Evaluation of different aspects of maximum entropy for niche-based modeling

摘要

Biodiversity conservation is a world challenge that needs attention and efficient strategies for its success. Modeling of geographic distributions of species is used in assorted applications related to biodiversity conservation. Maximum entropy (maxent) is a technique recently applied to modeling of geographic distributions of species and is being largely used by biologists. The aim is to evaluate different viewpoints of this technique. The first evaluation is concerned with the performance of the algorithm. A parallel version of the maxent-based algorithm available in openModeller is presented. openModeller is a set of tools provided for researchers interested in modeling of geographic distributions of species. The second evaluation is focused on tuning the regularization parameter, since it can severely affect the performance of the algorithm and can take a long time to be adjusted. In addition, the algorithm was evaluated without the use of a regularization parameter and with an adaptive maximum entropy approach. This approach was evaluated as a replacement of the regularization parameter. The validation of the assessments was based on a dataset with 20 species. The results show: an improvement in the algorithm performance using parallelism, considering only the running time; the regularization parameter does not depend on the number of samples, or on the number of iterations in training; species with the same number of samples fit better with different values of the regularization parameter (different magnitude order); the adaptive approach cannot replace the regularization parameter.