Comparability across and within groups
Amphibian, dragonfly and reptile maps have separate drivers of spatial bias, and their biases differ both spatially and temporally as a consequence of differing methodologies. Consequently, the main benefit of using a single data source is entirely lost, as this inconsistency precludes attributing differences between regions or taxa to genuine differences rather than to methodological differences. For example, reptile distribution maps frequently relied on either one-degree or half-degree grids, with these boundaries clearly detectable on maps of species range boundaries and turnover, with, for example, >71% of range boundaries shared by >5 species falling on either a political boundary or on a half-degree grid. The fact that grids and political areas and other features are variably used only further complicates these issues. Even odonata, where more biologically-relevant river basin boundaries are used, the near-universal use of such features (i.e. 92% of formerly mapped ranges are on river basin boundaries) may still result in inaccurate maps with both type one and type two errors. Thus, whilst trimming ERMs with appropriate filters could be applied to most taxa, assessments of gridding for reptiles before such an approach could be usefully applied. For other taxa, once possible political boundaries have been assessed, it may be possible to trim species ranges based on clear assessments of habitat needs.
Regional biases are an especially notable issue. For example, only 31% of US county boundary area (width=1km) has no amphibian boundary data, whereas 98% of the land more than 500m from a boundary has no amphibian range boundaries (Figure S2a). This is because of a specific initiative within the US, and such inconsistent standards makes comparable analysis between regions impossible (Blackburn et al., 2002). Once US Amphibians which have all their borders on county borders are removed, only 40% of species remain, and the remainder includes range-restricted and invasive species. As county limits do not typically follow ecological boundaries, they clearly do not represent species boundaries, thus, most forms of trimming could not be used to accurately map ranges or diversity of such a group.
In many cases, the development of published IUCN distributions contradicts their general guidance (IUCN guidelines: https://www.iucnredlist.org/resources/guidelines-for-appropriate-uses-of-red-list-data). Despite the fact that former IUCN training material (https://www.iucn.org/content/iucn-red-list-training-course-now-online) explicitly said not to draw buffers around single locality points for building distributions, this has clearly been done for hundreds of species across taxa, extensively for some groups (Figure S2d). In addition to other inconsistent biases both within and across groups, these datasets lose the invaluable comparability that would normally be expected from methods that are standardized temporally, spatially and taxonomically. Consequently, further validation and refinement with verified point data is necessary before these data are used for formal conservation management.