In later GBIF snapshots, the reported 2006–2015 loss of recorded bee species fell from 22% to 0% (all records) and from 26% to 10% (specimens), May 2020 to October 2026

Zattara and Aizen (2021) used a May 2020 GBIF download to report that about 25% fewer bee species were recorded in 2006–2015 than before the 1990s. They removed 2016–2019 because of data-entry lag, but they did not test the earlier years. We repeated their filters and their Chao1 richness metric on the same collection years in GBIF AWS snapshots from 2021 to October 2026. Our code reproduces their result: a 2006–2015 loss of 22.1% (all records) and 26.1% (preserved specimens) relative to 1946–1995, and a GLS year effect of −29.5 ± 10.8 species per year (published value −31.9 ± 11.0). In later snapshots the loss became smaller. In October 2026 it was 0.0% for all records and 9.8% for specimens. Specimen records for 2006–2015 increased 2.15 times after May 2020, and 1,492 of the 6,436 species recorded in 2006–2015 come only from datasets that were not in the 2020 download. The covariate-adjusted GLS trend for all records is no longer significant (−15.3 ± 11.7, p = 0.20). The specimen-only trend remains negative (−23.3 ± 9.0, p = 0.017) but is smaller than in 2020 (−40.5). A bootstrap over GBIF datasets gives wide 95% intervals for the loss, also in the original data (−41% to +41%). At equal numbers of records, the 2006–2015 records still contain 28–31% fewer species. But the 20 most-recorded species now supply 33–38% of these records, against 13–19% before 1996, so changed recording practice can explain this result. The 2021 snapshot holds only CC0/CC BY georeferenced records, so we do not compare it directly. Conclusion: most of the published decline came from data that GBIF did not yet hold in 2020. The GBIF counts do not support a global decline in bee species richness, and they also do not show that no decline occurred.

bees · Anthophila · GBIF · data mobilization lag · species richness · biodiversity decline · museum specimens · digitization · Chao1 · data vintage · reproducibility · pollinators