Soil and atmospheric drought trigger early leaf senescence and increase subsequent canopy mortality risk in temperate forests

Abstract

Early leaf senescence - an advanced decline in leaf functionality before the typical autumn period - has emerged as recurring drought response in temperate forests. Yet its large-scale patterns, drivers, and impacts on tree vitality remain poorly understood. Using six years (2018-2023) of Sentinel-2 time series across Germany, we estimated the onset of leaf senescence in European beech (Fagus sylvatica) and oak (Quercus robur, Q. petraea) forests. Early leaf senescence was most widespread during exceptional drought years (2018, 2019, and 2022) and showed pronounced spatial variability. Sustained high atmospheric demand (VPD) and critically low soil water potential ({psi}soil) in summer were key drivers identified by logistic regression analysis (F1-score = 0.57), with beech responding more strongly to soil drought and oak to atmospheric drought. Thresholds triggering early leaf senescence (probability >50%) were six weeks of daily maximum VPD [≥]1.9 kPa in beech and [≥]2.1 kPa in oak, or two weeks of root zone {psi}soil [≤]-0.8 MPa and [≤]-0.9 MPa, respectively. Elevated spring VPD further amplified early senescence risk in both species, indicating seasonal legacy effects that may reduce hydraulic function or leaf longevity. Forests with very early and repeated early leaf senescence exhibited elevated canopy mortality in subsequent years, highlighting them as potential warning signals of drought-induced forest decline. K-means clustering of forest stands based on senescence patterns, mortality, and drought sensitivity classified three drought response types - resistant, sensitive, and vulnerable - shaped mainly by local soil water availability and stand structure. Our findings demonstrate that early leaf senescence reflects cumulative drought stress and indicates increased mortality risk, rather than a stress-avoidance strategy. We propose it as a mechanistic indicator that can be incorporated into ecosystem models to improve predictions of forest vulnerability and carbon dynamics under climate extremes, and guide forest management to anticipate climate-induced forest decline.