Species distribution models (SDM) are computer algorithms that relate species distribution, occurrence or abundance, with information on environmental conditions and spatial characteristics of locations where the species has been, or is suspected to be, found. These models can be used to predict or to have a better understanding of the species distribution, and are widely used in ecology, evolutionary biology and conservation.
With climate emergency becoming a reality, in recent years there has been an increased interest in understanding how future environmental changes may impact species distribution. As a result, novel methods to better predict environmental suitability for species have been developed, accompanied by a drive to improve performance of existing models.
A variety of statistical methods have been applied to species distribution modelling, from regression-based models or neural networks to maximum entropy models. The problem with most of these methods is that species responses along environmental gradients are not ecologically meaningful or otherwise difficult to interpret.
It seems evident that species distribution models need a stronger theoretical background. One key point is clarifying the relationship between species distribution models and the concept of ecological niche, resulting in a statistical model that should be ecologically plausible.
According to ecological niche theory, species distributions should have a single peak with respect to environmental gradients, i.e., they are unimodal relationships. This is a technical way of saying that when environmental conditions become less favourable, for example, various stages of the life cycle (feeding, growth and reproduction) are affected, resulting in lower presence of the species. If species data are scarce and there is a heterogeneous distribution of species occurrence along gradients, we find the problem of the models providing multimodal and ecologically non-meaningful relationships with environmental variables. Thus, species distribution models need to combine environmental variables that are expected to meet the ecological niche theory with other explanatory variables that do not interact dynamically with the species and hence are not affected by species abundance, the so-called scenopoetic variables, which have no shape (modality) restrictions.
Commonly used methods to build species distribution models in the ecological niche theory framework include regression-based methods, such as Generalized Linear Models (GLMs) and Generalized Additive Models (GAMs). Shape-constrained generalized additive models (SC-GAMs) are based on the same statistical framework as GLMs and GAMs regression methods, but they allow, as the name suggests, to incorporate shape-constraints. Imposing some constraints should be an effective alternative to fitting certain response curves, while retaining the unimodality constraint, required by ecological niche theory, for direct variables and limiting factors.
Now, a team of researchers has assessed 1 the performance of SC-GAMs in fitting species distribution models under the ecological niche theory in comparison with other approaches. Two different implementations of SC-GAMs were evaluated: the maximum likelihood implementation from the scam R Core Team package and the component-wise boosting approach from the mboost R package. Both methods have been also tested in two different real case studies.
They conclude that the proposed SC-GAMs can be readily applied for fitting distribution models and are useful tools for modelling communities of large number of species, as they result in a good balance between goodness of fit and agreement with ecological niche theory. They can incorporate multiple explanatory variables with or without interaction, both shape-constrained and unconstrained, depending on the nature of the variables involved.
Therefore, SC-GAMs offer the possibility of investigating, for example, the effect of climate change on multiple species without requiring sophisticated and time-consuming mechanistic models that depend on detailed knowledge of vital rates and life traits for each species.
Author: César Tomé López is a science writer and the editor of Mapping Ignorance
Disclaimer: Parts of this article may be copied verbatim or almost verbatim from the referenced research paper.