Yearly fits with flowering dates cut mean squared fruit-date error by 13% for two elder species and rowan in Swiss records, 2011–2025

We tested whether observed flowering dates improve later-year predictions of fruit maturity for black elder, red elder and rowan in the Swiss Phenology Network. Stations were selected from 1996–2010 records only. The test used 2,448 species-station-year pairs at 97 distinct stations in 2011–2025. Each year, linear models were fitted to earlier years only. Adding a fitted flowering term to station effects and a time trend reduced the equal-species mean squared error by 12.9%, with a nominal conditional 95% interval of 5.3% to 19.5%. The pooled root error fell from 15.61 to 14.57 days. Every species had lower point error. A model with its flowering slope fixed at one gave a smaller, uncertain mean gain. A model kept fixed after 2010 also gave an uncertain gain. The intervals resample test stations and year blocks while keeping fitted predictions fixed. They omit training-parameter uncertainty. The result applies to yearly updated linear models and selected paired records. It does not identify a causal flowering effect or measure fruit amount. The package includes unchanged source files, provenance, code, predictions, checks and figures.

phenology · fruit maturity · flowering · MeteoSwiss · chronological prediction · Sambucus · Sorbus · open data