German spring phenology kept its elevation gradient from 1952 to 2025 (2.75 and 2.76 days per 100 m), while thermal-time models with station temperatures predict a 10-12% decrease

Question: did the elevation gradient of spring phenology (days of delay per 100 m of station elevation) become smaller in Germany, as reported for the Alps, and do temperature-driven models predict the change? Data: Deutscher Wetterdienst (DWD) annual-reporter network, 32 spring phases of 28 wild plants, 1,956,717 dates at 1,900 stations (0-1,142 m), 1952-2025, with HYRAS-DE 1-km daily temperature at each station. Method: we estimated the gradient in each year within natural-region-by-year cells, and its trend with station fixed effects. We applied the same estimator to dates predicted by three thermal-time models with fixed parameters (forcing with station thresholds, forcing with one threshold, chilling-forcing), fitted per phase on 1952-1987. Intervals are moving-block bootstraps of years. Results: the observed gradient was 2.75 days per 100 m in 1952-1961 and 2.76 in 2016-2025. It rose to 3.53 in 1972-1981 and then fell. The models reproduce this rise and fall (year-to-year correlation 0.76), but they predict a 10-12% net decrease. The observed trend was -0.022 days per 100 m per decade (95% interval -0.102 to +0.035). The model trend was -0.084 (-0.161 to -0.025). The difference, +0.062 (+0.027 to +0.088), was positive in 24 of 32 phases. The difference stayed between +0.05 and +0.08 when we fitted the models on all years or only on 1989-2025, used 1-degree cells, excluded stations above 800 m, or used only western states. Uniform warming of +1 degC lowers the model gradient by 8% (range 3-14%), so compression does not need elevation-dependent warming. HYRAS and 153 DWD climate stations agree that March-April warming was only about 0.01 degC per 100 m per decade larger at higher sites. In 1988-2025 the model error was larger at low stations (3.7 days too late at 0 m, 1.7 days at 800 m). Opposite to the Alpine report, warmer winters came with a steeper gradient. Limitations: the station elevations only reach 1,142 m. Observer, urban or local land-use change at low stations is a possible cause that this data cannot separate. The bootstrap keeps the model parameters fixed.

phenology · elevation gradient · Hopkins bioclimatic law · thermal-time model · growing degree days · climate change · Germany · DWD · HYRAS · elevation-dependent warming · space-for-time · spring leaf-out · flowering