Chapter 4
Cluster Analysis
4.1 Objectives
In most cases, data exploration (Chap. 2) and the computation of association
matrices (Chap. 3) are preliminary steps towards deeper analyses. In this chapter
you will go further by experimenting one of the large groups of analytical methods
used in ecology: clustering. Practically, you will:
• learn how to choose among various clustering methods and compute them;
• apply these techniques to the Doubs River data to identify groups of sites and fish
species.
• explore two methods of constrained clustering, a powerful modelling approach
where the clustering process is constrained by an external data set.
4.2 Clustering Overview
The objective of clustering is to recognize discontinuous subsets in an environment
that is sometimes discrete (as in taxonomy), but most often perceived as continuous
in ecology. This requires some degree of abstraction, but ecologists may want to get
a simplified, structured view of their data; generating a typology is one way to
achieve that goal. In some instances, typologies are compared to independent
classifications (based on theory or on other typologies obtained from independent
data). What we present here is a collection of methods used to decide whether objects
are similar enough to be allocated to a group, and identify the distinctions or
separations between groups.
Clustering consists in partitioning the collection of objects (or descriptors in
R-mode) under study. A hard partition is a division of a set (collection) into subsets,
such that each object or descriptor belongs to one and only one subset for that
partition (Legendre and Rogers 1972). For instance, a species cannot be
© Springer International Publishing AG, part of Springer Nature 2018
D. Borcard et al., Numerical Ecology with R, Use R!,
https://doi.org/10.1007/978-3-319-71404-2_4
59
Cluster Analysis
4.1 Objectives
In most cases, data exploration (Chap. 2) and the computation of association
matrices (Chap. 3) are preliminary steps towards deeper analyses. In this chapter
you will go further by experimenting one of the large groups of analytical methods
used in ecology: clustering. Practically, you will:
• learn how to choose among various clustering methods and compute them;
• apply these techniques to the Doubs River data to identify groups of sites and fish
species.
• explore two methods of constrained clustering, a powerful modelling approach
where the clustering process is constrained by an external data set.
4.2 Clustering Overview
The objective of clustering is to recognize discontinuous subsets in an environment
that is sometimes discrete (as in taxonomy), but most often perceived as continuous
in ecology. This requires some degree of abstraction, but ecologists may want to get
a simplified, structured view of their data; generating a typology is one way to
achieve that goal. In some instances, typologies are compared to independent
classifications (based on theory or on other typologies obtained from independent
data). What we present here is a collection of methods used to decide whether objects
are similar enough to be allocated to a group, and identify the distinctions or
separations between groups.
Clustering consists in partitioning the collection of objects (or descriptors in
R-mode) under study. A hard partition is a division of a set (collection) into subsets,
such that each object or descriptor belongs to one and only one subset for that
partition (Legendre and Rogers 1972). For instance, a species cannot be
© Springer International Publishing AG, part of Springer Nature 2018
D. Borcard et al., Numerical Ecology with R, Use R!,
https://doi.org/10.1007/978-3-319-71404-2_4
59