Using Self-Organizing Maps to Explore Patterns in Species Richness and Protection

Abstract

The combination of species distributions with abiotic and landscape variables using Geographic Information Systems can prioritize areas for biodiversity protection by identifying areas of high richness, although the number of variables and complexity of the relationships between them can prove difficult for traditional statistical methods. The use of these methods, which commonly assume linearity and low correlation between independent variables, can obscure even strong relationships and patterns. Self-Organizing Maps (SOM) is a heuristic statistical tool based on machine learning methods that can be used to explore patterns in large, complex datasets for linear and nonlinear patterns. Here we use SOM to visualize broad patterns in species richness by taxonomic group (birds, mammals, reptiles, and amphibians) and 78 habitat, landscape and environmental variables using data from the Gap analysis project for West Virginia, USA. Soil and habitat variables demonstrated clear relationships with species richness; areas with high species richness occurred in areas with high soil richness. Landscape metrics were less important, although habitat diversity and evenness indices were positively related to species richness in some taxonomic groups. Current coverage of protected areas (e.g., National Forests and state parks) appeared to be insufficient to cover most of the areas of high species richness, especially for reptiles; many of the polygons with the highest richness were not covered by these areas. The identification of polygons with high richness and low protection can be used to focus conservation efforts in those areas.