0. Brief introduction
This tutorial aims at describing the different features of the R package bioregion. The main purpose of the bioregion‘s package is to propose a transparent methodological framework to compare bioregionalization methods. Below is the typical flow chart of bioregions’ identification based on a site-species bipartite network or co-occurrence matrix with bioregion (Figure 1). This workflow can be divided into four main steps:
- Preprocess the data (matrix or network formats)
- Compute similarity/dissimilarity metrics between sites based on species composition
- Run the different algorithms to identify different set of bioregions
- Evaluate and visualize the results
Figure 1: Workflow of the bioregion’s package.
3. Pairwise similarity/dissimilarity metrics
The functions similarity and dissimilarity compute respectively pairwise similarity and dissimilarity metrics based on a (site-species) co-occurrence matrix. The resulting data.frame is stored in a bioregion.pairwise.metric object containing all requested metrics between each pair of sites.
The functions dissimilarity_to_similarity and similarity_to_dissimilarity can be used to transform a similarity object into a dissimilarity object and vice versa.
Please have a look at this tutorial page to better understand how these functions work.
4. Bioregionalization algorithms
The bioregion R package gathers several methods allowing to group sites and species into similar entities called bioregions. All these methods can lead to several partitions of sites and species, i.e. to different bioregionalizations.
Bioregionalization methods can be based on hierarchical clustering algorithms, non-hierarchical clustering algorithms or network algorithms.
The functions in the package are related to each of these three families and produce output that have a specific class, namely the bioregion.clusters class.
4.1 Hierarchical clustering
The functions relying on hierarchical clustering start with the prefix hclu_. With these algorithms, the bioregions are placed into a dendrogram that ranges from two extremes: all sites belong to the same bioregion (top of the tree) or all sites belong to a different bioregion (bottom of the tree).
See the following tutorial page for more details.
4.2 Non-hierarchical clustering
The functions relying on hierarchical clustering start with the prefix nhclu_. For most of these algorithms, the user needs to predefine the number of clusters, although this number can be determined by estimating the optimal bioregionalization.
See this tutorial page for more details.
4.3 Network clustering
The functions relying on network clustering start with the prefix netclu_. Site-species matrices can be seen as (bipartite) networks where the nodes are either the sites or the species and the links between them are the occurrences of species within sites.
With networks, modularity algorithms can be applied, leading to bioregionalization.
The following tutorial page details more each clustering functions relying on a network algorithm.
4.4 Microbenchmark
The different bioregionalization methods listed in the package rely on more or less computationally intensive algorithms.
The following page estimates the time required to run each method on data sets of different sizes.
5.1 Visualization
If sites have geographic coordinates, then each bioregionalization can be visualized with the function map_bioregions().
This tutorial page details different ways to plot your bioregionalization.
5.3 Summary metrics
In this section, we compute summary statistics at different scales, either at the bioregion or at the site or species level. Related functions are detailed in this page.