One of the most common questions in beekeeping is: when will the acacia bloom this year? The calendar gives an answer, but spring temperatures vary from year to year. Pollination Lab links flowering in its nectar flow score not to the calendar, but to heat accumulation (GDD). We tested this approach against Europe’s joint phenology database PEP725.

Key results

  • Acacia: The GDD method outperformed the calendar prediction adjusted for latitude and elevation. Mean absolute error dropped from 8.5 days to 6.2 days.
  • Linden: A small but consistent improvement (RMSE from 11.3 to 10.5 days).
  • Dandelion, willow and sunflower: GDD did not provide a clear advantage (only a very small improvement for willow with a 0 °C base). These species need different models.
  • Our current GDD ranges (e.g. Vienna and Kyiv) are consistent with these observations.
  • Results are valid for Central Europe. Transfer to Turkey and Azerbaijan remains an assumption for now.

Data used

PEP725 brings together plant phenology observations recorded by national meteorological services and volunteer observers across Europe. We obtained flowering start records from the PEP725 team: 237,507 observations, 5,289 stations, 1991–2026. For the analysis we used the 100 cells with the densest observations (0.25° boxes, mostly in Germany): 30,365 observations between 2000 and 2025. Temperature data were taken from the Open-Meteo archive (ERA5-based).

Method

  1. For each observation we calculated the accumulated GDD from 1 January to the flowering day at that station in that year. The formula is the same as in the application: the portion of the daily (max + min)/2 temperature above the base temperature.
  2. We tested base temperatures of 0, 3, 5, 7 and 10 °C.
  3. From the training data we derived a GDD threshold (median) for each species and predicted the flowering day as the day the threshold was exceeded.
  4. We tested the predictions on locations and years the model had never seen (5-fold cross-validation). The error metric was RMSE. The comparison baseline was a calendar prediction adjusted for latitude and elevation.

Results

Species Calendar + lat. + elev. (RMSE, days) GDD base 5 °C (RMSE, days) Outcome
Acacia 10.9 9.2 GDD better
Linden 11.3 10.5 Small gain
Dandelion 10.9 11.3 No gain
Willow 13.9 18.4 No gain (13.5 with base 0 °C)
Sunflower 14.7 15.6 No gain

For acacia, base temperatures of 5–7 °C performed best; 10 °C was worse. So the 5 °C base used in the application is appropriate.

GDD at flowering start (base 5 °C, median and interquartile range):

  • Acacia: 390 (351–434)
  • Linden: 620 (563–691)
  • Willow: 40 (26–55)
  • Dandelion: 101 (79–124)

We compared these with the current ranges in Pollination Lab. For example, the “May: acacia” line for Vienna covers 301–606 GDD, and the linden peak line starts at 607. The observed acacia start (351–434) falls inside the first range, and the linden start (around 620) is at the beginning of the second. Thus the ranges derived from climate normals are consistent with independent observations.

Limitations

  • Germany-heavy data. Outside Central Europe, especially for Turkey and Azerbaijan, the validity of these thresholds has not been confirmed.
  • Observer difference. Even with station-specific thresholds the error remains around 9 days. A large part of this comes from variation between observers in deciding “it has flowered”. This method cannot go below that floor.
  • Start of flowering, not peak nectar flow. The data measure the first day of flowering. Intense nectar flow usually comes later.
  • One model is not enough for every species. For very early species such as dandelion and willow, chilling (winter dormancy) may also play a role. These species need an additional model.
  • Limited open phenology data in Turkey. The phenology maps of the Turkish State Meteorological Service cover only fruit trees, olive and citrus and do not share tabular data. Validation for Turkey has therefore not yet been possible.

What this means for beekeepers

Heat accumulation tracks acacia flowering better than the calendar, but it does not give an exact date: typical uncertainty is about one week. Use the score as a supporting indicator alongside your own observations.

Frequently Asked Questions

What is GDD? Growing Degree Days is the accumulated excess of daily mean temperature above a base temperature. Plants develop according to accumulated heat, not the calendar.
What is PEP725? An open database that combines plant phenology records from national meteorological services and observation networks across Europe.
Are these results valid for Turkey as well? Not yet validated. The data come from Central Europe. We are looking for local observation data for Turkey.
Why were dandelion and willow not well predicted? These species flower very early and their flowering may also depend on winter chilling. Heat accumulation alone was not sufficient.

Try the nectar flow score

👉 lab.polinasyon.org/nektar

Related articles: Nectar Flow Score Engine Updated: GDD, VPD and Pine Honeydew · How the Nectar Flow Score is Calculated with GDD

Acknowledgements and sources

Data were provided by the members of the PEP725 project.

Altıntaş, F. (2026). Can Acacia and Linden Flowering Be Predicted with GDD? Pollination Lab Academy. https://www.polinasyon.org/blog/acacia-linden-flowering-gdd-pep725