Climate change-induced shifts in leaf phenology of trees: past and future trends

Forests and atmosphere interact, for example when trees assimilate and respire CO₂ during photosynthesis. Depending on their productivity and net assimilation, forests may function as a carbon sink or source. Further, the amounts of synthesized sugars assigned to growth, reproduction, and defense affect the fitness of the trees and eventually the distribution of species. Climate strongly impacts such forest-atmosphere interactions, which affect forests and in turn feed back to the climate.

For deciduous trees, the beginning and end of the photosynthetically active pe­riod (i.e., the ‘growing season’) relate to spring and autumn leaf phe­nology (i.e., leaf unfolding and leaf senescence), respectively. The climatic conditions during the growing season (i.e., the ‘bioclimate’) are directly and indirectly influenced by climate change, as climate change alters the timing and length of the growing season through shifts in leaf unfolding and leaf senescence. While past changes in leaf unfolding and leaf senescence can be analyzed by in-situ observations, the underlying drivers and particularly possible future changes are often studied with pro­cess-oriented models. However, these models may suffer from a bias to­wards the mean (BTM), which causes the simulated values to be closer to the average than the observations and could result in overly flat simulated trends.

Here I discuss (1) changes in the growing season and bioclimate in Switzerland during recent decades, (2) impacts of the calibration approach on the performance of 21 leaf se­nescence models tested with observations from Central Europe, and (3) effects of the BTM in these models on their performance and projections. Growing seasons have predominately lengthened at elevation-specific rates, which was primarily caused by changes in leaf senescence and increased the number of days with a negative atmospheric water balance at low elevations. Calibrations with the Gen­eralized Simulated Annealing algorithm and with systematically balanced or stratified samples yielded the best performing leaf senescence models, while their performance was most influenced by their structure. The BTM caused the performance of current leaf senescence models to be only slightly better than the performance of a null model that constantly simulates the average of the calibration sample. Standard model comparisons favored models with stronger BTM, while models with weaker BTM projected smaller backward shifts in future leaf senescence. The latter is counter-intuitive, since smaller shifts result from flatter trends and are therefore associated with stronger rather than weaker BTM.

I conclude that (1) the effects of phenological changes on the bioclimate should be considered when studying past and future forest productivity and species composition, (2) inference from process-oriented models to the underlying processes and drivers of leaf senescence is valid, and (3) current leaf senescence projections are highly uncertain and thus unreliable. While it is likely that current projections of future biosphere behavior under global change are distorted by erroneous state-of-the-art leaf senescence models, there is ample need and potential for the development of more accurate process-oriented models.