Big data, changing taxonomy, and ghost records: Permanent preservation of collected specimens is essential for insect monitoring

Successful long-term biodiversity monitoring requires consistent identification of all specimens, both those newly collected and those collected earlier. Consistent identification requires re-examining and re-identification of older specimens because taxonomists replace older species definitions with modern concepts. Therefore, the loss or destruction of physical specimens may severely limit the accuracy of monitoring and biodiversity datasets. We demonstrate this problem using a case study of 3 datasets sourced from the USGS Bee Inventory and Monitoring Lab (BIML), including data from the biodiversity repository GBIF. We focus on the common bee genera Ceratina and Lasioglossum, which have both undergone recent revisions. BIML data contains numerous taxonomic inaccuracies for these genera due to changes in taxonomic concepts and identification criteria over time. We estimate that 25% to 28% of all pre-2010 records are inaccurate and must be re-examined because they are based on Ceratina or Lasioglossum identifications that have since been invalidated. However, the need for re-examination encounters a second, unfixable problem: re-examination requires that all collected specimens be preserved, but only a small synoptic collection has been saved, with the rest of the specimens given away or destroyed. As a result, these data are largely made up of "ghost records"that are unverifiable and inappropriate for monitoring. We discuss cascading issues, such as how destroyed specimens are acceptably reported as "preserved specimens"on GBIF, and we call on researchers monitoring insects to apply best practices for specimen preservation. Finally, we highlight natural history museums as a solution to many of these issues.