Aging and stress explain the earlier start of leaf senescence in trees in warmer years: translating the latest findings on senescence regulation into the DP3 model (v1.1)

Abstract. Leaf senescence ends the growing season of deciduous trees, affecting the amount of atmospheric CO2 sequestered by forests. Therefore, some climate models integrate projected leaf senescence dates to simulate the carbon cycle. Here, we developed a process-oriented model of leaf senescence (the “DP3 model”) by testing 34 formulations of the leaf development process based on the latest findings on the regulation of leaf aging and senescence. The period between leaf unfolding and leaf senescence was separated into the subsequent young, mature, and old leaf phases, with particular reactions to leaf aging and cold stress, photoperiod stress, and dry stress. The DP3 model simulates daily rates of aging and stress to predict dates of transition from young to mature to old leaf, senescence induction dates, and leaf senescence dates. This allows new hypotheses regarding the regulation of leaf senescence to be tested. For example, the DP3 model predicted an earlier onset of senescence in warmer conditions, likely due to earlier leaf unfolding (aging) and increased cold and dry stress in spring, together with longer-lasting senescence, likely due to the later accumulation of photoperiod stress relative to leaf development and decreased cold stress in summer and fall, which can be validated through experiments and in situ observations. The DP3 model and compared previous models were equally accurate but less accurate than the Null model (average senescence date observed in the calibration sample). This lower accuracy of the DP3 and compared models is likely due to noise in the visually observed leaf senescence data, which blurs the signal of the leaf senescence process, and to incorrect model formulations. The model errors were similarly affected by climate conditions and location among compared models (including the Null model) and varied mostly due to the leaf senescence data. Noisy leaf senescence data likely force the models to resort to the mean observation, impeding inferences from accuracy-based model comparisons about the leaf senescence process. This calls for revised observation protocols and methods that measure rather than estimate different senescence stages, such as senescence induction and 50 % of the leaves having changed color, e.g., based on greenness, involving digital cameras and automated image assessment.