GBIF records added after 2020 removed most of the reported 2006–2015 drop in recorded bee species (−26% to −8%), and the drop moved to the newest decade, so this index does not show global bee decline

Zattara & Aizen (2021, One Earth) used GBIF occurrence records to report a global decline in bee species richness: about 25% fewer bee species had records in 2006–2015 than before the 1990s. We repeated their analysis on their own 2020 GBIF download (doi:10.15468/dl.ysjm4x, 9.18 million Hymenoptera records). Then we applied the same method to the 1 October 2026 GBIF snapshot (3.95 billion records read). The reproduction is exact: Chao1 richness of the 2006–2015 bin is 22.1% (specimens and human observations) and 26.2% (specimens only) below the 1946–1985 mean. In the 2026 snapshot, the same years and filters give +1.8% and −8.2%. Between the two data versions, species recorded per year increased by a factor of 1.26 for 1946–1985 and 1.68 for 2006–2015. Datasets registered after the 2020 download supply 26% of the 2006–2015 specimen records. Without them, the 2026 change is −25.9%, the same as in 2020. The drop moved forward in time: in the 2026 snapshot, specimen richness is 29.6% below the baseline in 2016–2025, and 27.4% below it in 2013–2022, which has the same data age as 2006–2015 in 2020. With equal numbers of specimen records from 21 countries, the 2006–2015 change is −0.1%. For the USA, the change was −27.6% in 2020 and is −3.2% now. Among 173 insect families, 46% of non-bee families also have a specimen richness drop, and bee families are not unusual (mean percentile rank 0.37, permutation p = 0.16). When we resample the 577 contributing datasets, the 95% interval of the original −26.2% is −47.9% to +43.0%. Thus the GBIF 'species recorded' index for bees has a drop at the newest decade of each data version, because publication of specimen data lags collection. The index is not reliable evidence of a global bee decline. Limits: the test cannot show that bee diversity is stable. Andrenidae and Colletidae still have drops of about 20%, which can be real or can come from lag. The two data versions use different GBIF taxonomies. We did not repeat the GLS, continent and evenness analyses. Code, aggregate count tables and results are included, and they reproduce all results byte for byte.

bees · Anthophila · GBIF · species richness · biodiversity trends · data mobilization lag · museum specimens · Chao1 · reanalysis · pollinator decline · occurrence records · sampling bias