Seminar: Christian Reimers

Plants mediate exchanges of water, carbon, and energy between land and atmosphere, so accurately representing vegetation phenology is essential for reliable climate projections. This study addresses four related questions: how start, end, and length of the growing season respond to temperature, how near-start-of-season frost events affect subsequent timing and total greenness, the relative importance of current meteorological conditions versus memory (lagged) effects and which meteorological variables matter most at different seasonal phases.Combining near-surface PhenoCam GCC observations with daily Daymet meteorological forcings across North America, we develop a novel deep-learning architecture of two transformers: one transforms daily meteorology into a stress signal and the other maps that stress time series to daily greenness. The model substantially outperforms competing approaches (−9% MSE, +10% R2) and more accurately captures temperature–phenology sensitivities.Using this model we can answer the questions posed above. Warming generally lengthens the growing season for many plant functional types (PFTs), for example, deciduous broadleaf forests, which lengthen by 5.57 days °C−1 via earlier springs and later autumns. Late spring frosts delay peak greenness across PFTs (average delay 8.36 days for deciduous broadleaf), with vulnerability concentrated within about two days of season start. Phenological dynamics depend on both immediate weather and memory: current meteorology is most influential at season start and end, while lagged effects dominate during the growing and non-growing seasons. Soil temperature triggers spring green-up, atmospheric dryness drives within-season declines, and soil/air temperature plus chill and moisture conditions shape autumn and winter greenness.