G. Evelyn Hutchinson defined the niche as a set of environments – specifically the set of environments in which species can reproduce and persist. Much ecological theory is based on this definition and practical applications abound, especially the modeling of species geographic distributions. To test theories founded on Hutchinson’s conception of the niche requires an operational means to separate niche from non-niche environments, i.e. to be able to draw a boundary around the species niche. But most contemporary approaches to niche modeling do not do this. Instead, they seek to estimate the center of a species niche (which could be the same for a specialist and a generalist species despite the fact that they have very different niche “widths”) or to statistically measure the conditional probability of finding a species at a particular location in the environmental space. I argue that “centrality based” and “probabilistic” niche models do not provide the representations needed to test ecological theories that draw on Hutchinson’s concept of the niche as a set of environments (although they may represent other ecological concepts, such as environmental suitability). Moreover, these methods can lead to counter-intuitive conclusions, such as the conclusion that a particular habitat is irrelevant to the persistence of a population when in fact it is essential. The idea of concept learning pertains to the recognition of classes of objects from observations of instances of those objects. This study uses concept learning for identifying the boundaries of ecological niches. I present some novel algorithms, which are easy to use and devised specifically to represent Hutchinson’s concept of the ecological niche. The empirical performance of of these methods is compared with popular probabilistic methods.
Results/Conclusions
Boundary estimation methods for ecological niche modeling perform comparably to statistical methods and are conceptually preferable in the sense that they more directly represent key ecological concepts. The utility of boundary estimation for ecological niche modeling suggests that concept leaning may be useful to the identification of other ecological concepts, too, particularly in community ecology and ecosystem ecology.