Previous autumn colouring gave no clear gain in linear spring predictions at 88 Swiss beech stations in 2011–2025 with weather through February

A tree's autumn state can affect its next spring. We tested whether the previous autumn's 50% leaf-colouring date improves spring leaf-unfolding predictions for European beech at Swiss stations. We fitted two linear models on 3,099 observations in 1962–2010. Both used station effects, a time trend and six monthly temperatures through February. One also used the autumn date. We tested the models on the same 1,173 observations at 88 stations in 2011–2025. Root mean squared error was 7.9861 days for the baseline and 7.9866 days with autumn. The change was +0.0005 days, with a 95% common three-year-block interval of −0.054 to +0.068 days. A check that fitted the models again in sampled training years gave −0.068 to +0.108 days. The historical autumn coefficient was positive, but it gave no clear gain in later predictions. The test uses retrospective weather with an information cut at 1 March. It applies to these additive linear models and long station records. It does not establish that trees have no memory across winter. The package preserves source files, hashes, the full data ledger, predictions, uncertainty draws, figures and offline repeat code.

phenology · European beech · autumn colouring · spring leaf unfolding · prediction · Swiss Phenology Network · temporal validation · open data