GBIF cleaning
GBIF data are useful for testing species distributions. To ensure exclusion of inaccurate localities, we filtered them stepwise before assessing ERMs. Firstly, oceanic records were removed with a terrestrial mask. An ecoregion map (https://ecoregions2017.appspot.com/) was then used to filter samples clearly in the wrong localities using the realms species occupy according to IUCN data. Corrections were made when listed IUCN realms were regularly inconsistent with distribution maps, and further analysis to assay what realm the IUCN range maps were needed assay genuine realms IUCN maps fell into, to develop matching filters and using those realms to filter GBIF data from the same realms.
As GBIF data includes some synonyms, these were also corrected prior to their use. Synonym lists were developed via IUCN-lists, for birds Clements bird checklist (https://www.worldbirdnames.org/ioc-lists/master-list-2/). As IUCN lists sometimes gave species as both synonyms and true species, any species listed as both was corrected during filtering. Given the slow rate of taxonomic updates on the redlist e.g only 45% of amphibian species described between 2004- 2016 were assessed by the IUCN (Tapley et al., 2018) and GBIF efforts to update data filters, our approaches are an appropriate resolution and accuracy to assess under-estimation in species ranges from their polygon data. Following GBIF filtering, percentages of points within appropriate polygons were calculated. Analyses of relationships between species range boundaries and administrative boundaries (coastal, non-coastal) were made in R.