GBIF record dates show a 3.4 times larger 1980–2020 flowering advance than the DWD network for 27 German plant species, and fern records also advance, so recording changes cause most of the trend

Question: do flowering trends from the dates of GBIF occurrence records in Germany agree with trends from the standardized Deutscher Wetterdienst (DWD) phenological network for the same species? Freimuth et al. (2022, Proc. R. Soc. B) used record dates and reported a plant advance of 7.2 days per decade in 1980–2020. Data: all 35.7 million German vascular plant records in the GBIF snapshot of 1 October 2026 (17.98 million dated to the day, 1980–2025) and DWD flowering-onset records for 27 wild species. Method: per-species OLS trends of record day of year with latitude and longitude (the published method), with dataset fixed effects, with a target-group correction, and as annual medians in a fixed flowering window; a two-way fixed-effects DWD index; species bootstrap intervals; a fern negative control; detrended year-to-year correlations. Results: the published method gave a mean trend of −6.7 days per decade for 1,864 species, near the reported −7.2. For the 27 DWD species, it gave −12.0 days per decade, but the DWD trend was −3.5 (difference −8.5, 95% CI −11.8 to −5.6; 25 of 27 species more negative). Fifty-two fern and lycophyte species have no flowers, but their mean record-date trend was −3.5 days per decade. Bavarian summer floristic surveys supplied 93% of dated records in the 1980s and 7% in the 2020s, while spring-biased app and citizen-science sources grew. The mean record date of all plant records moved from day 220 to day 173. Without the Bavarian surveys, the bias was −3.6 days per decade (−6.2 to −1.1). Dataset fixed effects decreased the bias to −2.1 (−4.7 to 0.1). In a flowering window with dataset effects, detrended GBIF and DWD anomalies had a mean r of 0.50 (all 27 positive; highest for conspicuous spring flowers), but GBIF anomalies were smaller (slope 0.31). Conclusion: raw record dates in Germany do not give a valid long-term flowering trend; changes in sources and recording season cause most of the published trend. Limitations: one country, 27 mostly spring-flowering species, simple estimators, a newer snapshot than the original study, and an imperfect fern control.

phenology · GBIF · occurrence records · citizen science · sampling bias · flowering time · DWD · Germany · climate change · replication · observer effort · data sources