An Evolutionary Algorithm for the Selection of Geographically Informative Species

The geographic distribution of species is of interest in making conservation plans, designating biosphere reserves, and in understanding the different range sizes of species. In this paper, an evolutionary algorithm is used to classify species of freshwater crustacean zooplankton as geographically informative or geographically non-informative. Geographically non-informative species tend to have a broad distribution, while geographically informative species exhibit a locality that suggests a greater degree of ecological vulnerability. The data under analysis were mined from the literature from 1930 to 1992 and are in the form of presence/absence data for each species. An evolutionary algorithm is used to maximize the correlation of the geographic distances between ponds with the Hamming distances between ponds computed from the species presence/absence data. Maximization is over the set of species selected for computation of the Hamming distance. The representation used is a binary-gene evolutionary algorithm that selects which species are used. The evolutionary algorithm has highly consistent results between runs. All the runs select the same large group (meaning those species are geographically informative) and fail to select from another large group (meaning those are geographically non-informative); a small number of species are ambiguous, sometimes selected and sometimes not. A second set of experiments in which a large number of short runs are made demonstrates that the fitness landscape of the optimization problem in this study is not a single multi-dimensional hill. Non-trivial interaction between different loci in the representation is found. This study demonstrates that the scheme presented for searching for a purely biologically-based distance that mimics geographic distance is both practical and a non-trivial evolutionary computation problem