of 12
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
1/12
Multivariate and geostatistical analysis of wetland soil salinityin nested areas of the Yellow River Delta
M. YangA
, S. L. LiuA,C
, Z. F. YangA
, T. SunA
, and Robert BeazleyB
ASchool of Environment, State Key Laboratory of Water Environment Simulation,
Beijing Normal University, Beijing 100875, China.BDepartment of Natural Resources, College of Agriculture and Life Sciences, Fernow Hall 302,
Cornell University, Ithaca, NY 14853, USA.CCorresponding author. Email: [email protected]
Abstract. This study investigated scale dependency of certain soil salinity ions in topsoil horizons in the Yellow RiverDelta in north-east Shandong province, China. Factorial kriging analysis (FKA) was used to analyse spatial variability of
soil salinity ions (Na+, K+, Mg2+, Ca2+, Cl, SO42) sampled at 3 nested areas over a geologically contrasting region.
Correlation analysis and principal component analysis (PCA) were performed on the logarithmic variables, thenmultivariate geostatistics was used to investigate scale dependency of soil salinity spatial variability and auto- and
cross-variograms exhibited by 3 spatial structures: nugget effect, short-range, and long-range structures. Statistical analysis
showed that NaCl was the main salinity type over the 3 nested sample areas. In addition, the variables were random and
regional, which implied that a linear model of coregionalisation was feasible for the analysis of their spatial variability. The
coefficients of the coregionalisation matrix showed that the short-range structures of auto- and cross-correlation for soil
salinity were dominant at the large and middle-sized sample areas, while the long-range structure dominated at the small
area. The resulting structural correlation coefficients showed strong correlations between variables changing as a function
of spatial scale. These relationships between soil salinity variables at different spatial structures were not acquired by the
linear correlation coefficients.
PCA was then performed on the coregionalisation matrices at each sample area to summarise the relationships among
variables at different spatial structures. From the synthetic analysis of coregionalisation matrices, correlation matrices, and
principal components, we concluded that soil genesis and parent material may act on short-range variation of soil salinity,
while climate and topography may influence long-range structure at the large sample area. At the middle-sized sample area,
variations were mostly affected by mineral fertilisation at the short-range structure, while human activities such as irrigationand drainage in wetland restorations influenced the soil salinity spatial variability at the long-range scale. Vegetation and
groundwater table may also be important factors influencing the spatial variability of soil salinity at different spatial
structures at the small sample areas.
Additional keywords: spatial variability, salinity, factory kriging, PCA, coregionalisation.
Introduction
Spatial variability of soil properties results from complex
interactions between natural processes and management
practices at different spatial and temporal scales (Castrignanet al. 2000b; Lin et al. 2002). Soil-forming factors, such as
parent materials, biota, climate, and topography (Jenny 1941),
explain most of the general characteristic variability; however,
management practices may also affect soil variability
significantly (Marx et al. 1988; Dobermann 1994; Bocchi
et al. 2000; Mohawesh et al. 2008). Some of the several
possible factors that govern soil variability are likely to have
a short-range action, whereas others operate at longer distances.
As a consequence, soil variables are expected to be correlated in
a way that is scale-dependent (Castrignan et al. 2000a).
Despite the amount of literature published in the past 3
decades, knowledge about soil variability is still dispersed
and requires further synthesis (Heuvelink and Webster 2001;
Lin et al. 2005). In particular, there is a need to quantify soil
variability across multiple scales, which will enhance the use of
soil information in diverse applications. To determine scaledependency, factorial kriging analysis (FKA) developed by
Matheron (1982) was previously used in soil science
(Goovaerts 1992). FKA is a variant of kriging that aims to
estimate and map different sources of spatial variability
determined from the experimental variograms (Goovaerts
1992, 1998). This multivariate geostatistical technique allows
description of the spatial relationships, as well as separating
the sources of variation according to the spatial scales at which
they operate (Imrie et al. 2008).
Previously, FKA has been successfully applied in various
fields including remote sensing (Oliver et al . 2000; van
Meirvenne and Goovaerts 2002; Rodgers and Oliver 2007;
CSIRO 2009 10.1071/SR08211 0004-9573/09/050486
CSIRO PUBLISHING
www.publish.csiro.au/journals/ajsr Australian Journal of Soil Research, 2009, 47, 486497
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
2/12
Tarnavsky et al. 2008), hydrogeology (Wang et al. 2001; Linet al. 2004, 2006; Ryu et al. 2006), ecology (Castrignan et al.
2000b; Hernndez et al. 2007), and landscape (Bishop and Lark
2006). Within the field of soil science, it has been used to assess
variation in soil properties for crop management (Dobermannet al. 1995; Bocchi et al. 2000; Casa and Castrignan 2008),
geochemical exploration (Jimnez-Espinosa and Chica-Olmo
1999; Batista et al. 2001; Reis et al. 2004), pollution (Einax
and Soldt 1998; Lin et al. 2002; Rodrguez et al. 2008), and
analyses of underlying geochemical processes (Bourennane
et al. 2003; Xu and Tao 2004). Considerable work has been
done to investigate variability of soil basic properties (Brdossy
and Lehmann 1998; Qiu et al. 2001) and soil pollutants (Li et al.
2006; Tavares et al. 2008), particularly heavy metals (Reis et al.
2003; Xu and Tao 2004). However, quantification of soil salinity
variability at multiple scales is often desirable, especially in
wetlands where salinity accumulates in great amount and
threatens plants and animals.
Accumulation of soluble salts in soils has caused serious
problems in relation to agricultural development and naturalresources management. As soils become more saline, soil
moisture becomes less available to plants, until at higher
salinities water is drawn from the roots back into the soil
(Brady and Weil 2002). The classic conditions that promote
increasing soil salinity are drier climates, where irrigation is
irregular and evapotranspiration allows salts to become
concentrated in the upper part of the soil profile. However,
low-lying coasts, such as deltas where surface drainage is poor
andflooding during extremely high tides or storms is common,
can also foster conditions for high levels of soil salinity (Fang
et al. 2005). Salinities of 0.7 dS/m are less stressful to mostplants. Above this threshold, salt toxicity occurs, with different
plants becoming susceptible to perceptible salt stress at salinities
as low as 0.8 dS/m.The Yellow River Delta, the only large delta in China to
undergo extensive development, is characterised by extensive
coverage of saline soils (Liu and Drost 1997). A high proportion
of these occur in the most actively prograding areas inconjunction with recently formed estuarine wetlands. Despite
plans of the Chinese central government to enlarge the
agricultural production base in the Yellow River Delta, a lack
of information on the salinisation potential of regional soils
remains an impediment to developing a balanced and
ecologically sound plan to achieve this goal.
The main objective of this work was to study the spatial
variability of soil salinity over 3 nested areas in the Yellow River
Delta and seek possible explanations for their distributions inthe light of the statistical evidence. Here, the soil salinity was
characterised by calcium (Ca2+), potassium (K+), sodium (Na+),
magnesium (Mg2+), chloride (Cl), and sulfate (SO42) ions.
Each ion had its own distinctive distribution and it was difficult
to discern common patterns and to seek common causes for
them. Therefore, we deemed that an analysis of
coregionalisation would be more revealing than a univariate
geostatistical analysis. We examined the scale-dependent
correlation structure of some soil properties, proposing that it
can reflect the different sources of variability. Information about
soil variability is important in ecological modelling,
environmental prediction, precision agriculture, and natural
resources management (Lin et al. 2005). Consequently, studyon soil salinity variability in the Yellow River Delta can
provide important scientific data for natural resources
management and wetland restorations.
Materials and methodsThe study area
The study area is located at 11880701198100E and
378200388120 N in northern Shandong Province, China
(Fig. 1). It encompasses an area >6000 km2, from Ninghai,extending north-east to the Taoerhe River estuary, south-east
to the Xiaoqinghe River estuary, and eastward forming a fan
shape. This region lies within the newly created wetlands of the
Yellow River Delta, which extends from the mouth of YellowRiver to the Bohai Sea. The current course of Yellow River was
formed artificially in 1976 by changing the old course from the
Diaokou River to the Qingshui Gully. The soils in the delta
formed on marine sediments as a result of deposition of a large
amount of sand and mud transported by the Yellow River,together with lateral sea seepage. The Yellow River Delta is
characterised by extensive coverage of saline soils, a high
proportion of which occurs in the most actively prograding
areas in conjunction with recently formed estuarine wetlands.
The area has a monsoon climate of the warm-temperate zone.
The average annual temperature is 11.712.68C. The average
annual precipitation is 530630 mm, of which 70% is rainfall
during summer (MayJuly), and the ratio of evaporation to
precipitation is 3.22 on annual average. The groundwater
table in the delta is high, in general ranging from 1.6 to
2.4 m and the mineralisation degree is 32.4 g/L on average.
Soil samples and analytical methodsOur analysis is based on 118 samples collected in October and
November 2007. Soil samples were taken at 3 nested areas:
(1) the large area of the whole Yellow River Delta, (2) the middle
area of The Dawenliu Nature Reserve, and (3) the small area of
the core of The Dawenliui Nature Reserve (Fig. 1). Collected
soil samples were air-dried and crushed to pass through a 2-mm
mesh. Fifty-g subsamples were ground in a mortar to pass
through a 0.25-mm sieve. Samples were subsequently
transported to the laboratory for determination of soil salt
content, including Ca2+, K+, Na+, Mg2+, Cl, and SO42. Na+
and K+ which were determined by frame photometry, and Ca2+,
Mg2+, Cl, and SO42 by ion chromatography.
Multivariate geostatistical analysis
The theory underpinning factorial kriging analysis (FKA) has
been published in several texts (Goovaerts 1997; Wackernagel
1998; Pardo-Igzquiza and Dowd 2002). A brief outline is
provided below. The first step involves the definition of a
linear model of coregionalisation (LMC). Geostatistical
techniques rely on variograms, which are convenient
representations of the auto- and cross-correlation structures in
a spatially distributed dataset. Spatial scales of variation in our
context are related to different ranges observed in the
experimental semi-variogram. An experimental variogram is
calculated as follows:
Analysis of wetland soil salinity Australian Journal of Soil Research 487
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
3/12
gh 12Nh
XNh
i 1
zxi zxi h2 1
where N is the number of data pairs approximately separated
by the vectorh, andz(xi) is the value of the regionalised variable
zof interest at location xi. Cross-variograms between 2 variables
za and zb may be calculated as follows:
gabh 1
2Nh
XNh
i 1
zaxi zaxi h zbxi zbxi h
2
Valid models which are commonly fitted to the experimentalvariograms include spherical, Gaussian, and exponential
functions. These are characterised by a sill, which represents
the covariance accounted for by the model, and a range, which
signifies the extent of spatial correlation. The value of the
variograms where the model approaches the abscissa is
referred to as the nugget effect. This encompasses the micro-
scale variation and any errors due to analytical, sampling, or
location measurements (Imrie et al. 2008).
The experimental semi-variograms and semi-cross-
variograms are modelled with nested structures, with each
structure representing a particular scale of variation, e.g. a
nugget effect, a short-range structure, and a long-range
structure (Pardo-Igzquiza and Dowd 2002). Such a
covariogram is a linear combination of Ns component
functions gu(h):
gabh XNsu 1
guabh XNsu 1
buab guh 3
where the buab are coefficients which represent the importance
of each spatial scale u on the relationships between the variables.
This LMC can be expressed in matrix terms:
Gh XNsu 1
Buguh 4
where G(h) is the p p variogram matrix and Bu
is a positivesemi-definite matrix of the coefficients babu . A measure of the
correlation between the variables za andzb at the spatial scale u
is given by:
ruab buabffiffiffiffiffiffiffiffiffiffiffiffiffiffibuaab
ubb
q 5
The structural correlation coefficients rabu differ from the
traditional product-moment correlation coefficients in that
they focus on specific spatial scales, filtering out the
processes operating over different distances (Castrignan
et al. 2000b).
Legend
Legend
Large scope
Middle scope
Small scope
Small scope
The delta
Core area
0 5 10 20 km
0 500 1000 2000 km
Yellow River
Yangtze River
The Yellow River Delta
0 15 30 60 m
N
N
N
Fig. 1. Location of the Yellow River Delta and samples at 3 nested areas.
488 Australian Journal of Soil Research M. Yang et al.
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
4/12
Principal component analysis (PCA) can be performed on the
Bu matrices, in order to aid the identification of the major
processes operating at the spatial scales identified. The
interpretation of the resulting factors is based on their
correlation with the original variables.
Results and discussionSummary statistics
Table 1 summarises the statistics of all measured data. The
distribution of the soil salinity ions at different sample areas was
acquired by geostatistical module in ArcGIS software. The
distributions of the original variables at the 3 areas were
asymmetric, only Na+ at the middle area and K+ and Ca2+
ions at the small area had skewness coefficient
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
5/12
Extracting the latent roots and vectors from the correlationmatrices may reveal the relations among the variables more
clearly. The first 2 eigenvalues are listed in Table 2. At the large
and middle areas, the first accounted for ~50% of the variance,
and the second for 19.32% and 24.82%, respectively. However,the first accounted for 45.23% and the second for 28.22% at
the small sample area. All the eigenvalues obtained at the 3 areas
confirmed the results of the above correlation analyses that
the correlations between the variables are complicated and
dispersive. The correlation coefficients between the variables
and principal components are reported in Table 3.
The variable Na+ and Cl, which determined the
salinealkaline types of the soils, appeared closely related,
positively loading on the first principal component at the
large area, whereas K+ was strongly correlated with the
second principal component. This seemed to suggest that soil
salinisation was formed on marine sediments due to lateral sea
seepage in the prograding process of the Yellow River Delta
(Weng and Gong 2006). Geological process and monsoon
climate may also be important factors.
The same principal components were present at the middleand small areas. Na+ and Clwere positively associated with the
first principal component and also showed a strong link with
SO42, Mg2+, and Ca2+, which were the second principal
components compared to the large area. This may beattributed to some other processes superimposing on soil
salinity characteristics.
In this context, Na+ and Cl were strongly correlated,
positively loading on the first principal component, which
Table 4. Coefficients of double-spherical model fitted to the principal
components
Area Component Nugget Sill1 Sill2 Range1 (m) Range2 (m)
Large Factor1 0.216 0.954 0.774 2120 4080
Factor2 0 0.686 0.259 2120 4080
Middle Factor1 1.38 1.5 0.39 1080 2050
Factor2 0.39 1.005 0.525 1080 2050
Small Factor1 0.784 1.512 0.532 45 105
Factor2 0.238 1.377 0.544 45 105
0 20 40 60 80 100 120
0
0.3
0.6
0.9
1.2
1.5
1.8
2.1
2.4
lhl (m)
0 20 40 60 80 100 120
0
0.4
0.8
1.2
1.6
2.0
2.4
2.8
3.2
0 300 600 900 1200 1500 1800 2100 2400 2700
0
0.3
0.6
0.9
1.2
1.5
1.8
2.1
2.4
0 300 600 900 1200 1500 1800 2100 2400
0
0.4
0.8
1.2
1.6
2.0
2.4
2.8
3.2
(|h|)
0 500 1000 1500 2000 2500 3000 3500 4000 4500
0
0.3
0.6
0.9
1.2
1.5
1.8
2.1
2.4
2.7
0 500 1000 1500 2000 2500 3000 3500 4000 4500
0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Large sample scope
Middle sample scope
Small sample scope
Fig. 3. Spherical auto-variograms of the principal components plotted as points. Solid lines are those of the fitted models of linear
coregionalisation.
490 Australian Journal of Soil Research M. Yang et al.
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
6/12
suggested that NaCl may be the main salinity type in the YellowRiver Delta, but there were different variables related with the
second principal component, K+ at the large area and Mg2+ and
Ca2+ at the middle and small area. This showed different salinity
characteristics formed from many different processes, such asmineral fertilisation, geological genesis, climate, vegetation,
lateral seepage, channel change, groundwater tables, and
human activities.
Spatial analysis and coregionalisation
Classical statistical analysis cannot separate the different sources
of spatial variability affecting soil salinity at the site surveyed.
This required a particular statistical approach that combinesclassical factor analysis for describing the correlation
structure of a multivariate dataset with geostatistics, to take
into account the regionalised nature of the variables.
Consequently coregionalisation analysis was performed(Castrignan et al. 2000b).
Coregionalisation analysis
Before attempting to fit a linear model of coregionalisation,we calculated the experimental auto- and cross-variograms of
the 6 salinity ions based on the correlation matrices and the PCA
aforementioned. We calculated the scores of the principal
components at each sample area, and then calculated thevariograms of the first 2 principal components of the
variables. The 6 graphs of the first 2 principal components at
the 3 nested areas displayed a steady increase in semi-variance
with increasing lag distance to ~2120 m at the large sample area,
1080 m at the middle sample area, and 45 m at the small sample
area. Then it reached a maximum more slowly at>4080, 2050,and 105 m, respectively. To represent the whole variograms
with both short- and long-range structures of each factor we
fitted a double-spherical model with a nugget (except for the
factor 2 at the large sample area). The values of the coefficients
are given in Table 4. The solid lines in Fig. 3 represents these
fitted models.
0 500 1000 1500 2000 2500 3000 3500 4000 45000
0.05
0.45
0.40
SO42
Na+
0 700 1400 2100 2800 3500 4200 4900 5600 6300
0
0.002
0.004
0.006
0.008
0.012
0.010
0.014
0.016
0 700 1400 2100 2800 3500 4200 4900 5600 6300
0
0.0024
0.0020
0.0028
0.0032
0.0036 K+
0 700 1400 2100 2800 3500 4200 4900 5600 6300 0
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0.018
(|h|)
Mg2+
0 700 1400 2100 2800 3500 4200 4900 5600 6300
0
0.000
0.001
0.002
0.003
0.004
0.005
0.006
0.007
0.004
Ca2+
0 700 1400 2100 2800 3500 4200 4900 5600 6300
0
0.02
0.04
0.06
0.08
0.10
0.12
0.140.16
|h| (m)
Cl
0.35
0.30
0.25
0.10
0.15
0.20
0.0016
0.0012
0.0008
0.0004
Fig. 4. Spherical auto-variograms of the soil standardised variables plotted as points at the large scale. Solid lines are those of the
fitted models of linear coregionalisation.
Analysis of wetland soil salinity Australian Journal of Soil Research 491
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
7/12
Although each soil attribute has its own more-or-lessdistinctive distribution, there are some similarities. Therefore,
an analysis of coregionalisation would be more revealing with
common spatial patterns and aid in seeking common causes
for them at different spatial structures (Castrignan et al.2000b).
Figure 4 shows the linear models of coregionalisation as it
fitted the 6 direct auto-variograms superimposed on the
coefficients at the large sample area (Table 5; auto-variogram
at the middle and small area and cross-variograms are not
reported). They show that the short-range structure of auto-
correlation was dominant in almost all the variables at the
3 areas. The short-range structure dominated the cross-
variograms between Na+ and K+ (negative), Na+ and Cl
(positive), K+ and Mg2+ (positive), Mg2+ and SO42
(negative), and Ca2+ and SO42 (positive) at the large area;
Na+ and SO42(negative), K+ and Mg2+ (negative), K+ and Ca2+
(positive), and Cl and SO42 (negative) at the middle area; and
Na+ and Cl (positive), K+ and Cl (positive), and Ca2+ and
SO42
(negative) at the small area. However, the long-rangestructure dominated mainly that of Na+ and K+ (negative) at the
middle sample area, and Na+ and Mg2+ (negative), Na+ and Ca2+
(negative), Na+ and SO42 (negative), K+ and Mg2+ (positive),
Mg2+ and Ca2+ (negative), and Mg2+ and SO42(negative) at the
small area.
By synthetic analyses, we found that there were differentspatial structures for the cross-varigrams between different soil
variables. The short-range structure dominated at the large
sample area, long-range structure at the small sample area,
and at the middle sample area, the dominant spatial structureswere between the short-range and long-range structures by
different auto- and cross-variograms. Previous studies mostly
examining only one study area showed that the main spatial
structure was different; some exhibited long-range structure
(Bourennane et al . 2003), and some showed short-range
structure (Imrie et al. 2008). The importance of each structure
reflected the influence of the corresponding spatial variable or
pair of variables on soil variation. However, the whole set of
multivariate spatial relations can be better represented by their
nugget and structural correlation coefficients, which allows
focusing on a specific spatial structure, filtering the effects of
the other structures of variation. This difference showed that
sampling interval and area may influence spatial variability of
soil salinity to a certain extent, and the right amount of samples
was important for exact analysis of spatial variability and itssources at different spatial structures (Pye et al. 2006).
Structural correlation coefficients
In this case study, the simple productmoment correlation
coefficients did not reveal the real relationships among the
Table 5. Coefficients of double spherical coregionlisation models for variograms
Area Structure Na+ K+ Mg2+ Ca2+ Cl SO42
Large Nugget 0.0006 0.00016 0.00153 0.00144 0 0
Sill1 0.08 0.00174 0.0085 0.00258 0.0756 0.284
Sill2 0.0034 0.00082 0.00423 0.00198 0.0585 0.064Middle Nugget 0.018 0.0252 0.0008 0 0.06 0.02
Sill1 0.12 0.133 0.0736 0.09 0.26 0.144
Sill2 0.075 0.0273 0.0192 0.034 0.092 0.022
Small Nugget 0.0234 0.0198 0.0096 0.0098 0.05 0.0035
Sill1 0.0459 0.0282 0.0189 0.0096 0.06 0.029
Sill2 0.0261 0.0048 0.0054 0.0018 0.044 0.012
Na+K+ Na+Mg2+ Na+Ca2+ Na+Cl Na+SO42 K+Mg2+ K+Ca2+
Large Nugget 0 0 0 0 0 0 0.0005
Sill1 0.003 0.0039 0.004 0.011 0.0282 0.0028 0.0019
Sill2 0.0011 0.0034 0.0029 0.0053 0.0102 0.0009 0.0015
Middle Nugget 0.016 0 0 0.09 0.028 0.0054 0
Sill1 0.048 0.02 0.0639 0.068 0.084 0.0166 0.0135
Sill2 0.054 0.0134 0.036 0.044 0.012 0.0068 0.0081
Small Nugget
0.0024 0 0 0.026 0 0
0.004Sill1 0.0426 0.0114 0.0072 0.066 0.01 0.0033 0.0544
Sill2 0.0168 0.0118 0.0086 0.016 0.02 0.004 0.02
K+Cl K+SO42 Mg2+Ca2+ Mg2+Cl Mg2+SO4
2 Ca2+Cl Ca2+SO42 ClSO4
2
Large Nugget 0 0 0.00215 0.00052 0.018 0 0.012 0
Sill1 0.0065 0.0067 0.0035 0.007 0.164 0.0095 0.123 0.009
Sill2 0.0025 0.0042 0.0013 0.004 0.0735 0.0074 0.006 0.006
Middle Nugget 0.0486 0.012 0.009 0 0 0 0.0084 0.028
Sill1 0.0414 0.0358 0.0378 0.0255 0.0148 0.064 0.0438 0.146
Sill2 0.0162 0.014 0.0114 0.0138 0.0058 0.02 0.012 0.062
Small Nugget 0 0 0.0016 0.0036 0.0016 0 0 0.0084
Sill1 0.0183 0.0108 0.005 0.0102 0.005 0.0069 0.005 0.0312
Sill2 0.0066 0.0054 0.0056 0.0026 0.0056 0.0035 0.0015 0.0126
492 Australian Journal of Soil Research M. Yang et al.
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
8/12
variables, since it averaged out distinct changes in the correlationstructures occurring at different spatial scales. Pooling all the
spatial structures, the only significant correlations between the
variables were observed for the pairs of Na+ and K+, Cl, SO42
at each sample area (Fig. 2). In contrast, filtering the differentcomponents disclosed interesting correlations between the
variables, changing as a function of spatial scales
(Castrignan et al. 2000b).
At the large sample area, in the shortest spatial structure
(nugget effect), the most relevant correlation appeared between
Ca2+ and Mg2+ (negative), and to a less extent between Cl and
SO42 (positive). As the nugget effect always comprised an
unknown variance caused by procedural errors, we chose to
direct our attention to correlation structures at short- and long-
range scales. At plot-size level, Ca2+ and Mg2+ revealed strong
correlations, whereas at longer spatial structure, high
correlations were between Na+ and K+, as well as Na+ and
Mg2+. Cl appeared correlated, but to a lesser extent, with Ca2+
(Fig. 5a).
At the middle sample area, the correlation structures lookedsomewhat different, and the short-range structure showed
SO42 was strongly correlated with Cl, and slightly with Na+
and Ca2+. However, strong correlations were shown between
Na+ and K+ atP< 0.01, and Cland Ca2+ aswell asCa2+ and Na+
at P< 0.05 (Fig. 5b).At the small sample area, the strong correlations between
Na+ and Cl and Cl and SO42 showed at the 3 spatial scales,
while there were also strong correlations between K+ and Na+,
Cl, SO42, and between Mg2+ and Ca2+, at the short- and long-
range spatial structures (Fig. 5c).
From the analyses at 3 sample areas, there were strong
correlations between Na+ and Cl, K+ at the large and small
areas, Cl and Ca2+ at the large and middle area, as well as
between other variables. In fact, there were some correlations
between the soil base-exchange ions which could be found in
vegetation ecology (Guo and Tang 1999; Li et al. 2002).
Therefore, from the estimation of the structural correlation
coefficients, we were able to summarise the causes of the
(a) Large sample area
(b) Middle sample area
(c) Small sample area
Nugget effect
Na+
K+
Mg2+
Ca2+
Cl
SO42
Na+
K+
Ca2+
Mg
Cl
SO42
Na+
K+
Mg2+
Ca2+
Cl
SO42
Na+
K+
Mg2+
Ca2+
Cl
SO42
Na+
K+
Mg2+
Ca2+
Cl
SO42
Na+
K+
Ca2+
Mg
Cl
SO42
Na+
K+
Mg2+
Ca2+
Cl
SO42
Na+
K+
Mg2+
Ca2+
Cl
SO42
Na+ K+ Mg2+ Ca2+ Cl SO42
Na+
K+
Ca2+
Mg
Cl
SO42
Na+ K+ Mg2+ Ca2+ Cl SO42 Na+ K+ Mg2+ Ca2+ Cl SO42
Short-range Long-range
Fig. 5. Correlation matrices for each spatial structure at the 3 areas.
Analysis of wetland soil salinity Australian Journal of Soil Research 493
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
9/12
variability of soil salinity at each spatial structure at the 3 nestedsample areas.
At the large sample area, plot-size spatial variation was
mostly affected by environmental factors such as climate and
topography, which controlled soil genesis on the whole(Bruelheide and Udelhoven 2005). At the long-range scale,
the monsoon climate, characterised by high evaporation
discharge and relatively low precipitation, and rolling
topography imposing on the vertical and lateral flow, may
have played some role in the spatial structure of soil salinity.
At short-range structure, spatial variation was likely
dominated by parent material (river alluvium and marine
sediments in the bottom) being influenced by tides and the
Yellow River.
At the middle sample area, spatial variability at short-rangestructure was mostly influenced by mineral fertilisation, which
carried a lot of soluble salts accumulating in the soil and
consisting principally in various proportions of sodium and
calcium cations, as well as chloride and sulfate anions (Wang
et al. 2006). Long-range spatial variability was potentially
affected by irrigation and draining in wetland restoration
processes, which influenced the salinisation and leaching of
cations in soils (Shan 2007).
Based on the analyses above, at the small sample area, plot-
size spatial variation was mostly caused by the distribution of
vegetation, especially by halo-tolerant plants, which influenced
soil salinity by secreting or absorbing salts in soils, and
alleviated salinity stress (Song et al. 2003). Additionally
SO42
K+
Cl
Mg2+
Ca2+
Na+
35.2%
37.8% Mg2+
Ca2+
SO42
K+
Cl
Na+
27.7%
54.2%
Na+Mg2+
K+
SO42
Cl
Ca2+
35.3%
58%
SO42
Na+
Cl
Mg2+
K+
Ca2+
67%
24.3%
SO42
K+Na+Cl Mg
2+
Ca2+
63%
17.2 %
Na+
K+
Ca2+Mg2+
Cl
SO42
74.3%
13.4%
Na+
K+
Ca2+
SO42
Mg2+
Cl
72.3%
25.5%
Ca+
Na+
SO42
Mg2+
Cl
K+
51.1%
27.3%
Na+
K+
Mg2+
Cl
Ca2+
SO42
75.3%
18.9%
Large sample area
Middle sample area
Small sample area
Nugget effect Short-range Long-range
Fig. 6. Correlation circles for the structural variation at the 3 scales.
494 Australian Journal of Soil Research M. Yang et al.
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
10/12
groundwater tables may be the cause of spatial variability at thelong-range scale. Alluvial soils in the delta were subject to
seepage or flooding during high tides and coastal storms, and
their poor drainage from high groundwater tables allowed salts
to accumulate due to inadequate leaching (Fang et al. 2005).Because the 3-sample area was nested, the spatial structure at the
3 areas had some overlaps.
Regional factors analyses
In order to summarise the relationships among the variables
at different spatial structures, a PCA was performed on the
regionalisation matrices for each sample area. The spatial
interrelations among the variables, as described above by
coregionalisation matrices for each sample area, could be
clearly displayed in the circles of correlations corresponding
to different spatial scales at each sample area (Saporta 1990)
(Fig. 6), where the pair of coordinates of each variable was
determined by the pair of correlation coefficients between the
spatial component of the variables and the first 2 regionalisedfactors.
At the small area, the first 2 factors explained ~78% of nugget
variance: the first one was highly correlated with Na+ and Ca2+
(negatively) and also with SO42 (positively), whereas the
second was positively correlated with Mg2+ and Cl
. Thissuggested that there were various salinity types at a very
short range and basic salt is predominant, which is at
variance with the results of Weng and Gong (2006), who
thought that Na+ and Cl were significantly correlated and themain salinity type was sodium chloride. However, at this spatial
structure components also included procedural errors and hence
one should interpret them with caution (Castrignan et al.
2000b).The first 2 components of the short-range structure explained
>97% of the overall variance at this spatial scale, where the Na+,Ca2+, and K+ are strongly correlated with the first component,
whereas Clwas weighing alone slightly over the second factor.
At long-range structure, the components weighing over the
first 2 components were similar to that at the short-range
structure, except for the exchange of Cl and Ca2+ loading on
the first 2 components. This seemed to suggest that salinity
variation at the short-range structure can be characterised by
calcareous ingredients (Xu 2003), and at the long-range structure
by soluble ions.
Proceeding to the larger sample area, no differences of
importance are observed, except for the different performance
of Ca2+
and Cl
at the nugget and plot-size level and Mg2+
at thelong-range structure. At the nugget and plot-size, Ca2+ and Cl
had a reversed impact on the first 2 components, which suggests
that different processes may influence spatial variability of Ca2+
and Cl in different sample scales, as for Mg2+. Some of these
processes may include the infiltration of sea water owing to the
proximity of the study site to the shore, watering and draining in
wetland restoration, the geological nature of the rock, the rise of
the groundwater table to within few meters of soil surface, and
the previous accumulation of salt (Shao et al. 2008).
In extending our sample area great differences were
observed. In the nugget variance, the first is highly correlated
with K+ and SO42, whereas the second is negatively correlated
with Mg2+ and positively with Ca2+. At short-range structure,Cl influenced spatial variation independently from other
variables on the first component, and also, slightly, together
with Na+. This may be attributed to several factors such as low
groundwater tables caused by down-draught of sea water, andhigh evaporation. SO4
2 was highly correlated with the Mg2+
and Ca2+, which suggested that the parent material was likely
alkaline (Wang et al. 2004). Cl was correlated with Ca2+, and
weighed commonly over the first components, whereas other
variables controlled the second component.
Conclusions
The FKA described above enabled the factors underlying the
variation in soil salinity to be examined at 3 different spatial
scales at 3 nested sample areas.
At the large area, sampled in the whole Yellow River Delta,
spatial variability of soil salinity was affected by historic or
intrinsic environment conditions, such as climate, geology,
topography, soil genesis, and parent material, which acted ondifferent spatial structures. Because there was an unknown
variance caused by procedural errors at the scales shorter
than the sampling density, we focused our attention on the
short- and long-range structures. At the short-range structure,
variations in parent material geology of the major structural
divisions of the continent to sea were detected. At the long-range
scale the influence of monsoon climate and topography became
apparent.
At the middle sample area, variations were mostly affected by
mineral fertilisation at the short-range structure, while human
activities such as irrigation and drainage in wetland restorations
influenced the variations of soil salinity at the long-range scale.
At the small sample area, an area of 200 m by 200 m, vegetation
and groundwater table may influence the spatial variability ofsoil salinity at the short and long structures.
PCA showed that salinity types were slightly different atdifferent spatial structures and significantly at the 3 sample areas.
Compared to our analysis using univariate geostatistics, which
showed that NaCl was the main salinity type, analysis using
multivariate geostatistics showed similar results, but exhibitedmore detailed information at different spatial structures. With
this type of analysis it is possible to have better management of
spatial and temporal variability associated with all aspects of
wetland restoration for the purpose of improving soil propertiesand environmental quality.
Multivariate geostatistical techniques such as factorial
kriging analysis have facilitated the separation of the different
sources of spatial variation at different scales. This method has
proved very useful for several main reasons (Bocchi et al. 2000;
Castrignan et al. 2000a) and is likely to be applicable to the
future datasets that will emerge as part of the Global
Geochemical Reference Network (Imrie et al. 2008).
Acknowledgements
This study was supported by the National Key and Important Program for
Basic Research of China (No. 2006CB403303) and the National Natural
Sciences Foundation of China (No. 40871237; No. 40501067). The authors
would like to thank instructor Liu R M for guiding in operation of
multivariate geostatistical techniques.
Analysis of wetland soil salinity Australian Journal of Soil Research 495
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
11/12
References
Brdossy A, Lehmann W (1998) Spatial distribution of soil moisture in a
small catchment. Part 1: Geostatistical analysis. Journal of Hydrology
206, 115. doi: 10.1016/S0022-1694(97)00152-2
Batista AC, Sousa AJ, Batista MJ, Viegas L (2001) Factorial kriging with
external drift: a case study on the Penedono Region, Portugal. AppliedGeochemistry 16, 921929. doi: 10.1016/S0883-2927(00)00069-X
Bishop TFA, Lark RM (2006) The geostatistical analysis of experiments at
the landscape-scale. Geoderma 133, 87106. doi: 10.1016/j.
geoderma.2006.03.039
Bocchi S, Castrigan A, Fornaro F, Maggiore T (2000) Application of
factorial kriging for mapping soil variation at field scale.
European Journal of Agronomy 13, 295308. doi: 10.1016/S1161-
0301(00)00061-7
Bourennane H, Salvador-Blanes S, Cornu S, King D (2003) Scale of spatial
dependence between chemical properties of topsoil and subsoil over a
geologically contrasted area (Massif central, France). Geoderma 112,
235251. doi: 10.1016/S0016-7061(02)00309-9
Brady NC, Weil R (2002) The nature and properties of soils. (Prentice-
Hall: Upper Saddle River, NJ)
Bruelheide H, Udelhoven P (2005) Correspondence of the fine-scale spatialvariation in soil chemistry and the herb layer vegetation in beech forests.
Forest Ecology and Management 210, 205223. doi: 10.1016/j.foreco.
2005.02.050
Casa R, Castrignan A (2008) Analysis of spatial relationships between soil
and crop variables in a durum wheat field using a multivariate
geostatistical approach. European Journal of Agronomy 28, 331342.
doi: 10.1016/j.eja.2007.10.001
Castrignan A, Giugliarini L, Risaliti R, Martinelli N (2000 b) Study of
spatial relationships among some soil physico-chemical properties of a
field in central Italy using multivariate geostatistics. Geoderma 97,
3960. doi: 10.1016/S0016-7061(00)00025-2
Castrignan A, Goovaerts P, Lulli L, Bragato G (2000 a) A geostatistical
approach to estimate probability of occurrence of Tuber melanosporum
in relation to some soil properties. Geoderma 98, 95113. doi: 10.1016/
S0016-7061(00)00054-9Dobermann A (1994) Factors causingfield variation of direct-seededflooded
rice. Geoderma 62, 125150. doi: 10.1016/0016-7061(94)90032-9
Dobermann A, Goovaerts P, George T (1995) Sources of soil variation in an
acid Ultisol of the Philippines. Geoderma 68, 173191. doi: 10.1016/
0016-7061(95)00035-M
Einax JW, Soldt U (1998) Multivariate geostatistical analysis of soil
contaminations. Fresenius Journal of Analytical Chemistry 361,
1014. doi: 10.1007/s002160050826
Fang HL, LiuGH, Kearney M (2005) Georelational analysis of soil type, soil
salt content, landform, and land use in the Yellow River Delta, China.
Environmental Management35(1), 7283. doi: 10.1007/s00267-004-
3066-2
Goovaerts P (1992) Factorial kriging analysis: a useful tool for exploring
the structure of multivariate spatial information. Journal of Soil Science
43, 597
619. doi: 10.1111/j.1365-2389.1992.tb00163.xGoovaerts P (1997) Geostatistics for natural resources evaluation. (Oxford
University Press: New York)
Goovaerts P (1998) Geostatistical tools for characterizing the spatial
variability of microbiological and physico-chemical soil properties.
Biology and Fertility of Soils 27, 315334. doi: 10.1007/s003740050439
Guo FQ, Tang ZC (1999) Difference in Na+, K+ accumulation in the salt
tolerant mutant and the wild type of wheat during exposure to NaCl
stress. Acta Botanica Sinica 41(5), 515518.
Hernndez P, Fernndez R, Novo M, Trigo D, Diaz Cosin DJ (2007)
Geostatistical and multivariate analysis of the horizontal distribution
of an earthworm community in El Molar (Madrid, Spain). Pedobiologia
51, 1321. doi: 10.1016/j.pedobi.2006.11.002
Heuvelink GBM, Webster R (2001) Modelling soil variation: past, present,
and future. Geoderma 100, 269301. doi: 10.1016/S0016-7061(01)
00025-8
Imrie CE, Korre A, Munoz-Melendez G, Thornton I, Durucan S (2008)
Application of factorial kriging analysis to the FOREGS European
topsoil geochemistry database. The Science of the Total Environment
393, 96
110. doi: 10.1016/j.scitotenv.2007.12.012Jenny H (1941) A system of quantitative pedology. In Factors of soil
formation. (McGraw Hill: New York)
Jimnez-Espinosa R, Chica-Olmo M (1999) Application of geostatistics to
identify gold-rich areas in the FinisterreFervenza region, NW Spain.
Applied Geochemistry 14, 133145. doi: 10.1016/S0883-2927(98)
00035-3
Li BG, Cao J, Liu WX, Shen WR, Wang XJ, Tao S (2006) Geostatistical
analysis and kriging of Hexachlorocyclohexane residues in topsoil from
Tianjin, China. Environmental Pollution 142, 567575. doi: 10.1016/
j.envpol.2005.10.039
Li SH, Xu X, Hui HX, Mi HL, Ma XQ (2002) Study on K+, Na+ selective
absorption of different organs of spring wheat under soil saline sodic
stress in different growth season. Acta Botanica Boreal-Occident Sinica
22(3), 587594.
Lin HS, Wheeler D, Bell J, Wilding L (2005) Assessment of soil spatialvariability at multiple scales. Ecological Modelling 182, 271290.
doi: 10.1016/j.ecolmodel.2004.04.006
Lin YB, Lin YP, Lui CW, Tan YC (2006) Mapping of spatial multi-scale
sources of arsenic variation in groundwater on ChiaNan floodplain of
Taiwan. The Science of the Total Environment 370, 168181.
doi: 10.1016/j.scitotenv.2006.07.002
Lin YB, Tan YC, Lin YP, Liu CW, Hung CJ (2004) Geostatistical method to
delineate anomalies of multi-scale spatial variation in hydrogeological
changes due to the ChiChi earthquake in the ChouShui River alluvial fan
in Taiwan. Environmental Geology 47, 102118. doi: 10.1007/s00254-
004-1138-5
Lin YP, Chang TK, Shi CW, Tseng CH (2002) Factorial and indicator
kriging methods using a geographic information system to delineate
spatial variation and pollution sources of soil heavy metals.
Environmental Geology 42, 900
909. doi: 10.1007/s00254-002-0600-5Liu G, Drost HJ (1997) Atlas of the Yellow River Delta. (House of
Surveying and Mapping Publishing: Beijing)
Marx DB, Gilmour JT, Scott HD, Ferguson JA (1988) Effects of long-term
water management in a humid region on spatial variability of soil
chemical status. Soil Science 145, 188193. doi: 10.1097/00010694-
198803000-00005
Matheron G (1982) Pour une analyse Krigeante des donnees regionalisees.
Report 732. Centre de Geostatistique, Fontainebleau, France.
Mohawesh O, Ishida T, Fukumura K, Yoshino K (2008) Assessment of
spatial variability of penetration resistance and hardpan characteristics in
a cassava field. Australian Journal of Soil Research 46, 210218.
doi: 10.1071/SR07118
Oliver MA, Webster R, Slocum K (2000) Filtering SPOTimageryby kriging
analysis. International Journal of Remote Sensing 21, 735752.
doi: 10.1080/014311600210542Pardo-Igzquiza E, Dowd PA (2002) FACTOR 2D: a computer program for
factorial cokriging. Computers & Geosciences 28, 857875.
doi: 10.1016/S0098-3004(02)00003-1
Pye K, Blott SJ, Croft DJ, Carter JF (2006) Forensic comparison of soil
samples: Assessment of small-scale spatial variability in element
composition, carbon and nitrogen isotope ratios, color, and particle
size distribution. Forensic Science International 163, 5980.
doi: 10.1016/j.forsciint.2005.11.008
Qiu Y, Fu BJ, Wang J, Chen LD (2001) Spatial variability of soil moisture
content and its relation to environmental indices in a semi-arid gully
catchment of the Loess Plateau, China. Journal of Arid Environments 49,
723750. doi: 10.1006/jare.2001.0828
496 Australian Journal of Soil Research M. Yang et al.
8/2/2019 Yang, Liu, Yang,Sun, & Beazley 2009 Multivariate and Geostatistical Analysis of Soil Salinity in Nested Areas of the Y
12/12
Reis AP, Sousa AJ, Cardoso Fonseca E (2003) Application of geostatistical
methods in gold geochemichal anomalies identification (Motemor-o-
Novo, Portugal). Journal of Geochemical Exploration 77(1), 4563.
doi: 10.1016/S0375-6742(02)00269-8
Reis AP, Sousa AJ, Ferreira da Silva E, Patinha C, Fonseca EC (2004)
Combining multiple correspondence analysis with factorial kriging
analysis for geochemical mapping of the gold-silver deposit atMarrancos (Portugal). Applied Geochemistry 19, 623631.
doi: 10.1016/j.apgeochem.2003.09.003
Rodgers SE, Oliver MA (2007) A geostatistical analysis of soil, vegetation,
and image data characterizing land surface variation. Geographical
Analysis 39, 195216. doi: 10.1111/j.1538-4632.2007.00701.x
Rodrguez JA, Nanos N, Grau JM, Gil L, Lpez-Arias M (2008) Multiscale
analysis of heavy metal contents in Spanish agricultural topsoils.
Chemosphere 70, 10851096. doi: 10.1016/j.chemosphere.2007.07.056
Ryu JS, Lee KS, Kim JH,AhnKH, Chang HW(2006)Geostatistical analysis
for hydrogeochemical characterization of the Han River, Korea:
Identification of major factors governing water chemistry. Bulletin of
Environmental Contamination and Toxicology 76, 17. doi: 10.1007/
s00128-005-0882-x
Saporta A (1990) Probabilits, Analyse Des Donnes et Statistique, Technip,
Paris.Shan K (2007) Theory, methodology and practices of wetland ecological
restoration in Yellow River Delta Nature Reserve. Wetland Science and
Management 3(4), 1620. [In Chinese]
Shao HB, Chu LY, Shao MA (2008) Calcium as a versatile plant signal
transducer under soil water stress. BioEssays 30, 634641. doi: 10.1002/
bies.20770
Song YM, Zhang JF, Xing SJ, Xing SJ, Xi JB (2003) Features of plant
community and its restoration techniques in Yellow River Delta Region.
Journal of Northeast Forestry University 31(6), 8789.
Tarnavsky E, Garrigues S, Brown ME (2008) Multiscale geostatistical
analysis of AVHRR, SPOT-VGT, and MODIS global NDVI
products. Remote Sensing of Environment 112, 535549.
doi: 10.1016/j.rse.2007.05.008
Tavares MT, Sousa AJ, Abreu MM (2008) Ordinary kriging and indicator
kriging in the cartography of trace elements contamination in Sao
Domingos mining site (Alentejo, Portugal). Journal of Geochemical
Exploration 98(12), 4356. doi: 10.1016/j.gexplo.2007.10.002
van Meirvenne M, Goovaerts P (2002) Accounting for spatial dependence in
the processing of multi-temporal SAR images using factorial kriging.
International Journal of Remote Sensing 23, 371
387. doi: 10.1080/01431160010014800
Wackernagel H (1998) Multivariate geostatistics: An introduction with
applications. (Springer Publishing: Berlin)
Wang FY, Liu RJ, Lin XG, Zhou JM (2004) Arbuscular mycorrhizal status
of wild plants in saline-alkaline soils of the Yellow River Delta.
Mycorrhiza 14, 133137. doi: 10.1007/s00572-003-0248-3
Wang SJ, Hassan MA, Xie XP (2006) Relationship between suspended
sediment load, channel geometry and land area increment in the Yellow
River Delta. Catena 65, 302314. doi: 10.1016/j.catena.2006.01.003
Wang Y, Ma T, Luo Z (2001) Geostatistical and geochemical analysis of
surface water leakage into groundwater on a regional scale: a case study
in the Liulin karst system, northwestern China. Journal of Hydrology
246, 223234. doi: 10.1016/S0022-1694(01)00376-6
Weng YL, Gong P (2006) Soil salinity measurements on the Yellow River
Delta. Journal of Nanjing University (Natural Science) 42(6), 602
610.Xu JX (2003) Sediment flux to the sea as influenced by changing human
activities and precipitation: Example of the Yellow River, China.
Environmental Management32(3), 328341.
Xu S, Tao S (2004) Coregionalisation analysis of heavy metals in the surface
soil of Inner Mongolia. The Science of the Total Environment 320,
7387. doi: 10.1016/S0048-9697(03)00450-9
Manuscript received 22 September 2008, accepted 30 April 2009
Analysis of wetland soil salinity Australian Journal of Soil Research 497
http://www.publish.csiro.au/journals/ajsr