spatial distribution of nematodes in a heavy metal contamibated nature reserve thesis 2013
TRANSCRIPT
Promoter: Prof. dr. Ron de Goede
Co-promoter: dr. ir. Gerard Heuvelink
Israel Onoja Ikoyi
Spatial distribution of nematodes in a heavy metal-contaminated nature
reserve
Academic year 2012 – 2013
Wageningen University
Department of Soil Quality
Soil Biology and Biological Soil
Quality Group
&
Wageningen University
Department of Soil Geography and
Landscape
Thesis submitted to obtain the degree of European Master of Science in Nematology
1 | P a g e
Spatial distribution of nematodes in a heavy
metal-contaminated nature reserve
Israel O. IKOYI
Department of Soil Quality, Department of Soil Geography and Landscape, Wageningen
University, Droevendaalsesteeg 4, 6708PB, Wageningen, The Netherlands
Department of Biology, Nematology Section, Ghent University, K.L. Ledeganckstraat 35, B-
9000 Ghent, Belgium
Summary- Soil organisms are often distributed in patches indicating spatial aggregation in the
populations of these organisms. Studies have shown that field distribution of nematodes is aggregated indicating that nematodes also do not occur at random but exhibit spatial dependency. Nematodes are
seen as promising indicators of soil quality and health due to their high diversity and diversity of
ecological roles. This research was set up to evaluate the spatial structure in the distribution of
nematode community indices in a heavy metal-contaminated nature reserve. Also, regression models were built to establish relationships between nematode community indices (maturity index, structure
index, diversity indices, abundances of life history and feeding groups) and a selection of
environmental factors (pH, clay, moisture, organic matter, cadmium and zinc concentrations) to contribute to a better understanding of the possible drivers behind the spatial arrangement of the
nematode communities. Results obtained indicated that about 41 to 86% (relative structure) of the
variation in the distribution of the nematode community indices in the study area was not random but exhibited spatial patterns over distances ranging from 80 to 180 m. The regression models obtained
were significant indicating that the spatial structure in the environmental factors reflected in the spatial
structure of the nematode data. These models explained about 21-50% of the variation in the nematode
data. The nematode community indices showed significant negative correlation with most of the environmental factors and the regression models indicated that soil pH was almost always (significant
in 11 out of the 13 models) a significant factor affecting the distribution of the nematode community
indices in the study area. Detailed analysis of the result showed that factors such as pH and moisture content were the possible drivers behind the spatial structure of the nematode community indices
rather than heavy metal pollution. The regression models were used in predicting and mapping the
spatial distribution of the nematode community indices in a regression kriging approach. The results indicated that the selected environmental factors could not explain all the spatially related variation in
the nematode data (as shown by the spatial structure in the residual data of the regression models of
some of the nematode community indices). This suggests that inclusion of factors other than the ones
considered in the current study could further improve the explanation of the spatial structure of the nematode community indices. There is the need to validate the accuracy of the resulting maps obtained
in this study. Further studies could also explore the use of other forms of non-linear models such as
random forests to see if such models better explain the relationship between nematode community indices and environmental factors.
Keywords –kriging, regression models, spatial heterogeneity, nematode community structure,
cadmium, zinc.
2 | P a g e
In terrestrial ecosystems, the soil biodiversity consists mainly of microorganisms and small
invertebrates (Lavelle & Spain, 2001). Studies have shown that the distribution of these
organisms is not random but display rather spatial patterns which are aggregated over scales
ranging from square millimeters to hectares (Gorres et al., 1998; Stoyan et al., 2000; Ettema
et al., 2000). Competition, configuration of the habitat and historical effects have been
identified as possible factors influencing the spatial structure of communities of micro- and
macro organisms (Martiny et al., 2006; Horner-Devine et al., 2007; Vos & Velicer, 2008). As
a result, populations become isolated and therefore differentiations occur at local spatial
scales. Spatial patterns of soil biota are vertical, horizontal or both. Nematodes are among the
small invertebrates living in the soil.
Nematodes have been reported to be the most abundant metazoans on the Earth‟s surface with
about 108 individuals distributed in a square meter (Lambshed, 2004; Decraemer & Hunt,
2006). They occur in all environments, in every type of soil, under every climatic condition
and in all ecological niches, varying from undisturbed to disturbed ones (Bongers & Ferris,
1999). Different nematode taxa utilize specific food sources and shifts in these feeding groups
often reflect changes in the soil foodweb (Ferris et al., 2001; Yeates et al., 2009). For
instance, members of the Plectidae and Cephalobidae families feed on bacteria but never on
higher plants or fungi. Their transparent body allows easy observation of the internal
structures and determination of feeding habit. They have varying lifespan ranging from a few
days to years. In addition, nematodes have differential sensitivities to different forms of stress
and disturbance including heavy metal contamination (Korthals et al., 1996; Georgieva et al.,
2002; Tenuta & Ferris, 2004; Sanchez-Moreno & Navas, 2007). The combinations of feeding
habit, life span and sensitivity to environmental disturbance have allowed the classification of
nematode taxa with similar response characteristics into functional guilds (Bongers &
Bongers, 1998; Ferris et al., 2001; Ferris & Bongers, 2006). Because of their high diversity
and ecological roles, nematodes are seen as promising bioindicators of environmental health,
soil quality and ecosystem resilience (Bongers 1990; Ferris et al., 2001; Neher, 2001; Hoss &
Traunspurger, 2003; Yeates, 2003; Mulder et al., 2005; Schratzberger et al., 2006; Heininger
et al., 2007). Several studies have reported the effects of soil disturbance on nematode faunal
assemblage and diversity (Hanel, 2003; Kardol et al., 2005; Korenko & Schmidt, 2006).
Recently, lower nematode abundance and trophic diversity (a measure of the relative
abundance and evenness of the occurrence of nematode trophic groups) have been reported
from heavy metal contaminated sites (Park et al., 2011; Salamun et al., 2012).
3 | P a g e
Field distribution of nematode taxa has been described as aggregated indicating that nematode
data are spatially dependent. Considerable variability in the spatial patterns of species
composition and abundance of plant parasitic nematodes has been reported from agricultural
fields (which are perceived to have relatively homogenous soils) (Robertson & Freckman,
1995). Previous studies showed that nematode distribution and soil properties often are
correlated and can have spatial patterns (Ettema et al., 1998; Liang et al., 2005; Monroy et al.,
2012; Park, 2012). There is however limited information on models that relate nematode
distribution with environmental variables. Such models are needed for a better understanding
of the spatial patterns of soil biota such as nematodes in natural ecosystems which are far less
homogenous than agricultural soils. These spatial patterns can have important effects on the
patterns of above ground plant community structure (Ettema & Wardle, 2002). The reverse
relationship is also true. In addition, data from sites with useful gradients in natural and
anthropogenic soil characteristics presents possibilities to study „natural‟ relationships
between nematodes and their environment and also the effects of artificially introduced
stresses.
The floodplains of the „Dommel‟ river (tributary of the river Meuse) in the south of The
Netherlands are heavily contaminated by heavy metals such as cadmium and zinc, due to
former upstream mining activities. The highest topsoil Cd concentrations (up to 100 mg/kg) in
The Netherlands are found in the Malpiebeemden nature reserve. The distribution of the
contamination in the floodplains is very heterogeneous, resulting in large gradients of heavy
metals (Bleeker & Gestel, 2007). The river Dommel is a rainfed river being almost always
flooded after spring rains. Upstream, close to the Dutch-Belgian border, is a factory that
produces zinc and cadmium. Since 1888 waste discharge into the Dommel resulted in serious
heavy metal contamination in the downstream area. In the mid-1990s, about 1-3 kg cadmium
and 50-200 kg zinc were transported daily by the river. At present the concentrations in the
river water are lower, but concentrations in the flood plains are still elevated (Postma, 1995;
Bleeker & Gestel, 2007). In 2008, soil moisture, pH, SOM, Total and available Cd and Zn
concentrations, Earthworm number and Biomass were measured from a spatial grid of 100
sampling points but no data on nematodes were considered.
Against this background, there is therefore the need to study the spatial distribution of
nematodes in relation to environmental properties (including heavy metal-contamination) as
this will make sampling for bioindication surveys more efficient.
This study was designed to achieve the following objectives:
4 | P a g e
1. To evaluate the spatial structure of the nematode faunal assemblage along the Dommel
river.
2. To establish the relationships between nematode community indices and
environmental factors including heavy metal concentrations.
3. Use these relationships in mapping the spatial distribution of the nematodes in a
regression kriging approach.
The research questions addressed in this study include: a) Does spatial heterogeneity exist in
the distribution of soil nematodes in the contaminated site? b) How are the nematode
distribution and other environmental factors including soil heavy metal content related? c)
Can geostatistical models be developed and used to predict the spatial distribution of
nematodes at unsampled locations in the Malpiebeemden area?
The hypotheses considered in this study include:
1. Spatial heterogeneities exist in the distribution of nematode taxa in the contaminated
site and this will be associated with specific patterns in their life-history
characteristics. The c-p2 nematodes will aggregate and be dominant even in spots with
higher heavy metal concentration, while persisters (K-selected species) will have
lower incidences and relative abundances in such spots as a result of their sensitivity to
disturbances, such as heavy metal contamination.
2. An increased complexity of the nematode assemblage (as indicated by taxa richness
and diversity) will be associated with spots with low levels of disturbance, because
stressed conditions (heavy metal pollution, moisture and oxygen stress) favour mainly
the tolerant taxa (c-p2).
3. The selected abiotic environmental (soil) properties that were measured in the
Malpiebeemden area are sufficient to explain the spatial structure of the nematode
fauna. The remaining, i.e. unexplained, residual variation will have no spatial
component.
5 | P a g e
Materials and methods:
STUDY AREA DESCRIPTION
This study was conducted in the Malpiebeemden nature reserve located in the south of The
Netherlands, in the province of North-Brabant, in between the municipality of Valkenswaard
and the Dutch-Belgian border. The area of 122 ha comprises grassland, bog, forested bog with
several ponds, and is bordered at the east by the river Dommel. The area is owned by
Staatsbosbeheer (National Forest Conservation Organisation), but it is managed by
Natuurmonumenten (Organization for the Conservation of Nature Reserves). Figure 1a is the
map of the Netherlands showing the location of Wageningen and the study area.
The Malpiebeemden area was selected for this research because not only do the spatial
distribution of Cd and Zn differ, but also other environmental properties such as soil texture,
organic matter content, pH, moisture and elevation are spatially variable (Bleeker & Gestel,
2007). This research focused on a relative small part within the Malpiebeemden (5o27‟EL;
51o18‟NB), comprising grassland and small forest patches (Figure 1b) south of Groot Malpie-
ven and West of the river Dommel. The size of the studied area is about 23 hectares. The area
is characterized by four different soil types (Figure 1c).
6 | P a g e
Figure 1. Map of a) the Netherlands indicating Wageningen and the study area, b) the study
area and the forest patches marked with yellow and green outline respectively, c) the study
area showing the different soil types (world classification system) according to De Bakker,
(1979).
SOIL SAMPLING
The sampling design was a stratified random sampling because this yields a fairly uniform
spread over the area while still having data for doing sound validation using sampling theory
from statistics (De Gruijter et al., 2006). It also yields short-distance comparisons useful for
variogram fitting. The area was divided into eight strata of equal size using the spcosa
7 | P a g e
package (in the R statistical software). The spcosa-package provides algorithms for spatial
coverage sampling and for random sampling from compact geographical strata (De Gruijter et
al., 2006, Walvoort et al., 2010). Prior to stratification, the forested patches (Figure 1b) were
masked in order to have soil samples from uniform vegetation (grassland). Ten random soil
sampling points were chosen from each stratum making a total of 80 soil sampling points
distributed over the entire study area (Figure 2). Also, before going to the field, five extra
random soil sampling points per stratum were selected as a precaution in case any of the
selected main sampling points turns out to be in an unsuitable location (i.e. inside the forest
patches) in the field. These five additional soil sampling points had a ranking order in such a
way that the rank is taken into consideration when any of the main sampling points is to be
replaced. In the field, the positions of the soil sampling points were located with a
Geographical Positioning System (GPS) based on the beforehand selected coordinates. At
each of the sampling locations, one soil sample was taken from 0-20 cm soil depth using a 4
cm diameter soil corer.
NEMATODE EXTRACTION AND IDENTIFICATION
Nematodes were extracted from 100 g fresh soil samples using the Oostenbrink elutriator
(Oostenbrink, 1960). They were counted (10% of the total extracted volume) using a high
magnification microscope at 100-400 times magnification. The nematodes were heat killed
and fixed in 4% formaldehyde and identified to the genus or family based on morphological
features and identification keys (Bongers, 1988; Andrassy, 1984; Siddiqi, 1986; Jairajpuri &
Ahmad, 1992). This was the highest achievable taxonomic resolution and is typically the
finest resolution used in studies that report composition of soil nematode communities.
8 | P a g e
Figure 2. Study area showing sampling points.
CHARACTERIZATION OF NEMATODE FAUNAL ASSEMBLAGE
The nematodes were categorized into the various feeding groups (Yeates et al., 1993) and c-p
groups (Bongers & Bongers, 1998). The total nematode abundance (individuals/100g soil),
abundance of feeding groups (individuals/100g soil), abundance of c-p groups
(individuals/100g soil), Genera richness (GR), Shannon-Weiner (H'), maturity, enrichment,
channel and structure indices were calculated for each of the 80 samples. Genera richness and
Shannon-Weiner index were selected as diversity indices based on their wide usage in most
ecological studies (Neher & Darby, 2009).
Maturity index was calculated (for the free-living taxa) as:
9 | P a g e
∑
(1)
Where vi is the colonizer-persister (c-p) value assigned to the family i, fi is the frequency of
family i in the sample and n is the total number of individuals of the free-living taxa in the
sample.
Food web indices were calculated following Ferris et al. (2001) and Ferris & Matute (2003).
All nematode genera were assigned to functional guilds associated with a specific weight for
each group. Basal (b), enrichment (e) and the structure (s) component of the nematode
assemblage were calculated as:
( ) (2)
( ) ( ) (3)
( ) ( ) ( ) ( ) ( ) (4)
where Ba1&2, Fu2, Ca2, Can and Omn are bacterial-feeding, fungal-feeding, carnivorous and
omnivorous nematodes belonging to 1, 2 and n c-p groups, respectively; n is the sum of the
c-p groups 3, 4 and 5; W1=3.2, W2=0.8, W3=1.8, W4=3.2 and W5=5.0. In this way, enrichment
and structure indices were calculated as:
(
) (5)
(
) (6)
Channel index ( ) was calculated as the proportion of fungal-feeding nematodes in c-p2
(Fu2) within the opportunistic decomposer guilds i.e. fungal-feeding nematodes in c-p2 and
the bacterial-feeding nematodes in c-p1 (Ba1). Taking into consideration the specific weights
of each guild, CI was calculated as:
(
) (7)
These calculations were done for each of the 80 samples. Pictorial presentation of the faunal
composition was summarized in a food web diagram (Ferris et al., 2001). For the food web
diagram, the 80 sample were divided into 3 pollution gradients (based on their total zinc
concentrations): group A (unpolluted samples with total Zinc concentration less than or equal
10 | P a g e
to the Dutch target value of 140 mg/kg), group B (moderately polluted with total Zinc
concentration above the target value but less than the Dutch intervention value of 720 mg/kg)
and group C (heavily polluted with total Zinc concentration above the Dutch intervention
value of 720 mg/kg).
KRIGING SOIL PHYSICAL AND CHEMICAL PROPERTIES
Kriging is a geostatistical interpolation or mapping approach where the value of a random
variable at one or more unsampled locations is predicted by calculating some weighted
average of the observations (Webster & Oliver, 2007; Hengl et al., 2007). The general kriging
formula is given as:
( ) ∑ ( )
( )
where ( ) is the predicted value of the target variable at an unvisited location given its
map coordinates, the sample data ( ) ( )...., ( ) and their coordinates. The weights ( )
are selected in such a way that the error associated with the prediction variance is minimized
producing weights that depend on the spatial autocorrelation structure of the variable. This
type of interpolation approach is often referred to as ordinary kriging (OK) in most literature.
The expected error associated with the prediction is given as:
( ( ) ( )). (9)
For each kriged estimate, there is a variance associated with the kriging which is calculated
as:
( ) { ( ( ) ( ))}
2 . (10)
The standard deviation is calculated by taking the square root of the variance and used in
plotting the kriging standard deviation map. A crucial step in kriging is the calculation and
fitting of the semivariogram. This involves calculating the semivariance (half of the squared
difference between values at paired locations) for all pairs of locations separated by distance
(h).
The general formula for calculating the semivariance as given in Ettema & Wardle (2002) is:
11 | P a g e
( )
( ) ∑[ ( ) ( )]
( )
where N(h) is the number of pairs observations separated by distance h, z(si) is the value of
the variable of interest at location si, and z(si+h) is its value at a location at distance h from si.
This is often presented as a graph of the averaged semivariances (plotted against distance)
known as the semivariogram. The semivariogram presents a picture of the spatial dependency
of the data based on the principle of geography that states: closer things are more similar than
distant things (Ettema & Wardle, 2002). A hypothetical semivariogram is shown in Figure 3.
Figure 3. Generalized semivariogram showing the various properties.
The range is the distance at which the data exhibit spatial dependency and beyond which the
data is independent. The Sill (C) is an estimate of the total population variance. The nugget
(intercept Co) is the variance due to sampling error or spatial heterogeneity at scales not
sampled, while the difference between the sill and the nugget gives the partial sill (spatially
structured variance). In a situation where there are no spatial relationships between samples
(i.e. the association is entirely random), it is called a “pure nugget effect”. In such cases, the
fitted model (the solid line) will be a straight horizontal line as the semivariances will be
scattered just around the total sample variance (the sill).
In 2008, soil samples collected from 100 sampling points in the study area were analyzed for
pH, organic matter, clay and moisture contents, and total (extraction in 2 M HNO3) and
available (extraction in 0.01 M CaCl2) concentrations of Cadmium and Zinc. The various soil
physical and chemical properties were mapped using ordinary kriging (OK). The results of the
OK predictions of soil physical and chemical properties at the 80 locations for our study were
obtained by overlaying these locations on the OK prediction maps. All this was done in the R
statistical software. A sample R script used for this analysis is attached in Appendix 1.
12 | P a g e
STATISTICAL ANALYSES
Estimation of spatial heterogeneity in the distribution of nematode community indices
Geostatistical tools in R (R Development Core Team, 2013) were used to calculate and fit the
semivariograms of the nematode data to test for spatial heterogeneity. The semivariances
were calculated according to equation 11. The relative structure (C/C+Co) of the
semivariograms i.e. the proportion of the sample variance that is spatially autocorrelated was
calculated for each of the nematode parameters that showed spatial heterogeneity (Robertson
& Freckman, 1995). To meet the statistical assumptions of normality, the nematode count
data were log-transformed prior to the semivariogram fitting. Nematode data that showed no
spatial structure (Pure nugget effect) were excluded from further analyses (modelling and
mapping). The R scripts for these analyses are presented in Appendix 2.
Modelling relationships between the nematode community indices and soil properties
Multiple linear regression models were used to establish the relationship between the
dimensionless nematode indices i.e. maturity index, structure index, Shannon index and
genera richness (as dependent variables) and soil properties (pH, organic matter, Moisture,
clay content, Cd and Zn concentrations) as explanatory variables. As a result of a nearly
perfect positive correlation between the total and available concentrations of Cd and Zn,
models were built using only the total concentrations of the metals and the other soil
properties as explanatory variables. The regression was done in a stepwise manner to select
the best predictors based on the Akaike Information Criterion (AIC). The AIC is a method of
selecting a model from a set of candidate models in which the chosen model is the one that
minimizes the Kullback-Leibler distance (a measure of the difference between two probability
distributions) between the model and the truth (Burnham & Anderson, 2002). In this way,
over-fitting of the model is avoided. Models with the lowest AIC are considered to be the
best. The goodness of fit of the models was examined by a plot of the observed values against
the fitted values. Also, the normality plot of the residuals was examined as the residuals of a
multiple linear regression model are assumed to have a normal distribution (See Appendix 3
for sample R script). The general form of a multiple linear regression model is:
(12)
13 | P a g e
Where Y is the dependent variable, is the intercept, are the coefficients of
the predictors (explanatory variables) and while is the residual.
Poisson regression is often used for modelling the relationship between environmental
properties and counts of organisms because the statistical distributions of the counts of
organisms are generally skewed, and hence not normally distributed. The assumption of the
Poisson distribution is that the variance is equal to the mean. However, it is often observed
that the amount of variation for each sampling unit is typically higher than that expected by a
pure Poisson process. This situation is referred to as over-dispersion (where the conditional
variance is greater than the conditional mean) making estimates from such models to be
erroneously interpreted. For our data, over-dispersion was observed in the Poisson regression
model. Although there are various recommendations for modelling over-dispersed data, a
more formal way to accommodate over-dispersion in a count data regression model is to use a
negative binomial model (Zeileis et al., 2008). The negative binomial regression analysis was
used to model the relationship between counts of nematode abundance (total abundance,
abundance of feeding groups and c-p groups) and environmental factors. This was done in R
using the glm.nb() function in the MASS package (Venables & Ripley, 2002). Unlike the
Poisson regression model, the negative binomial model assumes that the conditional mean are
not equal to the conditional variance and captures this by estimating a dispersion parameter.
One advantage of the negative binomial model over other methods recommended for
modelling over-dispersed count data is that it allows for stepwise variable selection based on
information criteria such as the AIC. The negative binomial regression can be seen as a form
of Poisson regression in which the log of the dependent variable is predicted with a linear
combination of the variables. The general form of the negative binomial regression model is:
( ) (13)
Where ( ) is the log of predicted counts, is the intercept while are the
coefficients of the predictors (explanatory variables) and . The output of the negative
binomial regression analyses also includes a null deviance (the amount of variation in the
model using only the intercept as predictor) and a Residual deviance (the amount of variation
in the model predictions when other predictors are added to the model). It is often expected
that the addition of other predictors to the null model (intercept-only) should reduce the null
deviance if the predictors are statistically significant to be included in the model.
14 | P a g e
The goodness of fit of all models was examined by plotting the observed values against the
fitted values. The R scripts used for these determinations are provided in Appendix 4.
kriging of the nematode community indices
Geostatistical kriging of the nematode community indices was done in a regression kriging
approach. The generic model for the regression kriging prediction is given as:
(14)
Where is the dependent variable, is the outcome of the regression analyses (trend)
and is the difference between the measured and predicted values (residual). The residuals
are checked for possible spatial dependency. If the residuals are spatially dependent, they are
then added to the regression models for improved explanation of the variation in the
dependent variable.
Geostatistical tools in R (R Development Core Team, 2013) were used to calculate and fit the
semivariograms of the residuals from the regression model to test for spatial autocorrelation.
The relative structure (C/C+Co) of the semivariograms i.e. the proportion of the sample
variance that is spatially autocorrelated was calculated for each of the residuals that showed
spatial autocorrelation (Robertson & Freckman, 1995). For the count data of nematode
abundances, the residuals were square root transformed prior to the semivariogram fitting
since they did not meet the assumptions of normality. The properties of the semivariograms
were observed for spatial dependency and where this was true; interpolation of the residuals
over the entire study area was done using a simple kriging approach (like ordinary kriging but
here the mean is assumed to be constant and fixed at 0). The interpolated values of the
residuals were then back-transformed and added to the predicted values from the regression
models. The summed predictions were then used in mapping the nematode spatial
distributions. This approach is referred to as the regression kriging method (Odeh et al., 1994;
1995). The algorithm and steps involved in regression kriging are described in Hengl et al.
(2007). A summary of the regression kriging steps is given below:
1. Select explanatory variables and fit regression model (estimate regression
coefficients).
2. Compute residuals (by subtracting the fitted trend from the observations) at
observation locations and compute from them a semivariogram.
15 | P a g e
3. Apply the regression model to all unobserved locations.
4. Krige the residuals (mostly done using simple kriging).
5. Add up the results of steps 3 and 4.
The kriging standard deviations (the square root of the kriging variance) were also computed.
The ratio of the kriging standard deviations to the predicted values indicates the coefficient of
variation which was used as a measure of the accuracy of predictions.
Results
NEMATODE FAUNAL ASSEMBLAGE
A total of 69 genera belonging to 33 families were identified from the area. The total number
of nematodes varied from 75 individuals to about 10,000 individuals per 100 g of fresh soil.
The most prevalent genera were Plectus (occuring in 96% of the sample), Aphelenchoides and
Eucephalobus (94%). These three genera belong to the c-p2 nematode grouping. The
nematode maturity index ranged between 1.3 and 2.8. Figure 4 shows box plots of the various
nematode community indices. The Shanon index ranged from 1.1 to 3.0 while the number of
genera per sample (genera richness) was between 8 and 32. The structure, enrichment and
channel indices ranged from 0 to 88, 4 to 93 and 0 to 100% respectively. For the nematode
life-history groups, the c-p2 nematodes were present in all samples and constituted about 65%
of the total nematode abundance. The relative proportions of c-pl, c-p3, c-p4 and c-p5 were
about 10, 20, 4 and 1% respectively. Among the feeding group distributions, the relative
proportion of bacterial feeders was highest (about 50% of the total nematode abundance). The
omnivores and predators were the least abundant with a relative proportion of about 3% of the
total nematode abundance.
The division of the samples based on the heavy metal pollution gradient showed that 70% of
the samples could be considered unpolluted (group A with total Zn concentration <140
mg/kg- the Dutch target value for Zn in soils), 21% moderately polluted (group B with total
Zn concentrations greater than the Dutch target value but less than the Dutch intervention
value for Zn in soils-720 mg/kg) and the remaining 9% as heavily polluted (group C having
Zn concentrations 720 mg/kg-the Dutch intervention value for Zn concentration in soils). As
can be seen from the foodweb analysis diagram (Figure 5), groups A and B were found in
each of the 4 quadrants. The heavily polluted samples (although there were few points) had
16 | P a g e
SI< 60%. Also, some of the samples showed some kind of eutrophication having relatively
high levels enrichment.
17 | P a g e
Figure 4. Box Plots of (a) maturity index (MI) (b) Shannon index (H'), (c) genera richness
(GR), (d) structure index (SI), enrichment index (EI) and Channel index (CI), (e) total
nematode abundance (TNEM), abundance of c-p groups and feeding groups (BF=bacterial
FF=fungal feeders, PF=plant feeders, OMV=omnivores, PRED=predators,
Ectop=ectoparasite subgroup of plant feeders, RHF=epidermal cell and root hair feeders
(subgroup of the plant parasites).
18 | P a g e
Figure 5. Food web diagnostic diagram: A=unpolluted, B=moderately polluted and
C=heavily polluted.
SEMIVARIOGRAMS OF THE DISTRIBUTION OF NEMATODE COMMUNITY INDICES
Semivariograms were computed and plotted to test for spatial heterogeneity in the distribution
of the nematode community indices. The results are shown in Figures 6-8. Most of the
nematode community indices showed spatial heterogeneities as evidenced from the
semivariograms. Table 2 summarizes the semivariogram characteristics. About 43-86% of the
distribution of nematode community indices was spatially dependent ranging over distances
between 80-180 m. The semivariograms of the log-transformed abundances of c-p1 (Figure
6b) nematodes, the root hair feeders (Figure 7g), and the enrichment and channel indices
(Figures. 8e&f) showed a pure nugget effect.
19 | P a g e
Figure 6. Semivariograms of the distributions of (a) total nematode abundance, (b) c-p1
nematodes, (c)c-p2 nematodes and (d)c-p3-5 nematodes.
20 | P a g e
Figure 7. Semivariograms of distributions of feeding groups (a) bacterial feeders, (b) fungal
feeders, (c) plant feeders, (d) omnivores, (e) predators and (f) ectoparasites (g) root hair
feeders.
21 | P a g e
Figure 8. Semivariograms of the distributions of (a) maturity index, (b) structure index, (c)
Shannon index, (d) Genera richness, (e) enrichment index and (f) channel index.
22 | P a g e
Table 1. Semivariogram characteristics of the distribution of the nematode community
indices.
Nematode
parameter
Model Co (nugget) C(partial sill) C/C+Co
(relative
structure)
Range
Total
nematode
abundance
Spherical 0.4 0.7 0.64 180
c-p1 Pure Nugget
c-p2 Spherical 0.6 1.0 0.63 165
c-p3-5 Spherical 1.2 2.0 0.63 140
Bacterial
feeders
Spherical 0.8 1.3 0.62 160
Fungal
feeders
Spherical 0.3 1.1 0.79 99
Plant feeders Spherical 0.5 0.35 0.41 100
Omnivores Spherical 1.6 1.8 0.53 80
Predators Spherical 0.6 2.3 0.79 90
Ectoparasites Spherical 1.2 3.0 0.71 130
Root hair
feeders
Pure Nugget
Maturity
index
Spherical 0.04 0.03 0.43 100
Enrichment
index
Pure Nugget
Structure
index
Spherical 158.0 402.0 0.72 158
Channel
index
Pure Nugget
Shannon
index
Spherical 0.05 0.09 0.64 140
Genera
richness
Spherical 3.0 19.0 0.86 130
23 | P a g e
KRIGED SOIL PHYSICAL AND CHEMICAL PROPERTIES
The results of the kriged soil physical and chemical properties used in this study and the
summary statistics is shown in Table 2. The pH was in the acidic range.
Table 2. Descriptive statistics of soil physical and chemical properties used in this study.
pH Moisture
(%dwt)
Clay
(%)
Organic
matter
(%dwt)
Cd total
(mg/kg)
Cd
available
(mg/kg)
Zn total
(mg/kg)
Zn
available
(mg/kg)
Min. 3.7 11.92 0.53 2.10 0.07 0.07 5.3 3.4
1st Qu. 4.3 18.43 0.65 3.69 0.44 0.22 19.9 11.7
Median 4.5 35.16 1.19 6.39 1.66 0.40 66.4 23.7
Mean 4.4 38.12 2.47 7.12 10.29 2.35 295.1 60.5
3rd Qu. 4.6 52.76 2.54 10.33 13.53 2.57 298.7 77.9
Max. 4.9 79.27 10.53 16.20 68.90 14.14 1329.0 328.5
The semivariograms of the soil properties (Figure 9) indicated that the soil properties were
spatially dependent. The kriged maps of the soil physical and chemical properties that were
selected for this study are shown in Figure 10. The highest values of most of the soil
properties are located at spots closer to the river (Eastern part of the maps).
24 | P a g e
Figure 9. Semivariograms of the soil properties.
25 | P a g e
Figure 10. Maps of the distribution of the soil properties used as predictors.
26 | P a g e
CORRELATION ANALYSES OUTPUT
Most of the soil properties (especially the total and available concentrations of Cd and Zn)
were strongly, positively cross-correlated (Figure 11). Most of the nematode parameters were
significantly (P<0.05) negatively correlated with soil properties (Table 3). The total number
of nematodes (log transformed), Maturity index, structure index and Shannon-Weiner Index,
genera richness, the log-transformed abundances of c-p3-5, plant feeders, omnivores and
predators were significantly negatively correlated with all the soil properties. The
bacterivorous nematodes were significantly negatively correlated with all the soil properties
with the exception of the clay content where no significant correlation was found. The log-
transformed abundances of c-p2, fungivorous and ectoparasitic nematodes were significantly
negatively correlated with only soil organic matter, moisture and pH. The enrichment index
and channel Index, the log-transformed abundances of c-p1 nematodes and root hair feeders
did not show any statistically significant correlation with the soil properties. As a result of
this, they were excluded from modelling and regression kriging predictions analyses.
27 | P a g e
Figure 11. Scatter plot-correlation matrix between soil properties.
28 | P a g e
Table 3. Pearson‟s correlation coefficients between the nematode community indices and soil
properties.
Clay SOM Cdtot Zntot Moisture pH
Total numbers -0.23 -0.45 -0.29 -0.30 -0.50 -0.61
Maturity Index (MI) -0.29 -0.42 -0.28 -0.33 -0.42 -0.44
Structure Index (SI) -0.38 -0.46 -0.36 -0.40 -0.45 -0.48 Enrichment Index (EI) 0.01 0.07 0.01 0.04 0.04 0.03
Channel Index (CI) 0.10 0.10 0.07 0.06 0.16 0.23
Shannon-Weiner Index (H') -0.31 -0.43 -0.26 -0.30 -0.46 -0.46 Genera richness -0.35 -0.48 -0.33 -0.34 -0.49 -0.51
log c-p1 -0.08 0.10 -0.08 -0.02 -0.02 -0.03
log c-p2 -0.14 -0.36 -0.21 -0.21 -0.42 -0.52
log c-p3-5 -0.38 -0.56 -0.37 -0.40 -0.59 -0.63 logFungivores(FF) -0.08 -0.30 -0.14 -0.13 -0.33 -0.37
logBacterivores(BF) -0.17 -0.39 -0.22 -0.23 -0.45 -0.56 logPlant-Feeders (PF) -0.33 -0.41 -0.34 -0.37 -0.41 -0.45
Omnivores -0.31 -0.39 -0.35 -0.35 -0.45 -0.40
Predators -0.27 -0.45 -0.29 -0.32 -0.49 -0.58
logRoot-hair feeders(RHF) -0.09 -0.01 -0.03 -0.03 -0.04 0.05
logEctoparasites -0.17 -0.39 -0.20 -0.24 -0.44 -0.46 *Values in bold shows significant correlation (p<0.05); SOM=soil organic matter content; Cdtot, and Zntot =total Cd and Zn concentrations
RESULTS OF REGRESSION MODELLING ANALYSIS
All the nematode data that showed spatial autocorrelation and significant correlations with the
soil properties were included in the regression modelling analyses. A stepwise regression
analyses was performed to model the relationship between the nematode parameters and soil
properties (Clay, Organic matter, Moisture, pH, total concentrations of Cd and Zn) as
predictors. The summary of the final model output is given in Tables 4&5. The estimated
model for the MI had pH and the total concentrations of the heavy metal (Cd and Zn) as its
predictors. This model explained about 21% of the variation in MI. The overall relationship
between MI and the predictors was significant (F3,76=7.80, p<0.0005). The estimated model
for the SI (with Organic matter, moisture and pH as predictors) explained about 25% of the
variation in the SI and the overall model was significant (F3,76=9.91, p<0.0005). For the
diversity indices, the Shannon index (H') had pH and moisture content of the soils as its
predictors with about 22% of the variation in the H‟ explained by this model (a significant
model with F2,77=12.14, p<0.0005) while the genera richness (GR) model (having pH, clay
and total Zn concentration as its predictors) explained about 27% of the total variation in the
GR (a significant model with F3,76=10.52, p<0.0005). Soil pH was a significant predictor in
all these 4 models.
29 | P a g e
As shown in Table 4 (given the values for the estimated regression coefficient of the
predictors), the estimated models relating the dimensionless nematode indices with the soil
properties are:
(15)
(16)
(17)
(18)
Table 4. Summary of multiple linear regression analysis for the dimensionless nematode
indices.
MI=maturity index, SI=structure index, H'=Shannon index, GR=genera richness, Cdtot=total cadmium concentration, OrgMat=soil organic
matter content, Zntot=total zinc concentration, Std. error= standard error of coefficients and Pr(>|t|)=probability value associated with the t
test.
The estimated negative binomial regression model (Table 5) for the total number of
nematodes observed in the study had pH, Organic matter and Clay contents as its predictors.
This model was significant as it reduced the null deviance (intercept-only model) by a factor
of 67.17 (about 44% reduction in the variance). For the log counts of c-p2, c-p3-5, bacterial
feeders, fungal feeders, omnivores and ectoparasites, there were significant reductions in the
null deviance due to the addition of the predictors by 38, 50, 40, 26, 24, 31, 40 and 27%
respectively (these percentages are the difference between the null deviance and the residual
Indices Model
parameters
Estimate Std.
error
t value Pr(>|t|) Adjusted
R2
MI Intercept 3.66 0.56 6.53 <0.001 0.2053
pH
Cdtot
-0.31
0.02
0.001
0.01
-2.39
1.85
0.02
0.07
Zntots -0.001 0.001 -1.99 0.05
SI Intercept 216.7 54.21 4.00 <0.001 0.2527
pH
OrgMat
-36.80
-5.72
13.60
3.14
-2.71
-1.82
0.008
0.07
Moisture 0.99 0.69 1.43 0.14
H' Intercept 4.10 0.87 4.73 <0.001 0.2199
pH
Moisture
-0.36
-0.01
0.22
0.003
-1.66
-1.66
0.10
0.09
GR Intercept 58.13 8.78 6.62 <0.001 0.2655
pH
Clay
-8.57
-0.79
2.01
0.40
-4.25
-1.96
<0.001
0.05
Zntot 0.01 0.004 1.54 0.13
30 | P a g e
deviance). Soil pH was an important predictor in 7 out of the 9 models. As shown in Table 5
(the estimated coefficients of the predictors), the estimated negative binomial regression
models relating the count of nematode abundances with soil properties are:
( ) (19)
( ) (20)
( ) (21)
( ) (22)
( )
(23)
( ) (24)
( ) (25)
( ) (26)
( ) (27)
In general, soil pH was selected as a significant predictor in 11 out of the 13 models.
31 | P a g e
Table 5. Summary of negative binomial regression output for the nematode counts
BF=bacterial feeders, FF=fungal feeders, PF=Plant feeders, Cdtot=total cadmium concentration, OrgMat=soil organic matter content,
Zntot=total zinc concentration, Std. error= standard error of coefficients and Pr(>|z|)=probability value associated with the significance of
predictors.
Indices Model
parameters
Estimate Std.
error
t
value
Pr(>|z|) Null
deviance
Residual
deviance
Total
nematode
number
Intercept 14.91 1.69 8.83 <0.001 154.08 86.91
pH -1.61 0.42 -3.82 <0.001
OrgMat -0.11 0.05 -2.33 0.02
Clay 0.12 0.06 2.09 0.04
CP2 Intercept 12.50 2.31 5.40 <0.001 145.96 89.92
pH -1.23 0.59 -2.11 0.03
Moisture -0.04 0.01 -2.98 0.003
Clay 0.19 0.06 3.10 0.002
CP3-5 Intercept 15.71 2.52 6.25 <0.001 190.65 95.07
pH -2.01 0.63 -3.22 0.001
OrgMat -0.26 0.07 -3.58 <0.001
Clay 0.16 0.08 1.88 0.06
BF Intercept 14.51 2.46 5.90 <0.001 151.52 91.15
pH -1.61 0.62 -2.59 0.01
Moisture -0.03 0.01 -2.64 0.008
Clay 0.18 0.07 2.75 0.006
FF Intercept 10.14 2.64 3.85 <0.001 123.31 91.56
pH -1.17 0.65 -1.80 0.07
OrgMat -0.15 0.07 -2.28 0.02
Cdtot -0.09 0.04 -2.47 0.01
Zntot 0.004 0.002 1.81 0.07
Clay 0.27 0.13 2.17 0.03
PF Intercept 11.12 1.79 6.22 <0.001 115.83 88.11
pH -0.98 0.44 -2.22 0.03
OrgMat -0.06 0.03 -1.84 0.07
Omnivores Intercept 5.01 0.68 7.41 <0.001 102.51 70.77
OrgMat 0.98 0.37 2.70 0.007
Cdtot -0.09 0.03 -2.83 0.005
Moisture -0.22 0.07 -3.22 0.001
Predators Intercept 15.10 6.09 2.48 0.01 93.68 56.04
pH -2.44 1.50 -1.62 0.10
OrgMat -0.40 0.12 -3.21 0.001
Ectoparasites Intercept 7.10 0.42 17.08 <0.001 133.27 97.71
Cdtot 0.14 0.06 2.58 0.01
Zntot -0.01 0.003 -2.45 0.01
Moisture -0.07 0.01 -5.18 <0.001
Clay 0.29 0.16 1.87 0.06
32 | P a g e
ANALYSIS OF SPATIAL STRUCTURE IN RESIDUALS
The semivariograms of the residuals from the regression modelling analyses of the nematode
community indices are shown in Figures 12-14. Most of the semivariogram of the residuals
exhibited a pure nugget effect (having their semivariances scattered around the total sample
variance-the sill) except for structure index (Figure 12b), Genera richness (Figure 12d), c-p2
nematodes (Figure 13b), bacterial feeders (Figure 14a) and fungal feeders (Figure 14b). The
characteristics of the semivariogram of residuals of all the nematode parameters that showed
spatial autocorrelation are shown in Table 6. The relative structure (C/C+Co) is the proportion
of sample variance that is spatially autocorrelated. For these five models of the nematode
community indices, 63-95% of the residual variance was still spatially dependent over
distances ranging from 60-150 m.
Table 6. Semivariogram characteristics of the regression model residuals that showed spatial
autocorrelation.
Nematode
parameter
Model Co (nugget) C(partial sill) C/C+Co
(relative
structure)
Range
SI Matheron 23.14 413.51 0.95 87
GR Matheron 3.00 14.00 0.82 80
c-p2 Spherical 0.05 0.15 0.75 150
BF Matheron 0.06 0.18 0.75 60
FF Spherical 0.07 0.12 0.63 100 SI=structure index, GR=Genera richness, BF=bacterial feeders, FF=fungal feeders
33 | P a g e
Figure 12. Semivariograms of residuals from the regression models for (a) Maturity index,
(b) Structure index, (c) Shannon-Weiner index and (d) Genera richness.
34 | P a g e
Figure 13. Semivariograms for residuals of the regression models of (a) Total nematode
number, (b) c-p2 nematodes and (c) c-p3-5 nematodes.
35 | P a g e
Figure 14. Semivariograms for residuals of regression models of (a) Bacterivorous (b)
Fungivorous (c) Plant feeding (d)Omnivorous (e) Predatory and (f) Ectoparasitic nematodes
REGRESSION KRIGING AND ASSOCIATED STANDARD DEVIATION MAPS OF PREDICTIONS OF
DISTRIBUTION OF NEMATODE COMMUNITY INDICES
Dimensionless nematode indices
The semivariograms of the residuals from the multiple linear regression analyses of maturity
index and Shannon index were found to be spatially independent (Figures 12a&c) and so
predictions of these indices were done with the regression model alone. For the structure
index and Genera richness, the semivariograms were found to show spatial autocorrelation
(Figures 12b&d). These residual semivariograms were interpolated over the entire study area
(Figures 16b and 18b) and then summed up with the regression model predictions in a
regression kriging approach to map the distribution of SI (Figure 16c) and GR (Figure 18c)
36 | P a g e
over the study area. A visual comparison of the maps of the SI and GR from the regression
model only with the regression kriging maps (regression plus the interpolated residuals)
showed that the addition of the maps of the interpolated residuals further improved the maps
(Figures 16a and 18a) as patterns that were not initially visible became differentiated and
visible. In general, the lowest MI predictions (<2.0) were obtained at locations closest to the
river while MI predictions higher than 2.4 were obtained at locations farther from the river. A
similar trend was observed for the SI where predictions higher than 60% were obtained
farther away from the river. The standard deviation maps (a measure of the error associated
with the regression predictions) was also obtained. For the maturity index, using the standard
deviation map (Figure 15b), the coefficient of variation (CV) associated with the regression
predictions ranged from 10-11% while that for the structure index ranged from 31 to 46%.
The blue spots seen in the standard deviation maps are overlapping with areas with a
relatively high density of observation points where the kriging variance is small. The
interpolation maps (Figures 17a&18c) showed a higher nematode diversity (H'>2.6) and
number of taxa (GR>25) at locations farther away from the river as compared to lower
diversity (H'<2.0) and number of taxa (GR<15) obtained at spots closest to the river. The
coefficient of variation for the predictions of the Shannon index and Genera richness (Figures.
16b & 17d) ranged from 12-15 and 11-21 %, respectively. In general, there was a high
variability associated with the prediction of the structure index compared to all the other
nematode parameters.
37 | P a g e
Figure 15. Maps of (a) predictions of maturity index from the regression model, (b) standard
deviations of the predictions (c) maturity index values at observation points.
38 | P a g e
Figure 16. Maps of (a) predictions of Structure index from regression model-only, (b)
interpolated residuals of the regression, (c) regression + residuals {regression-kriging} and (d)
regression-kriging standard deviations. This is to illustrate how regression kriging works.
39 | P a g e
Figure 17. Maps of (a) Shannon index predictions from the regression model, (b) standard
deviations associated with predictions and (c) values of Shannon index at sample locations.
40 | P a g e
Figure 18. Maps of (a) Genera richness predictions from the regression model (b) the
residuals of the regression model (c) the combination of the regression model predictions with
the residuals and (d) regression-kriging standard deviations.
Total nematode abundance and abundance of life history groups
The total nematode abundance as predicted by the regression kriging ranged from <1000 to
about 7000 with highest abundances (indicated with pink and yellow pattern) obtained at the
western parts of the study area (Figure 19a) which correspond to locations farther away from
the river. The standard deviation associated with this prediction ranged from 0.24 to 0.36
(Figure 18b). A visual look at the maps of the abundances of c-p2 (Figure 20c) and c-p3-5
(Figure 21a), lower abundances (<500) were predicted at locations closest to the river. The
standard deviations for the c-p2 abundance predictions ranged from 0.26 to 0.44 (Figure 20b)
while that for the c-p3-5 ranged from 0.43 to 0.47 (Figure 21b).
41 | P a g e
Figure 19. Maps of (a) the total nematode abundance predictions from the regression model
(b) standard deviations associated with the prediction and (c) plot of the observed values of
total nematode abundance.
42 | P a g e
Figure 20. Maps of (a) c-p2 nematode abundance predictions from regression-kriging and (b)
regression-kriging standard deviations
Figure 21. Maps of (a) c-p3-5 nematode abundance predictions from the regression model
and (b) standard deviations.
Abundance of feeding groups
The maps of the prediction of the abundances of the various trophic groups and the associated
standard deviations are shown in Figures 22-27. Generally, it can be seen from these maps
that bacterial feeding nematodes had the highest abundance in the study area ranging from
43 | P a g e
<1000 to 5000 (Figure 22a), with most of the highest abundances recorded at points farther
away from the river. For the fungal feeding nematodes, the abundance was between 0 and
350 (Figure 23a). This was far much lower compared with the abundance of the bacterial
feeders. The abundance of the plant feeding nematodes ranged from 200 to about 1600
(Figure 24a), with the highest abundances occurring farther from the river (Western part of
the area). It can be seen from Figures 25a and 26a that the omnivorous and predatory
nematodes had the least abundance in the entire area with the abundances ranging from 0 to
350 (Figure 25a) and 0 to 120 (Figure 26a) for the omnivorous and predatory nematodes,
respectively. For the ectoparasitic nematodes, a higher abundance was recorded at locations
farther from the river (Figure 27a).
Figure 22. Maps of (a) bacterial-feeding nematode abundance predictions from regression-
kriging (i.e. combined prediction from the regression model and the residual) and (b)
regression-kriging standard deviations.
44 | P a g e
Figure 23. Maps of (a) fungal-feeding nematode abundance predictions from regression-
kriging and (b) regression-kriging standard deviations.
Figure 24. Maps of (a) plant-feeding nematode abundance predictions the regression model
and (b) standard deviations.
45 | P a g e
Figure 25. Maps of (a) omnivorous nematode abundance predictions from the regression
model and (b) standard deviations.
Figure 26. Maps of (a) predatory nematode abundance predictions from the regression model
and (b) standard deviations.
46 | P a g e
Figure 27. Maps of (a) ectoparasitic nematode abundance predictions from the regression
models (log scale) and (b) standard deviations.
Discussion
The present study was focused on the spatial distribution of soil nematode communities; their
dependence on and relationship with environmental factors including heavy metal
concentrations. The first hypothesis of this study states that spatial heterogeneity exists in the
distribution of nematode community indices (maturity index, structure index, Shannon index,
Genera richness, abundance of feeding and life history groups) in the contaminated site. To
test this hypothesis, semivariograms were computed for the nematode data obtained from the
area. The results showed that the distributions of the nematode community indices were not
random but exhibited spatial patterns over distances ranging from 80 to 234 m (Figures 6-8).
These patterns of nematode distribution could be attributed to patterns in their life history
characteristics and response to environmental factors. This finding is in agreement with the
findings of Monroy et al. (2012), who found that spatial aggregation was a characteristic
feature of both bacteria and nematode communities. Other studies also showed that at various
sampling scales, nematode populations are spatially patterned (Ettema et al, 1998; Liang et al.
2005; Park, 2012). Even in an agricultural soil which is considered relatively homogenous,
Robertson & Freckman (1995) detected spatial patterns in the distribution of nematode
groups. The semivariograms of the log-transformed abundances of c-p1 (Figure 6b)
nematodes, the root hair feeders (Figure 7g), and the enrichment and channel indices (Figures
8e&f) showed a pure nugget effect indicating that the distribution of these nematode
community indices was either random or that their distribution was not spatially structured at
47 | P a g e
the scales studied. These indices also did not show any significant relationship with the
environmental properties (Table 2). For the c-p1 nematodes, this could be attributed to their
low proportion recorded in the study area (about 10% of the total nematode abundance). The
c-p1 nematodes are bacterivores that increase significantly in abundance in response to
eutrophication events. In the study site, the food web diagram shows signs of eutrophication
in a few of the samples only. The EI is calculated based on the proportion of c-p1 nematode
group (consisting of bacterial feeders) that increase rapidly upon an increase in microbial
activity as a result of organic input (Ferris et al., 2001). The low abundance of the c-p1
nematodes could have resulted in the high variability in the EI (Figure. 4d). The finding of
this study supports the findings of Bert et al. (2009), who also did not find significant
relationships between c-p1 nematode groups and the historical pollution.
To establish the relationships between the nematode community indices and environmental
factors, stepwise regression analysis was performed in line with the second objective of this
study. About 21-50% of the variation in these indices was explained by the soil properties;
although different predictors tended to predict the different indices (Tables 4&5). Soil pH was
almost always a significant predictor in all models. In most of the models were pH and
moisture were significant predictors, they had negative relationship with the nematode
community indices. The heavy metals were selected as predictors in 5 out of the 13 models.
This however does not mean that the other environmental factors that were not selected in
some models were not affecting the nematode distribution but that the strong cross-
correlations between the environmental factors (Figure 11) could have led to the selection of
other cross-correlated factors. Also, Soil nematodes may become stressed by a number of
environmental factors such as pH, salinity, moisture, redox potentials etc. thereby masking the
impacts of other factors such as heavy metal contamination (Sochova et al., 2006). The
Pearson‟s correlation analysis between the nematode community indices and the
environmental factors (Table 2) showed that the total nematode abundance, maturity index,
structure index, Shannon index, genera richness, the abundances of CP3-5, plant feeders,
omnivores and predators were significantly negatively correlated with all the soil properties.
This negative relationship was expected for the heavy metals as toxic effects of these metals
on total nematode abundance are often reported in literature (Korthals et al., 1996; Yeates et
al., 2003; Nagy et al., 2004; Zhang et al., 2007, Shao et al. 2008). Moreover, Van Vliet & De
Goede (2008) reported a significant negative correlation of MI, trophic diversity, proportion
of bacterial and fungal feeders with clay, organic matter, Cd and Zn contents. Correlations of
48 | P a g e
these nematode parameters with pH were however not reported in their study. The negative
correlation with pH is quite surprising as it was contrary to the general expectation that
increase in pH will have a positive impact on these nematode parameters. This observation
could be partly explained by the short range of pH found in the study area (between 3.7 and
4.9), which was also in the acidic range. The observation of the negative relationship could
also be due to the interaction between pH and other environmental variables. Shukurov et al.
(2005) for instance, reported negative correlation between total nematode abundance and pH.
The finding of this study is however in contrast to the findings of De Goede & Bongers
(1994) who reported a significant positive correlation between pH and the proportion of taxa
that were sensitive to environmental disturbance.
Although, the regression models were all significant (p<0.0005) indicating the importance of
the selected predictors in predicting the response variables, a large portion of the variation in
the nematode community indices still remained unexplained. An examination of the
semivariograms of the regression model residuals of the nematode community indices showed
that the residuals were mostly spatially randomly distributed, thereby adding no further spatial
explanation to the model. Thus, the third hypothesis that the selected abiotic properties that
were measured in the Malpiebeemden area were sufficient to explain the spatial structure of
the nematode fauna was partially true. For the few cases where the regression model residuals
showed spatial dependency (structure index, Genera richness, c-p2 nematodes, bacterial and
fungal feeders), the spatial autocorrelation of the residuals indicated that these nematode
parameters were affected by other factors not measured in this study. The other factors could
include other biotic parameters such as plant species identity and intrinsic population
processes such as competition for resources and reproduction (Ettema & Wardle, 2002), but
only if these factors are spatially dependent. For instance, De Deyn et al. (2004) and Viketoft
et al. (2005) reported that plant species identity can influence soil nematode assemblage
composition.
The third objective of this study was to include the relationship between the nematode
community indices and environmental variables (regression models) in mapping the spatial
distribution of the nematode community indices. This was done in a regression kriging
approach. The resulting maps are presented in Figures 15-27. For the maturity and structure
indices, the lowest values were generally obtained at locations closest to the river (Eastern
part of the study area). These locations correspond to where the highest values of the
environmental factors (pH, organic matter, clay, moisture Cd and Zinc contents) were
49 | P a g e
measured (see Figure 10 for the maps of the distribution of soil properties). The MI and SI
increased with increasing distance from the river and the highest values were obtained at the
farthest distances and highest elevations in the Malpiebeemden area. Combining the
correlation with a visual analysis of the map, it is however unclear if these patterns were
causally related to the pollution stress alone. To obtain a better understanding of the effects of
the heavy metal pollution, the samples were divided into 3 groups based on pollution
gradients (unpolluted, moderately polluted and heavily polluted samples). Correlation and
regression analyses were performed on the unpolluted samples which constituted about 70%
of the total number of samples. The results indicated that pH, organic matter and moisture
seemed to be the major factors affecting the nematode assemblage and were in all cases
showing negative relationships. The soil moisture distribution could be a possible explanation
for the trend observed in the distribution of the nematode community indices. Though
nematodes can survive under low oxygen conditions in saturated soils (Poinar, 1983), the
population growth of many taxa may be affected by accumulation of products of anaerobic
metabolism. This might be the case for this study as most of the locations nearest to the river
were in saturated conditions. All nematode groups were least abundant in these areas with
high moisture. Similar effect of high moisture on nematode taxa was reported by Ettema et al.
(1998). Higher MI and SI indicate that soils are in a stable condition and not stressed
(Bongers 1990, Ferris et al, 2001; Vitekoft et al., 2011). Similar trends were observed for the
diversity indices (Figures 17 & 18) and abundances of the nematode groups. Regarding the
second hypothesis of an increased diversity being associated with low levels of disturbance,
the observations in this study seemed to suggest that the nematode diversity is affected by
stress conditions like high moisture content rather than the heavy metal concentrations. This
claim is supported by the grouping of the samples based on levels of pollution and the
resulting foodweb diagram (Figure 5). Interestingly, in all models where moisture was
selected as a significant predictor, it had a negative relationship with the nematode
community indices. To further investigate the effect of moisture, samples were divided into
two moisture content classes: (Group A = those with moisture content at field capacity i.e.
moisture content less or equal to 25% while Group B = those samples with moisture content
at saturation point i.e. moisture content greater than 25%). The food web diagram that resulted
from this classification is shown in Appendix 5. The result indicated that most of the samples
having moisture content at saturation point had SI<60%.
50 | P a g e
The maps obtained in this study revealed similar abundances in both c-p2 (Figure 20) and c-
p3-5 (Figure 21) nematode groups at some locations even though the c-p3-5 group are
expected to be relatively more sensitive to most forms of stress (Zhang et al, 2007; Shao et
al., 2008). A possible explanation for this could be due to the dominance of c-p3 bacterial
feeders belonging to the family Teratocephalidae and Prismatolaimidae in the c-p3-5 group at
such locations and not the abundance of the stronger K-selected species (c-p4 and 5). Further
support is given to this explanation as the abundance of omnivores and predators (mainly
classified as c-p4 or c-p5) were relatively low (0-350 and 0-120, respectively, as seen in
Figures 25 and 26). For the feeding group distributions, bacterial feeding nematodes were the
most abundant. This corroborates the results of several other studies in European grasslands
(De Goede & Bongers, 1994; Wasilewska, 1994; Hanel, 1996; Ekschmitt et al., 2001; Zolda,
2006). Fungal feeders made up a smaller proportion compared to plant feeders similar to the
reports of Ekschmitt et al. (2001) and Hanel (2003). Decomposition pathways in grasslands
are mostly mediated by bacteria, and bacterial feeding nematodes are to be expected as
dominant secondary decomposers (McSorley and Frederick, 2000; Rues, 2003).
The standard deviation maps give some level of confidence on the accuracy of the regression
kriging maps. The highest coefficient of variation was obtained for the prediction of the
structure index (46%). In general, most of these maps showed that the coefficients of variation
associated with the predictions of nematode parameters at unsampled locations are within the
acceptable range for field experiments (Patel et al., 2001). For instance, for the maturity
index, the coefficient of variation ranged between 9 and 11 %.
Conclusions
This study was designed with the aim to evaluate the spatial structure of nematode community
indices, and examine the relationships between nematode community indices and
environmental factors including effects of heavy metal pollution. Also, another objective was
to use these relationships in mapping the spatial distribution of the nematode community
indices in a regression kriging approach. The results of this study indicated that spatial
heterogeneity exists in the distribution of nematodes in the Malpiebeemden study area. The
patterns obtained from the semivariograms of the observations and the interpolation maps of
the regression models support the claim that the distribution of nematodes in this study site
was not random but exhibited some spatial patterns. Despite its exploratory nature, this study
offers some insights into the relationships between nematode community indices and
51 | P a g e
environmental factors. Geostatistical models relating the nematode community indices with
environmental factors were developed and used in producing interpolation maps at unsampled
locations. Regression kriging approach seemed to be a valid technique in mapping spatial
distributions of nematode community indices only in a few cases where the residuals from the
regression models exhibited spatial dependency.
Although this study could not adequately explain any direct influence of heavy metals (Cd
and Zn) on the nematode faunal assemblage and distribution, studies such as this that
measures soil properties including heavy metals in polluted sites are a holistic approach on
factors that affect nematode faunal structure rather than reporting data on heavy metals alone.
It is also possible that the effects of the heavy metal contamination were not evident as only a
few samples were heavily polluted (heavy metal concentrations above the Dutch Intervention
value for Zn concentration in soils- 720 mg/kg).
There is need to validate the performance of the interpolated maps using interpolation and
validation based on random sample extraction technique. In fact, this was an objective for this
study but was not executed due to time constraints. The validation could be obtained by
removing two sample points from each stratum (for the eight strata=16). Regression models
could then be built and maps developed based on the remaining 64 sampling points.
Repeating this e.g. five times gives a total of 80 randomly distributed validation points. True
prediction accuracy can then be evaluated by comparing estimated values with actual
observations at validation points to assess systematic error, calculated as mean prediction
error; and accuracy of prediction, calculated as root mean square prediction error. The set-up
of the sampling design used in this study completely suits such a validation. There is also the
need for further investigation on other forms of models that could relate nematode community
indices with environmental factors. Suggestion could be the use of random forests or other
forms of non-linear models to see if such models could better explain the spatial variations in
these nematode community indices.
Acknowledgements
I would like to thank the European commission for the Erasmus Mundus grant,
EUMAINE Consortium, Soil Quality Research Group of Wageningen University (The
Netherlands) for accepting me to conduct my thesis research and Mr. Tamas Salanki for
helping with the field sampling and other technical assistance. Many thanks to my promoters:
Professors Ron de Goede and Gerard Heuvelink for their guidance, help, valuable advice and
52 | P a g e
for patiently supervising me throughout this work. Finally, I am grateful to my family for the
relentless prayers and unending support.
References
Andrassy, I. (1984). Klasse Nematoda (Ordungen Monhysterida, Desmoscolecida,
Areolaimida, Chromadorida, Rhabditida). Akademie-Verlag, Berlin, 288 pp.
Bert, W., Manhout, J., Van Colen, C., Borgonie, G. & Decraemer, W. (2009). Nematode
assemblages in a nature reserve with historical pollution. Belgian Journal of Zoology
139, 1-15.
Bleeker, E. A. J. & Van Gestel, C. A. M. (2007). Effects of spatial and temporal variation in
metal availability on earthworms in floodplain soils of the river Dommel, The
Netherlands. Environmental Pollution 148, 824-832.
Bongers, T. & Bongers, M. (1998). Functional diversity of nematodes. Applied Soil Ecology
10, 239–251.
Bongers, T. & Ferris, H. (1999). Nematode community structure as a bioindicator in
environmental monitoring. Trends in Ecology & Evolution 14, 224-228.
Bongers, T. (1988). De Nematoden van Nederland. KNNV Bibliotheek Uitgeverij Pirola,
Schoorl, 408pp.
Bongers, T. (1990). The maturity index: an ecological measure of environmental disturbance
based on nematode species composition. Oecologia 83, 14-19.
Burnham, K. P., & Anderson D. R. (2002). Model selection and multimodel inference: a
practical information-theoretic approach. Springer, New York.
De Bakker, H. ( 1979). Major soils and soil regions in the Netherlands. Junk , The Hague and
PUDOC, Wageningen, published. 203pp.
De Deyn, G.B., Raaijmakers, C.E., van Ruijven, J., Berendse, F. & Van der Putten, W.H.
(2004). Plant species identity and diversity effects on different trophic levels of
nematodes in the soil food web. Oikos 106, 576–586
De Goede, R.G.M. & Bongers, T. (1994). Nematode community structure in relation to soil
and vegetation characteristics. Applied Soil Ecology 1, 29-44.
De Gruijter, J., Brus, D., Bierkens, M. & Knotters, M. (2006). Sampling for Natural Resource
Monitoring. Springer, Berlin.
Decraemer, W. & Hunt, D. (2006). Structure and classification. In: Perry, R. & Moens, M.
(Eds). Plant Nematology. Wallingford, UK, CABI publishing. pp. 1-33.
Ekschmitt, K., Bakonyi, G., Bongers, M., Bongers, T., Bostrom, S., Dogan, H., Harrison, A.,
Nagy, P., O‟Donnel, A.G., Papatheodorou, E.M., Sohlenius, B., Stamou, G.P. &
53 | P a g e
Wolters, V. (2001). Nematode community structure as indicator of soil functioning in
European grassland soils. European Journal of Soil Biology 37, 263–268.
Ettema, C. H. & Wardle, D. A. (2002). Spatial soil ecology. Trends in Ecology & Evolution
17, 177-183.
Ettema, C. H., Coleman, D. C., Vellidis, G., Lowrance, R. & Rathbun, S. L. (1998).
Spatiotemporal distributions of bacterivorous nematodes and soil resources in a
restored riparian wetland. Ecology 79, 2721-2734.
Ettema, C. H., Rathbun, S. L. & Coleman, D. C. (2000). On spatiotemporal patchiness and the
coexistence of five species of Chronogaster (Nematoda: Chronogasteridae) in a
riparian wetland. Oecologia 125, 444-452.
Ferris, H. & Bongers, T. (2006). Nematode indicators of organic enrichment. Journal of
Nematology 38, 3-12.
Ferris, H., & Matute, M. M. (2003). Structural and functional succession in the nematode
fauna of a soil food web. Applied Soil Ecology 23, 93–110.
Ferris, H., Bongers, T., & De Goede, R.G.M. (2001). A framework for soil food web
diagnostics: extension of the nematode faunal analysis concept. Applied Soil Ecology
18, 13–29.
Ferris, H., Venette, R.C. & Scow, K.M. (2004). Soil management to enhance bacterivore and
fungivore nematode populations and their nitrogen mineralisation function. Applied
Soil Ecology 25, 19–35.
Georgieva, S. S., Mcgrath, S. P., Hooper, D. J. & Chambers, B. S. (2002). Nematode
communities under stress: the long-term effects of heavy metals in soil treated with
sewage sludge. Applied Soil Ecology 20, 27-42.
Gorres, J. H., Dichiaro, M. J., Lyons, J. B. & Amador, J. A. (1998). Spatial and temporal
patterns of soil biological activity in a forest and an old field. Soil Biology and
Biochemistry 30, 219-230.
Hanel, L. (1996). Composition and seasonal changes of soil nematode community in a south
Bohemian meadow. Acta Societatis Zoologicae Bohemicae 60, 103–114.
Hanel, L. (2003). Recovery of soil nematode populations from cropping stress by natural
secondary succession to meadow land. Applied Soil Ecology 22, 255-270.
Heininger, P., Höss, S., Claus, E., Pelzer, J. & Traunspurger, W. (2007). Nematode
communities in contaminated river sediments. Environmental Pollution 146, 64–76
Hengl, T., Heuvelink, G.B.M., & Rossiter, D.G. (2007). Hengl, T., Heuvelink, G.B.M., and
Rossiter, D.G. (2007). About regression kriging: From equations to case studies.
Computers and Geosciences 33, 1301-1315.
54 | P a g e
Horner-Devine, M. C., Silver, J. M., Leibold, M. A., Bohannan, B. J. M., Colwell, R. K.,
Fuhrman, J. A., Green, J. L., Kuske, C. R., Martiny, J. B. H. & Muyzer, G. (2007). A
comparison of taxon co-occurrence patterns for macro-and microorganisms. Ecology
88, 1345-1353.
Hoss, S. & Traunspurger, W. (2003). Nematodes. In: Markert, B.A., Breure, A.M. &
Zechmeister, H.G. (Eds.), Bioindicators and Biomonitors. Elsevier, Oxford, pp. 529–
554.
Jairajpuri, M. S. & Ahmad, W. (1992). Dorylaimida. Free-living, Predaceous and Plant-
parasitic Nematodes. E.J. Brill, Leiden, 458 pp.
Kardol, P., Bezemer, T., Van Der Wal, A. & Van Der Putten, W. (2005). Successional
trajectories of soil nematode and plant communities in a chronosequence of ex-arable
lands. Biological Conservation 126, 317-327.
Korenko, V. & Schmidt, C. (2006). Effects of agricultural practices in the rice crop system on
nematode communities in Uruguay. Nematologia Mediterranea 34, 151-159.
Korthals, G. W., Ende, A., Megen, H., Lexmond, T. M., Kammenga, J. E. & Bongers, T.
(1996). Short-term effects of cadmium, copper, nickel and zinc on soil nematodes
from different feeding and life-history strategy groups. Applied Soil Ecology 4, 107-
117.
Lambshead, P.J.D. (2004). Marine Nematode Biodiversity. In: Chen, Z.X., Chen, S.Y. &
Dickson, D.W. (Eds). Nematology: Advances and Perspectives. Wallingford, CABI
Publishing, pp.439-468.
Lavelle, P. & Spain, A. (2001). Soil ecology. Springer, pp. 203-213.
Liang, W., Jiang, Y., Li, Q., Liu, Y. & Wen, D. (2005). Spatial distribution of bacterivorous
nematodes in a Chinese Ecosystem Research Network (CERN) site. Ecological
Research 20, 481-486.
Martiny, J. B. H., Bohannan, B. J. M., Brown, J. H., Colwell, R. K., Fuhrman, J. A., Green, J.
L., Horner-Devine, M. C., Kane, M., Krumins, J. A. & Kuske, C. R. (2006). Microbial
biogeography: putting microorganisms on the map. Nature Reviews Microbiology 4,
102-112.
McSorley, R. & Frederick, J.J. (2000). Short-term effects of cattle grazing on nematode
communities in Florida pastures. Nematropica 30, 211–221.
Monroy, F., Van Der Putten, W. H., Yergeau, E., Mortimer, S. R., Duyts, H. & Bezemer, T.
M. (2012). Community patterns of soil bacteria and nematodes in relation to
geographic distance. Soil Biology and Biochemistry 45, 1-7
55 | P a g e
Mulder, C., Schoutena, A.J., Hund-Rinkeb, K., & Breurea, A.M. (2005). The use of
nematodes in ecological soil classification and assessment concepts. Ecotoxicology
and Environmental Safety 62, 278–289.
Nagy, P., Bakonyi, G., Bongers, T., Kadar, I., Fabian, M. & Kiss, I. (2004). Effects of
microelements on soil nematode assemblages seven years after contaminating an
agricultural field. Science of the total environment 320, 131-143.
Neher, D. A. (2001). Role of nematodes in soil health and their use as indicators. Journal of
Nematology 33, 161-168.
Neher, D.A. & Darby, B.J. (2009). General community indices that can be used for analysis
of nematode assemblages. In: Nematodes as Environmental Indicators. Wallingford:
CAB International, pp. 110-123.
Odeh, I., McBratney, A. & Chittleborough, D. (1994). Spatial prediction of soil properties
from landform attributes derived from a digital elevation model. Geoderma 63, 197–
214.
Odeh, I., McBratney, A. & Chittleborough, D. (1995). Further results on prediction of soil
properties from terrain attributes: heterotopic cokriging and regression-kriging.
Geoderma 67, 215– 226.
Oostenbrink, M. (1960). Estimating nematode populations by some selected methods. In:
Nematology (J.N. Sasser, W.R. Jenkins, eds.), pp. 85-102. The University of North
Carolina Press, Chapel Hill.
Park, B. Y., Lee, J. K., Ro, H. M. & Kim, Y. H. (2011). Effects of heavy metal contamination
from an abandoned mine on nematode community structure as an indicator of soil
ecosystem health. Applied Soil Ecology 51, 17-24.
Park, J. J. (2012). Spatial pattern analysis of entomopathogenic and other free‐living
nematodes at landscape scales. Entomological Research 42, 104-110.
Patel, J.K., Patel, N.M. & Shiyani, R.L. (2001). Coefficient of variation in field experiments
and yardstick thereof – An empirical study. Current Science 81, 1163-1164.
Poinar, G. O., Jr. (1983). The natural history of nematodes. Prentice-Hall, Englewood Cliffs,
New Jersey, USA.
Postma, J.F. (1995). Adaptation to metals in the midge Chironomus riparius. PhD thesis,
University of Amsterdam, Amsterdam, The Netherlands.
R Development Core Team (2013). R: A language and environment for statistical computing.
R Foundation for Statistical Computing, Vienna, Austria.
Robertson, G. P., & Freckman, D. W. (1995). The spatial distribution of nematode trophic
groups across a cultivated ecosystem. Ecology 76,1425–1432.
56 | P a g e
Ruess, L., (2003). Nematode soil faunal analysis of decomposition pathways in different
ecosystems. Nematology 5, 179–181.
Šalamún, P., Renčo, M., Kucanová, E., Brázová, T., Papajová, I., Miklisová, D. & Hanzelová,
V. (2012). Nematodes as bioindicators of soil degradation due to heavy metals.
Ecotoxicology 21, 2319-2330.
Sánchez-Moreno, S. & Navas, A. (2007). Nematode diversity and food web condition in
heavy metal polluted soils in a river basin in southern Spain. European Journal of Soil
Biology 43, 166-179.
Schratzberger, M., Bolam, S., Whomersley, P. & Warr, K. (2006) Differential response of
nematode colonist communities to the intertidal placement of dredged material.
Journal of Experimental Marine Biology and Ecology 334, 244–255.
Shao, Y., Zhang, W., Shen, J., Zhou, L., Xia, H., Shu, W., Ferris, H. & Fu, S. (2008).
Nematodes as indicators of soil recovery in tailings of a lead/zinc mine. Soil Biology
& Biochemistry 40, 2040–2046.
Shukurov, N., Pen-Mouratov, S. & Steinberger, Y. (2005). The impact of Almalyk industrial
complex on soil chemical and biological properties. Environmental Pollution 136, 331
– 340.
Siddiqi, M.R. (1986). Tylenchida. Parasites of Plants and Insects. Commonwealth
Agricultural Bureaux, Farnham Royal, Slough, 645 pp.
Sochova, I., Hofman, J. & Holoubek, I. (2006). Using nematodes in soil ecotoxicology.
Environment International 32, 374–383.
Stoyan, H., De-Polli, H., Bohm, S., Robertson, G. P. & Paul, E. A. (2000). Spatial
heterogeneity of soil respiration and related properties at the plant scale. Plant and
Soil 222, 203-214.
Tenuta, M. & Ferris, H. (2004). Relationship between nematode life-history classification and
sensitivity to stressors: ionic and osmotic effects of nitrogenous solutions. Journal of
Nematology 36, 85–94.
Van Vliet, P.C.J. & De Goede, R.G.M. (2008). Nematode-based risk assessment of mixture
toxicity in a moderately polluted river floodplain in The Netherlands. Science of the
Total Environment 406, 449-454.
Venables W.N. and Ripley B.D. (2002). Modern Applied Statistics with S. Springer-Verlag,
New York, 4th edition.
Viketoft, M., Palmborg, C., Sohlenius, B., Huss-Danell, K. & Bengtsson, J. (2005) Plant
species effects on soil nematode communities in experimental grasslands. Applied Soil
Ecology 30, 90–103.
57 | P a g e
Viketoft, M., Sohlenius, B., Boström, S., Palmborg, C., Bengtsson, J., Berg, M.P. & Huss-
Danell, K. (2011). Temporal dynamics of soil nematode communities in a grassland
plant diversity experiment. Soil Biology & Biochemistry 43, 1063-1070.
Vos, M. & Velicer, G. J. (2008). Isolation by Distance in the Spore-Forming Soil Bacterium
Myxococcus xanthus. Current Biology 18, 386-391.
Walvoort, D. J. J., Brus, D. J., & De Gruijter, J. J. (2010). An R package for spatial coverage
sampling and random sampling from compact geographical strata by k-means.
Computers & Geosciences 36, 1261-1267.
Wasilewska, L. (1994). The effect of age of meadows on succession and diversity in soil
nematode communities. Pedobiologia 38, 1–11.
Webster, R. & Oliver, M. A. (2007). Geostatistics for Environmental Scientists, Second
Edition. John Wiley & Sons, Chichester, 307pp.
Yeates, G.W., Bongers, T., De Goede, R.G.M., Freckman, D.W. & Georgieva, S.S. (1993).
Feeding habits in soil nematode families and genera – an outline for soil ecologists.
Journal of Nematology 25, 315–331.
Yeates, G.W., Ferris, H., Moens, T. & Van Der Putten, W.H. (2009). The role of nematodes
in ecosystems. In: Nematodes as Environmental Indicators. Wallingford: CAB
International, pp. 1-44.
Yeates, G.W., Percival, H.J. & Parshotam, A. (2003). Soil nematode responses to year-to-year
variation of low levels of heavy metals. Australian Journal of Soil Research 41, 613–
625.
Zeileis, A.; Kleiber, C. & Jackman, S. (2008). Regresssion models for count data in R.
Journal of Statistical Software 27, 1-25.
Zhang, W., Wang, X., Li, Q., Jiang, Y. & Liang, W. (2007). Soil nematode responses to
heavy metal stress. Helminthologia 44, 87-91.
Zolda, P. (2006). Nematode communities of grazed and ungrazed semi-natural steppe
grasslands in Eastern Austria. Pedobiologia 50, 11-22.
58 | P a g e
Appendix 1. Sample R script for ordinary kriging predictions of soil physical properties
from 2008 measurements
# Extrapolation of soil properties from 2008 measurements
rm(list = ls()) # clean memory
graphics.off() # close graphic windows
# load libraries:
library(foreign)
library(RColorBrewer)
library(rgdal) library(sp)
library(maptools)
library(gstat)
# read data file:
malpie = read.table("malpie_data.txt", header = TRUE)
class(malpie)
names(malpie)
# translation to get locations at the right spot:
malpie$Coord.X = malpie$Coord.X + 255 malpie$Coord.Y = malpie$Coord.Y + 95
# make spatial:
coordinates(malpie)=~Coord.X+Coord.Y
# read boundary of study area:
border = readShapePoly("border.shp")
# plot observations:
# windows(width=5, height=8)
spplot(malpie, zcol="Cdtot.s", scales=list(draw=T), cex=1.4, main="Total Cadmium concentration topsoil",
col.regions=brewer.pal(4, "Oranges"),
sp.layout=list("sp.polygons", border))
#log transformation to remove skewness
malpie$logCdtot.s = log(malpie$Cdtot.s + 1, 10)
spplot(malpie, zcol="logCdtot.s", scales=list(draw=T), cex=1.4,
main="10log of total Cadmium concentration topsoil",
col.regions=brewer.pal(4, "Oranges"),
sp.layout=list("sp.polygons", border))
hist(malpie$logCdtot.s)
hist(malpie$Cdtot.s)
# read and plot DEM:
grid = readGDAL("ahn_malpie.asc")
grid$elev = grid$band1
grid$band1 = NULL
windows(width = 6, height = 6)
spplot(grid, zcol = "elev", col.regions = bpy.colors(), scales=list(draw=T))
# read and plot distance to river:
temp = readGDAL("dist_to_river.asc")
grid$dist = temp$band1 rm(temp)
windows(width = 6, height = 6)
spplot(grid, zcol = "dist", col.regions = bpy.colors(), scales=list(draw=T))
59 | P a g e
# read and display mask:
temp = readGDAL("mask.asc")
grid$mask = temp$band1
rm(temp)
windows(width = 6, height = 6)
spplot(grid, zcol = "mask", col.regions = bpy.colors(), scales=list(draw=T))
names(grid)
# read and plot rasterized soil map: temp = readGDAL("bod50rast.asc")
grid$soil = temp$band1
rm(temp)
table(grid$soil)
class(grid$soil)
windows(width = 6, height = 6)
spplot(grid, zcol = "soil", col.regions=brewer.pal(6, "Set2"), scales=list(draw=T),
at = seq(from = -0.5, to = 5.5, by = 1))
soilrastdbf = read.dbf("bod50rast.dbf")
soilrastdbf
# zoom in on mask:
zgrid = grid[33:198,99:191]
# masking all maps:
zgrid$elev = 0*zgrid$mask + zgrid$elev
zgrid$soil = 0*zgrid$mask + zgrid$soil
zgrid$dist = 0*zgrid$mask + zgrid$dist
# plotting zoomed maps with data points:
pts = list("sp.points", malpie, pch = 3, lwd = 1.5, col = "grey")
# windows(width = 5, height = 7) spplot(zgrid, zcol = "elev", col.regions = bpy.colors(),
scales=list(draw=T), sp.layout = list(pts),
xlim = c(159275,159740), ylim = c(368440,369275))
# windows(width = 5, height = 7)
spplot(zgrid, zcol = "dist", col.regions = bpy.colors(),
scales=list(draw=T), sp.layout = list(pts),
xlim = c(159275,159740), ylim = c(368440,369275))
table(zgrid$soil)
# windows(width = 5, height = 7)
spplot(zgrid, zcol = "soil", col.regions = brewer.pal(4, "Set2"), scales=list(draw=T), sp.layout = list(pts),
at = seq(from = -0.5, to = 3.5, by = 1),
xlim = c(159275,159740), ylim = c(368440,369275))
# image(zgrid,4)
# points(159400,368800)
# extending dataset with spatial properties:
temp <- over(geometry(malpie),grid)
malpie$elev = temp$elev
malpie$dist = temp$dist
malpie$soil = temp$soil
names(malpie)
table(malpie$soil)
60 | P a g e
# relationships with elev and dist:
windows(width = 6, height = 6)
plot(malpie$elev, log(malpie$Cdtot.s)+1)
windows(width = 6, height = 6)
plot(malpie$dist, log(malpie$Cdtot.s)+1)
# geostatistics for Cdtot.s :
# define gstat object and compute variogram:
# malpie = subset(malpie, !is.na(malpie$logCdtot.s)) # needed if missing values
g = gstat(id = c("logCdtot.s"), formula = logCdtot.s~1, data = malpie)
vg = variogram(g, cutoff=500)
vg = variogram(g, boundaries = c(30,70,120,200,300,450))
windows(width = 8, height = 6)
plot(vg, plot.numbers = TRUE)
# choose initial variogram model and plot:
vgm = vgm(nugget=0, psill=0.5, range=140, model="Mat", kappa = 0.9)
plot(vg, vgm)
# fit variogram model:
vgm = fit.variogram(vg, vgm)
windows(width = 8, height = 6)
plot(vg, vgm, main = "Variogram 10logCdtot.s")
vgm
# ordinary kriging:
malpie.krig = krige(logCdtot.s~1, malpie, newdata = zgrid, vgm)
names(malpie.krig)[1] = "ok.pred"
names(malpie.krig)[2] = "ok.var"
# plot ordinary kriging map:
min(malpie.krig$ok.pred, na.rm = TRUE)
max(malpie.krig$ok.pred, na.rm = TRUE)
windows(width = 6, height = 9)
spplot(malpie.krig, zcol = "ok.pred", col.regions = bpy.colors(),
main = " Ordinary kriging logCdtot.s",
scales=list(draw=T), at = seq(from = -0.02, to = 2.2, by = 0.1))
savePlot(filename="logCdtot_okpred",type="png")
savePlot(filename="logCdtot_okpred",type="pdf")
malpie.krig$Cdtot.mean <- 10^(malpie.krig$ok.pred + 0.5*malpie.krig$ok.var) - 1
writeAsciiGrid(malpie.krig, attr="Cdtot.mean", "Cdtot.asc") writeAsciiGrid(malpie.krig, "logCdtot_okpred.asc", "ok.pred")
#Back transformation of the log transformed values
malpie.krig$Cdtot.median <- 10^(malpie.krig$ok.pred) - 1
malpie.krig$Cdtot.mean <- 10^(malpie.krig$ok.pred + 0.5*malpie.krig$ok.var) - 1
spplot(malpie.krig, zcol = "Cdtot.median", col.regions = bpy.colors())
spplot(malpie.krig, zcol = c("Cdtot.median","Cdtot.mean"), col.regions = bpy.colors())
names(malpie.krig)
# ordinary kriging to point locations and back transformations to mean and median:
cdtots = read.table("israelocs.csv", sep=",",header = TRUE) cdtots
class(cdtots)
cdtots.sp = cdtots
coordinates(cdtots.sp)=~Coord.X+Coord.Y
cdtots.krig = krige(logCdtot.s~1, malpie, newdata = cdtots.sp, vgm)
cdtots.krig$var1.sd = sqrt(cdtots.krig$var1.var)
61 | P a g e
cdtots.krig$Cdtot.median <- 10^(cdtots.krig$var1.pred) - 1
cdtots.krig$Cdtot.mean <- 10^(cdtots.krig$var1.pred + 0.5*cdtots.krig$var1.var) - 1
cdtots.krig
write.table(cdtots.krig, file = "cdtots.csv", sep = ",", row.names=F)
Appendix 2. Sample R script for geostatistical computation of semivariograms of
observed data (Total nematode abundance)
# plot of observations and semivariograms of nematode data
rm(list = ls()) # clean memory
graphics.off() # close graphic windows
# load libraries:
library(foreign)
library(RColorBrewer)
library(rgdal)
library(sp) library(maptools)
library(gstat)
# read data file:
malpie = read.table("malpie_data.txt", header = TRUE)
class(malpie)
names(malpie)
# make spatial:
coordinates(malpie)=~Coord.X+Coord.Y
# read boundary of study area:
border = readShapePoly("border.shp")
#log transformation of nematode counts to remove skewness
malpie$logTNEM=log(malpie$TNEM+1)
# plot observations for:
# windows(width=5, height=8)
spplot(malpie, zcol="TNEM", scales=list(draw=T), cex=1.4, cuts=c(0,500,1000,2000,5000,10000),
main="Total nematode abundance",
col.regions=brewer.pal(4, "Oranges"), sp.layout=list("sp.polygons", border))
savePlot(filename="obsptsTNEM", type="png")
#manipulate variogram plot characteristics
trellis.par.get("fontsize")->fontsize
fontsize$default<-16
fontsize$points<-16
fontsize$text<-20
trellis.par.set("fontsize",fontsize)
trellis.par.get("fontsize")
# geostatistics for log TNEM :
# define gstat object and compute variogram:
# malpie = subset(malpie, !is.na(malpie$logTNEM)) # needed if missing values
g = gstat(id = c("logTNEM"), formula = logTNEM~1, data = malpie)
vg = variogram(g, cutoff=500)
vg = variogram(g, boundaries = c(30,70,120,200,300,450))
windows(width = 8, height = 6)
62 | P a g e
plot(vg, plot.numbers = TRUE)
# choose initial variogram model and plot:
vgm = vgm(nugget=0.6, psill=0.5, range=200, model="Sph")
plot(vg, vgm)
# fit variogram model:
vgm = fit.variogram(vg, vgm)
windows(width = 8, height = 6) plot(vg, vgm, ylim=c(0,1.4), pch=19, lwd=2)
vgm
savePlot(filename="semivarTNEM", type="png")
Appendix 3. R script for multiple linear regression analyses
#Modelling relationships using multiple linear regression
#Modelling relationship between MI and soil properties in a stepwise selection manner
rm(list = ls()) # clean memory
graphics.off() # close graphic windows
#Read the txt table containing the dataset
malpie = read.table("malpie_data.txt", header = TRUE)
dim(malpie)
names(malpie)
summary(malpie)
library(MASS)
attach(malpie)
#regression model of MI using the explanatory variables (only total heavy metal conc) modell1MI<-lm(MI~Clay+OrgMat+Moisture+Cdtot+Zntots+pH)
step <- stepAIC(modell1MI, direction="backward")
step$anova
#using the final model (best model) after the stepwise regression
fit1MI<- lm(MI ~ Cdtot + Zntots + pH)
summary(fit1MI)
#Assessing Goodness of fit using 4 plots
par(mfrow=c(2,2))
plot(fit1MI)
hist(fit1MI$resid)
limits = c(min(MI, fitted(fit1MI)), max(MI, fitted(fit1MI))) windows(width = 5, height = 5)
plot(MI, fitted(fit1MI), xlab = "Observed", ylab = "Fitted",
xlim = limits, ylim = limits, col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2, main = "Maturity Index")
abline(a=0, b=1, col = "turquoise3", lwd = 2)
plot(fitted(fit1MI), residuals(fit1MI), xlab = "Fitted",
ylab = "Residuals", col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2)
abline(h=0)
#Modelling relationship between SI and soil properties in a stepwise selection manner
modell1SI<-lm(SI~Clay+OrgMat+Moisture+Cdtot+Zntots+pH)
step <- stepAIC(modell1SI, direction="backward")
step$anova
63 | P a g e
#using the final model (best predictors) after the stepwise regression
attach(malpie)
fit1SI<- lm(SI ~ OrgMat+Moisture + pH)
summary(fit1SI)
#Assessing Goodness of fit using 4 plots
par(mfrow=c(2,2))
plot(fit1SI)
hist(fit1SI$resid)
limits = c(min(SI, fitted(fit1SI)), max(SI, fitted(fit1SI)))
windows(width = 5, height = 5)
plot(SI, fitted(fit1SI), xlab = "Observed", ylab = "Fitted",
xlim = limits, ylim = limits, col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2, main = "Structure Index")
abline(a=0, b=1, col = "turquoise3", lwd = 2)
plot(fitted(fit1SI), residuals(fit1SI), xlab = "Fitted",
ylab = "Residuals", col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2)
abline(h=0)
#regression model of SI using the explanatory variables (only available heavy metal conc)
modell2SI<-lm(SI~Clay+OrgMat+Moisture+Cdav+Znav+pH)
step <- stepAIC(modell2SI, direction="backward")
step$anova
#using the final model (best model) after the stepwise regression
fit2SI<- lm(SI ~ OrgMat+Moisture+pH)
summary(fit2SI)
#Assessing Goodness of fit using 4 plots
par(mfrow=c(2,2)) plot(fit2SI)
hist(fit2SI$resid)
limits = c(min(SI, fitted(fit1SI)), max(SI, fitted(fit1SI)))
windows(width = 5, height = 5)
plot(MI, fitted(fit1SI), xlab = "Observed", ylab = "Fitted",
xlim = limits, ylim = limits, col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2, main = "Structure Index")
abline(a=0, b=1, col = "turquoise3", lwd = 2)
plot(fitted(fit1SI), residuals(fit1MI), xlab = "Fitted",
ylab = "Residuals", col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2)
abline(h=0)
#Modelling relationship between HI and soil properties in a stepwise selection manner
modell1HI<-lm(HI~Clay+OrgMat+Moisture+pH+Cdtot+Zntots)
step <- stepAIC(modell1HI, direction="backward")
step$anova
#using the final model (best predictors) after the stepwise ##regression
attach(malpie)
fit1HI<- lm(HI ~ Moisture + pH) summary(fit1HI)
#assessing the fitness of the model
par(mfrow=c(2,2))
plot(fit1HI)
hist(fit1HI$resid)
64 | P a g e
limits = c(min(HI, fitted(fit1HI)), max(HI, fitted(fit1HI)))
windows(width = 5, height = 5)
plot(HI, fitted(fit1HI), xlab = "Observed", ylab = "Fitted",
xlim = limits, ylim = limits, col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2, main = "Shannon Index")
abline(a=0, b=1, col = "turquoise3", lwd = 2)
plot(fitted(fit1HI), residuals(fit1HI), xlab = "Fitted",
ylab = "Residuals", col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2) abline(h=0)
#Modelling relationship between GR and soil properties in a stepwise selection manner
modell1GR<-lm(GR~Clay+OrgMat+Moisture+Cdtot+Zntots+pH)
step <- stepAIC(modell1GR, direction="backward")
step$anova
#using the final model (best predictors) after the stepwise regression
fit1GR<- lm(GR ~ Clay + Zntots + pH)
summary(fit1GR) #assessing the fitness of the model
par(mfrow=c(2,2))
plot(fit1GR)
hist(fit1GR$resid)
limits = c(min(GR, fitted(fit1GR)), max(GR, fitted(fit1GR)))
windows(width = 5, height = 5)
plot(GR, fitted(fit1GR), xlab = "Observed", ylab = "Fitted",
xlim = limits, ylim = limits, col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2, main = "Genera Richness")
abline(a=0, b=1, col = "turquoise3", lwd = 2) plot(fitted(fit1GR), residuals(fit1GR), xlab = "Fitted",
ylab = "Residuals", col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2)
abline(h=0)
Appendix 4. R script for negative binomial regression analyses
#poisson regression modelling(for count data)
rm(list = ls()) # clean memory
graphics.off() # close graphic windows
#Read the txt table containing the dataset
malpie = read.table("malpie_data.txt", header = TRUE)
dim(malpie)
names(malpie)
summary(malpie)
#Total Nematode abundance (TNEM)
attach(malpie)
modellTNEM <- glm(TNEM~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie,
family=poisson())
step <- stepAIC(modellTNEM, direction="backward")
step$anova ##All predictors were found to be important
attach(malpie)
modellTNEM <- glm(TNEM~pH + OrgMat +Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
65 | P a g e
summary(modellTNEM)
##Negative binomial regression recommended for overdispersed data
attach(malpie)
modellTNEM <- glm.nb(TNEM~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie)
step <- stepAIC(modellTNEM, direction="backward")
step$anova
#Negative binomial modelling for best predictors
attach(malpie) modellTNEM <- glm.nb(TNEM~pH + OrgMat + Clay, data=malpie)
summary(modellTNEM)
##Assesing of goodness of fit
limits = c(min(TNEM, fitted(modellTNEM)), max(TNEM, fitted(modellTNEM)))
windows(width = 5, height = 5)
plot(TNEM, fitted(modellTNEM), xlab = "Observed", ylab = "Fitted",
xlim = limits, ylim = limits, col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2, main = "Total nematode count")
abline(a=0, b=1, col = "turquoise3", lwd = 2)
plot(fitted(modellTNEM), residuals(modellTNEM), xlab = "Fitted",
ylab = "Residuals", col = "green", pch = 21, bg = "blue", cex = 1.2, cex.lab = 1.2, cex.axis = 1.2)
abline(h=0)
#CP2
attach(malpie)
modellCP2 <- glm(CP2~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
step <- stepAIC(modellCP2, direction="backward")
step$anova
## All predictors were found to be important
attach(malpie) modellCP2 <- glm(CP2~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
summary(modellCP2)
##Negative binomial regression also recommended for overdispersed data
attach(malpie)
modellCP2 <- glm.nb(CP2~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie)
step <- stepAIC(modellCP2, direction="backward")
step$anova
## Negative binomial modelling for best predictors
##Final Model: CP2 ~ pH + Moisture + Clay
attach(malpie) modellCP2 <- glm.nb(CP2~pH+Moisture+Clay, data=malpie)
summary(modellCP2)
##Assesing of goodness of fit
limits = c(min(CP2, fitted(modellCP2)), max(CP2, fitted(modellCP2)))
windows(width = 5, height = 5)
plot(CP2, fitted(modellCP2), xlab = "Observed", ylab = "Fitted",
xlim = limits, ylim = limits, col = "green", pch = 21, bg = "blue",
cex = 1.2, cex.lab = 1.2, cex.axis = 1.2, main = "CP2 nematode number")
abline(a=0, b=1, col = "turquoise3", lwd = 2)
plot(fitted(modellCP2), residuals(modellCP2), xlab = "Fitted",
ylab = "Residuals", col = "green", pch = 21, bg = "blue", cex = 1.2, cex.lab = 1.2, cex.axis = 1.2)
abline(h=0)
#CP3-5
attach(malpie)
66 | P a g e
modellCP35 <- glm(CP35~pH + OrgMat+Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
step <- stepAIC(modellCP35, direction="backward")
step$anova
## All predictors were found to be important
attach(malpie)
modellCP35 <- glm(CP35~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
summary(modellCP35)
##overdispersion was observed ##Negative binomial regression also recommended for overdispersed data
attach(malpie)
modellCP35 <- glm.nb(CP35~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie)
step <- stepAIC(modellCP35, direction="backward")
step$anova
# Negative binomial modelling for best predictors
#Final Model:CP35 ~ pH + OrgMat + Clay
attach(malpie)
modellCP35 <- glm.nb(CP35~pH+OrgMat+Clay, data=malpie) summary(modellCP35)
#FF
attach(malpie)
modellFF <- glm(FF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
step <- stepAIC(modellFF, direction="backward")
step$anova
## All predictors were found to be important
attach(malpie)
modellFF <- glm(FF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
summary(modellFF)
##overdispersion was observed
##Negative binomial regression also recommended for overdispersed data
attach(malpie)
modellFF<- glm.nb(FF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie)
step <- stepAIC(modellFF, direction="backward")
step$anova
## Negative binomial modelling for best predictors
## Final Model:FF ~ pH + OrgMat + Cdtot + Zntots + Clay
attach(malpie) modellFF <-glm.nb(FF~pH+OrgMat+Cdtot+Zntots+Clay, data=malpie)
summary(modellFF)
###BF
attach(malpie)
modellBF <- glm(BF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
step <- stepAIC(modellBF, direction="backwards")
step$anova
## All predictors were found to be important
attach(malpie)
modellBF <- glm(BF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson()) summary(modellBF)
#overdispersion was observed
#Negative binomial regression also recommended for overdispersed count data
attach(malpie)
modellBF<- glm.nb(BF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie)
67 | P a g e
step <- stepAIC(modellBF, direction="backward")
step$anova
# Negative binomial modelling for best predictors
#Final Model: BF ~ pH + Moisture + Clay
attach(malpie)
modellBF <-glm.nb(BF~pH + Moisture + Clay, data=malpie)
summary(modellBF)
###PF
attach(malpie)
modellPF <- glm(PF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
step <- stepAIC(modellPF, direction="backward")
step$anova
## All predictors were found to be important
attach(malpie)
modellPF <- glm(PF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
summary(modellPF)
##overdispersion was observed
##Negative binomial regression also recommended for overdispersed count data attach(malpie)
modellPF<- glm.nb(PF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie)
step <- stepAIC(modellPF, direction="backward")
step$anova
## Negative binomial modelling for best predictors
## Final Model:PF ~ pH + OrgMat
attach(malpie)
modellPF <-glm.nb(PF ~ pH + OrgMat, data=malpie)
summary(modellPF)
###RHF attach(malpie)
modellRHF <- glm(RHF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
step <- stepAIC(modellRHF, direction="backward")
step$anova
## All predictors were found to be important
attach(malpie)
modellRHF <- glm(RHF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie,
family=quasipoisson())
summary(modellRHF)
##overdispersion was observed
##Negative binomial regression recommended for overdispersed count data attach(malpie)
modellRHF<- glm.nb(RHF~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie)
step <- stepAIC(modellRHF, direction="backward")
step$anova
## Negative binomial modelling for some predictors
## Final Model:RHF ~ 1
attach(malpie)
modellRHF <-glm.nb(RHF~ pH+OrgMat+Moisture, data=malpie)
summary(modellRHF)
##OMV
attach(malpie)
modellOMV <- glm(OMV~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
step <- stepAIC(modellOMV, direction="backward")
step$anova
68 | P a g e
## All predictors were found to be important
attach(malpie)
modellOMV <- glm(OMV~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
summary(modellOMV)
##overdispersion was observed
modellOMV<- glm.nb(OMV~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie)
step <- stepAIC(modellOMV, direction="backward")
step$anova ## Negative binomial modelling for best predictors
## Final Model:OMV ~ OrgMat+Cdtot+Moisture
attach(malpie)
modellOMV <-glm.nb(OMV~OrgMat+Cdtot+Moisture, data=malpie)
summary(modellOMV)
###PRED
attach(malpie)
modellPRED <- glm(PRED~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
step <- stepAIC(modellPRED, direction="backward")
step$anova ## All predictors were found to be important
attach(malpie)
modellPRED <- glm(PRED~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
summary(modellPRED)
##overdispersion was observed
##Negative binomial regression recommended for overdispersed count data
attach(malpie)
modellPRED<- glm.nb(PRED~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie)
step <- stepAIC(modellPRED, direction="backward")
step$anova
## Negative binomial modelling for best predictors
## Final Model: PRED ~ pH + OrgMat attach(malpie)
modellPRED <-glm.nb(PRED ~ pH + OrgMat, data=malpie)
summary(modellPRED)
###ECTOP
attach(malpie)
modellEctop <- glm(Ectop~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
step <- stepAIC(modellEctop, direction="backward")
step$anova ## All Predictors were found to be important
attach(malpie)
modellEctop <- glm(Ectop~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie, family=poisson())
summary(modellEctop)
##overdispersion was observed
##Negative binomial regression also recommended for overdispersed count data
attach(malpie)
modellEctop<- glm.nb(Ectop~pH + OrgMat + Cdtot + Zntots + Moisture + Clay, data=malpie)
step <- stepAIC(modellEctop, direction="backward")
step$anova
## Negative binomial modelling for best Predictors
## Final Model: Ectop ~ Cdtot + Zntots + Moisture + Clay
attach(malpie)
modellEctop <-glm.nb(Ectop ~ Cdtot + Zntots + Moisture + Clay, data=malpie)
summary(modellEctop)
69 | P a g e
Appendix 5. Food web analysis graph based on moisture gradient classifications
(A=moisture content at field capacity, B=moisture content at saturation point)