31 ­26.67 1.24 4.70 1.93
TOC/TN 12.20 10.83 3.84 31.46 1.73
Sand % 17.23 2.35 24.69 143.25 1.28
Silt % 54.88 59.65 18.84 34.33 ­0.92
Clay % 27.60 26.98 12.06 43.69 ­0.08
Variable Unit (b)
Mean Median SD CV Skewness
SiO2 % 22.74 22.75 0.67 2.95 ­0.27
TiO2 % 0.69 0.69 0.05 7.92 0.54
Al2O3 % 4.01 4.05 0.19 4.71 ­0.81
Fe2O3 % 3.49 3.46 0.20 5.77 0.21
MnO % 0.16 0.15 0.04 22.12 3.53
MgO % 3.63 3.62 0.32 8.74 0.43
CaO % 6.31 6.26 0.81 12.91 0.16
Na2O % 0.67 0.67 0.06 9.35 1.33
K2O % 0.29 0.29 0.06 19.04 0.82
P2O5 % 0.03 0.03 0.01 21.70 0.56
Loi % 5.41 5.50 1.18 21.78 ­0.32
Ba ppm 253.76 247.00 40.34 15.90 0.71
Ce ppm 72.75 72.00 14.28 19.63 0.39
Co ppm 49.37 49.00 4.46 9.04 0.13
Cr ppm 160.33 154.00 22.70 14.16 0.96
Cu ppm 95.91 96.00 33.86 35.30 ­0.61
Nb ppm 12.43 12.00 1.97 15.88 0.61
Ni ppm 88.12 89.00 9.84 11.17 ­0.79
Pb ppm 30.88 22.50 19.16 62.07 1.40
Rb ppm 22.72 22.00 5.29 23.27 0.55
Sr ppm 226.31 227.00 19.26 8.51 ­0.14
V ppm 242.94 234.00 30.79 12.67 0.62
Y ppm 15.49 15.00 2.20 14.23 0.47
Zn ppm 94.57 93.00 12.89 13.63 1.36
Zr ppm 98.51 98.00 11.18 11.35 0.24
Aquat. Sci. Vol. 67, 2005 Research Article 91
field-measured quantities (Kumke et al., 1999; 2002).
The mean to median ratio were approximately 1.0, indi-
cating no extreme values. The TOC/TN ratio had mean
and median values that were in the range typical of au-
tochthonous-derived organic matter (Meyers and Ishiwa-
tari, 1995). However, there were some localities with
values >20 that might be caused by the introduction of
allochthonous material (Meyers and Ishiwatari, 1995).
The geochemical element distribution had, in general,
relatively low CVs with a few exceptions. The ratio be-
tween the mean and median is near unity for most ele-
ments except for Cu and Pb. Both, Cu and Pb had extreme
values at some localities in the eastern basin, thus explain-
ing their crelatively large CVs.
Biplot analysis
The PCA of all sedimentological measurements resulted
in a distance biplot of the first two principal axes shown
in Fig. 2. The first and second axis explained 34% and
18% of the variance, respectively. Sites were classified
according to their spatial position in the basins of the
lake.There were no structuring of sites along the first axis
in respect to that classification. However, an interesting
feature was that most sampling sites from the western
basin were located in the second and third quadrant of the
plot, indicating the western basin has sediments high in
silt and clay as well as being rich in some biogeochemical
components such as biogenic silica (Siam) and TN. Va-
riables that contribute most to the first axis were particle
size fractions, Siam, TN, the ratio of TOC/TN, and the
major elements Na2O, TiO2, CaO, MgO, Fe2O3 and P2O5.
Principal component scores were well-structured accord-
ing to the classification criterium along the second axis.
Almost every sample from the western basin had a posi-
tive principal component score for PC2, whereas samples
of the other three basins were located in the third and
forth quadrant of the plot. Thus, basin morphology and
water depths determined a large extent of the second axis.
Additionally, there was a significant negative correlation
between principal component scores of the second axis
and water depth (r = ­0.36, P  0.05) supporting this
observation. The most important variables, which covary
with the second axis, were d13
Corg, K2O, SiO2, the ratio of
TOC/TN, and the trace elements Ba, Zr, Nb, Ce and Cu.
Some of these (such as d13
Corg and Ba) are partially cor-
related to water depth.
The third and the fourth principal axes represented
9.6% and 6.2% of the explained variance, respectively.
However, biplot analyses using the third and fourth axes
(not shown) yielded no structure.
Variogram analysis
Using principal component scores of the first two axes, a
variogram analysis was carried out. Both variograms are
spatially structured (Fig. 3) and have typical variogram
features, i.e., a range (part of the curve where the semi-
variances increase with increasing lag distance), a sill
(part of the curve where the semi-variance is at its maxi-
mum and independent of h) and a nugget variance (i.e.,
semi-variance at h = 0). The variogram of the first axis
scores has small semi-variances at short lags that rise to
values independent of h at a length of about 5 km. There
is a decrease in semi-variances from 30 km to 36 km fol-
lowed again by an increase. This phenomenon is the so-
called `hole effect' (Webster and Oliver, 2000) and most
often has some physical reason. Variogram analysis re-
vealed that sampling points in the eastern tip and eastern
basin near river mouths (locations with large positive
principal component scores) contributed to the low semi-
variances at distances of 30 and 32 km. At lags of about
34 and 36 km, sampling points in the eastern basin (loca-
tions with negative principal component scores) separated
by these distances contributed largely to the semi-vari-
ances. Physical processes leading to the increased simi-
larities in sediment properties at these lags are not clear
but they are partly related to riverine input of material.
The nugget variance is considerably high, leading to a
nugget to sill ratio of 34%. Thus, about one third of the
Figure 2. Distance biplot of PCA conducted on sediment proper-
ties of Lake Lama. The first and second PCs are plotted alongside
with the contribution of each variable. All variables are shown that
fit into the variable space with 10% of the variance explained.
Arabian numbers indicate the number of the quadrant used in the
text.
92 T. Kumke, A. Schoonderwaldt and U. Kienel Spatial variability of sediment composition
variation is related to random fluctuations, measurement
errors or small-scale processes beyond our scale of mea-
surement (Webster and Oliver, 2000; Kienel and Kumke,
2002). Since there are some biogeochemical variables
such as biogenic silica and TN contributing to that vario-
gram, biotic processes like small-scale variations in prim-
ary productivity or nutrient uptake by aquatic organisms
might be responsible for the nugget variance. The vario-
gram of the second axis has a long range of spatial cor-
relation of approximately 61 km. This long range is prob-
ably caused by water depth since there is a significant
correlation between water depth and component scores of
the second axis. A similar phenomenon was observed by
the spatial analysis of diatom assemblages in the same
lake (Kienel and Kumke, 2002). However, unlike the
spatial structure in diatom assemblages, this variogram
has a noticeable relative nugget variance of about 21%.
The number of pairs used for the calculation of vario-
grams range between 50 and 396, sufficiently large to
estimate a reliable variogram. Variogram analysis could
not be extended into different directions due to the lack of
available pairs. We fitted spherical models to both vario-
grams; their model parameters as well as the results of
the cross-validation procedure are given in Table 2. The
prediction error (MSPE) is low, especially for the model
of the second axis. The residual error (RE) is close to
zero, indicating no systematic error in the models (Russo,
1984). Variograms of the third and forth axes also were
estimated (not shown), but they showed no spatial struc-
turing. For the variogram of the PCs of axis 3, a nugget
model could be fitted that indicates every variability of
the site scores is generated by random processes.
Spatial distribution of the sediment characteristics
in Lake Lama
Block-kriged estimates of principal component scores of
the first two axes (Fig. 4) revealed a distinct spatial struc-
turing in sediment properties in Lake Lama. The spatial
distribution in the estimation error (not presented)
showed no systematic features. Thus, estimates are non-
related to their error (Isaaks and Srivastava, 1989; Web-
ster and Oliver, 2000).
The distribution of site score estimates for axis 1
varied at relatively small scales (Fig. 4a). There are some
steep spatial gradients in estimates that occur near river
inflows in the eastern tip as well as in the north-south con-
nected basin, towards the northern shore of the western
basin and in the shallow part near the outflow of the
western basin. High positive estimates occurring near river
mouths, e.g., the Rivers Bek Chikaj, Kuraanach, Dyema
and partly Bucharama, are associated with large fractions
of sand and higher values in TOC/TN ratio and CaO con-
tent. The large sand fractions at river inlets are most pro-
bably caused by the transport and subsequent sedimenta-
tion of coarse-grained material in the lake. Sediments at
sample locations containing large fractions of sand also
have a distinctly higher percentage of diatoms typical of
rivers (Kienel and Kumke, 2002). In addition, the sand
fraction was positively correlated with CaO (r = 0.76,
P  0.05). Higher values ofTOC/TN ratios at river mouths
mostly are significantly larger than the whole lake aver-
age of about 12.2, indicating that the sediments are affect-
ed more by material from land plants than elsewhere in
the lake. Apart from these locations, TOC/TN values of
sediments in Lake Lama point to a dominance of auto-
chthonous production (Meyers and Ishiwatari, 1995).
Figure 3. Variogram of the PCs of the first two principal axes and
their fitted models. A summary of model parameters is given in
Table 2.
Table 2. Fitted standardized variogram model parameters, mean
squared prediction error (MSPE), residual error (RE) and its stan-
dard deviation for the object scores of PCA axes 1 and 2.
PCA axis 1 PCA axis 2
Model spherical spherical
Range (km) 4.88 61.0
Sill 1.03 1.79
Nugget 0.35 0.37
MSPE 0.24 0.01
RE 0.01 ­0.05
Aquat. Sci. Vol. 67, 2005 Research Article 93
A second prominent feature in the spatial gradients
in principal component scores of axis 1 is a region in the
middle of the western basin. Here the principal compo-
nent scores changed from the maximum negative values
to positive values up to 0.3 from the southern shore to-
wards the northern shore (Fig. 4a). This variation in prin-
cipal component scores is caused mainly by a decrease in
biogenic silica and partly by an increase in CaO in the
northern shore. The decrease in biogenic silica indicates
a change in primary productivity of siliceous algae that
is related to diatom assemblage change with decreasing
water depth (Kienel and Kumke, 2002).
The most distinct variation in estimates of principal
component scores of axis 1 occurred in the shallow water
areas of the western end of the western basin. Local mini-
mum and maximum values are separated by only 5 km.
Sedimentological properties mainly responsible for these
anomalies and the resulting spatial gradients are increas-
Figure 4. Spatial distribution of kriged estimates of the (a) first and (b) second PCs in Lake Lama.
(a)
(b)
94 T. Kumke, A. Schoonderwaldt and U. Kienel Spatial variability of sediment composition
ing sand fractions towards the lake outflow and high values
of biogenic silica that decrease in the outflow direction.
Other sediment components contributing to axis 1 vary
little. The increase in biogenic silica is about twofold
compared to other values in the region, indicating a
higher production of diatoms. Sedimentary diatom as-
semblages, however, are dominated by periphytic taxa
in that part of the lake and the spatial structure of the
assemblages (cf. Kienel and Kumke, 2002) is not related
to principal component scores. The large sand fractions
(up to 83%) and low values of biogenic silica (around
one-third of the lake average) near the lake outflow, caus-
ing high positive principal component scores, are pro-
bably the result of re-suspension and transport of parti-
cles by erosion. The shallow water area near the lake out-
flow is, hence, not a sediment accumulation area.
Additionally, this region is subjected to reinforced hydro-
dynamic processes that affect re-suspension. At the spa-
tial scale of 100s of meters to km, advective currents and
wind current patterns are processes influencing the lake
ecosystem and, subsequently, the sediments at shallow
water depths (Pinel-Alloul, 1995).
The spatial distribution in principal component scores
of axis 2 reveals large-scale patterns (Fig. 4b). Values are
negative and relatively stable in the eastern tip and the
eastern basin of the lake, increase gradually in the north-
south link, and become positive in the western basin.
Steep spatial gradients only occur in the western part of
the western basin where water depth decreases. The spa-
tial structure of the second principal component is related
to basin morphology; principal component scores are
significantly correlated with water depth (r = ­0.36,
P  0.05). Basin morphology is a feature typically influ-
encing lakes at large spatial scales (Pinel-Alloul, 1995;
Kienel and Kumke, 2002). The gradual increase in prin-
cipal component scores is caused by a number of sedi-
ment components that slightly, but not significantly, in-
crease from east to west as revealed by ANOVA. Among
these components are most of those that contributed to
the second principal axis (see Fig. 3). The TOC/TN ratio
and Cu showed the opposite trend since they contribute to
the opposite direction of the principal axis. However,
there are no significant differences between basin aver-
ages of the components, apart from d13
Corg and Zr that are
significantly lighter or higher at P0.05, respectively. In
addition, the most distinct differences occurred between
the means in TOC/TN ratios of the eastern tip/eastern
basin (about 13.5) and the north-south link/western basin
(about 11.5). As noted earlier, the largest values in TOC/
TN ratio are located near river inlets, pointing to an intro-
duction of material derived from land plants. Differences
among basin means could be caused by the higher energy
of rivers contributing to Lake Lama in the eastern tip and
eastern basin as well as by an increased sediment input by
erosion since the morphology of the catchment in the
eastern part is characterized by much steeper slopes than
in the western part. Hence, coarse material introduced by
rivers could be transported over longer distances by cur-
rents and influence a larger area of sediment in the eastern
than western part of the lake. In addition, a slumping of
coarse material along the steep slopes in the eastern basin
could occur and might explain the larger basin mean in
sand fraction.
Another prominent feature is the steep spatial gradient
in the western part of the western basin. Sedimentary
components responsible for these variations are d13
Corg,
Zr and partly Nb. Values of Zr and Nb are significantly
correlated (r = 0.56, P  0.05) and increase in the western
part of the western basin. Their increase in the sediments
of Lake Lama might be related to increased atmospheric
deposition since Zr especially is regarded to be a first fall-
out from dust due to its inability to be transported over
long distances (Sirocko et al., 2000). Increased eolian
deposition in the western part of Lake Lama likely occurs
because of its vicinity to the city of Norilsk west of Lake
Lama (see Fig. 1). Hagedorn et al. (1999) concluded that
any anthropogenic deposits, proven to have influenced
recent deposits in Lake Lama in their study, can only be
transported by wind since there is no discharge of waste-
water into the lake.Values of d13
Corg increase in the western
part of the western basin, leading to statistically signifi-
cant differences between the western basin and other
parts of Lake Lama. In addition, there was a significant
correlation between d13
Corg and water depth (r = ­0.45,
P  0.05). The isotopically heavier signatures of organic
carbon in the sediments support the assumption that
erosional processes are dominating in this part of the
lake, leading to exposure of Late Weichselian glaciola-
custrine sediments. In this case, sediments were enriched
in carbonates from the underlying sedimentary rocks
and, consequently, water is harder (Harwart et al., 1999).
Assimilation pathways of organisms would have changed
from CO2 to HCO3, leading to an increase of d13
Corg by
9 (Meyers and Lallier-Verges, 1999). Additionally,
Silurian rocks containing limestone are outcropping di-
rectly and uncovered at the western lake border and could
increase the hardness of the water. The closest sampling
point to the outcrops had the maximum value in d13
Corg.
However, since no measurements of pH in the lake are
available from this location, these explanations are only
probable processes affecting the spatial structure in sedi-
ment properties in that part of the lake. The statistical
relationship between d13
Corg and water depth indicates
that, in general, additional processes might influence the
carbon isotope signatures in Lake Lama that cannot be
identified by the data.
Aquat. Sci. Vol. 67, 2005 Research Article 95
Conclusions
Lake Lama is an excellent example showing how physi-
cal, chemical and biological lake processes are reflected
in lake sediments and how they affect the sediment com-
position in lakes. The general view that the number of
processes at different spatial scales increases with lake
size and complexity of the lake system (e.g., Pinel-Alloul,
1995) was observed in the sediments of Lake Lama. Mea-
sured sediment properties in this lake are influenced by a
variety of processes; e.g., lake morphology and the water
depth, the number, size and position of river inflows,
catchment morphology and hydrodynamical processes.
Each process is characterized by its own spatial scale re-
sulting in a large spatial heterogeneity in sediment pro-
perties of Lake Lama. The statistical techniques applied
in this study (i.e., principal components analysis and geo-
statistical analysis) were suitable for extracting and ex-
plaining the spatial structure in lake sediments.The use of
the first two principal components assured that the major
patterns, capturing more than half of the variance, were
preserved for geostatistical analysis. Variables that con-
tributed most to the spatial patterns were the particle size,
biogenic silica, CaO, TOC/TN ratio, d13
Corg and Zr. Vario-
grams of the first two principal components were quite
different in their shape and parameters. The variogram
of the first component was characterized by a range of
spatial correlation of about 5 km and a relative nugget
variance of 34%, whereas the variogram of the second
component had a range of approximately 60 km and a
relative nugget variance of 21%. Both variograms de-
scribe the spatial structure of the respective components
and allow one to suggest some processes responsible
for that structure. For the first component, we conclude
that the dominating processes are related to the material
transported and deposited by rivers which strongly affect
particle size distributions at river mouths. The second
component is largely influenced by lake morphology
and the water depth. However, the regionalized, block-
kriged maps of the components revealed some more fea-
tures in sediment spatial structure. For example, the most
distinct changes in sediment composition occured at rel-
atively small spatial scales in the western part of the
western basin. These variations are affected by hydrody-
namical processes such wind-induced currents and re-
suspension of material. However, the considerable
amount of small-scale variability of both principal com-
ponents indicates the presence of processes beyond our
sampling design.
The results showed the importance of a proper spatial
analysis of lake sediments to obtain a representative pic-
ture of sediment properties in lakes. The importance is
likely to increase with lake size and complexity of the
catchment. We, therefore, recommend such analyses prior
to conducting paleoenvironmental or paleolimnological
studies.
Acknowledgements
We thank the participants of the expedition Norilsk 97 of
the Alfred Wegener Institute for providing us with sample
material and field data. We are grateful to T. Boettger
for helping us to interpret the carbon isotope distribution.
We would like to thank J. Baier, D. Subetto and an anony-
mous reviewer for helpful comments on earlier drafts of
the manuscript.
References
Anderson, N. J., 1990. Spatial pattern of recent sediment and
diatom accumulation in a small, eutrophic lake. J. Paleolim. 3:
143­160.
Anderson, N. J., 1998. Variability of diatom-inferred phosphorus
profiles in a small lake basin and its implications for histories
of lake eutrophication. J. Paleolim. 20: 47­55.
Andreev, A. A., P. E. Tarasov, V. A. Klimanov, M. Melles, O. M.
Lisitsina and H.-W. Hubberten, 2004. Vegetation and climate
changes around the Lama Lake, Taymyr Peninsula during the
Late Pleistocene and Holocene reconstructed from pollen
records. Quat. Int. 122: 69­84.
Ashley, G. M., 1995. Glaciolacustrine Environments. In: J. Menzies
(ed.), Modern Glacial Environments ­ Processes, Dynamics
and Sediments. Butterworth-Heinemann Ltd., Oxford, pp. 417­
444.
Ellsworth, T. R. and C. W. Boast, 1996. Spatial structure of solute
transport variability in an unsaturated field soil. Soil Sci. Soc.
Am. J. 60: 1355­1367.
Englund, E. and A. Sparks, 1991. GEO-EAS 1.2.1. Geostatistical
Environmental Assessment Software ­ User's guide, US En-
vironmental Protection Agency, Las Vegas.
Galaziy, G. J. and Y. P. Paramuzin, 1981. Lakes in the North-West
of the Siberian Platform (in Russian), Nauka, St. Petersburg.
Hagedorn, B., S. Harwart, M. M. R. van der Loeff and M. Melles,
1999. Lead-210 dating and heavy metal concentration in
recent sediments of Lama Lake (Norilsk Area, Siberia). In:
H. Kassens, H. A. Bauch, I. A. Dimitrenko, H. Eicken, H.-W.
Hubberten, M. Melles, J. Thiede and L. A. Timokhov (eds.),
Land-Ocean Systems in the Siberian Arctic: Dynamics and
History, Springer Verlag, Berlin, pp. 361­376.
Hkansson, L. and M. Jansson, 1983. Principles of Lake Sedi-
mentology, Springer Verlag, Berlin.
Harwart, S., 1999. Geochemical weathering trends of a basaltic
source rock after the retreat of Late Pleistocene glaciers on the
Taymyr Peninsula (Central Siberia) ­ reconstruction using sedi-
mentary sequences of Lake Lama, PhD thesis, University of
Potsdam.
Isaaks, E. and M. Srivastava, 1989. An Introduction to Applied
Geostatistics. Oxford University Press, New York.
Kajak, Z., R. Kacprzak and R. Polkowski, 1965. Tubulat bottom
sampler. Ekol. Polska B 11: 159­165.
Kienel, U. and T. Kumke, 2002. Combining ordination techniques
and geostatistics to determine the patterns of diatom distri-
butions at Lake Lama, Central Siberia. J. Paleolim. 28: 181­
194.
Korsman, T., M. B. Nilsson, K. Landgren and I. Renberg, 1999.
Spatial variability in surface sediment composition charac-
terised by near-infrared (NIR) reflectance spectroscopy. J.
Paleolim. 21: 61­71.
Kumke, T., 1999. Ion transport through agriculturally used soils of
a catchment at the Uelzen Basin ­ Spatial variability of trans-
port and exchange parameters (in German). Wissenschaft &
Technik Verlag, Berlin.
96 T. Kumke, A. Schoonderwaldt and U. Kienel Spatial variability of sediment composition
Kumke, T., T. Streck and J. Richter, 1999. Regional pattern of the
mobile water fraction in soils as determined by disc infiltro-
meter experiments. J. Plant Nutr. Soil Sci. 162: 393­400.
Kumke, T., T. Streck and J. Richter, 2002. Ion transport through
unsaturated soils: field experiments and regional simulations.
Eur. J. Soil Sci. 53: 57­70.
Lebo, M. E. and J. E. Reuter, 1995. Spatial variability in sediment
composition and evidence for resuspension in a large, deep lake.
Mar. Freshwater Res. 46: 321­326.
Legendre, P. and M.-J. Fortin, 1989. Spatial pattern and ecological
analysis. Vegetatio 80: 107­138.
Legendre, P. and L. Legendre, 1998. Numerical Ecology, Elsevier,
Amsterdam.
Lightfoot, P. C., A. J. Naldrett, N. S. Gorbachev, W. Doherty and V.
A. Fedorenko, 1990. Geochemistry of the Siberian Trap of the
Noril'sk area, USSR, with implications for the relative contri-
butions of crust and mantle to flood basalt magmatism. Contrib.
Mineral. Petr. 104: 631­644.
Meyers, P. A. and R. Ishiwatari, 1995. Organic matter accumulation
records in lake sediments. In: A. Lerman, D. M. Imboden and
J. R. Gat (eds.), Physics and Chemistry of Lakes. Springer Ver-
lag, Berlin, pp. 279­329.
Meyers, P. A. and E. Lallier-Vergés, 1999. Lacustrine sedimentary
organic matter records of Late Quaternary paleoclimates. J. Pa-
leolim. 21: 345­372.
Naldrett, A. J., P. C. Lightfoot, V. A. Fedorenko, W. Doherty and N.
S. Gorbachev, 1992. Geology and geochemistry of intrusions
and flood basalts of the Norilsk region, USSR, with implica-
tions for the origin of the Ni-Cu ores. Econ. Geol. 87: 975­
1004.
Pinel-Alloul, B., 1995. Spatial heterogeneity as a multiscale char-
acteristic of zooplankton community. Hydrobiologia 301:
17­42.
Russo, D., 1984. A geostatistical approach to solute transport in
heterogeneous fields and its applications to salinity manage-
ment. Water Resour. Res. 20: 1260­1270.
Sirocko, F., D. Garbe-Schönberg and C. Devey, 2000. Processes
controlling trace element geochemistry of Arabian Sea sedi-
ments during the last 25,000 years. Global Planet. Chang. 26:
217­303.
Ter Braak, C. J. F. and C. Prentice, 1988. A theory of gradient ana-
lysis. Adv. Ecol. Res. 18: 271­317.
Webster, R., 1985. Quantitative spatial analysis of soil in the field.
Adv. Soil Sci. 3: 1­70.
Webster, R. and M. A. Oliver, 2000. Geostatistics for Environmen-
tal Scientists, Wiley, Chichester.
Yates, S. R. and M. V. Yates, 1990. Geostatistics for waste manage-
ment: User's Manual for the GEOPACK (Version 1.0) Geo-
statistical Software System, US Environmental Protection
Agency, Ada.