Pre-solstice warming advanced and post-solstice warming delayed leaf colouring of four German tree species (1951–2025), but these effects did not predict later colouring in 2016–2025

We tested two claims about autumn leaf senescence with 466,016 station-years of leaf unfolding and leaf colouring (50 % of leaves coloured) for Betula pendula, Aesculus hippocastanum, Fagus sylvatica and Quercus robur in the open German Weather Service (DWD) phenology network, 1951–2025, with 1-km monthly temperature grids. The original studies used data that end near 2015, so 2016–2025 is an out-of-sample test. We wrote the analysis plan before model fitting. Station fixed-effects models showed that warmer May–June advanced colouring (1951–2015: −1.1 to −2.2 days/°C; 2016–2025: −1.5 to −2.1 days/°C, wild cluster bootstrap p = 0.006–0.032) and warmer July–September delayed it (+1.1 to +2.4 days/°C; p = 0.03–0.25 in 2016–2025). Leaf-out of the same species at neighbouring stations, which carries the weather signal but not the observer error, shifted colouring by about 0.1 day per day (0.15–0.30 in 2016–2025). The local part of the spring–autumn correlation had the same size when we used the leaf-out of a different species at the same station, so it is mostly an observer or site effect. Models fitted on 1951–2015 followed national year-to-year anomalies in 2016–2025 (r = 0.64–0.90), but did not predict the mean shift. At the same stations, colouring in 2016–2025 was 1.8 to 5.0 days later than in 1951–2015. The pre-specified model predicted −1.6 to +0.4 days, because pre- and post-solstice warming effects cancelled. The unexplained delay was 1.1 to 4.8 days across all tested model variants and grew gradually from the 1990s. Thus, the sign pattern of the solstice hypothesis replicated, but interannual temperature effects and spring carryover did not explain the long-term colouring delay, and we saw no net advance. Limits: volunteer observers are not identified in the open data; changes in observation practice, artificial light, CO2, or long-term acclimation can each cause part of the residual delay; inference for 2016–2025 has 10 year clusters.

phenology · leaf senescence · autumn leaf colouring · summer solstice · spring carryover · out-of-sample test · DWD · Germany · Fagus sylvatica · Quercus robur · Betula pendula · Aesculus hippocastanum · observer effects · climate change