Does ecology predict taxonomy? How ecological differentiation can be used to spatially infer intra-specific diversity

Assessing the true dimension of biodiversity is a major challenge. Many
species hide within them a diversity that is now being uncovered using
molecular data. However, population genetic studies tend to be
resource-consuming and more difficult to apply to a broader range of taxa,
limiting scalability. Moreover, the growing shortage of trained
taxonomists makes it difficult to rely on comparative morphological
studies to assess divergence and speciation processes for the vast
majority of species-rich taxonomic groups. Here, we explore the usefulness
of the “ecological speciation” concept and explore how these hidden
lineages tend to occupy a distinct environmental niche that can be used to
identify natural groups in the geographical space. From a total of 298
species complexes, we assess the accuracy of five clustering methods and
Random Forest models for correctly classifying the occurrences of the
different subspecies based only on environmental data. For the best
performing clustering method (Gaussian Mixture Models), we obtained that
species can be predicted above random classification with a median
Adjusted Rand Index of 0.37, only by their environmental profile. Random
Forest, on the other hand, showed high accuracy for most of the species
(>0.75). We believe accuracy could be further improved by using
species-specific climatic variables, although this study focuses on a
widely applicable method. Our goal is to demonstrate that clustering
methods can be used on a large scale to reveal the true diversity hidden
within taxonomic complexes, thereby reducing the time and budget required
for exploratory analysis. We also aim to demonstrate the extent to which
different taxa are determined and delimited by the environment.