These categories are not represented equally in the ecologist’s toolbox. Most
methods presented below are sequential, agglomerative and hierarchical (Sects. 4.3,
4.3.1, 4.3.2, 4.4, 4.5 and 4.6), but others, like k-means partitioning, are divisive and
non-hierarchical (Sect. 4.8). Two methods are of special interest: Ward’s hierarchical
clustering and k-means partitioning are both least-squares methods. That characteristic relates them to the linear model. In addition, we will explore two methods of
constrained clustering, one with sequential constraint (Sect. 4.14) and the other,
more general, called multivariate regression tree analysis (MRT, Sect. 4.12).
Hierarchical clustering results are generally represented as dendrograms or similar tree-like graphs. Non-hierarchical procedures produce groups of objects
(or variables), which may either be used in further analyses, presented as
end-results (for instance species associations) or, when the project has a spatial
component, mapped on the area under study.
A clustering of the sites of a species data set is informative by itself, but ecologists
often want to interpret it by means of external, environmental variables. We will
explore two ways of achieving this goal (Sect. 4.9).
Although most methods in this chapter will be applied to sites, clustering can also
be applied to species in order to define species assemblages. Section 4.10 addresses
this topic.
The search for indicator or characteristic species in groups of sites is a particularly
important question in fundamental and applied ecology. It is presented in Sect. 4.11.
Finally, a brief section (Sect. 4.15) will be devoted to two methods of fuzzy
clustering, a non-hierarchical approach that considers partial memberships of objects
to clusters.
Before entering the subject, let us prepare our R session by loading the necessary
packages and preparing the data tables.
# Load the required packages
library(ade4)
library(adespatial)
library(vegan)
library(gclus)
library(cluster)
library(pvclust)
library(RColorBrewer)
library(labdsv)
library(rioja)
library(indicspecies)
library(mvpart)
library(MVPARTwrap)
library(dendextend)
library(vegclust)
library(colorspace)
library(agricolae)
library(picante)
4.2 Clustering Overview
61
methods presented below are sequential, agglomerative and hierarchical (Sects. 4.3,
4.3.1, 4.3.2, 4.4, 4.5 and 4.6), but others, like k-means partitioning, are divisive and
non-hierarchical (Sect. 4.8). Two methods are of special interest: Ward’s hierarchical
clustering and k-means partitioning are both least-squares methods. That characteristic relates them to the linear model. In addition, we will explore two methods of
constrained clustering, one with sequential constraint (Sect. 4.14) and the other,
more general, called multivariate regression tree analysis (MRT, Sect. 4.12).
Hierarchical clustering results are generally represented as dendrograms or similar tree-like graphs. Non-hierarchical procedures produce groups of objects
(or variables), which may either be used in further analyses, presented as
end-results (for instance species associations) or, when the project has a spatial
component, mapped on the area under study.
A clustering of the sites of a species data set is informative by itself, but ecologists
often want to interpret it by means of external, environmental variables. We will
explore two ways of achieving this goal (Sect. 4.9).
Although most methods in this chapter will be applied to sites, clustering can also
be applied to species in order to define species assemblages. Section 4.10 addresses
this topic.
The search for indicator or characteristic species in groups of sites is a particularly
important question in fundamental and applied ecology. It is presented in Sect. 4.11.
Finally, a brief section (Sect. 4.15) will be devoted to two methods of fuzzy
clustering, a non-hierarchical approach that considers partial memberships of objects
to clusters.
Before entering the subject, let us prepare our R session by loading the necessary
packages and preparing the data tables.
# Load the required packages
library(ade4)
library(adespatial)
library(vegan)
library(gclus)
library(cluster)
library(pvclust)
library(RColorBrewer)
library(labdsv)
library(rioja)
library(indicspecies)
library(mvpart)
library(MVPARTwrap)
library(dendextend)
library(vegclust)
library(colorspace)
library(agricolae)
library(picante)
4.2 Clustering Overview
61