Climate change-induced shifts in leaf phenology of trees: past and future trends
OPEN ACCESS
Loading...
Author / Producer
Date
2023
Publication Type
Doctoral Thesis
ETH Bibliography
yes
Citations
Altmetric
OPEN ACCESS
Data
Rights / License
Abstract
Forests are coupled to the atmosphere, for example by the carbon and water cycles, which results in forest-atmosphere interactions. Regarding the carbon cycle, forests assimilate and respire CO2 during processes such as photosynthesis and decomposition. Thereby, forests may function as a carbon sink or source, depending on their productivity and net assimilation. In addition, sugars are synthesised during photosynthesis and assigned to growth, reproduction, and defence, which affects the fitness of the trees and eventually the distribution of species. Climate change has strong impacts on these processes, altering the productivity of forests and the distribution of species, which in turn feeds back to the climate. To improve our understanding of these interactions, I studied changes in the photosynthetically active period of trees (i.e., the ‘growing season’) and changes in the climatic conditions to which trees are exposed during their growing season (i.e., the ‘bioclimate’).
For deciduous trees, the beginning and end of the growing season relate to spring and autumn leaf phenology (i.e., leaf unfolding and leaf senescence), respectively. Climate change alters the timing of leaf unfolding and leaf senescence, which in turn influences the timing and length of the growing season. Thus, the bioclimate is directly influenced by climate change and indirectly by changing growing seasons. While past changes in leaf phenology can be analysed by in-situ observations, the underlying drivers and particularly possible future changes can be determined by evaluating simulations with process-oriented models. In my doctoral thesis, I assessed (1) changes in the growing season and bioclimate in Switzerland during recent decades, (2) the impact of the calibration approach on simulated leaf senescence for the past and the future in Central Europe, and (3) biases in these model simulations, together with the resulting distortions in their accuracy and projections.
Chapter 1 focuses on past trends in the growing season and bioclimate, which were estimated with linear-mixed effects models. I studied leaf unfolding and leaf senescence as well as the resulting length of the growing season, while the bioclimate was described by sunshine duration, degree-days, precipitation, and a drought index (i.e., the number of days with a negative atmospheric water balance). The growing season has predominately lengthened, depending on elevation, which was primarily caused by changes in leaf senescence. Drought intensity has generally increased, and the increase has been amplified by changes in leaf unfolding and leaf senescence at low elevations, but weakened at high elevations.
Chapter 2 analyses past and future simulations of leaf senescence by paying special attention to the impact of the calibration approach (i.e., optimisation algorithm and sampling design). The simulations were based on 21 state-of-the-art process-oriented models that were calibrated with five optimisation algorithms, each executed in a normal and an extensive mode, based on at least seven sampling designs. Since some models are driven by leaf unfolding, I assumed a moderate advancement of 0.2 days per year for the projections. To assess model performance, I calculated the root mean squared error. To assess model projections, I calculated the shift between the observations during 1980-1999 and the simulation during 2080-2099. The structure of the leaf senescence models had the largest influence on model performance, followed by optimisation algorithms. Best performance was obtained with the Generalized Simulated Annealing algorithm and with systematically balanced or stratified samples. Compared to 1980-1999, the models predicted a shift in leaf senescence between -13 and +20 days.
Chapter 3 concentrates on distortions in model performance and model projections due to the bias towards the mean in state-of-the-art process-oriented models. This bias results in simulated values that are closer to the average than the observations. Therefore, I expected the model error to increase with increasing difference in the average leaf senescence between the calibration and validation sample. Similarly, I expected the projected shifts to decrease with increasing difference in average leaf senescence between the calibration sample and the observations that served as reference for the shifts. Model performance was assessed by the mean bias, root mean squared error, and Nash-Sutcliffe efficiency. Model projections were assessed by the shift between reference observations (1980-1999) and future simulations (2080-2099). As expected, model performance eroded at model-specific rates with increasing difference between the calibration and validation sample. Different models performed best on validation samples with identical vs. very different average leaf senescence to the calibration sample. The projected shifts of leaf senescence became significantly smaller when the difference between the calibration sample and the reference observations increased. Undistorted shifts were 10-13 days larger than distorted shifts and revealed that leaf senescence was mostly delayed under moderate and extreme climate warming.
Overall, my thesis extends our knowledge on the combined impact of climate and leaf phenology on the bioclimate during the growing season of trees and on the impacts of model structure, calibration approach and bias towards the mean on simulated leaf senescence. Chapter 1 demonstrates that the effect of phenological changes on the bioclimate should be included when studying past and future forest productivity and species composition. Chapter 2 supports conclusions from process-oriented models on underlying processes and drivers of leaf senescence and shows which calibration approaches lead to the most accurate simulations. Chapter 3 illustrates the need and potential for the development of more accurate process-oriented models. Until then, I have to conclude that current state-of-the-art models are unsuitable for reliable projections of leaf senescence.
Permanent link
Publication status
published
External links
Editor
Contributors
Book title
Journal / series
Volume
Pages / Article No.
Publisher
ETH Zurich
Event
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Organisational unit
03535 - Bugmann, Harald / Bugmann, Harald