How Do Absence Data Predict the Performance of Species Distribution Models?

Species distribution maps are a fundamental data source for ecologists and evolutionary biologist that connect more broadly into policy, management, and society. These maps are not unfamiliar to the general public, often found in field guides and park brochures. Today, species distribution mapping and modeling can be done at much finer-scale resolution than previously, facilitated by new Geographic Information System (GIS) tools and techniques and the availability of species presence data and environmental data. However, most broad-scale species distribution models only utilize species presence data. Here, I explore how absence data can help improve species maps. The goal for my work is to show how species inventories can provide a means to generate probabilistic assessment of absences. In order to generate absences, I helped develop a means to assess the completeness of an “area inventory” (e.g. park or protected area surveys) by capturing metadata of published inventories. For those inventories assessed to be “mostly” complete, I compared the list of species present with a broader-scale list generated from range maps (utilizing a species list tool already available in Map of Life). If species were present in the Map of Life species list but not the inventory list, this was considered an inferred absence. Finally, I selected 25 species presumed to be absent based on my comparisons, to assess the performance of habitat models. I was able to document that these habitat models were generally making errors of commission (predicting suitable habitat where they should predict absence). Further refining these maps can further help (i) our knowledge in biodiversity, (ii) in conservation strategy for endangered species, which can lead to management and policy, and in (iii) educating the public about species and their distribution